Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * PCI Bus Services, see include/linux/pci.h for further explanation.
4 *
5 * Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
6 * David Mosberger-Tang
7 *
8 * Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
9 */
10
11#include <linux/acpi.h>
12#include <linux/kernel.h>
13#include <linux/delay.h>
14#include <linux/dmi.h>
15#include <linux/init.h>
16#include <linux/of.h>
17#include <linux/of_pci.h>
18#include <linux/pci.h>
19#include <linux/pm.h>
20#include <linux/slab.h>
21#include <linux/module.h>
22#include <linux/spinlock.h>
23#include <linux/string.h>
24#include <linux/log2.h>
25#include <linux/logic_pio.h>
26#include <linux/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
38DEFINE_MUTEX(pci_slot_mutex);
39
40const char *pci_power_names[] = {
41 "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
42};
43EXPORT_SYMBOL_GPL(pci_power_names);
44
45int isa_dma_bridge_buggy;
46EXPORT_SYMBOL(isa_dma_bridge_buggy);
47
48int pci_pci_problems;
49EXPORT_SYMBOL(pci_pci_problems);
50
51unsigned int pci_pm_d3_delay;
52
53static void pci_pme_list_scan(struct work_struct *work);
54
55static LIST_HEAD(pci_pme_list);
56static DEFINE_MUTEX(pci_pme_list_mutex);
57static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
58
59struct pci_pme_device {
60 struct list_head list;
61 struct pci_dev *dev;
62};
63
64#define PME_TIMEOUT 1000 /* How long between PME checks */
65
66static void pci_dev_d3_sleep(struct pci_dev *dev)
67{
68 unsigned int delay = dev->d3_delay;
69
70 if (delay < pci_pm_d3_delay)
71 delay = pci_pm_d3_delay;
72
73 if (delay)
74 msleep(delay);
75}
76
77#ifdef CONFIG_PCI_DOMAINS
78int pci_domains_supported = 1;
79#endif
80
81#define DEFAULT_CARDBUS_IO_SIZE (256)
82#define DEFAULT_CARDBUS_MEM_SIZE (64*1024*1024)
83/* pci=cbmemsize=nnM,cbiosize=nn can override this */
84unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
85unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
86
87#define DEFAULT_HOTPLUG_IO_SIZE (256)
88#define DEFAULT_HOTPLUG_MEM_SIZE (2*1024*1024)
89/* pci=hpmemsize=nnM,hpiosize=nn can override this */
90unsigned long pci_hotplug_io_size = DEFAULT_HOTPLUG_IO_SIZE;
91unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
92
93#define DEFAULT_HOTPLUG_BUS_SIZE 1
94unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
95
96enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
97
98/*
99 * The default CLS is used if arch didn't set CLS explicitly and not
100 * all pci devices agree on the same value. Arch can override either
101 * the dfl or actual value as it sees fit. Don't forget this is
102 * measured in 32-bit words, not bytes.
103 */
104u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
105u8 pci_cache_line_size;
106
107/*
108 * If we set up a device for bus mastering, we need to check the latency
109 * timer as certain BIOSes forget to set it properly.
110 */
111unsigned int pcibios_max_latency = 255;
112
113/* If set, the PCIe ARI capability will not be used. */
114static bool pcie_ari_disabled;
115
116/* If set, the PCIe ATS capability will not be used. */
117static bool pcie_ats_disabled;
118
119/* If set, the PCI config space of each device is printed during boot. */
120bool pci_early_dump;
121
122bool pci_ats_disabled(void)
123{
124 return pcie_ats_disabled;
125}
126
127/* Disable bridge_d3 for all PCIe ports */
128static bool pci_bridge_d3_disable;
129/* Force bridge_d3 for all PCIe ports */
130static bool pci_bridge_d3_force;
131
132static int __init pcie_port_pm_setup(char *str)
133{
134 if (!strcmp(str, "off"))
135 pci_bridge_d3_disable = true;
136 else if (!strcmp(str, "force"))
137 pci_bridge_d3_force = true;
138 return 1;
139}
140__setup("pcie_port_pm=", pcie_port_pm_setup);
141
142/* Time to wait after a reset for device to become responsive */
143#define PCIE_RESET_READY_POLL_MS 60000
144
145/**
146 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
147 * @bus: pointer to PCI bus structure to search
148 *
149 * Given a PCI bus, returns the highest PCI bus number present in the set
150 * including the given PCI bus and its list of child PCI buses.
151 */
152unsigned char pci_bus_max_busnr(struct pci_bus *bus)
153{
154 struct pci_bus *tmp;
155 unsigned char max, n;
156
157 max = bus->busn_res.end;
158 list_for_each_entry(tmp, &bus->children, node) {
159 n = pci_bus_max_busnr(tmp);
160 if (n > max)
161 max = n;
162 }
163 return max;
164}
165EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
166
167#ifdef CONFIG_HAS_IOMEM
168void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
169{
170 struct resource *res = &pdev->resource[bar];
171
172 /*
173 * Make sure the BAR is actually a memory resource, not an IO resource
174 */
175 if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
176 pci_warn(pdev, "can't ioremap BAR %d: %pR\n", bar, res);
177 return NULL;
178 }
179 return ioremap_nocache(res->start, resource_size(res));
180}
181EXPORT_SYMBOL_GPL(pci_ioremap_bar);
182
183void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
184{
185 /*
186 * Make sure the BAR is actually a memory resource, not an IO resource
187 */
188 if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
189 WARN_ON(1);
190 return NULL;
191 }
192 return ioremap_wc(pci_resource_start(pdev, bar),
193 pci_resource_len(pdev, bar));
194}
195EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
196#endif
197
198/**
199 * pci_dev_str_match_path - test if a path string matches a device
200 * @dev: the PCI device to test
201 * @path: string to match the device against
202 * @endptr: pointer to the string after the match
203 *
204 * Test if a string (typically from a kernel parameter) formatted as a
205 * path of device/function addresses matches a PCI device. The string must
206 * be of the form:
207 *
208 * [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
209 *
210 * A path for a device can be obtained using 'lspci -t'. Using a path
211 * is more robust against bus renumbering than using only a single bus,
212 * device and function address.
213 *
214 * Returns 1 if the string matches the device, 0 if it does not and
215 * a negative error code if it fails to parse the string.
216 */
217static int pci_dev_str_match_path(struct pci_dev *dev, const char *path,
218 const char **endptr)
219{
220 int ret;
221 int seg, bus, slot, func;
222 char *wpath, *p;
223 char end;
224
225 *endptr = strchrnul(path, ';');
226
227 wpath = kmemdup_nul(path, *endptr - path, GFP_KERNEL);
228 if (!wpath)
229 return -ENOMEM;
230
231 while (1) {
232 p = strrchr(wpath, '/');
233 if (!p)
234 break;
235 ret = sscanf(p, "/%x.%x%c", &slot, &func, &end);
236 if (ret != 2) {
237 ret = -EINVAL;
238 goto free_and_exit;
239 }
240
241 if (dev->devfn != PCI_DEVFN(slot, func)) {
242 ret = 0;
243 goto free_and_exit;
244 }
245
246 /*
247 * Note: we don't need to get a reference to the upstream
248 * bridge because we hold a reference to the top level
249 * device which should hold a reference to the bridge,
250 * and so on.
251 */
252 dev = pci_upstream_bridge(dev);
253 if (!dev) {
254 ret = 0;
255 goto free_and_exit;
256 }
257
258 *p = 0;
259 }
260
261 ret = sscanf(wpath, "%x:%x:%x.%x%c", &seg, &bus, &slot,
262 &func, &end);
263 if (ret != 4) {
264 seg = 0;
265 ret = sscanf(wpath, "%x:%x.%x%c", &bus, &slot, &func, &end);
266 if (ret != 3) {
267 ret = -EINVAL;
268 goto free_and_exit;
269 }
270 }
271
272 ret = (seg == pci_domain_nr(dev->bus) &&
273 bus == dev->bus->number &&
274 dev->devfn == PCI_DEVFN(slot, func));
275
276free_and_exit:
277 kfree(wpath);
278 return ret;
279}
280
281/**
282 * pci_dev_str_match - test if a string matches a device
283 * @dev: the PCI device to test
284 * @p: string to match the device against
285 * @endptr: pointer to the string after the match
286 *
287 * Test if a string (typically from a kernel parameter) matches a specified
288 * PCI device. The string may be of one of the following formats:
289 *
290 * [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
291 * pci:<vendor>:<device>[:<subvendor>:<subdevice>]
292 *
293 * The first format specifies a PCI bus/device/function address which
294 * may change if new hardware is inserted, if motherboard firmware changes,
295 * or due to changes caused in kernel parameters. If the domain is
296 * left unspecified, it is taken to be 0. In order to be robust against
297 * bus renumbering issues, a path of PCI device/function numbers may be used
298 * to address the specific device. The path for a device can be determined
299 * through the use of 'lspci -t'.
300 *
301 * The second format matches devices using IDs in the configuration
302 * space which may match multiple devices in the system. A value of 0
303 * for any field will match all devices. (Note: this differs from
304 * in-kernel code that uses PCI_ANY_ID which is ~0; this is for
305 * legacy reasons and convenience so users don't have to specify
306 * FFFFFFFFs on the command line.)
307 *
308 * Returns 1 if the string matches the device, 0 if it does not and
309 * a negative error code if the string cannot be parsed.
310 */
311static int pci_dev_str_match(struct pci_dev *dev, const char *p,
312 const char **endptr)
313{
314 int ret;
315 int count;
316 unsigned short vendor, device, subsystem_vendor, subsystem_device;
317
318 if (strncmp(p, "pci:", 4) == 0) {
319 /* PCI vendor/device (subvendor/subdevice) IDs are specified */
320 p += 4;
321 ret = sscanf(p, "%hx:%hx:%hx:%hx%n", &vendor, &device,
322 &subsystem_vendor, &subsystem_device, &count);
323 if (ret != 4) {
324 ret = sscanf(p, "%hx:%hx%n", &vendor, &device, &count);
325 if (ret != 2)
326 return -EINVAL;
327
328 subsystem_vendor = 0;
329 subsystem_device = 0;
330 }
331
332 p += count;
333
334 if ((!vendor || vendor == dev->vendor) &&
335 (!device || device == dev->device) &&
336 (!subsystem_vendor ||
337 subsystem_vendor == dev->subsystem_vendor) &&
338 (!subsystem_device ||
339 subsystem_device == dev->subsystem_device))
340 goto found;
341 } else {
342 /*
343 * PCI Bus, Device, Function IDs are specified
344 * (optionally, may include a path of devfns following it)
345 */
346 ret = pci_dev_str_match_path(dev, p, &p);
347 if (ret < 0)
348 return ret;
349 else if (ret)
350 goto found;
351 }
352
353 *endptr = p;
354 return 0;
355
356found:
357 *endptr = p;
358 return 1;
359}
360
361static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
362 u8 pos, int cap, int *ttl)
363{
364 u8 id;
365 u16 ent;
366
367 pci_bus_read_config_byte(bus, devfn, pos, &pos);
368
369 while ((*ttl)--) {
370 if (pos < 0x40)
371 break;
372 pos &= ~3;
373 pci_bus_read_config_word(bus, devfn, pos, &ent);
374
375 id = ent & 0xff;
376 if (id == 0xff)
377 break;
378 if (id == cap)
379 return pos;
380 pos = (ent >> 8);
381 }
382 return 0;
383}
384
385static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
386 u8 pos, int cap)
387{
388 int ttl = PCI_FIND_CAP_TTL;
389
390 return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
391}
392
393int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
394{
395 return __pci_find_next_cap(dev->bus, dev->devfn,
396 pos + PCI_CAP_LIST_NEXT, cap);
397}
398EXPORT_SYMBOL_GPL(pci_find_next_capability);
399
400static int __pci_bus_find_cap_start(struct pci_bus *bus,
401 unsigned int devfn, u8 hdr_type)
402{
403 u16 status;
404
405 pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
406 if (!(status & PCI_STATUS_CAP_LIST))
407 return 0;
408
409 switch (hdr_type) {
410 case PCI_HEADER_TYPE_NORMAL:
411 case PCI_HEADER_TYPE_BRIDGE:
412 return PCI_CAPABILITY_LIST;
413 case PCI_HEADER_TYPE_CARDBUS:
414 return PCI_CB_CAPABILITY_LIST;
415 }
416
417 return 0;
418}
419
420/**
421 * pci_find_capability - query for devices' capabilities
422 * @dev: PCI device to query
423 * @cap: capability code
424 *
425 * Tell if a device supports a given PCI capability.
426 * Returns the address of the requested capability structure within the
427 * device's PCI configuration space or 0 in case the device does not
428 * support it. Possible values for @cap include:
429 *
430 * %PCI_CAP_ID_PM Power Management
431 * %PCI_CAP_ID_AGP Accelerated Graphics Port
432 * %PCI_CAP_ID_VPD Vital Product Data
433 * %PCI_CAP_ID_SLOTID Slot Identification
434 * %PCI_CAP_ID_MSI Message Signalled Interrupts
435 * %PCI_CAP_ID_CHSWP CompactPCI HotSwap
436 * %PCI_CAP_ID_PCIX PCI-X
437 * %PCI_CAP_ID_EXP PCI Express
438 */
439int pci_find_capability(struct pci_dev *dev, int cap)
440{
441 int pos;
442
443 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
444 if (pos)
445 pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
446
447 return pos;
448}
449EXPORT_SYMBOL(pci_find_capability);
450
451/**
452 * pci_bus_find_capability - query for devices' capabilities
453 * @bus: the PCI bus to query
454 * @devfn: PCI device to query
455 * @cap: capability code
456 *
457 * Like pci_find_capability() but works for PCI devices that do not have a
458 * pci_dev structure set up yet.
459 *
460 * Returns the address of the requested capability structure within the
461 * device's PCI configuration space or 0 in case the device does not
462 * support it.
463 */
464int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
465{
466 int pos;
467 u8 hdr_type;
468
469 pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
470
471 pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
472 if (pos)
473 pos = __pci_find_next_cap(bus, devfn, pos, cap);
474
475 return pos;
476}
477EXPORT_SYMBOL(pci_bus_find_capability);
478
479/**
480 * pci_find_next_ext_capability - Find an extended capability
481 * @dev: PCI device to query
482 * @start: address at which to start looking (0 to start at beginning of list)
483 * @cap: capability code
484 *
485 * Returns the address of the next matching extended capability structure
486 * within the device's PCI configuration space or 0 if the device does
487 * not support it. Some capabilities can occur several times, e.g., the
488 * vendor-specific capability, and this provides a way to find them all.
489 */
490int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
491{
492 u32 header;
493 int ttl;
494 int pos = PCI_CFG_SPACE_SIZE;
495
496 /* minimum 8 bytes per capability */
497 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
498
499 if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
500 return 0;
501
502 if (start)
503 pos = start;
504
505 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
506 return 0;
507
508 /*
509 * If we have no capabilities, this is indicated by cap ID,
510 * cap version and next pointer all being 0.
511 */
512 if (header == 0)
513 return 0;
514
515 while (ttl-- > 0) {
516 if (PCI_EXT_CAP_ID(header) == cap && pos != start)
517 return pos;
518
519 pos = PCI_EXT_CAP_NEXT(header);
520 if (pos < PCI_CFG_SPACE_SIZE)
521 break;
522
523 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
524 break;
525 }
526
527 return 0;
528}
529EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
530
531/**
532 * pci_find_ext_capability - Find an extended capability
533 * @dev: PCI device to query
534 * @cap: capability code
535 *
536 * Returns the address of the requested extended capability structure
537 * within the device's PCI configuration space or 0 if the device does
538 * not support it. Possible values for @cap include:
539 *
540 * %PCI_EXT_CAP_ID_ERR Advanced Error Reporting
541 * %PCI_EXT_CAP_ID_VC Virtual Channel
542 * %PCI_EXT_CAP_ID_DSN Device Serial Number
543 * %PCI_EXT_CAP_ID_PWR Power Budgeting
544 */
545int pci_find_ext_capability(struct pci_dev *dev, int cap)
546{
547 return pci_find_next_ext_capability(dev, 0, cap);
548}
549EXPORT_SYMBOL_GPL(pci_find_ext_capability);
550
551static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
552{
553 int rc, ttl = PCI_FIND_CAP_TTL;
554 u8 cap, mask;
555
556 if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
557 mask = HT_3BIT_CAP_MASK;
558 else
559 mask = HT_5BIT_CAP_MASK;
560
561 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
562 PCI_CAP_ID_HT, &ttl);
563 while (pos) {
564 rc = pci_read_config_byte(dev, pos + 3, &cap);
565 if (rc != PCIBIOS_SUCCESSFUL)
566 return 0;
567
568 if ((cap & mask) == ht_cap)
569 return pos;
570
571 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
572 pos + PCI_CAP_LIST_NEXT,
573 PCI_CAP_ID_HT, &ttl);
574 }
575
576 return 0;
577}
578/**
579 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
580 * @dev: PCI device to query
581 * @pos: Position from which to continue searching
582 * @ht_cap: Hypertransport capability code
583 *
584 * To be used in conjunction with pci_find_ht_capability() to search for
585 * all capabilities matching @ht_cap. @pos should always be a value returned
586 * from pci_find_ht_capability().
587 *
588 * NB. To be 100% safe against broken PCI devices, the caller should take
589 * steps to avoid an infinite loop.
590 */
591int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
592{
593 return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
594}
595EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
596
597/**
598 * pci_find_ht_capability - query a device's Hypertransport capabilities
599 * @dev: PCI device to query
600 * @ht_cap: Hypertransport capability code
601 *
602 * Tell if a device supports a given Hypertransport capability.
603 * Returns an address within the device's PCI configuration space
604 * or 0 in case the device does not support the request capability.
605 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
606 * which has a Hypertransport capability matching @ht_cap.
607 */
608int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
609{
610 int pos;
611
612 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
613 if (pos)
614 pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
615
616 return pos;
617}
618EXPORT_SYMBOL_GPL(pci_find_ht_capability);
619
620/**
621 * pci_find_parent_resource - return resource region of parent bus of given
622 * region
623 * @dev: PCI device structure contains resources to be searched
624 * @res: child resource record for which parent is sought
625 *
626 * For given resource region of given device, return the resource region of
627 * parent bus the given region is contained in.
628 */
629struct resource *pci_find_parent_resource(const struct pci_dev *dev,
630 struct resource *res)
631{
632 const struct pci_bus *bus = dev->bus;
633 struct resource *r;
634 int i;
635
636 pci_bus_for_each_resource(bus, r, i) {
637 if (!r)
638 continue;
639 if (resource_contains(r, res)) {
640
641 /*
642 * If the window is prefetchable but the BAR is
643 * not, the allocator made a mistake.
644 */
645 if (r->flags & IORESOURCE_PREFETCH &&
646 !(res->flags & IORESOURCE_PREFETCH))
647 return NULL;
648
649 /*
650 * If we're below a transparent bridge, there may
651 * be both a positively-decoded aperture and a
652 * subtractively-decoded region that contain the BAR.
653 * We want the positively-decoded one, so this depends
654 * on pci_bus_for_each_resource() giving us those
655 * first.
656 */
657 return r;
658 }
659 }
660 return NULL;
661}
662EXPORT_SYMBOL(pci_find_parent_resource);
663
664/**
665 * pci_find_resource - Return matching PCI device resource
666 * @dev: PCI device to query
667 * @res: Resource to look for
668 *
669 * Goes over standard PCI resources (BARs) and checks if the given resource
670 * is partially or fully contained in any of them. In that case the
671 * matching resource is returned, %NULL otherwise.
672 */
673struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
674{
675 int i;
676
677 for (i = 0; i < PCI_ROM_RESOURCE; i++) {
678 struct resource *r = &dev->resource[i];
679
680 if (r->start && resource_contains(r, res))
681 return r;
682 }
683
684 return NULL;
685}
686EXPORT_SYMBOL(pci_find_resource);
687
688/**
689 * pci_find_pcie_root_port - return PCIe Root Port
690 * @dev: PCI device to query
691 *
692 * Traverse up the parent chain and return the PCIe Root Port PCI Device
693 * for a given PCI Device.
694 */
695struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
696{
697 struct pci_dev *bridge, *highest_pcie_bridge = dev;
698
699 bridge = pci_upstream_bridge(dev);
700 while (bridge && pci_is_pcie(bridge)) {
701 highest_pcie_bridge = bridge;
702 bridge = pci_upstream_bridge(bridge);
703 }
704
705 if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
706 return NULL;
707
708 return highest_pcie_bridge;
709}
710EXPORT_SYMBOL(pci_find_pcie_root_port);
711
712/**
713 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
714 * @dev: the PCI device to operate on
715 * @pos: config space offset of status word
716 * @mask: mask of bit(s) to care about in status word
717 *
718 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
719 */
720int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
721{
722 int i;
723
724 /* Wait for Transaction Pending bit clean */
725 for (i = 0; i < 4; i++) {
726 u16 status;
727 if (i)
728 msleep((1 << (i - 1)) * 100);
729
730 pci_read_config_word(dev, pos, &status);
731 if (!(status & mask))
732 return 1;
733 }
734
735 return 0;
736}
737
738/**
739 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
740 * @dev: PCI device to have its BARs restored
741 *
742 * Restore the BAR values for a given device, so as to make it
743 * accessible by its driver.
744 */
745static void pci_restore_bars(struct pci_dev *dev)
746{
747 int i;
748
749 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
750 pci_update_resource(dev, i);
751}
752
753static const struct pci_platform_pm_ops *pci_platform_pm;
754
755int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
756{
757 if (!ops->is_manageable || !ops->set_state || !ops->get_state ||
758 !ops->choose_state || !ops->set_wakeup || !ops->need_resume)
759 return -EINVAL;
760 pci_platform_pm = ops;
761 return 0;
762}
763
764static inline bool platform_pci_power_manageable(struct pci_dev *dev)
765{
766 return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
767}
768
769static inline int platform_pci_set_power_state(struct pci_dev *dev,
770 pci_power_t t)
771{
772 return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
773}
774
775static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
776{
777 return pci_platform_pm ? pci_platform_pm->get_state(dev) : PCI_UNKNOWN;
778}
779
780static inline void platform_pci_refresh_power_state(struct pci_dev *dev)
781{
782 if (pci_platform_pm && pci_platform_pm->refresh_state)
783 pci_platform_pm->refresh_state(dev);
784}
785
786static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
787{
788 return pci_platform_pm ?
789 pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
790}
791
792static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
793{
794 return pci_platform_pm ?
795 pci_platform_pm->set_wakeup(dev, enable) : -ENODEV;
796}
797
798static inline bool platform_pci_need_resume(struct pci_dev *dev)
799{
800 return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
801}
802
803static inline bool platform_pci_bridge_d3(struct pci_dev *dev)
804{
805 return pci_platform_pm ? pci_platform_pm->bridge_d3(dev) : false;
806}
807
808/**
809 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
810 * given PCI device
811 * @dev: PCI device to handle.
812 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
813 *
814 * RETURN VALUE:
815 * -EINVAL if the requested state is invalid.
816 * -EIO if device does not support PCI PM or its PM capabilities register has a
817 * wrong version, or device doesn't support the requested state.
818 * 0 if device already is in the requested state.
819 * 0 if device's power state has been successfully changed.
820 */
821static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
822{
823 u16 pmcsr;
824 bool need_restore = false;
825
826 /* Check if we're already there */
827 if (dev->current_state == state)
828 return 0;
829
830 if (!dev->pm_cap)
831 return -EIO;
832
833 if (state < PCI_D0 || state > PCI_D3hot)
834 return -EINVAL;
835
836 /*
837 * Validate current state:
838 * Can enter D0 from any state, but if we can only go deeper
839 * to sleep if we're already in a low power state
840 */
841 if (state != PCI_D0 && dev->current_state <= PCI_D3cold
842 && dev->current_state > state) {
843 pci_err(dev, "invalid power transition (from state %d to %d)\n",
844 dev->current_state, state);
845 return -EINVAL;
846 }
847
848 /* Check if this device supports the desired state */
849 if ((state == PCI_D1 && !dev->d1_support)
850 || (state == PCI_D2 && !dev->d2_support))
851 return -EIO;
852
853 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
854
855 /*
856 * If we're (effectively) in D3, force entire word to 0.
857 * This doesn't affect PME_Status, disables PME_En, and
858 * sets PowerState to 0.
859 */
860 switch (dev->current_state) {
861 case PCI_D0:
862 case PCI_D1:
863 case PCI_D2:
864 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
865 pmcsr |= state;
866 break;
867 case PCI_D3hot:
868 case PCI_D3cold:
869 case PCI_UNKNOWN: /* Boot-up */
870 if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
871 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
872 need_restore = true;
873 /* Fall-through - force to D0 */
874 default:
875 pmcsr = 0;
876 break;
877 }
878
879 /* Enter specified state */
880 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
881
882 /*
883 * Mandatory power management transition delays; see PCI PM 1.1
884 * 5.6.1 table 18
885 */
886 if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
887 pci_dev_d3_sleep(dev);
888 else if (state == PCI_D2 || dev->current_state == PCI_D2)
889 udelay(PCI_PM_D2_DELAY);
890
891 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
892 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
893 if (dev->current_state != state)
894 pci_info_ratelimited(dev, "Refused to change power state, currently in D%d\n",
895 dev->current_state);
896
897 /*
898 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
899 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
900 * from D3hot to D0 _may_ perform an internal reset, thereby
901 * going to "D0 Uninitialized" rather than "D0 Initialized".
902 * For example, at least some versions of the 3c905B and the
903 * 3c556B exhibit this behaviour.
904 *
905 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
906 * devices in a D3hot state at boot. Consequently, we need to
907 * restore at least the BARs so that the device will be
908 * accessible to its driver.
909 */
910 if (need_restore)
911 pci_restore_bars(dev);
912
913 if (dev->bus->self)
914 pcie_aspm_pm_state_change(dev->bus->self);
915
916 return 0;
917}
918
919/**
920 * pci_update_current_state - Read power state of given device and cache it
921 * @dev: PCI device to handle.
922 * @state: State to cache in case the device doesn't have the PM capability
923 *
924 * The power state is read from the PMCSR register, which however is
925 * inaccessible in D3cold. The platform firmware is therefore queried first
926 * to detect accessibility of the register. In case the platform firmware
927 * reports an incorrect state or the device isn't power manageable by the
928 * platform at all, we try to detect D3cold by testing accessibility of the
929 * vendor ID in config space.
930 */
931void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
932{
933 if (platform_pci_get_power_state(dev) == PCI_D3cold ||
934 !pci_device_is_present(dev)) {
935 dev->current_state = PCI_D3cold;
936 } else if (dev->pm_cap) {
937 u16 pmcsr;
938
939 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
940 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
941 } else {
942 dev->current_state = state;
943 }
944}
945
946/**
947 * pci_refresh_power_state - Refresh the given device's power state data
948 * @dev: Target PCI device.
949 *
950 * Ask the platform to refresh the devices power state information and invoke
951 * pci_update_current_state() to update its current PCI power state.
952 */
953void pci_refresh_power_state(struct pci_dev *dev)
954{
955 if (platform_pci_power_manageable(dev))
956 platform_pci_refresh_power_state(dev);
957
958 pci_update_current_state(dev, dev->current_state);
959}
960
961/**
962 * pci_platform_power_transition - Use platform to change device power state
963 * @dev: PCI device to handle.
964 * @state: State to put the device into.
965 */
966static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
967{
968 int error;
969
970 if (platform_pci_power_manageable(dev)) {
971 error = platform_pci_set_power_state(dev, state);
972 if (!error)
973 pci_update_current_state(dev, state);
974 } else
975 error = -ENODEV;
976
977 if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
978 dev->current_state = PCI_D0;
979
980 return error;
981}
982
983/**
984 * pci_wakeup - Wake up a PCI device
985 * @pci_dev: Device to handle.
986 * @ign: ignored parameter
987 */
988static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
989{
990 pci_wakeup_event(pci_dev);
991 pm_request_resume(&pci_dev->dev);
992 return 0;
993}
994
995/**
996 * pci_wakeup_bus - Walk given bus and wake up devices on it
997 * @bus: Top bus of the subtree to walk.
998 */
999void pci_wakeup_bus(struct pci_bus *bus)
1000{
1001 if (bus)
1002 pci_walk_bus(bus, pci_wakeup, NULL);
1003}
1004
1005/**
1006 * __pci_start_power_transition - Start power transition of a PCI device
1007 * @dev: PCI device to handle.
1008 * @state: State to put the device into.
1009 */
1010static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
1011{
1012 if (state == PCI_D0) {
1013 pci_platform_power_transition(dev, PCI_D0);
1014 /*
1015 * Mandatory power management transition delays, see
1016 * PCI Express Base Specification Revision 2.0 Section
1017 * 6.6.1: Conventional Reset. Do not delay for
1018 * devices powered on/off by corresponding bridge,
1019 * because have already delayed for the bridge.
1020 */
1021 if (dev->runtime_d3cold) {
1022 if (dev->d3cold_delay && !dev->imm_ready)
1023 msleep(dev->d3cold_delay);
1024 /*
1025 * When powering on a bridge from D3cold, the
1026 * whole hierarchy may be powered on into
1027 * D0uninitialized state, resume them to give
1028 * them a chance to suspend again
1029 */
1030 pci_wakeup_bus(dev->subordinate);
1031 }
1032 }
1033}
1034
1035/**
1036 * __pci_dev_set_current_state - Set current state of a PCI device
1037 * @dev: Device to handle
1038 * @data: pointer to state to be set
1039 */
1040static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
1041{
1042 pci_power_t state = *(pci_power_t *)data;
1043
1044 dev->current_state = state;
1045 return 0;
1046}
1047
1048/**
1049 * pci_bus_set_current_state - Walk given bus and set current state of devices
1050 * @bus: Top bus of the subtree to walk.
1051 * @state: state to be set
1052 */
1053void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
1054{
1055 if (bus)
1056 pci_walk_bus(bus, __pci_dev_set_current_state, &state);
1057}
1058
1059/**
1060 * __pci_complete_power_transition - Complete power transition of a PCI device
1061 * @dev: PCI device to handle.
1062 * @state: State to put the device into.
1063 *
1064 * This function should not be called directly by device drivers.
1065 */
1066int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
1067{
1068 int ret;
1069
1070 if (state <= PCI_D0)
1071 return -EINVAL;
1072 ret = pci_platform_power_transition(dev, state);
1073 /* Power off the bridge may power off the whole hierarchy */
1074 if (!ret && state == PCI_D3cold)
1075 pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
1076 return ret;
1077}
1078EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
1079
1080/**
1081 * pci_set_power_state - Set the power state of a PCI device
1082 * @dev: PCI device to handle.
1083 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
1084 *
1085 * Transition a device to a new power state, using the platform firmware and/or
1086 * the device's PCI PM registers.
1087 *
1088 * RETURN VALUE:
1089 * -EINVAL if the requested state is invalid.
1090 * -EIO if device does not support PCI PM or its PM capabilities register has a
1091 * wrong version, or device doesn't support the requested state.
1092 * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
1093 * 0 if device already is in the requested state.
1094 * 0 if the transition is to D3 but D3 is not supported.
1095 * 0 if device's power state has been successfully changed.
1096 */
1097int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
1098{
1099 int error;
1100
1101 /* Bound the state we're entering */
1102 if (state > PCI_D3cold)
1103 state = PCI_D3cold;
1104 else if (state < PCI_D0)
1105 state = PCI_D0;
1106 else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
1107
1108 /*
1109 * If the device or the parent bridge do not support PCI
1110 * PM, ignore the request if we're doing anything other
1111 * than putting it into D0 (which would only happen on
1112 * boot).
1113 */
1114 return 0;
1115
1116 /* Check if we're already there */
1117 if (dev->current_state == state)
1118 return 0;
1119
1120 __pci_start_power_transition(dev, state);
1121
1122 /*
1123 * This device is quirked not to be put into D3, so don't put it in
1124 * D3
1125 */
1126 if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
1127 return 0;
1128
1129 /*
1130 * To put device in D3cold, we put device into D3hot in native
1131 * way, then put device into D3cold with platform ops
1132 */
1133 error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
1134 PCI_D3hot : state);
1135
1136 if (!__pci_complete_power_transition(dev, state))
1137 error = 0;
1138
1139 return error;
1140}
1141EXPORT_SYMBOL(pci_set_power_state);
1142
1143/**
1144 * pci_power_up - Put the given device into D0 forcibly
1145 * @dev: PCI device to power up
1146 */
1147void pci_power_up(struct pci_dev *dev)
1148{
1149 __pci_start_power_transition(dev, PCI_D0);
1150 pci_raw_set_power_state(dev, PCI_D0);
1151 pci_update_current_state(dev, PCI_D0);
1152}
1153
1154/**
1155 * pci_choose_state - Choose the power state of a PCI device
1156 * @dev: PCI device to be suspended
1157 * @state: target sleep state for the whole system. This is the value
1158 * that is passed to suspend() function.
1159 *
1160 * Returns PCI power state suitable for given device and given system
1161 * message.
1162 */
1163pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
1164{
1165 pci_power_t ret;
1166
1167 if (!dev->pm_cap)
1168 return PCI_D0;
1169
1170 ret = platform_pci_choose_state(dev);
1171 if (ret != PCI_POWER_ERROR)
1172 return ret;
1173
1174 switch (state.event) {
1175 case PM_EVENT_ON:
1176 return PCI_D0;
1177 case PM_EVENT_FREEZE:
1178 case PM_EVENT_PRETHAW:
1179 /* REVISIT both freeze and pre-thaw "should" use D0 */
1180 case PM_EVENT_SUSPEND:
1181 case PM_EVENT_HIBERNATE:
1182 return PCI_D3hot;
1183 default:
1184 pci_info(dev, "unrecognized suspend event %d\n",
1185 state.event);
1186 BUG();
1187 }
1188 return PCI_D0;
1189}
1190EXPORT_SYMBOL(pci_choose_state);
1191
1192#define PCI_EXP_SAVE_REGS 7
1193
1194static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
1195 u16 cap, bool extended)
1196{
1197 struct pci_cap_saved_state *tmp;
1198
1199 hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
1200 if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
1201 return tmp;
1202 }
1203 return NULL;
1204}
1205
1206struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
1207{
1208 return _pci_find_saved_cap(dev, cap, false);
1209}
1210
1211struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
1212{
1213 return _pci_find_saved_cap(dev, cap, true);
1214}
1215
1216static int pci_save_pcie_state(struct pci_dev *dev)
1217{
1218 int i = 0;
1219 struct pci_cap_saved_state *save_state;
1220 u16 *cap;
1221
1222 if (!pci_is_pcie(dev))
1223 return 0;
1224
1225 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1226 if (!save_state) {
1227 pci_err(dev, "buffer not found in %s\n", __func__);
1228 return -ENOMEM;
1229 }
1230
1231 cap = (u16 *)&save_state->cap.data[0];
1232 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1233 pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1234 pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1235 pcie_capability_read_word(dev, PCI_EXP_RTCTL, &cap[i++]);
1236 pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1237 pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1238 pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1239
1240 return 0;
1241}
1242
1243static void pci_restore_pcie_state(struct pci_dev *dev)
1244{
1245 int i = 0;
1246 struct pci_cap_saved_state *save_state;
1247 u16 *cap;
1248
1249 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1250 if (!save_state)
1251 return;
1252
1253 cap = (u16 *)&save_state->cap.data[0];
1254 pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1255 pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1256 pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1257 pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1258 pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1259 pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1260 pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1261}
1262
1263static int pci_save_pcix_state(struct pci_dev *dev)
1264{
1265 int pos;
1266 struct pci_cap_saved_state *save_state;
1267
1268 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1269 if (!pos)
1270 return 0;
1271
1272 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1273 if (!save_state) {
1274 pci_err(dev, "buffer not found in %s\n", __func__);
1275 return -ENOMEM;
1276 }
1277
1278 pci_read_config_word(dev, pos + PCI_X_CMD,
1279 (u16 *)save_state->cap.data);
1280
1281 return 0;
1282}
1283
1284static void pci_restore_pcix_state(struct pci_dev *dev)
1285{
1286 int i = 0, pos;
1287 struct pci_cap_saved_state *save_state;
1288 u16 *cap;
1289
1290 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1291 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1292 if (!save_state || !pos)
1293 return;
1294 cap = (u16 *)&save_state->cap.data[0];
1295
1296 pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1297}
1298
1299static void pci_save_ltr_state(struct pci_dev *dev)
1300{
1301 int ltr;
1302 struct pci_cap_saved_state *save_state;
1303 u16 *cap;
1304
1305 if (!pci_is_pcie(dev))
1306 return;
1307
1308 ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1309 if (!ltr)
1310 return;
1311
1312 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1313 if (!save_state) {
1314 pci_err(dev, "no suspend buffer for LTR; ASPM issues possible after resume\n");
1315 return;
1316 }
1317
1318 cap = (u16 *)&save_state->cap.data[0];
1319 pci_read_config_word(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, cap++);
1320 pci_read_config_word(dev, ltr + PCI_LTR_MAX_NOSNOOP_LAT, cap++);
1321}
1322
1323static void pci_restore_ltr_state(struct pci_dev *dev)
1324{
1325 struct pci_cap_saved_state *save_state;
1326 int ltr;
1327 u16 *cap;
1328
1329 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1330 ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1331 if (!save_state || !ltr)
1332 return;
1333
1334 cap = (u16 *)&save_state->cap.data[0];
1335 pci_write_config_word(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, *cap++);
1336 pci_write_config_word(dev, ltr + PCI_LTR_MAX_NOSNOOP_LAT, *cap++);
1337}
1338
1339/**
1340 * pci_save_state - save the PCI configuration space of a device before
1341 * suspending
1342 * @dev: PCI device that we're dealing with
1343 */
1344int pci_save_state(struct pci_dev *dev)
1345{
1346 int i;
1347 /* XXX: 100% dword access ok here? */
1348 for (i = 0; i < 16; i++)
1349 pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1350 dev->state_saved = true;
1351
1352 i = pci_save_pcie_state(dev);
1353 if (i != 0)
1354 return i;
1355
1356 i = pci_save_pcix_state(dev);
1357 if (i != 0)
1358 return i;
1359
1360 pci_save_ltr_state(dev);
1361 pci_save_dpc_state(dev);
1362 return pci_save_vc_state(dev);
1363}
1364EXPORT_SYMBOL(pci_save_state);
1365
1366static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1367 u32 saved_val, int retry, bool force)
1368{
1369 u32 val;
1370
1371 pci_read_config_dword(pdev, offset, &val);
1372 if (!force && val == saved_val)
1373 return;
1374
1375 for (;;) {
1376 pci_dbg(pdev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1377 offset, val, saved_val);
1378 pci_write_config_dword(pdev, offset, saved_val);
1379 if (retry-- <= 0)
1380 return;
1381
1382 pci_read_config_dword(pdev, offset, &val);
1383 if (val == saved_val)
1384 return;
1385
1386 mdelay(1);
1387 }
1388}
1389
1390static void pci_restore_config_space_range(struct pci_dev *pdev,
1391 int start, int end, int retry,
1392 bool force)
1393{
1394 int index;
1395
1396 for (index = end; index >= start; index--)
1397 pci_restore_config_dword(pdev, 4 * index,
1398 pdev->saved_config_space[index],
1399 retry, force);
1400}
1401
1402static void pci_restore_config_space(struct pci_dev *pdev)
1403{
1404 if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1405 pci_restore_config_space_range(pdev, 10, 15, 0, false);
1406 /* Restore BARs before the command register. */
1407 pci_restore_config_space_range(pdev, 4, 9, 10, false);
1408 pci_restore_config_space_range(pdev, 0, 3, 0, false);
1409 } else if (pdev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1410 pci_restore_config_space_range(pdev, 12, 15, 0, false);
1411
1412 /*
1413 * Force rewriting of prefetch registers to avoid S3 resume
1414 * issues on Intel PCI bridges that occur when these
1415 * registers are not explicitly written.
1416 */
1417 pci_restore_config_space_range(pdev, 9, 11, 0, true);
1418 pci_restore_config_space_range(pdev, 0, 8, 0, false);
1419 } else {
1420 pci_restore_config_space_range(pdev, 0, 15, 0, false);
1421 }
1422}
1423
1424static void pci_restore_rebar_state(struct pci_dev *pdev)
1425{
1426 unsigned int pos, nbars, i;
1427 u32 ctrl;
1428
1429 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
1430 if (!pos)
1431 return;
1432
1433 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1434 nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
1435 PCI_REBAR_CTRL_NBAR_SHIFT;
1436
1437 for (i = 0; i < nbars; i++, pos += 8) {
1438 struct resource *res;
1439 int bar_idx, size;
1440
1441 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1442 bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
1443 res = pdev->resource + bar_idx;
1444 size = ilog2(resource_size(res)) - 20;
1445 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
1446 ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
1447 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
1448 }
1449}
1450
1451/**
1452 * pci_restore_state - Restore the saved state of a PCI device
1453 * @dev: PCI device that we're dealing with
1454 */
1455void pci_restore_state(struct pci_dev *dev)
1456{
1457 if (!dev->state_saved)
1458 return;
1459
1460 /*
1461 * Restore max latencies (in the LTR capability) before enabling
1462 * LTR itself (in the PCIe capability).
1463 */
1464 pci_restore_ltr_state(dev);
1465
1466 pci_restore_pcie_state(dev);
1467 pci_restore_pasid_state(dev);
1468 pci_restore_pri_state(dev);
1469 pci_restore_ats_state(dev);
1470 pci_restore_vc_state(dev);
1471 pci_restore_rebar_state(dev);
1472 pci_restore_dpc_state(dev);
1473
1474 pci_cleanup_aer_error_status_regs(dev);
1475
1476 pci_restore_config_space(dev);
1477
1478 pci_restore_pcix_state(dev);
1479 pci_restore_msi_state(dev);
1480
1481 /* Restore ACS and IOV configuration state */
1482 pci_enable_acs(dev);
1483 pci_restore_iov_state(dev);
1484
1485 dev->state_saved = false;
1486}
1487EXPORT_SYMBOL(pci_restore_state);
1488
1489struct pci_saved_state {
1490 u32 config_space[16];
1491 struct pci_cap_saved_data cap[0];
1492};
1493
1494/**
1495 * pci_store_saved_state - Allocate and return an opaque struct containing
1496 * the device saved state.
1497 * @dev: PCI device that we're dealing with
1498 *
1499 * Return NULL if no state or error.
1500 */
1501struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1502{
1503 struct pci_saved_state *state;
1504 struct pci_cap_saved_state *tmp;
1505 struct pci_cap_saved_data *cap;
1506 size_t size;
1507
1508 if (!dev->state_saved)
1509 return NULL;
1510
1511 size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1512
1513 hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1514 size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1515
1516 state = kzalloc(size, GFP_KERNEL);
1517 if (!state)
1518 return NULL;
1519
1520 memcpy(state->config_space, dev->saved_config_space,
1521 sizeof(state->config_space));
1522
1523 cap = state->cap;
1524 hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1525 size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1526 memcpy(cap, &tmp->cap, len);
1527 cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1528 }
1529 /* Empty cap_save terminates list */
1530
1531 return state;
1532}
1533EXPORT_SYMBOL_GPL(pci_store_saved_state);
1534
1535/**
1536 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1537 * @dev: PCI device that we're dealing with
1538 * @state: Saved state returned from pci_store_saved_state()
1539 */
1540int pci_load_saved_state(struct pci_dev *dev,
1541 struct pci_saved_state *state)
1542{
1543 struct pci_cap_saved_data *cap;
1544
1545 dev->state_saved = false;
1546
1547 if (!state)
1548 return 0;
1549
1550 memcpy(dev->saved_config_space, state->config_space,
1551 sizeof(state->config_space));
1552
1553 cap = state->cap;
1554 while (cap->size) {
1555 struct pci_cap_saved_state *tmp;
1556
1557 tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1558 if (!tmp || tmp->cap.size != cap->size)
1559 return -EINVAL;
1560
1561 memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1562 cap = (struct pci_cap_saved_data *)((u8 *)cap +
1563 sizeof(struct pci_cap_saved_data) + cap->size);
1564 }
1565
1566 dev->state_saved = true;
1567 return 0;
1568}
1569EXPORT_SYMBOL_GPL(pci_load_saved_state);
1570
1571/**
1572 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1573 * and free the memory allocated for it.
1574 * @dev: PCI device that we're dealing with
1575 * @state: Pointer to saved state returned from pci_store_saved_state()
1576 */
1577int pci_load_and_free_saved_state(struct pci_dev *dev,
1578 struct pci_saved_state **state)
1579{
1580 int ret = pci_load_saved_state(dev, *state);
1581 kfree(*state);
1582 *state = NULL;
1583 return ret;
1584}
1585EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1586
1587int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1588{
1589 return pci_enable_resources(dev, bars);
1590}
1591
1592static int do_pci_enable_device(struct pci_dev *dev, int bars)
1593{
1594 int err;
1595 struct pci_dev *bridge;
1596 u16 cmd;
1597 u8 pin;
1598
1599 err = pci_set_power_state(dev, PCI_D0);
1600 if (err < 0 && err != -EIO)
1601 return err;
1602
1603 bridge = pci_upstream_bridge(dev);
1604 if (bridge)
1605 pcie_aspm_powersave_config_link(bridge);
1606
1607 err = pcibios_enable_device(dev, bars);
1608 if (err < 0)
1609 return err;
1610 pci_fixup_device(pci_fixup_enable, dev);
1611
1612 if (dev->msi_enabled || dev->msix_enabled)
1613 return 0;
1614
1615 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1616 if (pin) {
1617 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1618 if (cmd & PCI_COMMAND_INTX_DISABLE)
1619 pci_write_config_word(dev, PCI_COMMAND,
1620 cmd & ~PCI_COMMAND_INTX_DISABLE);
1621 }
1622
1623 return 0;
1624}
1625
1626/**
1627 * pci_reenable_device - Resume abandoned device
1628 * @dev: PCI device to be resumed
1629 *
1630 * NOTE: This function is a backend of pci_default_resume() and is not supposed
1631 * to be called by normal code, write proper resume handler and use it instead.
1632 */
1633int pci_reenable_device(struct pci_dev *dev)
1634{
1635 if (pci_is_enabled(dev))
1636 return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1637 return 0;
1638}
1639EXPORT_SYMBOL(pci_reenable_device);
1640
1641static void pci_enable_bridge(struct pci_dev *dev)
1642{
1643 struct pci_dev *bridge;
1644 int retval;
1645
1646 bridge = pci_upstream_bridge(dev);
1647 if (bridge)
1648 pci_enable_bridge(bridge);
1649
1650 if (pci_is_enabled(dev)) {
1651 if (!dev->is_busmaster)
1652 pci_set_master(dev);
1653 return;
1654 }
1655
1656 retval = pci_enable_device(dev);
1657 if (retval)
1658 pci_err(dev, "Error enabling bridge (%d), continuing\n",
1659 retval);
1660 pci_set_master(dev);
1661}
1662
1663static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1664{
1665 struct pci_dev *bridge;
1666 int err;
1667 int i, bars = 0;
1668
1669 /*
1670 * Power state could be unknown at this point, either due to a fresh
1671 * boot or a device removal call. So get the current power state
1672 * so that things like MSI message writing will behave as expected
1673 * (e.g. if the device really is in D0 at enable time).
1674 */
1675 if (dev->pm_cap) {
1676 u16 pmcsr;
1677 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1678 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1679 }
1680
1681 if (atomic_inc_return(&dev->enable_cnt) > 1)
1682 return 0; /* already enabled */
1683
1684 bridge = pci_upstream_bridge(dev);
1685 if (bridge)
1686 pci_enable_bridge(bridge);
1687
1688 /* only skip sriov related */
1689 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1690 if (dev->resource[i].flags & flags)
1691 bars |= (1 << i);
1692 for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1693 if (dev->resource[i].flags & flags)
1694 bars |= (1 << i);
1695
1696 err = do_pci_enable_device(dev, bars);
1697 if (err < 0)
1698 atomic_dec(&dev->enable_cnt);
1699 return err;
1700}
1701
1702/**
1703 * pci_enable_device_io - Initialize a device for use with IO space
1704 * @dev: PCI device to be initialized
1705 *
1706 * Initialize device before it's used by a driver. Ask low-level code
1707 * to enable I/O resources. Wake up the device if it was suspended.
1708 * Beware, this function can fail.
1709 */
1710int pci_enable_device_io(struct pci_dev *dev)
1711{
1712 return pci_enable_device_flags(dev, IORESOURCE_IO);
1713}
1714EXPORT_SYMBOL(pci_enable_device_io);
1715
1716/**
1717 * pci_enable_device_mem - Initialize a device for use with Memory space
1718 * @dev: PCI device to be initialized
1719 *
1720 * Initialize device before it's used by a driver. Ask low-level code
1721 * to enable Memory resources. Wake up the device if it was suspended.
1722 * Beware, this function can fail.
1723 */
1724int pci_enable_device_mem(struct pci_dev *dev)
1725{
1726 return pci_enable_device_flags(dev, IORESOURCE_MEM);
1727}
1728EXPORT_SYMBOL(pci_enable_device_mem);
1729
1730/**
1731 * pci_enable_device - Initialize device before it's used by a driver.
1732 * @dev: PCI device to be initialized
1733 *
1734 * Initialize device before it's used by a driver. Ask low-level code
1735 * to enable I/O and memory. Wake up the device if it was suspended.
1736 * Beware, this function can fail.
1737 *
1738 * Note we don't actually enable the device many times if we call
1739 * this function repeatedly (we just increment the count).
1740 */
1741int pci_enable_device(struct pci_dev *dev)
1742{
1743 return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1744}
1745EXPORT_SYMBOL(pci_enable_device);
1746
1747/*
1748 * Managed PCI resources. This manages device on/off, INTx/MSI/MSI-X
1749 * on/off and BAR regions. pci_dev itself records MSI/MSI-X status, so
1750 * there's no need to track it separately. pci_devres is initialized
1751 * when a device is enabled using managed PCI device enable interface.
1752 */
1753struct pci_devres {
1754 unsigned int enabled:1;
1755 unsigned int pinned:1;
1756 unsigned int orig_intx:1;
1757 unsigned int restore_intx:1;
1758 unsigned int mwi:1;
1759 u32 region_mask;
1760};
1761
1762static void pcim_release(struct device *gendev, void *res)
1763{
1764 struct pci_dev *dev = to_pci_dev(gendev);
1765 struct pci_devres *this = res;
1766 int i;
1767
1768 if (dev->msi_enabled)
1769 pci_disable_msi(dev);
1770 if (dev->msix_enabled)
1771 pci_disable_msix(dev);
1772
1773 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1774 if (this->region_mask & (1 << i))
1775 pci_release_region(dev, i);
1776
1777 if (this->mwi)
1778 pci_clear_mwi(dev);
1779
1780 if (this->restore_intx)
1781 pci_intx(dev, this->orig_intx);
1782
1783 if (this->enabled && !this->pinned)
1784 pci_disable_device(dev);
1785}
1786
1787static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1788{
1789 struct pci_devres *dr, *new_dr;
1790
1791 dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1792 if (dr)
1793 return dr;
1794
1795 new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1796 if (!new_dr)
1797 return NULL;
1798 return devres_get(&pdev->dev, new_dr, NULL, NULL);
1799}
1800
1801static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1802{
1803 if (pci_is_managed(pdev))
1804 return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1805 return NULL;
1806}
1807
1808/**
1809 * pcim_enable_device - Managed pci_enable_device()
1810 * @pdev: PCI device to be initialized
1811 *
1812 * Managed pci_enable_device().
1813 */
1814int pcim_enable_device(struct pci_dev *pdev)
1815{
1816 struct pci_devres *dr;
1817 int rc;
1818
1819 dr = get_pci_dr(pdev);
1820 if (unlikely(!dr))
1821 return -ENOMEM;
1822 if (dr->enabled)
1823 return 0;
1824
1825 rc = pci_enable_device(pdev);
1826 if (!rc) {
1827 pdev->is_managed = 1;
1828 dr->enabled = 1;
1829 }
1830 return rc;
1831}
1832EXPORT_SYMBOL(pcim_enable_device);
1833
1834/**
1835 * pcim_pin_device - Pin managed PCI device
1836 * @pdev: PCI device to pin
1837 *
1838 * Pin managed PCI device @pdev. Pinned device won't be disabled on
1839 * driver detach. @pdev must have been enabled with
1840 * pcim_enable_device().
1841 */
1842void pcim_pin_device(struct pci_dev *pdev)
1843{
1844 struct pci_devres *dr;
1845
1846 dr = find_pci_dr(pdev);
1847 WARN_ON(!dr || !dr->enabled);
1848 if (dr)
1849 dr->pinned = 1;
1850}
1851EXPORT_SYMBOL(pcim_pin_device);
1852
1853/*
1854 * pcibios_add_device - provide arch specific hooks when adding device dev
1855 * @dev: the PCI device being added
1856 *
1857 * Permits the platform to provide architecture specific functionality when
1858 * devices are added. This is the default implementation. Architecture
1859 * implementations can override this.
1860 */
1861int __weak pcibios_add_device(struct pci_dev *dev)
1862{
1863 return 0;
1864}
1865
1866/**
1867 * pcibios_release_device - provide arch specific hooks when releasing
1868 * device dev
1869 * @dev: the PCI device being released
1870 *
1871 * Permits the platform to provide architecture specific functionality when
1872 * devices are released. This is the default implementation. Architecture
1873 * implementations can override this.
1874 */
1875void __weak pcibios_release_device(struct pci_dev *dev) {}
1876
1877/**
1878 * pcibios_disable_device - disable arch specific PCI resources for device dev
1879 * @dev: the PCI device to disable
1880 *
1881 * Disables architecture specific PCI resources for the device. This
1882 * is the default implementation. Architecture implementations can
1883 * override this.
1884 */
1885void __weak pcibios_disable_device(struct pci_dev *dev) {}
1886
1887/**
1888 * pcibios_penalize_isa_irq - penalize an ISA IRQ
1889 * @irq: ISA IRQ to penalize
1890 * @active: IRQ active or not
1891 *
1892 * Permits the platform to provide architecture-specific functionality when
1893 * penalizing ISA IRQs. This is the default implementation. Architecture
1894 * implementations can override this.
1895 */
1896void __weak pcibios_penalize_isa_irq(int irq, int active) {}
1897
1898static void do_pci_disable_device(struct pci_dev *dev)
1899{
1900 u16 pci_command;
1901
1902 pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1903 if (pci_command & PCI_COMMAND_MASTER) {
1904 pci_command &= ~PCI_COMMAND_MASTER;
1905 pci_write_config_word(dev, PCI_COMMAND, pci_command);
1906 }
1907
1908 pcibios_disable_device(dev);
1909}
1910
1911/**
1912 * pci_disable_enabled_device - Disable device without updating enable_cnt
1913 * @dev: PCI device to disable
1914 *
1915 * NOTE: This function is a backend of PCI power management routines and is
1916 * not supposed to be called drivers.
1917 */
1918void pci_disable_enabled_device(struct pci_dev *dev)
1919{
1920 if (pci_is_enabled(dev))
1921 do_pci_disable_device(dev);
1922}
1923
1924/**
1925 * pci_disable_device - Disable PCI device after use
1926 * @dev: PCI device to be disabled
1927 *
1928 * Signal to the system that the PCI device is not in use by the system
1929 * anymore. This only involves disabling PCI bus-mastering, if active.
1930 *
1931 * Note we don't actually disable the device until all callers of
1932 * pci_enable_device() have called pci_disable_device().
1933 */
1934void pci_disable_device(struct pci_dev *dev)
1935{
1936 struct pci_devres *dr;
1937
1938 dr = find_pci_dr(dev);
1939 if (dr)
1940 dr->enabled = 0;
1941
1942 dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
1943 "disabling already-disabled device");
1944
1945 if (atomic_dec_return(&dev->enable_cnt) != 0)
1946 return;
1947
1948 do_pci_disable_device(dev);
1949
1950 dev->is_busmaster = 0;
1951}
1952EXPORT_SYMBOL(pci_disable_device);
1953
1954/**
1955 * pcibios_set_pcie_reset_state - set reset state for device dev
1956 * @dev: the PCIe device reset
1957 * @state: Reset state to enter into
1958 *
1959 * Set the PCIe reset state for the device. This is the default
1960 * implementation. Architecture implementations can override this.
1961 */
1962int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1963 enum pcie_reset_state state)
1964{
1965 return -EINVAL;
1966}
1967
1968/**
1969 * pci_set_pcie_reset_state - set reset state for device dev
1970 * @dev: the PCIe device reset
1971 * @state: Reset state to enter into
1972 *
1973 * Sets the PCI reset state for the device.
1974 */
1975int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1976{
1977 return pcibios_set_pcie_reset_state(dev, state);
1978}
1979EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
1980
1981/**
1982 * pcie_clear_root_pme_status - Clear root port PME interrupt status.
1983 * @dev: PCIe root port or event collector.
1984 */
1985void pcie_clear_root_pme_status(struct pci_dev *dev)
1986{
1987 pcie_capability_set_dword(dev, PCI_EXP_RTSTA, PCI_EXP_RTSTA_PME);
1988}
1989
1990/**
1991 * pci_check_pme_status - Check if given device has generated PME.
1992 * @dev: Device to check.
1993 *
1994 * Check the PME status of the device and if set, clear it and clear PME enable
1995 * (if set). Return 'true' if PME status and PME enable were both set or
1996 * 'false' otherwise.
1997 */
1998bool pci_check_pme_status(struct pci_dev *dev)
1999{
2000 int pmcsr_pos;
2001 u16 pmcsr;
2002 bool ret = false;
2003
2004 if (!dev->pm_cap)
2005 return false;
2006
2007 pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
2008 pci_read_config_word(dev, pmcsr_pos, &pmcsr);
2009 if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
2010 return false;
2011
2012 /* Clear PME status. */
2013 pmcsr |= PCI_PM_CTRL_PME_STATUS;
2014 if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
2015 /* Disable PME to avoid interrupt flood. */
2016 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2017 ret = true;
2018 }
2019
2020 pci_write_config_word(dev, pmcsr_pos, pmcsr);
2021
2022 return ret;
2023}
2024
2025/**
2026 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
2027 * @dev: Device to handle.
2028 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
2029 *
2030 * Check if @dev has generated PME and queue a resume request for it in that
2031 * case.
2032 */
2033static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
2034{
2035 if (pme_poll_reset && dev->pme_poll)
2036 dev->pme_poll = false;
2037
2038 if (pci_check_pme_status(dev)) {
2039 pci_wakeup_event(dev);
2040 pm_request_resume(&dev->dev);
2041 }
2042 return 0;
2043}
2044
2045/**
2046 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
2047 * @bus: Top bus of the subtree to walk.
2048 */
2049void pci_pme_wakeup_bus(struct pci_bus *bus)
2050{
2051 if (bus)
2052 pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
2053}
2054
2055
2056/**
2057 * pci_pme_capable - check the capability of PCI device to generate PME#
2058 * @dev: PCI device to handle.
2059 * @state: PCI state from which device will issue PME#.
2060 */
2061bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
2062{
2063 if (!dev->pm_cap)
2064 return false;
2065
2066 return !!(dev->pme_support & (1 << state));
2067}
2068EXPORT_SYMBOL(pci_pme_capable);
2069
2070static void pci_pme_list_scan(struct work_struct *work)
2071{
2072 struct pci_pme_device *pme_dev, *n;
2073
2074 mutex_lock(&pci_pme_list_mutex);
2075 list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
2076 if (pme_dev->dev->pme_poll) {
2077 struct pci_dev *bridge;
2078
2079 bridge = pme_dev->dev->bus->self;
2080 /*
2081 * If bridge is in low power state, the
2082 * configuration space of subordinate devices
2083 * may be not accessible
2084 */
2085 if (bridge && bridge->current_state != PCI_D0)
2086 continue;
2087 /*
2088 * If the device is in D3cold it should not be
2089 * polled either.
2090 */
2091 if (pme_dev->dev->current_state == PCI_D3cold)
2092 continue;
2093
2094 pci_pme_wakeup(pme_dev->dev, NULL);
2095 } else {
2096 list_del(&pme_dev->list);
2097 kfree(pme_dev);
2098 }
2099 }
2100 if (!list_empty(&pci_pme_list))
2101 queue_delayed_work(system_freezable_wq, &pci_pme_work,
2102 msecs_to_jiffies(PME_TIMEOUT));
2103 mutex_unlock(&pci_pme_list_mutex);
2104}
2105
2106static void __pci_pme_active(struct pci_dev *dev, bool enable)
2107{
2108 u16 pmcsr;
2109
2110 if (!dev->pme_support)
2111 return;
2112
2113 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2114 /* Clear PME_Status by writing 1 to it and enable PME# */
2115 pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
2116 if (!enable)
2117 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2118
2119 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2120}
2121
2122/**
2123 * pci_pme_restore - Restore PME configuration after config space restore.
2124 * @dev: PCI device to update.
2125 */
2126void pci_pme_restore(struct pci_dev *dev)
2127{
2128 u16 pmcsr;
2129
2130 if (!dev->pme_support)
2131 return;
2132
2133 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2134 if (dev->wakeup_prepared) {
2135 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2136 pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
2137 } else {
2138 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2139 pmcsr |= PCI_PM_CTRL_PME_STATUS;
2140 }
2141 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2142}
2143
2144/**
2145 * pci_pme_active - enable or disable PCI device's PME# function
2146 * @dev: PCI device to handle.
2147 * @enable: 'true' to enable PME# generation; 'false' to disable it.
2148 *
2149 * The caller must verify that the device is capable of generating PME# before
2150 * calling this function with @enable equal to 'true'.
2151 */
2152void pci_pme_active(struct pci_dev *dev, bool enable)
2153{
2154 __pci_pme_active(dev, enable);
2155
2156 /*
2157 * PCI (as opposed to PCIe) PME requires that the device have
2158 * its PME# line hooked up correctly. Not all hardware vendors
2159 * do this, so the PME never gets delivered and the device
2160 * remains asleep. The easiest way around this is to
2161 * periodically walk the list of suspended devices and check
2162 * whether any have their PME flag set. The assumption is that
2163 * we'll wake up often enough anyway that this won't be a huge
2164 * hit, and the power savings from the devices will still be a
2165 * win.
2166 *
2167 * Although PCIe uses in-band PME message instead of PME# line
2168 * to report PME, PME does not work for some PCIe devices in
2169 * reality. For example, there are devices that set their PME
2170 * status bits, but don't really bother to send a PME message;
2171 * there are PCI Express Root Ports that don't bother to
2172 * trigger interrupts when they receive PME messages from the
2173 * devices below. So PME poll is used for PCIe devices too.
2174 */
2175
2176 if (dev->pme_poll) {
2177 struct pci_pme_device *pme_dev;
2178 if (enable) {
2179 pme_dev = kmalloc(sizeof(struct pci_pme_device),
2180 GFP_KERNEL);
2181 if (!pme_dev) {
2182 pci_warn(dev, "can't enable PME#\n");
2183 return;
2184 }
2185 pme_dev->dev = dev;
2186 mutex_lock(&pci_pme_list_mutex);
2187 list_add(&pme_dev->list, &pci_pme_list);
2188 if (list_is_singular(&pci_pme_list))
2189 queue_delayed_work(system_freezable_wq,
2190 &pci_pme_work,
2191 msecs_to_jiffies(PME_TIMEOUT));
2192 mutex_unlock(&pci_pme_list_mutex);
2193 } else {
2194 mutex_lock(&pci_pme_list_mutex);
2195 list_for_each_entry(pme_dev, &pci_pme_list, list) {
2196 if (pme_dev->dev == dev) {
2197 list_del(&pme_dev->list);
2198 kfree(pme_dev);
2199 break;
2200 }
2201 }
2202 mutex_unlock(&pci_pme_list_mutex);
2203 }
2204 }
2205
2206 pci_dbg(dev, "PME# %s\n", enable ? "enabled" : "disabled");
2207}
2208EXPORT_SYMBOL(pci_pme_active);
2209
2210/**
2211 * __pci_enable_wake - enable PCI device as wakeup event source
2212 * @dev: PCI device affected
2213 * @state: PCI state from which device will issue wakeup events
2214 * @enable: True to enable event generation; false to disable
2215 *
2216 * This enables the device as a wakeup event source, or disables it.
2217 * When such events involves platform-specific hooks, those hooks are
2218 * called automatically by this routine.
2219 *
2220 * Devices with legacy power management (no standard PCI PM capabilities)
2221 * always require such platform hooks.
2222 *
2223 * RETURN VALUE:
2224 * 0 is returned on success
2225 * -EINVAL is returned if device is not supposed to wake up the system
2226 * Error code depending on the platform is returned if both the platform and
2227 * the native mechanism fail to enable the generation of wake-up events
2228 */
2229static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
2230{
2231 int ret = 0;
2232
2233 /*
2234 * Bridges that are not power-manageable directly only signal
2235 * wakeup on behalf of subordinate devices which is set up
2236 * elsewhere, so skip them. However, bridges that are
2237 * power-manageable may signal wakeup for themselves (for example,
2238 * on a hotplug event) and they need to be covered here.
2239 */
2240 if (!pci_power_manageable(dev))
2241 return 0;
2242
2243 /* Don't do the same thing twice in a row for one device. */
2244 if (!!enable == !!dev->wakeup_prepared)
2245 return 0;
2246
2247 /*
2248 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
2249 * Anderson we should be doing PME# wake enable followed by ACPI wake
2250 * enable. To disable wake-up we call the platform first, for symmetry.
2251 */
2252
2253 if (enable) {
2254 int error;
2255
2256 if (pci_pme_capable(dev, state))
2257 pci_pme_active(dev, true);
2258 else
2259 ret = 1;
2260 error = platform_pci_set_wakeup(dev, true);
2261 if (ret)
2262 ret = error;
2263 if (!ret)
2264 dev->wakeup_prepared = true;
2265 } else {
2266 platform_pci_set_wakeup(dev, false);
2267 pci_pme_active(dev, false);
2268 dev->wakeup_prepared = false;
2269 }
2270
2271 return ret;
2272}
2273
2274/**
2275 * pci_enable_wake - change wakeup settings for a PCI device
2276 * @pci_dev: Target device
2277 * @state: PCI state from which device will issue wakeup events
2278 * @enable: Whether or not to enable event generation
2279 *
2280 * If @enable is set, check device_may_wakeup() for the device before calling
2281 * __pci_enable_wake() for it.
2282 */
2283int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
2284{
2285 if (enable && !device_may_wakeup(&pci_dev->dev))
2286 return -EINVAL;
2287
2288 return __pci_enable_wake(pci_dev, state, enable);
2289}
2290EXPORT_SYMBOL(pci_enable_wake);
2291
2292/**
2293 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
2294 * @dev: PCI device to prepare
2295 * @enable: True to enable wake-up event generation; false to disable
2296 *
2297 * Many drivers want the device to wake up the system from D3_hot or D3_cold
2298 * and this function allows them to set that up cleanly - pci_enable_wake()
2299 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
2300 * ordering constraints.
2301 *
2302 * This function only returns error code if the device is not allowed to wake
2303 * up the system from sleep or it is not capable of generating PME# from both
2304 * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
2305 */
2306int pci_wake_from_d3(struct pci_dev *dev, bool enable)
2307{
2308 return pci_pme_capable(dev, PCI_D3cold) ?
2309 pci_enable_wake(dev, PCI_D3cold, enable) :
2310 pci_enable_wake(dev, PCI_D3hot, enable);
2311}
2312EXPORT_SYMBOL(pci_wake_from_d3);
2313
2314/**
2315 * pci_target_state - find an appropriate low power state for a given PCI dev
2316 * @dev: PCI device
2317 * @wakeup: Whether or not wakeup functionality will be enabled for the device.
2318 *
2319 * Use underlying platform code to find a supported low power state for @dev.
2320 * If the platform can't manage @dev, return the deepest state from which it
2321 * can generate wake events, based on any available PME info.
2322 */
2323static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
2324{
2325 pci_power_t target_state = PCI_D3hot;
2326
2327 if (platform_pci_power_manageable(dev)) {
2328 /*
2329 * Call the platform to find the target state for the device.
2330 */
2331 pci_power_t state = platform_pci_choose_state(dev);
2332
2333 switch (state) {
2334 case PCI_POWER_ERROR:
2335 case PCI_UNKNOWN:
2336 break;
2337 case PCI_D1:
2338 case PCI_D2:
2339 if (pci_no_d1d2(dev))
2340 break;
2341 /* else, fall through */
2342 default:
2343 target_state = state;
2344 }
2345
2346 return target_state;
2347 }
2348
2349 if (!dev->pm_cap)
2350 target_state = PCI_D0;
2351
2352 /*
2353 * If the device is in D3cold even though it's not power-manageable by
2354 * the platform, it may have been powered down by non-standard means.
2355 * Best to let it slumber.
2356 */
2357 if (dev->current_state == PCI_D3cold)
2358 target_state = PCI_D3cold;
2359
2360 if (wakeup) {
2361 /*
2362 * Find the deepest state from which the device can generate
2363 * PME#.
2364 */
2365 if (dev->pme_support) {
2366 while (target_state
2367 && !(dev->pme_support & (1 << target_state)))
2368 target_state--;
2369 }
2370 }
2371
2372 return target_state;
2373}
2374
2375/**
2376 * pci_prepare_to_sleep - prepare PCI device for system-wide transition
2377 * into a sleep state
2378 * @dev: Device to handle.
2379 *
2380 * Choose the power state appropriate for the device depending on whether
2381 * it can wake up the system and/or is power manageable by the platform
2382 * (PCI_D3hot is the default) and put the device into that state.
2383 */
2384int pci_prepare_to_sleep(struct pci_dev *dev)
2385{
2386 bool wakeup = device_may_wakeup(&dev->dev);
2387 pci_power_t target_state = pci_target_state(dev, wakeup);
2388 int error;
2389
2390 if (target_state == PCI_POWER_ERROR)
2391 return -EIO;
2392
2393 pci_enable_wake(dev, target_state, wakeup);
2394
2395 error = pci_set_power_state(dev, target_state);
2396
2397 if (error)
2398 pci_enable_wake(dev, target_state, false);
2399
2400 return error;
2401}
2402EXPORT_SYMBOL(pci_prepare_to_sleep);
2403
2404/**
2405 * pci_back_from_sleep - turn PCI device on during system-wide transition
2406 * into working state
2407 * @dev: Device to handle.
2408 *
2409 * Disable device's system wake-up capability and put it into D0.
2410 */
2411int pci_back_from_sleep(struct pci_dev *dev)
2412{
2413 pci_enable_wake(dev, PCI_D0, false);
2414 return pci_set_power_state(dev, PCI_D0);
2415}
2416EXPORT_SYMBOL(pci_back_from_sleep);
2417
2418/**
2419 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2420 * @dev: PCI device being suspended.
2421 *
2422 * Prepare @dev to generate wake-up events at run time and put it into a low
2423 * power state.
2424 */
2425int pci_finish_runtime_suspend(struct pci_dev *dev)
2426{
2427 pci_power_t target_state;
2428 int error;
2429
2430 target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2431 if (target_state == PCI_POWER_ERROR)
2432 return -EIO;
2433
2434 dev->runtime_d3cold = target_state == PCI_D3cold;
2435
2436 __pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2437
2438 error = pci_set_power_state(dev, target_state);
2439
2440 if (error) {
2441 pci_enable_wake(dev, target_state, false);
2442 dev->runtime_d3cold = false;
2443 }
2444
2445 return error;
2446}
2447
2448/**
2449 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2450 * @dev: Device to check.
2451 *
2452 * Return true if the device itself is capable of generating wake-up events
2453 * (through the platform or using the native PCIe PME) or if the device supports
2454 * PME and one of its upstream bridges can generate wake-up events.
2455 */
2456bool pci_dev_run_wake(struct pci_dev *dev)
2457{
2458 struct pci_bus *bus = dev->bus;
2459
2460 if (!dev->pme_support)
2461 return false;
2462
2463 /* PME-capable in principle, but not from the target power state */
2464 if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2465 return false;
2466
2467 if (device_can_wakeup(&dev->dev))
2468 return true;
2469
2470 while (bus->parent) {
2471 struct pci_dev *bridge = bus->self;
2472
2473 if (device_can_wakeup(&bridge->dev))
2474 return true;
2475
2476 bus = bus->parent;
2477 }
2478
2479 /* We have reached the root bus. */
2480 if (bus->bridge)
2481 return device_can_wakeup(bus->bridge);
2482
2483 return false;
2484}
2485EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2486
2487/**
2488 * pci_dev_need_resume - Check if it is necessary to resume the device.
2489 * @pci_dev: Device to check.
2490 *
2491 * Return 'true' if the device is not runtime-suspended or it has to be
2492 * reconfigured due to wakeup settings difference between system and runtime
2493 * suspend, or the current power state of it is not suitable for the upcoming
2494 * (system-wide) transition.
2495 */
2496bool pci_dev_need_resume(struct pci_dev *pci_dev)
2497{
2498 struct device *dev = &pci_dev->dev;
2499 pci_power_t target_state;
2500
2501 if (!pm_runtime_suspended(dev) || platform_pci_need_resume(pci_dev))
2502 return true;
2503
2504 target_state = pci_target_state(pci_dev, device_may_wakeup(dev));
2505
2506 /*
2507 * If the earlier platform check has not triggered, D3cold is just power
2508 * removal on top of D3hot, so no need to resume the device in that
2509 * case.
2510 */
2511 return target_state != pci_dev->current_state &&
2512 target_state != PCI_D3cold &&
2513 pci_dev->current_state != PCI_D3hot;
2514}
2515
2516/**
2517 * pci_dev_adjust_pme - Adjust PME setting for a suspended device.
2518 * @pci_dev: Device to check.
2519 *
2520 * If the device is suspended and it is not configured for system wakeup,
2521 * disable PME for it to prevent it from waking up the system unnecessarily.
2522 *
2523 * Note that if the device's power state is D3cold and the platform check in
2524 * pci_dev_need_resume() has not triggered, the device's configuration need not
2525 * be changed.
2526 */
2527void pci_dev_adjust_pme(struct pci_dev *pci_dev)
2528{
2529 struct device *dev = &pci_dev->dev;
2530
2531 spin_lock_irq(&dev->power.lock);
2532
2533 if (pm_runtime_suspended(dev) && !device_may_wakeup(dev) &&
2534 pci_dev->current_state < PCI_D3cold)
2535 __pci_pme_active(pci_dev, false);
2536
2537 spin_unlock_irq(&dev->power.lock);
2538}
2539
2540/**
2541 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2542 * @pci_dev: Device to handle.
2543 *
2544 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2545 * it might have been disabled during the prepare phase of system suspend if
2546 * the device was not configured for system wakeup.
2547 */
2548void pci_dev_complete_resume(struct pci_dev *pci_dev)
2549{
2550 struct device *dev = &pci_dev->dev;
2551
2552 if (!pci_dev_run_wake(pci_dev))
2553 return;
2554
2555 spin_lock_irq(&dev->power.lock);
2556
2557 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2558 __pci_pme_active(pci_dev, true);
2559
2560 spin_unlock_irq(&dev->power.lock);
2561}
2562
2563void pci_config_pm_runtime_get(struct pci_dev *pdev)
2564{
2565 struct device *dev = &pdev->dev;
2566 struct device *parent = dev->parent;
2567
2568 if (parent)
2569 pm_runtime_get_sync(parent);
2570 pm_runtime_get_noresume(dev);
2571 /*
2572 * pdev->current_state is set to PCI_D3cold during suspending,
2573 * so wait until suspending completes
2574 */
2575 pm_runtime_barrier(dev);
2576 /*
2577 * Only need to resume devices in D3cold, because config
2578 * registers are still accessible for devices suspended but
2579 * not in D3cold.
2580 */
2581 if (pdev->current_state == PCI_D3cold)
2582 pm_runtime_resume(dev);
2583}
2584
2585void pci_config_pm_runtime_put(struct pci_dev *pdev)
2586{
2587 struct device *dev = &pdev->dev;
2588 struct device *parent = dev->parent;
2589
2590 pm_runtime_put(dev);
2591 if (parent)
2592 pm_runtime_put_sync(parent);
2593}
2594
2595static const struct dmi_system_id bridge_d3_blacklist[] = {
2596#ifdef CONFIG_X86
2597 {
2598 /*
2599 * Gigabyte X299 root port is not marked as hotplug capable
2600 * which allows Linux to power manage it. However, this
2601 * confuses the BIOS SMI handler so don't power manage root
2602 * ports on that system.
2603 */
2604 .ident = "X299 DESIGNARE EX-CF",
2605 .matches = {
2606 DMI_MATCH(DMI_BOARD_VENDOR, "Gigabyte Technology Co., Ltd."),
2607 DMI_MATCH(DMI_BOARD_NAME, "X299 DESIGNARE EX-CF"),
2608 },
2609 },
2610#endif
2611 { }
2612};
2613
2614/**
2615 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2616 * @bridge: Bridge to check
2617 *
2618 * This function checks if it is possible to move the bridge to D3.
2619 * Currently we only allow D3 for recent enough PCIe ports and Thunderbolt.
2620 */
2621bool pci_bridge_d3_possible(struct pci_dev *bridge)
2622{
2623 if (!pci_is_pcie(bridge))
2624 return false;
2625
2626 switch (pci_pcie_type(bridge)) {
2627 case PCI_EXP_TYPE_ROOT_PORT:
2628 case PCI_EXP_TYPE_UPSTREAM:
2629 case PCI_EXP_TYPE_DOWNSTREAM:
2630 if (pci_bridge_d3_disable)
2631 return false;
2632
2633 /*
2634 * Hotplug ports handled by firmware in System Management Mode
2635 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2636 */
2637 if (bridge->is_hotplug_bridge && !pciehp_is_native(bridge))
2638 return false;
2639
2640 if (pci_bridge_d3_force)
2641 return true;
2642
2643 /* Even the oldest 2010 Thunderbolt controller supports D3. */
2644 if (bridge->is_thunderbolt)
2645 return true;
2646
2647 /* Platform might know better if the bridge supports D3 */
2648 if (platform_pci_bridge_d3(bridge))
2649 return true;
2650
2651 /*
2652 * Hotplug ports handled natively by the OS were not validated
2653 * by vendors for runtime D3 at least until 2018 because there
2654 * was no OS support.
2655 */
2656 if (bridge->is_hotplug_bridge)
2657 return false;
2658
2659 if (dmi_check_system(bridge_d3_blacklist))
2660 return false;
2661
2662 /*
2663 * It should be safe to put PCIe ports from 2015 or newer
2664 * to D3.
2665 */
2666 if (dmi_get_bios_year() >= 2015)
2667 return true;
2668 break;
2669 }
2670
2671 return false;
2672}
2673
2674static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2675{
2676 bool *d3cold_ok = data;
2677
2678 if (/* The device needs to be allowed to go D3cold ... */
2679 dev->no_d3cold || !dev->d3cold_allowed ||
2680
2681 /* ... and if it is wakeup capable to do so from D3cold. */
2682 (device_may_wakeup(&dev->dev) &&
2683 !pci_pme_capable(dev, PCI_D3cold)) ||
2684
2685 /* If it is a bridge it must be allowed to go to D3. */
2686 !pci_power_manageable(dev))
2687
2688 *d3cold_ok = false;
2689
2690 return !*d3cold_ok;
2691}
2692
2693/*
2694 * pci_bridge_d3_update - Update bridge D3 capabilities
2695 * @dev: PCI device which is changed
2696 *
2697 * Update upstream bridge PM capabilities accordingly depending on if the
2698 * device PM configuration was changed or the device is being removed. The
2699 * change is also propagated upstream.
2700 */
2701void pci_bridge_d3_update(struct pci_dev *dev)
2702{
2703 bool remove = !device_is_registered(&dev->dev);
2704 struct pci_dev *bridge;
2705 bool d3cold_ok = true;
2706
2707 bridge = pci_upstream_bridge(dev);
2708 if (!bridge || !pci_bridge_d3_possible(bridge))
2709 return;
2710
2711 /*
2712 * If D3 is currently allowed for the bridge, removing one of its
2713 * children won't change that.
2714 */
2715 if (remove && bridge->bridge_d3)
2716 return;
2717
2718 /*
2719 * If D3 is currently allowed for the bridge and a child is added or
2720 * changed, disallowance of D3 can only be caused by that child, so
2721 * we only need to check that single device, not any of its siblings.
2722 *
2723 * If D3 is currently not allowed for the bridge, checking the device
2724 * first may allow us to skip checking its siblings.
2725 */
2726 if (!remove)
2727 pci_dev_check_d3cold(dev, &d3cold_ok);
2728
2729 /*
2730 * If D3 is currently not allowed for the bridge, this may be caused
2731 * either by the device being changed/removed or any of its siblings,
2732 * so we need to go through all children to find out if one of them
2733 * continues to block D3.
2734 */
2735 if (d3cold_ok && !bridge->bridge_d3)
2736 pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
2737 &d3cold_ok);
2738
2739 if (bridge->bridge_d3 != d3cold_ok) {
2740 bridge->bridge_d3 = d3cold_ok;
2741 /* Propagate change to upstream bridges */
2742 pci_bridge_d3_update(bridge);
2743 }
2744}
2745
2746/**
2747 * pci_d3cold_enable - Enable D3cold for device
2748 * @dev: PCI device to handle
2749 *
2750 * This function can be used in drivers to enable D3cold from the device
2751 * they handle. It also updates upstream PCI bridge PM capabilities
2752 * accordingly.
2753 */
2754void pci_d3cold_enable(struct pci_dev *dev)
2755{
2756 if (dev->no_d3cold) {
2757 dev->no_d3cold = false;
2758 pci_bridge_d3_update(dev);
2759 }
2760}
2761EXPORT_SYMBOL_GPL(pci_d3cold_enable);
2762
2763/**
2764 * pci_d3cold_disable - Disable D3cold for device
2765 * @dev: PCI device to handle
2766 *
2767 * This function can be used in drivers to disable D3cold from the device
2768 * they handle. It also updates upstream PCI bridge PM capabilities
2769 * accordingly.
2770 */
2771void pci_d3cold_disable(struct pci_dev *dev)
2772{
2773 if (!dev->no_d3cold) {
2774 dev->no_d3cold = true;
2775 pci_bridge_d3_update(dev);
2776 }
2777}
2778EXPORT_SYMBOL_GPL(pci_d3cold_disable);
2779
2780/**
2781 * pci_pm_init - Initialize PM functions of given PCI device
2782 * @dev: PCI device to handle.
2783 */
2784void pci_pm_init(struct pci_dev *dev)
2785{
2786 int pm;
2787 u16 status;
2788 u16 pmc;
2789
2790 pm_runtime_forbid(&dev->dev);
2791 pm_runtime_set_active(&dev->dev);
2792 pm_runtime_enable(&dev->dev);
2793 device_enable_async_suspend(&dev->dev);
2794 dev->wakeup_prepared = false;
2795
2796 dev->pm_cap = 0;
2797 dev->pme_support = 0;
2798
2799 /* find PCI PM capability in list */
2800 pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2801 if (!pm)
2802 return;
2803 /* Check device's ability to generate PME# */
2804 pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2805
2806 if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2807 pci_err(dev, "unsupported PM cap regs version (%u)\n",
2808 pmc & PCI_PM_CAP_VER_MASK);
2809 return;
2810 }
2811
2812 dev->pm_cap = pm;
2813 dev->d3_delay = PCI_PM_D3_WAIT;
2814 dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2815 dev->bridge_d3 = pci_bridge_d3_possible(dev);
2816 dev->d3cold_allowed = true;
2817
2818 dev->d1_support = false;
2819 dev->d2_support = false;
2820 if (!pci_no_d1d2(dev)) {
2821 if (pmc & PCI_PM_CAP_D1)
2822 dev->d1_support = true;
2823 if (pmc & PCI_PM_CAP_D2)
2824 dev->d2_support = true;
2825
2826 if (dev->d1_support || dev->d2_support)
2827 pci_info(dev, "supports%s%s\n",
2828 dev->d1_support ? " D1" : "",
2829 dev->d2_support ? " D2" : "");
2830 }
2831
2832 pmc &= PCI_PM_CAP_PME_MASK;
2833 if (pmc) {
2834 pci_info(dev, "PME# supported from%s%s%s%s%s\n",
2835 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2836 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2837 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2838 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2839 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2840 dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2841 dev->pme_poll = true;
2842 /*
2843 * Make device's PM flags reflect the wake-up capability, but
2844 * let the user space enable it to wake up the system as needed.
2845 */
2846 device_set_wakeup_capable(&dev->dev, true);
2847 /* Disable the PME# generation functionality */
2848 pci_pme_active(dev, false);
2849 }
2850
2851 pci_read_config_word(dev, PCI_STATUS, &status);
2852 if (status & PCI_STATUS_IMM_READY)
2853 dev->imm_ready = 1;
2854}
2855
2856static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2857{
2858 unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
2859
2860 switch (prop) {
2861 case PCI_EA_P_MEM:
2862 case PCI_EA_P_VF_MEM:
2863 flags |= IORESOURCE_MEM;
2864 break;
2865 case PCI_EA_P_MEM_PREFETCH:
2866 case PCI_EA_P_VF_MEM_PREFETCH:
2867 flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2868 break;
2869 case PCI_EA_P_IO:
2870 flags |= IORESOURCE_IO;
2871 break;
2872 default:
2873 return 0;
2874 }
2875
2876 return flags;
2877}
2878
2879static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2880 u8 prop)
2881{
2882 if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2883 return &dev->resource[bei];
2884#ifdef CONFIG_PCI_IOV
2885 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2886 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2887 return &dev->resource[PCI_IOV_RESOURCES +
2888 bei - PCI_EA_BEI_VF_BAR0];
2889#endif
2890 else if (bei == PCI_EA_BEI_ROM)
2891 return &dev->resource[PCI_ROM_RESOURCE];
2892 else
2893 return NULL;
2894}
2895
2896/* Read an Enhanced Allocation (EA) entry */
2897static int pci_ea_read(struct pci_dev *dev, int offset)
2898{
2899 struct resource *res;
2900 int ent_size, ent_offset = offset;
2901 resource_size_t start, end;
2902 unsigned long flags;
2903 u32 dw0, bei, base, max_offset;
2904 u8 prop;
2905 bool support_64 = (sizeof(resource_size_t) >= 8);
2906
2907 pci_read_config_dword(dev, ent_offset, &dw0);
2908 ent_offset += 4;
2909
2910 /* Entry size field indicates DWORDs after 1st */
2911 ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2912
2913 if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2914 goto out;
2915
2916 bei = (dw0 & PCI_EA_BEI) >> 4;
2917 prop = (dw0 & PCI_EA_PP) >> 8;
2918
2919 /*
2920 * If the Property is in the reserved range, try the Secondary
2921 * Property instead.
2922 */
2923 if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2924 prop = (dw0 & PCI_EA_SP) >> 16;
2925 if (prop > PCI_EA_P_BRIDGE_IO)
2926 goto out;
2927
2928 res = pci_ea_get_resource(dev, bei, prop);
2929 if (!res) {
2930 pci_err(dev, "Unsupported EA entry BEI: %u\n", bei);
2931 goto out;
2932 }
2933
2934 flags = pci_ea_flags(dev, prop);
2935 if (!flags) {
2936 pci_err(dev, "Unsupported EA properties: %#x\n", prop);
2937 goto out;
2938 }
2939
2940 /* Read Base */
2941 pci_read_config_dword(dev, ent_offset, &base);
2942 start = (base & PCI_EA_FIELD_MASK);
2943 ent_offset += 4;
2944
2945 /* Read MaxOffset */
2946 pci_read_config_dword(dev, ent_offset, &max_offset);
2947 ent_offset += 4;
2948
2949 /* Read Base MSBs (if 64-bit entry) */
2950 if (base & PCI_EA_IS_64) {
2951 u32 base_upper;
2952
2953 pci_read_config_dword(dev, ent_offset, &base_upper);
2954 ent_offset += 4;
2955
2956 flags |= IORESOURCE_MEM_64;
2957
2958 /* entry starts above 32-bit boundary, can't use */
2959 if (!support_64 && base_upper)
2960 goto out;
2961
2962 if (support_64)
2963 start |= ((u64)base_upper << 32);
2964 }
2965
2966 end = start + (max_offset | 0x03);
2967
2968 /* Read MaxOffset MSBs (if 64-bit entry) */
2969 if (max_offset & PCI_EA_IS_64) {
2970 u32 max_offset_upper;
2971
2972 pci_read_config_dword(dev, ent_offset, &max_offset_upper);
2973 ent_offset += 4;
2974
2975 flags |= IORESOURCE_MEM_64;
2976
2977 /* entry too big, can't use */
2978 if (!support_64 && max_offset_upper)
2979 goto out;
2980
2981 if (support_64)
2982 end += ((u64)max_offset_upper << 32);
2983 }
2984
2985 if (end < start) {
2986 pci_err(dev, "EA Entry crosses address boundary\n");
2987 goto out;
2988 }
2989
2990 if (ent_size != ent_offset - offset) {
2991 pci_err(dev, "EA Entry Size (%d) does not match length read (%d)\n",
2992 ent_size, ent_offset - offset);
2993 goto out;
2994 }
2995
2996 res->name = pci_name(dev);
2997 res->start = start;
2998 res->end = end;
2999 res->flags = flags;
3000
3001 if (bei <= PCI_EA_BEI_BAR5)
3002 pci_info(dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
3003 bei, res, prop);
3004 else if (bei == PCI_EA_BEI_ROM)
3005 pci_info(dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
3006 res, prop);
3007 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
3008 pci_info(dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
3009 bei - PCI_EA_BEI_VF_BAR0, res, prop);
3010 else
3011 pci_info(dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
3012 bei, res, prop);
3013
3014out:
3015 return offset + ent_size;
3016}
3017
3018/* Enhanced Allocation Initialization */
3019void pci_ea_init(struct pci_dev *dev)
3020{
3021 int ea;
3022 u8 num_ent;
3023 int offset;
3024 int i;
3025
3026 /* find PCI EA capability in list */
3027 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
3028 if (!ea)
3029 return;
3030
3031 /* determine the number of entries */
3032 pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
3033 &num_ent);
3034 num_ent &= PCI_EA_NUM_ENT_MASK;
3035
3036 offset = ea + PCI_EA_FIRST_ENT;
3037
3038 /* Skip DWORD 2 for type 1 functions */
3039 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
3040 offset += 4;
3041
3042 /* parse each EA entry */
3043 for (i = 0; i < num_ent; ++i)
3044 offset = pci_ea_read(dev, offset);
3045}
3046
3047static void pci_add_saved_cap(struct pci_dev *pci_dev,
3048 struct pci_cap_saved_state *new_cap)
3049{
3050 hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
3051}
3052
3053/**
3054 * _pci_add_cap_save_buffer - allocate buffer for saving given
3055 * capability registers
3056 * @dev: the PCI device
3057 * @cap: the capability to allocate the buffer for
3058 * @extended: Standard or Extended capability ID
3059 * @size: requested size of the buffer
3060 */
3061static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
3062 bool extended, unsigned int size)
3063{
3064 int pos;
3065 struct pci_cap_saved_state *save_state;
3066
3067 if (extended)
3068 pos = pci_find_ext_capability(dev, cap);
3069 else
3070 pos = pci_find_capability(dev, cap);
3071
3072 if (!pos)
3073 return 0;
3074
3075 save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
3076 if (!save_state)
3077 return -ENOMEM;
3078
3079 save_state->cap.cap_nr = cap;
3080 save_state->cap.cap_extended = extended;
3081 save_state->cap.size = size;
3082 pci_add_saved_cap(dev, save_state);
3083
3084 return 0;
3085}
3086
3087int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
3088{
3089 return _pci_add_cap_save_buffer(dev, cap, false, size);
3090}
3091
3092int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
3093{
3094 return _pci_add_cap_save_buffer(dev, cap, true, size);
3095}
3096
3097/**
3098 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
3099 * @dev: the PCI device
3100 */
3101void pci_allocate_cap_save_buffers(struct pci_dev *dev)
3102{
3103 int error;
3104
3105 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
3106 PCI_EXP_SAVE_REGS * sizeof(u16));
3107 if (error)
3108 pci_err(dev, "unable to preallocate PCI Express save buffer\n");
3109
3110 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
3111 if (error)
3112 pci_err(dev, "unable to preallocate PCI-X save buffer\n");
3113
3114 error = pci_add_ext_cap_save_buffer(dev, PCI_EXT_CAP_ID_LTR,
3115 2 * sizeof(u16));
3116 if (error)
3117 pci_err(dev, "unable to allocate suspend buffer for LTR\n");
3118
3119 pci_allocate_vc_save_buffers(dev);
3120}
3121
3122void pci_free_cap_save_buffers(struct pci_dev *dev)
3123{
3124 struct pci_cap_saved_state *tmp;
3125 struct hlist_node *n;
3126
3127 hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
3128 kfree(tmp);
3129}
3130
3131/**
3132 * pci_configure_ari - enable or disable ARI forwarding
3133 * @dev: the PCI device
3134 *
3135 * If @dev and its upstream bridge both support ARI, enable ARI in the
3136 * bridge. Otherwise, disable ARI in the bridge.
3137 */
3138void pci_configure_ari(struct pci_dev *dev)
3139{
3140 u32 cap;
3141 struct pci_dev *bridge;
3142
3143 if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
3144 return;
3145
3146 bridge = dev->bus->self;
3147 if (!bridge)
3148 return;
3149
3150 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3151 if (!(cap & PCI_EXP_DEVCAP2_ARI))
3152 return;
3153
3154 if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
3155 pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
3156 PCI_EXP_DEVCTL2_ARI);
3157 bridge->ari_enabled = 1;
3158 } else {
3159 pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
3160 PCI_EXP_DEVCTL2_ARI);
3161 bridge->ari_enabled = 0;
3162 }
3163}
3164
3165static int pci_acs_enable;
3166
3167/**
3168 * pci_request_acs - ask for ACS to be enabled if supported
3169 */
3170void pci_request_acs(void)
3171{
3172 pci_acs_enable = 1;
3173}
3174
3175static const char *disable_acs_redir_param;
3176
3177/**
3178 * pci_disable_acs_redir - disable ACS redirect capabilities
3179 * @dev: the PCI device
3180 *
3181 * For only devices specified in the disable_acs_redir parameter.
3182 */
3183static void pci_disable_acs_redir(struct pci_dev *dev)
3184{
3185 int ret = 0;
3186 const char *p;
3187 int pos;
3188 u16 ctrl;
3189
3190 if (!disable_acs_redir_param)
3191 return;
3192
3193 p = disable_acs_redir_param;
3194 while (*p) {
3195 ret = pci_dev_str_match(dev, p, &p);
3196 if (ret < 0) {
3197 pr_info_once("PCI: Can't parse disable_acs_redir parameter: %s\n",
3198 disable_acs_redir_param);
3199
3200 break;
3201 } else if (ret == 1) {
3202 /* Found a match */
3203 break;
3204 }
3205
3206 if (*p != ';' && *p != ',') {
3207 /* End of param or invalid format */
3208 break;
3209 }
3210 p++;
3211 }
3212
3213 if (ret != 1)
3214 return;
3215
3216 if (!pci_dev_specific_disable_acs_redir(dev))
3217 return;
3218
3219 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3220 if (!pos) {
3221 pci_warn(dev, "cannot disable ACS redirect for this hardware as it does not have ACS capabilities\n");
3222 return;
3223 }
3224
3225 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
3226
3227 /* P2P Request & Completion Redirect */
3228 ctrl &= ~(PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC);
3229
3230 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
3231
3232 pci_info(dev, "disabled ACS redirect\n");
3233}
3234
3235/**
3236 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilities
3237 * @dev: the PCI device
3238 */
3239static void pci_std_enable_acs(struct pci_dev *dev)
3240{
3241 int pos;
3242 u16 cap;
3243 u16 ctrl;
3244
3245 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3246 if (!pos)
3247 return;
3248
3249 pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
3250 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
3251
3252 /* Source Validation */
3253 ctrl |= (cap & PCI_ACS_SV);
3254
3255 /* P2P Request Redirect */
3256 ctrl |= (cap & PCI_ACS_RR);
3257
3258 /* P2P Completion Redirect */
3259 ctrl |= (cap & PCI_ACS_CR);
3260
3261 /* Upstream Forwarding */
3262 ctrl |= (cap & PCI_ACS_UF);
3263
3264 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
3265}
3266
3267/**
3268 * pci_enable_acs - enable ACS if hardware support it
3269 * @dev: the PCI device
3270 */
3271void pci_enable_acs(struct pci_dev *dev)
3272{
3273 if (!pci_acs_enable)
3274 goto disable_acs_redir;
3275
3276 if (!pci_dev_specific_enable_acs(dev))
3277 goto disable_acs_redir;
3278
3279 pci_std_enable_acs(dev);
3280
3281disable_acs_redir:
3282 /*
3283 * Note: pci_disable_acs_redir() must be called even if ACS was not
3284 * enabled by the kernel because it may have been enabled by
3285 * platform firmware. So if we are told to disable it, we should
3286 * always disable it after setting the kernel's default
3287 * preferences.
3288 */
3289 pci_disable_acs_redir(dev);
3290}
3291
3292static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
3293{
3294 int pos;
3295 u16 cap, ctrl;
3296
3297 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
3298 if (!pos)
3299 return false;
3300
3301 /*
3302 * Except for egress control, capabilities are either required
3303 * or only required if controllable. Features missing from the
3304 * capability field can therefore be assumed as hard-wired enabled.
3305 */
3306 pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
3307 acs_flags &= (cap | PCI_ACS_EC);
3308
3309 pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
3310 return (ctrl & acs_flags) == acs_flags;
3311}
3312
3313/**
3314 * pci_acs_enabled - test ACS against required flags for a given device
3315 * @pdev: device to test
3316 * @acs_flags: required PCI ACS flags
3317 *
3318 * Return true if the device supports the provided flags. Automatically
3319 * filters out flags that are not implemented on multifunction devices.
3320 *
3321 * Note that this interface checks the effective ACS capabilities of the
3322 * device rather than the actual capabilities. For instance, most single
3323 * function endpoints are not required to support ACS because they have no
3324 * opportunity for peer-to-peer access. We therefore return 'true'
3325 * regardless of whether the device exposes an ACS capability. This makes
3326 * it much easier for callers of this function to ignore the actual type
3327 * or topology of the device when testing ACS support.
3328 */
3329bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
3330{
3331 int ret;
3332
3333 ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
3334 if (ret >= 0)
3335 return ret > 0;
3336
3337 /*
3338 * Conventional PCI and PCI-X devices never support ACS, either
3339 * effectively or actually. The shared bus topology implies that
3340 * any device on the bus can receive or snoop DMA.
3341 */
3342 if (!pci_is_pcie(pdev))
3343 return false;
3344
3345 switch (pci_pcie_type(pdev)) {
3346 /*
3347 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
3348 * but since their primary interface is PCI/X, we conservatively
3349 * handle them as we would a non-PCIe device.
3350 */
3351 case PCI_EXP_TYPE_PCIE_BRIDGE:
3352 /*
3353 * PCIe 3.0, 6.12.1 excludes ACS on these devices. "ACS is never
3354 * applicable... must never implement an ACS Extended Capability...".
3355 * This seems arbitrary, but we take a conservative interpretation
3356 * of this statement.
3357 */
3358 case PCI_EXP_TYPE_PCI_BRIDGE:
3359 case PCI_EXP_TYPE_RC_EC:
3360 return false;
3361 /*
3362 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
3363 * implement ACS in order to indicate their peer-to-peer capabilities,
3364 * regardless of whether they are single- or multi-function devices.
3365 */
3366 case PCI_EXP_TYPE_DOWNSTREAM:
3367 case PCI_EXP_TYPE_ROOT_PORT:
3368 return pci_acs_flags_enabled(pdev, acs_flags);
3369 /*
3370 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
3371 * implemented by the remaining PCIe types to indicate peer-to-peer
3372 * capabilities, but only when they are part of a multifunction
3373 * device. The footnote for section 6.12 indicates the specific
3374 * PCIe types included here.
3375 */
3376 case PCI_EXP_TYPE_ENDPOINT:
3377 case PCI_EXP_TYPE_UPSTREAM:
3378 case PCI_EXP_TYPE_LEG_END:
3379 case PCI_EXP_TYPE_RC_END:
3380 if (!pdev->multifunction)
3381 break;
3382
3383 return pci_acs_flags_enabled(pdev, acs_flags);
3384 }
3385
3386 /*
3387 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
3388 * to single function devices with the exception of downstream ports.
3389 */
3390 return true;
3391}
3392
3393/**
3394 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
3395 * @start: starting downstream device
3396 * @end: ending upstream device or NULL to search to the root bus
3397 * @acs_flags: required flags
3398 *
3399 * Walk up a device tree from start to end testing PCI ACS support. If
3400 * any step along the way does not support the required flags, return false.
3401 */
3402bool pci_acs_path_enabled(struct pci_dev *start,
3403 struct pci_dev *end, u16 acs_flags)
3404{
3405 struct pci_dev *pdev, *parent = start;
3406
3407 do {
3408 pdev = parent;
3409
3410 if (!pci_acs_enabled(pdev, acs_flags))
3411 return false;
3412
3413 if (pci_is_root_bus(pdev->bus))
3414 return (end == NULL);
3415
3416 parent = pdev->bus->self;
3417 } while (pdev != end);
3418
3419 return true;
3420}
3421
3422/**
3423 * pci_rebar_find_pos - find position of resize ctrl reg for BAR
3424 * @pdev: PCI device
3425 * @bar: BAR to find
3426 *
3427 * Helper to find the position of the ctrl register for a BAR.
3428 * Returns -ENOTSUPP if resizable BARs are not supported at all.
3429 * Returns -ENOENT if no ctrl register for the BAR could be found.
3430 */
3431static int pci_rebar_find_pos(struct pci_dev *pdev, int bar)
3432{
3433 unsigned int pos, nbars, i;
3434 u32 ctrl;
3435
3436 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
3437 if (!pos)
3438 return -ENOTSUPP;
3439
3440 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3441 nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
3442 PCI_REBAR_CTRL_NBAR_SHIFT;
3443
3444 for (i = 0; i < nbars; i++, pos += 8) {
3445 int bar_idx;
3446
3447 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3448 bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
3449 if (bar_idx == bar)
3450 return pos;
3451 }
3452
3453 return -ENOENT;
3454}
3455
3456/**
3457 * pci_rebar_get_possible_sizes - get possible sizes for BAR
3458 * @pdev: PCI device
3459 * @bar: BAR to query
3460 *
3461 * Get the possible sizes of a resizable BAR as bitmask defined in the spec
3462 * (bit 0=1MB, bit 19=512GB). Returns 0 if BAR isn't resizable.
3463 */
3464u32 pci_rebar_get_possible_sizes(struct pci_dev *pdev, int bar)
3465{
3466 int pos;
3467 u32 cap;
3468
3469 pos = pci_rebar_find_pos(pdev, bar);
3470 if (pos < 0)
3471 return 0;
3472
3473 pci_read_config_dword(pdev, pos + PCI_REBAR_CAP, &cap);
3474 return (cap & PCI_REBAR_CAP_SIZES) >> 4;
3475}
3476
3477/**
3478 * pci_rebar_get_current_size - get the current size of a BAR
3479 * @pdev: PCI device
3480 * @bar: BAR to set size to
3481 *
3482 * Read the size of a BAR from the resizable BAR config.
3483 * Returns size if found or negative error code.
3484 */
3485int pci_rebar_get_current_size(struct pci_dev *pdev, int bar)
3486{
3487 int pos;
3488 u32 ctrl;
3489
3490 pos = pci_rebar_find_pos(pdev, bar);
3491 if (pos < 0)
3492 return pos;
3493
3494 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3495 return (ctrl & PCI_REBAR_CTRL_BAR_SIZE) >> PCI_REBAR_CTRL_BAR_SHIFT;
3496}
3497
3498/**
3499 * pci_rebar_set_size - set a new size for a BAR
3500 * @pdev: PCI device
3501 * @bar: BAR to set size to
3502 * @size: new size as defined in the spec (0=1MB, 19=512GB)
3503 *
3504 * Set the new size of a BAR as defined in the spec.
3505 * Returns zero if resizing was successful, error code otherwise.
3506 */
3507int pci_rebar_set_size(struct pci_dev *pdev, int bar, int size)
3508{
3509 int pos;
3510 u32 ctrl;
3511
3512 pos = pci_rebar_find_pos(pdev, bar);
3513 if (pos < 0)
3514 return pos;
3515
3516 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3517 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
3518 ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
3519 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
3520 return 0;
3521}
3522
3523/**
3524 * pci_enable_atomic_ops_to_root - enable AtomicOp requests to root port
3525 * @dev: the PCI device
3526 * @cap_mask: mask of desired AtomicOp sizes, including one or more of:
3527 * PCI_EXP_DEVCAP2_ATOMIC_COMP32
3528 * PCI_EXP_DEVCAP2_ATOMIC_COMP64
3529 * PCI_EXP_DEVCAP2_ATOMIC_COMP128
3530 *
3531 * Return 0 if all upstream bridges support AtomicOp routing, egress
3532 * blocking is disabled on all upstream ports, and the root port supports
3533 * the requested completion capabilities (32-bit, 64-bit and/or 128-bit
3534 * AtomicOp completion), or negative otherwise.
3535 */
3536int pci_enable_atomic_ops_to_root(struct pci_dev *dev, u32 cap_mask)
3537{
3538 struct pci_bus *bus = dev->bus;
3539 struct pci_dev *bridge;
3540 u32 cap, ctl2;
3541
3542 if (!pci_is_pcie(dev))
3543 return -EINVAL;
3544
3545 /*
3546 * Per PCIe r4.0, sec 6.15, endpoints and root ports may be
3547 * AtomicOp requesters. For now, we only support endpoints as
3548 * requesters and root ports as completers. No endpoints as
3549 * completers, and no peer-to-peer.
3550 */
3551
3552 switch (pci_pcie_type(dev)) {
3553 case PCI_EXP_TYPE_ENDPOINT:
3554 case PCI_EXP_TYPE_LEG_END:
3555 case PCI_EXP_TYPE_RC_END:
3556 break;
3557 default:
3558 return -EINVAL;
3559 }
3560
3561 while (bus->parent) {
3562 bridge = bus->self;
3563
3564 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3565
3566 switch (pci_pcie_type(bridge)) {
3567 /* Ensure switch ports support AtomicOp routing */
3568 case PCI_EXP_TYPE_UPSTREAM:
3569 case PCI_EXP_TYPE_DOWNSTREAM:
3570 if (!(cap & PCI_EXP_DEVCAP2_ATOMIC_ROUTE))
3571 return -EINVAL;
3572 break;
3573
3574 /* Ensure root port supports all the sizes we care about */
3575 case PCI_EXP_TYPE_ROOT_PORT:
3576 if ((cap & cap_mask) != cap_mask)
3577 return -EINVAL;
3578 break;
3579 }
3580
3581 /* Ensure upstream ports don't block AtomicOps on egress */
3582 if (pci_pcie_type(bridge) == PCI_EXP_TYPE_UPSTREAM) {
3583 pcie_capability_read_dword(bridge, PCI_EXP_DEVCTL2,
3584 &ctl2);
3585 if (ctl2 & PCI_EXP_DEVCTL2_ATOMIC_EGRESS_BLOCK)
3586 return -EINVAL;
3587 }
3588
3589 bus = bus->parent;
3590 }
3591
3592 pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
3593 PCI_EXP_DEVCTL2_ATOMIC_REQ);
3594 return 0;
3595}
3596EXPORT_SYMBOL(pci_enable_atomic_ops_to_root);
3597
3598/**
3599 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
3600 * @dev: the PCI device
3601 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
3602 *
3603 * Perform INTx swizzling for a device behind one level of bridge. This is
3604 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
3605 * behind bridges on add-in cards. For devices with ARI enabled, the slot
3606 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
3607 * the PCI Express Base Specification, Revision 2.1)
3608 */
3609u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
3610{
3611 int slot;
3612
3613 if (pci_ari_enabled(dev->bus))
3614 slot = 0;
3615 else
3616 slot = PCI_SLOT(dev->devfn);
3617
3618 return (((pin - 1) + slot) % 4) + 1;
3619}
3620
3621int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3622{
3623 u8 pin;
3624
3625 pin = dev->pin;
3626 if (!pin)
3627 return -1;
3628
3629 while (!pci_is_root_bus(dev->bus)) {
3630 pin = pci_swizzle_interrupt_pin(dev, pin);
3631 dev = dev->bus->self;
3632 }
3633 *bridge = dev;
3634 return pin;
3635}
3636
3637/**
3638 * pci_common_swizzle - swizzle INTx all the way to root bridge
3639 * @dev: the PCI device
3640 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3641 *
3642 * Perform INTx swizzling for a device. This traverses through all PCI-to-PCI
3643 * bridges all the way up to a PCI root bus.
3644 */
3645u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3646{
3647 u8 pin = *pinp;
3648
3649 while (!pci_is_root_bus(dev->bus)) {
3650 pin = pci_swizzle_interrupt_pin(dev, pin);
3651 dev = dev->bus->self;
3652 }
3653 *pinp = pin;
3654 return PCI_SLOT(dev->devfn);
3655}
3656EXPORT_SYMBOL_GPL(pci_common_swizzle);
3657
3658/**
3659 * pci_release_region - Release a PCI bar
3660 * @pdev: PCI device whose resources were previously reserved by
3661 * pci_request_region()
3662 * @bar: BAR to release
3663 *
3664 * Releases the PCI I/O and memory resources previously reserved by a
3665 * successful call to pci_request_region(). Call this function only
3666 * after all use of the PCI regions has ceased.
3667 */
3668void pci_release_region(struct pci_dev *pdev, int bar)
3669{
3670 struct pci_devres *dr;
3671
3672 if (pci_resource_len(pdev, bar) == 0)
3673 return;
3674 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3675 release_region(pci_resource_start(pdev, bar),
3676 pci_resource_len(pdev, bar));
3677 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3678 release_mem_region(pci_resource_start(pdev, bar),
3679 pci_resource_len(pdev, bar));
3680
3681 dr = find_pci_dr(pdev);
3682 if (dr)
3683 dr->region_mask &= ~(1 << bar);
3684}
3685EXPORT_SYMBOL(pci_release_region);
3686
3687/**
3688 * __pci_request_region - Reserved PCI I/O and memory resource
3689 * @pdev: PCI device whose resources are to be reserved
3690 * @bar: BAR to be reserved
3691 * @res_name: Name to be associated with resource.
3692 * @exclusive: whether the region access is exclusive or not
3693 *
3694 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3695 * being reserved by owner @res_name. Do not access any
3696 * address inside the PCI regions unless this call returns
3697 * successfully.
3698 *
3699 * If @exclusive is set, then the region is marked so that userspace
3700 * is explicitly not allowed to map the resource via /dev/mem or
3701 * sysfs MMIO access.
3702 *
3703 * Returns 0 on success, or %EBUSY on error. A warning
3704 * message is also printed on failure.
3705 */
3706static int __pci_request_region(struct pci_dev *pdev, int bar,
3707 const char *res_name, int exclusive)
3708{
3709 struct pci_devres *dr;
3710
3711 if (pci_resource_len(pdev, bar) == 0)
3712 return 0;
3713
3714 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3715 if (!request_region(pci_resource_start(pdev, bar),
3716 pci_resource_len(pdev, bar), res_name))
3717 goto err_out;
3718 } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3719 if (!__request_mem_region(pci_resource_start(pdev, bar),
3720 pci_resource_len(pdev, bar), res_name,
3721 exclusive))
3722 goto err_out;
3723 }
3724
3725 dr = find_pci_dr(pdev);
3726 if (dr)
3727 dr->region_mask |= 1 << bar;
3728
3729 return 0;
3730
3731err_out:
3732 pci_warn(pdev, "BAR %d: can't reserve %pR\n", bar,
3733 &pdev->resource[bar]);
3734 return -EBUSY;
3735}
3736
3737/**
3738 * pci_request_region - Reserve PCI I/O and memory resource
3739 * @pdev: PCI device whose resources are to be reserved
3740 * @bar: BAR to be reserved
3741 * @res_name: Name to be associated with resource
3742 *
3743 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3744 * being reserved by owner @res_name. Do not access any
3745 * address inside the PCI regions unless this call returns
3746 * successfully.
3747 *
3748 * Returns 0 on success, or %EBUSY on error. A warning
3749 * message is also printed on failure.
3750 */
3751int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3752{
3753 return __pci_request_region(pdev, bar, res_name, 0);
3754}
3755EXPORT_SYMBOL(pci_request_region);
3756
3757/**
3758 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3759 * @pdev: PCI device whose resources were previously reserved
3760 * @bars: Bitmask of BARs to be released
3761 *
3762 * Release selected PCI I/O and memory resources previously reserved.
3763 * Call this function only after all use of the PCI regions has ceased.
3764 */
3765void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3766{
3767 int i;
3768
3769 for (i = 0; i < 6; i++)
3770 if (bars & (1 << i))
3771 pci_release_region(pdev, i);
3772}
3773EXPORT_SYMBOL(pci_release_selected_regions);
3774
3775static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3776 const char *res_name, int excl)
3777{
3778 int i;
3779
3780 for (i = 0; i < 6; i++)
3781 if (bars & (1 << i))
3782 if (__pci_request_region(pdev, i, res_name, excl))
3783 goto err_out;
3784 return 0;
3785
3786err_out:
3787 while (--i >= 0)
3788 if (bars & (1 << i))
3789 pci_release_region(pdev, i);
3790
3791 return -EBUSY;
3792}
3793
3794
3795/**
3796 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3797 * @pdev: PCI device whose resources are to be reserved
3798 * @bars: Bitmask of BARs to be requested
3799 * @res_name: Name to be associated with resource
3800 */
3801int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3802 const char *res_name)
3803{
3804 return __pci_request_selected_regions(pdev, bars, res_name, 0);
3805}
3806EXPORT_SYMBOL(pci_request_selected_regions);
3807
3808int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3809 const char *res_name)
3810{
3811 return __pci_request_selected_regions(pdev, bars, res_name,
3812 IORESOURCE_EXCLUSIVE);
3813}
3814EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3815
3816/**
3817 * pci_release_regions - Release reserved PCI I/O and memory resources
3818 * @pdev: PCI device whose resources were previously reserved by
3819 * pci_request_regions()
3820 *
3821 * Releases all PCI I/O and memory resources previously reserved by a
3822 * successful call to pci_request_regions(). Call this function only
3823 * after all use of the PCI regions has ceased.
3824 */
3825
3826void pci_release_regions(struct pci_dev *pdev)
3827{
3828 pci_release_selected_regions(pdev, (1 << 6) - 1);
3829}
3830EXPORT_SYMBOL(pci_release_regions);
3831
3832/**
3833 * pci_request_regions - Reserve PCI I/O and memory resources
3834 * @pdev: PCI device whose resources are to be reserved
3835 * @res_name: Name to be associated with resource.
3836 *
3837 * Mark all PCI regions associated with PCI device @pdev as
3838 * being reserved by owner @res_name. Do not access any
3839 * address inside the PCI regions unless this call returns
3840 * successfully.
3841 *
3842 * Returns 0 on success, or %EBUSY on error. A warning
3843 * message is also printed on failure.
3844 */
3845int pci_request_regions(struct pci_dev *pdev, const char *res_name)
3846{
3847 return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
3848}
3849EXPORT_SYMBOL(pci_request_regions);
3850
3851/**
3852 * pci_request_regions_exclusive - Reserve PCI I/O and memory resources
3853 * @pdev: PCI device whose resources are to be reserved
3854 * @res_name: Name to be associated with resource.
3855 *
3856 * Mark all PCI regions associated with PCI device @pdev as being reserved
3857 * by owner @res_name. Do not access any address inside the PCI regions
3858 * unless this call returns successfully.
3859 *
3860 * pci_request_regions_exclusive() will mark the region so that /dev/mem
3861 * and the sysfs MMIO access will not be allowed.
3862 *
3863 * Returns 0 on success, or %EBUSY on error. A warning message is also
3864 * printed on failure.
3865 */
3866int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3867{
3868 return pci_request_selected_regions_exclusive(pdev,
3869 ((1 << 6) - 1), res_name);
3870}
3871EXPORT_SYMBOL(pci_request_regions_exclusive);
3872
3873/*
3874 * Record the PCI IO range (expressed as CPU physical address + size).
3875 * Return a negative value if an error has occurred, zero otherwise
3876 */
3877int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
3878 resource_size_t size)
3879{
3880 int ret = 0;
3881#ifdef PCI_IOBASE
3882 struct logic_pio_hwaddr *range;
3883
3884 if (!size || addr + size < addr)
3885 return -EINVAL;
3886
3887 range = kzalloc(sizeof(*range), GFP_ATOMIC);
3888 if (!range)
3889 return -ENOMEM;
3890
3891 range->fwnode = fwnode;
3892 range->size = size;
3893 range->hw_start = addr;
3894 range->flags = LOGIC_PIO_CPU_MMIO;
3895
3896 ret = logic_pio_register_range(range);
3897 if (ret)
3898 kfree(range);
3899#endif
3900
3901 return ret;
3902}
3903
3904phys_addr_t pci_pio_to_address(unsigned long pio)
3905{
3906 phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3907
3908#ifdef PCI_IOBASE
3909 if (pio >= MMIO_UPPER_LIMIT)
3910 return address;
3911
3912 address = logic_pio_to_hwaddr(pio);
3913#endif
3914
3915 return address;
3916}
3917
3918unsigned long __weak pci_address_to_pio(phys_addr_t address)
3919{
3920#ifdef PCI_IOBASE
3921 return logic_pio_trans_cpuaddr(address);
3922#else
3923 if (address > IO_SPACE_LIMIT)
3924 return (unsigned long)-1;
3925
3926 return (unsigned long) address;
3927#endif
3928}
3929
3930/**
3931 * pci_remap_iospace - Remap the memory mapped I/O space
3932 * @res: Resource describing the I/O space
3933 * @phys_addr: physical address of range to be mapped
3934 *
3935 * Remap the memory mapped I/O space described by the @res and the CPU
3936 * physical address @phys_addr into virtual address space. Only
3937 * architectures that have memory mapped IO functions defined (and the
3938 * PCI_IOBASE value defined) should call this function.
3939 */
3940int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3941{
3942#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3943 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3944
3945 if (!(res->flags & IORESOURCE_IO))
3946 return -EINVAL;
3947
3948 if (res->end > IO_SPACE_LIMIT)
3949 return -EINVAL;
3950
3951 return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3952 pgprot_device(PAGE_KERNEL));
3953#else
3954 /*
3955 * This architecture does not have memory mapped I/O space,
3956 * so this function should never be called
3957 */
3958 WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3959 return -ENODEV;
3960#endif
3961}
3962EXPORT_SYMBOL(pci_remap_iospace);
3963
3964/**
3965 * pci_unmap_iospace - Unmap the memory mapped I/O space
3966 * @res: resource to be unmapped
3967 *
3968 * Unmap the CPU virtual address @res from virtual address space. Only
3969 * architectures that have memory mapped IO functions defined (and the
3970 * PCI_IOBASE value defined) should call this function.
3971 */
3972void pci_unmap_iospace(struct resource *res)
3973{
3974#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3975 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3976
3977 unmap_kernel_range(vaddr, resource_size(res));
3978#endif
3979}
3980EXPORT_SYMBOL(pci_unmap_iospace);
3981
3982static void devm_pci_unmap_iospace(struct device *dev, void *ptr)
3983{
3984 struct resource **res = ptr;
3985
3986 pci_unmap_iospace(*res);
3987}
3988
3989/**
3990 * devm_pci_remap_iospace - Managed pci_remap_iospace()
3991 * @dev: Generic device to remap IO address for
3992 * @res: Resource describing the I/O space
3993 * @phys_addr: physical address of range to be mapped
3994 *
3995 * Managed pci_remap_iospace(). Map is automatically unmapped on driver
3996 * detach.
3997 */
3998int devm_pci_remap_iospace(struct device *dev, const struct resource *res,
3999 phys_addr_t phys_addr)
4000{
4001 const struct resource **ptr;
4002 int error;
4003
4004 ptr = devres_alloc(devm_pci_unmap_iospace, sizeof(*ptr), GFP_KERNEL);
4005 if (!ptr)
4006 return -ENOMEM;
4007
4008 error = pci_remap_iospace(res, phys_addr);
4009 if (error) {
4010 devres_free(ptr);
4011 } else {
4012 *ptr = res;
4013 devres_add(dev, ptr);
4014 }
4015
4016 return error;
4017}
4018EXPORT_SYMBOL(devm_pci_remap_iospace);
4019
4020/**
4021 * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
4022 * @dev: Generic device to remap IO address for
4023 * @offset: Resource address to map
4024 * @size: Size of map
4025 *
4026 * Managed pci_remap_cfgspace(). Map is automatically unmapped on driver
4027 * detach.
4028 */
4029void __iomem *devm_pci_remap_cfgspace(struct device *dev,
4030 resource_size_t offset,
4031 resource_size_t size)
4032{
4033 void __iomem **ptr, *addr;
4034
4035 ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
4036 if (!ptr)
4037 return NULL;
4038
4039 addr = pci_remap_cfgspace(offset, size);
4040 if (addr) {
4041 *ptr = addr;
4042 devres_add(dev, ptr);
4043 } else
4044 devres_free(ptr);
4045
4046 return addr;
4047}
4048EXPORT_SYMBOL(devm_pci_remap_cfgspace);
4049
4050/**
4051 * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
4052 * @dev: generic device to handle the resource for
4053 * @res: configuration space resource to be handled
4054 *
4055 * Checks that a resource is a valid memory region, requests the memory
4056 * region and ioremaps with pci_remap_cfgspace() API that ensures the
4057 * proper PCI configuration space memory attributes are guaranteed.
4058 *
4059 * All operations are managed and will be undone on driver detach.
4060 *
4061 * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
4062 * on failure. Usage example::
4063 *
4064 * res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
4065 * base = devm_pci_remap_cfg_resource(&pdev->dev, res);
4066 * if (IS_ERR(base))
4067 * return PTR_ERR(base);
4068 */
4069void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
4070 struct resource *res)
4071{
4072 resource_size_t size;
4073 const char *name;
4074 void __iomem *dest_ptr;
4075
4076 BUG_ON(!dev);
4077
4078 if (!res || resource_type(res) != IORESOURCE_MEM) {
4079 dev_err(dev, "invalid resource\n");
4080 return IOMEM_ERR_PTR(-EINVAL);
4081 }
4082
4083 size = resource_size(res);
4084 name = res->name ?: dev_name(dev);
4085
4086 if (!devm_request_mem_region(dev, res->start, size, name)) {
4087 dev_err(dev, "can't request region for resource %pR\n", res);
4088 return IOMEM_ERR_PTR(-EBUSY);
4089 }
4090
4091 dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
4092 if (!dest_ptr) {
4093 dev_err(dev, "ioremap failed for resource %pR\n", res);
4094 devm_release_mem_region(dev, res->start, size);
4095 dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
4096 }
4097
4098 return dest_ptr;
4099}
4100EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
4101
4102static void __pci_set_master(struct pci_dev *dev, bool enable)
4103{
4104 u16 old_cmd, cmd;
4105
4106 pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
4107 if (enable)
4108 cmd = old_cmd | PCI_COMMAND_MASTER;
4109 else
4110 cmd = old_cmd & ~PCI_COMMAND_MASTER;
4111 if (cmd != old_cmd) {
4112 pci_dbg(dev, "%s bus mastering\n",
4113 enable ? "enabling" : "disabling");
4114 pci_write_config_word(dev, PCI_COMMAND, cmd);
4115 }
4116 dev->is_busmaster = enable;
4117}
4118
4119/**
4120 * pcibios_setup - process "pci=" kernel boot arguments
4121 * @str: string used to pass in "pci=" kernel boot arguments
4122 *
4123 * Process kernel boot arguments. This is the default implementation.
4124 * Architecture specific implementations can override this as necessary.
4125 */
4126char * __weak __init pcibios_setup(char *str)
4127{
4128 return str;
4129}
4130
4131/**
4132 * pcibios_set_master - enable PCI bus-mastering for device dev
4133 * @dev: the PCI device to enable
4134 *
4135 * Enables PCI bus-mastering for the device. This is the default
4136 * implementation. Architecture specific implementations can override
4137 * this if necessary.
4138 */
4139void __weak pcibios_set_master(struct pci_dev *dev)
4140{
4141 u8 lat;
4142
4143 /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
4144 if (pci_is_pcie(dev))
4145 return;
4146
4147 pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
4148 if (lat < 16)
4149 lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
4150 else if (lat > pcibios_max_latency)
4151 lat = pcibios_max_latency;
4152 else
4153 return;
4154
4155 pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
4156}
4157
4158/**
4159 * pci_set_master - enables bus-mastering for device dev
4160 * @dev: the PCI device to enable
4161 *
4162 * Enables bus-mastering on the device and calls pcibios_set_master()
4163 * to do the needed arch specific settings.
4164 */
4165void pci_set_master(struct pci_dev *dev)
4166{
4167 __pci_set_master(dev, true);
4168 pcibios_set_master(dev);
4169}
4170EXPORT_SYMBOL(pci_set_master);
4171
4172/**
4173 * pci_clear_master - disables bus-mastering for device dev
4174 * @dev: the PCI device to disable
4175 */
4176void pci_clear_master(struct pci_dev *dev)
4177{
4178 __pci_set_master(dev, false);
4179}
4180EXPORT_SYMBOL(pci_clear_master);
4181
4182/**
4183 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
4184 * @dev: the PCI device for which MWI is to be enabled
4185 *
4186 * Helper function for pci_set_mwi.
4187 * Originally copied from drivers/net/acenic.c.
4188 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
4189 *
4190 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4191 */
4192int pci_set_cacheline_size(struct pci_dev *dev)
4193{
4194 u8 cacheline_size;
4195
4196 if (!pci_cache_line_size)
4197 return -EINVAL;
4198
4199 /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
4200 equal to or multiple of the right value. */
4201 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4202 if (cacheline_size >= pci_cache_line_size &&
4203 (cacheline_size % pci_cache_line_size) == 0)
4204 return 0;
4205
4206 /* Write the correct value. */
4207 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
4208 /* Read it back. */
4209 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4210 if (cacheline_size == pci_cache_line_size)
4211 return 0;
4212
4213 pci_info(dev, "cache line size of %d is not supported\n",
4214 pci_cache_line_size << 2);
4215
4216 return -EINVAL;
4217}
4218EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
4219
4220/**
4221 * pci_set_mwi - enables memory-write-invalidate PCI transaction
4222 * @dev: the PCI device for which MWI is enabled
4223 *
4224 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4225 *
4226 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4227 */
4228int pci_set_mwi(struct pci_dev *dev)
4229{
4230#ifdef PCI_DISABLE_MWI
4231 return 0;
4232#else
4233 int rc;
4234 u16 cmd;
4235
4236 rc = pci_set_cacheline_size(dev);
4237 if (rc)
4238 return rc;
4239
4240 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4241 if (!(cmd & PCI_COMMAND_INVALIDATE)) {
4242 pci_dbg(dev, "enabling Mem-Wr-Inval\n");
4243 cmd |= PCI_COMMAND_INVALIDATE;
4244 pci_write_config_word(dev, PCI_COMMAND, cmd);
4245 }
4246 return 0;
4247#endif
4248}
4249EXPORT_SYMBOL(pci_set_mwi);
4250
4251/**
4252 * pcim_set_mwi - a device-managed pci_set_mwi()
4253 * @dev: the PCI device for which MWI is enabled
4254 *
4255 * Managed pci_set_mwi().
4256 *
4257 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4258 */
4259int pcim_set_mwi(struct pci_dev *dev)
4260{
4261 struct pci_devres *dr;
4262
4263 dr = find_pci_dr(dev);
4264 if (!dr)
4265 return -ENOMEM;
4266
4267 dr->mwi = 1;
4268 return pci_set_mwi(dev);
4269}
4270EXPORT_SYMBOL(pcim_set_mwi);
4271
4272/**
4273 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
4274 * @dev: the PCI device for which MWI is enabled
4275 *
4276 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4277 * Callers are not required to check the return value.
4278 *
4279 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4280 */
4281int pci_try_set_mwi(struct pci_dev *dev)
4282{
4283#ifdef PCI_DISABLE_MWI
4284 return 0;
4285#else
4286 return pci_set_mwi(dev);
4287#endif
4288}
4289EXPORT_SYMBOL(pci_try_set_mwi);
4290
4291/**
4292 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
4293 * @dev: the PCI device to disable
4294 *
4295 * Disables PCI Memory-Write-Invalidate transaction on the device
4296 */
4297void pci_clear_mwi(struct pci_dev *dev)
4298{
4299#ifndef PCI_DISABLE_MWI
4300 u16 cmd;
4301
4302 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4303 if (cmd & PCI_COMMAND_INVALIDATE) {
4304 cmd &= ~PCI_COMMAND_INVALIDATE;
4305 pci_write_config_word(dev, PCI_COMMAND, cmd);
4306 }
4307#endif
4308}
4309EXPORT_SYMBOL(pci_clear_mwi);
4310
4311/**
4312 * pci_intx - enables/disables PCI INTx for device dev
4313 * @pdev: the PCI device to operate on
4314 * @enable: boolean: whether to enable or disable PCI INTx
4315 *
4316 * Enables/disables PCI INTx for device @pdev
4317 */
4318void pci_intx(struct pci_dev *pdev, int enable)
4319{
4320 u16 pci_command, new;
4321
4322 pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
4323
4324 if (enable)
4325 new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
4326 else
4327 new = pci_command | PCI_COMMAND_INTX_DISABLE;
4328
4329 if (new != pci_command) {
4330 struct pci_devres *dr;
4331
4332 pci_write_config_word(pdev, PCI_COMMAND, new);
4333
4334 dr = find_pci_dr(pdev);
4335 if (dr && !dr->restore_intx) {
4336 dr->restore_intx = 1;
4337 dr->orig_intx = !enable;
4338 }
4339 }
4340}
4341EXPORT_SYMBOL_GPL(pci_intx);
4342
4343static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
4344{
4345 struct pci_bus *bus = dev->bus;
4346 bool mask_updated = true;
4347 u32 cmd_status_dword;
4348 u16 origcmd, newcmd;
4349 unsigned long flags;
4350 bool irq_pending;
4351
4352 /*
4353 * We do a single dword read to retrieve both command and status.
4354 * Document assumptions that make this possible.
4355 */
4356 BUILD_BUG_ON(PCI_COMMAND % 4);
4357 BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
4358
4359 raw_spin_lock_irqsave(&pci_lock, flags);
4360
4361 bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
4362
4363 irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
4364
4365 /*
4366 * Check interrupt status register to see whether our device
4367 * triggered the interrupt (when masking) or the next IRQ is
4368 * already pending (when unmasking).
4369 */
4370 if (mask != irq_pending) {
4371 mask_updated = false;
4372 goto done;
4373 }
4374
4375 origcmd = cmd_status_dword;
4376 newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
4377 if (mask)
4378 newcmd |= PCI_COMMAND_INTX_DISABLE;
4379 if (newcmd != origcmd)
4380 bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
4381
4382done:
4383 raw_spin_unlock_irqrestore(&pci_lock, flags);
4384
4385 return mask_updated;
4386}
4387
4388/**
4389 * pci_check_and_mask_intx - mask INTx on pending interrupt
4390 * @dev: the PCI device to operate on
4391 *
4392 * Check if the device dev has its INTx line asserted, mask it and return
4393 * true in that case. False is returned if no interrupt was pending.
4394 */
4395bool pci_check_and_mask_intx(struct pci_dev *dev)
4396{
4397 return pci_check_and_set_intx_mask(dev, true);
4398}
4399EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
4400
4401/**
4402 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
4403 * @dev: the PCI device to operate on
4404 *
4405 * Check if the device dev has its INTx line asserted, unmask it if not and
4406 * return true. False is returned and the mask remains active if there was
4407 * still an interrupt pending.
4408 */
4409bool pci_check_and_unmask_intx(struct pci_dev *dev)
4410{
4411 return pci_check_and_set_intx_mask(dev, false);
4412}
4413EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
4414
4415/**
4416 * pci_wait_for_pending_transaction - wait for pending transaction
4417 * @dev: the PCI device to operate on
4418 *
4419 * Return 0 if transaction is pending 1 otherwise.
4420 */
4421int pci_wait_for_pending_transaction(struct pci_dev *dev)
4422{
4423 if (!pci_is_pcie(dev))
4424 return 1;
4425
4426 return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
4427 PCI_EXP_DEVSTA_TRPND);
4428}
4429EXPORT_SYMBOL(pci_wait_for_pending_transaction);
4430
4431static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
4432{
4433 int delay = 1;
4434 u32 id;
4435
4436 /*
4437 * After reset, the device should not silently discard config
4438 * requests, but it may still indicate that it needs more time by
4439 * responding to them with CRS completions. The Root Port will
4440 * generally synthesize ~0 data to complete the read (except when
4441 * CRS SV is enabled and the read was for the Vendor ID; in that
4442 * case it synthesizes 0x0001 data).
4443 *
4444 * Wait for the device to return a non-CRS completion. Read the
4445 * Command register instead of Vendor ID so we don't have to
4446 * contend with the CRS SV value.
4447 */
4448 pci_read_config_dword(dev, PCI_COMMAND, &id);
4449 while (id == ~0) {
4450 if (delay > timeout) {
4451 pci_warn(dev, "not ready %dms after %s; giving up\n",
4452 delay - 1, reset_type);
4453 return -ENOTTY;
4454 }
4455
4456 if (delay > 1000)
4457 pci_info(dev, "not ready %dms after %s; waiting\n",
4458 delay - 1, reset_type);
4459
4460 msleep(delay);
4461 delay *= 2;
4462 pci_read_config_dword(dev, PCI_COMMAND, &id);
4463 }
4464
4465 if (delay > 1000)
4466 pci_info(dev, "ready %dms after %s\n", delay - 1,
4467 reset_type);
4468
4469 return 0;
4470}
4471
4472/**
4473 * pcie_has_flr - check if a device supports function level resets
4474 * @dev: device to check
4475 *
4476 * Returns true if the device advertises support for PCIe function level
4477 * resets.
4478 */
4479bool pcie_has_flr(struct pci_dev *dev)
4480{
4481 u32 cap;
4482
4483 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4484 return false;
4485
4486 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
4487 return cap & PCI_EXP_DEVCAP_FLR;
4488}
4489EXPORT_SYMBOL_GPL(pcie_has_flr);
4490
4491/**
4492 * pcie_flr - initiate a PCIe function level reset
4493 * @dev: device to reset
4494 *
4495 * Initiate a function level reset on @dev. The caller should ensure the
4496 * device supports FLR before calling this function, e.g. by using the
4497 * pcie_has_flr() helper.
4498 */
4499int pcie_flr(struct pci_dev *dev)
4500{
4501 if (!pci_wait_for_pending_transaction(dev))
4502 pci_err(dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
4503
4504 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
4505
4506 if (dev->imm_ready)
4507 return 0;
4508
4509 /*
4510 * Per PCIe r4.0, sec 6.6.2, a device must complete an FLR within
4511 * 100ms, but may silently discard requests while the FLR is in
4512 * progress. Wait 100ms before trying to access the device.
4513 */
4514 msleep(100);
4515
4516 return pci_dev_wait(dev, "FLR", PCIE_RESET_READY_POLL_MS);
4517}
4518EXPORT_SYMBOL_GPL(pcie_flr);
4519
4520static int pci_af_flr(struct pci_dev *dev, int probe)
4521{
4522 int pos;
4523 u8 cap;
4524
4525 pos = pci_find_capability(dev, PCI_CAP_ID_AF);
4526 if (!pos)
4527 return -ENOTTY;
4528
4529 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4530 return -ENOTTY;
4531
4532 pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
4533 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
4534 return -ENOTTY;
4535
4536 if (probe)
4537 return 0;
4538
4539 /*
4540 * Wait for Transaction Pending bit to clear. A word-aligned test
4541 * is used, so we use the control offset rather than status and shift
4542 * the test bit to match.
4543 */
4544 if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
4545 PCI_AF_STATUS_TP << 8))
4546 pci_err(dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
4547
4548 pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
4549
4550 if (dev->imm_ready)
4551 return 0;
4552
4553 /*
4554 * Per Advanced Capabilities for Conventional PCI ECN, 13 April 2006,
4555 * updated 27 July 2006; a device must complete an FLR within
4556 * 100ms, but may silently discard requests while the FLR is in
4557 * progress. Wait 100ms before trying to access the device.
4558 */
4559 msleep(100);
4560
4561 return pci_dev_wait(dev, "AF_FLR", PCIE_RESET_READY_POLL_MS);
4562}
4563
4564/**
4565 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
4566 * @dev: Device to reset.
4567 * @probe: If set, only check if the device can be reset this way.
4568 *
4569 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
4570 * unset, it will be reinitialized internally when going from PCI_D3hot to
4571 * PCI_D0. If that's the case and the device is not in a low-power state
4572 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
4573 *
4574 * NOTE: This causes the caller to sleep for twice the device power transition
4575 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
4576 * by default (i.e. unless the @dev's d3_delay field has a different value).
4577 * Moreover, only devices in D0 can be reset by this function.
4578 */
4579static int pci_pm_reset(struct pci_dev *dev, int probe)
4580{
4581 u16 csr;
4582
4583 if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
4584 return -ENOTTY;
4585
4586 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
4587 if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
4588 return -ENOTTY;
4589
4590 if (probe)
4591 return 0;
4592
4593 if (dev->current_state != PCI_D0)
4594 return -EINVAL;
4595
4596 csr &= ~PCI_PM_CTRL_STATE_MASK;
4597 csr |= PCI_D3hot;
4598 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4599 pci_dev_d3_sleep(dev);
4600
4601 csr &= ~PCI_PM_CTRL_STATE_MASK;
4602 csr |= PCI_D0;
4603 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4604 pci_dev_d3_sleep(dev);
4605
4606 return pci_dev_wait(dev, "PM D3->D0", PCIE_RESET_READY_POLL_MS);
4607}
4608/**
4609 * pcie_wait_for_link - Wait until link is active or inactive
4610 * @pdev: Bridge device
4611 * @active: waiting for active or inactive?
4612 *
4613 * Use this to wait till link becomes active or inactive.
4614 */
4615bool pcie_wait_for_link(struct pci_dev *pdev, bool active)
4616{
4617 int timeout = 1000;
4618 bool ret;
4619 u16 lnk_status;
4620
4621 /*
4622 * Some controllers might not implement link active reporting. In this
4623 * case, we wait for 1000 + 100 ms.
4624 */
4625 if (!pdev->link_active_reporting) {
4626 msleep(1100);
4627 return true;
4628 }
4629
4630 /*
4631 * PCIe r4.0 sec 6.6.1, a component must enter LTSSM Detect within 20ms,
4632 * after which we should expect an link active if the reset was
4633 * successful. If so, software must wait a minimum 100ms before sending
4634 * configuration requests to devices downstream this port.
4635 *
4636 * If the link fails to activate, either the device was physically
4637 * removed or the link is permanently failed.
4638 */
4639 if (active)
4640 msleep(20);
4641 for (;;) {
4642 pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
4643 ret = !!(lnk_status & PCI_EXP_LNKSTA_DLLLA);
4644 if (ret == active)
4645 break;
4646 if (timeout <= 0)
4647 break;
4648 msleep(10);
4649 timeout -= 10;
4650 }
4651 if (active && ret)
4652 msleep(100);
4653 else if (ret != active)
4654 pci_info(pdev, "Data Link Layer Link Active not %s in 1000 msec\n",
4655 active ? "set" : "cleared");
4656 return ret == active;
4657}
4658
4659void pci_reset_secondary_bus(struct pci_dev *dev)
4660{
4661 u16 ctrl;
4662
4663 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
4664 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
4665 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4666
4667 /*
4668 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms. Double
4669 * this to 2ms to ensure that we meet the minimum requirement.
4670 */
4671 msleep(2);
4672
4673 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4674 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4675
4676 /*
4677 * Trhfa for conventional PCI is 2^25 clock cycles.
4678 * Assuming a minimum 33MHz clock this results in a 1s
4679 * delay before we can consider subordinate devices to
4680 * be re-initialized. PCIe has some ways to shorten this,
4681 * but we don't make use of them yet.
4682 */
4683 ssleep(1);
4684}
4685
4686void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4687{
4688 pci_reset_secondary_bus(dev);
4689}
4690
4691/**
4692 * pci_bridge_secondary_bus_reset - Reset the secondary bus on a PCI bridge.
4693 * @dev: Bridge device
4694 *
4695 * Use the bridge control register to assert reset on the secondary bus.
4696 * Devices on the secondary bus are left in power-on state.
4697 */
4698int pci_bridge_secondary_bus_reset(struct pci_dev *dev)
4699{
4700 pcibios_reset_secondary_bus(dev);
4701
4702 return pci_dev_wait(dev, "bus reset", PCIE_RESET_READY_POLL_MS);
4703}
4704EXPORT_SYMBOL_GPL(pci_bridge_secondary_bus_reset);
4705
4706static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
4707{
4708 struct pci_dev *pdev;
4709
4710 if (pci_is_root_bus(dev->bus) || dev->subordinate ||
4711 !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4712 return -ENOTTY;
4713
4714 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4715 if (pdev != dev)
4716 return -ENOTTY;
4717
4718 if (probe)
4719 return 0;
4720
4721 return pci_bridge_secondary_bus_reset(dev->bus->self);
4722}
4723
4724static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
4725{
4726 int rc = -ENOTTY;
4727
4728 if (!hotplug || !try_module_get(hotplug->owner))
4729 return rc;
4730
4731 if (hotplug->ops->reset_slot)
4732 rc = hotplug->ops->reset_slot(hotplug, probe);
4733
4734 module_put(hotplug->owner);
4735
4736 return rc;
4737}
4738
4739static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
4740{
4741 struct pci_dev *pdev;
4742
4743 if (dev->subordinate || !dev->slot ||
4744 dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4745 return -ENOTTY;
4746
4747 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4748 if (pdev != dev && pdev->slot == dev->slot)
4749 return -ENOTTY;
4750
4751 return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
4752}
4753
4754static void pci_dev_lock(struct pci_dev *dev)
4755{
4756 pci_cfg_access_lock(dev);
4757 /* block PM suspend, driver probe, etc. */
4758 device_lock(&dev->dev);
4759}
4760
4761/* Return 1 on successful lock, 0 on contention */
4762static int pci_dev_trylock(struct pci_dev *dev)
4763{
4764 if (pci_cfg_access_trylock(dev)) {
4765 if (device_trylock(&dev->dev))
4766 return 1;
4767 pci_cfg_access_unlock(dev);
4768 }
4769
4770 return 0;
4771}
4772
4773static void pci_dev_unlock(struct pci_dev *dev)
4774{
4775 device_unlock(&dev->dev);
4776 pci_cfg_access_unlock(dev);
4777}
4778
4779static void pci_dev_save_and_disable(struct pci_dev *dev)
4780{
4781 const struct pci_error_handlers *err_handler =
4782 dev->driver ? dev->driver->err_handler : NULL;
4783
4784 /*
4785 * dev->driver->err_handler->reset_prepare() is protected against
4786 * races with ->remove() by the device lock, which must be held by
4787 * the caller.
4788 */
4789 if (err_handler && err_handler->reset_prepare)
4790 err_handler->reset_prepare(dev);
4791
4792 /*
4793 * Wake-up device prior to save. PM registers default to D0 after
4794 * reset and a simple register restore doesn't reliably return
4795 * to a non-D0 state anyway.
4796 */
4797 pci_set_power_state(dev, PCI_D0);
4798
4799 pci_save_state(dev);
4800 /*
4801 * Disable the device by clearing the Command register, except for
4802 * INTx-disable which is set. This not only disables MMIO and I/O port
4803 * BARs, but also prevents the device from being Bus Master, preventing
4804 * DMA from the device including MSI/MSI-X interrupts. For PCI 2.3
4805 * compliant devices, INTx-disable prevents legacy interrupts.
4806 */
4807 pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
4808}
4809
4810static void pci_dev_restore(struct pci_dev *dev)
4811{
4812 const struct pci_error_handlers *err_handler =
4813 dev->driver ? dev->driver->err_handler : NULL;
4814
4815 pci_restore_state(dev);
4816
4817 /*
4818 * dev->driver->err_handler->reset_done() is protected against
4819 * races with ->remove() by the device lock, which must be held by
4820 * the caller.
4821 */
4822 if (err_handler && err_handler->reset_done)
4823 err_handler->reset_done(dev);
4824}
4825
4826/**
4827 * __pci_reset_function_locked - reset a PCI device function while holding
4828 * the @dev mutex lock.
4829 * @dev: PCI device to reset
4830 *
4831 * Some devices allow an individual function to be reset without affecting
4832 * other functions in the same device. The PCI device must be responsive
4833 * to PCI config space in order to use this function.
4834 *
4835 * The device function is presumed to be unused and the caller is holding
4836 * the device mutex lock when this function is called.
4837 *
4838 * Resetting the device will make the contents of PCI configuration space
4839 * random, so any caller of this must be prepared to reinitialise the
4840 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4841 * etc.
4842 *
4843 * Returns 0 if the device function was successfully reset or negative if the
4844 * device doesn't support resetting a single function.
4845 */
4846int __pci_reset_function_locked(struct pci_dev *dev)
4847{
4848 int rc;
4849
4850 might_sleep();
4851
4852 /*
4853 * A reset method returns -ENOTTY if it doesn't support this device
4854 * and we should try the next method.
4855 *
4856 * If it returns 0 (success), we're finished. If it returns any
4857 * other error, we're also finished: this indicates that further
4858 * reset mechanisms might be broken on the device.
4859 */
4860 rc = pci_dev_specific_reset(dev, 0);
4861 if (rc != -ENOTTY)
4862 return rc;
4863 if (pcie_has_flr(dev)) {
4864 rc = pcie_flr(dev);
4865 if (rc != -ENOTTY)
4866 return rc;
4867 }
4868 rc = pci_af_flr(dev, 0);
4869 if (rc != -ENOTTY)
4870 return rc;
4871 rc = pci_pm_reset(dev, 0);
4872 if (rc != -ENOTTY)
4873 return rc;
4874 rc = pci_dev_reset_slot_function(dev, 0);
4875 if (rc != -ENOTTY)
4876 return rc;
4877 return pci_parent_bus_reset(dev, 0);
4878}
4879EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
4880
4881/**
4882 * pci_probe_reset_function - check whether the device can be safely reset
4883 * @dev: PCI device to reset
4884 *
4885 * Some devices allow an individual function to be reset without affecting
4886 * other functions in the same device. The PCI device must be responsive
4887 * to PCI config space in order to use this function.
4888 *
4889 * Returns 0 if the device function can be reset or negative if the
4890 * device doesn't support resetting a single function.
4891 */
4892int pci_probe_reset_function(struct pci_dev *dev)
4893{
4894 int rc;
4895
4896 might_sleep();
4897
4898 rc = pci_dev_specific_reset(dev, 1);
4899 if (rc != -ENOTTY)
4900 return rc;
4901 if (pcie_has_flr(dev))
4902 return 0;
4903 rc = pci_af_flr(dev, 1);
4904 if (rc != -ENOTTY)
4905 return rc;
4906 rc = pci_pm_reset(dev, 1);
4907 if (rc != -ENOTTY)
4908 return rc;
4909 rc = pci_dev_reset_slot_function(dev, 1);
4910 if (rc != -ENOTTY)
4911 return rc;
4912
4913 return pci_parent_bus_reset(dev, 1);
4914}
4915
4916/**
4917 * pci_reset_function - quiesce and reset a PCI device function
4918 * @dev: PCI device to reset
4919 *
4920 * Some devices allow an individual function to be reset without affecting
4921 * other functions in the same device. The PCI device must be responsive
4922 * to PCI config space in order to use this function.
4923 *
4924 * This function does not just reset the PCI portion of a device, but
4925 * clears all the state associated with the device. This function differs
4926 * from __pci_reset_function_locked() in that it saves and restores device state
4927 * over the reset and takes the PCI device lock.
4928 *
4929 * Returns 0 if the device function was successfully reset or negative if the
4930 * device doesn't support resetting a single function.
4931 */
4932int pci_reset_function(struct pci_dev *dev)
4933{
4934 int rc;
4935
4936 if (!dev->reset_fn)
4937 return -ENOTTY;
4938
4939 pci_dev_lock(dev);
4940 pci_dev_save_and_disable(dev);
4941
4942 rc = __pci_reset_function_locked(dev);
4943
4944 pci_dev_restore(dev);
4945 pci_dev_unlock(dev);
4946
4947 return rc;
4948}
4949EXPORT_SYMBOL_GPL(pci_reset_function);
4950
4951/**
4952 * pci_reset_function_locked - quiesce and reset a PCI device function
4953 * @dev: PCI device to reset
4954 *
4955 * Some devices allow an individual function to be reset without affecting
4956 * other functions in the same device. The PCI device must be responsive
4957 * to PCI config space in order to use this function.
4958 *
4959 * This function does not just reset the PCI portion of a device, but
4960 * clears all the state associated with the device. This function differs
4961 * from __pci_reset_function_locked() in that it saves and restores device state
4962 * over the reset. It also differs from pci_reset_function() in that it
4963 * requires the PCI device lock to be held.
4964 *
4965 * Returns 0 if the device function was successfully reset or negative if the
4966 * device doesn't support resetting a single function.
4967 */
4968int pci_reset_function_locked(struct pci_dev *dev)
4969{
4970 int rc;
4971
4972 if (!dev->reset_fn)
4973 return -ENOTTY;
4974
4975 pci_dev_save_and_disable(dev);
4976
4977 rc = __pci_reset_function_locked(dev);
4978
4979 pci_dev_restore(dev);
4980
4981 return rc;
4982}
4983EXPORT_SYMBOL_GPL(pci_reset_function_locked);
4984
4985/**
4986 * pci_try_reset_function - quiesce and reset a PCI device function
4987 * @dev: PCI device to reset
4988 *
4989 * Same as above, except return -EAGAIN if unable to lock device.
4990 */
4991int pci_try_reset_function(struct pci_dev *dev)
4992{
4993 int rc;
4994
4995 if (!dev->reset_fn)
4996 return -ENOTTY;
4997
4998 if (!pci_dev_trylock(dev))
4999 return -EAGAIN;
5000
5001 pci_dev_save_and_disable(dev);
5002 rc = __pci_reset_function_locked(dev);
5003 pci_dev_restore(dev);
5004 pci_dev_unlock(dev);
5005
5006 return rc;
5007}
5008EXPORT_SYMBOL_GPL(pci_try_reset_function);
5009
5010/* Do any devices on or below this bus prevent a bus reset? */
5011static bool pci_bus_resetable(struct pci_bus *bus)
5012{
5013 struct pci_dev *dev;
5014
5015
5016 if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5017 return false;
5018
5019 list_for_each_entry(dev, &bus->devices, bus_list) {
5020 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5021 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
5022 return false;
5023 }
5024
5025 return true;
5026}
5027
5028/* Lock devices from the top of the tree down */
5029static void pci_bus_lock(struct pci_bus *bus)
5030{
5031 struct pci_dev *dev;
5032
5033 list_for_each_entry(dev, &bus->devices, bus_list) {
5034 pci_dev_lock(dev);
5035 if (dev->subordinate)
5036 pci_bus_lock(dev->subordinate);
5037 }
5038}
5039
5040/* Unlock devices from the bottom of the tree up */
5041static void pci_bus_unlock(struct pci_bus *bus)
5042{
5043 struct pci_dev *dev;
5044
5045 list_for_each_entry(dev, &bus->devices, bus_list) {
5046 if (dev->subordinate)
5047 pci_bus_unlock(dev->subordinate);
5048 pci_dev_unlock(dev);
5049 }
5050}
5051
5052/* Return 1 on successful lock, 0 on contention */
5053static int pci_bus_trylock(struct pci_bus *bus)
5054{
5055 struct pci_dev *dev;
5056
5057 list_for_each_entry(dev, &bus->devices, bus_list) {
5058 if (!pci_dev_trylock(dev))
5059 goto unlock;
5060 if (dev->subordinate) {
5061 if (!pci_bus_trylock(dev->subordinate)) {
5062 pci_dev_unlock(dev);
5063 goto unlock;
5064 }
5065 }
5066 }
5067 return 1;
5068
5069unlock:
5070 list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
5071 if (dev->subordinate)
5072 pci_bus_unlock(dev->subordinate);
5073 pci_dev_unlock(dev);
5074 }
5075 return 0;
5076}
5077
5078/* Do any devices on or below this slot prevent a bus reset? */
5079static bool pci_slot_resetable(struct pci_slot *slot)
5080{
5081 struct pci_dev *dev;
5082
5083 if (slot->bus->self &&
5084 (slot->bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5085 return false;
5086
5087 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5088 if (!dev->slot || dev->slot != slot)
5089 continue;
5090 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5091 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
5092 return false;
5093 }
5094
5095 return true;
5096}
5097
5098/* Lock devices from the top of the tree down */
5099static void pci_slot_lock(struct pci_slot *slot)
5100{
5101 struct pci_dev *dev;
5102
5103 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5104 if (!dev->slot || dev->slot != slot)
5105 continue;
5106 pci_dev_lock(dev);
5107 if (dev->subordinate)
5108 pci_bus_lock(dev->subordinate);
5109 }
5110}
5111
5112/* Unlock devices from the bottom of the tree up */
5113static void pci_slot_unlock(struct pci_slot *slot)
5114{
5115 struct pci_dev *dev;
5116
5117 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5118 if (!dev->slot || dev->slot != slot)
5119 continue;
5120 if (dev->subordinate)
5121 pci_bus_unlock(dev->subordinate);
5122 pci_dev_unlock(dev);
5123 }
5124}
5125
5126/* Return 1 on successful lock, 0 on contention */
5127static int pci_slot_trylock(struct pci_slot *slot)
5128{
5129 struct pci_dev *dev;
5130
5131 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5132 if (!dev->slot || dev->slot != slot)
5133 continue;
5134 if (!pci_dev_trylock(dev))
5135 goto unlock;
5136 if (dev->subordinate) {
5137 if (!pci_bus_trylock(dev->subordinate)) {
5138 pci_dev_unlock(dev);
5139 goto unlock;
5140 }
5141 }
5142 }
5143 return 1;
5144
5145unlock:
5146 list_for_each_entry_continue_reverse(dev,
5147 &slot->bus->devices, bus_list) {
5148 if (!dev->slot || dev->slot != slot)
5149 continue;
5150 if (dev->subordinate)
5151 pci_bus_unlock(dev->subordinate);
5152 pci_dev_unlock(dev);
5153 }
5154 return 0;
5155}
5156
5157/*
5158 * Save and disable devices from the top of the tree down while holding
5159 * the @dev mutex lock for the entire tree.
5160 */
5161static void pci_bus_save_and_disable_locked(struct pci_bus *bus)
5162{
5163 struct pci_dev *dev;
5164
5165 list_for_each_entry(dev, &bus->devices, bus_list) {
5166 pci_dev_save_and_disable(dev);
5167 if (dev->subordinate)
5168 pci_bus_save_and_disable_locked(dev->subordinate);
5169 }
5170}
5171
5172/*
5173 * Restore devices from top of the tree down while holding @dev mutex lock
5174 * for the entire tree. Parent bridges need to be restored before we can
5175 * get to subordinate devices.
5176 */
5177static void pci_bus_restore_locked(struct pci_bus *bus)
5178{
5179 struct pci_dev *dev;
5180
5181 list_for_each_entry(dev, &bus->devices, bus_list) {
5182 pci_dev_restore(dev);
5183 if (dev->subordinate)
5184 pci_bus_restore_locked(dev->subordinate);
5185 }
5186}
5187
5188/*
5189 * Save and disable devices from the top of the tree down while holding
5190 * the @dev mutex lock for the entire tree.
5191 */
5192static void pci_slot_save_and_disable_locked(struct pci_slot *slot)
5193{
5194 struct pci_dev *dev;
5195
5196 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5197 if (!dev->slot || dev->slot != slot)
5198 continue;
5199 pci_dev_save_and_disable(dev);
5200 if (dev->subordinate)
5201 pci_bus_save_and_disable_locked(dev->subordinate);
5202 }
5203}
5204
5205/*
5206 * Restore devices from top of the tree down while holding @dev mutex lock
5207 * for the entire tree. Parent bridges need to be restored before we can
5208 * get to subordinate devices.
5209 */
5210static void pci_slot_restore_locked(struct pci_slot *slot)
5211{
5212 struct pci_dev *dev;
5213
5214 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5215 if (!dev->slot || dev->slot != slot)
5216 continue;
5217 pci_dev_restore(dev);
5218 if (dev->subordinate)
5219 pci_bus_restore_locked(dev->subordinate);
5220 }
5221}
5222
5223static int pci_slot_reset(struct pci_slot *slot, int probe)
5224{
5225 int rc;
5226
5227 if (!slot || !pci_slot_resetable(slot))
5228 return -ENOTTY;
5229
5230 if (!probe)
5231 pci_slot_lock(slot);
5232
5233 might_sleep();
5234
5235 rc = pci_reset_hotplug_slot(slot->hotplug, probe);
5236
5237 if (!probe)
5238 pci_slot_unlock(slot);
5239
5240 return rc;
5241}
5242
5243/**
5244 * pci_probe_reset_slot - probe whether a PCI slot can be reset
5245 * @slot: PCI slot to probe
5246 *
5247 * Return 0 if slot can be reset, negative if a slot reset is not supported.
5248 */
5249int pci_probe_reset_slot(struct pci_slot *slot)
5250{
5251 return pci_slot_reset(slot, 1);
5252}
5253EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
5254
5255/**
5256 * __pci_reset_slot - Try to reset a PCI slot
5257 * @slot: PCI slot to reset
5258 *
5259 * A PCI bus may host multiple slots, each slot may support a reset mechanism
5260 * independent of other slots. For instance, some slots may support slot power
5261 * control. In the case of a 1:1 bus to slot architecture, this function may
5262 * wrap the bus reset to avoid spurious slot related events such as hotplug.
5263 * Generally a slot reset should be attempted before a bus reset. All of the
5264 * function of the slot and any subordinate buses behind the slot are reset
5265 * through this function. PCI config space of all devices in the slot and
5266 * behind the slot is saved before and restored after reset.
5267 *
5268 * Same as above except return -EAGAIN if the slot cannot be locked
5269 */
5270static int __pci_reset_slot(struct pci_slot *slot)
5271{
5272 int rc;
5273
5274 rc = pci_slot_reset(slot, 1);
5275 if (rc)
5276 return rc;
5277
5278 if (pci_slot_trylock(slot)) {
5279 pci_slot_save_and_disable_locked(slot);
5280 might_sleep();
5281 rc = pci_reset_hotplug_slot(slot->hotplug, 0);
5282 pci_slot_restore_locked(slot);
5283 pci_slot_unlock(slot);
5284 } else
5285 rc = -EAGAIN;
5286
5287 return rc;
5288}
5289
5290static int pci_bus_reset(struct pci_bus *bus, int probe)
5291{
5292 int ret;
5293
5294 if (!bus->self || !pci_bus_resetable(bus))
5295 return -ENOTTY;
5296
5297 if (probe)
5298 return 0;
5299
5300 pci_bus_lock(bus);
5301
5302 might_sleep();
5303
5304 ret = pci_bridge_secondary_bus_reset(bus->self);
5305
5306 pci_bus_unlock(bus);
5307
5308 return ret;
5309}
5310
5311/**
5312 * pci_bus_error_reset - reset the bridge's subordinate bus
5313 * @bridge: The parent device that connects to the bus to reset
5314 *
5315 * This function will first try to reset the slots on this bus if the method is
5316 * available. If slot reset fails or is not available, this will fall back to a
5317 * secondary bus reset.
5318 */
5319int pci_bus_error_reset(struct pci_dev *bridge)
5320{
5321 struct pci_bus *bus = bridge->subordinate;
5322 struct pci_slot *slot;
5323
5324 if (!bus)
5325 return -ENOTTY;
5326
5327 mutex_lock(&pci_slot_mutex);
5328 if (list_empty(&bus->slots))
5329 goto bus_reset;
5330
5331 list_for_each_entry(slot, &bus->slots, list)
5332 if (pci_probe_reset_slot(slot))
5333 goto bus_reset;
5334
5335 list_for_each_entry(slot, &bus->slots, list)
5336 if (pci_slot_reset(slot, 0))
5337 goto bus_reset;
5338
5339 mutex_unlock(&pci_slot_mutex);
5340 return 0;
5341bus_reset:
5342 mutex_unlock(&pci_slot_mutex);
5343 return pci_bus_reset(bridge->subordinate, 0);
5344}
5345
5346/**
5347 * pci_probe_reset_bus - probe whether a PCI bus can be reset
5348 * @bus: PCI bus to probe
5349 *
5350 * Return 0 if bus can be reset, negative if a bus reset is not supported.
5351 */
5352int pci_probe_reset_bus(struct pci_bus *bus)
5353{
5354 return pci_bus_reset(bus, 1);
5355}
5356EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
5357
5358/**
5359 * __pci_reset_bus - Try to reset a PCI bus
5360 * @bus: top level PCI bus to reset
5361 *
5362 * Same as above except return -EAGAIN if the bus cannot be locked
5363 */
5364static int __pci_reset_bus(struct pci_bus *bus)
5365{
5366 int rc;
5367
5368 rc = pci_bus_reset(bus, 1);
5369 if (rc)
5370 return rc;
5371
5372 if (pci_bus_trylock(bus)) {
5373 pci_bus_save_and_disable_locked(bus);
5374 might_sleep();
5375 rc = pci_bridge_secondary_bus_reset(bus->self);
5376 pci_bus_restore_locked(bus);
5377 pci_bus_unlock(bus);
5378 } else
5379 rc = -EAGAIN;
5380
5381 return rc;
5382}
5383
5384/**
5385 * pci_reset_bus - Try to reset a PCI bus
5386 * @pdev: top level PCI device to reset via slot/bus
5387 *
5388 * Same as above except return -EAGAIN if the bus cannot be locked
5389 */
5390int pci_reset_bus(struct pci_dev *pdev)
5391{
5392 return (!pci_probe_reset_slot(pdev->slot)) ?
5393 __pci_reset_slot(pdev->slot) : __pci_reset_bus(pdev->bus);
5394}
5395EXPORT_SYMBOL_GPL(pci_reset_bus);
5396
5397/**
5398 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
5399 * @dev: PCI device to query
5400 *
5401 * Returns mmrbc: maximum designed memory read count in bytes or
5402 * appropriate error value.
5403 */
5404int pcix_get_max_mmrbc(struct pci_dev *dev)
5405{
5406 int cap;
5407 u32 stat;
5408
5409 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5410 if (!cap)
5411 return -EINVAL;
5412
5413 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5414 return -EINVAL;
5415
5416 return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
5417}
5418EXPORT_SYMBOL(pcix_get_max_mmrbc);
5419
5420/**
5421 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
5422 * @dev: PCI device to query
5423 *
5424 * Returns mmrbc: maximum memory read count in bytes or appropriate error
5425 * value.
5426 */
5427int pcix_get_mmrbc(struct pci_dev *dev)
5428{
5429 int cap;
5430 u16 cmd;
5431
5432 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5433 if (!cap)
5434 return -EINVAL;
5435
5436 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5437 return -EINVAL;
5438
5439 return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
5440}
5441EXPORT_SYMBOL(pcix_get_mmrbc);
5442
5443/**
5444 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
5445 * @dev: PCI device to query
5446 * @mmrbc: maximum memory read count in bytes
5447 * valid values are 512, 1024, 2048, 4096
5448 *
5449 * If possible sets maximum memory read byte count, some bridges have errata
5450 * that prevent this.
5451 */
5452int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
5453{
5454 int cap;
5455 u32 stat, v, o;
5456 u16 cmd;
5457
5458 if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
5459 return -EINVAL;
5460
5461 v = ffs(mmrbc) - 10;
5462
5463 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5464 if (!cap)
5465 return -EINVAL;
5466
5467 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5468 return -EINVAL;
5469
5470 if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
5471 return -E2BIG;
5472
5473 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5474 return -EINVAL;
5475
5476 o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
5477 if (o != v) {
5478 if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
5479 return -EIO;
5480
5481 cmd &= ~PCI_X_CMD_MAX_READ;
5482 cmd |= v << 2;
5483 if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
5484 return -EIO;
5485 }
5486 return 0;
5487}
5488EXPORT_SYMBOL(pcix_set_mmrbc);
5489
5490/**
5491 * pcie_get_readrq - get PCI Express read request size
5492 * @dev: PCI device to query
5493 *
5494 * Returns maximum memory read request in bytes or appropriate error value.
5495 */
5496int pcie_get_readrq(struct pci_dev *dev)
5497{
5498 u16 ctl;
5499
5500 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5501
5502 return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
5503}
5504EXPORT_SYMBOL(pcie_get_readrq);
5505
5506/**
5507 * pcie_set_readrq - set PCI Express maximum memory read request
5508 * @dev: PCI device to query
5509 * @rq: maximum memory read count in bytes
5510 * valid values are 128, 256, 512, 1024, 2048, 4096
5511 *
5512 * If possible sets maximum memory read request in bytes
5513 */
5514int pcie_set_readrq(struct pci_dev *dev, int rq)
5515{
5516 u16 v;
5517
5518 if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
5519 return -EINVAL;
5520
5521 /*
5522 * If using the "performance" PCIe config, we clamp the read rq
5523 * size to the max packet size to keep the host bridge from
5524 * generating requests larger than we can cope with.
5525 */
5526 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
5527 int mps = pcie_get_mps(dev);
5528
5529 if (mps < rq)
5530 rq = mps;
5531 }
5532
5533 v = (ffs(rq) - 8) << 12;
5534
5535 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5536 PCI_EXP_DEVCTL_READRQ, v);
5537}
5538EXPORT_SYMBOL(pcie_set_readrq);
5539
5540/**
5541 * pcie_get_mps - get PCI Express maximum payload size
5542 * @dev: PCI device to query
5543 *
5544 * Returns maximum payload size in bytes
5545 */
5546int pcie_get_mps(struct pci_dev *dev)
5547{
5548 u16 ctl;
5549
5550 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5551
5552 return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
5553}
5554EXPORT_SYMBOL(pcie_get_mps);
5555
5556/**
5557 * pcie_set_mps - set PCI Express maximum payload size
5558 * @dev: PCI device to query
5559 * @mps: maximum payload size in bytes
5560 * valid values are 128, 256, 512, 1024, 2048, 4096
5561 *
5562 * If possible sets maximum payload size
5563 */
5564int pcie_set_mps(struct pci_dev *dev, int mps)
5565{
5566 u16 v;
5567
5568 if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
5569 return -EINVAL;
5570
5571 v = ffs(mps) - 8;
5572 if (v > dev->pcie_mpss)
5573 return -EINVAL;
5574 v <<= 5;
5575
5576 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5577 PCI_EXP_DEVCTL_PAYLOAD, v);
5578}
5579EXPORT_SYMBOL(pcie_set_mps);
5580
5581/**
5582 * pcie_bandwidth_available - determine minimum link settings of a PCIe
5583 * device and its bandwidth limitation
5584 * @dev: PCI device to query
5585 * @limiting_dev: storage for device causing the bandwidth limitation
5586 * @speed: storage for speed of limiting device
5587 * @width: storage for width of limiting device
5588 *
5589 * Walk up the PCI device chain and find the point where the minimum
5590 * bandwidth is available. Return the bandwidth available there and (if
5591 * limiting_dev, speed, and width pointers are supplied) information about
5592 * that point. The bandwidth returned is in Mb/s, i.e., megabits/second of
5593 * raw bandwidth.
5594 */
5595u32 pcie_bandwidth_available(struct pci_dev *dev, struct pci_dev **limiting_dev,
5596 enum pci_bus_speed *speed,
5597 enum pcie_link_width *width)
5598{
5599 u16 lnksta;
5600 enum pci_bus_speed next_speed;
5601 enum pcie_link_width next_width;
5602 u32 bw, next_bw;
5603
5604 if (speed)
5605 *speed = PCI_SPEED_UNKNOWN;
5606 if (width)
5607 *width = PCIE_LNK_WIDTH_UNKNOWN;
5608
5609 bw = 0;
5610
5611 while (dev) {
5612 pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
5613
5614 next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
5615 next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
5616 PCI_EXP_LNKSTA_NLW_SHIFT;
5617
5618 next_bw = next_width * PCIE_SPEED2MBS_ENC(next_speed);
5619
5620 /* Check if current device limits the total bandwidth */
5621 if (!bw || next_bw <= bw) {
5622 bw = next_bw;
5623
5624 if (limiting_dev)
5625 *limiting_dev = dev;
5626 if (speed)
5627 *speed = next_speed;
5628 if (width)
5629 *width = next_width;
5630 }
5631
5632 dev = pci_upstream_bridge(dev);
5633 }
5634
5635 return bw;
5636}
5637EXPORT_SYMBOL(pcie_bandwidth_available);
5638
5639/**
5640 * pcie_get_speed_cap - query for the PCI device's link speed capability
5641 * @dev: PCI device to query
5642 *
5643 * Query the PCI device speed capability. Return the maximum link speed
5644 * supported by the device.
5645 */
5646enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
5647{
5648 u32 lnkcap2, lnkcap;
5649
5650 /*
5651 * Link Capabilities 2 was added in PCIe r3.0, sec 7.8.18. The
5652 * implementation note there recommends using the Supported Link
5653 * Speeds Vector in Link Capabilities 2 when supported.
5654 *
5655 * Without Link Capabilities 2, i.e., prior to PCIe r3.0, software
5656 * should use the Supported Link Speeds field in Link Capabilities,
5657 * where only 2.5 GT/s and 5.0 GT/s speeds were defined.
5658 */
5659 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP2, &lnkcap2);
5660 if (lnkcap2) { /* PCIe r3.0-compliant */
5661 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_32_0GB)
5662 return PCIE_SPEED_32_0GT;
5663 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_16_0GB)
5664 return PCIE_SPEED_16_0GT;
5665 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
5666 return PCIE_SPEED_8_0GT;
5667 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
5668 return PCIE_SPEED_5_0GT;
5669 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
5670 return PCIE_SPEED_2_5GT;
5671 return PCI_SPEED_UNKNOWN;
5672 }
5673
5674 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5675 if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_5_0GB)
5676 return PCIE_SPEED_5_0GT;
5677 else if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_2_5GB)
5678 return PCIE_SPEED_2_5GT;
5679
5680 return PCI_SPEED_UNKNOWN;
5681}
5682EXPORT_SYMBOL(pcie_get_speed_cap);
5683
5684/**
5685 * pcie_get_width_cap - query for the PCI device's link width capability
5686 * @dev: PCI device to query
5687 *
5688 * Query the PCI device width capability. Return the maximum link width
5689 * supported by the device.
5690 */
5691enum pcie_link_width pcie_get_width_cap(struct pci_dev *dev)
5692{
5693 u32 lnkcap;
5694
5695 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5696 if (lnkcap)
5697 return (lnkcap & PCI_EXP_LNKCAP_MLW) >> 4;
5698
5699 return PCIE_LNK_WIDTH_UNKNOWN;
5700}
5701EXPORT_SYMBOL(pcie_get_width_cap);
5702
5703/**
5704 * pcie_bandwidth_capable - calculate a PCI device's link bandwidth capability
5705 * @dev: PCI device
5706 * @speed: storage for link speed
5707 * @width: storage for link width
5708 *
5709 * Calculate a PCI device's link bandwidth by querying for its link speed
5710 * and width, multiplying them, and applying encoding overhead. The result
5711 * is in Mb/s, i.e., megabits/second of raw bandwidth.
5712 */
5713u32 pcie_bandwidth_capable(struct pci_dev *dev, enum pci_bus_speed *speed,
5714 enum pcie_link_width *width)
5715{
5716 *speed = pcie_get_speed_cap(dev);
5717 *width = pcie_get_width_cap(dev);
5718
5719 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
5720 return 0;
5721
5722 return *width * PCIE_SPEED2MBS_ENC(*speed);
5723}
5724
5725/**
5726 * __pcie_print_link_status - Report the PCI device's link speed and width
5727 * @dev: PCI device to query
5728 * @verbose: Print info even when enough bandwidth is available
5729 *
5730 * If the available bandwidth at the device is less than the device is
5731 * capable of, report the device's maximum possible bandwidth and the
5732 * upstream link that limits its performance. If @verbose, always print
5733 * the available bandwidth, even if the device isn't constrained.
5734 */
5735void __pcie_print_link_status(struct pci_dev *dev, bool verbose)
5736{
5737 enum pcie_link_width width, width_cap;
5738 enum pci_bus_speed speed, speed_cap;
5739 struct pci_dev *limiting_dev = NULL;
5740 u32 bw_avail, bw_cap;
5741
5742 bw_cap = pcie_bandwidth_capable(dev, &speed_cap, &width_cap);
5743 bw_avail = pcie_bandwidth_available(dev, &limiting_dev, &speed, &width);
5744
5745 if (bw_avail >= bw_cap && verbose)
5746 pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth (%s x%d link)\n",
5747 bw_cap / 1000, bw_cap % 1000,
5748 PCIE_SPEED2STR(speed_cap), width_cap);
5749 else if (bw_avail < bw_cap)
5750 pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth, limited by %s x%d link at %s (capable of %u.%03u Gb/s with %s x%d link)\n",
5751 bw_avail / 1000, bw_avail % 1000,
5752 PCIE_SPEED2STR(speed), width,
5753 limiting_dev ? pci_name(limiting_dev) : "<unknown>",
5754 bw_cap / 1000, bw_cap % 1000,
5755 PCIE_SPEED2STR(speed_cap), width_cap);
5756}
5757
5758/**
5759 * pcie_print_link_status - Report the PCI device's link speed and width
5760 * @dev: PCI device to query
5761 *
5762 * Report the available bandwidth at the device.
5763 */
5764void pcie_print_link_status(struct pci_dev *dev)
5765{
5766 __pcie_print_link_status(dev, true);
5767}
5768EXPORT_SYMBOL(pcie_print_link_status);
5769
5770/**
5771 * pci_select_bars - Make BAR mask from the type of resource
5772 * @dev: the PCI device for which BAR mask is made
5773 * @flags: resource type mask to be selected
5774 *
5775 * This helper routine makes bar mask from the type of resource.
5776 */
5777int pci_select_bars(struct pci_dev *dev, unsigned long flags)
5778{
5779 int i, bars = 0;
5780 for (i = 0; i < PCI_NUM_RESOURCES; i++)
5781 if (pci_resource_flags(dev, i) & flags)
5782 bars |= (1 << i);
5783 return bars;
5784}
5785EXPORT_SYMBOL(pci_select_bars);
5786
5787/* Some architectures require additional programming to enable VGA */
5788static arch_set_vga_state_t arch_set_vga_state;
5789
5790void __init pci_register_set_vga_state(arch_set_vga_state_t func)
5791{
5792 arch_set_vga_state = func; /* NULL disables */
5793}
5794
5795static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
5796 unsigned int command_bits, u32 flags)
5797{
5798 if (arch_set_vga_state)
5799 return arch_set_vga_state(dev, decode, command_bits,
5800 flags);
5801 return 0;
5802}
5803
5804/**
5805 * pci_set_vga_state - set VGA decode state on device and parents if requested
5806 * @dev: the PCI device
5807 * @decode: true = enable decoding, false = disable decoding
5808 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
5809 * @flags: traverse ancestors and change bridges
5810 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
5811 */
5812int pci_set_vga_state(struct pci_dev *dev, bool decode,
5813 unsigned int command_bits, u32 flags)
5814{
5815 struct pci_bus *bus;
5816 struct pci_dev *bridge;
5817 u16 cmd;
5818 int rc;
5819
5820 WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
5821
5822 /* ARCH specific VGA enables */
5823 rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
5824 if (rc)
5825 return rc;
5826
5827 if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
5828 pci_read_config_word(dev, PCI_COMMAND, &cmd);
5829 if (decode == true)
5830 cmd |= command_bits;
5831 else
5832 cmd &= ~command_bits;
5833 pci_write_config_word(dev, PCI_COMMAND, cmd);
5834 }
5835
5836 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
5837 return 0;
5838
5839 bus = dev->bus;
5840 while (bus) {
5841 bridge = bus->self;
5842 if (bridge) {
5843 pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
5844 &cmd);
5845 if (decode == true)
5846 cmd |= PCI_BRIDGE_CTL_VGA;
5847 else
5848 cmd &= ~PCI_BRIDGE_CTL_VGA;
5849 pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
5850 cmd);
5851 }
5852 bus = bus->parent;
5853 }
5854 return 0;
5855}
5856
5857/**
5858 * pci_add_dma_alias - Add a DMA devfn alias for a device
5859 * @dev: the PCI device for which alias is added
5860 * @devfn: alias slot and function
5861 *
5862 * This helper encodes an 8-bit devfn as a bit number in dma_alias_mask
5863 * which is used to program permissible bus-devfn source addresses for DMA
5864 * requests in an IOMMU. These aliases factor into IOMMU group creation
5865 * and are useful for devices generating DMA requests beyond or different
5866 * from their logical bus-devfn. Examples include device quirks where the
5867 * device simply uses the wrong devfn, as well as non-transparent bridges
5868 * where the alias may be a proxy for devices in another domain.
5869 *
5870 * IOMMU group creation is performed during device discovery or addition,
5871 * prior to any potential DMA mapping and therefore prior to driver probing
5872 * (especially for userspace assigned devices where IOMMU group definition
5873 * cannot be left as a userspace activity). DMA aliases should therefore
5874 * be configured via quirks, such as the PCI fixup header quirk.
5875 */
5876void pci_add_dma_alias(struct pci_dev *dev, u8 devfn)
5877{
5878 if (!dev->dma_alias_mask)
5879 dev->dma_alias_mask = bitmap_zalloc(U8_MAX, GFP_KERNEL);
5880 if (!dev->dma_alias_mask) {
5881 pci_warn(dev, "Unable to allocate DMA alias mask\n");
5882 return;
5883 }
5884
5885 set_bit(devfn, dev->dma_alias_mask);
5886 pci_info(dev, "Enabling fixed DMA alias to %02x.%d\n",
5887 PCI_SLOT(devfn), PCI_FUNC(devfn));
5888}
5889
5890bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
5891{
5892 return (dev1->dma_alias_mask &&
5893 test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
5894 (dev2->dma_alias_mask &&
5895 test_bit(dev1->devfn, dev2->dma_alias_mask));
5896}
5897
5898bool pci_device_is_present(struct pci_dev *pdev)
5899{
5900 u32 v;
5901
5902 if (pci_dev_is_disconnected(pdev))
5903 return false;
5904 return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
5905}
5906EXPORT_SYMBOL_GPL(pci_device_is_present);
5907
5908void pci_ignore_hotplug(struct pci_dev *dev)
5909{
5910 struct pci_dev *bridge = dev->bus->self;
5911
5912 dev->ignore_hotplug = 1;
5913 /* Propagate the "ignore hotplug" setting to the parent bridge. */
5914 if (bridge)
5915 bridge->ignore_hotplug = 1;
5916}
5917EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
5918
5919resource_size_t __weak pcibios_default_alignment(void)
5920{
5921 return 0;
5922}
5923
5924/*
5925 * Arches that don't want to expose struct resource to userland as-is in
5926 * sysfs and /proc can implement their own pci_resource_to_user().
5927 */
5928void __weak pci_resource_to_user(const struct pci_dev *dev, int bar,
5929 const struct resource *rsrc,
5930 resource_size_t *start, resource_size_t *end)
5931{
5932 *start = rsrc->start;
5933 *end = rsrc->end;
5934}
5935
5936static char *resource_alignment_param;
5937static DEFINE_SPINLOCK(resource_alignment_lock);
5938
5939/**
5940 * pci_specified_resource_alignment - get resource alignment specified by user.
5941 * @dev: the PCI device to get
5942 * @resize: whether or not to change resources' size when reassigning alignment
5943 *
5944 * RETURNS: Resource alignment if it is specified.
5945 * Zero if it is not specified.
5946 */
5947static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
5948 bool *resize)
5949{
5950 int align_order, count;
5951 resource_size_t align = pcibios_default_alignment();
5952 const char *p;
5953 int ret;
5954
5955 spin_lock(&resource_alignment_lock);
5956 p = resource_alignment_param;
5957 if (!p || !*p)
5958 goto out;
5959 if (pci_has_flag(PCI_PROBE_ONLY)) {
5960 align = 0;
5961 pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
5962 goto out;
5963 }
5964
5965 while (*p) {
5966 count = 0;
5967 if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
5968 p[count] == '@') {
5969 p += count + 1;
5970 } else {
5971 align_order = -1;
5972 }
5973
5974 ret = pci_dev_str_match(dev, p, &p);
5975 if (ret == 1) {
5976 *resize = true;
5977 if (align_order == -1)
5978 align = PAGE_SIZE;
5979 else
5980 align = 1 << align_order;
5981 break;
5982 } else if (ret < 0) {
5983 pr_err("PCI: Can't parse resource_alignment parameter: %s\n",
5984 p);
5985 break;
5986 }
5987
5988 if (*p != ';' && *p != ',') {
5989 /* End of param or invalid format */
5990 break;
5991 }
5992 p++;
5993 }
5994out:
5995 spin_unlock(&resource_alignment_lock);
5996 return align;
5997}
5998
5999static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
6000 resource_size_t align, bool resize)
6001{
6002 struct resource *r = &dev->resource[bar];
6003 resource_size_t size;
6004
6005 if (!(r->flags & IORESOURCE_MEM))
6006 return;
6007
6008 if (r->flags & IORESOURCE_PCI_FIXED) {
6009 pci_info(dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
6010 bar, r, (unsigned long long)align);
6011 return;
6012 }
6013
6014 size = resource_size(r);
6015 if (size >= align)
6016 return;
6017
6018 /*
6019 * Increase the alignment of the resource. There are two ways we
6020 * can do this:
6021 *
6022 * 1) Increase the size of the resource. BARs are aligned on their
6023 * size, so when we reallocate space for this resource, we'll
6024 * allocate it with the larger alignment. This also prevents
6025 * assignment of any other BARs inside the alignment region, so
6026 * if we're requesting page alignment, this means no other BARs
6027 * will share the page.
6028 *
6029 * The disadvantage is that this makes the resource larger than
6030 * the hardware BAR, which may break drivers that compute things
6031 * based on the resource size, e.g., to find registers at a
6032 * fixed offset before the end of the BAR.
6033 *
6034 * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
6035 * set r->start to the desired alignment. By itself this
6036 * doesn't prevent other BARs being put inside the alignment
6037 * region, but if we realign *every* resource of every device in
6038 * the system, none of them will share an alignment region.
6039 *
6040 * When the user has requested alignment for only some devices via
6041 * the "pci=resource_alignment" argument, "resize" is true and we
6042 * use the first method. Otherwise we assume we're aligning all
6043 * devices and we use the second.
6044 */
6045
6046 pci_info(dev, "BAR%d %pR: requesting alignment to %#llx\n",
6047 bar, r, (unsigned long long)align);
6048
6049 if (resize) {
6050 r->start = 0;
6051 r->end = align - 1;
6052 } else {
6053 r->flags &= ~IORESOURCE_SIZEALIGN;
6054 r->flags |= IORESOURCE_STARTALIGN;
6055 r->start = align;
6056 r->end = r->start + size - 1;
6057 }
6058 r->flags |= IORESOURCE_UNSET;
6059}
6060
6061/*
6062 * This function disables memory decoding and releases memory resources
6063 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
6064 * It also rounds up size to specified alignment.
6065 * Later on, the kernel will assign page-aligned memory resource back
6066 * to the device.
6067 */
6068void pci_reassigndev_resource_alignment(struct pci_dev *dev)
6069{
6070 int i;
6071 struct resource *r;
6072 resource_size_t align;
6073 u16 command;
6074 bool resize = false;
6075
6076 /*
6077 * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
6078 * 3.4.1.11. Their resources are allocated from the space
6079 * described by the VF BARx register in the PF's SR-IOV capability.
6080 * We can't influence their alignment here.
6081 */
6082 if (dev->is_virtfn)
6083 return;
6084
6085 /* check if specified PCI is target device to reassign */
6086 align = pci_specified_resource_alignment(dev, &resize);
6087 if (!align)
6088 return;
6089
6090 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
6091 (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
6092 pci_warn(dev, "Can't reassign resources to host bridge\n");
6093 return;
6094 }
6095
6096 pci_read_config_word(dev, PCI_COMMAND, &command);
6097 command &= ~PCI_COMMAND_MEMORY;
6098 pci_write_config_word(dev, PCI_COMMAND, command);
6099
6100 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
6101 pci_request_resource_alignment(dev, i, align, resize);
6102
6103 /*
6104 * Need to disable bridge's resource window,
6105 * to enable the kernel to reassign new resource
6106 * window later on.
6107 */
6108 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
6109 for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
6110 r = &dev->resource[i];
6111 if (!(r->flags & IORESOURCE_MEM))
6112 continue;
6113 r->flags |= IORESOURCE_UNSET;
6114 r->end = resource_size(r) - 1;
6115 r->start = 0;
6116 }
6117 pci_disable_bridge_window(dev);
6118 }
6119}
6120
6121static ssize_t resource_alignment_show(struct bus_type *bus, char *buf)
6122{
6123 size_t count = 0;
6124
6125 spin_lock(&resource_alignment_lock);
6126 if (resource_alignment_param)
6127 count = snprintf(buf, PAGE_SIZE, "%s", resource_alignment_param);
6128 spin_unlock(&resource_alignment_lock);
6129
6130 /*
6131 * When set by the command line, resource_alignment_param will not
6132 * have a trailing line feed, which is ugly. So conditionally add
6133 * it here.
6134 */
6135 if (count >= 2 && buf[count - 2] != '\n' && count < PAGE_SIZE - 1) {
6136 buf[count - 1] = '\n';
6137 buf[count++] = 0;
6138 }
6139
6140 return count;
6141}
6142
6143static ssize_t resource_alignment_store(struct bus_type *bus,
6144 const char *buf, size_t count)
6145{
6146 char *param = kstrndup(buf, count, GFP_KERNEL);
6147
6148 if (!param)
6149 return -ENOMEM;
6150
6151 spin_lock(&resource_alignment_lock);
6152 kfree(resource_alignment_param);
6153 resource_alignment_param = param;
6154 spin_unlock(&resource_alignment_lock);
6155 return count;
6156}
6157
6158static BUS_ATTR_RW(resource_alignment);
6159
6160static int __init pci_resource_alignment_sysfs_init(void)
6161{
6162 return bus_create_file(&pci_bus_type,
6163 &bus_attr_resource_alignment);
6164}
6165late_initcall(pci_resource_alignment_sysfs_init);
6166
6167static void pci_no_domains(void)
6168{
6169#ifdef CONFIG_PCI_DOMAINS
6170 pci_domains_supported = 0;
6171#endif
6172}
6173
6174#ifdef CONFIG_PCI_DOMAINS_GENERIC
6175static atomic_t __domain_nr = ATOMIC_INIT(-1);
6176
6177static int pci_get_new_domain_nr(void)
6178{
6179 return atomic_inc_return(&__domain_nr);
6180}
6181
6182static int of_pci_bus_find_domain_nr(struct device *parent)
6183{
6184 static int use_dt_domains = -1;
6185 int domain = -1;
6186
6187 if (parent)
6188 domain = of_get_pci_domain_nr(parent->of_node);
6189
6190 /*
6191 * Check DT domain and use_dt_domains values.
6192 *
6193 * If DT domain property is valid (domain >= 0) and
6194 * use_dt_domains != 0, the DT assignment is valid since this means
6195 * we have not previously allocated a domain number by using
6196 * pci_get_new_domain_nr(); we should also update use_dt_domains to
6197 * 1, to indicate that we have just assigned a domain number from
6198 * DT.
6199 *
6200 * If DT domain property value is not valid (ie domain < 0), and we
6201 * have not previously assigned a domain number from DT
6202 * (use_dt_domains != 1) we should assign a domain number by
6203 * using the:
6204 *
6205 * pci_get_new_domain_nr()
6206 *
6207 * API and update the use_dt_domains value to keep track of method we
6208 * are using to assign domain numbers (use_dt_domains = 0).
6209 *
6210 * All other combinations imply we have a platform that is trying
6211 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
6212 * which is a recipe for domain mishandling and it is prevented by
6213 * invalidating the domain value (domain = -1) and printing a
6214 * corresponding error.
6215 */
6216 if (domain >= 0 && use_dt_domains) {
6217 use_dt_domains = 1;
6218 } else if (domain < 0 && use_dt_domains != 1) {
6219 use_dt_domains = 0;
6220 domain = pci_get_new_domain_nr();
6221 } else {
6222 if (parent)
6223 pr_err("Node %pOF has ", parent->of_node);
6224 pr_err("Inconsistent \"linux,pci-domain\" property in DT\n");
6225 domain = -1;
6226 }
6227
6228 return domain;
6229}
6230
6231int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
6232{
6233 return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
6234 acpi_pci_bus_find_domain_nr(bus);
6235}
6236#endif
6237
6238/**
6239 * pci_ext_cfg_avail - can we access extended PCI config space?
6240 *
6241 * Returns 1 if we can access PCI extended config space (offsets
6242 * greater than 0xff). This is the default implementation. Architecture
6243 * implementations can override this.
6244 */
6245int __weak pci_ext_cfg_avail(void)
6246{
6247 return 1;
6248}
6249
6250void __weak pci_fixup_cardbus(struct pci_bus *bus)
6251{
6252}
6253EXPORT_SYMBOL(pci_fixup_cardbus);
6254
6255static int __init pci_setup(char *str)
6256{
6257 while (str) {
6258 char *k = strchr(str, ',');
6259 if (k)
6260 *k++ = 0;
6261 if (*str && (str = pcibios_setup(str)) && *str) {
6262 if (!strcmp(str, "nomsi")) {
6263 pci_no_msi();
6264 } else if (!strncmp(str, "noats", 5)) {
6265 pr_info("PCIe: ATS is disabled\n");
6266 pcie_ats_disabled = true;
6267 } else if (!strcmp(str, "noaer")) {
6268 pci_no_aer();
6269 } else if (!strcmp(str, "earlydump")) {
6270 pci_early_dump = true;
6271 } else if (!strncmp(str, "realloc=", 8)) {
6272 pci_realloc_get_opt(str + 8);
6273 } else if (!strncmp(str, "realloc", 7)) {
6274 pci_realloc_get_opt("on");
6275 } else if (!strcmp(str, "nodomains")) {
6276 pci_no_domains();
6277 } else if (!strncmp(str, "noari", 5)) {
6278 pcie_ari_disabled = true;
6279 } else if (!strncmp(str, "cbiosize=", 9)) {
6280 pci_cardbus_io_size = memparse(str + 9, &str);
6281 } else if (!strncmp(str, "cbmemsize=", 10)) {
6282 pci_cardbus_mem_size = memparse(str + 10, &str);
6283 } else if (!strncmp(str, "resource_alignment=", 19)) {
6284 resource_alignment_param = str + 19;
6285 } else if (!strncmp(str, "ecrc=", 5)) {
6286 pcie_ecrc_get_policy(str + 5);
6287 } else if (!strncmp(str, "hpiosize=", 9)) {
6288 pci_hotplug_io_size = memparse(str + 9, &str);
6289 } else if (!strncmp(str, "hpmemsize=", 10)) {
6290 pci_hotplug_mem_size = memparse(str + 10, &str);
6291 } else if (!strncmp(str, "hpbussize=", 10)) {
6292 pci_hotplug_bus_size =
6293 simple_strtoul(str + 10, &str, 0);
6294 if (pci_hotplug_bus_size > 0xff)
6295 pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
6296 } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
6297 pcie_bus_config = PCIE_BUS_TUNE_OFF;
6298 } else if (!strncmp(str, "pcie_bus_safe", 13)) {
6299 pcie_bus_config = PCIE_BUS_SAFE;
6300 } else if (!strncmp(str, "pcie_bus_perf", 13)) {
6301 pcie_bus_config = PCIE_BUS_PERFORMANCE;
6302 } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
6303 pcie_bus_config = PCIE_BUS_PEER2PEER;
6304 } else if (!strncmp(str, "pcie_scan_all", 13)) {
6305 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
6306 } else if (!strncmp(str, "disable_acs_redir=", 18)) {
6307 disable_acs_redir_param = str + 18;
6308 } else {
6309 pr_err("PCI: Unknown option `%s'\n", str);
6310 }
6311 }
6312 str = k;
6313 }
6314 return 0;
6315}
6316early_param("pci", pci_setup);
6317
6318/*
6319 * 'resource_alignment_param' and 'disable_acs_redir_param' are initialized
6320 * in pci_setup(), above, to point to data in the __initdata section which
6321 * will be freed after the init sequence is complete. We can't allocate memory
6322 * in pci_setup() because some architectures do not have any memory allocation
6323 * service available during an early_param() call. So we allocate memory and
6324 * copy the variable here before the init section is freed.
6325 *
6326 */
6327static int __init pci_realloc_setup_params(void)
6328{
6329 resource_alignment_param = kstrdup(resource_alignment_param,
6330 GFP_KERNEL);
6331 disable_acs_redir_param = kstrdup(disable_acs_redir_param, GFP_KERNEL);
6332
6333 return 0;
6334}
6335pure_initcall(pci_realloc_setup_params);
1/*
2 * PCI Bus Services, see include/linux/pci.h for further explanation.
3 *
4 * Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
5 * David Mosberger-Tang
6 *
7 * Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
8 */
9
10#include <linux/kernel.h>
11#include <linux/delay.h>
12#include <linux/init.h>
13#include <linux/of.h>
14#include <linux/of_pci.h>
15#include <linux/pci.h>
16#include <linux/pm.h>
17#include <linux/slab.h>
18#include <linux/module.h>
19#include <linux/spinlock.h>
20#include <linux/string.h>
21#include <linux/log2.h>
22#include <linux/pci-aspm.h>
23#include <linux/pm_wakeup.h>
24#include <linux/interrupt.h>
25#include <linux/device.h>
26#include <linux/pm_runtime.h>
27#include <linux/pci_hotplug.h>
28#include <asm/setup.h>
29#include <linux/aer.h>
30#include "pci.h"
31
32const char *pci_power_names[] = {
33 "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
34};
35EXPORT_SYMBOL_GPL(pci_power_names);
36
37int isa_dma_bridge_buggy;
38EXPORT_SYMBOL(isa_dma_bridge_buggy);
39
40int pci_pci_problems;
41EXPORT_SYMBOL(pci_pci_problems);
42
43unsigned int pci_pm_d3_delay;
44
45static void pci_pme_list_scan(struct work_struct *work);
46
47static LIST_HEAD(pci_pme_list);
48static DEFINE_MUTEX(pci_pme_list_mutex);
49static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
50
51struct pci_pme_device {
52 struct list_head list;
53 struct pci_dev *dev;
54};
55
56#define PME_TIMEOUT 1000 /* How long between PME checks */
57
58static void pci_dev_d3_sleep(struct pci_dev *dev)
59{
60 unsigned int delay = dev->d3_delay;
61
62 if (delay < pci_pm_d3_delay)
63 delay = pci_pm_d3_delay;
64
65 msleep(delay);
66}
67
68#ifdef CONFIG_PCI_DOMAINS
69int pci_domains_supported = 1;
70#endif
71
72#define DEFAULT_CARDBUS_IO_SIZE (256)
73#define DEFAULT_CARDBUS_MEM_SIZE (64*1024*1024)
74/* pci=cbmemsize=nnM,cbiosize=nn can override this */
75unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
76unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
77
78#define DEFAULT_HOTPLUG_IO_SIZE (256)
79#define DEFAULT_HOTPLUG_MEM_SIZE (2*1024*1024)
80/* pci=hpmemsize=nnM,hpiosize=nn can override this */
81unsigned long pci_hotplug_io_size = DEFAULT_HOTPLUG_IO_SIZE;
82unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
83
84enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
85
86/*
87 * The default CLS is used if arch didn't set CLS explicitly and not
88 * all pci devices agree on the same value. Arch can override either
89 * the dfl or actual value as it sees fit. Don't forget this is
90 * measured in 32-bit words, not bytes.
91 */
92u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
93u8 pci_cache_line_size;
94
95/*
96 * If we set up a device for bus mastering, we need to check the latency
97 * timer as certain BIOSes forget to set it properly.
98 */
99unsigned int pcibios_max_latency = 255;
100
101/* If set, the PCIe ARI capability will not be used. */
102static bool pcie_ari_disabled;
103
104/**
105 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
106 * @bus: pointer to PCI bus structure to search
107 *
108 * Given a PCI bus, returns the highest PCI bus number present in the set
109 * including the given PCI bus and its list of child PCI buses.
110 */
111unsigned char pci_bus_max_busnr(struct pci_bus *bus)
112{
113 struct pci_bus *tmp;
114 unsigned char max, n;
115
116 max = bus->busn_res.end;
117 list_for_each_entry(tmp, &bus->children, node) {
118 n = pci_bus_max_busnr(tmp);
119 if (n > max)
120 max = n;
121 }
122 return max;
123}
124EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
125
126#ifdef CONFIG_HAS_IOMEM
127void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
128{
129 struct resource *res = &pdev->resource[bar];
130
131 /*
132 * Make sure the BAR is actually a memory resource, not an IO resource
133 */
134 if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
135 dev_warn(&pdev->dev, "can't ioremap BAR %d: %pR\n", bar, res);
136 return NULL;
137 }
138 return ioremap_nocache(res->start, resource_size(res));
139}
140EXPORT_SYMBOL_GPL(pci_ioremap_bar);
141
142void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
143{
144 /*
145 * Make sure the BAR is actually a memory resource, not an IO resource
146 */
147 if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
148 WARN_ON(1);
149 return NULL;
150 }
151 return ioremap_wc(pci_resource_start(pdev, bar),
152 pci_resource_len(pdev, bar));
153}
154EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
155#endif
156
157
158static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
159 u8 pos, int cap, int *ttl)
160{
161 u8 id;
162 u16 ent;
163
164 pci_bus_read_config_byte(bus, devfn, pos, &pos);
165
166 while ((*ttl)--) {
167 if (pos < 0x40)
168 break;
169 pos &= ~3;
170 pci_bus_read_config_word(bus, devfn, pos, &ent);
171
172 id = ent & 0xff;
173 if (id == 0xff)
174 break;
175 if (id == cap)
176 return pos;
177 pos = (ent >> 8);
178 }
179 return 0;
180}
181
182static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
183 u8 pos, int cap)
184{
185 int ttl = PCI_FIND_CAP_TTL;
186
187 return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
188}
189
190int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
191{
192 return __pci_find_next_cap(dev->bus, dev->devfn,
193 pos + PCI_CAP_LIST_NEXT, cap);
194}
195EXPORT_SYMBOL_GPL(pci_find_next_capability);
196
197static int __pci_bus_find_cap_start(struct pci_bus *bus,
198 unsigned int devfn, u8 hdr_type)
199{
200 u16 status;
201
202 pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
203 if (!(status & PCI_STATUS_CAP_LIST))
204 return 0;
205
206 switch (hdr_type) {
207 case PCI_HEADER_TYPE_NORMAL:
208 case PCI_HEADER_TYPE_BRIDGE:
209 return PCI_CAPABILITY_LIST;
210 case PCI_HEADER_TYPE_CARDBUS:
211 return PCI_CB_CAPABILITY_LIST;
212 }
213
214 return 0;
215}
216
217/**
218 * pci_find_capability - query for devices' capabilities
219 * @dev: PCI device to query
220 * @cap: capability code
221 *
222 * Tell if a device supports a given PCI capability.
223 * Returns the address of the requested capability structure within the
224 * device's PCI configuration space or 0 in case the device does not
225 * support it. Possible values for @cap:
226 *
227 * %PCI_CAP_ID_PM Power Management
228 * %PCI_CAP_ID_AGP Accelerated Graphics Port
229 * %PCI_CAP_ID_VPD Vital Product Data
230 * %PCI_CAP_ID_SLOTID Slot Identification
231 * %PCI_CAP_ID_MSI Message Signalled Interrupts
232 * %PCI_CAP_ID_CHSWP CompactPCI HotSwap
233 * %PCI_CAP_ID_PCIX PCI-X
234 * %PCI_CAP_ID_EXP PCI Express
235 */
236int pci_find_capability(struct pci_dev *dev, int cap)
237{
238 int pos;
239
240 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
241 if (pos)
242 pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
243
244 return pos;
245}
246EXPORT_SYMBOL(pci_find_capability);
247
248/**
249 * pci_bus_find_capability - query for devices' capabilities
250 * @bus: the PCI bus to query
251 * @devfn: PCI device to query
252 * @cap: capability code
253 *
254 * Like pci_find_capability() but works for pci devices that do not have a
255 * pci_dev structure set up yet.
256 *
257 * Returns the address of the requested capability structure within the
258 * device's PCI configuration space or 0 in case the device does not
259 * support it.
260 */
261int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
262{
263 int pos;
264 u8 hdr_type;
265
266 pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
267
268 pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
269 if (pos)
270 pos = __pci_find_next_cap(bus, devfn, pos, cap);
271
272 return pos;
273}
274EXPORT_SYMBOL(pci_bus_find_capability);
275
276/**
277 * pci_find_next_ext_capability - Find an extended capability
278 * @dev: PCI device to query
279 * @start: address at which to start looking (0 to start at beginning of list)
280 * @cap: capability code
281 *
282 * Returns the address of the next matching extended capability structure
283 * within the device's PCI configuration space or 0 if the device does
284 * not support it. Some capabilities can occur several times, e.g., the
285 * vendor-specific capability, and this provides a way to find them all.
286 */
287int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
288{
289 u32 header;
290 int ttl;
291 int pos = PCI_CFG_SPACE_SIZE;
292
293 /* minimum 8 bytes per capability */
294 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
295
296 if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
297 return 0;
298
299 if (start)
300 pos = start;
301
302 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
303 return 0;
304
305 /*
306 * If we have no capabilities, this is indicated by cap ID,
307 * cap version and next pointer all being 0.
308 */
309 if (header == 0)
310 return 0;
311
312 while (ttl-- > 0) {
313 if (PCI_EXT_CAP_ID(header) == cap && pos != start)
314 return pos;
315
316 pos = PCI_EXT_CAP_NEXT(header);
317 if (pos < PCI_CFG_SPACE_SIZE)
318 break;
319
320 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
321 break;
322 }
323
324 return 0;
325}
326EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
327
328/**
329 * pci_find_ext_capability - Find an extended capability
330 * @dev: PCI device to query
331 * @cap: capability code
332 *
333 * Returns the address of the requested extended capability structure
334 * within the device's PCI configuration space or 0 if the device does
335 * not support it. Possible values for @cap:
336 *
337 * %PCI_EXT_CAP_ID_ERR Advanced Error Reporting
338 * %PCI_EXT_CAP_ID_VC Virtual Channel
339 * %PCI_EXT_CAP_ID_DSN Device Serial Number
340 * %PCI_EXT_CAP_ID_PWR Power Budgeting
341 */
342int pci_find_ext_capability(struct pci_dev *dev, int cap)
343{
344 return pci_find_next_ext_capability(dev, 0, cap);
345}
346EXPORT_SYMBOL_GPL(pci_find_ext_capability);
347
348static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
349{
350 int rc, ttl = PCI_FIND_CAP_TTL;
351 u8 cap, mask;
352
353 if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
354 mask = HT_3BIT_CAP_MASK;
355 else
356 mask = HT_5BIT_CAP_MASK;
357
358 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
359 PCI_CAP_ID_HT, &ttl);
360 while (pos) {
361 rc = pci_read_config_byte(dev, pos + 3, &cap);
362 if (rc != PCIBIOS_SUCCESSFUL)
363 return 0;
364
365 if ((cap & mask) == ht_cap)
366 return pos;
367
368 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
369 pos + PCI_CAP_LIST_NEXT,
370 PCI_CAP_ID_HT, &ttl);
371 }
372
373 return 0;
374}
375/**
376 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
377 * @dev: PCI device to query
378 * @pos: Position from which to continue searching
379 * @ht_cap: Hypertransport capability code
380 *
381 * To be used in conjunction with pci_find_ht_capability() to search for
382 * all capabilities matching @ht_cap. @pos should always be a value returned
383 * from pci_find_ht_capability().
384 *
385 * NB. To be 100% safe against broken PCI devices, the caller should take
386 * steps to avoid an infinite loop.
387 */
388int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
389{
390 return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
391}
392EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
393
394/**
395 * pci_find_ht_capability - query a device's Hypertransport capabilities
396 * @dev: PCI device to query
397 * @ht_cap: Hypertransport capability code
398 *
399 * Tell if a device supports a given Hypertransport capability.
400 * Returns an address within the device's PCI configuration space
401 * or 0 in case the device does not support the request capability.
402 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
403 * which has a Hypertransport capability matching @ht_cap.
404 */
405int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
406{
407 int pos;
408
409 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
410 if (pos)
411 pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
412
413 return pos;
414}
415EXPORT_SYMBOL_GPL(pci_find_ht_capability);
416
417/**
418 * pci_find_parent_resource - return resource region of parent bus of given region
419 * @dev: PCI device structure contains resources to be searched
420 * @res: child resource record for which parent is sought
421 *
422 * For given resource region of given device, return the resource
423 * region of parent bus the given region is contained in.
424 */
425struct resource *pci_find_parent_resource(const struct pci_dev *dev,
426 struct resource *res)
427{
428 const struct pci_bus *bus = dev->bus;
429 struct resource *r;
430 int i;
431
432 pci_bus_for_each_resource(bus, r, i) {
433 if (!r)
434 continue;
435 if (res->start && resource_contains(r, res)) {
436
437 /*
438 * If the window is prefetchable but the BAR is
439 * not, the allocator made a mistake.
440 */
441 if (r->flags & IORESOURCE_PREFETCH &&
442 !(res->flags & IORESOURCE_PREFETCH))
443 return NULL;
444
445 /*
446 * If we're below a transparent bridge, there may
447 * be both a positively-decoded aperture and a
448 * subtractively-decoded region that contain the BAR.
449 * We want the positively-decoded one, so this depends
450 * on pci_bus_for_each_resource() giving us those
451 * first.
452 */
453 return r;
454 }
455 }
456 return NULL;
457}
458EXPORT_SYMBOL(pci_find_parent_resource);
459
460/**
461 * pci_find_pcie_root_port - return PCIe Root Port
462 * @dev: PCI device to query
463 *
464 * Traverse up the parent chain and return the PCIe Root Port PCI Device
465 * for a given PCI Device.
466 */
467struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
468{
469 struct pci_dev *bridge, *highest_pcie_bridge = NULL;
470
471 bridge = pci_upstream_bridge(dev);
472 while (bridge && pci_is_pcie(bridge)) {
473 highest_pcie_bridge = bridge;
474 bridge = pci_upstream_bridge(bridge);
475 }
476
477 if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
478 return NULL;
479
480 return highest_pcie_bridge;
481}
482EXPORT_SYMBOL(pci_find_pcie_root_port);
483
484/**
485 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
486 * @dev: the PCI device to operate on
487 * @pos: config space offset of status word
488 * @mask: mask of bit(s) to care about in status word
489 *
490 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
491 */
492int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
493{
494 int i;
495
496 /* Wait for Transaction Pending bit clean */
497 for (i = 0; i < 4; i++) {
498 u16 status;
499 if (i)
500 msleep((1 << (i - 1)) * 100);
501
502 pci_read_config_word(dev, pos, &status);
503 if (!(status & mask))
504 return 1;
505 }
506
507 return 0;
508}
509
510/**
511 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
512 * @dev: PCI device to have its BARs restored
513 *
514 * Restore the BAR values for a given device, so as to make it
515 * accessible by its driver.
516 */
517static void pci_restore_bars(struct pci_dev *dev)
518{
519 int i;
520
521 /* Per SR-IOV spec 3.4.1.11, VF BARs are RO zero */
522 if (dev->is_virtfn)
523 return;
524
525 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
526 pci_update_resource(dev, i);
527}
528
529static const struct pci_platform_pm_ops *pci_platform_pm;
530
531int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
532{
533 if (!ops->is_manageable || !ops->set_state || !ops->choose_state
534 || !ops->sleep_wake)
535 return -EINVAL;
536 pci_platform_pm = ops;
537 return 0;
538}
539
540static inline bool platform_pci_power_manageable(struct pci_dev *dev)
541{
542 return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
543}
544
545static inline int platform_pci_set_power_state(struct pci_dev *dev,
546 pci_power_t t)
547{
548 return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
549}
550
551static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
552{
553 return pci_platform_pm ?
554 pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
555}
556
557static inline int platform_pci_sleep_wake(struct pci_dev *dev, bool enable)
558{
559 return pci_platform_pm ?
560 pci_platform_pm->sleep_wake(dev, enable) : -ENODEV;
561}
562
563static inline int platform_pci_run_wake(struct pci_dev *dev, bool enable)
564{
565 return pci_platform_pm ?
566 pci_platform_pm->run_wake(dev, enable) : -ENODEV;
567}
568
569static inline bool platform_pci_need_resume(struct pci_dev *dev)
570{
571 return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
572}
573
574/**
575 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
576 * given PCI device
577 * @dev: PCI device to handle.
578 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
579 *
580 * RETURN VALUE:
581 * -EINVAL if the requested state is invalid.
582 * -EIO if device does not support PCI PM or its PM capabilities register has a
583 * wrong version, or device doesn't support the requested state.
584 * 0 if device already is in the requested state.
585 * 0 if device's power state has been successfully changed.
586 */
587static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
588{
589 u16 pmcsr;
590 bool need_restore = false;
591
592 /* Check if we're already there */
593 if (dev->current_state == state)
594 return 0;
595
596 if (!dev->pm_cap)
597 return -EIO;
598
599 if (state < PCI_D0 || state > PCI_D3hot)
600 return -EINVAL;
601
602 /* Validate current state:
603 * Can enter D0 from any state, but if we can only go deeper
604 * to sleep if we're already in a low power state
605 */
606 if (state != PCI_D0 && dev->current_state <= PCI_D3cold
607 && dev->current_state > state) {
608 dev_err(&dev->dev, "invalid power transition (from state %d to %d)\n",
609 dev->current_state, state);
610 return -EINVAL;
611 }
612
613 /* check if this device supports the desired state */
614 if ((state == PCI_D1 && !dev->d1_support)
615 || (state == PCI_D2 && !dev->d2_support))
616 return -EIO;
617
618 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
619
620 /* If we're (effectively) in D3, force entire word to 0.
621 * This doesn't affect PME_Status, disables PME_En, and
622 * sets PowerState to 0.
623 */
624 switch (dev->current_state) {
625 case PCI_D0:
626 case PCI_D1:
627 case PCI_D2:
628 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
629 pmcsr |= state;
630 break;
631 case PCI_D3hot:
632 case PCI_D3cold:
633 case PCI_UNKNOWN: /* Boot-up */
634 if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
635 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
636 need_restore = true;
637 /* Fall-through: force to D0 */
638 default:
639 pmcsr = 0;
640 break;
641 }
642
643 /* enter specified state */
644 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
645
646 /* Mandatory power management transition delays */
647 /* see PCI PM 1.1 5.6.1 table 18 */
648 if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
649 pci_dev_d3_sleep(dev);
650 else if (state == PCI_D2 || dev->current_state == PCI_D2)
651 udelay(PCI_PM_D2_DELAY);
652
653 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
654 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
655 if (dev->current_state != state && printk_ratelimit())
656 dev_info(&dev->dev, "Refused to change power state, currently in D%d\n",
657 dev->current_state);
658
659 /*
660 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
661 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
662 * from D3hot to D0 _may_ perform an internal reset, thereby
663 * going to "D0 Uninitialized" rather than "D0 Initialized".
664 * For example, at least some versions of the 3c905B and the
665 * 3c556B exhibit this behaviour.
666 *
667 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
668 * devices in a D3hot state at boot. Consequently, we need to
669 * restore at least the BARs so that the device will be
670 * accessible to its driver.
671 */
672 if (need_restore)
673 pci_restore_bars(dev);
674
675 if (dev->bus->self)
676 pcie_aspm_pm_state_change(dev->bus->self);
677
678 return 0;
679}
680
681/**
682 * pci_update_current_state - Read PCI power state of given device from its
683 * PCI PM registers and cache it
684 * @dev: PCI device to handle.
685 * @state: State to cache in case the device doesn't have the PM capability
686 */
687void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
688{
689 if (dev->pm_cap) {
690 u16 pmcsr;
691
692 /*
693 * Configuration space is not accessible for device in
694 * D3cold, so just keep or set D3cold for safety
695 */
696 if (dev->current_state == PCI_D3cold)
697 return;
698 if (state == PCI_D3cold) {
699 dev->current_state = PCI_D3cold;
700 return;
701 }
702 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
703 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
704 } else {
705 dev->current_state = state;
706 }
707}
708
709/**
710 * pci_power_up - Put the given device into D0 forcibly
711 * @dev: PCI device to power up
712 */
713void pci_power_up(struct pci_dev *dev)
714{
715 if (platform_pci_power_manageable(dev))
716 platform_pci_set_power_state(dev, PCI_D0);
717
718 pci_raw_set_power_state(dev, PCI_D0);
719 pci_update_current_state(dev, PCI_D0);
720}
721
722/**
723 * pci_platform_power_transition - Use platform to change device power state
724 * @dev: PCI device to handle.
725 * @state: State to put the device into.
726 */
727static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
728{
729 int error;
730
731 if (platform_pci_power_manageable(dev)) {
732 error = platform_pci_set_power_state(dev, state);
733 if (!error)
734 pci_update_current_state(dev, state);
735 } else
736 error = -ENODEV;
737
738 if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
739 dev->current_state = PCI_D0;
740
741 return error;
742}
743
744/**
745 * pci_wakeup - Wake up a PCI device
746 * @pci_dev: Device to handle.
747 * @ign: ignored parameter
748 */
749static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
750{
751 pci_wakeup_event(pci_dev);
752 pm_request_resume(&pci_dev->dev);
753 return 0;
754}
755
756/**
757 * pci_wakeup_bus - Walk given bus and wake up devices on it
758 * @bus: Top bus of the subtree to walk.
759 */
760static void pci_wakeup_bus(struct pci_bus *bus)
761{
762 if (bus)
763 pci_walk_bus(bus, pci_wakeup, NULL);
764}
765
766/**
767 * __pci_start_power_transition - Start power transition of a PCI device
768 * @dev: PCI device to handle.
769 * @state: State to put the device into.
770 */
771static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
772{
773 if (state == PCI_D0) {
774 pci_platform_power_transition(dev, PCI_D0);
775 /*
776 * Mandatory power management transition delays, see
777 * PCI Express Base Specification Revision 2.0 Section
778 * 6.6.1: Conventional Reset. Do not delay for
779 * devices powered on/off by corresponding bridge,
780 * because have already delayed for the bridge.
781 */
782 if (dev->runtime_d3cold) {
783 msleep(dev->d3cold_delay);
784 /*
785 * When powering on a bridge from D3cold, the
786 * whole hierarchy may be powered on into
787 * D0uninitialized state, resume them to give
788 * them a chance to suspend again
789 */
790 pci_wakeup_bus(dev->subordinate);
791 }
792 }
793}
794
795/**
796 * __pci_dev_set_current_state - Set current state of a PCI device
797 * @dev: Device to handle
798 * @data: pointer to state to be set
799 */
800static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
801{
802 pci_power_t state = *(pci_power_t *)data;
803
804 dev->current_state = state;
805 return 0;
806}
807
808/**
809 * __pci_bus_set_current_state - Walk given bus and set current state of devices
810 * @bus: Top bus of the subtree to walk.
811 * @state: state to be set
812 */
813static void __pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
814{
815 if (bus)
816 pci_walk_bus(bus, __pci_dev_set_current_state, &state);
817}
818
819/**
820 * __pci_complete_power_transition - Complete power transition of a PCI device
821 * @dev: PCI device to handle.
822 * @state: State to put the device into.
823 *
824 * This function should not be called directly by device drivers.
825 */
826int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
827{
828 int ret;
829
830 if (state <= PCI_D0)
831 return -EINVAL;
832 ret = pci_platform_power_transition(dev, state);
833 /* Power off the bridge may power off the whole hierarchy */
834 if (!ret && state == PCI_D3cold)
835 __pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
836 return ret;
837}
838EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
839
840/**
841 * pci_set_power_state - Set the power state of a PCI device
842 * @dev: PCI device to handle.
843 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
844 *
845 * Transition a device to a new power state, using the platform firmware and/or
846 * the device's PCI PM registers.
847 *
848 * RETURN VALUE:
849 * -EINVAL if the requested state is invalid.
850 * -EIO if device does not support PCI PM or its PM capabilities register has a
851 * wrong version, or device doesn't support the requested state.
852 * 0 if device already is in the requested state.
853 * 0 if device's power state has been successfully changed.
854 */
855int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
856{
857 int error;
858
859 /* bound the state we're entering */
860 if (state > PCI_D3cold)
861 state = PCI_D3cold;
862 else if (state < PCI_D0)
863 state = PCI_D0;
864 else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
865 /*
866 * If the device or the parent bridge do not support PCI PM,
867 * ignore the request if we're doing anything other than putting
868 * it into D0 (which would only happen on boot).
869 */
870 return 0;
871
872 /* Check if we're already there */
873 if (dev->current_state == state)
874 return 0;
875
876 __pci_start_power_transition(dev, state);
877
878 /* This device is quirked not to be put into D3, so
879 don't put it in D3 */
880 if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
881 return 0;
882
883 /*
884 * To put device in D3cold, we put device into D3hot in native
885 * way, then put device into D3cold with platform ops
886 */
887 error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
888 PCI_D3hot : state);
889
890 if (!__pci_complete_power_transition(dev, state))
891 error = 0;
892
893 return error;
894}
895EXPORT_SYMBOL(pci_set_power_state);
896
897/**
898 * pci_choose_state - Choose the power state of a PCI device
899 * @dev: PCI device to be suspended
900 * @state: target sleep state for the whole system. This is the value
901 * that is passed to suspend() function.
902 *
903 * Returns PCI power state suitable for given device and given system
904 * message.
905 */
906
907pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
908{
909 pci_power_t ret;
910
911 if (!dev->pm_cap)
912 return PCI_D0;
913
914 ret = platform_pci_choose_state(dev);
915 if (ret != PCI_POWER_ERROR)
916 return ret;
917
918 switch (state.event) {
919 case PM_EVENT_ON:
920 return PCI_D0;
921 case PM_EVENT_FREEZE:
922 case PM_EVENT_PRETHAW:
923 /* REVISIT both freeze and pre-thaw "should" use D0 */
924 case PM_EVENT_SUSPEND:
925 case PM_EVENT_HIBERNATE:
926 return PCI_D3hot;
927 default:
928 dev_info(&dev->dev, "unrecognized suspend event %d\n",
929 state.event);
930 BUG();
931 }
932 return PCI_D0;
933}
934EXPORT_SYMBOL(pci_choose_state);
935
936#define PCI_EXP_SAVE_REGS 7
937
938static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
939 u16 cap, bool extended)
940{
941 struct pci_cap_saved_state *tmp;
942
943 hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
944 if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
945 return tmp;
946 }
947 return NULL;
948}
949
950struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
951{
952 return _pci_find_saved_cap(dev, cap, false);
953}
954
955struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
956{
957 return _pci_find_saved_cap(dev, cap, true);
958}
959
960static int pci_save_pcie_state(struct pci_dev *dev)
961{
962 int i = 0;
963 struct pci_cap_saved_state *save_state;
964 u16 *cap;
965
966 if (!pci_is_pcie(dev))
967 return 0;
968
969 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
970 if (!save_state) {
971 dev_err(&dev->dev, "buffer not found in %s\n", __func__);
972 return -ENOMEM;
973 }
974
975 cap = (u16 *)&save_state->cap.data[0];
976 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
977 pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
978 pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
979 pcie_capability_read_word(dev, PCI_EXP_RTCTL, &cap[i++]);
980 pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
981 pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
982 pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
983
984 return 0;
985}
986
987static void pci_restore_pcie_state(struct pci_dev *dev)
988{
989 int i = 0;
990 struct pci_cap_saved_state *save_state;
991 u16 *cap;
992
993 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
994 if (!save_state)
995 return;
996
997 cap = (u16 *)&save_state->cap.data[0];
998 pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
999 pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1000 pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1001 pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1002 pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1003 pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1004 pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1005}
1006
1007
1008static int pci_save_pcix_state(struct pci_dev *dev)
1009{
1010 int pos;
1011 struct pci_cap_saved_state *save_state;
1012
1013 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1014 if (!pos)
1015 return 0;
1016
1017 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1018 if (!save_state) {
1019 dev_err(&dev->dev, "buffer not found in %s\n", __func__);
1020 return -ENOMEM;
1021 }
1022
1023 pci_read_config_word(dev, pos + PCI_X_CMD,
1024 (u16 *)save_state->cap.data);
1025
1026 return 0;
1027}
1028
1029static void pci_restore_pcix_state(struct pci_dev *dev)
1030{
1031 int i = 0, pos;
1032 struct pci_cap_saved_state *save_state;
1033 u16 *cap;
1034
1035 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1036 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1037 if (!save_state || !pos)
1038 return;
1039 cap = (u16 *)&save_state->cap.data[0];
1040
1041 pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1042}
1043
1044
1045/**
1046 * pci_save_state - save the PCI configuration space of a device before suspending
1047 * @dev: - PCI device that we're dealing with
1048 */
1049int pci_save_state(struct pci_dev *dev)
1050{
1051 int i;
1052 /* XXX: 100% dword access ok here? */
1053 for (i = 0; i < 16; i++)
1054 pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1055 dev->state_saved = true;
1056
1057 i = pci_save_pcie_state(dev);
1058 if (i != 0)
1059 return i;
1060
1061 i = pci_save_pcix_state(dev);
1062 if (i != 0)
1063 return i;
1064
1065 return pci_save_vc_state(dev);
1066}
1067EXPORT_SYMBOL(pci_save_state);
1068
1069static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1070 u32 saved_val, int retry)
1071{
1072 u32 val;
1073
1074 pci_read_config_dword(pdev, offset, &val);
1075 if (val == saved_val)
1076 return;
1077
1078 for (;;) {
1079 dev_dbg(&pdev->dev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1080 offset, val, saved_val);
1081 pci_write_config_dword(pdev, offset, saved_val);
1082 if (retry-- <= 0)
1083 return;
1084
1085 pci_read_config_dword(pdev, offset, &val);
1086 if (val == saved_val)
1087 return;
1088
1089 mdelay(1);
1090 }
1091}
1092
1093static void pci_restore_config_space_range(struct pci_dev *pdev,
1094 int start, int end, int retry)
1095{
1096 int index;
1097
1098 for (index = end; index >= start; index--)
1099 pci_restore_config_dword(pdev, 4 * index,
1100 pdev->saved_config_space[index],
1101 retry);
1102}
1103
1104static void pci_restore_config_space(struct pci_dev *pdev)
1105{
1106 if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1107 pci_restore_config_space_range(pdev, 10, 15, 0);
1108 /* Restore BARs before the command register. */
1109 pci_restore_config_space_range(pdev, 4, 9, 10);
1110 pci_restore_config_space_range(pdev, 0, 3, 0);
1111 } else {
1112 pci_restore_config_space_range(pdev, 0, 15, 0);
1113 }
1114}
1115
1116/**
1117 * pci_restore_state - Restore the saved state of a PCI device
1118 * @dev: - PCI device that we're dealing with
1119 */
1120void pci_restore_state(struct pci_dev *dev)
1121{
1122 if (!dev->state_saved)
1123 return;
1124
1125 /* PCI Express register must be restored first */
1126 pci_restore_pcie_state(dev);
1127 pci_restore_ats_state(dev);
1128 pci_restore_vc_state(dev);
1129
1130 pci_cleanup_aer_error_status_regs(dev);
1131
1132 pci_restore_config_space(dev);
1133
1134 pci_restore_pcix_state(dev);
1135 pci_restore_msi_state(dev);
1136
1137 /* Restore ACS and IOV configuration state */
1138 pci_enable_acs(dev);
1139 pci_restore_iov_state(dev);
1140
1141 dev->state_saved = false;
1142}
1143EXPORT_SYMBOL(pci_restore_state);
1144
1145struct pci_saved_state {
1146 u32 config_space[16];
1147 struct pci_cap_saved_data cap[0];
1148};
1149
1150/**
1151 * pci_store_saved_state - Allocate and return an opaque struct containing
1152 * the device saved state.
1153 * @dev: PCI device that we're dealing with
1154 *
1155 * Return NULL if no state or error.
1156 */
1157struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1158{
1159 struct pci_saved_state *state;
1160 struct pci_cap_saved_state *tmp;
1161 struct pci_cap_saved_data *cap;
1162 size_t size;
1163
1164 if (!dev->state_saved)
1165 return NULL;
1166
1167 size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1168
1169 hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1170 size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1171
1172 state = kzalloc(size, GFP_KERNEL);
1173 if (!state)
1174 return NULL;
1175
1176 memcpy(state->config_space, dev->saved_config_space,
1177 sizeof(state->config_space));
1178
1179 cap = state->cap;
1180 hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1181 size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1182 memcpy(cap, &tmp->cap, len);
1183 cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1184 }
1185 /* Empty cap_save terminates list */
1186
1187 return state;
1188}
1189EXPORT_SYMBOL_GPL(pci_store_saved_state);
1190
1191/**
1192 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1193 * @dev: PCI device that we're dealing with
1194 * @state: Saved state returned from pci_store_saved_state()
1195 */
1196int pci_load_saved_state(struct pci_dev *dev,
1197 struct pci_saved_state *state)
1198{
1199 struct pci_cap_saved_data *cap;
1200
1201 dev->state_saved = false;
1202
1203 if (!state)
1204 return 0;
1205
1206 memcpy(dev->saved_config_space, state->config_space,
1207 sizeof(state->config_space));
1208
1209 cap = state->cap;
1210 while (cap->size) {
1211 struct pci_cap_saved_state *tmp;
1212
1213 tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1214 if (!tmp || tmp->cap.size != cap->size)
1215 return -EINVAL;
1216
1217 memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1218 cap = (struct pci_cap_saved_data *)((u8 *)cap +
1219 sizeof(struct pci_cap_saved_data) + cap->size);
1220 }
1221
1222 dev->state_saved = true;
1223 return 0;
1224}
1225EXPORT_SYMBOL_GPL(pci_load_saved_state);
1226
1227/**
1228 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1229 * and free the memory allocated for it.
1230 * @dev: PCI device that we're dealing with
1231 * @state: Pointer to saved state returned from pci_store_saved_state()
1232 */
1233int pci_load_and_free_saved_state(struct pci_dev *dev,
1234 struct pci_saved_state **state)
1235{
1236 int ret = pci_load_saved_state(dev, *state);
1237 kfree(*state);
1238 *state = NULL;
1239 return ret;
1240}
1241EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1242
1243int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1244{
1245 return pci_enable_resources(dev, bars);
1246}
1247
1248static int do_pci_enable_device(struct pci_dev *dev, int bars)
1249{
1250 int err;
1251 struct pci_dev *bridge;
1252 u16 cmd;
1253 u8 pin;
1254
1255 err = pci_set_power_state(dev, PCI_D0);
1256 if (err < 0 && err != -EIO)
1257 return err;
1258
1259 bridge = pci_upstream_bridge(dev);
1260 if (bridge)
1261 pcie_aspm_powersave_config_link(bridge);
1262
1263 err = pcibios_enable_device(dev, bars);
1264 if (err < 0)
1265 return err;
1266 pci_fixup_device(pci_fixup_enable, dev);
1267
1268 if (dev->msi_enabled || dev->msix_enabled)
1269 return 0;
1270
1271 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1272 if (pin) {
1273 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1274 if (cmd & PCI_COMMAND_INTX_DISABLE)
1275 pci_write_config_word(dev, PCI_COMMAND,
1276 cmd & ~PCI_COMMAND_INTX_DISABLE);
1277 }
1278
1279 return 0;
1280}
1281
1282/**
1283 * pci_reenable_device - Resume abandoned device
1284 * @dev: PCI device to be resumed
1285 *
1286 * Note this function is a backend of pci_default_resume and is not supposed
1287 * to be called by normal code, write proper resume handler and use it instead.
1288 */
1289int pci_reenable_device(struct pci_dev *dev)
1290{
1291 if (pci_is_enabled(dev))
1292 return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1293 return 0;
1294}
1295EXPORT_SYMBOL(pci_reenable_device);
1296
1297static void pci_enable_bridge(struct pci_dev *dev)
1298{
1299 struct pci_dev *bridge;
1300 int retval;
1301
1302 bridge = pci_upstream_bridge(dev);
1303 if (bridge)
1304 pci_enable_bridge(bridge);
1305
1306 if (pci_is_enabled(dev)) {
1307 if (!dev->is_busmaster)
1308 pci_set_master(dev);
1309 return;
1310 }
1311
1312 retval = pci_enable_device(dev);
1313 if (retval)
1314 dev_err(&dev->dev, "Error enabling bridge (%d), continuing\n",
1315 retval);
1316 pci_set_master(dev);
1317}
1318
1319static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1320{
1321 struct pci_dev *bridge;
1322 int err;
1323 int i, bars = 0;
1324
1325 /*
1326 * Power state could be unknown at this point, either due to a fresh
1327 * boot or a device removal call. So get the current power state
1328 * so that things like MSI message writing will behave as expected
1329 * (e.g. if the device really is in D0 at enable time).
1330 */
1331 if (dev->pm_cap) {
1332 u16 pmcsr;
1333 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1334 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1335 }
1336
1337 if (atomic_inc_return(&dev->enable_cnt) > 1)
1338 return 0; /* already enabled */
1339
1340 bridge = pci_upstream_bridge(dev);
1341 if (bridge)
1342 pci_enable_bridge(bridge);
1343
1344 /* only skip sriov related */
1345 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1346 if (dev->resource[i].flags & flags)
1347 bars |= (1 << i);
1348 for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1349 if (dev->resource[i].flags & flags)
1350 bars |= (1 << i);
1351
1352 err = do_pci_enable_device(dev, bars);
1353 if (err < 0)
1354 atomic_dec(&dev->enable_cnt);
1355 return err;
1356}
1357
1358/**
1359 * pci_enable_device_io - Initialize a device for use with IO space
1360 * @dev: PCI device to be initialized
1361 *
1362 * Initialize device before it's used by a driver. Ask low-level code
1363 * to enable I/O resources. Wake up the device if it was suspended.
1364 * Beware, this function can fail.
1365 */
1366int pci_enable_device_io(struct pci_dev *dev)
1367{
1368 return pci_enable_device_flags(dev, IORESOURCE_IO);
1369}
1370EXPORT_SYMBOL(pci_enable_device_io);
1371
1372/**
1373 * pci_enable_device_mem - Initialize a device for use with Memory space
1374 * @dev: PCI device to be initialized
1375 *
1376 * Initialize device before it's used by a driver. Ask low-level code
1377 * to enable Memory resources. Wake up the device if it was suspended.
1378 * Beware, this function can fail.
1379 */
1380int pci_enable_device_mem(struct pci_dev *dev)
1381{
1382 return pci_enable_device_flags(dev, IORESOURCE_MEM);
1383}
1384EXPORT_SYMBOL(pci_enable_device_mem);
1385
1386/**
1387 * pci_enable_device - Initialize device before it's used by a driver.
1388 * @dev: PCI device to be initialized
1389 *
1390 * Initialize device before it's used by a driver. Ask low-level code
1391 * to enable I/O and memory. Wake up the device if it was suspended.
1392 * Beware, this function can fail.
1393 *
1394 * Note we don't actually enable the device many times if we call
1395 * this function repeatedly (we just increment the count).
1396 */
1397int pci_enable_device(struct pci_dev *dev)
1398{
1399 return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1400}
1401EXPORT_SYMBOL(pci_enable_device);
1402
1403/*
1404 * Managed PCI resources. This manages device on/off, intx/msi/msix
1405 * on/off and BAR regions. pci_dev itself records msi/msix status, so
1406 * there's no need to track it separately. pci_devres is initialized
1407 * when a device is enabled using managed PCI device enable interface.
1408 */
1409struct pci_devres {
1410 unsigned int enabled:1;
1411 unsigned int pinned:1;
1412 unsigned int orig_intx:1;
1413 unsigned int restore_intx:1;
1414 u32 region_mask;
1415};
1416
1417static void pcim_release(struct device *gendev, void *res)
1418{
1419 struct pci_dev *dev = to_pci_dev(gendev);
1420 struct pci_devres *this = res;
1421 int i;
1422
1423 if (dev->msi_enabled)
1424 pci_disable_msi(dev);
1425 if (dev->msix_enabled)
1426 pci_disable_msix(dev);
1427
1428 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1429 if (this->region_mask & (1 << i))
1430 pci_release_region(dev, i);
1431
1432 if (this->restore_intx)
1433 pci_intx(dev, this->orig_intx);
1434
1435 if (this->enabled && !this->pinned)
1436 pci_disable_device(dev);
1437}
1438
1439static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1440{
1441 struct pci_devres *dr, *new_dr;
1442
1443 dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1444 if (dr)
1445 return dr;
1446
1447 new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1448 if (!new_dr)
1449 return NULL;
1450 return devres_get(&pdev->dev, new_dr, NULL, NULL);
1451}
1452
1453static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1454{
1455 if (pci_is_managed(pdev))
1456 return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1457 return NULL;
1458}
1459
1460/**
1461 * pcim_enable_device - Managed pci_enable_device()
1462 * @pdev: PCI device to be initialized
1463 *
1464 * Managed pci_enable_device().
1465 */
1466int pcim_enable_device(struct pci_dev *pdev)
1467{
1468 struct pci_devres *dr;
1469 int rc;
1470
1471 dr = get_pci_dr(pdev);
1472 if (unlikely(!dr))
1473 return -ENOMEM;
1474 if (dr->enabled)
1475 return 0;
1476
1477 rc = pci_enable_device(pdev);
1478 if (!rc) {
1479 pdev->is_managed = 1;
1480 dr->enabled = 1;
1481 }
1482 return rc;
1483}
1484EXPORT_SYMBOL(pcim_enable_device);
1485
1486/**
1487 * pcim_pin_device - Pin managed PCI device
1488 * @pdev: PCI device to pin
1489 *
1490 * Pin managed PCI device @pdev. Pinned device won't be disabled on
1491 * driver detach. @pdev must have been enabled with
1492 * pcim_enable_device().
1493 */
1494void pcim_pin_device(struct pci_dev *pdev)
1495{
1496 struct pci_devres *dr;
1497
1498 dr = find_pci_dr(pdev);
1499 WARN_ON(!dr || !dr->enabled);
1500 if (dr)
1501 dr->pinned = 1;
1502}
1503EXPORT_SYMBOL(pcim_pin_device);
1504
1505/*
1506 * pcibios_add_device - provide arch specific hooks when adding device dev
1507 * @dev: the PCI device being added
1508 *
1509 * Permits the platform to provide architecture specific functionality when
1510 * devices are added. This is the default implementation. Architecture
1511 * implementations can override this.
1512 */
1513int __weak pcibios_add_device(struct pci_dev *dev)
1514{
1515 return 0;
1516}
1517
1518/**
1519 * pcibios_release_device - provide arch specific hooks when releasing device dev
1520 * @dev: the PCI device being released
1521 *
1522 * Permits the platform to provide architecture specific functionality when
1523 * devices are released. This is the default implementation. Architecture
1524 * implementations can override this.
1525 */
1526void __weak pcibios_release_device(struct pci_dev *dev) {}
1527
1528/**
1529 * pcibios_disable_device - disable arch specific PCI resources for device dev
1530 * @dev: the PCI device to disable
1531 *
1532 * Disables architecture specific PCI resources for the device. This
1533 * is the default implementation. Architecture implementations can
1534 * override this.
1535 */
1536void __weak pcibios_disable_device(struct pci_dev *dev) {}
1537
1538/**
1539 * pcibios_penalize_isa_irq - penalize an ISA IRQ
1540 * @irq: ISA IRQ to penalize
1541 * @active: IRQ active or not
1542 *
1543 * Permits the platform to provide architecture-specific functionality when
1544 * penalizing ISA IRQs. This is the default implementation. Architecture
1545 * implementations can override this.
1546 */
1547void __weak pcibios_penalize_isa_irq(int irq, int active) {}
1548
1549static void do_pci_disable_device(struct pci_dev *dev)
1550{
1551 u16 pci_command;
1552
1553 pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1554 if (pci_command & PCI_COMMAND_MASTER) {
1555 pci_command &= ~PCI_COMMAND_MASTER;
1556 pci_write_config_word(dev, PCI_COMMAND, pci_command);
1557 }
1558
1559 pcibios_disable_device(dev);
1560}
1561
1562/**
1563 * pci_disable_enabled_device - Disable device without updating enable_cnt
1564 * @dev: PCI device to disable
1565 *
1566 * NOTE: This function is a backend of PCI power management routines and is
1567 * not supposed to be called drivers.
1568 */
1569void pci_disable_enabled_device(struct pci_dev *dev)
1570{
1571 if (pci_is_enabled(dev))
1572 do_pci_disable_device(dev);
1573}
1574
1575/**
1576 * pci_disable_device - Disable PCI device after use
1577 * @dev: PCI device to be disabled
1578 *
1579 * Signal to the system that the PCI device is not in use by the system
1580 * anymore. This only involves disabling PCI bus-mastering, if active.
1581 *
1582 * Note we don't actually disable the device until all callers of
1583 * pci_enable_device() have called pci_disable_device().
1584 */
1585void pci_disable_device(struct pci_dev *dev)
1586{
1587 struct pci_devres *dr;
1588
1589 dr = find_pci_dr(dev);
1590 if (dr)
1591 dr->enabled = 0;
1592
1593 dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
1594 "disabling already-disabled device");
1595
1596 if (atomic_dec_return(&dev->enable_cnt) != 0)
1597 return;
1598
1599 do_pci_disable_device(dev);
1600
1601 dev->is_busmaster = 0;
1602}
1603EXPORT_SYMBOL(pci_disable_device);
1604
1605/**
1606 * pcibios_set_pcie_reset_state - set reset state for device dev
1607 * @dev: the PCIe device reset
1608 * @state: Reset state to enter into
1609 *
1610 *
1611 * Sets the PCIe reset state for the device. This is the default
1612 * implementation. Architecture implementations can override this.
1613 */
1614int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1615 enum pcie_reset_state state)
1616{
1617 return -EINVAL;
1618}
1619
1620/**
1621 * pci_set_pcie_reset_state - set reset state for device dev
1622 * @dev: the PCIe device reset
1623 * @state: Reset state to enter into
1624 *
1625 *
1626 * Sets the PCI reset state for the device.
1627 */
1628int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1629{
1630 return pcibios_set_pcie_reset_state(dev, state);
1631}
1632EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
1633
1634/**
1635 * pci_check_pme_status - Check if given device has generated PME.
1636 * @dev: Device to check.
1637 *
1638 * Check the PME status of the device and if set, clear it and clear PME enable
1639 * (if set). Return 'true' if PME status and PME enable were both set or
1640 * 'false' otherwise.
1641 */
1642bool pci_check_pme_status(struct pci_dev *dev)
1643{
1644 int pmcsr_pos;
1645 u16 pmcsr;
1646 bool ret = false;
1647
1648 if (!dev->pm_cap)
1649 return false;
1650
1651 pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
1652 pci_read_config_word(dev, pmcsr_pos, &pmcsr);
1653 if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
1654 return false;
1655
1656 /* Clear PME status. */
1657 pmcsr |= PCI_PM_CTRL_PME_STATUS;
1658 if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
1659 /* Disable PME to avoid interrupt flood. */
1660 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1661 ret = true;
1662 }
1663
1664 pci_write_config_word(dev, pmcsr_pos, pmcsr);
1665
1666 return ret;
1667}
1668
1669/**
1670 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
1671 * @dev: Device to handle.
1672 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
1673 *
1674 * Check if @dev has generated PME and queue a resume request for it in that
1675 * case.
1676 */
1677static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
1678{
1679 if (pme_poll_reset && dev->pme_poll)
1680 dev->pme_poll = false;
1681
1682 if (pci_check_pme_status(dev)) {
1683 pci_wakeup_event(dev);
1684 pm_request_resume(&dev->dev);
1685 }
1686 return 0;
1687}
1688
1689/**
1690 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
1691 * @bus: Top bus of the subtree to walk.
1692 */
1693void pci_pme_wakeup_bus(struct pci_bus *bus)
1694{
1695 if (bus)
1696 pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
1697}
1698
1699
1700/**
1701 * pci_pme_capable - check the capability of PCI device to generate PME#
1702 * @dev: PCI device to handle.
1703 * @state: PCI state from which device will issue PME#.
1704 */
1705bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
1706{
1707 if (!dev->pm_cap)
1708 return false;
1709
1710 return !!(dev->pme_support & (1 << state));
1711}
1712EXPORT_SYMBOL(pci_pme_capable);
1713
1714static void pci_pme_list_scan(struct work_struct *work)
1715{
1716 struct pci_pme_device *pme_dev, *n;
1717
1718 mutex_lock(&pci_pme_list_mutex);
1719 list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
1720 if (pme_dev->dev->pme_poll) {
1721 struct pci_dev *bridge;
1722
1723 bridge = pme_dev->dev->bus->self;
1724 /*
1725 * If bridge is in low power state, the
1726 * configuration space of subordinate devices
1727 * may be not accessible
1728 */
1729 if (bridge && bridge->current_state != PCI_D0)
1730 continue;
1731 pci_pme_wakeup(pme_dev->dev, NULL);
1732 } else {
1733 list_del(&pme_dev->list);
1734 kfree(pme_dev);
1735 }
1736 }
1737 if (!list_empty(&pci_pme_list))
1738 schedule_delayed_work(&pci_pme_work,
1739 msecs_to_jiffies(PME_TIMEOUT));
1740 mutex_unlock(&pci_pme_list_mutex);
1741}
1742
1743static void __pci_pme_active(struct pci_dev *dev, bool enable)
1744{
1745 u16 pmcsr;
1746
1747 if (!dev->pme_support)
1748 return;
1749
1750 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1751 /* Clear PME_Status by writing 1 to it and enable PME# */
1752 pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
1753 if (!enable)
1754 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1755
1756 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1757}
1758
1759/**
1760 * pci_pme_active - enable or disable PCI device's PME# function
1761 * @dev: PCI device to handle.
1762 * @enable: 'true' to enable PME# generation; 'false' to disable it.
1763 *
1764 * The caller must verify that the device is capable of generating PME# before
1765 * calling this function with @enable equal to 'true'.
1766 */
1767void pci_pme_active(struct pci_dev *dev, bool enable)
1768{
1769 __pci_pme_active(dev, enable);
1770
1771 /*
1772 * PCI (as opposed to PCIe) PME requires that the device have
1773 * its PME# line hooked up correctly. Not all hardware vendors
1774 * do this, so the PME never gets delivered and the device
1775 * remains asleep. The easiest way around this is to
1776 * periodically walk the list of suspended devices and check
1777 * whether any have their PME flag set. The assumption is that
1778 * we'll wake up often enough anyway that this won't be a huge
1779 * hit, and the power savings from the devices will still be a
1780 * win.
1781 *
1782 * Although PCIe uses in-band PME message instead of PME# line
1783 * to report PME, PME does not work for some PCIe devices in
1784 * reality. For example, there are devices that set their PME
1785 * status bits, but don't really bother to send a PME message;
1786 * there are PCI Express Root Ports that don't bother to
1787 * trigger interrupts when they receive PME messages from the
1788 * devices below. So PME poll is used for PCIe devices too.
1789 */
1790
1791 if (dev->pme_poll) {
1792 struct pci_pme_device *pme_dev;
1793 if (enable) {
1794 pme_dev = kmalloc(sizeof(struct pci_pme_device),
1795 GFP_KERNEL);
1796 if (!pme_dev) {
1797 dev_warn(&dev->dev, "can't enable PME#\n");
1798 return;
1799 }
1800 pme_dev->dev = dev;
1801 mutex_lock(&pci_pme_list_mutex);
1802 list_add(&pme_dev->list, &pci_pme_list);
1803 if (list_is_singular(&pci_pme_list))
1804 schedule_delayed_work(&pci_pme_work,
1805 msecs_to_jiffies(PME_TIMEOUT));
1806 mutex_unlock(&pci_pme_list_mutex);
1807 } else {
1808 mutex_lock(&pci_pme_list_mutex);
1809 list_for_each_entry(pme_dev, &pci_pme_list, list) {
1810 if (pme_dev->dev == dev) {
1811 list_del(&pme_dev->list);
1812 kfree(pme_dev);
1813 break;
1814 }
1815 }
1816 mutex_unlock(&pci_pme_list_mutex);
1817 }
1818 }
1819
1820 dev_dbg(&dev->dev, "PME# %s\n", enable ? "enabled" : "disabled");
1821}
1822EXPORT_SYMBOL(pci_pme_active);
1823
1824/**
1825 * __pci_enable_wake - enable PCI device as wakeup event source
1826 * @dev: PCI device affected
1827 * @state: PCI state from which device will issue wakeup events
1828 * @runtime: True if the events are to be generated at run time
1829 * @enable: True to enable event generation; false to disable
1830 *
1831 * This enables the device as a wakeup event source, or disables it.
1832 * When such events involves platform-specific hooks, those hooks are
1833 * called automatically by this routine.
1834 *
1835 * Devices with legacy power management (no standard PCI PM capabilities)
1836 * always require such platform hooks.
1837 *
1838 * RETURN VALUE:
1839 * 0 is returned on success
1840 * -EINVAL is returned if device is not supposed to wake up the system
1841 * Error code depending on the platform is returned if both the platform and
1842 * the native mechanism fail to enable the generation of wake-up events
1843 */
1844int __pci_enable_wake(struct pci_dev *dev, pci_power_t state,
1845 bool runtime, bool enable)
1846{
1847 int ret = 0;
1848
1849 if (enable && !runtime && !device_may_wakeup(&dev->dev))
1850 return -EINVAL;
1851
1852 /* Don't do the same thing twice in a row for one device. */
1853 if (!!enable == !!dev->wakeup_prepared)
1854 return 0;
1855
1856 /*
1857 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
1858 * Anderson we should be doing PME# wake enable followed by ACPI wake
1859 * enable. To disable wake-up we call the platform first, for symmetry.
1860 */
1861
1862 if (enable) {
1863 int error;
1864
1865 if (pci_pme_capable(dev, state))
1866 pci_pme_active(dev, true);
1867 else
1868 ret = 1;
1869 error = runtime ? platform_pci_run_wake(dev, true) :
1870 platform_pci_sleep_wake(dev, true);
1871 if (ret)
1872 ret = error;
1873 if (!ret)
1874 dev->wakeup_prepared = true;
1875 } else {
1876 if (runtime)
1877 platform_pci_run_wake(dev, false);
1878 else
1879 platform_pci_sleep_wake(dev, false);
1880 pci_pme_active(dev, false);
1881 dev->wakeup_prepared = false;
1882 }
1883
1884 return ret;
1885}
1886EXPORT_SYMBOL(__pci_enable_wake);
1887
1888/**
1889 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
1890 * @dev: PCI device to prepare
1891 * @enable: True to enable wake-up event generation; false to disable
1892 *
1893 * Many drivers want the device to wake up the system from D3_hot or D3_cold
1894 * and this function allows them to set that up cleanly - pci_enable_wake()
1895 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
1896 * ordering constraints.
1897 *
1898 * This function only returns error code if the device is not capable of
1899 * generating PME# from both D3_hot and D3_cold, and the platform is unable to
1900 * enable wake-up power for it.
1901 */
1902int pci_wake_from_d3(struct pci_dev *dev, bool enable)
1903{
1904 return pci_pme_capable(dev, PCI_D3cold) ?
1905 pci_enable_wake(dev, PCI_D3cold, enable) :
1906 pci_enable_wake(dev, PCI_D3hot, enable);
1907}
1908EXPORT_SYMBOL(pci_wake_from_d3);
1909
1910/**
1911 * pci_target_state - find an appropriate low power state for a given PCI dev
1912 * @dev: PCI device
1913 *
1914 * Use underlying platform code to find a supported low power state for @dev.
1915 * If the platform can't manage @dev, return the deepest state from which it
1916 * can generate wake events, based on any available PME info.
1917 */
1918static pci_power_t pci_target_state(struct pci_dev *dev)
1919{
1920 pci_power_t target_state = PCI_D3hot;
1921
1922 if (platform_pci_power_manageable(dev)) {
1923 /*
1924 * Call the platform to choose the target state of the device
1925 * and enable wake-up from this state if supported.
1926 */
1927 pci_power_t state = platform_pci_choose_state(dev);
1928
1929 switch (state) {
1930 case PCI_POWER_ERROR:
1931 case PCI_UNKNOWN:
1932 break;
1933 case PCI_D1:
1934 case PCI_D2:
1935 if (pci_no_d1d2(dev))
1936 break;
1937 default:
1938 target_state = state;
1939 }
1940 } else if (!dev->pm_cap) {
1941 target_state = PCI_D0;
1942 } else if (device_may_wakeup(&dev->dev)) {
1943 /*
1944 * Find the deepest state from which the device can generate
1945 * wake-up events, make it the target state and enable device
1946 * to generate PME#.
1947 */
1948 if (dev->pme_support) {
1949 while (target_state
1950 && !(dev->pme_support & (1 << target_state)))
1951 target_state--;
1952 }
1953 }
1954
1955 return target_state;
1956}
1957
1958/**
1959 * pci_prepare_to_sleep - prepare PCI device for system-wide transition into a sleep state
1960 * @dev: Device to handle.
1961 *
1962 * Choose the power state appropriate for the device depending on whether
1963 * it can wake up the system and/or is power manageable by the platform
1964 * (PCI_D3hot is the default) and put the device into that state.
1965 */
1966int pci_prepare_to_sleep(struct pci_dev *dev)
1967{
1968 pci_power_t target_state = pci_target_state(dev);
1969 int error;
1970
1971 if (target_state == PCI_POWER_ERROR)
1972 return -EIO;
1973
1974 pci_enable_wake(dev, target_state, device_may_wakeup(&dev->dev));
1975
1976 error = pci_set_power_state(dev, target_state);
1977
1978 if (error)
1979 pci_enable_wake(dev, target_state, false);
1980
1981 return error;
1982}
1983EXPORT_SYMBOL(pci_prepare_to_sleep);
1984
1985/**
1986 * pci_back_from_sleep - turn PCI device on during system-wide transition into working state
1987 * @dev: Device to handle.
1988 *
1989 * Disable device's system wake-up capability and put it into D0.
1990 */
1991int pci_back_from_sleep(struct pci_dev *dev)
1992{
1993 pci_enable_wake(dev, PCI_D0, false);
1994 return pci_set_power_state(dev, PCI_D0);
1995}
1996EXPORT_SYMBOL(pci_back_from_sleep);
1997
1998/**
1999 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2000 * @dev: PCI device being suspended.
2001 *
2002 * Prepare @dev to generate wake-up events at run time and put it into a low
2003 * power state.
2004 */
2005int pci_finish_runtime_suspend(struct pci_dev *dev)
2006{
2007 pci_power_t target_state = pci_target_state(dev);
2008 int error;
2009
2010 if (target_state == PCI_POWER_ERROR)
2011 return -EIO;
2012
2013 dev->runtime_d3cold = target_state == PCI_D3cold;
2014
2015 __pci_enable_wake(dev, target_state, true, pci_dev_run_wake(dev));
2016
2017 error = pci_set_power_state(dev, target_state);
2018
2019 if (error) {
2020 __pci_enable_wake(dev, target_state, true, false);
2021 dev->runtime_d3cold = false;
2022 }
2023
2024 return error;
2025}
2026
2027/**
2028 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2029 * @dev: Device to check.
2030 *
2031 * Return true if the device itself is capable of generating wake-up events
2032 * (through the platform or using the native PCIe PME) or if the device supports
2033 * PME and one of its upstream bridges can generate wake-up events.
2034 */
2035bool pci_dev_run_wake(struct pci_dev *dev)
2036{
2037 struct pci_bus *bus = dev->bus;
2038
2039 if (device_run_wake(&dev->dev))
2040 return true;
2041
2042 if (!dev->pme_support)
2043 return false;
2044
2045 while (bus->parent) {
2046 struct pci_dev *bridge = bus->self;
2047
2048 if (device_run_wake(&bridge->dev))
2049 return true;
2050
2051 bus = bus->parent;
2052 }
2053
2054 /* We have reached the root bus. */
2055 if (bus->bridge)
2056 return device_run_wake(bus->bridge);
2057
2058 return false;
2059}
2060EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2061
2062/**
2063 * pci_dev_keep_suspended - Check if the device can stay in the suspended state.
2064 * @pci_dev: Device to check.
2065 *
2066 * Return 'true' if the device is runtime-suspended, it doesn't have to be
2067 * reconfigured due to wakeup settings difference between system and runtime
2068 * suspend and the current power state of it is suitable for the upcoming
2069 * (system) transition.
2070 *
2071 * If the device is not configured for system wakeup, disable PME for it before
2072 * returning 'true' to prevent it from waking up the system unnecessarily.
2073 */
2074bool pci_dev_keep_suspended(struct pci_dev *pci_dev)
2075{
2076 struct device *dev = &pci_dev->dev;
2077
2078 if (!pm_runtime_suspended(dev)
2079 || pci_target_state(pci_dev) != pci_dev->current_state
2080 || platform_pci_need_resume(pci_dev))
2081 return false;
2082
2083 /*
2084 * At this point the device is good to go unless it's been configured
2085 * to generate PME at the runtime suspend time, but it is not supposed
2086 * to wake up the system. In that case, simply disable PME for it
2087 * (it will have to be re-enabled on exit from system resume).
2088 *
2089 * If the device's power state is D3cold and the platform check above
2090 * hasn't triggered, the device's configuration is suitable and we don't
2091 * need to manipulate it at all.
2092 */
2093 spin_lock_irq(&dev->power.lock);
2094
2095 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold &&
2096 !device_may_wakeup(dev))
2097 __pci_pme_active(pci_dev, false);
2098
2099 spin_unlock_irq(&dev->power.lock);
2100 return true;
2101}
2102
2103/**
2104 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2105 * @pci_dev: Device to handle.
2106 *
2107 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2108 * it might have been disabled during the prepare phase of system suspend if
2109 * the device was not configured for system wakeup.
2110 */
2111void pci_dev_complete_resume(struct pci_dev *pci_dev)
2112{
2113 struct device *dev = &pci_dev->dev;
2114
2115 if (!pci_dev_run_wake(pci_dev))
2116 return;
2117
2118 spin_lock_irq(&dev->power.lock);
2119
2120 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2121 __pci_pme_active(pci_dev, true);
2122
2123 spin_unlock_irq(&dev->power.lock);
2124}
2125
2126void pci_config_pm_runtime_get(struct pci_dev *pdev)
2127{
2128 struct device *dev = &pdev->dev;
2129 struct device *parent = dev->parent;
2130
2131 if (parent)
2132 pm_runtime_get_sync(parent);
2133 pm_runtime_get_noresume(dev);
2134 /*
2135 * pdev->current_state is set to PCI_D3cold during suspending,
2136 * so wait until suspending completes
2137 */
2138 pm_runtime_barrier(dev);
2139 /*
2140 * Only need to resume devices in D3cold, because config
2141 * registers are still accessible for devices suspended but
2142 * not in D3cold.
2143 */
2144 if (pdev->current_state == PCI_D3cold)
2145 pm_runtime_resume(dev);
2146}
2147
2148void pci_config_pm_runtime_put(struct pci_dev *pdev)
2149{
2150 struct device *dev = &pdev->dev;
2151 struct device *parent = dev->parent;
2152
2153 pm_runtime_put(dev);
2154 if (parent)
2155 pm_runtime_put_sync(parent);
2156}
2157
2158/**
2159 * pci_pm_init - Initialize PM functions of given PCI device
2160 * @dev: PCI device to handle.
2161 */
2162void pci_pm_init(struct pci_dev *dev)
2163{
2164 int pm;
2165 u16 pmc;
2166
2167 pm_runtime_forbid(&dev->dev);
2168 pm_runtime_set_active(&dev->dev);
2169 pm_runtime_enable(&dev->dev);
2170 device_enable_async_suspend(&dev->dev);
2171 dev->wakeup_prepared = false;
2172
2173 dev->pm_cap = 0;
2174 dev->pme_support = 0;
2175
2176 /* find PCI PM capability in list */
2177 pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2178 if (!pm)
2179 return;
2180 /* Check device's ability to generate PME# */
2181 pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2182
2183 if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2184 dev_err(&dev->dev, "unsupported PM cap regs version (%u)\n",
2185 pmc & PCI_PM_CAP_VER_MASK);
2186 return;
2187 }
2188
2189 dev->pm_cap = pm;
2190 dev->d3_delay = PCI_PM_D3_WAIT;
2191 dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2192 dev->d3cold_allowed = true;
2193
2194 dev->d1_support = false;
2195 dev->d2_support = false;
2196 if (!pci_no_d1d2(dev)) {
2197 if (pmc & PCI_PM_CAP_D1)
2198 dev->d1_support = true;
2199 if (pmc & PCI_PM_CAP_D2)
2200 dev->d2_support = true;
2201
2202 if (dev->d1_support || dev->d2_support)
2203 dev_printk(KERN_DEBUG, &dev->dev, "supports%s%s\n",
2204 dev->d1_support ? " D1" : "",
2205 dev->d2_support ? " D2" : "");
2206 }
2207
2208 pmc &= PCI_PM_CAP_PME_MASK;
2209 if (pmc) {
2210 dev_printk(KERN_DEBUG, &dev->dev,
2211 "PME# supported from%s%s%s%s%s\n",
2212 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2213 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2214 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2215 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2216 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2217 dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2218 dev->pme_poll = true;
2219 /*
2220 * Make device's PM flags reflect the wake-up capability, but
2221 * let the user space enable it to wake up the system as needed.
2222 */
2223 device_set_wakeup_capable(&dev->dev, true);
2224 /* Disable the PME# generation functionality */
2225 pci_pme_active(dev, false);
2226 }
2227}
2228
2229static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2230{
2231 unsigned long flags = IORESOURCE_PCI_FIXED;
2232
2233 switch (prop) {
2234 case PCI_EA_P_MEM:
2235 case PCI_EA_P_VF_MEM:
2236 flags |= IORESOURCE_MEM;
2237 break;
2238 case PCI_EA_P_MEM_PREFETCH:
2239 case PCI_EA_P_VF_MEM_PREFETCH:
2240 flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2241 break;
2242 case PCI_EA_P_IO:
2243 flags |= IORESOURCE_IO;
2244 break;
2245 default:
2246 return 0;
2247 }
2248
2249 return flags;
2250}
2251
2252static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2253 u8 prop)
2254{
2255 if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2256 return &dev->resource[bei];
2257#ifdef CONFIG_PCI_IOV
2258 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2259 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2260 return &dev->resource[PCI_IOV_RESOURCES +
2261 bei - PCI_EA_BEI_VF_BAR0];
2262#endif
2263 else if (bei == PCI_EA_BEI_ROM)
2264 return &dev->resource[PCI_ROM_RESOURCE];
2265 else
2266 return NULL;
2267}
2268
2269/* Read an Enhanced Allocation (EA) entry */
2270static int pci_ea_read(struct pci_dev *dev, int offset)
2271{
2272 struct resource *res;
2273 int ent_size, ent_offset = offset;
2274 resource_size_t start, end;
2275 unsigned long flags;
2276 u32 dw0, bei, base, max_offset;
2277 u8 prop;
2278 bool support_64 = (sizeof(resource_size_t) >= 8);
2279
2280 pci_read_config_dword(dev, ent_offset, &dw0);
2281 ent_offset += 4;
2282
2283 /* Entry size field indicates DWORDs after 1st */
2284 ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2285
2286 if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2287 goto out;
2288
2289 bei = (dw0 & PCI_EA_BEI) >> 4;
2290 prop = (dw0 & PCI_EA_PP) >> 8;
2291
2292 /*
2293 * If the Property is in the reserved range, try the Secondary
2294 * Property instead.
2295 */
2296 if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2297 prop = (dw0 & PCI_EA_SP) >> 16;
2298 if (prop > PCI_EA_P_BRIDGE_IO)
2299 goto out;
2300
2301 res = pci_ea_get_resource(dev, bei, prop);
2302 if (!res) {
2303 dev_err(&dev->dev, "Unsupported EA entry BEI: %u\n", bei);
2304 goto out;
2305 }
2306
2307 flags = pci_ea_flags(dev, prop);
2308 if (!flags) {
2309 dev_err(&dev->dev, "Unsupported EA properties: %#x\n", prop);
2310 goto out;
2311 }
2312
2313 /* Read Base */
2314 pci_read_config_dword(dev, ent_offset, &base);
2315 start = (base & PCI_EA_FIELD_MASK);
2316 ent_offset += 4;
2317
2318 /* Read MaxOffset */
2319 pci_read_config_dword(dev, ent_offset, &max_offset);
2320 ent_offset += 4;
2321
2322 /* Read Base MSBs (if 64-bit entry) */
2323 if (base & PCI_EA_IS_64) {
2324 u32 base_upper;
2325
2326 pci_read_config_dword(dev, ent_offset, &base_upper);
2327 ent_offset += 4;
2328
2329 flags |= IORESOURCE_MEM_64;
2330
2331 /* entry starts above 32-bit boundary, can't use */
2332 if (!support_64 && base_upper)
2333 goto out;
2334
2335 if (support_64)
2336 start |= ((u64)base_upper << 32);
2337 }
2338
2339 end = start + (max_offset | 0x03);
2340
2341 /* Read MaxOffset MSBs (if 64-bit entry) */
2342 if (max_offset & PCI_EA_IS_64) {
2343 u32 max_offset_upper;
2344
2345 pci_read_config_dword(dev, ent_offset, &max_offset_upper);
2346 ent_offset += 4;
2347
2348 flags |= IORESOURCE_MEM_64;
2349
2350 /* entry too big, can't use */
2351 if (!support_64 && max_offset_upper)
2352 goto out;
2353
2354 if (support_64)
2355 end += ((u64)max_offset_upper << 32);
2356 }
2357
2358 if (end < start) {
2359 dev_err(&dev->dev, "EA Entry crosses address boundary\n");
2360 goto out;
2361 }
2362
2363 if (ent_size != ent_offset - offset) {
2364 dev_err(&dev->dev,
2365 "EA Entry Size (%d) does not match length read (%d)\n",
2366 ent_size, ent_offset - offset);
2367 goto out;
2368 }
2369
2370 res->name = pci_name(dev);
2371 res->start = start;
2372 res->end = end;
2373 res->flags = flags;
2374
2375 if (bei <= PCI_EA_BEI_BAR5)
2376 dev_printk(KERN_DEBUG, &dev->dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2377 bei, res, prop);
2378 else if (bei == PCI_EA_BEI_ROM)
2379 dev_printk(KERN_DEBUG, &dev->dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
2380 res, prop);
2381 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
2382 dev_printk(KERN_DEBUG, &dev->dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2383 bei - PCI_EA_BEI_VF_BAR0, res, prop);
2384 else
2385 dev_printk(KERN_DEBUG, &dev->dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
2386 bei, res, prop);
2387
2388out:
2389 return offset + ent_size;
2390}
2391
2392/* Enhanced Allocation Initalization */
2393void pci_ea_init(struct pci_dev *dev)
2394{
2395 int ea;
2396 u8 num_ent;
2397 int offset;
2398 int i;
2399
2400 /* find PCI EA capability in list */
2401 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
2402 if (!ea)
2403 return;
2404
2405 /* determine the number of entries */
2406 pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
2407 &num_ent);
2408 num_ent &= PCI_EA_NUM_ENT_MASK;
2409
2410 offset = ea + PCI_EA_FIRST_ENT;
2411
2412 /* Skip DWORD 2 for type 1 functions */
2413 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
2414 offset += 4;
2415
2416 /* parse each EA entry */
2417 for (i = 0; i < num_ent; ++i)
2418 offset = pci_ea_read(dev, offset);
2419}
2420
2421static void pci_add_saved_cap(struct pci_dev *pci_dev,
2422 struct pci_cap_saved_state *new_cap)
2423{
2424 hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
2425}
2426
2427/**
2428 * _pci_add_cap_save_buffer - allocate buffer for saving given
2429 * capability registers
2430 * @dev: the PCI device
2431 * @cap: the capability to allocate the buffer for
2432 * @extended: Standard or Extended capability ID
2433 * @size: requested size of the buffer
2434 */
2435static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
2436 bool extended, unsigned int size)
2437{
2438 int pos;
2439 struct pci_cap_saved_state *save_state;
2440
2441 if (extended)
2442 pos = pci_find_ext_capability(dev, cap);
2443 else
2444 pos = pci_find_capability(dev, cap);
2445
2446 if (!pos)
2447 return 0;
2448
2449 save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
2450 if (!save_state)
2451 return -ENOMEM;
2452
2453 save_state->cap.cap_nr = cap;
2454 save_state->cap.cap_extended = extended;
2455 save_state->cap.size = size;
2456 pci_add_saved_cap(dev, save_state);
2457
2458 return 0;
2459}
2460
2461int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
2462{
2463 return _pci_add_cap_save_buffer(dev, cap, false, size);
2464}
2465
2466int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
2467{
2468 return _pci_add_cap_save_buffer(dev, cap, true, size);
2469}
2470
2471/**
2472 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
2473 * @dev: the PCI device
2474 */
2475void pci_allocate_cap_save_buffers(struct pci_dev *dev)
2476{
2477 int error;
2478
2479 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
2480 PCI_EXP_SAVE_REGS * sizeof(u16));
2481 if (error)
2482 dev_err(&dev->dev,
2483 "unable to preallocate PCI Express save buffer\n");
2484
2485 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
2486 if (error)
2487 dev_err(&dev->dev,
2488 "unable to preallocate PCI-X save buffer\n");
2489
2490 pci_allocate_vc_save_buffers(dev);
2491}
2492
2493void pci_free_cap_save_buffers(struct pci_dev *dev)
2494{
2495 struct pci_cap_saved_state *tmp;
2496 struct hlist_node *n;
2497
2498 hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
2499 kfree(tmp);
2500}
2501
2502/**
2503 * pci_configure_ari - enable or disable ARI forwarding
2504 * @dev: the PCI device
2505 *
2506 * If @dev and its upstream bridge both support ARI, enable ARI in the
2507 * bridge. Otherwise, disable ARI in the bridge.
2508 */
2509void pci_configure_ari(struct pci_dev *dev)
2510{
2511 u32 cap;
2512 struct pci_dev *bridge;
2513
2514 if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
2515 return;
2516
2517 bridge = dev->bus->self;
2518 if (!bridge)
2519 return;
2520
2521 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
2522 if (!(cap & PCI_EXP_DEVCAP2_ARI))
2523 return;
2524
2525 if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
2526 pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
2527 PCI_EXP_DEVCTL2_ARI);
2528 bridge->ari_enabled = 1;
2529 } else {
2530 pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
2531 PCI_EXP_DEVCTL2_ARI);
2532 bridge->ari_enabled = 0;
2533 }
2534}
2535
2536static int pci_acs_enable;
2537
2538/**
2539 * pci_request_acs - ask for ACS to be enabled if supported
2540 */
2541void pci_request_acs(void)
2542{
2543 pci_acs_enable = 1;
2544}
2545
2546/**
2547 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilites
2548 * @dev: the PCI device
2549 */
2550static int pci_std_enable_acs(struct pci_dev *dev)
2551{
2552 int pos;
2553 u16 cap;
2554 u16 ctrl;
2555
2556 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
2557 if (!pos)
2558 return -ENODEV;
2559
2560 pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
2561 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
2562
2563 /* Source Validation */
2564 ctrl |= (cap & PCI_ACS_SV);
2565
2566 /* P2P Request Redirect */
2567 ctrl |= (cap & PCI_ACS_RR);
2568
2569 /* P2P Completion Redirect */
2570 ctrl |= (cap & PCI_ACS_CR);
2571
2572 /* Upstream Forwarding */
2573 ctrl |= (cap & PCI_ACS_UF);
2574
2575 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
2576
2577 return 0;
2578}
2579
2580/**
2581 * pci_enable_acs - enable ACS if hardware support it
2582 * @dev: the PCI device
2583 */
2584void pci_enable_acs(struct pci_dev *dev)
2585{
2586 if (!pci_acs_enable)
2587 return;
2588
2589 if (!pci_std_enable_acs(dev))
2590 return;
2591
2592 pci_dev_specific_enable_acs(dev);
2593}
2594
2595static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
2596{
2597 int pos;
2598 u16 cap, ctrl;
2599
2600 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
2601 if (!pos)
2602 return false;
2603
2604 /*
2605 * Except for egress control, capabilities are either required
2606 * or only required if controllable. Features missing from the
2607 * capability field can therefore be assumed as hard-wired enabled.
2608 */
2609 pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
2610 acs_flags &= (cap | PCI_ACS_EC);
2611
2612 pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
2613 return (ctrl & acs_flags) == acs_flags;
2614}
2615
2616/**
2617 * pci_acs_enabled - test ACS against required flags for a given device
2618 * @pdev: device to test
2619 * @acs_flags: required PCI ACS flags
2620 *
2621 * Return true if the device supports the provided flags. Automatically
2622 * filters out flags that are not implemented on multifunction devices.
2623 *
2624 * Note that this interface checks the effective ACS capabilities of the
2625 * device rather than the actual capabilities. For instance, most single
2626 * function endpoints are not required to support ACS because they have no
2627 * opportunity for peer-to-peer access. We therefore return 'true'
2628 * regardless of whether the device exposes an ACS capability. This makes
2629 * it much easier for callers of this function to ignore the actual type
2630 * or topology of the device when testing ACS support.
2631 */
2632bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
2633{
2634 int ret;
2635
2636 ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
2637 if (ret >= 0)
2638 return ret > 0;
2639
2640 /*
2641 * Conventional PCI and PCI-X devices never support ACS, either
2642 * effectively or actually. The shared bus topology implies that
2643 * any device on the bus can receive or snoop DMA.
2644 */
2645 if (!pci_is_pcie(pdev))
2646 return false;
2647
2648 switch (pci_pcie_type(pdev)) {
2649 /*
2650 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
2651 * but since their primary interface is PCI/X, we conservatively
2652 * handle them as we would a non-PCIe device.
2653 */
2654 case PCI_EXP_TYPE_PCIE_BRIDGE:
2655 /*
2656 * PCIe 3.0, 6.12.1 excludes ACS on these devices. "ACS is never
2657 * applicable... must never implement an ACS Extended Capability...".
2658 * This seems arbitrary, but we take a conservative interpretation
2659 * of this statement.
2660 */
2661 case PCI_EXP_TYPE_PCI_BRIDGE:
2662 case PCI_EXP_TYPE_RC_EC:
2663 return false;
2664 /*
2665 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
2666 * implement ACS in order to indicate their peer-to-peer capabilities,
2667 * regardless of whether they are single- or multi-function devices.
2668 */
2669 case PCI_EXP_TYPE_DOWNSTREAM:
2670 case PCI_EXP_TYPE_ROOT_PORT:
2671 return pci_acs_flags_enabled(pdev, acs_flags);
2672 /*
2673 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
2674 * implemented by the remaining PCIe types to indicate peer-to-peer
2675 * capabilities, but only when they are part of a multifunction
2676 * device. The footnote for section 6.12 indicates the specific
2677 * PCIe types included here.
2678 */
2679 case PCI_EXP_TYPE_ENDPOINT:
2680 case PCI_EXP_TYPE_UPSTREAM:
2681 case PCI_EXP_TYPE_LEG_END:
2682 case PCI_EXP_TYPE_RC_END:
2683 if (!pdev->multifunction)
2684 break;
2685
2686 return pci_acs_flags_enabled(pdev, acs_flags);
2687 }
2688
2689 /*
2690 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
2691 * to single function devices with the exception of downstream ports.
2692 */
2693 return true;
2694}
2695
2696/**
2697 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
2698 * @start: starting downstream device
2699 * @end: ending upstream device or NULL to search to the root bus
2700 * @acs_flags: required flags
2701 *
2702 * Walk up a device tree from start to end testing PCI ACS support. If
2703 * any step along the way does not support the required flags, return false.
2704 */
2705bool pci_acs_path_enabled(struct pci_dev *start,
2706 struct pci_dev *end, u16 acs_flags)
2707{
2708 struct pci_dev *pdev, *parent = start;
2709
2710 do {
2711 pdev = parent;
2712
2713 if (!pci_acs_enabled(pdev, acs_flags))
2714 return false;
2715
2716 if (pci_is_root_bus(pdev->bus))
2717 return (end == NULL);
2718
2719 parent = pdev->bus->self;
2720 } while (pdev != end);
2721
2722 return true;
2723}
2724
2725/**
2726 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
2727 * @dev: the PCI device
2728 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
2729 *
2730 * Perform INTx swizzling for a device behind one level of bridge. This is
2731 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
2732 * behind bridges on add-in cards. For devices with ARI enabled, the slot
2733 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
2734 * the PCI Express Base Specification, Revision 2.1)
2735 */
2736u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
2737{
2738 int slot;
2739
2740 if (pci_ari_enabled(dev->bus))
2741 slot = 0;
2742 else
2743 slot = PCI_SLOT(dev->devfn);
2744
2745 return (((pin - 1) + slot) % 4) + 1;
2746}
2747
2748int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
2749{
2750 u8 pin;
2751
2752 pin = dev->pin;
2753 if (!pin)
2754 return -1;
2755
2756 while (!pci_is_root_bus(dev->bus)) {
2757 pin = pci_swizzle_interrupt_pin(dev, pin);
2758 dev = dev->bus->self;
2759 }
2760 *bridge = dev;
2761 return pin;
2762}
2763
2764/**
2765 * pci_common_swizzle - swizzle INTx all the way to root bridge
2766 * @dev: the PCI device
2767 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
2768 *
2769 * Perform INTx swizzling for a device. This traverses through all PCI-to-PCI
2770 * bridges all the way up to a PCI root bus.
2771 */
2772u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
2773{
2774 u8 pin = *pinp;
2775
2776 while (!pci_is_root_bus(dev->bus)) {
2777 pin = pci_swizzle_interrupt_pin(dev, pin);
2778 dev = dev->bus->self;
2779 }
2780 *pinp = pin;
2781 return PCI_SLOT(dev->devfn);
2782}
2783EXPORT_SYMBOL_GPL(pci_common_swizzle);
2784
2785/**
2786 * pci_release_region - Release a PCI bar
2787 * @pdev: PCI device whose resources were previously reserved by pci_request_region
2788 * @bar: BAR to release
2789 *
2790 * Releases the PCI I/O and memory resources previously reserved by a
2791 * successful call to pci_request_region. Call this function only
2792 * after all use of the PCI regions has ceased.
2793 */
2794void pci_release_region(struct pci_dev *pdev, int bar)
2795{
2796 struct pci_devres *dr;
2797
2798 if (pci_resource_len(pdev, bar) == 0)
2799 return;
2800 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
2801 release_region(pci_resource_start(pdev, bar),
2802 pci_resource_len(pdev, bar));
2803 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
2804 release_mem_region(pci_resource_start(pdev, bar),
2805 pci_resource_len(pdev, bar));
2806
2807 dr = find_pci_dr(pdev);
2808 if (dr)
2809 dr->region_mask &= ~(1 << bar);
2810}
2811EXPORT_SYMBOL(pci_release_region);
2812
2813/**
2814 * __pci_request_region - Reserved PCI I/O and memory resource
2815 * @pdev: PCI device whose resources are to be reserved
2816 * @bar: BAR to be reserved
2817 * @res_name: Name to be associated with resource.
2818 * @exclusive: whether the region access is exclusive or not
2819 *
2820 * Mark the PCI region associated with PCI device @pdev BR @bar as
2821 * being reserved by owner @res_name. Do not access any
2822 * address inside the PCI regions unless this call returns
2823 * successfully.
2824 *
2825 * If @exclusive is set, then the region is marked so that userspace
2826 * is explicitly not allowed to map the resource via /dev/mem or
2827 * sysfs MMIO access.
2828 *
2829 * Returns 0 on success, or %EBUSY on error. A warning
2830 * message is also printed on failure.
2831 */
2832static int __pci_request_region(struct pci_dev *pdev, int bar,
2833 const char *res_name, int exclusive)
2834{
2835 struct pci_devres *dr;
2836
2837 if (pci_resource_len(pdev, bar) == 0)
2838 return 0;
2839
2840 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
2841 if (!request_region(pci_resource_start(pdev, bar),
2842 pci_resource_len(pdev, bar), res_name))
2843 goto err_out;
2844 } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
2845 if (!__request_mem_region(pci_resource_start(pdev, bar),
2846 pci_resource_len(pdev, bar), res_name,
2847 exclusive))
2848 goto err_out;
2849 }
2850
2851 dr = find_pci_dr(pdev);
2852 if (dr)
2853 dr->region_mask |= 1 << bar;
2854
2855 return 0;
2856
2857err_out:
2858 dev_warn(&pdev->dev, "BAR %d: can't reserve %pR\n", bar,
2859 &pdev->resource[bar]);
2860 return -EBUSY;
2861}
2862
2863/**
2864 * pci_request_region - Reserve PCI I/O and memory resource
2865 * @pdev: PCI device whose resources are to be reserved
2866 * @bar: BAR to be reserved
2867 * @res_name: Name to be associated with resource
2868 *
2869 * Mark the PCI region associated with PCI device @pdev BAR @bar as
2870 * being reserved by owner @res_name. Do not access any
2871 * address inside the PCI regions unless this call returns
2872 * successfully.
2873 *
2874 * Returns 0 on success, or %EBUSY on error. A warning
2875 * message is also printed on failure.
2876 */
2877int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
2878{
2879 return __pci_request_region(pdev, bar, res_name, 0);
2880}
2881EXPORT_SYMBOL(pci_request_region);
2882
2883/**
2884 * pci_request_region_exclusive - Reserved PCI I/O and memory resource
2885 * @pdev: PCI device whose resources are to be reserved
2886 * @bar: BAR to be reserved
2887 * @res_name: Name to be associated with resource.
2888 *
2889 * Mark the PCI region associated with PCI device @pdev BR @bar as
2890 * being reserved by owner @res_name. Do not access any
2891 * address inside the PCI regions unless this call returns
2892 * successfully.
2893 *
2894 * Returns 0 on success, or %EBUSY on error. A warning
2895 * message is also printed on failure.
2896 *
2897 * The key difference that _exclusive makes it that userspace is
2898 * explicitly not allowed to map the resource via /dev/mem or
2899 * sysfs.
2900 */
2901int pci_request_region_exclusive(struct pci_dev *pdev, int bar,
2902 const char *res_name)
2903{
2904 return __pci_request_region(pdev, bar, res_name, IORESOURCE_EXCLUSIVE);
2905}
2906EXPORT_SYMBOL(pci_request_region_exclusive);
2907
2908/**
2909 * pci_release_selected_regions - Release selected PCI I/O and memory resources
2910 * @pdev: PCI device whose resources were previously reserved
2911 * @bars: Bitmask of BARs to be released
2912 *
2913 * Release selected PCI I/O and memory resources previously reserved.
2914 * Call this function only after all use of the PCI regions has ceased.
2915 */
2916void pci_release_selected_regions(struct pci_dev *pdev, int bars)
2917{
2918 int i;
2919
2920 for (i = 0; i < 6; i++)
2921 if (bars & (1 << i))
2922 pci_release_region(pdev, i);
2923}
2924EXPORT_SYMBOL(pci_release_selected_regions);
2925
2926static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
2927 const char *res_name, int excl)
2928{
2929 int i;
2930
2931 for (i = 0; i < 6; i++)
2932 if (bars & (1 << i))
2933 if (__pci_request_region(pdev, i, res_name, excl))
2934 goto err_out;
2935 return 0;
2936
2937err_out:
2938 while (--i >= 0)
2939 if (bars & (1 << i))
2940 pci_release_region(pdev, i);
2941
2942 return -EBUSY;
2943}
2944
2945
2946/**
2947 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
2948 * @pdev: PCI device whose resources are to be reserved
2949 * @bars: Bitmask of BARs to be requested
2950 * @res_name: Name to be associated with resource
2951 */
2952int pci_request_selected_regions(struct pci_dev *pdev, int bars,
2953 const char *res_name)
2954{
2955 return __pci_request_selected_regions(pdev, bars, res_name, 0);
2956}
2957EXPORT_SYMBOL(pci_request_selected_regions);
2958
2959int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
2960 const char *res_name)
2961{
2962 return __pci_request_selected_regions(pdev, bars, res_name,
2963 IORESOURCE_EXCLUSIVE);
2964}
2965EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
2966
2967/**
2968 * pci_release_regions - Release reserved PCI I/O and memory resources
2969 * @pdev: PCI device whose resources were previously reserved by pci_request_regions
2970 *
2971 * Releases all PCI I/O and memory resources previously reserved by a
2972 * successful call to pci_request_regions. Call this function only
2973 * after all use of the PCI regions has ceased.
2974 */
2975
2976void pci_release_regions(struct pci_dev *pdev)
2977{
2978 pci_release_selected_regions(pdev, (1 << 6) - 1);
2979}
2980EXPORT_SYMBOL(pci_release_regions);
2981
2982/**
2983 * pci_request_regions - Reserved PCI I/O and memory resources
2984 * @pdev: PCI device whose resources are to be reserved
2985 * @res_name: Name to be associated with resource.
2986 *
2987 * Mark all PCI regions associated with PCI device @pdev as
2988 * being reserved by owner @res_name. Do not access any
2989 * address inside the PCI regions unless this call returns
2990 * successfully.
2991 *
2992 * Returns 0 on success, or %EBUSY on error. A warning
2993 * message is also printed on failure.
2994 */
2995int pci_request_regions(struct pci_dev *pdev, const char *res_name)
2996{
2997 return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
2998}
2999EXPORT_SYMBOL(pci_request_regions);
3000
3001/**
3002 * pci_request_regions_exclusive - Reserved PCI I/O and memory resources
3003 * @pdev: PCI device whose resources are to be reserved
3004 * @res_name: Name to be associated with resource.
3005 *
3006 * Mark all PCI regions associated with PCI device @pdev as
3007 * being reserved by owner @res_name. Do not access any
3008 * address inside the PCI regions unless this call returns
3009 * successfully.
3010 *
3011 * pci_request_regions_exclusive() will mark the region so that
3012 * /dev/mem and the sysfs MMIO access will not be allowed.
3013 *
3014 * Returns 0 on success, or %EBUSY on error. A warning
3015 * message is also printed on failure.
3016 */
3017int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3018{
3019 return pci_request_selected_regions_exclusive(pdev,
3020 ((1 << 6) - 1), res_name);
3021}
3022EXPORT_SYMBOL(pci_request_regions_exclusive);
3023
3024/**
3025 * pci_remap_iospace - Remap the memory mapped I/O space
3026 * @res: Resource describing the I/O space
3027 * @phys_addr: physical address of range to be mapped
3028 *
3029 * Remap the memory mapped I/O space described by the @res
3030 * and the CPU physical address @phys_addr into virtual address space.
3031 * Only architectures that have memory mapped IO functions defined
3032 * (and the PCI_IOBASE value defined) should call this function.
3033 */
3034int __weak pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3035{
3036#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3037 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3038
3039 if (!(res->flags & IORESOURCE_IO))
3040 return -EINVAL;
3041
3042 if (res->end > IO_SPACE_LIMIT)
3043 return -EINVAL;
3044
3045 return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3046 pgprot_device(PAGE_KERNEL));
3047#else
3048 /* this architecture does not have memory mapped I/O space,
3049 so this function should never be called */
3050 WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3051 return -ENODEV;
3052#endif
3053}
3054
3055static void __pci_set_master(struct pci_dev *dev, bool enable)
3056{
3057 u16 old_cmd, cmd;
3058
3059 pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
3060 if (enable)
3061 cmd = old_cmd | PCI_COMMAND_MASTER;
3062 else
3063 cmd = old_cmd & ~PCI_COMMAND_MASTER;
3064 if (cmd != old_cmd) {
3065 dev_dbg(&dev->dev, "%s bus mastering\n",
3066 enable ? "enabling" : "disabling");
3067 pci_write_config_word(dev, PCI_COMMAND, cmd);
3068 }
3069 dev->is_busmaster = enable;
3070}
3071
3072/**
3073 * pcibios_setup - process "pci=" kernel boot arguments
3074 * @str: string used to pass in "pci=" kernel boot arguments
3075 *
3076 * Process kernel boot arguments. This is the default implementation.
3077 * Architecture specific implementations can override this as necessary.
3078 */
3079char * __weak __init pcibios_setup(char *str)
3080{
3081 return str;
3082}
3083
3084/**
3085 * pcibios_set_master - enable PCI bus-mastering for device dev
3086 * @dev: the PCI device to enable
3087 *
3088 * Enables PCI bus-mastering for the device. This is the default
3089 * implementation. Architecture specific implementations can override
3090 * this if necessary.
3091 */
3092void __weak pcibios_set_master(struct pci_dev *dev)
3093{
3094 u8 lat;
3095
3096 /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
3097 if (pci_is_pcie(dev))
3098 return;
3099
3100 pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
3101 if (lat < 16)
3102 lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
3103 else if (lat > pcibios_max_latency)
3104 lat = pcibios_max_latency;
3105 else
3106 return;
3107
3108 pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
3109}
3110
3111/**
3112 * pci_set_master - enables bus-mastering for device dev
3113 * @dev: the PCI device to enable
3114 *
3115 * Enables bus-mastering on the device and calls pcibios_set_master()
3116 * to do the needed arch specific settings.
3117 */
3118void pci_set_master(struct pci_dev *dev)
3119{
3120 __pci_set_master(dev, true);
3121 pcibios_set_master(dev);
3122}
3123EXPORT_SYMBOL(pci_set_master);
3124
3125/**
3126 * pci_clear_master - disables bus-mastering for device dev
3127 * @dev: the PCI device to disable
3128 */
3129void pci_clear_master(struct pci_dev *dev)
3130{
3131 __pci_set_master(dev, false);
3132}
3133EXPORT_SYMBOL(pci_clear_master);
3134
3135/**
3136 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
3137 * @dev: the PCI device for which MWI is to be enabled
3138 *
3139 * Helper function for pci_set_mwi.
3140 * Originally copied from drivers/net/acenic.c.
3141 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
3142 *
3143 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3144 */
3145int pci_set_cacheline_size(struct pci_dev *dev)
3146{
3147 u8 cacheline_size;
3148
3149 if (!pci_cache_line_size)
3150 return -EINVAL;
3151
3152 /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
3153 equal to or multiple of the right value. */
3154 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3155 if (cacheline_size >= pci_cache_line_size &&
3156 (cacheline_size % pci_cache_line_size) == 0)
3157 return 0;
3158
3159 /* Write the correct value. */
3160 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
3161 /* Read it back. */
3162 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3163 if (cacheline_size == pci_cache_line_size)
3164 return 0;
3165
3166 dev_printk(KERN_DEBUG, &dev->dev, "cache line size of %d is not supported\n",
3167 pci_cache_line_size << 2);
3168
3169 return -EINVAL;
3170}
3171EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
3172
3173/**
3174 * pci_set_mwi - enables memory-write-invalidate PCI transaction
3175 * @dev: the PCI device for which MWI is enabled
3176 *
3177 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3178 *
3179 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3180 */
3181int pci_set_mwi(struct pci_dev *dev)
3182{
3183#ifdef PCI_DISABLE_MWI
3184 return 0;
3185#else
3186 int rc;
3187 u16 cmd;
3188
3189 rc = pci_set_cacheline_size(dev);
3190 if (rc)
3191 return rc;
3192
3193 pci_read_config_word(dev, PCI_COMMAND, &cmd);
3194 if (!(cmd & PCI_COMMAND_INVALIDATE)) {
3195 dev_dbg(&dev->dev, "enabling Mem-Wr-Inval\n");
3196 cmd |= PCI_COMMAND_INVALIDATE;
3197 pci_write_config_word(dev, PCI_COMMAND, cmd);
3198 }
3199 return 0;
3200#endif
3201}
3202EXPORT_SYMBOL(pci_set_mwi);
3203
3204/**
3205 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
3206 * @dev: the PCI device for which MWI is enabled
3207 *
3208 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3209 * Callers are not required to check the return value.
3210 *
3211 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3212 */
3213int pci_try_set_mwi(struct pci_dev *dev)
3214{
3215#ifdef PCI_DISABLE_MWI
3216 return 0;
3217#else
3218 return pci_set_mwi(dev);
3219#endif
3220}
3221EXPORT_SYMBOL(pci_try_set_mwi);
3222
3223/**
3224 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
3225 * @dev: the PCI device to disable
3226 *
3227 * Disables PCI Memory-Write-Invalidate transaction on the device
3228 */
3229void pci_clear_mwi(struct pci_dev *dev)
3230{
3231#ifndef PCI_DISABLE_MWI
3232 u16 cmd;
3233
3234 pci_read_config_word(dev, PCI_COMMAND, &cmd);
3235 if (cmd & PCI_COMMAND_INVALIDATE) {
3236 cmd &= ~PCI_COMMAND_INVALIDATE;
3237 pci_write_config_word(dev, PCI_COMMAND, cmd);
3238 }
3239#endif
3240}
3241EXPORT_SYMBOL(pci_clear_mwi);
3242
3243/**
3244 * pci_intx - enables/disables PCI INTx for device dev
3245 * @pdev: the PCI device to operate on
3246 * @enable: boolean: whether to enable or disable PCI INTx
3247 *
3248 * Enables/disables PCI INTx for device dev
3249 */
3250void pci_intx(struct pci_dev *pdev, int enable)
3251{
3252 u16 pci_command, new;
3253
3254 pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
3255
3256 if (enable)
3257 new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
3258 else
3259 new = pci_command | PCI_COMMAND_INTX_DISABLE;
3260
3261 if (new != pci_command) {
3262 struct pci_devres *dr;
3263
3264 pci_write_config_word(pdev, PCI_COMMAND, new);
3265
3266 dr = find_pci_dr(pdev);
3267 if (dr && !dr->restore_intx) {
3268 dr->restore_intx = 1;
3269 dr->orig_intx = !enable;
3270 }
3271 }
3272}
3273EXPORT_SYMBOL_GPL(pci_intx);
3274
3275/**
3276 * pci_intx_mask_supported - probe for INTx masking support
3277 * @dev: the PCI device to operate on
3278 *
3279 * Check if the device dev support INTx masking via the config space
3280 * command word.
3281 */
3282bool pci_intx_mask_supported(struct pci_dev *dev)
3283{
3284 bool mask_supported = false;
3285 u16 orig, new;
3286
3287 if (dev->broken_intx_masking)
3288 return false;
3289
3290 pci_cfg_access_lock(dev);
3291
3292 pci_read_config_word(dev, PCI_COMMAND, &orig);
3293 pci_write_config_word(dev, PCI_COMMAND,
3294 orig ^ PCI_COMMAND_INTX_DISABLE);
3295 pci_read_config_word(dev, PCI_COMMAND, &new);
3296
3297 /*
3298 * There's no way to protect against hardware bugs or detect them
3299 * reliably, but as long as we know what the value should be, let's
3300 * go ahead and check it.
3301 */
3302 if ((new ^ orig) & ~PCI_COMMAND_INTX_DISABLE) {
3303 dev_err(&dev->dev, "Command register changed from 0x%x to 0x%x: driver or hardware bug?\n",
3304 orig, new);
3305 } else if ((new ^ orig) & PCI_COMMAND_INTX_DISABLE) {
3306 mask_supported = true;
3307 pci_write_config_word(dev, PCI_COMMAND, orig);
3308 }
3309
3310 pci_cfg_access_unlock(dev);
3311 return mask_supported;
3312}
3313EXPORT_SYMBOL_GPL(pci_intx_mask_supported);
3314
3315static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
3316{
3317 struct pci_bus *bus = dev->bus;
3318 bool mask_updated = true;
3319 u32 cmd_status_dword;
3320 u16 origcmd, newcmd;
3321 unsigned long flags;
3322 bool irq_pending;
3323
3324 /*
3325 * We do a single dword read to retrieve both command and status.
3326 * Document assumptions that make this possible.
3327 */
3328 BUILD_BUG_ON(PCI_COMMAND % 4);
3329 BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
3330
3331 raw_spin_lock_irqsave(&pci_lock, flags);
3332
3333 bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
3334
3335 irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
3336
3337 /*
3338 * Check interrupt status register to see whether our device
3339 * triggered the interrupt (when masking) or the next IRQ is
3340 * already pending (when unmasking).
3341 */
3342 if (mask != irq_pending) {
3343 mask_updated = false;
3344 goto done;
3345 }
3346
3347 origcmd = cmd_status_dword;
3348 newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
3349 if (mask)
3350 newcmd |= PCI_COMMAND_INTX_DISABLE;
3351 if (newcmd != origcmd)
3352 bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
3353
3354done:
3355 raw_spin_unlock_irqrestore(&pci_lock, flags);
3356
3357 return mask_updated;
3358}
3359
3360/**
3361 * pci_check_and_mask_intx - mask INTx on pending interrupt
3362 * @dev: the PCI device to operate on
3363 *
3364 * Check if the device dev has its INTx line asserted, mask it and
3365 * return true in that case. False is returned if not interrupt was
3366 * pending.
3367 */
3368bool pci_check_and_mask_intx(struct pci_dev *dev)
3369{
3370 return pci_check_and_set_intx_mask(dev, true);
3371}
3372EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
3373
3374/**
3375 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
3376 * @dev: the PCI device to operate on
3377 *
3378 * Check if the device dev has its INTx line asserted, unmask it if not
3379 * and return true. False is returned and the mask remains active if
3380 * there was still an interrupt pending.
3381 */
3382bool pci_check_and_unmask_intx(struct pci_dev *dev)
3383{
3384 return pci_check_and_set_intx_mask(dev, false);
3385}
3386EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
3387
3388/**
3389 * pci_wait_for_pending_transaction - waits for pending transaction
3390 * @dev: the PCI device to operate on
3391 *
3392 * Return 0 if transaction is pending 1 otherwise.
3393 */
3394int pci_wait_for_pending_transaction(struct pci_dev *dev)
3395{
3396 if (!pci_is_pcie(dev))
3397 return 1;
3398
3399 return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
3400 PCI_EXP_DEVSTA_TRPND);
3401}
3402EXPORT_SYMBOL(pci_wait_for_pending_transaction);
3403
3404/*
3405 * We should only need to wait 100ms after FLR, but some devices take longer.
3406 * Wait for up to 1000ms for config space to return something other than -1.
3407 * Intel IGD requires this when an LCD panel is attached. We read the 2nd
3408 * dword because VFs don't implement the 1st dword.
3409 */
3410static void pci_flr_wait(struct pci_dev *dev)
3411{
3412 int i = 0;
3413 u32 id;
3414
3415 do {
3416 msleep(100);
3417 pci_read_config_dword(dev, PCI_COMMAND, &id);
3418 } while (i++ < 10 && id == ~0);
3419
3420 if (id == ~0)
3421 dev_warn(&dev->dev, "Failed to return from FLR\n");
3422 else if (i > 1)
3423 dev_info(&dev->dev, "Required additional %dms to return from FLR\n",
3424 (i - 1) * 100);
3425}
3426
3427static int pcie_flr(struct pci_dev *dev, int probe)
3428{
3429 u32 cap;
3430
3431 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
3432 if (!(cap & PCI_EXP_DEVCAP_FLR))
3433 return -ENOTTY;
3434
3435 if (probe)
3436 return 0;
3437
3438 if (!pci_wait_for_pending_transaction(dev))
3439 dev_err(&dev->dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
3440
3441 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
3442 pci_flr_wait(dev);
3443 return 0;
3444}
3445
3446static int pci_af_flr(struct pci_dev *dev, int probe)
3447{
3448 int pos;
3449 u8 cap;
3450
3451 pos = pci_find_capability(dev, PCI_CAP_ID_AF);
3452 if (!pos)
3453 return -ENOTTY;
3454
3455 pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
3456 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
3457 return -ENOTTY;
3458
3459 if (probe)
3460 return 0;
3461
3462 /*
3463 * Wait for Transaction Pending bit to clear. A word-aligned test
3464 * is used, so we use the conrol offset rather than status and shift
3465 * the test bit to match.
3466 */
3467 if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
3468 PCI_AF_STATUS_TP << 8))
3469 dev_err(&dev->dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
3470
3471 pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
3472 pci_flr_wait(dev);
3473 return 0;
3474}
3475
3476/**
3477 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
3478 * @dev: Device to reset.
3479 * @probe: If set, only check if the device can be reset this way.
3480 *
3481 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
3482 * unset, it will be reinitialized internally when going from PCI_D3hot to
3483 * PCI_D0. If that's the case and the device is not in a low-power state
3484 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
3485 *
3486 * NOTE: This causes the caller to sleep for twice the device power transition
3487 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
3488 * by default (i.e. unless the @dev's d3_delay field has a different value).
3489 * Moreover, only devices in D0 can be reset by this function.
3490 */
3491static int pci_pm_reset(struct pci_dev *dev, int probe)
3492{
3493 u16 csr;
3494
3495 if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
3496 return -ENOTTY;
3497
3498 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
3499 if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
3500 return -ENOTTY;
3501
3502 if (probe)
3503 return 0;
3504
3505 if (dev->current_state != PCI_D0)
3506 return -EINVAL;
3507
3508 csr &= ~PCI_PM_CTRL_STATE_MASK;
3509 csr |= PCI_D3hot;
3510 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3511 pci_dev_d3_sleep(dev);
3512
3513 csr &= ~PCI_PM_CTRL_STATE_MASK;
3514 csr |= PCI_D0;
3515 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3516 pci_dev_d3_sleep(dev);
3517
3518 return 0;
3519}
3520
3521void pci_reset_secondary_bus(struct pci_dev *dev)
3522{
3523 u16 ctrl;
3524
3525 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
3526 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
3527 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
3528 /*
3529 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms. Double
3530 * this to 2ms to ensure that we meet the minimum requirement.
3531 */
3532 msleep(2);
3533
3534 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
3535 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
3536
3537 /*
3538 * Trhfa for conventional PCI is 2^25 clock cycles.
3539 * Assuming a minimum 33MHz clock this results in a 1s
3540 * delay before we can consider subordinate devices to
3541 * be re-initialized. PCIe has some ways to shorten this,
3542 * but we don't make use of them yet.
3543 */
3544 ssleep(1);
3545}
3546
3547void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
3548{
3549 pci_reset_secondary_bus(dev);
3550}
3551
3552/**
3553 * pci_reset_bridge_secondary_bus - Reset the secondary bus on a PCI bridge.
3554 * @dev: Bridge device
3555 *
3556 * Use the bridge control register to assert reset on the secondary bus.
3557 * Devices on the secondary bus are left in power-on state.
3558 */
3559void pci_reset_bridge_secondary_bus(struct pci_dev *dev)
3560{
3561 pcibios_reset_secondary_bus(dev);
3562}
3563EXPORT_SYMBOL_GPL(pci_reset_bridge_secondary_bus);
3564
3565static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
3566{
3567 struct pci_dev *pdev;
3568
3569 if (pci_is_root_bus(dev->bus) || dev->subordinate ||
3570 !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
3571 return -ENOTTY;
3572
3573 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
3574 if (pdev != dev)
3575 return -ENOTTY;
3576
3577 if (probe)
3578 return 0;
3579
3580 pci_reset_bridge_secondary_bus(dev->bus->self);
3581
3582 return 0;
3583}
3584
3585static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
3586{
3587 int rc = -ENOTTY;
3588
3589 if (!hotplug || !try_module_get(hotplug->ops->owner))
3590 return rc;
3591
3592 if (hotplug->ops->reset_slot)
3593 rc = hotplug->ops->reset_slot(hotplug, probe);
3594
3595 module_put(hotplug->ops->owner);
3596
3597 return rc;
3598}
3599
3600static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
3601{
3602 struct pci_dev *pdev;
3603
3604 if (dev->subordinate || !dev->slot ||
3605 dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
3606 return -ENOTTY;
3607
3608 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
3609 if (pdev != dev && pdev->slot == dev->slot)
3610 return -ENOTTY;
3611
3612 return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
3613}
3614
3615static int __pci_dev_reset(struct pci_dev *dev, int probe)
3616{
3617 int rc;
3618
3619 might_sleep();
3620
3621 rc = pci_dev_specific_reset(dev, probe);
3622 if (rc != -ENOTTY)
3623 goto done;
3624
3625 rc = pcie_flr(dev, probe);
3626 if (rc != -ENOTTY)
3627 goto done;
3628
3629 rc = pci_af_flr(dev, probe);
3630 if (rc != -ENOTTY)
3631 goto done;
3632
3633 rc = pci_pm_reset(dev, probe);
3634 if (rc != -ENOTTY)
3635 goto done;
3636
3637 rc = pci_dev_reset_slot_function(dev, probe);
3638 if (rc != -ENOTTY)
3639 goto done;
3640
3641 rc = pci_parent_bus_reset(dev, probe);
3642done:
3643 return rc;
3644}
3645
3646static void pci_dev_lock(struct pci_dev *dev)
3647{
3648 pci_cfg_access_lock(dev);
3649 /* block PM suspend, driver probe, etc. */
3650 device_lock(&dev->dev);
3651}
3652
3653/* Return 1 on successful lock, 0 on contention */
3654static int pci_dev_trylock(struct pci_dev *dev)
3655{
3656 if (pci_cfg_access_trylock(dev)) {
3657 if (device_trylock(&dev->dev))
3658 return 1;
3659 pci_cfg_access_unlock(dev);
3660 }
3661
3662 return 0;
3663}
3664
3665static void pci_dev_unlock(struct pci_dev *dev)
3666{
3667 device_unlock(&dev->dev);
3668 pci_cfg_access_unlock(dev);
3669}
3670
3671/**
3672 * pci_reset_notify - notify device driver of reset
3673 * @dev: device to be notified of reset
3674 * @prepare: 'true' if device is about to be reset; 'false' if reset attempt
3675 * completed
3676 *
3677 * Must be called prior to device access being disabled and after device
3678 * access is restored.
3679 */
3680static void pci_reset_notify(struct pci_dev *dev, bool prepare)
3681{
3682 const struct pci_error_handlers *err_handler =
3683 dev->driver ? dev->driver->err_handler : NULL;
3684 if (err_handler && err_handler->reset_notify)
3685 err_handler->reset_notify(dev, prepare);
3686}
3687
3688static void pci_dev_save_and_disable(struct pci_dev *dev)
3689{
3690 pci_reset_notify(dev, true);
3691
3692 /*
3693 * Wake-up device prior to save. PM registers default to D0 after
3694 * reset and a simple register restore doesn't reliably return
3695 * to a non-D0 state anyway.
3696 */
3697 pci_set_power_state(dev, PCI_D0);
3698
3699 pci_save_state(dev);
3700 /*
3701 * Disable the device by clearing the Command register, except for
3702 * INTx-disable which is set. This not only disables MMIO and I/O port
3703 * BARs, but also prevents the device from being Bus Master, preventing
3704 * DMA from the device including MSI/MSI-X interrupts. For PCI 2.3
3705 * compliant devices, INTx-disable prevents legacy interrupts.
3706 */
3707 pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
3708}
3709
3710static void pci_dev_restore(struct pci_dev *dev)
3711{
3712 pci_restore_state(dev);
3713 pci_reset_notify(dev, false);
3714}
3715
3716static int pci_dev_reset(struct pci_dev *dev, int probe)
3717{
3718 int rc;
3719
3720 if (!probe)
3721 pci_dev_lock(dev);
3722
3723 rc = __pci_dev_reset(dev, probe);
3724
3725 if (!probe)
3726 pci_dev_unlock(dev);
3727
3728 return rc;
3729}
3730
3731/**
3732 * __pci_reset_function - reset a PCI device function
3733 * @dev: PCI device to reset
3734 *
3735 * Some devices allow an individual function to be reset without affecting
3736 * other functions in the same device. The PCI device must be responsive
3737 * to PCI config space in order to use this function.
3738 *
3739 * The device function is presumed to be unused when this function is called.
3740 * Resetting the device will make the contents of PCI configuration space
3741 * random, so any caller of this must be prepared to reinitialise the
3742 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
3743 * etc.
3744 *
3745 * Returns 0 if the device function was successfully reset or negative if the
3746 * device doesn't support resetting a single function.
3747 */
3748int __pci_reset_function(struct pci_dev *dev)
3749{
3750 return pci_dev_reset(dev, 0);
3751}
3752EXPORT_SYMBOL_GPL(__pci_reset_function);
3753
3754/**
3755 * __pci_reset_function_locked - reset a PCI device function while holding
3756 * the @dev mutex lock.
3757 * @dev: PCI device to reset
3758 *
3759 * Some devices allow an individual function to be reset without affecting
3760 * other functions in the same device. The PCI device must be responsive
3761 * to PCI config space in order to use this function.
3762 *
3763 * The device function is presumed to be unused and the caller is holding
3764 * the device mutex lock when this function is called.
3765 * Resetting the device will make the contents of PCI configuration space
3766 * random, so any caller of this must be prepared to reinitialise the
3767 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
3768 * etc.
3769 *
3770 * Returns 0 if the device function was successfully reset or negative if the
3771 * device doesn't support resetting a single function.
3772 */
3773int __pci_reset_function_locked(struct pci_dev *dev)
3774{
3775 return __pci_dev_reset(dev, 0);
3776}
3777EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
3778
3779/**
3780 * pci_probe_reset_function - check whether the device can be safely reset
3781 * @dev: PCI device to reset
3782 *
3783 * Some devices allow an individual function to be reset without affecting
3784 * other functions in the same device. The PCI device must be responsive
3785 * to PCI config space in order to use this function.
3786 *
3787 * Returns 0 if the device function can be reset or negative if the
3788 * device doesn't support resetting a single function.
3789 */
3790int pci_probe_reset_function(struct pci_dev *dev)
3791{
3792 return pci_dev_reset(dev, 1);
3793}
3794
3795/**
3796 * pci_reset_function - quiesce and reset a PCI device function
3797 * @dev: PCI device to reset
3798 *
3799 * Some devices allow an individual function to be reset without affecting
3800 * other functions in the same device. The PCI device must be responsive
3801 * to PCI config space in order to use this function.
3802 *
3803 * This function does not just reset the PCI portion of a device, but
3804 * clears all the state associated with the device. This function differs
3805 * from __pci_reset_function in that it saves and restores device state
3806 * over the reset.
3807 *
3808 * Returns 0 if the device function was successfully reset or negative if the
3809 * device doesn't support resetting a single function.
3810 */
3811int pci_reset_function(struct pci_dev *dev)
3812{
3813 int rc;
3814
3815 rc = pci_dev_reset(dev, 1);
3816 if (rc)
3817 return rc;
3818
3819 pci_dev_save_and_disable(dev);
3820
3821 rc = pci_dev_reset(dev, 0);
3822
3823 pci_dev_restore(dev);
3824
3825 return rc;
3826}
3827EXPORT_SYMBOL_GPL(pci_reset_function);
3828
3829/**
3830 * pci_try_reset_function - quiesce and reset a PCI device function
3831 * @dev: PCI device to reset
3832 *
3833 * Same as above, except return -EAGAIN if unable to lock device.
3834 */
3835int pci_try_reset_function(struct pci_dev *dev)
3836{
3837 int rc;
3838
3839 rc = pci_dev_reset(dev, 1);
3840 if (rc)
3841 return rc;
3842
3843 pci_dev_save_and_disable(dev);
3844
3845 if (pci_dev_trylock(dev)) {
3846 rc = __pci_dev_reset(dev, 0);
3847 pci_dev_unlock(dev);
3848 } else
3849 rc = -EAGAIN;
3850
3851 pci_dev_restore(dev);
3852
3853 return rc;
3854}
3855EXPORT_SYMBOL_GPL(pci_try_reset_function);
3856
3857/* Do any devices on or below this bus prevent a bus reset? */
3858static bool pci_bus_resetable(struct pci_bus *bus)
3859{
3860 struct pci_dev *dev;
3861
3862 list_for_each_entry(dev, &bus->devices, bus_list) {
3863 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
3864 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
3865 return false;
3866 }
3867
3868 return true;
3869}
3870
3871/* Lock devices from the top of the tree down */
3872static void pci_bus_lock(struct pci_bus *bus)
3873{
3874 struct pci_dev *dev;
3875
3876 list_for_each_entry(dev, &bus->devices, bus_list) {
3877 pci_dev_lock(dev);
3878 if (dev->subordinate)
3879 pci_bus_lock(dev->subordinate);
3880 }
3881}
3882
3883/* Unlock devices from the bottom of the tree up */
3884static void pci_bus_unlock(struct pci_bus *bus)
3885{
3886 struct pci_dev *dev;
3887
3888 list_for_each_entry(dev, &bus->devices, bus_list) {
3889 if (dev->subordinate)
3890 pci_bus_unlock(dev->subordinate);
3891 pci_dev_unlock(dev);
3892 }
3893}
3894
3895/* Return 1 on successful lock, 0 on contention */
3896static int pci_bus_trylock(struct pci_bus *bus)
3897{
3898 struct pci_dev *dev;
3899
3900 list_for_each_entry(dev, &bus->devices, bus_list) {
3901 if (!pci_dev_trylock(dev))
3902 goto unlock;
3903 if (dev->subordinate) {
3904 if (!pci_bus_trylock(dev->subordinate)) {
3905 pci_dev_unlock(dev);
3906 goto unlock;
3907 }
3908 }
3909 }
3910 return 1;
3911
3912unlock:
3913 list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
3914 if (dev->subordinate)
3915 pci_bus_unlock(dev->subordinate);
3916 pci_dev_unlock(dev);
3917 }
3918 return 0;
3919}
3920
3921/* Do any devices on or below this slot prevent a bus reset? */
3922static bool pci_slot_resetable(struct pci_slot *slot)
3923{
3924 struct pci_dev *dev;
3925
3926 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
3927 if (!dev->slot || dev->slot != slot)
3928 continue;
3929 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
3930 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
3931 return false;
3932 }
3933
3934 return true;
3935}
3936
3937/* Lock devices from the top of the tree down */
3938static void pci_slot_lock(struct pci_slot *slot)
3939{
3940 struct pci_dev *dev;
3941
3942 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
3943 if (!dev->slot || dev->slot != slot)
3944 continue;
3945 pci_dev_lock(dev);
3946 if (dev->subordinate)
3947 pci_bus_lock(dev->subordinate);
3948 }
3949}
3950
3951/* Unlock devices from the bottom of the tree up */
3952static void pci_slot_unlock(struct pci_slot *slot)
3953{
3954 struct pci_dev *dev;
3955
3956 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
3957 if (!dev->slot || dev->slot != slot)
3958 continue;
3959 if (dev->subordinate)
3960 pci_bus_unlock(dev->subordinate);
3961 pci_dev_unlock(dev);
3962 }
3963}
3964
3965/* Return 1 on successful lock, 0 on contention */
3966static int pci_slot_trylock(struct pci_slot *slot)
3967{
3968 struct pci_dev *dev;
3969
3970 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
3971 if (!dev->slot || dev->slot != slot)
3972 continue;
3973 if (!pci_dev_trylock(dev))
3974 goto unlock;
3975 if (dev->subordinate) {
3976 if (!pci_bus_trylock(dev->subordinate)) {
3977 pci_dev_unlock(dev);
3978 goto unlock;
3979 }
3980 }
3981 }
3982 return 1;
3983
3984unlock:
3985 list_for_each_entry_continue_reverse(dev,
3986 &slot->bus->devices, bus_list) {
3987 if (!dev->slot || dev->slot != slot)
3988 continue;
3989 if (dev->subordinate)
3990 pci_bus_unlock(dev->subordinate);
3991 pci_dev_unlock(dev);
3992 }
3993 return 0;
3994}
3995
3996/* Save and disable devices from the top of the tree down */
3997static void pci_bus_save_and_disable(struct pci_bus *bus)
3998{
3999 struct pci_dev *dev;
4000
4001 list_for_each_entry(dev, &bus->devices, bus_list) {
4002 pci_dev_save_and_disable(dev);
4003 if (dev->subordinate)
4004 pci_bus_save_and_disable(dev->subordinate);
4005 }
4006}
4007
4008/*
4009 * Restore devices from top of the tree down - parent bridges need to be
4010 * restored before we can get to subordinate devices.
4011 */
4012static void pci_bus_restore(struct pci_bus *bus)
4013{
4014 struct pci_dev *dev;
4015
4016 list_for_each_entry(dev, &bus->devices, bus_list) {
4017 pci_dev_restore(dev);
4018 if (dev->subordinate)
4019 pci_bus_restore(dev->subordinate);
4020 }
4021}
4022
4023/* Save and disable devices from the top of the tree down */
4024static void pci_slot_save_and_disable(struct pci_slot *slot)
4025{
4026 struct pci_dev *dev;
4027
4028 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4029 if (!dev->slot || dev->slot != slot)
4030 continue;
4031 pci_dev_save_and_disable(dev);
4032 if (dev->subordinate)
4033 pci_bus_save_and_disable(dev->subordinate);
4034 }
4035}
4036
4037/*
4038 * Restore devices from top of the tree down - parent bridges need to be
4039 * restored before we can get to subordinate devices.
4040 */
4041static void pci_slot_restore(struct pci_slot *slot)
4042{
4043 struct pci_dev *dev;
4044
4045 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4046 if (!dev->slot || dev->slot != slot)
4047 continue;
4048 pci_dev_restore(dev);
4049 if (dev->subordinate)
4050 pci_bus_restore(dev->subordinate);
4051 }
4052}
4053
4054static int pci_slot_reset(struct pci_slot *slot, int probe)
4055{
4056 int rc;
4057
4058 if (!slot || !pci_slot_resetable(slot))
4059 return -ENOTTY;
4060
4061 if (!probe)
4062 pci_slot_lock(slot);
4063
4064 might_sleep();
4065
4066 rc = pci_reset_hotplug_slot(slot->hotplug, probe);
4067
4068 if (!probe)
4069 pci_slot_unlock(slot);
4070
4071 return rc;
4072}
4073
4074/**
4075 * pci_probe_reset_slot - probe whether a PCI slot can be reset
4076 * @slot: PCI slot to probe
4077 *
4078 * Return 0 if slot can be reset, negative if a slot reset is not supported.
4079 */
4080int pci_probe_reset_slot(struct pci_slot *slot)
4081{
4082 return pci_slot_reset(slot, 1);
4083}
4084EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
4085
4086/**
4087 * pci_reset_slot - reset a PCI slot
4088 * @slot: PCI slot to reset
4089 *
4090 * A PCI bus may host multiple slots, each slot may support a reset mechanism
4091 * independent of other slots. For instance, some slots may support slot power
4092 * control. In the case of a 1:1 bus to slot architecture, this function may
4093 * wrap the bus reset to avoid spurious slot related events such as hotplug.
4094 * Generally a slot reset should be attempted before a bus reset. All of the
4095 * function of the slot and any subordinate buses behind the slot are reset
4096 * through this function. PCI config space of all devices in the slot and
4097 * behind the slot is saved before and restored after reset.
4098 *
4099 * Return 0 on success, non-zero on error.
4100 */
4101int pci_reset_slot(struct pci_slot *slot)
4102{
4103 int rc;
4104
4105 rc = pci_slot_reset(slot, 1);
4106 if (rc)
4107 return rc;
4108
4109 pci_slot_save_and_disable(slot);
4110
4111 rc = pci_slot_reset(slot, 0);
4112
4113 pci_slot_restore(slot);
4114
4115 return rc;
4116}
4117EXPORT_SYMBOL_GPL(pci_reset_slot);
4118
4119/**
4120 * pci_try_reset_slot - Try to reset a PCI slot
4121 * @slot: PCI slot to reset
4122 *
4123 * Same as above except return -EAGAIN if the slot cannot be locked
4124 */
4125int pci_try_reset_slot(struct pci_slot *slot)
4126{
4127 int rc;
4128
4129 rc = pci_slot_reset(slot, 1);
4130 if (rc)
4131 return rc;
4132
4133 pci_slot_save_and_disable(slot);
4134
4135 if (pci_slot_trylock(slot)) {
4136 might_sleep();
4137 rc = pci_reset_hotplug_slot(slot->hotplug, 0);
4138 pci_slot_unlock(slot);
4139 } else
4140 rc = -EAGAIN;
4141
4142 pci_slot_restore(slot);
4143
4144 return rc;
4145}
4146EXPORT_SYMBOL_GPL(pci_try_reset_slot);
4147
4148static int pci_bus_reset(struct pci_bus *bus, int probe)
4149{
4150 if (!bus->self || !pci_bus_resetable(bus))
4151 return -ENOTTY;
4152
4153 if (probe)
4154 return 0;
4155
4156 pci_bus_lock(bus);
4157
4158 might_sleep();
4159
4160 pci_reset_bridge_secondary_bus(bus->self);
4161
4162 pci_bus_unlock(bus);
4163
4164 return 0;
4165}
4166
4167/**
4168 * pci_probe_reset_bus - probe whether a PCI bus can be reset
4169 * @bus: PCI bus to probe
4170 *
4171 * Return 0 if bus can be reset, negative if a bus reset is not supported.
4172 */
4173int pci_probe_reset_bus(struct pci_bus *bus)
4174{
4175 return pci_bus_reset(bus, 1);
4176}
4177EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
4178
4179/**
4180 * pci_reset_bus - reset a PCI bus
4181 * @bus: top level PCI bus to reset
4182 *
4183 * Do a bus reset on the given bus and any subordinate buses, saving
4184 * and restoring state of all devices.
4185 *
4186 * Return 0 on success, non-zero on error.
4187 */
4188int pci_reset_bus(struct pci_bus *bus)
4189{
4190 int rc;
4191
4192 rc = pci_bus_reset(bus, 1);
4193 if (rc)
4194 return rc;
4195
4196 pci_bus_save_and_disable(bus);
4197
4198 rc = pci_bus_reset(bus, 0);
4199
4200 pci_bus_restore(bus);
4201
4202 return rc;
4203}
4204EXPORT_SYMBOL_GPL(pci_reset_bus);
4205
4206/**
4207 * pci_try_reset_bus - Try to reset a PCI bus
4208 * @bus: top level PCI bus to reset
4209 *
4210 * Same as above except return -EAGAIN if the bus cannot be locked
4211 */
4212int pci_try_reset_bus(struct pci_bus *bus)
4213{
4214 int rc;
4215
4216 rc = pci_bus_reset(bus, 1);
4217 if (rc)
4218 return rc;
4219
4220 pci_bus_save_and_disable(bus);
4221
4222 if (pci_bus_trylock(bus)) {
4223 might_sleep();
4224 pci_reset_bridge_secondary_bus(bus->self);
4225 pci_bus_unlock(bus);
4226 } else
4227 rc = -EAGAIN;
4228
4229 pci_bus_restore(bus);
4230
4231 return rc;
4232}
4233EXPORT_SYMBOL_GPL(pci_try_reset_bus);
4234
4235/**
4236 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
4237 * @dev: PCI device to query
4238 *
4239 * Returns mmrbc: maximum designed memory read count in bytes
4240 * or appropriate error value.
4241 */
4242int pcix_get_max_mmrbc(struct pci_dev *dev)
4243{
4244 int cap;
4245 u32 stat;
4246
4247 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4248 if (!cap)
4249 return -EINVAL;
4250
4251 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4252 return -EINVAL;
4253
4254 return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
4255}
4256EXPORT_SYMBOL(pcix_get_max_mmrbc);
4257
4258/**
4259 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
4260 * @dev: PCI device to query
4261 *
4262 * Returns mmrbc: maximum memory read count in bytes
4263 * or appropriate error value.
4264 */
4265int pcix_get_mmrbc(struct pci_dev *dev)
4266{
4267 int cap;
4268 u16 cmd;
4269
4270 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4271 if (!cap)
4272 return -EINVAL;
4273
4274 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4275 return -EINVAL;
4276
4277 return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
4278}
4279EXPORT_SYMBOL(pcix_get_mmrbc);
4280
4281/**
4282 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
4283 * @dev: PCI device to query
4284 * @mmrbc: maximum memory read count in bytes
4285 * valid values are 512, 1024, 2048, 4096
4286 *
4287 * If possible sets maximum memory read byte count, some bridges have erratas
4288 * that prevent this.
4289 */
4290int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
4291{
4292 int cap;
4293 u32 stat, v, o;
4294 u16 cmd;
4295
4296 if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
4297 return -EINVAL;
4298
4299 v = ffs(mmrbc) - 10;
4300
4301 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4302 if (!cap)
4303 return -EINVAL;
4304
4305 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4306 return -EINVAL;
4307
4308 if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
4309 return -E2BIG;
4310
4311 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4312 return -EINVAL;
4313
4314 o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
4315 if (o != v) {
4316 if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
4317 return -EIO;
4318
4319 cmd &= ~PCI_X_CMD_MAX_READ;
4320 cmd |= v << 2;
4321 if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
4322 return -EIO;
4323 }
4324 return 0;
4325}
4326EXPORT_SYMBOL(pcix_set_mmrbc);
4327
4328/**
4329 * pcie_get_readrq - get PCI Express read request size
4330 * @dev: PCI device to query
4331 *
4332 * Returns maximum memory read request in bytes
4333 * or appropriate error value.
4334 */
4335int pcie_get_readrq(struct pci_dev *dev)
4336{
4337 u16 ctl;
4338
4339 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
4340
4341 return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
4342}
4343EXPORT_SYMBOL(pcie_get_readrq);
4344
4345/**
4346 * pcie_set_readrq - set PCI Express maximum memory read request
4347 * @dev: PCI device to query
4348 * @rq: maximum memory read count in bytes
4349 * valid values are 128, 256, 512, 1024, 2048, 4096
4350 *
4351 * If possible sets maximum memory read request in bytes
4352 */
4353int pcie_set_readrq(struct pci_dev *dev, int rq)
4354{
4355 u16 v;
4356
4357 if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
4358 return -EINVAL;
4359
4360 /*
4361 * If using the "performance" PCIe config, we clamp the
4362 * read rq size to the max packet size to prevent the
4363 * host bridge generating requests larger than we can
4364 * cope with
4365 */
4366 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
4367 int mps = pcie_get_mps(dev);
4368
4369 if (mps < rq)
4370 rq = mps;
4371 }
4372
4373 v = (ffs(rq) - 8) << 12;
4374
4375 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
4376 PCI_EXP_DEVCTL_READRQ, v);
4377}
4378EXPORT_SYMBOL(pcie_set_readrq);
4379
4380/**
4381 * pcie_get_mps - get PCI Express maximum payload size
4382 * @dev: PCI device to query
4383 *
4384 * Returns maximum payload size in bytes
4385 */
4386int pcie_get_mps(struct pci_dev *dev)
4387{
4388 u16 ctl;
4389
4390 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
4391
4392 return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
4393}
4394EXPORT_SYMBOL(pcie_get_mps);
4395
4396/**
4397 * pcie_set_mps - set PCI Express maximum payload size
4398 * @dev: PCI device to query
4399 * @mps: maximum payload size in bytes
4400 * valid values are 128, 256, 512, 1024, 2048, 4096
4401 *
4402 * If possible sets maximum payload size
4403 */
4404int pcie_set_mps(struct pci_dev *dev, int mps)
4405{
4406 u16 v;
4407
4408 if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
4409 return -EINVAL;
4410
4411 v = ffs(mps) - 8;
4412 if (v > dev->pcie_mpss)
4413 return -EINVAL;
4414 v <<= 5;
4415
4416 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
4417 PCI_EXP_DEVCTL_PAYLOAD, v);
4418}
4419EXPORT_SYMBOL(pcie_set_mps);
4420
4421/**
4422 * pcie_get_minimum_link - determine minimum link settings of a PCI device
4423 * @dev: PCI device to query
4424 * @speed: storage for minimum speed
4425 * @width: storage for minimum width
4426 *
4427 * This function will walk up the PCI device chain and determine the minimum
4428 * link width and speed of the device.
4429 */
4430int pcie_get_minimum_link(struct pci_dev *dev, enum pci_bus_speed *speed,
4431 enum pcie_link_width *width)
4432{
4433 int ret;
4434
4435 *speed = PCI_SPEED_UNKNOWN;
4436 *width = PCIE_LNK_WIDTH_UNKNOWN;
4437
4438 while (dev) {
4439 u16 lnksta;
4440 enum pci_bus_speed next_speed;
4441 enum pcie_link_width next_width;
4442
4443 ret = pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
4444 if (ret)
4445 return ret;
4446
4447 next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
4448 next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
4449 PCI_EXP_LNKSTA_NLW_SHIFT;
4450
4451 if (next_speed < *speed)
4452 *speed = next_speed;
4453
4454 if (next_width < *width)
4455 *width = next_width;
4456
4457 dev = dev->bus->self;
4458 }
4459
4460 return 0;
4461}
4462EXPORT_SYMBOL(pcie_get_minimum_link);
4463
4464/**
4465 * pci_select_bars - Make BAR mask from the type of resource
4466 * @dev: the PCI device for which BAR mask is made
4467 * @flags: resource type mask to be selected
4468 *
4469 * This helper routine makes bar mask from the type of resource.
4470 */
4471int pci_select_bars(struct pci_dev *dev, unsigned long flags)
4472{
4473 int i, bars = 0;
4474 for (i = 0; i < PCI_NUM_RESOURCES; i++)
4475 if (pci_resource_flags(dev, i) & flags)
4476 bars |= (1 << i);
4477 return bars;
4478}
4479EXPORT_SYMBOL(pci_select_bars);
4480
4481/**
4482 * pci_resource_bar - get position of the BAR associated with a resource
4483 * @dev: the PCI device
4484 * @resno: the resource number
4485 * @type: the BAR type to be filled in
4486 *
4487 * Returns BAR position in config space, or 0 if the BAR is invalid.
4488 */
4489int pci_resource_bar(struct pci_dev *dev, int resno, enum pci_bar_type *type)
4490{
4491 int reg;
4492
4493 if (resno < PCI_ROM_RESOURCE) {
4494 *type = pci_bar_unknown;
4495 return PCI_BASE_ADDRESS_0 + 4 * resno;
4496 } else if (resno == PCI_ROM_RESOURCE) {
4497 *type = pci_bar_mem32;
4498 return dev->rom_base_reg;
4499 } else if (resno < PCI_BRIDGE_RESOURCES) {
4500 /* device specific resource */
4501 *type = pci_bar_unknown;
4502 reg = pci_iov_resource_bar(dev, resno);
4503 if (reg)
4504 return reg;
4505 }
4506
4507 dev_err(&dev->dev, "BAR %d: invalid resource\n", resno);
4508 return 0;
4509}
4510
4511/* Some architectures require additional programming to enable VGA */
4512static arch_set_vga_state_t arch_set_vga_state;
4513
4514void __init pci_register_set_vga_state(arch_set_vga_state_t func)
4515{
4516 arch_set_vga_state = func; /* NULL disables */
4517}
4518
4519static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
4520 unsigned int command_bits, u32 flags)
4521{
4522 if (arch_set_vga_state)
4523 return arch_set_vga_state(dev, decode, command_bits,
4524 flags);
4525 return 0;
4526}
4527
4528/**
4529 * pci_set_vga_state - set VGA decode state on device and parents if requested
4530 * @dev: the PCI device
4531 * @decode: true = enable decoding, false = disable decoding
4532 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
4533 * @flags: traverse ancestors and change bridges
4534 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
4535 */
4536int pci_set_vga_state(struct pci_dev *dev, bool decode,
4537 unsigned int command_bits, u32 flags)
4538{
4539 struct pci_bus *bus;
4540 struct pci_dev *bridge;
4541 u16 cmd;
4542 int rc;
4543
4544 WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
4545
4546 /* ARCH specific VGA enables */
4547 rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
4548 if (rc)
4549 return rc;
4550
4551 if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
4552 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4553 if (decode == true)
4554 cmd |= command_bits;
4555 else
4556 cmd &= ~command_bits;
4557 pci_write_config_word(dev, PCI_COMMAND, cmd);
4558 }
4559
4560 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
4561 return 0;
4562
4563 bus = dev->bus;
4564 while (bus) {
4565 bridge = bus->self;
4566 if (bridge) {
4567 pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
4568 &cmd);
4569 if (decode == true)
4570 cmd |= PCI_BRIDGE_CTL_VGA;
4571 else
4572 cmd &= ~PCI_BRIDGE_CTL_VGA;
4573 pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
4574 cmd);
4575 }
4576 bus = bus->parent;
4577 }
4578 return 0;
4579}
4580
4581bool pci_device_is_present(struct pci_dev *pdev)
4582{
4583 u32 v;
4584
4585 return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
4586}
4587EXPORT_SYMBOL_GPL(pci_device_is_present);
4588
4589void pci_ignore_hotplug(struct pci_dev *dev)
4590{
4591 struct pci_dev *bridge = dev->bus->self;
4592
4593 dev->ignore_hotplug = 1;
4594 /* Propagate the "ignore hotplug" setting to the parent bridge. */
4595 if (bridge)
4596 bridge->ignore_hotplug = 1;
4597}
4598EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
4599
4600#define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
4601static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
4602static DEFINE_SPINLOCK(resource_alignment_lock);
4603
4604/**
4605 * pci_specified_resource_alignment - get resource alignment specified by user.
4606 * @dev: the PCI device to get
4607 *
4608 * RETURNS: Resource alignment if it is specified.
4609 * Zero if it is not specified.
4610 */
4611static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev)
4612{
4613 int seg, bus, slot, func, align_order, count;
4614 resource_size_t align = 0;
4615 char *p;
4616
4617 spin_lock(&resource_alignment_lock);
4618 p = resource_alignment_param;
4619 while (*p) {
4620 count = 0;
4621 if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
4622 p[count] == '@') {
4623 p += count + 1;
4624 } else {
4625 align_order = -1;
4626 }
4627 if (sscanf(p, "%x:%x:%x.%x%n",
4628 &seg, &bus, &slot, &func, &count) != 4) {
4629 seg = 0;
4630 if (sscanf(p, "%x:%x.%x%n",
4631 &bus, &slot, &func, &count) != 3) {
4632 /* Invalid format */
4633 printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: %s\n",
4634 p);
4635 break;
4636 }
4637 }
4638 p += count;
4639 if (seg == pci_domain_nr(dev->bus) &&
4640 bus == dev->bus->number &&
4641 slot == PCI_SLOT(dev->devfn) &&
4642 func == PCI_FUNC(dev->devfn)) {
4643 if (align_order == -1)
4644 align = PAGE_SIZE;
4645 else
4646 align = 1 << align_order;
4647 /* Found */
4648 break;
4649 }
4650 if (*p != ';' && *p != ',') {
4651 /* End of param or invalid format */
4652 break;
4653 }
4654 p++;
4655 }
4656 spin_unlock(&resource_alignment_lock);
4657 return align;
4658}
4659
4660/*
4661 * This function disables memory decoding and releases memory resources
4662 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
4663 * It also rounds up size to specified alignment.
4664 * Later on, the kernel will assign page-aligned memory resource back
4665 * to the device.
4666 */
4667void pci_reassigndev_resource_alignment(struct pci_dev *dev)
4668{
4669 int i;
4670 struct resource *r;
4671 resource_size_t align, size;
4672 u16 command;
4673
4674 /* check if specified PCI is target device to reassign */
4675 align = pci_specified_resource_alignment(dev);
4676 if (!align)
4677 return;
4678
4679 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
4680 (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
4681 dev_warn(&dev->dev,
4682 "Can't reassign resources to host bridge.\n");
4683 return;
4684 }
4685
4686 dev_info(&dev->dev,
4687 "Disabling memory decoding and releasing memory resources.\n");
4688 pci_read_config_word(dev, PCI_COMMAND, &command);
4689 command &= ~PCI_COMMAND_MEMORY;
4690 pci_write_config_word(dev, PCI_COMMAND, command);
4691
4692 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++) {
4693 r = &dev->resource[i];
4694 if (!(r->flags & IORESOURCE_MEM))
4695 continue;
4696 size = resource_size(r);
4697 if (size < align) {
4698 size = align;
4699 dev_info(&dev->dev,
4700 "Rounding up size of resource #%d to %#llx.\n",
4701 i, (unsigned long long)size);
4702 }
4703 r->flags |= IORESOURCE_UNSET;
4704 r->end = size - 1;
4705 r->start = 0;
4706 }
4707 /* Need to disable bridge's resource window,
4708 * to enable the kernel to reassign new resource
4709 * window later on.
4710 */
4711 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE &&
4712 (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
4713 for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
4714 r = &dev->resource[i];
4715 if (!(r->flags & IORESOURCE_MEM))
4716 continue;
4717 r->flags |= IORESOURCE_UNSET;
4718 r->end = resource_size(r) - 1;
4719 r->start = 0;
4720 }
4721 pci_disable_bridge_window(dev);
4722 }
4723}
4724
4725static ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
4726{
4727 if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
4728 count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
4729 spin_lock(&resource_alignment_lock);
4730 strncpy(resource_alignment_param, buf, count);
4731 resource_alignment_param[count] = '\0';
4732 spin_unlock(&resource_alignment_lock);
4733 return count;
4734}
4735
4736static ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
4737{
4738 size_t count;
4739 spin_lock(&resource_alignment_lock);
4740 count = snprintf(buf, size, "%s", resource_alignment_param);
4741 spin_unlock(&resource_alignment_lock);
4742 return count;
4743}
4744
4745static ssize_t pci_resource_alignment_show(struct bus_type *bus, char *buf)
4746{
4747 return pci_get_resource_alignment_param(buf, PAGE_SIZE);
4748}
4749
4750static ssize_t pci_resource_alignment_store(struct bus_type *bus,
4751 const char *buf, size_t count)
4752{
4753 return pci_set_resource_alignment_param(buf, count);
4754}
4755
4756BUS_ATTR(resource_alignment, 0644, pci_resource_alignment_show,
4757 pci_resource_alignment_store);
4758
4759static int __init pci_resource_alignment_sysfs_init(void)
4760{
4761 return bus_create_file(&pci_bus_type,
4762 &bus_attr_resource_alignment);
4763}
4764late_initcall(pci_resource_alignment_sysfs_init);
4765
4766static void pci_no_domains(void)
4767{
4768#ifdef CONFIG_PCI_DOMAINS
4769 pci_domains_supported = 0;
4770#endif
4771}
4772
4773#ifdef CONFIG_PCI_DOMAINS
4774static atomic_t __domain_nr = ATOMIC_INIT(-1);
4775
4776int pci_get_new_domain_nr(void)
4777{
4778 return atomic_inc_return(&__domain_nr);
4779}
4780
4781#ifdef CONFIG_PCI_DOMAINS_GENERIC
4782void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
4783{
4784 static int use_dt_domains = -1;
4785 int domain = -1;
4786
4787 if (parent)
4788 domain = of_get_pci_domain_nr(parent->of_node);
4789 /*
4790 * Check DT domain and use_dt_domains values.
4791 *
4792 * If DT domain property is valid (domain >= 0) and
4793 * use_dt_domains != 0, the DT assignment is valid since this means
4794 * we have not previously allocated a domain number by using
4795 * pci_get_new_domain_nr(); we should also update use_dt_domains to
4796 * 1, to indicate that we have just assigned a domain number from
4797 * DT.
4798 *
4799 * If DT domain property value is not valid (ie domain < 0), and we
4800 * have not previously assigned a domain number from DT
4801 * (use_dt_domains != 1) we should assign a domain number by
4802 * using the:
4803 *
4804 * pci_get_new_domain_nr()
4805 *
4806 * API and update the use_dt_domains value to keep track of method we
4807 * are using to assign domain numbers (use_dt_domains = 0).
4808 *
4809 * All other combinations imply we have a platform that is trying
4810 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
4811 * which is a recipe for domain mishandling and it is prevented by
4812 * invalidating the domain value (domain = -1) and printing a
4813 * corresponding error.
4814 */
4815 if (domain >= 0 && use_dt_domains) {
4816 use_dt_domains = 1;
4817 } else if (domain < 0 && use_dt_domains != 1) {
4818 use_dt_domains = 0;
4819 domain = pci_get_new_domain_nr();
4820 } else {
4821 dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
4822 parent->of_node->full_name);
4823 domain = -1;
4824 }
4825
4826 bus->domain_nr = domain;
4827}
4828#endif
4829#endif
4830
4831/**
4832 * pci_ext_cfg_avail - can we access extended PCI config space?
4833 *
4834 * Returns 1 if we can access PCI extended config space (offsets
4835 * greater than 0xff). This is the default implementation. Architecture
4836 * implementations can override this.
4837 */
4838int __weak pci_ext_cfg_avail(void)
4839{
4840 return 1;
4841}
4842
4843void __weak pci_fixup_cardbus(struct pci_bus *bus)
4844{
4845}
4846EXPORT_SYMBOL(pci_fixup_cardbus);
4847
4848static int __init pci_setup(char *str)
4849{
4850 while (str) {
4851 char *k = strchr(str, ',');
4852 if (k)
4853 *k++ = 0;
4854 if (*str && (str = pcibios_setup(str)) && *str) {
4855 if (!strcmp(str, "nomsi")) {
4856 pci_no_msi();
4857 } else if (!strcmp(str, "noaer")) {
4858 pci_no_aer();
4859 } else if (!strncmp(str, "realloc=", 8)) {
4860 pci_realloc_get_opt(str + 8);
4861 } else if (!strncmp(str, "realloc", 7)) {
4862 pci_realloc_get_opt("on");
4863 } else if (!strcmp(str, "nodomains")) {
4864 pci_no_domains();
4865 } else if (!strncmp(str, "noari", 5)) {
4866 pcie_ari_disabled = true;
4867 } else if (!strncmp(str, "cbiosize=", 9)) {
4868 pci_cardbus_io_size = memparse(str + 9, &str);
4869 } else if (!strncmp(str, "cbmemsize=", 10)) {
4870 pci_cardbus_mem_size = memparse(str + 10, &str);
4871 } else if (!strncmp(str, "resource_alignment=", 19)) {
4872 pci_set_resource_alignment_param(str + 19,
4873 strlen(str + 19));
4874 } else if (!strncmp(str, "ecrc=", 5)) {
4875 pcie_ecrc_get_policy(str + 5);
4876 } else if (!strncmp(str, "hpiosize=", 9)) {
4877 pci_hotplug_io_size = memparse(str + 9, &str);
4878 } else if (!strncmp(str, "hpmemsize=", 10)) {
4879 pci_hotplug_mem_size = memparse(str + 10, &str);
4880 } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
4881 pcie_bus_config = PCIE_BUS_TUNE_OFF;
4882 } else if (!strncmp(str, "pcie_bus_safe", 13)) {
4883 pcie_bus_config = PCIE_BUS_SAFE;
4884 } else if (!strncmp(str, "pcie_bus_perf", 13)) {
4885 pcie_bus_config = PCIE_BUS_PERFORMANCE;
4886 } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
4887 pcie_bus_config = PCIE_BUS_PEER2PEER;
4888 } else if (!strncmp(str, "pcie_scan_all", 13)) {
4889 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
4890 } else {
4891 printk(KERN_ERR "PCI: Unknown option `%s'\n",
4892 str);
4893 }
4894 }
4895 str = k;
4896 }
4897 return 0;
4898}
4899early_param("pci", pci_setup);