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