]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/pci/pci.c
PCI: PM: Skip devices in D0 for suspend-to-idle
[mirror_ubuntu-bionic-kernel.git] / drivers / pci / pci.c
1 /*
2 * PCI Bus Services, see include/linux/pci.h for further explanation.
3 *
4 * Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
5 * David Mosberger-Tang
6 *
7 * Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
8 */
9
10 #include <linux/acpi.h>
11 #include <linux/kernel.h>
12 #include <linux/delay.h>
13 #include <linux/dmi.h>
14 #include <linux/init.h>
15 #include <linux/of.h>
16 #include <linux/of_pci.h>
17 #include <linux/pci.h>
18 #include <linux/pm.h>
19 #include <linux/slab.h>
20 #include <linux/module.h>
21 #include <linux/spinlock.h>
22 #include <linux/string.h>
23 #include <linux/log2.h>
24 #include <linux/logic_pio.h>
25 #include <linux/pci-aspm.h>
26 #include <linux/pm_wakeup.h>
27 #include <linux/interrupt.h>
28 #include <linux/device.h>
29 #include <linux/pm_runtime.h>
30 #include <linux/pci_hotplug.h>
31 #include <linux/vmalloc.h>
32 #include <linux/pci-ats.h>
33 #include <asm/setup.h>
34 #include <asm/dma.h>
35 #include <linux/aer.h>
36 #include "pci.h"
37
38 const char *pci_power_names[] = {
39 "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
40 };
41 EXPORT_SYMBOL_GPL(pci_power_names);
42
43 int isa_dma_bridge_buggy;
44 EXPORT_SYMBOL(isa_dma_bridge_buggy);
45
46 int pci_pci_problems;
47 EXPORT_SYMBOL(pci_pci_problems);
48
49 unsigned int pci_pm_d3_delay;
50
51 static void pci_pme_list_scan(struct work_struct *work);
52
53 static LIST_HEAD(pci_pme_list);
54 static DEFINE_MUTEX(pci_pme_list_mutex);
55 static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
56
57 struct pci_pme_device {
58 struct list_head list;
59 struct pci_dev *dev;
60 };
61
62 #define PME_TIMEOUT 1000 /* How long between PME checks */
63
64 static void pci_dev_d3_sleep(struct pci_dev *dev)
65 {
66 unsigned int delay = dev->d3_delay;
67
68 if (delay < pci_pm_d3_delay)
69 delay = pci_pm_d3_delay;
70
71 if (delay)
72 msleep(delay);
73 }
74
75 #ifdef CONFIG_PCI_DOMAINS
76 int pci_domains_supported = 1;
77 #endif
78
79 #define DEFAULT_CARDBUS_IO_SIZE (256)
80 #define DEFAULT_CARDBUS_MEM_SIZE (64*1024*1024)
81 /* pci=cbmemsize=nnM,cbiosize=nn can override this */
82 unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
83 unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
84
85 #define DEFAULT_HOTPLUG_IO_SIZE (256)
86 #define DEFAULT_HOTPLUG_MEM_SIZE (2*1024*1024)
87 /* pci=hpmemsize=nnM,hpiosize=nn can override this */
88 unsigned long pci_hotplug_io_size = DEFAULT_HOTPLUG_IO_SIZE;
89 unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
90
91 #define DEFAULT_HOTPLUG_BUS_SIZE 1
92 unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
93
94 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
95
96 /*
97 * The default CLS is used if arch didn't set CLS explicitly and not
98 * all pci devices agree on the same value. Arch can override either
99 * the dfl or actual value as it sees fit. Don't forget this is
100 * measured in 32-bit words, not bytes.
101 */
102 u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
103 u8 pci_cache_line_size;
104
105 /*
106 * If we set up a device for bus mastering, we need to check the latency
107 * timer as certain BIOSes forget to set it properly.
108 */
109 unsigned int pcibios_max_latency = 255;
110
111 /* If set, the PCIe ARI capability will not be used. */
112 static bool pcie_ari_disabled;
113
114 /* Disable bridge_d3 for all PCIe ports */
115 static bool pci_bridge_d3_disable;
116 /* Force bridge_d3 for all PCIe ports */
117 static bool pci_bridge_d3_force;
118
119 static int __init pcie_port_pm_setup(char *str)
120 {
121 if (!strcmp(str, "off"))
122 pci_bridge_d3_disable = true;
123 else if (!strcmp(str, "force"))
124 pci_bridge_d3_force = true;
125 return 1;
126 }
127 __setup("pcie_port_pm=", pcie_port_pm_setup);
128
129 /**
130 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
131 * @bus: pointer to PCI bus structure to search
132 *
133 * Given a PCI bus, returns the highest PCI bus number present in the set
134 * including the given PCI bus and its list of child PCI buses.
135 */
136 unsigned char pci_bus_max_busnr(struct pci_bus *bus)
137 {
138 struct pci_bus *tmp;
139 unsigned char max, n;
140
141 max = bus->busn_res.end;
142 list_for_each_entry(tmp, &bus->children, node) {
143 n = pci_bus_max_busnr(tmp);
144 if (n > max)
145 max = n;
146 }
147 return max;
148 }
149 EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
150
151 #ifdef CONFIG_HAS_IOMEM
152 void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
153 {
154 struct resource *res = &pdev->resource[bar];
155
156 /*
157 * Make sure the BAR is actually a memory resource, not an IO resource
158 */
159 if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
160 dev_warn(&pdev->dev, "can't ioremap BAR %d: %pR\n", bar, res);
161 return NULL;
162 }
163 return ioremap_nocache(res->start, resource_size(res));
164 }
165 EXPORT_SYMBOL_GPL(pci_ioremap_bar);
166
167 void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
168 {
169 /*
170 * Make sure the BAR is actually a memory resource, not an IO resource
171 */
172 if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
173 WARN_ON(1);
174 return NULL;
175 }
176 return ioremap_wc(pci_resource_start(pdev, bar),
177 pci_resource_len(pdev, bar));
178 }
179 EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
180 #endif
181
182
183 static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
184 u8 pos, int cap, int *ttl)
185 {
186 u8 id;
187 u16 ent;
188
189 pci_bus_read_config_byte(bus, devfn, pos, &pos);
190
191 while ((*ttl)--) {
192 if (pos < 0x40)
193 break;
194 pos &= ~3;
195 pci_bus_read_config_word(bus, devfn, pos, &ent);
196
197 id = ent & 0xff;
198 if (id == 0xff)
199 break;
200 if (id == cap)
201 return pos;
202 pos = (ent >> 8);
203 }
204 return 0;
205 }
206
207 static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
208 u8 pos, int cap)
209 {
210 int ttl = PCI_FIND_CAP_TTL;
211
212 return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
213 }
214
215 int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
216 {
217 return __pci_find_next_cap(dev->bus, dev->devfn,
218 pos + PCI_CAP_LIST_NEXT, cap);
219 }
220 EXPORT_SYMBOL_GPL(pci_find_next_capability);
221
222 static int __pci_bus_find_cap_start(struct pci_bus *bus,
223 unsigned int devfn, u8 hdr_type)
224 {
225 u16 status;
226
227 pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
228 if (!(status & PCI_STATUS_CAP_LIST))
229 return 0;
230
231 switch (hdr_type) {
232 case PCI_HEADER_TYPE_NORMAL:
233 case PCI_HEADER_TYPE_BRIDGE:
234 return PCI_CAPABILITY_LIST;
235 case PCI_HEADER_TYPE_CARDBUS:
236 return PCI_CB_CAPABILITY_LIST;
237 }
238
239 return 0;
240 }
241
242 /**
243 * pci_find_capability - query for devices' capabilities
244 * @dev: PCI device to query
245 * @cap: capability code
246 *
247 * Tell if a device supports a given PCI capability.
248 * Returns the address of the requested capability structure within the
249 * device's PCI configuration space or 0 in case the device does not
250 * support it. Possible values for @cap:
251 *
252 * %PCI_CAP_ID_PM Power Management
253 * %PCI_CAP_ID_AGP Accelerated Graphics Port
254 * %PCI_CAP_ID_VPD Vital Product Data
255 * %PCI_CAP_ID_SLOTID Slot Identification
256 * %PCI_CAP_ID_MSI Message Signalled Interrupts
257 * %PCI_CAP_ID_CHSWP CompactPCI HotSwap
258 * %PCI_CAP_ID_PCIX PCI-X
259 * %PCI_CAP_ID_EXP PCI Express
260 */
261 int pci_find_capability(struct pci_dev *dev, int cap)
262 {
263 int pos;
264
265 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
266 if (pos)
267 pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
268
269 return pos;
270 }
271 EXPORT_SYMBOL(pci_find_capability);
272
273 /**
274 * pci_bus_find_capability - query for devices' capabilities
275 * @bus: the PCI bus to query
276 * @devfn: PCI device to query
277 * @cap: capability code
278 *
279 * Like pci_find_capability() but works for pci devices that do not have a
280 * pci_dev structure set up yet.
281 *
282 * Returns the address of the requested capability structure within the
283 * device's PCI configuration space or 0 in case the device does not
284 * support it.
285 */
286 int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
287 {
288 int pos;
289 u8 hdr_type;
290
291 pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
292
293 pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
294 if (pos)
295 pos = __pci_find_next_cap(bus, devfn, pos, cap);
296
297 return pos;
298 }
299 EXPORT_SYMBOL(pci_bus_find_capability);
300
301 /**
302 * pci_find_next_ext_capability - Find an extended capability
303 * @dev: PCI device to query
304 * @start: address at which to start looking (0 to start at beginning of list)
305 * @cap: capability code
306 *
307 * Returns the address of the next matching extended capability structure
308 * within the device's PCI configuration space or 0 if the device does
309 * not support it. Some capabilities can occur several times, e.g., the
310 * vendor-specific capability, and this provides a way to find them all.
311 */
312 int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
313 {
314 u32 header;
315 int ttl;
316 int pos = PCI_CFG_SPACE_SIZE;
317
318 /* minimum 8 bytes per capability */
319 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
320
321 if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
322 return 0;
323
324 if (start)
325 pos = start;
326
327 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
328 return 0;
329
330 /*
331 * If we have no capabilities, this is indicated by cap ID,
332 * cap version and next pointer all being 0.
333 */
334 if (header == 0)
335 return 0;
336
337 while (ttl-- > 0) {
338 if (PCI_EXT_CAP_ID(header) == cap && pos != start)
339 return pos;
340
341 pos = PCI_EXT_CAP_NEXT(header);
342 if (pos < PCI_CFG_SPACE_SIZE)
343 break;
344
345 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
346 break;
347 }
348
349 return 0;
350 }
351 EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
352
353 /**
354 * pci_find_ext_capability - Find an extended capability
355 * @dev: PCI device to query
356 * @cap: capability code
357 *
358 * Returns the address of the requested extended capability structure
359 * within the device's PCI configuration space or 0 if the device does
360 * not support it. Possible values for @cap:
361 *
362 * %PCI_EXT_CAP_ID_ERR Advanced Error Reporting
363 * %PCI_EXT_CAP_ID_VC Virtual Channel
364 * %PCI_EXT_CAP_ID_DSN Device Serial Number
365 * %PCI_EXT_CAP_ID_PWR Power Budgeting
366 */
367 int pci_find_ext_capability(struct pci_dev *dev, int cap)
368 {
369 return pci_find_next_ext_capability(dev, 0, cap);
370 }
371 EXPORT_SYMBOL_GPL(pci_find_ext_capability);
372
373 static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
374 {
375 int rc, ttl = PCI_FIND_CAP_TTL;
376 u8 cap, mask;
377
378 if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
379 mask = HT_3BIT_CAP_MASK;
380 else
381 mask = HT_5BIT_CAP_MASK;
382
383 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
384 PCI_CAP_ID_HT, &ttl);
385 while (pos) {
386 rc = pci_read_config_byte(dev, pos + 3, &cap);
387 if (rc != PCIBIOS_SUCCESSFUL)
388 return 0;
389
390 if ((cap & mask) == ht_cap)
391 return pos;
392
393 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
394 pos + PCI_CAP_LIST_NEXT,
395 PCI_CAP_ID_HT, &ttl);
396 }
397
398 return 0;
399 }
400 /**
401 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
402 * @dev: PCI device to query
403 * @pos: Position from which to continue searching
404 * @ht_cap: Hypertransport capability code
405 *
406 * To be used in conjunction with pci_find_ht_capability() to search for
407 * all capabilities matching @ht_cap. @pos should always be a value returned
408 * from pci_find_ht_capability().
409 *
410 * NB. To be 100% safe against broken PCI devices, the caller should take
411 * steps to avoid an infinite loop.
412 */
413 int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
414 {
415 return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
416 }
417 EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
418
419 /**
420 * pci_find_ht_capability - query a device's Hypertransport capabilities
421 * @dev: PCI device to query
422 * @ht_cap: Hypertransport capability code
423 *
424 * Tell if a device supports a given Hypertransport capability.
425 * Returns an address within the device's PCI configuration space
426 * or 0 in case the device does not support the request capability.
427 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
428 * which has a Hypertransport capability matching @ht_cap.
429 */
430 int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
431 {
432 int pos;
433
434 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
435 if (pos)
436 pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
437
438 return pos;
439 }
440 EXPORT_SYMBOL_GPL(pci_find_ht_capability);
441
442 /**
443 * pci_find_parent_resource - return resource region of parent bus of given region
444 * @dev: PCI device structure contains resources to be searched
445 * @res: child resource record for which parent is sought
446 *
447 * For given resource region of given device, return the resource
448 * region of parent bus the given region is contained in.
449 */
450 struct resource *pci_find_parent_resource(const struct pci_dev *dev,
451 struct resource *res)
452 {
453 const struct pci_bus *bus = dev->bus;
454 struct resource *r;
455 int i;
456
457 pci_bus_for_each_resource(bus, r, i) {
458 if (!r)
459 continue;
460 if (resource_contains(r, res)) {
461
462 /*
463 * If the window is prefetchable but the BAR is
464 * not, the allocator made a mistake.
465 */
466 if (r->flags & IORESOURCE_PREFETCH &&
467 !(res->flags & IORESOURCE_PREFETCH))
468 return NULL;
469
470 /*
471 * If we're below a transparent bridge, there may
472 * be both a positively-decoded aperture and a
473 * subtractively-decoded region that contain the BAR.
474 * We want the positively-decoded one, so this depends
475 * on pci_bus_for_each_resource() giving us those
476 * first.
477 */
478 return r;
479 }
480 }
481 return NULL;
482 }
483 EXPORT_SYMBOL(pci_find_parent_resource);
484
485 /**
486 * pci_find_resource - Return matching PCI device resource
487 * @dev: PCI device to query
488 * @res: Resource to look for
489 *
490 * Goes over standard PCI resources (BARs) and checks if the given resource
491 * is partially or fully contained in any of them. In that case the
492 * matching resource is returned, %NULL otherwise.
493 */
494 struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
495 {
496 int i;
497
498 for (i = 0; i < PCI_ROM_RESOURCE; i++) {
499 struct resource *r = &dev->resource[i];
500
501 if (r->start && resource_contains(r, res))
502 return r;
503 }
504
505 return NULL;
506 }
507 EXPORT_SYMBOL(pci_find_resource);
508
509 /**
510 * pci_find_pcie_root_port - return PCIe Root Port
511 * @dev: PCI device to query
512 *
513 * Traverse up the parent chain and return the PCIe Root Port PCI Device
514 * for a given PCI Device.
515 */
516 struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
517 {
518 struct pci_dev *bridge, *highest_pcie_bridge = dev;
519
520 bridge = pci_upstream_bridge(dev);
521 while (bridge && pci_is_pcie(bridge)) {
522 highest_pcie_bridge = bridge;
523 bridge = pci_upstream_bridge(bridge);
524 }
525
526 if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
527 return NULL;
528
529 return highest_pcie_bridge;
530 }
531 EXPORT_SYMBOL(pci_find_pcie_root_port);
532
533 /**
534 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
535 * @dev: the PCI device to operate on
536 * @pos: config space offset of status word
537 * @mask: mask of bit(s) to care about in status word
538 *
539 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
540 */
541 int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
542 {
543 int i;
544
545 /* Wait for Transaction Pending bit clean */
546 for (i = 0; i < 4; i++) {
547 u16 status;
548 if (i)
549 msleep((1 << (i - 1)) * 100);
550
551 pci_read_config_word(dev, pos, &status);
552 if (!(status & mask))
553 return 1;
554 }
555
556 return 0;
557 }
558
559 /**
560 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
561 * @dev: PCI device to have its BARs restored
562 *
563 * Restore the BAR values for a given device, so as to make it
564 * accessible by its driver.
565 */
566 static void pci_restore_bars(struct pci_dev *dev)
567 {
568 int i;
569
570 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
571 pci_update_resource(dev, i);
572 }
573
574 static const struct pci_platform_pm_ops *pci_platform_pm;
575
576 int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
577 {
578 if (!ops->is_manageable || !ops->set_state || !ops->get_state ||
579 !ops->choose_state || !ops->set_wakeup || !ops->need_resume)
580 return -EINVAL;
581 pci_platform_pm = ops;
582 return 0;
583 }
584
585 static inline bool platform_pci_power_manageable(struct pci_dev *dev)
586 {
587 return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
588 }
589
590 static inline int platform_pci_set_power_state(struct pci_dev *dev,
591 pci_power_t t)
592 {
593 return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
594 }
595
596 static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
597 {
598 return pci_platform_pm ? pci_platform_pm->get_state(dev) : PCI_UNKNOWN;
599 }
600
601 static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
602 {
603 return pci_platform_pm ?
604 pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
605 }
606
607 static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
608 {
609 return pci_platform_pm ?
610 pci_platform_pm->set_wakeup(dev, enable) : -ENODEV;
611 }
612
613 static inline bool platform_pci_need_resume(struct pci_dev *dev)
614 {
615 return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
616 }
617
618 /**
619 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
620 * given PCI device
621 * @dev: PCI device to handle.
622 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
623 *
624 * RETURN VALUE:
625 * -EINVAL if the requested state is invalid.
626 * -EIO if device does not support PCI PM or its PM capabilities register has a
627 * wrong version, or device doesn't support the requested state.
628 * 0 if device already is in the requested state.
629 * 0 if device's power state has been successfully changed.
630 */
631 static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
632 {
633 u16 pmcsr;
634 bool need_restore = false;
635
636 /* Check if we're already there */
637 if (dev->current_state == state)
638 return 0;
639
640 if (!dev->pm_cap)
641 return -EIO;
642
643 if (state < PCI_D0 || state > PCI_D3hot)
644 return -EINVAL;
645
646 /* Validate current state:
647 * Can enter D0 from any state, but if we can only go deeper
648 * to sleep if we're already in a low power state
649 */
650 if (state != PCI_D0 && dev->current_state <= PCI_D3cold
651 && dev->current_state > state) {
652 dev_err(&dev->dev, "invalid power transition (from state %d to %d)\n",
653 dev->current_state, state);
654 return -EINVAL;
655 }
656
657 /* check if this device supports the desired state */
658 if ((state == PCI_D1 && !dev->d1_support)
659 || (state == PCI_D2 && !dev->d2_support))
660 return -EIO;
661
662 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
663
664 /* If we're (effectively) in D3, force entire word to 0.
665 * This doesn't affect PME_Status, disables PME_En, and
666 * sets PowerState to 0.
667 */
668 switch (dev->current_state) {
669 case PCI_D0:
670 case PCI_D1:
671 case PCI_D2:
672 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
673 pmcsr |= state;
674 break;
675 case PCI_D3hot:
676 case PCI_D3cold:
677 case PCI_UNKNOWN: /* Boot-up */
678 if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
679 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
680 need_restore = true;
681 /* Fall-through: force to D0 */
682 default:
683 pmcsr = 0;
684 break;
685 }
686
687 /* enter specified state */
688 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
689
690 /* Mandatory power management transition delays */
691 /* see PCI PM 1.1 5.6.1 table 18 */
692 if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
693 pci_dev_d3_sleep(dev);
694 else if (state == PCI_D2 || dev->current_state == PCI_D2)
695 udelay(PCI_PM_D2_DELAY);
696
697 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
698 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
699 if (dev->current_state != state && printk_ratelimit())
700 dev_info(&dev->dev, "Refused to change power state, currently in D%d\n",
701 dev->current_state);
702
703 /*
704 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
705 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
706 * from D3hot to D0 _may_ perform an internal reset, thereby
707 * going to "D0 Uninitialized" rather than "D0 Initialized".
708 * For example, at least some versions of the 3c905B and the
709 * 3c556B exhibit this behaviour.
710 *
711 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
712 * devices in a D3hot state at boot. Consequently, we need to
713 * restore at least the BARs so that the device will be
714 * accessible to its driver.
715 */
716 if (need_restore)
717 pci_restore_bars(dev);
718
719 if (dev->bus->self)
720 pcie_aspm_pm_state_change(dev->bus->self);
721
722 return 0;
723 }
724
725 /**
726 * pci_update_current_state - Read power state of given device and cache it
727 * @dev: PCI device to handle.
728 * @state: State to cache in case the device doesn't have the PM capability
729 *
730 * The power state is read from the PMCSR register, which however is
731 * inaccessible in D3cold. The platform firmware is therefore queried first
732 * to detect accessibility of the register. In case the platform firmware
733 * reports an incorrect state or the device isn't power manageable by the
734 * platform at all, we try to detect D3cold by testing accessibility of the
735 * vendor ID in config space.
736 */
737 void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
738 {
739 if (platform_pci_get_power_state(dev) == PCI_D3cold ||
740 !pci_device_is_present(dev)) {
741 dev->current_state = PCI_D3cold;
742 } else if (dev->pm_cap) {
743 u16 pmcsr;
744
745 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
746 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
747 } else {
748 dev->current_state = state;
749 }
750 }
751
752 /**
753 * pci_power_up - Put the given device into D0 forcibly
754 * @dev: PCI device to power up
755 */
756 void pci_power_up(struct pci_dev *dev)
757 {
758 if (platform_pci_power_manageable(dev))
759 platform_pci_set_power_state(dev, PCI_D0);
760
761 pci_raw_set_power_state(dev, PCI_D0);
762 pci_update_current_state(dev, PCI_D0);
763 }
764
765 /**
766 * pci_platform_power_transition - Use platform to change device power state
767 * @dev: PCI device to handle.
768 * @state: State to put the device into.
769 */
770 static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
771 {
772 int error;
773
774 if (platform_pci_power_manageable(dev)) {
775 error = platform_pci_set_power_state(dev, state);
776 if (!error)
777 pci_update_current_state(dev, state);
778 } else
779 error = -ENODEV;
780
781 if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
782 dev->current_state = PCI_D0;
783
784 return error;
785 }
786
787 /**
788 * pci_wakeup - Wake up a PCI device
789 * @pci_dev: Device to handle.
790 * @ign: ignored parameter
791 */
792 static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
793 {
794 pci_wakeup_event(pci_dev);
795 pm_request_resume(&pci_dev->dev);
796 return 0;
797 }
798
799 /**
800 * pci_wakeup_bus - Walk given bus and wake up devices on it
801 * @bus: Top bus of the subtree to walk.
802 */
803 static void pci_wakeup_bus(struct pci_bus *bus)
804 {
805 if (bus)
806 pci_walk_bus(bus, pci_wakeup, NULL);
807 }
808
809 /**
810 * __pci_start_power_transition - Start power transition of a PCI device
811 * @dev: PCI device to handle.
812 * @state: State to put the device into.
813 */
814 static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
815 {
816 if (state == PCI_D0) {
817 pci_platform_power_transition(dev, PCI_D0);
818 /*
819 * Mandatory power management transition delays, see
820 * PCI Express Base Specification Revision 2.0 Section
821 * 6.6.1: Conventional Reset. Do not delay for
822 * devices powered on/off by corresponding bridge,
823 * because have already delayed for the bridge.
824 */
825 if (dev->runtime_d3cold) {
826 if (dev->d3cold_delay)
827 msleep(dev->d3cold_delay);
828 /*
829 * When powering on a bridge from D3cold, the
830 * whole hierarchy may be powered on into
831 * D0uninitialized state, resume them to give
832 * them a chance to suspend again
833 */
834 pci_wakeup_bus(dev->subordinate);
835 }
836 }
837 }
838
839 /**
840 * __pci_dev_set_current_state - Set current state of a PCI device
841 * @dev: Device to handle
842 * @data: pointer to state to be set
843 */
844 static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
845 {
846 pci_power_t state = *(pci_power_t *)data;
847
848 dev->current_state = state;
849 return 0;
850 }
851
852 /**
853 * __pci_bus_set_current_state - Walk given bus and set current state of devices
854 * @bus: Top bus of the subtree to walk.
855 * @state: state to be set
856 */
857 static void __pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
858 {
859 if (bus)
860 pci_walk_bus(bus, __pci_dev_set_current_state, &state);
861 }
862
863 /**
864 * __pci_complete_power_transition - Complete power transition of a PCI device
865 * @dev: PCI device to handle.
866 * @state: State to put the device into.
867 *
868 * This function should not be called directly by device drivers.
869 */
870 int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
871 {
872 int ret;
873
874 if (state <= PCI_D0)
875 return -EINVAL;
876 ret = pci_platform_power_transition(dev, state);
877 /* Power off the bridge may power off the whole hierarchy */
878 if (!ret && state == PCI_D3cold)
879 __pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
880 return ret;
881 }
882 EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
883
884 /**
885 * pci_set_power_state - Set the power state of a PCI device
886 * @dev: PCI device to handle.
887 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
888 *
889 * Transition a device to a new power state, using the platform firmware and/or
890 * the device's PCI PM registers.
891 *
892 * RETURN VALUE:
893 * -EINVAL if the requested state is invalid.
894 * -EIO if device does not support PCI PM or its PM capabilities register has a
895 * wrong version, or device doesn't support the requested state.
896 * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
897 * 0 if device already is in the requested state.
898 * 0 if the transition is to D3 but D3 is not supported.
899 * 0 if device's power state has been successfully changed.
900 */
901 int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
902 {
903 int error;
904
905 /* bound the state we're entering */
906 if (state > PCI_D3cold)
907 state = PCI_D3cold;
908 else if (state < PCI_D0)
909 state = PCI_D0;
910 else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
911 /*
912 * If the device or the parent bridge do not support PCI PM,
913 * ignore the request if we're doing anything other than putting
914 * it into D0 (which would only happen on boot).
915 */
916 return 0;
917
918 /* Check if we're already there */
919 if (dev->current_state == state)
920 return 0;
921
922 __pci_start_power_transition(dev, state);
923
924 /* This device is quirked not to be put into D3, so
925 don't put it in D3 */
926 if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
927 return 0;
928
929 /*
930 * To put device in D3cold, we put device into D3hot in native
931 * way, then put device into D3cold with platform ops
932 */
933 error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
934 PCI_D3hot : state);
935
936 if (!__pci_complete_power_transition(dev, state))
937 error = 0;
938
939 return error;
940 }
941 EXPORT_SYMBOL(pci_set_power_state);
942
943 /**
944 * pci_choose_state - Choose the power state of a PCI device
945 * @dev: PCI device to be suspended
946 * @state: target sleep state for the whole system. This is the value
947 * that is passed to suspend() function.
948 *
949 * Returns PCI power state suitable for given device and given system
950 * message.
951 */
952
953 pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
954 {
955 pci_power_t ret;
956
957 if (!dev->pm_cap)
958 return PCI_D0;
959
960 ret = platform_pci_choose_state(dev);
961 if (ret != PCI_POWER_ERROR)
962 return ret;
963
964 switch (state.event) {
965 case PM_EVENT_ON:
966 return PCI_D0;
967 case PM_EVENT_FREEZE:
968 case PM_EVENT_PRETHAW:
969 /* REVISIT both freeze and pre-thaw "should" use D0 */
970 case PM_EVENT_SUSPEND:
971 case PM_EVENT_HIBERNATE:
972 return PCI_D3hot;
973 default:
974 dev_info(&dev->dev, "unrecognized suspend event %d\n",
975 state.event);
976 BUG();
977 }
978 return PCI_D0;
979 }
980 EXPORT_SYMBOL(pci_choose_state);
981
982 #define PCI_EXP_SAVE_REGS 7
983
984 static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
985 u16 cap, bool extended)
986 {
987 struct pci_cap_saved_state *tmp;
988
989 hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
990 if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
991 return tmp;
992 }
993 return NULL;
994 }
995
996 struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
997 {
998 return _pci_find_saved_cap(dev, cap, false);
999 }
1000
1001 struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
1002 {
1003 return _pci_find_saved_cap(dev, cap, true);
1004 }
1005
1006 static int pci_save_pcie_state(struct pci_dev *dev)
1007 {
1008 int i = 0;
1009 struct pci_cap_saved_state *save_state;
1010 u16 *cap;
1011
1012 if (!pci_is_pcie(dev))
1013 return 0;
1014
1015 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1016 if (!save_state) {
1017 dev_err(&dev->dev, "buffer not found in %s\n", __func__);
1018 return -ENOMEM;
1019 }
1020
1021 cap = (u16 *)&save_state->cap.data[0];
1022 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1023 pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1024 pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1025 pcie_capability_read_word(dev, PCI_EXP_RTCTL, &cap[i++]);
1026 pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1027 pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1028 pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1029
1030 return 0;
1031 }
1032
1033 static void pci_restore_pcie_state(struct pci_dev *dev)
1034 {
1035 int i = 0;
1036 struct pci_cap_saved_state *save_state;
1037 u16 *cap;
1038
1039 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1040 if (!save_state)
1041 return;
1042
1043 cap = (u16 *)&save_state->cap.data[0];
1044 pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1045 pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1046 pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1047 pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1048 pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1049 pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1050 pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1051 }
1052
1053
1054 static int pci_save_pcix_state(struct pci_dev *dev)
1055 {
1056 int pos;
1057 struct pci_cap_saved_state *save_state;
1058
1059 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1060 if (!pos)
1061 return 0;
1062
1063 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1064 if (!save_state) {
1065 dev_err(&dev->dev, "buffer not found in %s\n", __func__);
1066 return -ENOMEM;
1067 }
1068
1069 pci_read_config_word(dev, pos + PCI_X_CMD,
1070 (u16 *)save_state->cap.data);
1071
1072 return 0;
1073 }
1074
1075 static void pci_restore_pcix_state(struct pci_dev *dev)
1076 {
1077 int i = 0, pos;
1078 struct pci_cap_saved_state *save_state;
1079 u16 *cap;
1080
1081 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1082 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1083 if (!save_state || !pos)
1084 return;
1085 cap = (u16 *)&save_state->cap.data[0];
1086
1087 pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1088 }
1089
1090
1091 /**
1092 * pci_save_state - save the PCI configuration space of a device before suspending
1093 * @dev: - PCI device that we're dealing with
1094 */
1095 int pci_save_state(struct pci_dev *dev)
1096 {
1097 int i;
1098 /* XXX: 100% dword access ok here? */
1099 for (i = 0; i < 16; i++)
1100 pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1101 dev->state_saved = true;
1102
1103 i = pci_save_pcie_state(dev);
1104 if (i != 0)
1105 return i;
1106
1107 i = pci_save_pcix_state(dev);
1108 if (i != 0)
1109 return i;
1110
1111 return pci_save_vc_state(dev);
1112 }
1113 EXPORT_SYMBOL(pci_save_state);
1114
1115 static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1116 u32 saved_val, int retry, bool force)
1117 {
1118 u32 val;
1119
1120 pci_read_config_dword(pdev, offset, &val);
1121 if (!force && val == saved_val)
1122 return;
1123
1124 for (;;) {
1125 dev_dbg(&pdev->dev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1126 offset, val, saved_val);
1127 pci_write_config_dword(pdev, offset, saved_val);
1128 if (retry-- <= 0)
1129 return;
1130
1131 pci_read_config_dword(pdev, offset, &val);
1132 if (val == saved_val)
1133 return;
1134
1135 mdelay(1);
1136 }
1137 }
1138
1139 static void pci_restore_config_space_range(struct pci_dev *pdev,
1140 int start, int end, int retry,
1141 bool force)
1142 {
1143 int index;
1144
1145 for (index = end; index >= start; index--)
1146 pci_restore_config_dword(pdev, 4 * index,
1147 pdev->saved_config_space[index],
1148 retry, force);
1149 }
1150
1151 static void pci_restore_config_space(struct pci_dev *pdev)
1152 {
1153 if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1154 pci_restore_config_space_range(pdev, 10, 15, 0, false);
1155 /* Restore BARs before the command register. */
1156 pci_restore_config_space_range(pdev, 4, 9, 10, false);
1157 pci_restore_config_space_range(pdev, 0, 3, 0, false);
1158 } else if (pdev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1159 pci_restore_config_space_range(pdev, 12, 15, 0, false);
1160
1161 /*
1162 * Force rewriting of prefetch registers to avoid S3 resume
1163 * issues on Intel PCI bridges that occur when these
1164 * registers are not explicitly written.
1165 */
1166 pci_restore_config_space_range(pdev, 9, 11, 0, true);
1167 pci_restore_config_space_range(pdev, 0, 8, 0, false);
1168 } else {
1169 pci_restore_config_space_range(pdev, 0, 15, 0, false);
1170 }
1171 }
1172
1173 static void pci_restore_rebar_state(struct pci_dev *pdev)
1174 {
1175 unsigned int pos, nbars, i;
1176 u32 ctrl;
1177
1178 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
1179 if (!pos)
1180 return;
1181
1182 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1183 nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
1184 PCI_REBAR_CTRL_NBAR_SHIFT;
1185
1186 for (i = 0; i < nbars; i++, pos += 8) {
1187 struct resource *res;
1188 int bar_idx, size;
1189
1190 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1191 bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
1192 res = pdev->resource + bar_idx;
1193 size = order_base_2((resource_size(res) >> 20) | 1) - 1;
1194 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
1195 ctrl |= size << 8;
1196 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
1197 }
1198 }
1199
1200 /**
1201 * pci_restore_state - Restore the saved state of a PCI device
1202 * @dev: - PCI device that we're dealing with
1203 */
1204 void pci_restore_state(struct pci_dev *dev)
1205 {
1206 if (!dev->state_saved)
1207 return;
1208
1209 /* PCI Express register must be restored first */
1210 pci_restore_pcie_state(dev);
1211 pci_restore_pasid_state(dev);
1212 pci_restore_pri_state(dev);
1213 pci_restore_ats_state(dev);
1214 pci_restore_vc_state(dev);
1215 pci_restore_rebar_state(dev);
1216
1217 pci_cleanup_aer_error_status_regs(dev);
1218
1219 pci_restore_config_space(dev);
1220
1221 pci_restore_pcix_state(dev);
1222 pci_restore_msi_state(dev);
1223
1224 /* Restore ACS and IOV configuration state */
1225 pci_enable_acs(dev);
1226 pci_restore_iov_state(dev);
1227
1228 dev->state_saved = false;
1229 }
1230 EXPORT_SYMBOL(pci_restore_state);
1231
1232 struct pci_saved_state {
1233 u32 config_space[16];
1234 struct pci_cap_saved_data cap[0];
1235 };
1236
1237 /**
1238 * pci_store_saved_state - Allocate and return an opaque struct containing
1239 * the device saved state.
1240 * @dev: PCI device that we're dealing with
1241 *
1242 * Return NULL if no state or error.
1243 */
1244 struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1245 {
1246 struct pci_saved_state *state;
1247 struct pci_cap_saved_state *tmp;
1248 struct pci_cap_saved_data *cap;
1249 size_t size;
1250
1251 if (!dev->state_saved)
1252 return NULL;
1253
1254 size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1255
1256 hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1257 size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1258
1259 state = kzalloc(size, GFP_KERNEL);
1260 if (!state)
1261 return NULL;
1262
1263 memcpy(state->config_space, dev->saved_config_space,
1264 sizeof(state->config_space));
1265
1266 cap = state->cap;
1267 hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1268 size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1269 memcpy(cap, &tmp->cap, len);
1270 cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1271 }
1272 /* Empty cap_save terminates list */
1273
1274 return state;
1275 }
1276 EXPORT_SYMBOL_GPL(pci_store_saved_state);
1277
1278 /**
1279 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1280 * @dev: PCI device that we're dealing with
1281 * @state: Saved state returned from pci_store_saved_state()
1282 */
1283 int pci_load_saved_state(struct pci_dev *dev,
1284 struct pci_saved_state *state)
1285 {
1286 struct pci_cap_saved_data *cap;
1287
1288 dev->state_saved = false;
1289
1290 if (!state)
1291 return 0;
1292
1293 memcpy(dev->saved_config_space, state->config_space,
1294 sizeof(state->config_space));
1295
1296 cap = state->cap;
1297 while (cap->size) {
1298 struct pci_cap_saved_state *tmp;
1299
1300 tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1301 if (!tmp || tmp->cap.size != cap->size)
1302 return -EINVAL;
1303
1304 memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1305 cap = (struct pci_cap_saved_data *)((u8 *)cap +
1306 sizeof(struct pci_cap_saved_data) + cap->size);
1307 }
1308
1309 dev->state_saved = true;
1310 return 0;
1311 }
1312 EXPORT_SYMBOL_GPL(pci_load_saved_state);
1313
1314 /**
1315 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1316 * and free the memory allocated for it.
1317 * @dev: PCI device that we're dealing with
1318 * @state: Pointer to saved state returned from pci_store_saved_state()
1319 */
1320 int pci_load_and_free_saved_state(struct pci_dev *dev,
1321 struct pci_saved_state **state)
1322 {
1323 int ret = pci_load_saved_state(dev, *state);
1324 kfree(*state);
1325 *state = NULL;
1326 return ret;
1327 }
1328 EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1329
1330 int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1331 {
1332 return pci_enable_resources(dev, bars);
1333 }
1334
1335 static int do_pci_enable_device(struct pci_dev *dev, int bars)
1336 {
1337 int err;
1338 struct pci_dev *bridge;
1339 u16 cmd;
1340 u8 pin;
1341
1342 err = pci_set_power_state(dev, PCI_D0);
1343 if (err < 0 && err != -EIO)
1344 return err;
1345
1346 bridge = pci_upstream_bridge(dev);
1347 if (bridge)
1348 pcie_aspm_powersave_config_link(bridge);
1349
1350 err = pcibios_enable_device(dev, bars);
1351 if (err < 0)
1352 return err;
1353 pci_fixup_device(pci_fixup_enable, dev);
1354
1355 if (dev->msi_enabled || dev->msix_enabled)
1356 return 0;
1357
1358 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1359 if (pin) {
1360 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1361 if (cmd & PCI_COMMAND_INTX_DISABLE)
1362 pci_write_config_word(dev, PCI_COMMAND,
1363 cmd & ~PCI_COMMAND_INTX_DISABLE);
1364 }
1365
1366 return 0;
1367 }
1368
1369 /**
1370 * pci_reenable_device - Resume abandoned device
1371 * @dev: PCI device to be resumed
1372 *
1373 * Note this function is a backend of pci_default_resume and is not supposed
1374 * to be called by normal code, write proper resume handler and use it instead.
1375 */
1376 int pci_reenable_device(struct pci_dev *dev)
1377 {
1378 if (pci_is_enabled(dev))
1379 return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1380 return 0;
1381 }
1382 EXPORT_SYMBOL(pci_reenable_device);
1383
1384 static void pci_enable_bridge(struct pci_dev *dev)
1385 {
1386 struct pci_dev *bridge;
1387 int retval;
1388
1389 bridge = pci_upstream_bridge(dev);
1390 if (bridge)
1391 pci_enable_bridge(bridge);
1392
1393 if (pci_is_enabled(dev)) {
1394 if (!dev->is_busmaster)
1395 pci_set_master(dev);
1396 return;
1397 }
1398
1399 retval = pci_enable_device(dev);
1400 if (retval)
1401 dev_err(&dev->dev, "Error enabling bridge (%d), continuing\n",
1402 retval);
1403 pci_set_master(dev);
1404 }
1405
1406 static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1407 {
1408 struct pci_dev *bridge;
1409 int err;
1410 int i, bars = 0;
1411
1412 /*
1413 * Power state could be unknown at this point, either due to a fresh
1414 * boot or a device removal call. So get the current power state
1415 * so that things like MSI message writing will behave as expected
1416 * (e.g. if the device really is in D0 at enable time).
1417 */
1418 if (dev->pm_cap) {
1419 u16 pmcsr;
1420 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1421 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1422 }
1423
1424 if (atomic_inc_return(&dev->enable_cnt) > 1)
1425 return 0; /* already enabled */
1426
1427 bridge = pci_upstream_bridge(dev);
1428 if (bridge)
1429 pci_enable_bridge(bridge);
1430
1431 /* only skip sriov related */
1432 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1433 if (dev->resource[i].flags & flags)
1434 bars |= (1 << i);
1435 for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1436 if (dev->resource[i].flags & flags)
1437 bars |= (1 << i);
1438
1439 err = do_pci_enable_device(dev, bars);
1440 if (err < 0)
1441 atomic_dec(&dev->enable_cnt);
1442 return err;
1443 }
1444
1445 /**
1446 * pci_enable_device_io - Initialize a device for use with IO space
1447 * @dev: PCI device to be initialized
1448 *
1449 * Initialize device before it's used by a driver. Ask low-level code
1450 * to enable I/O resources. Wake up the device if it was suspended.
1451 * Beware, this function can fail.
1452 */
1453 int pci_enable_device_io(struct pci_dev *dev)
1454 {
1455 return pci_enable_device_flags(dev, IORESOURCE_IO);
1456 }
1457 EXPORT_SYMBOL(pci_enable_device_io);
1458
1459 /**
1460 * pci_enable_device_mem - Initialize a device for use with Memory space
1461 * @dev: PCI device to be initialized
1462 *
1463 * Initialize device before it's used by a driver. Ask low-level code
1464 * to enable Memory resources. Wake up the device if it was suspended.
1465 * Beware, this function can fail.
1466 */
1467 int pci_enable_device_mem(struct pci_dev *dev)
1468 {
1469 return pci_enable_device_flags(dev, IORESOURCE_MEM);
1470 }
1471 EXPORT_SYMBOL(pci_enable_device_mem);
1472
1473 /**
1474 * pci_enable_device - Initialize device before it's used by a driver.
1475 * @dev: PCI device to be initialized
1476 *
1477 * Initialize device before it's used by a driver. Ask low-level code
1478 * to enable I/O and memory. Wake up the device if it was suspended.
1479 * Beware, this function can fail.
1480 *
1481 * Note we don't actually enable the device many times if we call
1482 * this function repeatedly (we just increment the count).
1483 */
1484 int pci_enable_device(struct pci_dev *dev)
1485 {
1486 return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1487 }
1488 EXPORT_SYMBOL(pci_enable_device);
1489
1490 /*
1491 * Managed PCI resources. This manages device on/off, intx/msi/msix
1492 * on/off and BAR regions. pci_dev itself records msi/msix status, so
1493 * there's no need to track it separately. pci_devres is initialized
1494 * when a device is enabled using managed PCI device enable interface.
1495 */
1496 struct pci_devres {
1497 unsigned int enabled:1;
1498 unsigned int pinned:1;
1499 unsigned int orig_intx:1;
1500 unsigned int restore_intx:1;
1501 unsigned int mwi:1;
1502 u32 region_mask;
1503 };
1504
1505 static void pcim_release(struct device *gendev, void *res)
1506 {
1507 struct pci_dev *dev = to_pci_dev(gendev);
1508 struct pci_devres *this = res;
1509 int i;
1510
1511 if (dev->msi_enabled)
1512 pci_disable_msi(dev);
1513 if (dev->msix_enabled)
1514 pci_disable_msix(dev);
1515
1516 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1517 if (this->region_mask & (1 << i))
1518 pci_release_region(dev, i);
1519
1520 if (this->mwi)
1521 pci_clear_mwi(dev);
1522
1523 if (this->restore_intx)
1524 pci_intx(dev, this->orig_intx);
1525
1526 if (this->enabled && !this->pinned)
1527 pci_disable_device(dev);
1528 }
1529
1530 static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1531 {
1532 struct pci_devres *dr, *new_dr;
1533
1534 dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1535 if (dr)
1536 return dr;
1537
1538 new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1539 if (!new_dr)
1540 return NULL;
1541 return devres_get(&pdev->dev, new_dr, NULL, NULL);
1542 }
1543
1544 static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1545 {
1546 if (pci_is_managed(pdev))
1547 return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1548 return NULL;
1549 }
1550
1551 /**
1552 * pcim_enable_device - Managed pci_enable_device()
1553 * @pdev: PCI device to be initialized
1554 *
1555 * Managed pci_enable_device().
1556 */
1557 int pcim_enable_device(struct pci_dev *pdev)
1558 {
1559 struct pci_devres *dr;
1560 int rc;
1561
1562 dr = get_pci_dr(pdev);
1563 if (unlikely(!dr))
1564 return -ENOMEM;
1565 if (dr->enabled)
1566 return 0;
1567
1568 rc = pci_enable_device(pdev);
1569 if (!rc) {
1570 pdev->is_managed = 1;
1571 dr->enabled = 1;
1572 }
1573 return rc;
1574 }
1575 EXPORT_SYMBOL(pcim_enable_device);
1576
1577 /**
1578 * pcim_pin_device - Pin managed PCI device
1579 * @pdev: PCI device to pin
1580 *
1581 * Pin managed PCI device @pdev. Pinned device won't be disabled on
1582 * driver detach. @pdev must have been enabled with
1583 * pcim_enable_device().
1584 */
1585 void pcim_pin_device(struct pci_dev *pdev)
1586 {
1587 struct pci_devres *dr;
1588
1589 dr = find_pci_dr(pdev);
1590 WARN_ON(!dr || !dr->enabled);
1591 if (dr)
1592 dr->pinned = 1;
1593 }
1594 EXPORT_SYMBOL(pcim_pin_device);
1595
1596 /*
1597 * pcibios_add_device - provide arch specific hooks when adding device dev
1598 * @dev: the PCI device being added
1599 *
1600 * Permits the platform to provide architecture specific functionality when
1601 * devices are added. This is the default implementation. Architecture
1602 * implementations can override this.
1603 */
1604 int __weak pcibios_add_device(struct pci_dev *dev)
1605 {
1606 return 0;
1607 }
1608
1609 /**
1610 * pcibios_release_device - provide arch specific hooks when releasing device dev
1611 * @dev: the PCI device being released
1612 *
1613 * Permits the platform to provide architecture specific functionality when
1614 * devices are released. This is the default implementation. Architecture
1615 * implementations can override this.
1616 */
1617 void __weak pcibios_release_device(struct pci_dev *dev) {}
1618
1619 /**
1620 * pcibios_disable_device - disable arch specific PCI resources for device dev
1621 * @dev: the PCI device to disable
1622 *
1623 * Disables architecture specific PCI resources for the device. This
1624 * is the default implementation. Architecture implementations can
1625 * override this.
1626 */
1627 void __weak pcibios_disable_device(struct pci_dev *dev) {}
1628
1629 /**
1630 * pcibios_penalize_isa_irq - penalize an ISA IRQ
1631 * @irq: ISA IRQ to penalize
1632 * @active: IRQ active or not
1633 *
1634 * Permits the platform to provide architecture-specific functionality when
1635 * penalizing ISA IRQs. This is the default implementation. Architecture
1636 * implementations can override this.
1637 */
1638 void __weak pcibios_penalize_isa_irq(int irq, int active) {}
1639
1640 static void do_pci_disable_device(struct pci_dev *dev)
1641 {
1642 u16 pci_command;
1643
1644 pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1645 if (pci_command & PCI_COMMAND_MASTER) {
1646 pci_command &= ~PCI_COMMAND_MASTER;
1647 pci_write_config_word(dev, PCI_COMMAND, pci_command);
1648 }
1649
1650 pcibios_disable_device(dev);
1651 }
1652
1653 /**
1654 * pci_disable_enabled_device - Disable device without updating enable_cnt
1655 * @dev: PCI device to disable
1656 *
1657 * NOTE: This function is a backend of PCI power management routines and is
1658 * not supposed to be called drivers.
1659 */
1660 void pci_disable_enabled_device(struct pci_dev *dev)
1661 {
1662 if (pci_is_enabled(dev))
1663 do_pci_disable_device(dev);
1664 }
1665
1666 /**
1667 * pci_disable_device - Disable PCI device after use
1668 * @dev: PCI device to be disabled
1669 *
1670 * Signal to the system that the PCI device is not in use by the system
1671 * anymore. This only involves disabling PCI bus-mastering, if active.
1672 *
1673 * Note we don't actually disable the device until all callers of
1674 * pci_enable_device() have called pci_disable_device().
1675 */
1676 void pci_disable_device(struct pci_dev *dev)
1677 {
1678 struct pci_devres *dr;
1679
1680 dr = find_pci_dr(dev);
1681 if (dr)
1682 dr->enabled = 0;
1683
1684 dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
1685 "disabling already-disabled device");
1686
1687 if (atomic_dec_return(&dev->enable_cnt) != 0)
1688 return;
1689
1690 do_pci_disable_device(dev);
1691
1692 dev->is_busmaster = 0;
1693 }
1694 EXPORT_SYMBOL(pci_disable_device);
1695
1696 /**
1697 * pcibios_set_pcie_reset_state - set reset state for device dev
1698 * @dev: the PCIe device reset
1699 * @state: Reset state to enter into
1700 *
1701 *
1702 * Sets the PCIe reset state for the device. This is the default
1703 * implementation. Architecture implementations can override this.
1704 */
1705 int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1706 enum pcie_reset_state state)
1707 {
1708 return -EINVAL;
1709 }
1710
1711 /**
1712 * pci_set_pcie_reset_state - set reset state for device dev
1713 * @dev: the PCIe device reset
1714 * @state: Reset state to enter into
1715 *
1716 *
1717 * Sets the PCI reset state for the device.
1718 */
1719 int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1720 {
1721 return pcibios_set_pcie_reset_state(dev, state);
1722 }
1723 EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
1724
1725 /**
1726 * pci_check_pme_status - Check if given device has generated PME.
1727 * @dev: Device to check.
1728 *
1729 * Check the PME status of the device and if set, clear it and clear PME enable
1730 * (if set). Return 'true' if PME status and PME enable were both set or
1731 * 'false' otherwise.
1732 */
1733 bool pci_check_pme_status(struct pci_dev *dev)
1734 {
1735 int pmcsr_pos;
1736 u16 pmcsr;
1737 bool ret = false;
1738
1739 if (!dev->pm_cap)
1740 return false;
1741
1742 pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
1743 pci_read_config_word(dev, pmcsr_pos, &pmcsr);
1744 if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
1745 return false;
1746
1747 /* Clear PME status. */
1748 pmcsr |= PCI_PM_CTRL_PME_STATUS;
1749 if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
1750 /* Disable PME to avoid interrupt flood. */
1751 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1752 ret = true;
1753 }
1754
1755 pci_write_config_word(dev, pmcsr_pos, pmcsr);
1756
1757 return ret;
1758 }
1759
1760 /**
1761 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
1762 * @dev: Device to handle.
1763 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
1764 *
1765 * Check if @dev has generated PME and queue a resume request for it in that
1766 * case.
1767 */
1768 static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
1769 {
1770 if (pme_poll_reset && dev->pme_poll)
1771 dev->pme_poll = false;
1772
1773 if (pci_check_pme_status(dev)) {
1774 pci_wakeup_event(dev);
1775 pm_request_resume(&dev->dev);
1776 }
1777 return 0;
1778 }
1779
1780 /**
1781 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
1782 * @bus: Top bus of the subtree to walk.
1783 */
1784 void pci_pme_wakeup_bus(struct pci_bus *bus)
1785 {
1786 if (bus)
1787 pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
1788 }
1789
1790
1791 /**
1792 * pci_pme_capable - check the capability of PCI device to generate PME#
1793 * @dev: PCI device to handle.
1794 * @state: PCI state from which device will issue PME#.
1795 */
1796 bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
1797 {
1798 if (!dev->pm_cap)
1799 return false;
1800
1801 return !!(dev->pme_support & (1 << state));
1802 }
1803 EXPORT_SYMBOL(pci_pme_capable);
1804
1805 static void pci_pme_list_scan(struct work_struct *work)
1806 {
1807 struct pci_pme_device *pme_dev, *n;
1808
1809 mutex_lock(&pci_pme_list_mutex);
1810 list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
1811 if (pme_dev->dev->pme_poll) {
1812 struct pci_dev *bridge;
1813
1814 bridge = pme_dev->dev->bus->self;
1815 /*
1816 * If bridge is in low power state, the
1817 * configuration space of subordinate devices
1818 * may be not accessible
1819 */
1820 if (bridge && bridge->current_state != PCI_D0)
1821 continue;
1822 pci_pme_wakeup(pme_dev->dev, NULL);
1823 } else {
1824 list_del(&pme_dev->list);
1825 kfree(pme_dev);
1826 }
1827 }
1828 if (!list_empty(&pci_pme_list))
1829 queue_delayed_work(system_freezable_wq, &pci_pme_work,
1830 msecs_to_jiffies(PME_TIMEOUT));
1831 mutex_unlock(&pci_pme_list_mutex);
1832 }
1833
1834 static void __pci_pme_active(struct pci_dev *dev, bool enable)
1835 {
1836 u16 pmcsr;
1837
1838 if (!dev->pme_support)
1839 return;
1840
1841 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1842 /* Clear PME_Status by writing 1 to it and enable PME# */
1843 pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
1844 if (!enable)
1845 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1846
1847 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1848 }
1849
1850 /**
1851 * pci_pme_restore - Restore PME configuration after config space restore.
1852 * @dev: PCI device to update.
1853 */
1854 void pci_pme_restore(struct pci_dev *dev)
1855 {
1856 u16 pmcsr;
1857
1858 if (!dev->pme_support)
1859 return;
1860
1861 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1862 if (dev->wakeup_prepared) {
1863 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
1864 pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
1865 } else {
1866 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1867 pmcsr |= PCI_PM_CTRL_PME_STATUS;
1868 }
1869 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1870 }
1871
1872 /**
1873 * pci_pme_active - enable or disable PCI device's PME# function
1874 * @dev: PCI device to handle.
1875 * @enable: 'true' to enable PME# generation; 'false' to disable it.
1876 *
1877 * The caller must verify that the device is capable of generating PME# before
1878 * calling this function with @enable equal to 'true'.
1879 */
1880 void pci_pme_active(struct pci_dev *dev, bool enable)
1881 {
1882 __pci_pme_active(dev, enable);
1883
1884 /*
1885 * PCI (as opposed to PCIe) PME requires that the device have
1886 * its PME# line hooked up correctly. Not all hardware vendors
1887 * do this, so the PME never gets delivered and the device
1888 * remains asleep. The easiest way around this is to
1889 * periodically walk the list of suspended devices and check
1890 * whether any have their PME flag set. The assumption is that
1891 * we'll wake up often enough anyway that this won't be a huge
1892 * hit, and the power savings from the devices will still be a
1893 * win.
1894 *
1895 * Although PCIe uses in-band PME message instead of PME# line
1896 * to report PME, PME does not work for some PCIe devices in
1897 * reality. For example, there are devices that set their PME
1898 * status bits, but don't really bother to send a PME message;
1899 * there are PCI Express Root Ports that don't bother to
1900 * trigger interrupts when they receive PME messages from the
1901 * devices below. So PME poll is used for PCIe devices too.
1902 */
1903
1904 if (dev->pme_poll) {
1905 struct pci_pme_device *pme_dev;
1906 if (enable) {
1907 pme_dev = kmalloc(sizeof(struct pci_pme_device),
1908 GFP_KERNEL);
1909 if (!pme_dev) {
1910 dev_warn(&dev->dev, "can't enable PME#\n");
1911 return;
1912 }
1913 pme_dev->dev = dev;
1914 mutex_lock(&pci_pme_list_mutex);
1915 list_add(&pme_dev->list, &pci_pme_list);
1916 if (list_is_singular(&pci_pme_list))
1917 queue_delayed_work(system_freezable_wq,
1918 &pci_pme_work,
1919 msecs_to_jiffies(PME_TIMEOUT));
1920 mutex_unlock(&pci_pme_list_mutex);
1921 } else {
1922 mutex_lock(&pci_pme_list_mutex);
1923 list_for_each_entry(pme_dev, &pci_pme_list, list) {
1924 if (pme_dev->dev == dev) {
1925 list_del(&pme_dev->list);
1926 kfree(pme_dev);
1927 break;
1928 }
1929 }
1930 mutex_unlock(&pci_pme_list_mutex);
1931 }
1932 }
1933
1934 dev_dbg(&dev->dev, "PME# %s\n", enable ? "enabled" : "disabled");
1935 }
1936 EXPORT_SYMBOL(pci_pme_active);
1937
1938 /**
1939 * __pci_enable_wake - enable PCI device as wakeup event source
1940 * @dev: PCI device affected
1941 * @state: PCI state from which device will issue wakeup events
1942 * @enable: True to enable event generation; false to disable
1943 *
1944 * This enables the device as a wakeup event source, or disables it.
1945 * When such events involves platform-specific hooks, those hooks are
1946 * called automatically by this routine.
1947 *
1948 * Devices with legacy power management (no standard PCI PM capabilities)
1949 * always require such platform hooks.
1950 *
1951 * RETURN VALUE:
1952 * 0 is returned on success
1953 * -EINVAL is returned if device is not supposed to wake up the system
1954 * Error code depending on the platform is returned if both the platform and
1955 * the native mechanism fail to enable the generation of wake-up events
1956 */
1957 static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
1958 {
1959 int ret = 0;
1960
1961 /*
1962 * Bridges can only signal wakeup on behalf of subordinate devices,
1963 * but that is set up elsewhere, so skip them.
1964 */
1965 if (pci_has_subordinate(dev))
1966 return 0;
1967
1968 /* Don't do the same thing twice in a row for one device. */
1969 if (!!enable == !!dev->wakeup_prepared)
1970 return 0;
1971
1972 /*
1973 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
1974 * Anderson we should be doing PME# wake enable followed by ACPI wake
1975 * enable. To disable wake-up we call the platform first, for symmetry.
1976 */
1977
1978 if (enable) {
1979 int error;
1980
1981 if (pci_pme_capable(dev, state))
1982 pci_pme_active(dev, true);
1983 else
1984 ret = 1;
1985 error = platform_pci_set_wakeup(dev, true);
1986 if (ret)
1987 ret = error;
1988 if (!ret)
1989 dev->wakeup_prepared = true;
1990 } else {
1991 platform_pci_set_wakeup(dev, false);
1992 pci_pme_active(dev, false);
1993 dev->wakeup_prepared = false;
1994 }
1995
1996 return ret;
1997 }
1998
1999 /**
2000 * pci_enable_wake - change wakeup settings for a PCI device
2001 * @pci_dev: Target device
2002 * @state: PCI state from which device will issue wakeup events
2003 * @enable: Whether or not to enable event generation
2004 *
2005 * If @enable is set, check device_may_wakeup() for the device before calling
2006 * __pci_enable_wake() for it.
2007 */
2008 int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
2009 {
2010 if (enable && !device_may_wakeup(&pci_dev->dev))
2011 return -EINVAL;
2012
2013 return __pci_enable_wake(pci_dev, state, enable);
2014 }
2015 EXPORT_SYMBOL(pci_enable_wake);
2016
2017 /**
2018 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
2019 * @dev: PCI device to prepare
2020 * @enable: True to enable wake-up event generation; false to disable
2021 *
2022 * Many drivers want the device to wake up the system from D3_hot or D3_cold
2023 * and this function allows them to set that up cleanly - pci_enable_wake()
2024 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
2025 * ordering constraints.
2026 *
2027 * This function only returns error code if the device is not allowed to wake
2028 * up the system from sleep or it is not capable of generating PME# from both
2029 * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
2030 */
2031 int pci_wake_from_d3(struct pci_dev *dev, bool enable)
2032 {
2033 return pci_pme_capable(dev, PCI_D3cold) ?
2034 pci_enable_wake(dev, PCI_D3cold, enable) :
2035 pci_enable_wake(dev, PCI_D3hot, enable);
2036 }
2037 EXPORT_SYMBOL(pci_wake_from_d3);
2038
2039 /**
2040 * pci_target_state - find an appropriate low power state for a given PCI dev
2041 * @dev: PCI device
2042 * @wakeup: Whether or not wakeup functionality will be enabled for the device.
2043 *
2044 * Use underlying platform code to find a supported low power state for @dev.
2045 * If the platform can't manage @dev, return the deepest state from which it
2046 * can generate wake events, based on any available PME info.
2047 */
2048 static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
2049 {
2050 pci_power_t target_state = PCI_D3hot;
2051
2052 if (platform_pci_power_manageable(dev)) {
2053 /*
2054 * Call the platform to choose the target state of the device
2055 * and enable wake-up from this state if supported.
2056 */
2057 pci_power_t state = platform_pci_choose_state(dev);
2058
2059 switch (state) {
2060 case PCI_POWER_ERROR:
2061 case PCI_UNKNOWN:
2062 break;
2063 case PCI_D1:
2064 case PCI_D2:
2065 if (pci_no_d1d2(dev))
2066 break;
2067 default:
2068 target_state = state;
2069 }
2070
2071 return target_state;
2072 }
2073
2074 if (!dev->pm_cap)
2075 target_state = PCI_D0;
2076
2077 /*
2078 * If the device is in D3cold even though it's not power-manageable by
2079 * the platform, it may have been powered down by non-standard means.
2080 * Best to let it slumber.
2081 */
2082 if (dev->current_state == PCI_D3cold)
2083 target_state = PCI_D3cold;
2084
2085 if (wakeup) {
2086 /*
2087 * Find the deepest state from which the device can generate
2088 * wake-up events, make it the target state and enable device
2089 * to generate PME#.
2090 */
2091 if (dev->pme_support) {
2092 while (target_state
2093 && !(dev->pme_support & (1 << target_state)))
2094 target_state--;
2095 }
2096 }
2097
2098 return target_state;
2099 }
2100
2101 /**
2102 * pci_prepare_to_sleep - prepare PCI device for system-wide transition into a sleep state
2103 * @dev: Device to handle.
2104 *
2105 * Choose the power state appropriate for the device depending on whether
2106 * it can wake up the system and/or is power manageable by the platform
2107 * (PCI_D3hot is the default) and put the device into that state.
2108 */
2109 int pci_prepare_to_sleep(struct pci_dev *dev)
2110 {
2111 bool wakeup = device_may_wakeup(&dev->dev);
2112 pci_power_t target_state = pci_target_state(dev, wakeup);
2113 int error;
2114
2115 if (target_state == PCI_POWER_ERROR)
2116 return -EIO;
2117
2118 pci_enable_wake(dev, target_state, wakeup);
2119
2120 error = pci_set_power_state(dev, target_state);
2121
2122 if (error)
2123 pci_enable_wake(dev, target_state, false);
2124
2125 return error;
2126 }
2127 EXPORT_SYMBOL(pci_prepare_to_sleep);
2128
2129 /**
2130 * pci_back_from_sleep - turn PCI device on during system-wide transition into working state
2131 * @dev: Device to handle.
2132 *
2133 * Disable device's system wake-up capability and put it into D0.
2134 */
2135 int pci_back_from_sleep(struct pci_dev *dev)
2136 {
2137 pci_enable_wake(dev, PCI_D0, false);
2138 return pci_set_power_state(dev, PCI_D0);
2139 }
2140 EXPORT_SYMBOL(pci_back_from_sleep);
2141
2142 /**
2143 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2144 * @dev: PCI device being suspended.
2145 *
2146 * Prepare @dev to generate wake-up events at run time and put it into a low
2147 * power state.
2148 */
2149 int pci_finish_runtime_suspend(struct pci_dev *dev)
2150 {
2151 pci_power_t target_state;
2152 int error;
2153
2154 target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2155 if (target_state == PCI_POWER_ERROR)
2156 return -EIO;
2157
2158 dev->runtime_d3cold = target_state == PCI_D3cold;
2159
2160 __pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2161
2162 error = pci_set_power_state(dev, target_state);
2163
2164 if (error) {
2165 pci_enable_wake(dev, target_state, false);
2166 dev->runtime_d3cold = false;
2167 }
2168
2169 return error;
2170 }
2171
2172 /**
2173 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2174 * @dev: Device to check.
2175 *
2176 * Return true if the device itself is capable of generating wake-up events
2177 * (through the platform or using the native PCIe PME) or if the device supports
2178 * PME and one of its upstream bridges can generate wake-up events.
2179 */
2180 bool pci_dev_run_wake(struct pci_dev *dev)
2181 {
2182 struct pci_bus *bus = dev->bus;
2183
2184 if (!dev->pme_support)
2185 return false;
2186
2187 /* PME-capable in principle, but not from the target power state */
2188 if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2189 return false;
2190
2191 if (device_can_wakeup(&dev->dev))
2192 return true;
2193
2194 while (bus->parent) {
2195 struct pci_dev *bridge = bus->self;
2196
2197 if (device_can_wakeup(&bridge->dev))
2198 return true;
2199
2200 bus = bus->parent;
2201 }
2202
2203 /* We have reached the root bus. */
2204 if (bus->bridge)
2205 return device_can_wakeup(bus->bridge);
2206
2207 return false;
2208 }
2209 EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2210
2211 /**
2212 * pci_dev_keep_suspended - Check if the device can stay in the suspended state.
2213 * @pci_dev: Device to check.
2214 *
2215 * Return 'true' if the device is runtime-suspended, it doesn't have to be
2216 * reconfigured due to wakeup settings difference between system and runtime
2217 * suspend and the current power state of it is suitable for the upcoming
2218 * (system) transition.
2219 *
2220 * If the device is not configured for system wakeup, disable PME for it before
2221 * returning 'true' to prevent it from waking up the system unnecessarily.
2222 */
2223 bool pci_dev_keep_suspended(struct pci_dev *pci_dev)
2224 {
2225 struct device *dev = &pci_dev->dev;
2226 bool wakeup = device_may_wakeup(dev);
2227
2228 if (!pm_runtime_suspended(dev)
2229 || pci_target_state(pci_dev, wakeup) != pci_dev->current_state
2230 || platform_pci_need_resume(pci_dev))
2231 return false;
2232
2233 /*
2234 * At this point the device is good to go unless it's been configured
2235 * to generate PME at the runtime suspend time, but it is not supposed
2236 * to wake up the system. In that case, simply disable PME for it
2237 * (it will have to be re-enabled on exit from system resume).
2238 *
2239 * If the device's power state is D3cold and the platform check above
2240 * hasn't triggered, the device's configuration is suitable and we don't
2241 * need to manipulate it at all.
2242 */
2243 spin_lock_irq(&dev->power.lock);
2244
2245 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold &&
2246 !wakeup)
2247 __pci_pme_active(pci_dev, false);
2248
2249 spin_unlock_irq(&dev->power.lock);
2250 return true;
2251 }
2252
2253 /**
2254 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2255 * @pci_dev: Device to handle.
2256 *
2257 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2258 * it might have been disabled during the prepare phase of system suspend if
2259 * the device was not configured for system wakeup.
2260 */
2261 void pci_dev_complete_resume(struct pci_dev *pci_dev)
2262 {
2263 struct device *dev = &pci_dev->dev;
2264
2265 if (!pci_dev_run_wake(pci_dev))
2266 return;
2267
2268 spin_lock_irq(&dev->power.lock);
2269
2270 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2271 __pci_pme_active(pci_dev, true);
2272
2273 spin_unlock_irq(&dev->power.lock);
2274 }
2275
2276 void pci_config_pm_runtime_get(struct pci_dev *pdev)
2277 {
2278 struct device *dev = &pdev->dev;
2279 struct device *parent = dev->parent;
2280
2281 if (parent)
2282 pm_runtime_get_sync(parent);
2283 pm_runtime_get_noresume(dev);
2284 /*
2285 * pdev->current_state is set to PCI_D3cold during suspending,
2286 * so wait until suspending completes
2287 */
2288 pm_runtime_barrier(dev);
2289 /*
2290 * Only need to resume devices in D3cold, because config
2291 * registers are still accessible for devices suspended but
2292 * not in D3cold.
2293 */
2294 if (pdev->current_state == PCI_D3cold)
2295 pm_runtime_resume(dev);
2296 }
2297
2298 void pci_config_pm_runtime_put(struct pci_dev *pdev)
2299 {
2300 struct device *dev = &pdev->dev;
2301 struct device *parent = dev->parent;
2302
2303 pm_runtime_put(dev);
2304 if (parent)
2305 pm_runtime_put_sync(parent);
2306 }
2307
2308 /**
2309 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2310 * @bridge: Bridge to check
2311 *
2312 * This function checks if it is possible to move the bridge to D3.
2313 * Currently we only allow D3 for recent enough PCIe ports.
2314 */
2315 bool pci_bridge_d3_possible(struct pci_dev *bridge)
2316 {
2317 unsigned int year;
2318
2319 if (!pci_is_pcie(bridge))
2320 return false;
2321
2322 switch (pci_pcie_type(bridge)) {
2323 case PCI_EXP_TYPE_ROOT_PORT:
2324 case PCI_EXP_TYPE_UPSTREAM:
2325 case PCI_EXP_TYPE_DOWNSTREAM:
2326 if (pci_bridge_d3_disable)
2327 return false;
2328
2329 /*
2330 * Hotplug interrupts cannot be delivered if the link is down,
2331 * so parents of a hotplug port must stay awake. In addition,
2332 * hotplug ports handled by firmware in System Management Mode
2333 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2334 * For simplicity, disallow in general for now.
2335 */
2336 if (bridge->is_hotplug_bridge)
2337 return false;
2338
2339 if (pci_bridge_d3_force)
2340 return true;
2341
2342 /*
2343 * It should be safe to put PCIe ports from 2015 or newer
2344 * to D3.
2345 */
2346 if (dmi_get_date(DMI_BIOS_DATE, &year, NULL, NULL) &&
2347 year >= 2015) {
2348 return true;
2349 }
2350 break;
2351 }
2352
2353 return false;
2354 }
2355
2356 static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2357 {
2358 bool *d3cold_ok = data;
2359
2360 if (/* The device needs to be allowed to go D3cold ... */
2361 dev->no_d3cold || !dev->d3cold_allowed ||
2362
2363 /* ... and if it is wakeup capable to do so from D3cold. */
2364 (device_may_wakeup(&dev->dev) &&
2365 !pci_pme_capable(dev, PCI_D3cold)) ||
2366
2367 /* If it is a bridge it must be allowed to go to D3. */
2368 !pci_power_manageable(dev))
2369
2370 *d3cold_ok = false;
2371
2372 return !*d3cold_ok;
2373 }
2374
2375 /*
2376 * pci_bridge_d3_update - Update bridge D3 capabilities
2377 * @dev: PCI device which is changed
2378 *
2379 * Update upstream bridge PM capabilities accordingly depending on if the
2380 * device PM configuration was changed or the device is being removed. The
2381 * change is also propagated upstream.
2382 */
2383 void pci_bridge_d3_update(struct pci_dev *dev)
2384 {
2385 bool remove = !device_is_registered(&dev->dev);
2386 struct pci_dev *bridge;
2387 bool d3cold_ok = true;
2388
2389 bridge = pci_upstream_bridge(dev);
2390 if (!bridge || !pci_bridge_d3_possible(bridge))
2391 return;
2392
2393 /*
2394 * If D3 is currently allowed for the bridge, removing one of its
2395 * children won't change that.
2396 */
2397 if (remove && bridge->bridge_d3)
2398 return;
2399
2400 /*
2401 * If D3 is currently allowed for the bridge and a child is added or
2402 * changed, disallowance of D3 can only be caused by that child, so
2403 * we only need to check that single device, not any of its siblings.
2404 *
2405 * If D3 is currently not allowed for the bridge, checking the device
2406 * first may allow us to skip checking its siblings.
2407 */
2408 if (!remove)
2409 pci_dev_check_d3cold(dev, &d3cold_ok);
2410
2411 /*
2412 * If D3 is currently not allowed for the bridge, this may be caused
2413 * either by the device being changed/removed or any of its siblings,
2414 * so we need to go through all children to find out if one of them
2415 * continues to block D3.
2416 */
2417 if (d3cold_ok && !bridge->bridge_d3)
2418 pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
2419 &d3cold_ok);
2420
2421 if (bridge->bridge_d3 != d3cold_ok) {
2422 bridge->bridge_d3 = d3cold_ok;
2423 /* Propagate change to upstream bridges */
2424 pci_bridge_d3_update(bridge);
2425 }
2426 }
2427
2428 /**
2429 * pci_d3cold_enable - Enable D3cold for device
2430 * @dev: PCI device to handle
2431 *
2432 * This function can be used in drivers to enable D3cold from the device
2433 * they handle. It also updates upstream PCI bridge PM capabilities
2434 * accordingly.
2435 */
2436 void pci_d3cold_enable(struct pci_dev *dev)
2437 {
2438 if (dev->no_d3cold) {
2439 dev->no_d3cold = false;
2440 pci_bridge_d3_update(dev);
2441 }
2442 }
2443 EXPORT_SYMBOL_GPL(pci_d3cold_enable);
2444
2445 /**
2446 * pci_d3cold_disable - Disable D3cold for device
2447 * @dev: PCI device to handle
2448 *
2449 * This function can be used in drivers to disable D3cold from the device
2450 * they handle. It also updates upstream PCI bridge PM capabilities
2451 * accordingly.
2452 */
2453 void pci_d3cold_disable(struct pci_dev *dev)
2454 {
2455 if (!dev->no_d3cold) {
2456 dev->no_d3cold = true;
2457 pci_bridge_d3_update(dev);
2458 }
2459 }
2460 EXPORT_SYMBOL_GPL(pci_d3cold_disable);
2461
2462 /**
2463 * pci_pm_init - Initialize PM functions of given PCI device
2464 * @dev: PCI device to handle.
2465 */
2466 void pci_pm_init(struct pci_dev *dev)
2467 {
2468 int pm;
2469 u16 pmc;
2470
2471 pm_runtime_forbid(&dev->dev);
2472 pm_runtime_set_active(&dev->dev);
2473 pm_runtime_enable(&dev->dev);
2474 device_enable_async_suspend(&dev->dev);
2475 dev->wakeup_prepared = false;
2476
2477 dev->pm_cap = 0;
2478 dev->pme_support = 0;
2479
2480 /* find PCI PM capability in list */
2481 pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2482 if (!pm)
2483 return;
2484 /* Check device's ability to generate PME# */
2485 pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2486
2487 if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2488 dev_err(&dev->dev, "unsupported PM cap regs version (%u)\n",
2489 pmc & PCI_PM_CAP_VER_MASK);
2490 return;
2491 }
2492
2493 dev->pm_cap = pm;
2494 dev->d3_delay = PCI_PM_D3_WAIT;
2495 dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2496 dev->bridge_d3 = pci_bridge_d3_possible(dev);
2497 dev->d3cold_allowed = true;
2498
2499 dev->d1_support = false;
2500 dev->d2_support = false;
2501 if (!pci_no_d1d2(dev)) {
2502 if (pmc & PCI_PM_CAP_D1)
2503 dev->d1_support = true;
2504 if (pmc & PCI_PM_CAP_D2)
2505 dev->d2_support = true;
2506
2507 if (dev->d1_support || dev->d2_support)
2508 dev_printk(KERN_DEBUG, &dev->dev, "supports%s%s\n",
2509 dev->d1_support ? " D1" : "",
2510 dev->d2_support ? " D2" : "");
2511 }
2512
2513 pmc &= PCI_PM_CAP_PME_MASK;
2514 if (pmc) {
2515 dev_printk(KERN_DEBUG, &dev->dev,
2516 "PME# supported from%s%s%s%s%s\n",
2517 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2518 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2519 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2520 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2521 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2522 dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2523 dev->pme_poll = true;
2524 /*
2525 * Make device's PM flags reflect the wake-up capability, but
2526 * let the user space enable it to wake up the system as needed.
2527 */
2528 device_set_wakeup_capable(&dev->dev, true);
2529 /* Disable the PME# generation functionality */
2530 pci_pme_active(dev, false);
2531 }
2532 }
2533
2534 static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2535 {
2536 unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
2537
2538 switch (prop) {
2539 case PCI_EA_P_MEM:
2540 case PCI_EA_P_VF_MEM:
2541 flags |= IORESOURCE_MEM;
2542 break;
2543 case PCI_EA_P_MEM_PREFETCH:
2544 case PCI_EA_P_VF_MEM_PREFETCH:
2545 flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2546 break;
2547 case PCI_EA_P_IO:
2548 flags |= IORESOURCE_IO;
2549 break;
2550 default:
2551 return 0;
2552 }
2553
2554 return flags;
2555 }
2556
2557 static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2558 u8 prop)
2559 {
2560 if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2561 return &dev->resource[bei];
2562 #ifdef CONFIG_PCI_IOV
2563 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2564 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2565 return &dev->resource[PCI_IOV_RESOURCES +
2566 bei - PCI_EA_BEI_VF_BAR0];
2567 #endif
2568 else if (bei == PCI_EA_BEI_ROM)
2569 return &dev->resource[PCI_ROM_RESOURCE];
2570 else
2571 return NULL;
2572 }
2573
2574 /* Read an Enhanced Allocation (EA) entry */
2575 static int pci_ea_read(struct pci_dev *dev, int offset)
2576 {
2577 struct resource *res;
2578 int ent_size, ent_offset = offset;
2579 resource_size_t start, end;
2580 unsigned long flags;
2581 u32 dw0, bei, base, max_offset;
2582 u8 prop;
2583 bool support_64 = (sizeof(resource_size_t) >= 8);
2584
2585 pci_read_config_dword(dev, ent_offset, &dw0);
2586 ent_offset += 4;
2587
2588 /* Entry size field indicates DWORDs after 1st */
2589 ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2590
2591 if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2592 goto out;
2593
2594 bei = (dw0 & PCI_EA_BEI) >> 4;
2595 prop = (dw0 & PCI_EA_PP) >> 8;
2596
2597 /*
2598 * If the Property is in the reserved range, try the Secondary
2599 * Property instead.
2600 */
2601 if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2602 prop = (dw0 & PCI_EA_SP) >> 16;
2603 if (prop > PCI_EA_P_BRIDGE_IO)
2604 goto out;
2605
2606 res = pci_ea_get_resource(dev, bei, prop);
2607 if (!res) {
2608 dev_err(&dev->dev, "Unsupported EA entry BEI: %u\n", bei);
2609 goto out;
2610 }
2611
2612 flags = pci_ea_flags(dev, prop);
2613 if (!flags) {
2614 dev_err(&dev->dev, "Unsupported EA properties: %#x\n", prop);
2615 goto out;
2616 }
2617
2618 /* Read Base */
2619 pci_read_config_dword(dev, ent_offset, &base);
2620 start = (base & PCI_EA_FIELD_MASK);
2621 ent_offset += 4;
2622
2623 /* Read MaxOffset */
2624 pci_read_config_dword(dev, ent_offset, &max_offset);
2625 ent_offset += 4;
2626
2627 /* Read Base MSBs (if 64-bit entry) */
2628 if (base & PCI_EA_IS_64) {
2629 u32 base_upper;
2630
2631 pci_read_config_dword(dev, ent_offset, &base_upper);
2632 ent_offset += 4;
2633
2634 flags |= IORESOURCE_MEM_64;
2635
2636 /* entry starts above 32-bit boundary, can't use */
2637 if (!support_64 && base_upper)
2638 goto out;
2639
2640 if (support_64)
2641 start |= ((u64)base_upper << 32);
2642 }
2643
2644 end = start + (max_offset | 0x03);
2645
2646 /* Read MaxOffset MSBs (if 64-bit entry) */
2647 if (max_offset & PCI_EA_IS_64) {
2648 u32 max_offset_upper;
2649
2650 pci_read_config_dword(dev, ent_offset, &max_offset_upper);
2651 ent_offset += 4;
2652
2653 flags |= IORESOURCE_MEM_64;
2654
2655 /* entry too big, can't use */
2656 if (!support_64 && max_offset_upper)
2657 goto out;
2658
2659 if (support_64)
2660 end += ((u64)max_offset_upper << 32);
2661 }
2662
2663 if (end < start) {
2664 dev_err(&dev->dev, "EA Entry crosses address boundary\n");
2665 goto out;
2666 }
2667
2668 if (ent_size != ent_offset - offset) {
2669 dev_err(&dev->dev,
2670 "EA Entry Size (%d) does not match length read (%d)\n",
2671 ent_size, ent_offset - offset);
2672 goto out;
2673 }
2674
2675 res->name = pci_name(dev);
2676 res->start = start;
2677 res->end = end;
2678 res->flags = flags;
2679
2680 if (bei <= PCI_EA_BEI_BAR5)
2681 dev_printk(KERN_DEBUG, &dev->dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2682 bei, res, prop);
2683 else if (bei == PCI_EA_BEI_ROM)
2684 dev_printk(KERN_DEBUG, &dev->dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
2685 res, prop);
2686 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
2687 dev_printk(KERN_DEBUG, &dev->dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2688 bei - PCI_EA_BEI_VF_BAR0, res, prop);
2689 else
2690 dev_printk(KERN_DEBUG, &dev->dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
2691 bei, res, prop);
2692
2693 out:
2694 return offset + ent_size;
2695 }
2696
2697 /* Enhanced Allocation Initialization */
2698 void pci_ea_init(struct pci_dev *dev)
2699 {
2700 int ea;
2701 u8 num_ent;
2702 int offset;
2703 int i;
2704
2705 /* find PCI EA capability in list */
2706 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
2707 if (!ea)
2708 return;
2709
2710 /* determine the number of entries */
2711 pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
2712 &num_ent);
2713 num_ent &= PCI_EA_NUM_ENT_MASK;
2714
2715 offset = ea + PCI_EA_FIRST_ENT;
2716
2717 /* Skip DWORD 2 for type 1 functions */
2718 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
2719 offset += 4;
2720
2721 /* parse each EA entry */
2722 for (i = 0; i < num_ent; ++i)
2723 offset = pci_ea_read(dev, offset);
2724 }
2725
2726 static void pci_add_saved_cap(struct pci_dev *pci_dev,
2727 struct pci_cap_saved_state *new_cap)
2728 {
2729 hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
2730 }
2731
2732 /**
2733 * _pci_add_cap_save_buffer - allocate buffer for saving given
2734 * capability registers
2735 * @dev: the PCI device
2736 * @cap: the capability to allocate the buffer for
2737 * @extended: Standard or Extended capability ID
2738 * @size: requested size of the buffer
2739 */
2740 static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
2741 bool extended, unsigned int size)
2742 {
2743 int pos;
2744 struct pci_cap_saved_state *save_state;
2745
2746 if (extended)
2747 pos = pci_find_ext_capability(dev, cap);
2748 else
2749 pos = pci_find_capability(dev, cap);
2750
2751 if (!pos)
2752 return 0;
2753
2754 save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
2755 if (!save_state)
2756 return -ENOMEM;
2757
2758 save_state->cap.cap_nr = cap;
2759 save_state->cap.cap_extended = extended;
2760 save_state->cap.size = size;
2761 pci_add_saved_cap(dev, save_state);
2762
2763 return 0;
2764 }
2765
2766 int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
2767 {
2768 return _pci_add_cap_save_buffer(dev, cap, false, size);
2769 }
2770
2771 int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
2772 {
2773 return _pci_add_cap_save_buffer(dev, cap, true, size);
2774 }
2775
2776 /**
2777 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
2778 * @dev: the PCI device
2779 */
2780 void pci_allocate_cap_save_buffers(struct pci_dev *dev)
2781 {
2782 int error;
2783
2784 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
2785 PCI_EXP_SAVE_REGS * sizeof(u16));
2786 if (error)
2787 dev_err(&dev->dev,
2788 "unable to preallocate PCI Express save buffer\n");
2789
2790 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
2791 if (error)
2792 dev_err(&dev->dev,
2793 "unable to preallocate PCI-X save buffer\n");
2794
2795 pci_allocate_vc_save_buffers(dev);
2796 }
2797
2798 void pci_free_cap_save_buffers(struct pci_dev *dev)
2799 {
2800 struct pci_cap_saved_state *tmp;
2801 struct hlist_node *n;
2802
2803 hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
2804 kfree(tmp);
2805 }
2806
2807 /**
2808 * pci_configure_ari - enable or disable ARI forwarding
2809 * @dev: the PCI device
2810 *
2811 * If @dev and its upstream bridge both support ARI, enable ARI in the
2812 * bridge. Otherwise, disable ARI in the bridge.
2813 */
2814 void pci_configure_ari(struct pci_dev *dev)
2815 {
2816 u32 cap;
2817 struct pci_dev *bridge;
2818
2819 if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
2820 return;
2821
2822 bridge = dev->bus->self;
2823 if (!bridge)
2824 return;
2825
2826 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
2827 if (!(cap & PCI_EXP_DEVCAP2_ARI))
2828 return;
2829
2830 if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
2831 pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
2832 PCI_EXP_DEVCTL2_ARI);
2833 bridge->ari_enabled = 1;
2834 } else {
2835 pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
2836 PCI_EXP_DEVCTL2_ARI);
2837 bridge->ari_enabled = 0;
2838 }
2839 }
2840
2841 static int pci_acs_enable;
2842
2843 /**
2844 * pci_request_acs - ask for ACS to be enabled if supported
2845 */
2846 void pci_request_acs(void)
2847 {
2848 pci_acs_enable = 1;
2849 }
2850
2851 /**
2852 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilites
2853 * @dev: the PCI device
2854 */
2855 static void pci_std_enable_acs(struct pci_dev *dev)
2856 {
2857 int pos;
2858 u16 cap;
2859 u16 ctrl;
2860
2861 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
2862 if (!pos)
2863 return;
2864
2865 pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
2866 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
2867
2868 /* Source Validation */
2869 ctrl |= (cap & PCI_ACS_SV);
2870
2871 /* P2P Request Redirect */
2872 ctrl |= (cap & PCI_ACS_RR);
2873
2874 /* P2P Completion Redirect */
2875 ctrl |= (cap & PCI_ACS_CR);
2876
2877 /* Upstream Forwarding */
2878 ctrl |= (cap & PCI_ACS_UF);
2879
2880 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
2881 }
2882
2883 /**
2884 * pci_enable_acs - enable ACS if hardware support it
2885 * @dev: the PCI device
2886 */
2887 void pci_enable_acs(struct pci_dev *dev)
2888 {
2889 if (!pci_acs_enable)
2890 return;
2891
2892 if (!pci_dev_specific_enable_acs(dev))
2893 return;
2894
2895 pci_std_enable_acs(dev);
2896 }
2897
2898 static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
2899 {
2900 int pos;
2901 u16 cap, ctrl;
2902
2903 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
2904 if (!pos)
2905 return false;
2906
2907 /*
2908 * Except for egress control, capabilities are either required
2909 * or only required if controllable. Features missing from the
2910 * capability field can therefore be assumed as hard-wired enabled.
2911 */
2912 pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
2913 acs_flags &= (cap | PCI_ACS_EC);
2914
2915 pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
2916 return (ctrl & acs_flags) == acs_flags;
2917 }
2918
2919 /**
2920 * pci_acs_enabled - test ACS against required flags for a given device
2921 * @pdev: device to test
2922 * @acs_flags: required PCI ACS flags
2923 *
2924 * Return true if the device supports the provided flags. Automatically
2925 * filters out flags that are not implemented on multifunction devices.
2926 *
2927 * Note that this interface checks the effective ACS capabilities of the
2928 * device rather than the actual capabilities. For instance, most single
2929 * function endpoints are not required to support ACS because they have no
2930 * opportunity for peer-to-peer access. We therefore return 'true'
2931 * regardless of whether the device exposes an ACS capability. This makes
2932 * it much easier for callers of this function to ignore the actual type
2933 * or topology of the device when testing ACS support.
2934 */
2935 bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
2936 {
2937 int ret;
2938
2939 ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
2940 if (ret >= 0)
2941 return ret > 0;
2942
2943 /*
2944 * Conventional PCI and PCI-X devices never support ACS, either
2945 * effectively or actually. The shared bus topology implies that
2946 * any device on the bus can receive or snoop DMA.
2947 */
2948 if (!pci_is_pcie(pdev))
2949 return false;
2950
2951 switch (pci_pcie_type(pdev)) {
2952 /*
2953 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
2954 * but since their primary interface is PCI/X, we conservatively
2955 * handle them as we would a non-PCIe device.
2956 */
2957 case PCI_EXP_TYPE_PCIE_BRIDGE:
2958 /*
2959 * PCIe 3.0, 6.12.1 excludes ACS on these devices. "ACS is never
2960 * applicable... must never implement an ACS Extended Capability...".
2961 * This seems arbitrary, but we take a conservative interpretation
2962 * of this statement.
2963 */
2964 case PCI_EXP_TYPE_PCI_BRIDGE:
2965 case PCI_EXP_TYPE_RC_EC:
2966 return false;
2967 /*
2968 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
2969 * implement ACS in order to indicate their peer-to-peer capabilities,
2970 * regardless of whether they are single- or multi-function devices.
2971 */
2972 case PCI_EXP_TYPE_DOWNSTREAM:
2973 case PCI_EXP_TYPE_ROOT_PORT:
2974 return pci_acs_flags_enabled(pdev, acs_flags);
2975 /*
2976 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
2977 * implemented by the remaining PCIe types to indicate peer-to-peer
2978 * capabilities, but only when they are part of a multifunction
2979 * device. The footnote for section 6.12 indicates the specific
2980 * PCIe types included here.
2981 */
2982 case PCI_EXP_TYPE_ENDPOINT:
2983 case PCI_EXP_TYPE_UPSTREAM:
2984 case PCI_EXP_TYPE_LEG_END:
2985 case PCI_EXP_TYPE_RC_END:
2986 if (!pdev->multifunction)
2987 break;
2988
2989 return pci_acs_flags_enabled(pdev, acs_flags);
2990 }
2991
2992 /*
2993 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
2994 * to single function devices with the exception of downstream ports.
2995 */
2996 return true;
2997 }
2998
2999 /**
3000 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
3001 * @start: starting downstream device
3002 * @end: ending upstream device or NULL to search to the root bus
3003 * @acs_flags: required flags
3004 *
3005 * Walk up a device tree from start to end testing PCI ACS support. If
3006 * any step along the way does not support the required flags, return false.
3007 */
3008 bool pci_acs_path_enabled(struct pci_dev *start,
3009 struct pci_dev *end, u16 acs_flags)
3010 {
3011 struct pci_dev *pdev, *parent = start;
3012
3013 do {
3014 pdev = parent;
3015
3016 if (!pci_acs_enabled(pdev, acs_flags))
3017 return false;
3018
3019 if (pci_is_root_bus(pdev->bus))
3020 return (end == NULL);
3021
3022 parent = pdev->bus->self;
3023 } while (pdev != end);
3024
3025 return true;
3026 }
3027
3028 /**
3029 * pci_rebar_find_pos - find position of resize ctrl reg for BAR
3030 * @pdev: PCI device
3031 * @bar: BAR to find
3032 *
3033 * Helper to find the position of the ctrl register for a BAR.
3034 * Returns -ENOTSUPP if resizable BARs are not supported at all.
3035 * Returns -ENOENT if no ctrl register for the BAR could be found.
3036 */
3037 static int pci_rebar_find_pos(struct pci_dev *pdev, int bar)
3038 {
3039 unsigned int pos, nbars, i;
3040 u32 ctrl;
3041
3042 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
3043 if (!pos)
3044 return -ENOTSUPP;
3045
3046 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3047 nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
3048 PCI_REBAR_CTRL_NBAR_SHIFT;
3049
3050 for (i = 0; i < nbars; i++, pos += 8) {
3051 int bar_idx;
3052
3053 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3054 bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
3055 if (bar_idx == bar)
3056 return pos;
3057 }
3058
3059 return -ENOENT;
3060 }
3061
3062 /**
3063 * pci_rebar_get_possible_sizes - get possible sizes for BAR
3064 * @pdev: PCI device
3065 * @bar: BAR to query
3066 *
3067 * Get the possible sizes of a resizable BAR as bitmask defined in the spec
3068 * (bit 0=1MB, bit 19=512GB). Returns 0 if BAR isn't resizable.
3069 */
3070 u32 pci_rebar_get_possible_sizes(struct pci_dev *pdev, int bar)
3071 {
3072 int pos;
3073 u32 cap;
3074
3075 pos = pci_rebar_find_pos(pdev, bar);
3076 if (pos < 0)
3077 return 0;
3078
3079 pci_read_config_dword(pdev, pos + PCI_REBAR_CAP, &cap);
3080 return (cap & PCI_REBAR_CAP_SIZES) >> 4;
3081 }
3082
3083 /**
3084 * pci_rebar_get_current_size - get the current size of a BAR
3085 * @pdev: PCI device
3086 * @bar: BAR to set size to
3087 *
3088 * Read the size of a BAR from the resizable BAR config.
3089 * Returns size if found or negative error code.
3090 */
3091 int pci_rebar_get_current_size(struct pci_dev *pdev, int bar)
3092 {
3093 int pos;
3094 u32 ctrl;
3095
3096 pos = pci_rebar_find_pos(pdev, bar);
3097 if (pos < 0)
3098 return pos;
3099
3100 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3101 return (ctrl & PCI_REBAR_CTRL_BAR_SIZE) >> 8;
3102 }
3103
3104 /**
3105 * pci_rebar_set_size - set a new size for a BAR
3106 * @pdev: PCI device
3107 * @bar: BAR to set size to
3108 * @size: new size as defined in the spec (0=1MB, 19=512GB)
3109 *
3110 * Set the new size of a BAR as defined in the spec.
3111 * Returns zero if resizing was successful, error code otherwise.
3112 */
3113 int pci_rebar_set_size(struct pci_dev *pdev, int bar, int size)
3114 {
3115 int pos;
3116 u32 ctrl;
3117
3118 pos = pci_rebar_find_pos(pdev, bar);
3119 if (pos < 0)
3120 return pos;
3121
3122 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3123 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
3124 ctrl |= size << 8;
3125 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
3126 return 0;
3127 }
3128
3129 /**
3130 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
3131 * @dev: the PCI device
3132 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
3133 *
3134 * Perform INTx swizzling for a device behind one level of bridge. This is
3135 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
3136 * behind bridges on add-in cards. For devices with ARI enabled, the slot
3137 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
3138 * the PCI Express Base Specification, Revision 2.1)
3139 */
3140 u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
3141 {
3142 int slot;
3143
3144 if (pci_ari_enabled(dev->bus))
3145 slot = 0;
3146 else
3147 slot = PCI_SLOT(dev->devfn);
3148
3149 return (((pin - 1) + slot) % 4) + 1;
3150 }
3151
3152 int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3153 {
3154 u8 pin;
3155
3156 pin = dev->pin;
3157 if (!pin)
3158 return -1;
3159
3160 while (!pci_is_root_bus(dev->bus)) {
3161 pin = pci_swizzle_interrupt_pin(dev, pin);
3162 dev = dev->bus->self;
3163 }
3164 *bridge = dev;
3165 return pin;
3166 }
3167
3168 /**
3169 * pci_common_swizzle - swizzle INTx all the way to root bridge
3170 * @dev: the PCI device
3171 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3172 *
3173 * Perform INTx swizzling for a device. This traverses through all PCI-to-PCI
3174 * bridges all the way up to a PCI root bus.
3175 */
3176 u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3177 {
3178 u8 pin = *pinp;
3179
3180 while (!pci_is_root_bus(dev->bus)) {
3181 pin = pci_swizzle_interrupt_pin(dev, pin);
3182 dev = dev->bus->self;
3183 }
3184 *pinp = pin;
3185 return PCI_SLOT(dev->devfn);
3186 }
3187 EXPORT_SYMBOL_GPL(pci_common_swizzle);
3188
3189 /**
3190 * pci_release_region - Release a PCI bar
3191 * @pdev: PCI device whose resources were previously reserved by pci_request_region
3192 * @bar: BAR to release
3193 *
3194 * Releases the PCI I/O and memory resources previously reserved by a
3195 * successful call to pci_request_region. Call this function only
3196 * after all use of the PCI regions has ceased.
3197 */
3198 void pci_release_region(struct pci_dev *pdev, int bar)
3199 {
3200 struct pci_devres *dr;
3201
3202 if (pci_resource_len(pdev, bar) == 0)
3203 return;
3204 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3205 release_region(pci_resource_start(pdev, bar),
3206 pci_resource_len(pdev, bar));
3207 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3208 release_mem_region(pci_resource_start(pdev, bar),
3209 pci_resource_len(pdev, bar));
3210
3211 dr = find_pci_dr(pdev);
3212 if (dr)
3213 dr->region_mask &= ~(1 << bar);
3214 }
3215 EXPORT_SYMBOL(pci_release_region);
3216
3217 /**
3218 * __pci_request_region - Reserved PCI I/O and memory resource
3219 * @pdev: PCI device whose resources are to be reserved
3220 * @bar: BAR to be reserved
3221 * @res_name: Name to be associated with resource.
3222 * @exclusive: whether the region access is exclusive or not
3223 *
3224 * Mark the PCI region associated with PCI device @pdev BR @bar as
3225 * being reserved by owner @res_name. Do not access any
3226 * address inside the PCI regions unless this call returns
3227 * successfully.
3228 *
3229 * If @exclusive is set, then the region is marked so that userspace
3230 * is explicitly not allowed to map the resource via /dev/mem or
3231 * sysfs MMIO access.
3232 *
3233 * Returns 0 on success, or %EBUSY on error. A warning
3234 * message is also printed on failure.
3235 */
3236 static int __pci_request_region(struct pci_dev *pdev, int bar,
3237 const char *res_name, int exclusive)
3238 {
3239 struct pci_devres *dr;
3240
3241 if (pci_resource_len(pdev, bar) == 0)
3242 return 0;
3243
3244 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3245 if (!request_region(pci_resource_start(pdev, bar),
3246 pci_resource_len(pdev, bar), res_name))
3247 goto err_out;
3248 } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3249 if (!__request_mem_region(pci_resource_start(pdev, bar),
3250 pci_resource_len(pdev, bar), res_name,
3251 exclusive))
3252 goto err_out;
3253 }
3254
3255 dr = find_pci_dr(pdev);
3256 if (dr)
3257 dr->region_mask |= 1 << bar;
3258
3259 return 0;
3260
3261 err_out:
3262 dev_warn(&pdev->dev, "BAR %d: can't reserve %pR\n", bar,
3263 &pdev->resource[bar]);
3264 return -EBUSY;
3265 }
3266
3267 /**
3268 * pci_request_region - Reserve PCI I/O and memory resource
3269 * @pdev: PCI device whose resources are to be reserved
3270 * @bar: BAR to be reserved
3271 * @res_name: Name to be associated with resource
3272 *
3273 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3274 * being reserved by owner @res_name. Do not access any
3275 * address inside the PCI regions unless this call returns
3276 * successfully.
3277 *
3278 * Returns 0 on success, or %EBUSY on error. A warning
3279 * message is also printed on failure.
3280 */
3281 int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3282 {
3283 return __pci_request_region(pdev, bar, res_name, 0);
3284 }
3285 EXPORT_SYMBOL(pci_request_region);
3286
3287 /**
3288 * pci_request_region_exclusive - Reserved PCI I/O and memory resource
3289 * @pdev: PCI device whose resources are to be reserved
3290 * @bar: BAR to be reserved
3291 * @res_name: Name to be associated with resource.
3292 *
3293 * Mark the PCI region associated with PCI device @pdev BR @bar as
3294 * being reserved by owner @res_name. Do not access any
3295 * address inside the PCI regions unless this call returns
3296 * successfully.
3297 *
3298 * Returns 0 on success, or %EBUSY on error. A warning
3299 * message is also printed on failure.
3300 *
3301 * The key difference that _exclusive makes it that userspace is
3302 * explicitly not allowed to map the resource via /dev/mem or
3303 * sysfs.
3304 */
3305 int pci_request_region_exclusive(struct pci_dev *pdev, int bar,
3306 const char *res_name)
3307 {
3308 return __pci_request_region(pdev, bar, res_name, IORESOURCE_EXCLUSIVE);
3309 }
3310 EXPORT_SYMBOL(pci_request_region_exclusive);
3311
3312 /**
3313 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3314 * @pdev: PCI device whose resources were previously reserved
3315 * @bars: Bitmask of BARs to be released
3316 *
3317 * Release selected PCI I/O and memory resources previously reserved.
3318 * Call this function only after all use of the PCI regions has ceased.
3319 */
3320 void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3321 {
3322 int i;
3323
3324 for (i = 0; i < 6; i++)
3325 if (bars & (1 << i))
3326 pci_release_region(pdev, i);
3327 }
3328 EXPORT_SYMBOL(pci_release_selected_regions);
3329
3330 static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3331 const char *res_name, int excl)
3332 {
3333 int i;
3334
3335 for (i = 0; i < 6; i++)
3336 if (bars & (1 << i))
3337 if (__pci_request_region(pdev, i, res_name, excl))
3338 goto err_out;
3339 return 0;
3340
3341 err_out:
3342 while (--i >= 0)
3343 if (bars & (1 << i))
3344 pci_release_region(pdev, i);
3345
3346 return -EBUSY;
3347 }
3348
3349
3350 /**
3351 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3352 * @pdev: PCI device whose resources are to be reserved
3353 * @bars: Bitmask of BARs to be requested
3354 * @res_name: Name to be associated with resource
3355 */
3356 int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3357 const char *res_name)
3358 {
3359 return __pci_request_selected_regions(pdev, bars, res_name, 0);
3360 }
3361 EXPORT_SYMBOL(pci_request_selected_regions);
3362
3363 int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3364 const char *res_name)
3365 {
3366 return __pci_request_selected_regions(pdev, bars, res_name,
3367 IORESOURCE_EXCLUSIVE);
3368 }
3369 EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3370
3371 /**
3372 * pci_release_regions - Release reserved PCI I/O and memory resources
3373 * @pdev: PCI device whose resources were previously reserved by pci_request_regions
3374 *
3375 * Releases all PCI I/O and memory resources previously reserved by a
3376 * successful call to pci_request_regions. Call this function only
3377 * after all use of the PCI regions has ceased.
3378 */
3379
3380 void pci_release_regions(struct pci_dev *pdev)
3381 {
3382 pci_release_selected_regions(pdev, (1 << 6) - 1);
3383 }
3384 EXPORT_SYMBOL(pci_release_regions);
3385
3386 /**
3387 * pci_request_regions - Reserved PCI I/O and memory resources
3388 * @pdev: PCI device whose resources are to be reserved
3389 * @res_name: Name to be associated with resource.
3390 *
3391 * Mark all PCI regions associated with PCI device @pdev as
3392 * being reserved by owner @res_name. Do not access any
3393 * address inside the PCI regions unless this call returns
3394 * successfully.
3395 *
3396 * Returns 0 on success, or %EBUSY on error. A warning
3397 * message is also printed on failure.
3398 */
3399 int pci_request_regions(struct pci_dev *pdev, const char *res_name)
3400 {
3401 return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
3402 }
3403 EXPORT_SYMBOL(pci_request_regions);
3404
3405 /**
3406 * pci_request_regions_exclusive - Reserved PCI I/O and memory resources
3407 * @pdev: PCI device whose resources are to be reserved
3408 * @res_name: Name to be associated with resource.
3409 *
3410 * Mark all PCI regions associated with PCI device @pdev as
3411 * being reserved by owner @res_name. Do not access any
3412 * address inside the PCI regions unless this call returns
3413 * successfully.
3414 *
3415 * pci_request_regions_exclusive() will mark the region so that
3416 * /dev/mem and the sysfs MMIO access will not be allowed.
3417 *
3418 * Returns 0 on success, or %EBUSY on error. A warning
3419 * message is also printed on failure.
3420 */
3421 int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3422 {
3423 return pci_request_selected_regions_exclusive(pdev,
3424 ((1 << 6) - 1), res_name);
3425 }
3426 EXPORT_SYMBOL(pci_request_regions_exclusive);
3427
3428 /*
3429 * Record the PCI IO range (expressed as CPU physical address + size).
3430 * Return a negative value if an error has occured, zero otherwise
3431 */
3432 int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
3433 resource_size_t size)
3434 {
3435 int ret = 0;
3436 #ifdef PCI_IOBASE
3437 struct logic_pio_hwaddr *range;
3438
3439 if (!size || addr + size < addr)
3440 return -EINVAL;
3441
3442 range = kzalloc(sizeof(*range), GFP_ATOMIC);
3443 if (!range)
3444 return -ENOMEM;
3445
3446 range->fwnode = fwnode;
3447 range->size = size;
3448 range->hw_start = addr;
3449 range->flags = LOGIC_PIO_CPU_MMIO;
3450
3451 ret = logic_pio_register_range(range);
3452 if (ret)
3453 kfree(range);
3454 #endif
3455
3456 return ret;
3457 }
3458
3459 phys_addr_t pci_pio_to_address(unsigned long pio)
3460 {
3461 phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3462
3463 #ifdef PCI_IOBASE
3464 if (pio >= MMIO_UPPER_LIMIT)
3465 return address;
3466
3467 address = logic_pio_to_hwaddr(pio);
3468 #endif
3469
3470 return address;
3471 }
3472
3473 unsigned long __weak pci_address_to_pio(phys_addr_t address)
3474 {
3475 #ifdef PCI_IOBASE
3476 return logic_pio_trans_cpuaddr(address);
3477 #else
3478 if (address > IO_SPACE_LIMIT)
3479 return (unsigned long)-1;
3480
3481 return (unsigned long) address;
3482 #endif
3483 }
3484
3485 /**
3486 * pci_remap_iospace - Remap the memory mapped I/O space
3487 * @res: Resource describing the I/O space
3488 * @phys_addr: physical address of range to be mapped
3489 *
3490 * Remap the memory mapped I/O space described by the @res
3491 * and the CPU physical address @phys_addr into virtual address space.
3492 * Only architectures that have memory mapped IO functions defined
3493 * (and the PCI_IOBASE value defined) should call this function.
3494 */
3495 int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3496 {
3497 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3498 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3499
3500 if (!(res->flags & IORESOURCE_IO))
3501 return -EINVAL;
3502
3503 if (res->end > IO_SPACE_LIMIT)
3504 return -EINVAL;
3505
3506 return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3507 pgprot_device(PAGE_KERNEL));
3508 #else
3509 /* this architecture does not have memory mapped I/O space,
3510 so this function should never be called */
3511 WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3512 return -ENODEV;
3513 #endif
3514 }
3515 EXPORT_SYMBOL(pci_remap_iospace);
3516
3517 /**
3518 * pci_unmap_iospace - Unmap the memory mapped I/O space
3519 * @res: resource to be unmapped
3520 *
3521 * Unmap the CPU virtual address @res from virtual address space.
3522 * Only architectures that have memory mapped IO functions defined
3523 * (and the PCI_IOBASE value defined) should call this function.
3524 */
3525 void pci_unmap_iospace(struct resource *res)
3526 {
3527 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3528 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3529
3530 unmap_kernel_range(vaddr, resource_size(res));
3531 #endif
3532 }
3533 EXPORT_SYMBOL(pci_unmap_iospace);
3534
3535 /**
3536 * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
3537 * @dev: Generic device to remap IO address for
3538 * @offset: Resource address to map
3539 * @size: Size of map
3540 *
3541 * Managed pci_remap_cfgspace(). Map is automatically unmapped on driver
3542 * detach.
3543 */
3544 void __iomem *devm_pci_remap_cfgspace(struct device *dev,
3545 resource_size_t offset,
3546 resource_size_t size)
3547 {
3548 void __iomem **ptr, *addr;
3549
3550 ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
3551 if (!ptr)
3552 return NULL;
3553
3554 addr = pci_remap_cfgspace(offset, size);
3555 if (addr) {
3556 *ptr = addr;
3557 devres_add(dev, ptr);
3558 } else
3559 devres_free(ptr);
3560
3561 return addr;
3562 }
3563 EXPORT_SYMBOL(devm_pci_remap_cfgspace);
3564
3565 /**
3566 * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
3567 * @dev: generic device to handle the resource for
3568 * @res: configuration space resource to be handled
3569 *
3570 * Checks that a resource is a valid memory region, requests the memory
3571 * region and ioremaps with pci_remap_cfgspace() API that ensures the
3572 * proper PCI configuration space memory attributes are guaranteed.
3573 *
3574 * All operations are managed and will be undone on driver detach.
3575 *
3576 * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
3577 * on failure. Usage example::
3578 *
3579 * res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
3580 * base = devm_pci_remap_cfg_resource(&pdev->dev, res);
3581 * if (IS_ERR(base))
3582 * return PTR_ERR(base);
3583 */
3584 void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
3585 struct resource *res)
3586 {
3587 resource_size_t size;
3588 const char *name;
3589 void __iomem *dest_ptr;
3590
3591 BUG_ON(!dev);
3592
3593 if (!res || resource_type(res) != IORESOURCE_MEM) {
3594 dev_err(dev, "invalid resource\n");
3595 return IOMEM_ERR_PTR(-EINVAL);
3596 }
3597
3598 size = resource_size(res);
3599 name = res->name ?: dev_name(dev);
3600
3601 if (!devm_request_mem_region(dev, res->start, size, name)) {
3602 dev_err(dev, "can't request region for resource %pR\n", res);
3603 return IOMEM_ERR_PTR(-EBUSY);
3604 }
3605
3606 dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
3607 if (!dest_ptr) {
3608 dev_err(dev, "ioremap failed for resource %pR\n", res);
3609 devm_release_mem_region(dev, res->start, size);
3610 dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
3611 }
3612
3613 return dest_ptr;
3614 }
3615 EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
3616
3617 static void __pci_set_master(struct pci_dev *dev, bool enable)
3618 {
3619 u16 old_cmd, cmd;
3620
3621 pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
3622 if (enable)
3623 cmd = old_cmd | PCI_COMMAND_MASTER;
3624 else
3625 cmd = old_cmd & ~PCI_COMMAND_MASTER;
3626 if (cmd != old_cmd) {
3627 dev_dbg(&dev->dev, "%s bus mastering\n",
3628 enable ? "enabling" : "disabling");
3629 pci_write_config_word(dev, PCI_COMMAND, cmd);
3630 }
3631 dev->is_busmaster = enable;
3632 }
3633
3634 /**
3635 * pcibios_setup - process "pci=" kernel boot arguments
3636 * @str: string used to pass in "pci=" kernel boot arguments
3637 *
3638 * Process kernel boot arguments. This is the default implementation.
3639 * Architecture specific implementations can override this as necessary.
3640 */
3641 char * __weak __init pcibios_setup(char *str)
3642 {
3643 return str;
3644 }
3645
3646 /**
3647 * pcibios_set_master - enable PCI bus-mastering for device dev
3648 * @dev: the PCI device to enable
3649 *
3650 * Enables PCI bus-mastering for the device. This is the default
3651 * implementation. Architecture specific implementations can override
3652 * this if necessary.
3653 */
3654 void __weak pcibios_set_master(struct pci_dev *dev)
3655 {
3656 u8 lat;
3657
3658 /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
3659 if (pci_is_pcie(dev))
3660 return;
3661
3662 pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
3663 if (lat < 16)
3664 lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
3665 else if (lat > pcibios_max_latency)
3666 lat = pcibios_max_latency;
3667 else
3668 return;
3669
3670 pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
3671 }
3672
3673 /**
3674 * pci_set_master - enables bus-mastering for device dev
3675 * @dev: the PCI device to enable
3676 *
3677 * Enables bus-mastering on the device and calls pcibios_set_master()
3678 * to do the needed arch specific settings.
3679 */
3680 void pci_set_master(struct pci_dev *dev)
3681 {
3682 __pci_set_master(dev, true);
3683 pcibios_set_master(dev);
3684 }
3685 EXPORT_SYMBOL(pci_set_master);
3686
3687 /**
3688 * pci_clear_master - disables bus-mastering for device dev
3689 * @dev: the PCI device to disable
3690 */
3691 void pci_clear_master(struct pci_dev *dev)
3692 {
3693 __pci_set_master(dev, false);
3694 }
3695 EXPORT_SYMBOL(pci_clear_master);
3696
3697 /**
3698 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
3699 * @dev: the PCI device for which MWI is to be enabled
3700 *
3701 * Helper function for pci_set_mwi.
3702 * Originally copied from drivers/net/acenic.c.
3703 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
3704 *
3705 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3706 */
3707 int pci_set_cacheline_size(struct pci_dev *dev)
3708 {
3709 u8 cacheline_size;
3710
3711 if (!pci_cache_line_size)
3712 return -EINVAL;
3713
3714 /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
3715 equal to or multiple of the right value. */
3716 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3717 if (cacheline_size >= pci_cache_line_size &&
3718 (cacheline_size % pci_cache_line_size) == 0)
3719 return 0;
3720
3721 /* Write the correct value. */
3722 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
3723 /* Read it back. */
3724 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3725 if (cacheline_size == pci_cache_line_size)
3726 return 0;
3727
3728 dev_printk(KERN_DEBUG, &dev->dev, "cache line size of %d is not supported\n",
3729 pci_cache_line_size << 2);
3730
3731 return -EINVAL;
3732 }
3733 EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
3734
3735 /**
3736 * pci_set_mwi - enables memory-write-invalidate PCI transaction
3737 * @dev: the PCI device for which MWI is enabled
3738 *
3739 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3740 *
3741 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3742 */
3743 int pci_set_mwi(struct pci_dev *dev)
3744 {
3745 #ifdef PCI_DISABLE_MWI
3746 return 0;
3747 #else
3748 int rc;
3749 u16 cmd;
3750
3751 rc = pci_set_cacheline_size(dev);
3752 if (rc)
3753 return rc;
3754
3755 pci_read_config_word(dev, PCI_COMMAND, &cmd);
3756 if (!(cmd & PCI_COMMAND_INVALIDATE)) {
3757 dev_dbg(&dev->dev, "enabling Mem-Wr-Inval\n");
3758 cmd |= PCI_COMMAND_INVALIDATE;
3759 pci_write_config_word(dev, PCI_COMMAND, cmd);
3760 }
3761 return 0;
3762 #endif
3763 }
3764 EXPORT_SYMBOL(pci_set_mwi);
3765
3766 /**
3767 * pcim_set_mwi - a device-managed pci_set_mwi()
3768 * @dev: the PCI device for which MWI is enabled
3769 *
3770 * Managed pci_set_mwi().
3771 *
3772 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3773 */
3774 int pcim_set_mwi(struct pci_dev *dev)
3775 {
3776 struct pci_devres *dr;
3777
3778 dr = find_pci_dr(dev);
3779 if (!dr)
3780 return -ENOMEM;
3781
3782 dr->mwi = 1;
3783 return pci_set_mwi(dev);
3784 }
3785 EXPORT_SYMBOL(pcim_set_mwi);
3786
3787 /**
3788 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
3789 * @dev: the PCI device for which MWI is enabled
3790 *
3791 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3792 * Callers are not required to check the return value.
3793 *
3794 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3795 */
3796 int pci_try_set_mwi(struct pci_dev *dev)
3797 {
3798 #ifdef PCI_DISABLE_MWI
3799 return 0;
3800 #else
3801 return pci_set_mwi(dev);
3802 #endif
3803 }
3804 EXPORT_SYMBOL(pci_try_set_mwi);
3805
3806 /**
3807 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
3808 * @dev: the PCI device to disable
3809 *
3810 * Disables PCI Memory-Write-Invalidate transaction on the device
3811 */
3812 void pci_clear_mwi(struct pci_dev *dev)
3813 {
3814 #ifndef PCI_DISABLE_MWI
3815 u16 cmd;
3816
3817 pci_read_config_word(dev, PCI_COMMAND, &cmd);
3818 if (cmd & PCI_COMMAND_INVALIDATE) {
3819 cmd &= ~PCI_COMMAND_INVALIDATE;
3820 pci_write_config_word(dev, PCI_COMMAND, cmd);
3821 }
3822 #endif
3823 }
3824 EXPORT_SYMBOL(pci_clear_mwi);
3825
3826 /**
3827 * pci_intx - enables/disables PCI INTx for device dev
3828 * @pdev: the PCI device to operate on
3829 * @enable: boolean: whether to enable or disable PCI INTx
3830 *
3831 * Enables/disables PCI INTx for device dev
3832 */
3833 void pci_intx(struct pci_dev *pdev, int enable)
3834 {
3835 u16 pci_command, new;
3836
3837 pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
3838
3839 if (enable)
3840 new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
3841 else
3842 new = pci_command | PCI_COMMAND_INTX_DISABLE;
3843
3844 if (new != pci_command) {
3845 struct pci_devres *dr;
3846
3847 pci_write_config_word(pdev, PCI_COMMAND, new);
3848
3849 dr = find_pci_dr(pdev);
3850 if (dr && !dr->restore_intx) {
3851 dr->restore_intx = 1;
3852 dr->orig_intx = !enable;
3853 }
3854 }
3855 }
3856 EXPORT_SYMBOL_GPL(pci_intx);
3857
3858 static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
3859 {
3860 struct pci_bus *bus = dev->bus;
3861 bool mask_updated = true;
3862 u32 cmd_status_dword;
3863 u16 origcmd, newcmd;
3864 unsigned long flags;
3865 bool irq_pending;
3866
3867 /*
3868 * We do a single dword read to retrieve both command and status.
3869 * Document assumptions that make this possible.
3870 */
3871 BUILD_BUG_ON(PCI_COMMAND % 4);
3872 BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
3873
3874 raw_spin_lock_irqsave(&pci_lock, flags);
3875
3876 bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
3877
3878 irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
3879
3880 /*
3881 * Check interrupt status register to see whether our device
3882 * triggered the interrupt (when masking) or the next IRQ is
3883 * already pending (when unmasking).
3884 */
3885 if (mask != irq_pending) {
3886 mask_updated = false;
3887 goto done;
3888 }
3889
3890 origcmd = cmd_status_dword;
3891 newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
3892 if (mask)
3893 newcmd |= PCI_COMMAND_INTX_DISABLE;
3894 if (newcmd != origcmd)
3895 bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
3896
3897 done:
3898 raw_spin_unlock_irqrestore(&pci_lock, flags);
3899
3900 return mask_updated;
3901 }
3902
3903 /**
3904 * pci_check_and_mask_intx - mask INTx on pending interrupt
3905 * @dev: the PCI device to operate on
3906 *
3907 * Check if the device dev has its INTx line asserted, mask it and
3908 * return true in that case. False is returned if no interrupt was
3909 * pending.
3910 */
3911 bool pci_check_and_mask_intx(struct pci_dev *dev)
3912 {
3913 return pci_check_and_set_intx_mask(dev, true);
3914 }
3915 EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
3916
3917 /**
3918 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
3919 * @dev: the PCI device to operate on
3920 *
3921 * Check if the device dev has its INTx line asserted, unmask it if not
3922 * and return true. False is returned and the mask remains active if
3923 * there was still an interrupt pending.
3924 */
3925 bool pci_check_and_unmask_intx(struct pci_dev *dev)
3926 {
3927 return pci_check_and_set_intx_mask(dev, false);
3928 }
3929 EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
3930
3931 /**
3932 * pci_wait_for_pending_transaction - waits for pending transaction
3933 * @dev: the PCI device to operate on
3934 *
3935 * Return 0 if transaction is pending 1 otherwise.
3936 */
3937 int pci_wait_for_pending_transaction(struct pci_dev *dev)
3938 {
3939 if (!pci_is_pcie(dev))
3940 return 1;
3941
3942 return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
3943 PCI_EXP_DEVSTA_TRPND);
3944 }
3945 EXPORT_SYMBOL(pci_wait_for_pending_transaction);
3946
3947 static void pci_flr_wait(struct pci_dev *dev)
3948 {
3949 int delay = 1, timeout = 60000;
3950 u32 id;
3951
3952 /*
3953 * Per PCIe r3.1, sec 6.6.2, a device must complete an FLR within
3954 * 100ms, but may silently discard requests while the FLR is in
3955 * progress. Wait 100ms before trying to access the device.
3956 */
3957 msleep(100);
3958
3959 /*
3960 * After 100ms, the device should not silently discard config
3961 * requests, but it may still indicate that it needs more time by
3962 * responding to them with CRS completions. The Root Port will
3963 * generally synthesize ~0 data to complete the read (except when
3964 * CRS SV is enabled and the read was for the Vendor ID; in that
3965 * case it synthesizes 0x0001 data).
3966 *
3967 * Wait for the device to return a non-CRS completion. Read the
3968 * Command register instead of Vendor ID so we don't have to
3969 * contend with the CRS SV value.
3970 */
3971 pci_read_config_dword(dev, PCI_COMMAND, &id);
3972 while (id == ~0) {
3973 if (delay > timeout) {
3974 dev_warn(&dev->dev, "not ready %dms after FLR; giving up\n",
3975 100 + delay - 1);
3976 return;
3977 }
3978
3979 if (delay > 1000)
3980 dev_info(&dev->dev, "not ready %dms after FLR; waiting\n",
3981 100 + delay - 1);
3982
3983 msleep(delay);
3984 delay *= 2;
3985 pci_read_config_dword(dev, PCI_COMMAND, &id);
3986 }
3987
3988 if (delay > 1000)
3989 dev_info(&dev->dev, "ready %dms after FLR\n", 100 + delay - 1);
3990 }
3991
3992 /**
3993 * pcie_has_flr - check if a device supports function level resets
3994 * @dev: device to check
3995 *
3996 * Returns true if the device advertises support for PCIe function level
3997 * resets.
3998 */
3999 static bool pcie_has_flr(struct pci_dev *dev)
4000 {
4001 u32 cap;
4002
4003 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4004 return false;
4005
4006 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
4007 return cap & PCI_EXP_DEVCAP_FLR;
4008 }
4009
4010 /**
4011 * pcie_flr - initiate a PCIe function level reset
4012 * @dev: device to reset
4013 *
4014 * Initiate a function level reset on @dev. The caller should ensure the
4015 * device supports FLR before calling this function, e.g. by using the
4016 * pcie_has_flr() helper.
4017 */
4018 void pcie_flr(struct pci_dev *dev)
4019 {
4020 if (!pci_wait_for_pending_transaction(dev))
4021 dev_err(&dev->dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
4022
4023 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
4024 pci_flr_wait(dev);
4025 }
4026 EXPORT_SYMBOL_GPL(pcie_flr);
4027
4028 static int pci_af_flr(struct pci_dev *dev, int probe)
4029 {
4030 int pos;
4031 u8 cap;
4032
4033 pos = pci_find_capability(dev, PCI_CAP_ID_AF);
4034 if (!pos)
4035 return -ENOTTY;
4036
4037 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4038 return -ENOTTY;
4039
4040 pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
4041 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
4042 return -ENOTTY;
4043
4044 if (probe)
4045 return 0;
4046
4047 /*
4048 * Wait for Transaction Pending bit to clear. A word-aligned test
4049 * is used, so we use the conrol offset rather than status and shift
4050 * the test bit to match.
4051 */
4052 if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
4053 PCI_AF_STATUS_TP << 8))
4054 dev_err(&dev->dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
4055
4056 pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
4057 pci_flr_wait(dev);
4058 return 0;
4059 }
4060
4061 /**
4062 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
4063 * @dev: Device to reset.
4064 * @probe: If set, only check if the device can be reset this way.
4065 *
4066 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
4067 * unset, it will be reinitialized internally when going from PCI_D3hot to
4068 * PCI_D0. If that's the case and the device is not in a low-power state
4069 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
4070 *
4071 * NOTE: This causes the caller to sleep for twice the device power transition
4072 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
4073 * by default (i.e. unless the @dev's d3_delay field has a different value).
4074 * Moreover, only devices in D0 can be reset by this function.
4075 */
4076 static int pci_pm_reset(struct pci_dev *dev, int probe)
4077 {
4078 u16 csr;
4079
4080 if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
4081 return -ENOTTY;
4082
4083 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
4084 if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
4085 return -ENOTTY;
4086
4087 if (probe)
4088 return 0;
4089
4090 if (dev->current_state != PCI_D0)
4091 return -EINVAL;
4092
4093 csr &= ~PCI_PM_CTRL_STATE_MASK;
4094 csr |= PCI_D3hot;
4095 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4096 pci_dev_d3_sleep(dev);
4097
4098 csr &= ~PCI_PM_CTRL_STATE_MASK;
4099 csr |= PCI_D0;
4100 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4101 pci_dev_d3_sleep(dev);
4102
4103 return 0;
4104 }
4105
4106 void pci_reset_secondary_bus(struct pci_dev *dev)
4107 {
4108 u16 ctrl;
4109
4110 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
4111 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
4112 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4113 /*
4114 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms. Double
4115 * this to 2ms to ensure that we meet the minimum requirement.
4116 */
4117 msleep(2);
4118
4119 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4120 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4121
4122 /*
4123 * Trhfa for conventional PCI is 2^25 clock cycles.
4124 * Assuming a minimum 33MHz clock this results in a 1s
4125 * delay before we can consider subordinate devices to
4126 * be re-initialized. PCIe has some ways to shorten this,
4127 * but we don't make use of them yet.
4128 */
4129 ssleep(1);
4130 }
4131
4132 void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4133 {
4134 pci_reset_secondary_bus(dev);
4135 }
4136
4137 /**
4138 * pci_reset_bridge_secondary_bus - Reset the secondary bus on a PCI bridge.
4139 * @dev: Bridge device
4140 *
4141 * Use the bridge control register to assert reset on the secondary bus.
4142 * Devices on the secondary bus are left in power-on state.
4143 */
4144 void pci_reset_bridge_secondary_bus(struct pci_dev *dev)
4145 {
4146 pcibios_reset_secondary_bus(dev);
4147 }
4148 EXPORT_SYMBOL_GPL(pci_reset_bridge_secondary_bus);
4149
4150 static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
4151 {
4152 struct pci_dev *pdev;
4153
4154 if (pci_is_root_bus(dev->bus) || dev->subordinate ||
4155 !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4156 return -ENOTTY;
4157
4158 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4159 if (pdev != dev)
4160 return -ENOTTY;
4161
4162 if (probe)
4163 return 0;
4164
4165 pci_reset_bridge_secondary_bus(dev->bus->self);
4166
4167 return 0;
4168 }
4169
4170 static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
4171 {
4172 int rc = -ENOTTY;
4173
4174 if (!hotplug || !try_module_get(hotplug->ops->owner))
4175 return rc;
4176
4177 if (hotplug->ops->reset_slot)
4178 rc = hotplug->ops->reset_slot(hotplug, probe);
4179
4180 module_put(hotplug->ops->owner);
4181
4182 return rc;
4183 }
4184
4185 static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
4186 {
4187 struct pci_dev *pdev;
4188
4189 if (dev->subordinate || !dev->slot ||
4190 dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4191 return -ENOTTY;
4192
4193 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4194 if (pdev != dev && pdev->slot == dev->slot)
4195 return -ENOTTY;
4196
4197 return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
4198 }
4199
4200 static void pci_dev_lock(struct pci_dev *dev)
4201 {
4202 pci_cfg_access_lock(dev);
4203 /* block PM suspend, driver probe, etc. */
4204 device_lock(&dev->dev);
4205 }
4206
4207 /* Return 1 on successful lock, 0 on contention */
4208 static int pci_dev_trylock(struct pci_dev *dev)
4209 {
4210 if (pci_cfg_access_trylock(dev)) {
4211 if (device_trylock(&dev->dev))
4212 return 1;
4213 pci_cfg_access_unlock(dev);
4214 }
4215
4216 return 0;
4217 }
4218
4219 static void pci_dev_unlock(struct pci_dev *dev)
4220 {
4221 device_unlock(&dev->dev);
4222 pci_cfg_access_unlock(dev);
4223 }
4224
4225 static void pci_dev_save_and_disable(struct pci_dev *dev)
4226 {
4227 const struct pci_error_handlers *err_handler =
4228 dev->driver ? dev->driver->err_handler : NULL;
4229
4230 /*
4231 * dev->driver->err_handler->reset_prepare() is protected against
4232 * races with ->remove() by the device lock, which must be held by
4233 * the caller.
4234 */
4235 if (err_handler && err_handler->reset_prepare)
4236 err_handler->reset_prepare(dev);
4237
4238 /*
4239 * Wake-up device prior to save. PM registers default to D0 after
4240 * reset and a simple register restore doesn't reliably return
4241 * to a non-D0 state anyway.
4242 */
4243 pci_set_power_state(dev, PCI_D0);
4244
4245 pci_save_state(dev);
4246 /*
4247 * Disable the device by clearing the Command register, except for
4248 * INTx-disable which is set. This not only disables MMIO and I/O port
4249 * BARs, but also prevents the device from being Bus Master, preventing
4250 * DMA from the device including MSI/MSI-X interrupts. For PCI 2.3
4251 * compliant devices, INTx-disable prevents legacy interrupts.
4252 */
4253 pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
4254 }
4255
4256 static void pci_dev_restore(struct pci_dev *dev)
4257 {
4258 const struct pci_error_handlers *err_handler =
4259 dev->driver ? dev->driver->err_handler : NULL;
4260
4261 pci_restore_state(dev);
4262
4263 /*
4264 * dev->driver->err_handler->reset_done() is protected against
4265 * races with ->remove() by the device lock, which must be held by
4266 * the caller.
4267 */
4268 if (err_handler && err_handler->reset_done)
4269 err_handler->reset_done(dev);
4270 }
4271
4272 /**
4273 * __pci_reset_function_locked - reset a PCI device function while holding
4274 * the @dev mutex lock.
4275 * @dev: PCI device to reset
4276 *
4277 * Some devices allow an individual function to be reset without affecting
4278 * other functions in the same device. The PCI device must be responsive
4279 * to PCI config space in order to use this function.
4280 *
4281 * The device function is presumed to be unused and the caller is holding
4282 * the device mutex lock when this function is called.
4283 * Resetting the device will make the contents of PCI configuration space
4284 * random, so any caller of this must be prepared to reinitialise the
4285 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4286 * etc.
4287 *
4288 * Returns 0 if the device function was successfully reset or negative if the
4289 * device doesn't support resetting a single function.
4290 */
4291 int __pci_reset_function_locked(struct pci_dev *dev)
4292 {
4293 int rc;
4294
4295 might_sleep();
4296
4297 /*
4298 * A reset method returns -ENOTTY if it doesn't support this device
4299 * and we should try the next method.
4300 *
4301 * If it returns 0 (success), we're finished. If it returns any
4302 * other error, we're also finished: this indicates that further
4303 * reset mechanisms might be broken on the device.
4304 */
4305 rc = pci_dev_specific_reset(dev, 0);
4306 if (rc != -ENOTTY)
4307 return rc;
4308 if (pcie_has_flr(dev)) {
4309 pcie_flr(dev);
4310 return 0;
4311 }
4312 rc = pci_af_flr(dev, 0);
4313 if (rc != -ENOTTY)
4314 return rc;
4315 rc = pci_pm_reset(dev, 0);
4316 if (rc != -ENOTTY)
4317 return rc;
4318 rc = pci_dev_reset_slot_function(dev, 0);
4319 if (rc != -ENOTTY)
4320 return rc;
4321 return pci_parent_bus_reset(dev, 0);
4322 }
4323 EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
4324
4325 /**
4326 * pci_probe_reset_function - check whether the device can be safely reset
4327 * @dev: PCI device to reset
4328 *
4329 * Some devices allow an individual function to be reset without affecting
4330 * other functions in the same device. The PCI device must be responsive
4331 * to PCI config space in order to use this function.
4332 *
4333 * Returns 0 if the device function can be reset or negative if the
4334 * device doesn't support resetting a single function.
4335 */
4336 int pci_probe_reset_function(struct pci_dev *dev)
4337 {
4338 int rc;
4339
4340 might_sleep();
4341
4342 rc = pci_dev_specific_reset(dev, 1);
4343 if (rc != -ENOTTY)
4344 return rc;
4345 if (pcie_has_flr(dev))
4346 return 0;
4347 rc = pci_af_flr(dev, 1);
4348 if (rc != -ENOTTY)
4349 return rc;
4350 rc = pci_pm_reset(dev, 1);
4351 if (rc != -ENOTTY)
4352 return rc;
4353 rc = pci_dev_reset_slot_function(dev, 1);
4354 if (rc != -ENOTTY)
4355 return rc;
4356
4357 return pci_parent_bus_reset(dev, 1);
4358 }
4359
4360 /**
4361 * pci_reset_function - quiesce and reset a PCI device function
4362 * @dev: PCI device to reset
4363 *
4364 * Some devices allow an individual function to be reset without affecting
4365 * other functions in the same device. The PCI device must be responsive
4366 * to PCI config space in order to use this function.
4367 *
4368 * This function does not just reset the PCI portion of a device, but
4369 * clears all the state associated with the device. This function differs
4370 * from __pci_reset_function_locked() in that it saves and restores device state
4371 * over the reset and takes the PCI device lock.
4372 *
4373 * Returns 0 if the device function was successfully reset or negative if the
4374 * device doesn't support resetting a single function.
4375 */
4376 int pci_reset_function(struct pci_dev *dev)
4377 {
4378 int rc;
4379
4380 rc = pci_probe_reset_function(dev);
4381 if (rc)
4382 return rc;
4383
4384 pci_dev_lock(dev);
4385 pci_dev_save_and_disable(dev);
4386
4387 rc = __pci_reset_function_locked(dev);
4388
4389 pci_dev_restore(dev);
4390 pci_dev_unlock(dev);
4391
4392 return rc;
4393 }
4394 EXPORT_SYMBOL_GPL(pci_reset_function);
4395
4396 /**
4397 * pci_reset_function_locked - quiesce and reset a PCI device function
4398 * @dev: PCI device to reset
4399 *
4400 * Some devices allow an individual function to be reset without affecting
4401 * other functions in the same device. The PCI device must be responsive
4402 * to PCI config space in order to use this function.
4403 *
4404 * This function does not just reset the PCI portion of a device, but
4405 * clears all the state associated with the device. This function differs
4406 * from __pci_reset_function_locked() in that it saves and restores device state
4407 * over the reset. It also differs from pci_reset_function() in that it
4408 * requires the PCI device lock to be held.
4409 *
4410 * Returns 0 if the device function was successfully reset or negative if the
4411 * device doesn't support resetting a single function.
4412 */
4413 int pci_reset_function_locked(struct pci_dev *dev)
4414 {
4415 int rc;
4416
4417 rc = pci_probe_reset_function(dev);
4418 if (rc)
4419 return rc;
4420
4421 pci_dev_save_and_disable(dev);
4422
4423 rc = __pci_reset_function_locked(dev);
4424
4425 pci_dev_restore(dev);
4426
4427 return rc;
4428 }
4429 EXPORT_SYMBOL_GPL(pci_reset_function_locked);
4430
4431 /**
4432 * pci_try_reset_function - quiesce and reset a PCI device function
4433 * @dev: PCI device to reset
4434 *
4435 * Same as above, except return -EAGAIN if unable to lock device.
4436 */
4437 int pci_try_reset_function(struct pci_dev *dev)
4438 {
4439 int rc;
4440
4441 rc = pci_probe_reset_function(dev);
4442 if (rc)
4443 return rc;
4444
4445 if (!pci_dev_trylock(dev))
4446 return -EAGAIN;
4447
4448 pci_dev_save_and_disable(dev);
4449 rc = __pci_reset_function_locked(dev);
4450 pci_dev_unlock(dev);
4451
4452 pci_dev_restore(dev);
4453 return rc;
4454 }
4455 EXPORT_SYMBOL_GPL(pci_try_reset_function);
4456
4457 /* Do any devices on or below this bus prevent a bus reset? */
4458 static bool pci_bus_resetable(struct pci_bus *bus)
4459 {
4460 struct pci_dev *dev;
4461
4462
4463 if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
4464 return false;
4465
4466 list_for_each_entry(dev, &bus->devices, bus_list) {
4467 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4468 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4469 return false;
4470 }
4471
4472 return true;
4473 }
4474
4475 /* Lock devices from the top of the tree down */
4476 static void pci_bus_lock(struct pci_bus *bus)
4477 {
4478 struct pci_dev *dev;
4479
4480 list_for_each_entry(dev, &bus->devices, bus_list) {
4481 pci_dev_lock(dev);
4482 if (dev->subordinate)
4483 pci_bus_lock(dev->subordinate);
4484 }
4485 }
4486
4487 /* Unlock devices from the bottom of the tree up */
4488 static void pci_bus_unlock(struct pci_bus *bus)
4489 {
4490 struct pci_dev *dev;
4491
4492 list_for_each_entry(dev, &bus->devices, bus_list) {
4493 if (dev->subordinate)
4494 pci_bus_unlock(dev->subordinate);
4495 pci_dev_unlock(dev);
4496 }
4497 }
4498
4499 /* Return 1 on successful lock, 0 on contention */
4500 static int pci_bus_trylock(struct pci_bus *bus)
4501 {
4502 struct pci_dev *dev;
4503
4504 list_for_each_entry(dev, &bus->devices, bus_list) {
4505 if (!pci_dev_trylock(dev))
4506 goto unlock;
4507 if (dev->subordinate) {
4508 if (!pci_bus_trylock(dev->subordinate)) {
4509 pci_dev_unlock(dev);
4510 goto unlock;
4511 }
4512 }
4513 }
4514 return 1;
4515
4516 unlock:
4517 list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
4518 if (dev->subordinate)
4519 pci_bus_unlock(dev->subordinate);
4520 pci_dev_unlock(dev);
4521 }
4522 return 0;
4523 }
4524
4525 /* Do any devices on or below this slot prevent a bus reset? */
4526 static bool pci_slot_resetable(struct pci_slot *slot)
4527 {
4528 struct pci_dev *dev;
4529
4530 if (slot->bus->self &&
4531 (slot->bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
4532 return false;
4533
4534 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4535 if (!dev->slot || dev->slot != slot)
4536 continue;
4537 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4538 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4539 return false;
4540 }
4541
4542 return true;
4543 }
4544
4545 /* Lock devices from the top of the tree down */
4546 static void pci_slot_lock(struct pci_slot *slot)
4547 {
4548 struct pci_dev *dev;
4549
4550 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4551 if (!dev->slot || dev->slot != slot)
4552 continue;
4553 pci_dev_lock(dev);
4554 if (dev->subordinate)
4555 pci_bus_lock(dev->subordinate);
4556 }
4557 }
4558
4559 /* Unlock devices from the bottom of the tree up */
4560 static void pci_slot_unlock(struct pci_slot *slot)
4561 {
4562 struct pci_dev *dev;
4563
4564 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4565 if (!dev->slot || dev->slot != slot)
4566 continue;
4567 if (dev->subordinate)
4568 pci_bus_unlock(dev->subordinate);
4569 pci_dev_unlock(dev);
4570 }
4571 }
4572
4573 /* Return 1 on successful lock, 0 on contention */
4574 static int pci_slot_trylock(struct pci_slot *slot)
4575 {
4576 struct pci_dev *dev;
4577
4578 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4579 if (!dev->slot || dev->slot != slot)
4580 continue;
4581 if (!pci_dev_trylock(dev))
4582 goto unlock;
4583 if (dev->subordinate) {
4584 if (!pci_bus_trylock(dev->subordinate)) {
4585 pci_dev_unlock(dev);
4586 goto unlock;
4587 }
4588 }
4589 }
4590 return 1;
4591
4592 unlock:
4593 list_for_each_entry_continue_reverse(dev,
4594 &slot->bus->devices, bus_list) {
4595 if (!dev->slot || dev->slot != slot)
4596 continue;
4597 if (dev->subordinate)
4598 pci_bus_unlock(dev->subordinate);
4599 pci_dev_unlock(dev);
4600 }
4601 return 0;
4602 }
4603
4604 /* Save and disable devices from the top of the tree down */
4605 static void pci_bus_save_and_disable(struct pci_bus *bus)
4606 {
4607 struct pci_dev *dev;
4608
4609 list_for_each_entry(dev, &bus->devices, bus_list) {
4610 pci_dev_lock(dev);
4611 pci_dev_save_and_disable(dev);
4612 pci_dev_unlock(dev);
4613 if (dev->subordinate)
4614 pci_bus_save_and_disable(dev->subordinate);
4615 }
4616 }
4617
4618 /*
4619 * Restore devices from top of the tree down - parent bridges need to be
4620 * restored before we can get to subordinate devices.
4621 */
4622 static void pci_bus_restore(struct pci_bus *bus)
4623 {
4624 struct pci_dev *dev;
4625
4626 list_for_each_entry(dev, &bus->devices, bus_list) {
4627 pci_dev_lock(dev);
4628 pci_dev_restore(dev);
4629 pci_dev_unlock(dev);
4630 if (dev->subordinate)
4631 pci_bus_restore(dev->subordinate);
4632 }
4633 }
4634
4635 /* Save and disable devices from the top of the tree down */
4636 static void pci_slot_save_and_disable(struct pci_slot *slot)
4637 {
4638 struct pci_dev *dev;
4639
4640 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4641 if (!dev->slot || dev->slot != slot)
4642 continue;
4643 pci_dev_save_and_disable(dev);
4644 if (dev->subordinate)
4645 pci_bus_save_and_disable(dev->subordinate);
4646 }
4647 }
4648
4649 /*
4650 * Restore devices from top of the tree down - parent bridges need to be
4651 * restored before we can get to subordinate devices.
4652 */
4653 static void pci_slot_restore(struct pci_slot *slot)
4654 {
4655 struct pci_dev *dev;
4656
4657 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4658 if (!dev->slot || dev->slot != slot)
4659 continue;
4660 pci_dev_restore(dev);
4661 if (dev->subordinate)
4662 pci_bus_restore(dev->subordinate);
4663 }
4664 }
4665
4666 static int pci_slot_reset(struct pci_slot *slot, int probe)
4667 {
4668 int rc;
4669
4670 if (!slot || !pci_slot_resetable(slot))
4671 return -ENOTTY;
4672
4673 if (!probe)
4674 pci_slot_lock(slot);
4675
4676 might_sleep();
4677
4678 rc = pci_reset_hotplug_slot(slot->hotplug, probe);
4679
4680 if (!probe)
4681 pci_slot_unlock(slot);
4682
4683 return rc;
4684 }
4685
4686 /**
4687 * pci_probe_reset_slot - probe whether a PCI slot can be reset
4688 * @slot: PCI slot to probe
4689 *
4690 * Return 0 if slot can be reset, negative if a slot reset is not supported.
4691 */
4692 int pci_probe_reset_slot(struct pci_slot *slot)
4693 {
4694 return pci_slot_reset(slot, 1);
4695 }
4696 EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
4697
4698 /**
4699 * pci_reset_slot - reset a PCI slot
4700 * @slot: PCI slot to reset
4701 *
4702 * A PCI bus may host multiple slots, each slot may support a reset mechanism
4703 * independent of other slots. For instance, some slots may support slot power
4704 * control. In the case of a 1:1 bus to slot architecture, this function may
4705 * wrap the bus reset to avoid spurious slot related events such as hotplug.
4706 * Generally a slot reset should be attempted before a bus reset. All of the
4707 * function of the slot and any subordinate buses behind the slot are reset
4708 * through this function. PCI config space of all devices in the slot and
4709 * behind the slot is saved before and restored after reset.
4710 *
4711 * Return 0 on success, non-zero on error.
4712 */
4713 int pci_reset_slot(struct pci_slot *slot)
4714 {
4715 int rc;
4716
4717 rc = pci_slot_reset(slot, 1);
4718 if (rc)
4719 return rc;
4720
4721 pci_slot_save_and_disable(slot);
4722
4723 rc = pci_slot_reset(slot, 0);
4724
4725 pci_slot_restore(slot);
4726
4727 return rc;
4728 }
4729 EXPORT_SYMBOL_GPL(pci_reset_slot);
4730
4731 /**
4732 * pci_try_reset_slot - Try to reset a PCI slot
4733 * @slot: PCI slot to reset
4734 *
4735 * Same as above except return -EAGAIN if the slot cannot be locked
4736 */
4737 int pci_try_reset_slot(struct pci_slot *slot)
4738 {
4739 int rc;
4740
4741 rc = pci_slot_reset(slot, 1);
4742 if (rc)
4743 return rc;
4744
4745 pci_slot_save_and_disable(slot);
4746
4747 if (pci_slot_trylock(slot)) {
4748 might_sleep();
4749 rc = pci_reset_hotplug_slot(slot->hotplug, 0);
4750 pci_slot_unlock(slot);
4751 } else
4752 rc = -EAGAIN;
4753
4754 pci_slot_restore(slot);
4755
4756 return rc;
4757 }
4758 EXPORT_SYMBOL_GPL(pci_try_reset_slot);
4759
4760 static int pci_bus_reset(struct pci_bus *bus, int probe)
4761 {
4762 if (!bus->self || !pci_bus_resetable(bus))
4763 return -ENOTTY;
4764
4765 if (probe)
4766 return 0;
4767
4768 pci_bus_lock(bus);
4769
4770 might_sleep();
4771
4772 pci_reset_bridge_secondary_bus(bus->self);
4773
4774 pci_bus_unlock(bus);
4775
4776 return 0;
4777 }
4778
4779 /**
4780 * pci_probe_reset_bus - probe whether a PCI bus can be reset
4781 * @bus: PCI bus to probe
4782 *
4783 * Return 0 if bus can be reset, negative if a bus reset is not supported.
4784 */
4785 int pci_probe_reset_bus(struct pci_bus *bus)
4786 {
4787 return pci_bus_reset(bus, 1);
4788 }
4789 EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
4790
4791 /**
4792 * pci_reset_bus - reset a PCI bus
4793 * @bus: top level PCI bus to reset
4794 *
4795 * Do a bus reset on the given bus and any subordinate buses, saving
4796 * and restoring state of all devices.
4797 *
4798 * Return 0 on success, non-zero on error.
4799 */
4800 int pci_reset_bus(struct pci_bus *bus)
4801 {
4802 int rc;
4803
4804 rc = pci_bus_reset(bus, 1);
4805 if (rc)
4806 return rc;
4807
4808 pci_bus_save_and_disable(bus);
4809
4810 rc = pci_bus_reset(bus, 0);
4811
4812 pci_bus_restore(bus);
4813
4814 return rc;
4815 }
4816 EXPORT_SYMBOL_GPL(pci_reset_bus);
4817
4818 /**
4819 * pci_try_reset_bus - Try to reset a PCI bus
4820 * @bus: top level PCI bus to reset
4821 *
4822 * Same as above except return -EAGAIN if the bus cannot be locked
4823 */
4824 int pci_try_reset_bus(struct pci_bus *bus)
4825 {
4826 int rc;
4827
4828 rc = pci_bus_reset(bus, 1);
4829 if (rc)
4830 return rc;
4831
4832 pci_bus_save_and_disable(bus);
4833
4834 if (pci_bus_trylock(bus)) {
4835 might_sleep();
4836 pci_reset_bridge_secondary_bus(bus->self);
4837 pci_bus_unlock(bus);
4838 } else
4839 rc = -EAGAIN;
4840
4841 pci_bus_restore(bus);
4842
4843 return rc;
4844 }
4845 EXPORT_SYMBOL_GPL(pci_try_reset_bus);
4846
4847 /**
4848 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
4849 * @dev: PCI device to query
4850 *
4851 * Returns mmrbc: maximum designed memory read count in bytes
4852 * or appropriate error value.
4853 */
4854 int pcix_get_max_mmrbc(struct pci_dev *dev)
4855 {
4856 int cap;
4857 u32 stat;
4858
4859 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4860 if (!cap)
4861 return -EINVAL;
4862
4863 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4864 return -EINVAL;
4865
4866 return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
4867 }
4868 EXPORT_SYMBOL(pcix_get_max_mmrbc);
4869
4870 /**
4871 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
4872 * @dev: PCI device to query
4873 *
4874 * Returns mmrbc: maximum memory read count in bytes
4875 * or appropriate error value.
4876 */
4877 int pcix_get_mmrbc(struct pci_dev *dev)
4878 {
4879 int cap;
4880 u16 cmd;
4881
4882 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4883 if (!cap)
4884 return -EINVAL;
4885
4886 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4887 return -EINVAL;
4888
4889 return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
4890 }
4891 EXPORT_SYMBOL(pcix_get_mmrbc);
4892
4893 /**
4894 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
4895 * @dev: PCI device to query
4896 * @mmrbc: maximum memory read count in bytes
4897 * valid values are 512, 1024, 2048, 4096
4898 *
4899 * If possible sets maximum memory read byte count, some bridges have erratas
4900 * that prevent this.
4901 */
4902 int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
4903 {
4904 int cap;
4905 u32 stat, v, o;
4906 u16 cmd;
4907
4908 if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
4909 return -EINVAL;
4910
4911 v = ffs(mmrbc) - 10;
4912
4913 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4914 if (!cap)
4915 return -EINVAL;
4916
4917 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4918 return -EINVAL;
4919
4920 if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
4921 return -E2BIG;
4922
4923 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4924 return -EINVAL;
4925
4926 o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
4927 if (o != v) {
4928 if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
4929 return -EIO;
4930
4931 cmd &= ~PCI_X_CMD_MAX_READ;
4932 cmd |= v << 2;
4933 if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
4934 return -EIO;
4935 }
4936 return 0;
4937 }
4938 EXPORT_SYMBOL(pcix_set_mmrbc);
4939
4940 /**
4941 * pcie_get_readrq - get PCI Express read request size
4942 * @dev: PCI device to query
4943 *
4944 * Returns maximum memory read request in bytes
4945 * or appropriate error value.
4946 */
4947 int pcie_get_readrq(struct pci_dev *dev)
4948 {
4949 u16 ctl;
4950
4951 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
4952
4953 return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
4954 }
4955 EXPORT_SYMBOL(pcie_get_readrq);
4956
4957 /**
4958 * pcie_set_readrq - set PCI Express maximum memory read request
4959 * @dev: PCI device to query
4960 * @rq: maximum memory read count in bytes
4961 * valid values are 128, 256, 512, 1024, 2048, 4096
4962 *
4963 * If possible sets maximum memory read request in bytes
4964 */
4965 int pcie_set_readrq(struct pci_dev *dev, int rq)
4966 {
4967 u16 v;
4968
4969 if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
4970 return -EINVAL;
4971
4972 /*
4973 * If using the "performance" PCIe config, we clamp the
4974 * read rq size to the max packet size to prevent the
4975 * host bridge generating requests larger than we can
4976 * cope with
4977 */
4978 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
4979 int mps = pcie_get_mps(dev);
4980
4981 if (mps < rq)
4982 rq = mps;
4983 }
4984
4985 v = (ffs(rq) - 8) << 12;
4986
4987 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
4988 PCI_EXP_DEVCTL_READRQ, v);
4989 }
4990 EXPORT_SYMBOL(pcie_set_readrq);
4991
4992 /**
4993 * pcie_get_mps - get PCI Express maximum payload size
4994 * @dev: PCI device to query
4995 *
4996 * Returns maximum payload size in bytes
4997 */
4998 int pcie_get_mps(struct pci_dev *dev)
4999 {
5000 u16 ctl;
5001
5002 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5003
5004 return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
5005 }
5006 EXPORT_SYMBOL(pcie_get_mps);
5007
5008 /**
5009 * pcie_set_mps - set PCI Express maximum payload size
5010 * @dev: PCI device to query
5011 * @mps: maximum payload size in bytes
5012 * valid values are 128, 256, 512, 1024, 2048, 4096
5013 *
5014 * If possible sets maximum payload size
5015 */
5016 int pcie_set_mps(struct pci_dev *dev, int mps)
5017 {
5018 u16 v;
5019
5020 if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
5021 return -EINVAL;
5022
5023 v = ffs(mps) - 8;
5024 if (v > dev->pcie_mpss)
5025 return -EINVAL;
5026 v <<= 5;
5027
5028 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5029 PCI_EXP_DEVCTL_PAYLOAD, v);
5030 }
5031 EXPORT_SYMBOL(pcie_set_mps);
5032
5033 /**
5034 * pcie_get_minimum_link - determine minimum link settings of a PCI device
5035 * @dev: PCI device to query
5036 * @speed: storage for minimum speed
5037 * @width: storage for minimum width
5038 *
5039 * This function will walk up the PCI device chain and determine the minimum
5040 * link width and speed of the device.
5041 */
5042 int pcie_get_minimum_link(struct pci_dev *dev, enum pci_bus_speed *speed,
5043 enum pcie_link_width *width)
5044 {
5045 int ret;
5046
5047 *speed = PCI_SPEED_UNKNOWN;
5048 *width = PCIE_LNK_WIDTH_UNKNOWN;
5049
5050 while (dev) {
5051 u16 lnksta;
5052 enum pci_bus_speed next_speed;
5053 enum pcie_link_width next_width;
5054
5055 ret = pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
5056 if (ret)
5057 return ret;
5058
5059 next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
5060 next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
5061 PCI_EXP_LNKSTA_NLW_SHIFT;
5062
5063 if (next_speed < *speed)
5064 *speed = next_speed;
5065
5066 if (next_width < *width)
5067 *width = next_width;
5068
5069 dev = dev->bus->self;
5070 }
5071
5072 return 0;
5073 }
5074 EXPORT_SYMBOL(pcie_get_minimum_link);
5075
5076 /**
5077 * pci_select_bars - Make BAR mask from the type of resource
5078 * @dev: the PCI device for which BAR mask is made
5079 * @flags: resource type mask to be selected
5080 *
5081 * This helper routine makes bar mask from the type of resource.
5082 */
5083 int pci_select_bars(struct pci_dev *dev, unsigned long flags)
5084 {
5085 int i, bars = 0;
5086 for (i = 0; i < PCI_NUM_RESOURCES; i++)
5087 if (pci_resource_flags(dev, i) & flags)
5088 bars |= (1 << i);
5089 return bars;
5090 }
5091 EXPORT_SYMBOL(pci_select_bars);
5092
5093 /* Some architectures require additional programming to enable VGA */
5094 static arch_set_vga_state_t arch_set_vga_state;
5095
5096 void __init pci_register_set_vga_state(arch_set_vga_state_t func)
5097 {
5098 arch_set_vga_state = func; /* NULL disables */
5099 }
5100
5101 static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
5102 unsigned int command_bits, u32 flags)
5103 {
5104 if (arch_set_vga_state)
5105 return arch_set_vga_state(dev, decode, command_bits,
5106 flags);
5107 return 0;
5108 }
5109
5110 /**
5111 * pci_set_vga_state - set VGA decode state on device and parents if requested
5112 * @dev: the PCI device
5113 * @decode: true = enable decoding, false = disable decoding
5114 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
5115 * @flags: traverse ancestors and change bridges
5116 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
5117 */
5118 int pci_set_vga_state(struct pci_dev *dev, bool decode,
5119 unsigned int command_bits, u32 flags)
5120 {
5121 struct pci_bus *bus;
5122 struct pci_dev *bridge;
5123 u16 cmd;
5124 int rc;
5125
5126 WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
5127
5128 /* ARCH specific VGA enables */
5129 rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
5130 if (rc)
5131 return rc;
5132
5133 if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
5134 pci_read_config_word(dev, PCI_COMMAND, &cmd);
5135 if (decode == true)
5136 cmd |= command_bits;
5137 else
5138 cmd &= ~command_bits;
5139 pci_write_config_word(dev, PCI_COMMAND, cmd);
5140 }
5141
5142 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
5143 return 0;
5144
5145 bus = dev->bus;
5146 while (bus) {
5147 bridge = bus->self;
5148 if (bridge) {
5149 pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
5150 &cmd);
5151 if (decode == true)
5152 cmd |= PCI_BRIDGE_CTL_VGA;
5153 else
5154 cmd &= ~PCI_BRIDGE_CTL_VGA;
5155 pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
5156 cmd);
5157 }
5158 bus = bus->parent;
5159 }
5160 return 0;
5161 }
5162
5163 /**
5164 * pci_add_dma_alias - Add a DMA devfn alias for a device
5165 * @dev: the PCI device for which alias is added
5166 * @devfn: alias slot and function
5167 *
5168 * This helper encodes 8-bit devfn as bit number in dma_alias_mask.
5169 * It should be called early, preferably as PCI fixup header quirk.
5170 */
5171 void pci_add_dma_alias(struct pci_dev *dev, u8 devfn)
5172 {
5173 if (!dev->dma_alias_mask)
5174 dev->dma_alias_mask = kcalloc(BITS_TO_LONGS(U8_MAX),
5175 sizeof(long), GFP_KERNEL);
5176 if (!dev->dma_alias_mask) {
5177 dev_warn(&dev->dev, "Unable to allocate DMA alias mask\n");
5178 return;
5179 }
5180
5181 set_bit(devfn, dev->dma_alias_mask);
5182 dev_info(&dev->dev, "Enabling fixed DMA alias to %02x.%d\n",
5183 PCI_SLOT(devfn), PCI_FUNC(devfn));
5184 }
5185
5186 bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
5187 {
5188 return (dev1->dma_alias_mask &&
5189 test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
5190 (dev2->dma_alias_mask &&
5191 test_bit(dev1->devfn, dev2->dma_alias_mask));
5192 }
5193
5194 bool pci_device_is_present(struct pci_dev *pdev)
5195 {
5196 u32 v;
5197
5198 if (pci_dev_is_disconnected(pdev))
5199 return false;
5200 return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
5201 }
5202 EXPORT_SYMBOL_GPL(pci_device_is_present);
5203
5204 void pci_ignore_hotplug(struct pci_dev *dev)
5205 {
5206 struct pci_dev *bridge = dev->bus->self;
5207
5208 dev->ignore_hotplug = 1;
5209 /* Propagate the "ignore hotplug" setting to the parent bridge. */
5210 if (bridge)
5211 bridge->ignore_hotplug = 1;
5212 }
5213 EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
5214
5215 resource_size_t __weak pcibios_default_alignment(void)
5216 {
5217 return 0;
5218 }
5219
5220 #define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
5221 static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
5222 static DEFINE_SPINLOCK(resource_alignment_lock);
5223
5224 /**
5225 * pci_specified_resource_alignment - get resource alignment specified by user.
5226 * @dev: the PCI device to get
5227 * @resize: whether or not to change resources' size when reassigning alignment
5228 *
5229 * RETURNS: Resource alignment if it is specified.
5230 * Zero if it is not specified.
5231 */
5232 static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
5233 bool *resize)
5234 {
5235 int seg, bus, slot, func, align_order, count;
5236 unsigned short vendor, device, subsystem_vendor, subsystem_device;
5237 resource_size_t align = pcibios_default_alignment();
5238 char *p;
5239
5240 spin_lock(&resource_alignment_lock);
5241 p = resource_alignment_param;
5242 if (!*p && !align)
5243 goto out;
5244 if (pci_has_flag(PCI_PROBE_ONLY)) {
5245 align = 0;
5246 pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
5247 goto out;
5248 }
5249
5250 while (*p) {
5251 count = 0;
5252 if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
5253 p[count] == '@') {
5254 p += count + 1;
5255 } else {
5256 align_order = -1;
5257 }
5258 if (strncmp(p, "pci:", 4) == 0) {
5259 /* PCI vendor/device (subvendor/subdevice) ids are specified */
5260 p += 4;
5261 if (sscanf(p, "%hx:%hx:%hx:%hx%n",
5262 &vendor, &device, &subsystem_vendor, &subsystem_device, &count) != 4) {
5263 if (sscanf(p, "%hx:%hx%n", &vendor, &device, &count) != 2) {
5264 printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: pci:%s\n",
5265 p);
5266 break;
5267 }
5268 subsystem_vendor = subsystem_device = 0;
5269 }
5270 p += count;
5271 if ((!vendor || (vendor == dev->vendor)) &&
5272 (!device || (device == dev->device)) &&
5273 (!subsystem_vendor || (subsystem_vendor == dev->subsystem_vendor)) &&
5274 (!subsystem_device || (subsystem_device == dev->subsystem_device))) {
5275 *resize = true;
5276 if (align_order == -1)
5277 align = PAGE_SIZE;
5278 else
5279 align = 1 << align_order;
5280 /* Found */
5281 break;
5282 }
5283 }
5284 else {
5285 if (sscanf(p, "%x:%x:%x.%x%n",
5286 &seg, &bus, &slot, &func, &count) != 4) {
5287 seg = 0;
5288 if (sscanf(p, "%x:%x.%x%n",
5289 &bus, &slot, &func, &count) != 3) {
5290 /* Invalid format */
5291 printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: %s\n",
5292 p);
5293 break;
5294 }
5295 }
5296 p += count;
5297 if (seg == pci_domain_nr(dev->bus) &&
5298 bus == dev->bus->number &&
5299 slot == PCI_SLOT(dev->devfn) &&
5300 func == PCI_FUNC(dev->devfn)) {
5301 *resize = true;
5302 if (align_order == -1)
5303 align = PAGE_SIZE;
5304 else
5305 align = 1 << align_order;
5306 /* Found */
5307 break;
5308 }
5309 }
5310 if (*p != ';' && *p != ',') {
5311 /* End of param or invalid format */
5312 break;
5313 }
5314 p++;
5315 }
5316 out:
5317 spin_unlock(&resource_alignment_lock);
5318 return align;
5319 }
5320
5321 static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
5322 resource_size_t align, bool resize)
5323 {
5324 struct resource *r = &dev->resource[bar];
5325 resource_size_t size;
5326
5327 if (!(r->flags & IORESOURCE_MEM))
5328 return;
5329
5330 if (r->flags & IORESOURCE_PCI_FIXED) {
5331 dev_info(&dev->dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
5332 bar, r, (unsigned long long)align);
5333 return;
5334 }
5335
5336 size = resource_size(r);
5337 if (size >= align)
5338 return;
5339
5340 /*
5341 * Increase the alignment of the resource. There are two ways we
5342 * can do this:
5343 *
5344 * 1) Increase the size of the resource. BARs are aligned on their
5345 * size, so when we reallocate space for this resource, we'll
5346 * allocate it with the larger alignment. This also prevents
5347 * assignment of any other BARs inside the alignment region, so
5348 * if we're requesting page alignment, this means no other BARs
5349 * will share the page.
5350 *
5351 * The disadvantage is that this makes the resource larger than
5352 * the hardware BAR, which may break drivers that compute things
5353 * based on the resource size, e.g., to find registers at a
5354 * fixed offset before the end of the BAR.
5355 *
5356 * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
5357 * set r->start to the desired alignment. By itself this
5358 * doesn't prevent other BARs being put inside the alignment
5359 * region, but if we realign *every* resource of every device in
5360 * the system, none of them will share an alignment region.
5361 *
5362 * When the user has requested alignment for only some devices via
5363 * the "pci=resource_alignment" argument, "resize" is true and we
5364 * use the first method. Otherwise we assume we're aligning all
5365 * devices and we use the second.
5366 */
5367
5368 dev_info(&dev->dev, "BAR%d %pR: requesting alignment to %#llx\n",
5369 bar, r, (unsigned long long)align);
5370
5371 if (resize) {
5372 r->start = 0;
5373 r->end = align - 1;
5374 } else {
5375 r->flags &= ~IORESOURCE_SIZEALIGN;
5376 r->flags |= IORESOURCE_STARTALIGN;
5377 r->start = align;
5378 r->end = r->start + size - 1;
5379 }
5380 r->flags |= IORESOURCE_UNSET;
5381 }
5382
5383 /*
5384 * This function disables memory decoding and releases memory resources
5385 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
5386 * It also rounds up size to specified alignment.
5387 * Later on, the kernel will assign page-aligned memory resource back
5388 * to the device.
5389 */
5390 void pci_reassigndev_resource_alignment(struct pci_dev *dev)
5391 {
5392 int i;
5393 struct resource *r;
5394 resource_size_t align;
5395 u16 command;
5396 bool resize = false;
5397
5398 /*
5399 * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
5400 * 3.4.1.11. Their resources are allocated from the space
5401 * described by the VF BARx register in the PF's SR-IOV capability.
5402 * We can't influence their alignment here.
5403 */
5404 if (dev->is_virtfn)
5405 return;
5406
5407 /* check if specified PCI is target device to reassign */
5408 align = pci_specified_resource_alignment(dev, &resize);
5409 if (!align)
5410 return;
5411
5412 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
5413 (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
5414 dev_warn(&dev->dev,
5415 "Can't reassign resources to host bridge.\n");
5416 return;
5417 }
5418
5419 dev_info(&dev->dev,
5420 "Disabling memory decoding and releasing memory resources.\n");
5421 pci_read_config_word(dev, PCI_COMMAND, &command);
5422 command &= ~PCI_COMMAND_MEMORY;
5423 pci_write_config_word(dev, PCI_COMMAND, command);
5424
5425 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
5426 pci_request_resource_alignment(dev, i, align, resize);
5427
5428 /*
5429 * Need to disable bridge's resource window,
5430 * to enable the kernel to reassign new resource
5431 * window later on.
5432 */
5433 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE &&
5434 (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
5435 for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
5436 r = &dev->resource[i];
5437 if (!(r->flags & IORESOURCE_MEM))
5438 continue;
5439 r->flags |= IORESOURCE_UNSET;
5440 r->end = resource_size(r) - 1;
5441 r->start = 0;
5442 }
5443 pci_disable_bridge_window(dev);
5444 }
5445 }
5446
5447 static ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
5448 {
5449 if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
5450 count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
5451 spin_lock(&resource_alignment_lock);
5452 strncpy(resource_alignment_param, buf, count);
5453 resource_alignment_param[count] = '\0';
5454 spin_unlock(&resource_alignment_lock);
5455 return count;
5456 }
5457
5458 static ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
5459 {
5460 size_t count;
5461 spin_lock(&resource_alignment_lock);
5462 count = snprintf(buf, size, "%s", resource_alignment_param);
5463 spin_unlock(&resource_alignment_lock);
5464 return count;
5465 }
5466
5467 static ssize_t pci_resource_alignment_show(struct bus_type *bus, char *buf)
5468 {
5469 return pci_get_resource_alignment_param(buf, PAGE_SIZE);
5470 }
5471
5472 static ssize_t pci_resource_alignment_store(struct bus_type *bus,
5473 const char *buf, size_t count)
5474 {
5475 return pci_set_resource_alignment_param(buf, count);
5476 }
5477
5478 static BUS_ATTR(resource_alignment, 0644, pci_resource_alignment_show,
5479 pci_resource_alignment_store);
5480
5481 static int __init pci_resource_alignment_sysfs_init(void)
5482 {
5483 return bus_create_file(&pci_bus_type,
5484 &bus_attr_resource_alignment);
5485 }
5486 late_initcall(pci_resource_alignment_sysfs_init);
5487
5488 static void pci_no_domains(void)
5489 {
5490 #ifdef CONFIG_PCI_DOMAINS
5491 pci_domains_supported = 0;
5492 #endif
5493 }
5494
5495 #ifdef CONFIG_PCI_DOMAINS
5496 static atomic_t __domain_nr = ATOMIC_INIT(-1);
5497
5498 int pci_get_new_domain_nr(void)
5499 {
5500 return atomic_inc_return(&__domain_nr);
5501 }
5502
5503 #ifdef CONFIG_PCI_DOMAINS_GENERIC
5504 static int of_pci_bus_find_domain_nr(struct device *parent)
5505 {
5506 static int use_dt_domains = -1;
5507 int domain = -1;
5508
5509 if (parent)
5510 domain = of_get_pci_domain_nr(parent->of_node);
5511 /*
5512 * Check DT domain and use_dt_domains values.
5513 *
5514 * If DT domain property is valid (domain >= 0) and
5515 * use_dt_domains != 0, the DT assignment is valid since this means
5516 * we have not previously allocated a domain number by using
5517 * pci_get_new_domain_nr(); we should also update use_dt_domains to
5518 * 1, to indicate that we have just assigned a domain number from
5519 * DT.
5520 *
5521 * If DT domain property value is not valid (ie domain < 0), and we
5522 * have not previously assigned a domain number from DT
5523 * (use_dt_domains != 1) we should assign a domain number by
5524 * using the:
5525 *
5526 * pci_get_new_domain_nr()
5527 *
5528 * API and update the use_dt_domains value to keep track of method we
5529 * are using to assign domain numbers (use_dt_domains = 0).
5530 *
5531 * All other combinations imply we have a platform that is trying
5532 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
5533 * which is a recipe for domain mishandling and it is prevented by
5534 * invalidating the domain value (domain = -1) and printing a
5535 * corresponding error.
5536 */
5537 if (domain >= 0 && use_dt_domains) {
5538 use_dt_domains = 1;
5539 } else if (domain < 0 && use_dt_domains != 1) {
5540 use_dt_domains = 0;
5541 domain = pci_get_new_domain_nr();
5542 } else {
5543 dev_err(parent, "Node %pOF has inconsistent \"linux,pci-domain\" property in DT\n",
5544 parent->of_node);
5545 domain = -1;
5546 }
5547
5548 return domain;
5549 }
5550
5551 int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
5552 {
5553 return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
5554 acpi_pci_bus_find_domain_nr(bus);
5555 }
5556 #endif
5557 #endif
5558
5559 /**
5560 * pci_ext_cfg_avail - can we access extended PCI config space?
5561 *
5562 * Returns 1 if we can access PCI extended config space (offsets
5563 * greater than 0xff). This is the default implementation. Architecture
5564 * implementations can override this.
5565 */
5566 int __weak pci_ext_cfg_avail(void)
5567 {
5568 return 1;
5569 }
5570
5571 void __weak pci_fixup_cardbus(struct pci_bus *bus)
5572 {
5573 }
5574 EXPORT_SYMBOL(pci_fixup_cardbus);
5575
5576 static int __init pci_setup(char *str)
5577 {
5578 while (str) {
5579 char *k = strchr(str, ',');
5580 if (k)
5581 *k++ = 0;
5582 if (*str && (str = pcibios_setup(str)) && *str) {
5583 if (!strcmp(str, "nomsi")) {
5584 pci_no_msi();
5585 } else if (!strcmp(str, "noaer")) {
5586 pci_no_aer();
5587 } else if (!strncmp(str, "realloc=", 8)) {
5588 pci_realloc_get_opt(str + 8);
5589 } else if (!strncmp(str, "realloc", 7)) {
5590 pci_realloc_get_opt("on");
5591 } else if (!strcmp(str, "nodomains")) {
5592 pci_no_domains();
5593 } else if (!strncmp(str, "noari", 5)) {
5594 pcie_ari_disabled = true;
5595 } else if (!strncmp(str, "cbiosize=", 9)) {
5596 pci_cardbus_io_size = memparse(str + 9, &str);
5597 } else if (!strncmp(str, "cbmemsize=", 10)) {
5598 pci_cardbus_mem_size = memparse(str + 10, &str);
5599 } else if (!strncmp(str, "resource_alignment=", 19)) {
5600 pci_set_resource_alignment_param(str + 19,
5601 strlen(str + 19));
5602 } else if (!strncmp(str, "ecrc=", 5)) {
5603 pcie_ecrc_get_policy(str + 5);
5604 } else if (!strncmp(str, "hpiosize=", 9)) {
5605 pci_hotplug_io_size = memparse(str + 9, &str);
5606 } else if (!strncmp(str, "hpmemsize=", 10)) {
5607 pci_hotplug_mem_size = memparse(str + 10, &str);
5608 } else if (!strncmp(str, "hpbussize=", 10)) {
5609 pci_hotplug_bus_size =
5610 simple_strtoul(str + 10, &str, 0);
5611 if (pci_hotplug_bus_size > 0xff)
5612 pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
5613 } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
5614 pcie_bus_config = PCIE_BUS_TUNE_OFF;
5615 } else if (!strncmp(str, "pcie_bus_safe", 13)) {
5616 pcie_bus_config = PCIE_BUS_SAFE;
5617 } else if (!strncmp(str, "pcie_bus_perf", 13)) {
5618 pcie_bus_config = PCIE_BUS_PERFORMANCE;
5619 } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
5620 pcie_bus_config = PCIE_BUS_PEER2PEER;
5621 } else if (!strncmp(str, "pcie_scan_all", 13)) {
5622 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
5623 } else {
5624 printk(KERN_ERR "PCI: Unknown option `%s'\n",
5625 str);
5626 }
5627 }
5628 str = k;
5629 }
5630 return 0;
5631 }
5632 early_param("pci", pci_setup);