xref: /linux/drivers/pci/probe.c (revision 54ff328b46e58568c4b3350c2fa3223ef862e5a4)
1 /*
2  * probe.c - PCI detection and setup code
3  */
4 
5 #include <linux/kernel.h>
6 #include <linux/delay.h>
7 #include <linux/init.h>
8 #include <linux/pci.h>
9 #include <linux/slab.h>
10 #include <linux/module.h>
11 #include <linux/cpumask.h>
12 #include <linux/pci-aspm.h>
13 #include "pci.h"
14 
15 #define CARDBUS_LATENCY_TIMER	176	/* secondary latency timer */
16 #define CARDBUS_RESERVE_BUSNR	3
17 
18 /* Ugh.  Need to stop exporting this to modules. */
19 LIST_HEAD(pci_root_buses);
20 EXPORT_SYMBOL(pci_root_buses);
21 
22 
23 static int find_anything(struct device *dev, void *data)
24 {
25 	return 1;
26 }
27 
28 /*
29  * Some device drivers need know if pci is initiated.
30  * Basically, we think pci is not initiated when there
31  * is no device to be found on the pci_bus_type.
32  */
33 int no_pci_devices(void)
34 {
35 	struct device *dev;
36 	int no_devices;
37 
38 	dev = bus_find_device(&pci_bus_type, NULL, NULL, find_anything);
39 	no_devices = (dev == NULL);
40 	put_device(dev);
41 	return no_devices;
42 }
43 EXPORT_SYMBOL(no_pci_devices);
44 
45 /*
46  * PCI Bus Class Devices
47  */
48 static ssize_t pci_bus_show_cpuaffinity(struct device *dev,
49 					int type,
50 					struct device_attribute *attr,
51 					char *buf)
52 {
53 	int ret;
54 	const struct cpumask *cpumask;
55 
56 	cpumask = cpumask_of_pcibus(to_pci_bus(dev));
57 	ret = type?
58 		cpulist_scnprintf(buf, PAGE_SIZE-2, cpumask) :
59 		cpumask_scnprintf(buf, PAGE_SIZE-2, cpumask);
60 	buf[ret++] = '\n';
61 	buf[ret] = '\0';
62 	return ret;
63 }
64 
65 static ssize_t inline pci_bus_show_cpumaskaffinity(struct device *dev,
66 					struct device_attribute *attr,
67 					char *buf)
68 {
69 	return pci_bus_show_cpuaffinity(dev, 0, attr, buf);
70 }
71 
72 static ssize_t inline pci_bus_show_cpulistaffinity(struct device *dev,
73 					struct device_attribute *attr,
74 					char *buf)
75 {
76 	return pci_bus_show_cpuaffinity(dev, 1, attr, buf);
77 }
78 
79 DEVICE_ATTR(cpuaffinity,     S_IRUGO, pci_bus_show_cpumaskaffinity, NULL);
80 DEVICE_ATTR(cpulistaffinity, S_IRUGO, pci_bus_show_cpulistaffinity, NULL);
81 
82 /*
83  * PCI Bus Class
84  */
85 static void release_pcibus_dev(struct device *dev)
86 {
87 	struct pci_bus *pci_bus = to_pci_bus(dev);
88 
89 	if (pci_bus->bridge)
90 		put_device(pci_bus->bridge);
91 	kfree(pci_bus);
92 }
93 
94 static struct class pcibus_class = {
95 	.name		= "pci_bus",
96 	.dev_release	= &release_pcibus_dev,
97 };
98 
99 static int __init pcibus_class_init(void)
100 {
101 	return class_register(&pcibus_class);
102 }
103 postcore_initcall(pcibus_class_init);
104 
105 /*
106  * Translate the low bits of the PCI base
107  * to the resource type
108  */
109 static inline unsigned int pci_calc_resource_flags(unsigned int flags)
110 {
111 	if (flags & PCI_BASE_ADDRESS_SPACE_IO)
112 		return IORESOURCE_IO;
113 
114 	if (flags & PCI_BASE_ADDRESS_MEM_PREFETCH)
115 		return IORESOURCE_MEM | IORESOURCE_PREFETCH;
116 
117 	return IORESOURCE_MEM;
118 }
119 
120 static u64 pci_size(u64 base, u64 maxbase, u64 mask)
121 {
122 	u64 size = mask & maxbase;	/* Find the significant bits */
123 	if (!size)
124 		return 0;
125 
126 	/* Get the lowest of them to find the decode size, and
127 	   from that the extent.  */
128 	size = (size & ~(size-1)) - 1;
129 
130 	/* base == maxbase can be valid only if the BAR has
131 	   already been programmed with all 1s.  */
132 	if (base == maxbase && ((base | size) & mask) != mask)
133 		return 0;
134 
135 	return size;
136 }
137 
138 static inline enum pci_bar_type decode_bar(struct resource *res, u32 bar)
139 {
140 	if ((bar & PCI_BASE_ADDRESS_SPACE) == PCI_BASE_ADDRESS_SPACE_IO) {
141 		res->flags = bar & ~PCI_BASE_ADDRESS_IO_MASK;
142 		return pci_bar_io;
143 	}
144 
145 	res->flags = bar & ~PCI_BASE_ADDRESS_MEM_MASK;
146 
147 	if (res->flags & PCI_BASE_ADDRESS_MEM_TYPE_64)
148 		return pci_bar_mem64;
149 	return pci_bar_mem32;
150 }
151 
152 /**
153  * pci_read_base - read a PCI BAR
154  * @dev: the PCI device
155  * @type: type of the BAR
156  * @res: resource buffer to be filled in
157  * @pos: BAR position in the config space
158  *
159  * Returns 1 if the BAR is 64-bit, or 0 if 32-bit.
160  */
161 int __pci_read_base(struct pci_dev *dev, enum pci_bar_type type,
162 			struct resource *res, unsigned int pos)
163 {
164 	u32 l, sz, mask;
165 
166 	mask = type ? ~PCI_ROM_ADDRESS_ENABLE : ~0;
167 
168 	res->name = pci_name(dev);
169 
170 	pci_read_config_dword(dev, pos, &l);
171 	pci_write_config_dword(dev, pos, mask);
172 	pci_read_config_dword(dev, pos, &sz);
173 	pci_write_config_dword(dev, pos, l);
174 
175 	/*
176 	 * All bits set in sz means the device isn't working properly.
177 	 * If the BAR isn't implemented, all bits must be 0.  If it's a
178 	 * memory BAR or a ROM, bit 0 must be clear; if it's an io BAR, bit
179 	 * 1 must be clear.
180 	 */
181 	if (!sz || sz == 0xffffffff)
182 		goto fail;
183 
184 	/*
185 	 * I don't know how l can have all bits set.  Copied from old code.
186 	 * Maybe it fixes a bug on some ancient platform.
187 	 */
188 	if (l == 0xffffffff)
189 		l = 0;
190 
191 	if (type == pci_bar_unknown) {
192 		type = decode_bar(res, l);
193 		res->flags |= pci_calc_resource_flags(l) | IORESOURCE_SIZEALIGN;
194 		if (type == pci_bar_io) {
195 			l &= PCI_BASE_ADDRESS_IO_MASK;
196 			mask = PCI_BASE_ADDRESS_IO_MASK & 0xffff;
197 		} else {
198 			l &= PCI_BASE_ADDRESS_MEM_MASK;
199 			mask = (u32)PCI_BASE_ADDRESS_MEM_MASK;
200 		}
201 	} else {
202 		res->flags |= (l & IORESOURCE_ROM_ENABLE);
203 		l &= PCI_ROM_ADDRESS_MASK;
204 		mask = (u32)PCI_ROM_ADDRESS_MASK;
205 	}
206 
207 	if (type == pci_bar_mem64) {
208 		u64 l64 = l;
209 		u64 sz64 = sz;
210 		u64 mask64 = mask | (u64)~0 << 32;
211 
212 		pci_read_config_dword(dev, pos + 4, &l);
213 		pci_write_config_dword(dev, pos + 4, ~0);
214 		pci_read_config_dword(dev, pos + 4, &sz);
215 		pci_write_config_dword(dev, pos + 4, l);
216 
217 		l64 |= ((u64)l << 32);
218 		sz64 |= ((u64)sz << 32);
219 
220 		sz64 = pci_size(l64, sz64, mask64);
221 
222 		if (!sz64)
223 			goto fail;
224 
225 		if ((sizeof(resource_size_t) < 8) && (sz64 > 0x100000000ULL)) {
226 			dev_err(&dev->dev, "can't handle 64-bit BAR\n");
227 			goto fail;
228 		} else if ((sizeof(resource_size_t) < 8) && l) {
229 			/* Address above 32-bit boundary; disable the BAR */
230 			pci_write_config_dword(dev, pos, 0);
231 			pci_write_config_dword(dev, pos + 4, 0);
232 			res->start = 0;
233 			res->end = sz64;
234 		} else {
235 			res->start = l64;
236 			res->end = l64 + sz64;
237 			dev_printk(KERN_DEBUG, &dev->dev,
238 				"reg %x 64bit mmio: %pR\n", pos, res);
239 		}
240 	} else {
241 		sz = pci_size(l, sz, mask);
242 
243 		if (!sz)
244 			goto fail;
245 
246 		res->start = l;
247 		res->end = l + sz;
248 
249 		dev_printk(KERN_DEBUG, &dev->dev, "reg %x %s: %pR\n", pos,
250 			(res->flags & IORESOURCE_IO) ? "io port" : "32bit mmio",
251 			res);
252 	}
253 
254  out:
255 	return (type == pci_bar_mem64) ? 1 : 0;
256  fail:
257 	res->flags = 0;
258 	goto out;
259 }
260 
261 static void pci_read_bases(struct pci_dev *dev, unsigned int howmany, int rom)
262 {
263 	unsigned int pos, reg;
264 
265 	for (pos = 0; pos < howmany; pos++) {
266 		struct resource *res = &dev->resource[pos];
267 		reg = PCI_BASE_ADDRESS_0 + (pos << 2);
268 		pos += __pci_read_base(dev, pci_bar_unknown, res, reg);
269 	}
270 
271 	if (rom) {
272 		struct resource *res = &dev->resource[PCI_ROM_RESOURCE];
273 		dev->rom_base_reg = rom;
274 		res->flags = IORESOURCE_MEM | IORESOURCE_PREFETCH |
275 				IORESOURCE_READONLY | IORESOURCE_CACHEABLE |
276 				IORESOURCE_SIZEALIGN;
277 		__pci_read_base(dev, pci_bar_mem32, res, rom);
278 	}
279 }
280 
281 void __devinit pci_read_bridge_bases(struct pci_bus *child)
282 {
283 	struct pci_dev *dev = child->self;
284 	u8 io_base_lo, io_limit_lo;
285 	u16 mem_base_lo, mem_limit_lo;
286 	unsigned long base, limit;
287 	struct resource *res;
288 	int i;
289 
290 	if (!child->parent)	/* It's a host bus, nothing to read */
291 		return;
292 
293 	if (dev->transparent) {
294 		dev_info(&dev->dev, "transparent bridge\n");
295 		for(i = 3; i < PCI_BUS_NUM_RESOURCES; i++)
296 			child->resource[i] = child->parent->resource[i - 3];
297 	}
298 
299 	res = child->resource[0];
300 	pci_read_config_byte(dev, PCI_IO_BASE, &io_base_lo);
301 	pci_read_config_byte(dev, PCI_IO_LIMIT, &io_limit_lo);
302 	base = (io_base_lo & PCI_IO_RANGE_MASK) << 8;
303 	limit = (io_limit_lo & PCI_IO_RANGE_MASK) << 8;
304 
305 	if ((io_base_lo & PCI_IO_RANGE_TYPE_MASK) == PCI_IO_RANGE_TYPE_32) {
306 		u16 io_base_hi, io_limit_hi;
307 		pci_read_config_word(dev, PCI_IO_BASE_UPPER16, &io_base_hi);
308 		pci_read_config_word(dev, PCI_IO_LIMIT_UPPER16, &io_limit_hi);
309 		base |= (io_base_hi << 16);
310 		limit |= (io_limit_hi << 16);
311 	}
312 
313 	if (base <= limit) {
314 		res->flags = (io_base_lo & PCI_IO_RANGE_TYPE_MASK) | IORESOURCE_IO;
315 		if (!res->start)
316 			res->start = base;
317 		if (!res->end)
318 			res->end = limit + 0xfff;
319 		dev_printk(KERN_DEBUG, &dev->dev, "bridge io port: %pR\n", res);
320 	}
321 
322 	res = child->resource[1];
323 	pci_read_config_word(dev, PCI_MEMORY_BASE, &mem_base_lo);
324 	pci_read_config_word(dev, PCI_MEMORY_LIMIT, &mem_limit_lo);
325 	base = (mem_base_lo & PCI_MEMORY_RANGE_MASK) << 16;
326 	limit = (mem_limit_lo & PCI_MEMORY_RANGE_MASK) << 16;
327 	if (base <= limit) {
328 		res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM;
329 		res->start = base;
330 		res->end = limit + 0xfffff;
331 		dev_printk(KERN_DEBUG, &dev->dev, "bridge 32bit mmio: %pR\n",
332 			res);
333 	}
334 
335 	res = child->resource[2];
336 	pci_read_config_word(dev, PCI_PREF_MEMORY_BASE, &mem_base_lo);
337 	pci_read_config_word(dev, PCI_PREF_MEMORY_LIMIT, &mem_limit_lo);
338 	base = (mem_base_lo & PCI_PREF_RANGE_MASK) << 16;
339 	limit = (mem_limit_lo & PCI_PREF_RANGE_MASK) << 16;
340 
341 	if ((mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) {
342 		u32 mem_base_hi, mem_limit_hi;
343 		pci_read_config_dword(dev, PCI_PREF_BASE_UPPER32, &mem_base_hi);
344 		pci_read_config_dword(dev, PCI_PREF_LIMIT_UPPER32, &mem_limit_hi);
345 
346 		/*
347 		 * Some bridges set the base > limit by default, and some
348 		 * (broken) BIOSes do not initialize them.  If we find
349 		 * this, just assume they are not being used.
350 		 */
351 		if (mem_base_hi <= mem_limit_hi) {
352 #if BITS_PER_LONG == 64
353 			base |= ((long) mem_base_hi) << 32;
354 			limit |= ((long) mem_limit_hi) << 32;
355 #else
356 			if (mem_base_hi || mem_limit_hi) {
357 				dev_err(&dev->dev, "can't handle 64-bit "
358 					"address space for bridge\n");
359 				return;
360 			}
361 #endif
362 		}
363 	}
364 	if (base <= limit) {
365 		res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM | IORESOURCE_PREFETCH;
366 		res->start = base;
367 		res->end = limit + 0xfffff;
368 		dev_printk(KERN_DEBUG, &dev->dev, "bridge %sbit mmio pref: %pR\n",
369 			(res->flags & PCI_PREF_RANGE_TYPE_64) ? "64" : "32",
370 			res);
371 	}
372 }
373 
374 static struct pci_bus * pci_alloc_bus(void)
375 {
376 	struct pci_bus *b;
377 
378 	b = kzalloc(sizeof(*b), GFP_KERNEL);
379 	if (b) {
380 		INIT_LIST_HEAD(&b->node);
381 		INIT_LIST_HEAD(&b->children);
382 		INIT_LIST_HEAD(&b->devices);
383 		INIT_LIST_HEAD(&b->slots);
384 	}
385 	return b;
386 }
387 
388 static struct pci_bus *pci_alloc_child_bus(struct pci_bus *parent,
389 					   struct pci_dev *bridge, int busnr)
390 {
391 	struct pci_bus *child;
392 	int i;
393 
394 	/*
395 	 * Allocate a new bus, and inherit stuff from the parent..
396 	 */
397 	child = pci_alloc_bus();
398 	if (!child)
399 		return NULL;
400 
401 	child->parent = parent;
402 	child->ops = parent->ops;
403 	child->sysdata = parent->sysdata;
404 	child->bus_flags = parent->bus_flags;
405 
406 	/* initialize some portions of the bus device, but don't register it
407 	 * now as the parent is not properly set up yet.  This device will get
408 	 * registered later in pci_bus_add_devices()
409 	 */
410 	child->dev.class = &pcibus_class;
411 	dev_set_name(&child->dev, "%04x:%02x", pci_domain_nr(child), busnr);
412 
413 	/*
414 	 * Set up the primary, secondary and subordinate
415 	 * bus numbers.
416 	 */
417 	child->number = child->secondary = busnr;
418 	child->primary = parent->secondary;
419 	child->subordinate = 0xff;
420 
421 	if (!bridge)
422 		return child;
423 
424 	child->self = bridge;
425 	child->bridge = get_device(&bridge->dev);
426 
427 	/* Set up default resource pointers and names.. */
428 	for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++) {
429 		child->resource[i] = &bridge->resource[PCI_BRIDGE_RESOURCES+i];
430 		child->resource[i]->name = child->name;
431 	}
432 	bridge->subordinate = child;
433 
434 	return child;
435 }
436 
437 struct pci_bus *__ref pci_add_new_bus(struct pci_bus *parent, struct pci_dev *dev, int busnr)
438 {
439 	struct pci_bus *child;
440 
441 	child = pci_alloc_child_bus(parent, dev, busnr);
442 	if (child) {
443 		down_write(&pci_bus_sem);
444 		list_add_tail(&child->node, &parent->children);
445 		up_write(&pci_bus_sem);
446 	}
447 	return child;
448 }
449 
450 static void pci_fixup_parent_subordinate_busnr(struct pci_bus *child, int max)
451 {
452 	struct pci_bus *parent = child->parent;
453 
454 	/* Attempts to fix that up are really dangerous unless
455 	   we're going to re-assign all bus numbers. */
456 	if (!pcibios_assign_all_busses())
457 		return;
458 
459 	while (parent->parent && parent->subordinate < max) {
460 		parent->subordinate = max;
461 		pci_write_config_byte(parent->self, PCI_SUBORDINATE_BUS, max);
462 		parent = parent->parent;
463 	}
464 }
465 
466 /*
467  * If it's a bridge, configure it and scan the bus behind it.
468  * For CardBus bridges, we don't scan behind as the devices will
469  * be handled by the bridge driver itself.
470  *
471  * We need to process bridges in two passes -- first we scan those
472  * already configured by the BIOS and after we are done with all of
473  * them, we proceed to assigning numbers to the remaining buses in
474  * order to avoid overlaps between old and new bus numbers.
475  */
476 int __devinit pci_scan_bridge(struct pci_bus *bus, struct pci_dev *dev, int max, int pass)
477 {
478 	struct pci_bus *child;
479 	int is_cardbus = (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS);
480 	u32 buses, i, j = 0;
481 	u16 bctl;
482 	int broken = 0;
483 
484 	pci_read_config_dword(dev, PCI_PRIMARY_BUS, &buses);
485 
486 	dev_dbg(&dev->dev, "scanning behind bridge, config %06x, pass %d\n",
487 		buses & 0xffffff, pass);
488 
489 	/* Check if setup is sensible at all */
490 	if (!pass &&
491 	    ((buses & 0xff) != bus->number || ((buses >> 8) & 0xff) <= bus->number)) {
492 		dev_dbg(&dev->dev, "bus configuration invalid, reconfiguring\n");
493 		broken = 1;
494 	}
495 
496 	/* Disable MasterAbortMode during probing to avoid reporting
497 	   of bus errors (in some architectures) */
498 	pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl);
499 	pci_write_config_word(dev, PCI_BRIDGE_CONTROL,
500 			      bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT);
501 
502 	if ((buses & 0xffff00) && !pcibios_assign_all_busses() && !is_cardbus && !broken) {
503 		unsigned int cmax, busnr;
504 		/*
505 		 * Bus already configured by firmware, process it in the first
506 		 * pass and just note the configuration.
507 		 */
508 		if (pass)
509 			goto out;
510 		busnr = (buses >> 8) & 0xFF;
511 
512 		/*
513 		 * If we already got to this bus through a different bridge,
514 		 * don't re-add it. This can happen with the i450NX chipset.
515 		 *
516 		 * However, we continue to descend down the hierarchy and
517 		 * scan remaining child buses.
518 		 */
519 		child = pci_find_bus(pci_domain_nr(bus), busnr);
520 		if (!child) {
521 			child = pci_add_new_bus(bus, dev, busnr);
522 			if (!child)
523 				goto out;
524 			child->primary = buses & 0xFF;
525 			child->subordinate = (buses >> 16) & 0xFF;
526 			child->bridge_ctl = bctl;
527 		}
528 
529 		cmax = pci_scan_child_bus(child);
530 		if (cmax > max)
531 			max = cmax;
532 		if (child->subordinate > max)
533 			max = child->subordinate;
534 	} else {
535 		/*
536 		 * We need to assign a number to this bus which we always
537 		 * do in the second pass.
538 		 */
539 		if (!pass) {
540 			if (pcibios_assign_all_busses() || broken)
541 				/* Temporarily disable forwarding of the
542 				   configuration cycles on all bridges in
543 				   this bus segment to avoid possible
544 				   conflicts in the second pass between two
545 				   bridges programmed with overlapping
546 				   bus ranges. */
547 				pci_write_config_dword(dev, PCI_PRIMARY_BUS,
548 						       buses & ~0xffffff);
549 			goto out;
550 		}
551 
552 		/* Clear errors */
553 		pci_write_config_word(dev, PCI_STATUS, 0xffff);
554 
555 		/* Prevent assigning a bus number that already exists.
556 		 * This can happen when a bridge is hot-plugged */
557 		if (pci_find_bus(pci_domain_nr(bus), max+1))
558 			goto out;
559 		child = pci_add_new_bus(bus, dev, ++max);
560 		buses = (buses & 0xff000000)
561 		      | ((unsigned int)(child->primary)     <<  0)
562 		      | ((unsigned int)(child->secondary)   <<  8)
563 		      | ((unsigned int)(child->subordinate) << 16);
564 
565 		/*
566 		 * yenta.c forces a secondary latency timer of 176.
567 		 * Copy that behaviour here.
568 		 */
569 		if (is_cardbus) {
570 			buses &= ~0xff000000;
571 			buses |= CARDBUS_LATENCY_TIMER << 24;
572 		}
573 
574 		/*
575 		 * We need to blast all three values with a single write.
576 		 */
577 		pci_write_config_dword(dev, PCI_PRIMARY_BUS, buses);
578 
579 		if (!is_cardbus) {
580 			child->bridge_ctl = bctl;
581 			/*
582 			 * Adjust subordinate busnr in parent buses.
583 			 * We do this before scanning for children because
584 			 * some devices may not be detected if the bios
585 			 * was lazy.
586 			 */
587 			pci_fixup_parent_subordinate_busnr(child, max);
588 			/* Now we can scan all subordinate buses... */
589 			max = pci_scan_child_bus(child);
590 			/*
591 			 * now fix it up again since we have found
592 			 * the real value of max.
593 			 */
594 			pci_fixup_parent_subordinate_busnr(child, max);
595 		} else {
596 			/*
597 			 * For CardBus bridges, we leave 4 bus numbers
598 			 * as cards with a PCI-to-PCI bridge can be
599 			 * inserted later.
600 			 */
601 			for (i=0; i<CARDBUS_RESERVE_BUSNR; i++) {
602 				struct pci_bus *parent = bus;
603 				if (pci_find_bus(pci_domain_nr(bus),
604 							max+i+1))
605 					break;
606 				while (parent->parent) {
607 					if ((!pcibios_assign_all_busses()) &&
608 					    (parent->subordinate > max) &&
609 					    (parent->subordinate <= max+i)) {
610 						j = 1;
611 					}
612 					parent = parent->parent;
613 				}
614 				if (j) {
615 					/*
616 					 * Often, there are two cardbus bridges
617 					 * -- try to leave one valid bus number
618 					 * for each one.
619 					 */
620 					i /= 2;
621 					break;
622 				}
623 			}
624 			max += i;
625 			pci_fixup_parent_subordinate_busnr(child, max);
626 		}
627 		/*
628 		 * Set the subordinate bus number to its real value.
629 		 */
630 		child->subordinate = max;
631 		pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, max);
632 	}
633 
634 	sprintf(child->name,
635 		(is_cardbus ? "PCI CardBus %04x:%02x" : "PCI Bus %04x:%02x"),
636 		pci_domain_nr(bus), child->number);
637 
638 	/* Has only triggered on CardBus, fixup is in yenta_socket */
639 	while (bus->parent) {
640 		if ((child->subordinate > bus->subordinate) ||
641 		    (child->number > bus->subordinate) ||
642 		    (child->number < bus->number) ||
643 		    (child->subordinate < bus->number)) {
644 			pr_debug("PCI: Bus #%02x (-#%02x) is %s "
645 				"hidden behind%s bridge #%02x (-#%02x)\n",
646 				child->number, child->subordinate,
647 				(bus->number > child->subordinate &&
648 				 bus->subordinate < child->number) ?
649 					"wholly" : "partially",
650 				bus->self->transparent ? " transparent" : "",
651 				bus->number, bus->subordinate);
652 		}
653 		bus = bus->parent;
654 	}
655 
656 out:
657 	pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl);
658 
659 	return max;
660 }
661 
662 /*
663  * Read interrupt line and base address registers.
664  * The architecture-dependent code can tweak these, of course.
665  */
666 static void pci_read_irq(struct pci_dev *dev)
667 {
668 	unsigned char irq;
669 
670 	pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &irq);
671 	dev->pin = irq;
672 	if (irq)
673 		pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq);
674 	dev->irq = irq;
675 }
676 
677 static void set_pcie_port_type(struct pci_dev *pdev)
678 {
679 	int pos;
680 	u16 reg16;
681 
682 	pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
683 	if (!pos)
684 		return;
685 	pdev->is_pcie = 1;
686 	pci_read_config_word(pdev, pos + PCI_EXP_FLAGS, &reg16);
687 	pdev->pcie_type = (reg16 & PCI_EXP_FLAGS_TYPE) >> 4;
688 }
689 
690 #define LEGACY_IO_RESOURCE	(IORESOURCE_IO | IORESOURCE_PCI_FIXED)
691 
692 /**
693  * pci_setup_device - fill in class and map information of a device
694  * @dev: the device structure to fill
695  *
696  * Initialize the device structure with information about the device's
697  * vendor,class,memory and IO-space addresses,IRQ lines etc.
698  * Called at initialisation of the PCI subsystem and by CardBus services.
699  * Returns 0 on success and negative if unknown type of device (not normal,
700  * bridge or CardBus).
701  */
702 int pci_setup_device(struct pci_dev *dev)
703 {
704 	u32 class;
705 	u8 hdr_type;
706 	struct pci_slot *slot;
707 
708 	if (pci_read_config_byte(dev, PCI_HEADER_TYPE, &hdr_type))
709 		return -EIO;
710 
711 	dev->sysdata = dev->bus->sysdata;
712 	dev->dev.parent = dev->bus->bridge;
713 	dev->dev.bus = &pci_bus_type;
714 	dev->hdr_type = hdr_type & 0x7f;
715 	dev->multifunction = !!(hdr_type & 0x80);
716 	dev->error_state = pci_channel_io_normal;
717 	set_pcie_port_type(dev);
718 
719 	list_for_each_entry(slot, &dev->bus->slots, list)
720 		if (PCI_SLOT(dev->devfn) == slot->number)
721 			dev->slot = slot;
722 
723 	/* Assume 32-bit PCI; let 64-bit PCI cards (which are far rarer)
724 	   set this higher, assuming the system even supports it.  */
725 	dev->dma_mask = 0xffffffff;
726 
727 	dev_set_name(&dev->dev, "%04x:%02x:%02x.%d", pci_domain_nr(dev->bus),
728 		     dev->bus->number, PCI_SLOT(dev->devfn),
729 		     PCI_FUNC(dev->devfn));
730 
731 	pci_read_config_dword(dev, PCI_CLASS_REVISION, &class);
732 	dev->revision = class & 0xff;
733 	class >>= 8;				    /* upper 3 bytes */
734 	dev->class = class;
735 	class >>= 8;
736 
737 	dev_dbg(&dev->dev, "found [%04x:%04x] class %06x header type %02x\n",
738 		 dev->vendor, dev->device, class, dev->hdr_type);
739 
740 	/* need to have dev->class ready */
741 	dev->cfg_size = pci_cfg_space_size(dev);
742 
743 	/* "Unknown power state" */
744 	dev->current_state = PCI_UNKNOWN;
745 
746 	/* Early fixups, before probing the BARs */
747 	pci_fixup_device(pci_fixup_early, dev);
748 	/* device class may be changed after fixup */
749 	class = dev->class >> 8;
750 
751 	switch (dev->hdr_type) {		    /* header type */
752 	case PCI_HEADER_TYPE_NORMAL:		    /* standard header */
753 		if (class == PCI_CLASS_BRIDGE_PCI)
754 			goto bad;
755 		pci_read_irq(dev);
756 		pci_read_bases(dev, 6, PCI_ROM_ADDRESS);
757 		pci_read_config_word(dev, PCI_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
758 		pci_read_config_word(dev, PCI_SUBSYSTEM_ID, &dev->subsystem_device);
759 
760 		/*
761 		 *	Do the ugly legacy mode stuff here rather than broken chip
762 		 *	quirk code. Legacy mode ATA controllers have fixed
763 		 *	addresses. These are not always echoed in BAR0-3, and
764 		 *	BAR0-3 in a few cases contain junk!
765 		 */
766 		if (class == PCI_CLASS_STORAGE_IDE) {
767 			u8 progif;
768 			pci_read_config_byte(dev, PCI_CLASS_PROG, &progif);
769 			if ((progif & 1) == 0) {
770 				dev->resource[0].start = 0x1F0;
771 				dev->resource[0].end = 0x1F7;
772 				dev->resource[0].flags = LEGACY_IO_RESOURCE;
773 				dev->resource[1].start = 0x3F6;
774 				dev->resource[1].end = 0x3F6;
775 				dev->resource[1].flags = LEGACY_IO_RESOURCE;
776 			}
777 			if ((progif & 4) == 0) {
778 				dev->resource[2].start = 0x170;
779 				dev->resource[2].end = 0x177;
780 				dev->resource[2].flags = LEGACY_IO_RESOURCE;
781 				dev->resource[3].start = 0x376;
782 				dev->resource[3].end = 0x376;
783 				dev->resource[3].flags = LEGACY_IO_RESOURCE;
784 			}
785 		}
786 		break;
787 
788 	case PCI_HEADER_TYPE_BRIDGE:		    /* bridge header */
789 		if (class != PCI_CLASS_BRIDGE_PCI)
790 			goto bad;
791 		/* The PCI-to-PCI bridge spec requires that subtractive
792 		   decoding (i.e. transparent) bridge must have programming
793 		   interface code of 0x01. */
794 		pci_read_irq(dev);
795 		dev->transparent = ((dev->class & 0xff) == 1);
796 		pci_read_bases(dev, 2, PCI_ROM_ADDRESS1);
797 		break;
798 
799 	case PCI_HEADER_TYPE_CARDBUS:		    /* CardBus bridge header */
800 		if (class != PCI_CLASS_BRIDGE_CARDBUS)
801 			goto bad;
802 		pci_read_irq(dev);
803 		pci_read_bases(dev, 1, 0);
804 		pci_read_config_word(dev, PCI_CB_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
805 		pci_read_config_word(dev, PCI_CB_SUBSYSTEM_ID, &dev->subsystem_device);
806 		break;
807 
808 	default:				    /* unknown header */
809 		dev_err(&dev->dev, "unknown header type %02x, "
810 			"ignoring device\n", dev->hdr_type);
811 		return -EIO;
812 
813 	bad:
814 		dev_err(&dev->dev, "ignoring class %02x (doesn't match header "
815 			"type %02x)\n", class, dev->hdr_type);
816 		dev->class = PCI_CLASS_NOT_DEFINED;
817 	}
818 
819 	/* We found a fine healthy device, go go go... */
820 	return 0;
821 }
822 
823 static void pci_release_capabilities(struct pci_dev *dev)
824 {
825 	pci_vpd_release(dev);
826 	pci_iov_release(dev);
827 }
828 
829 /**
830  * pci_release_dev - free a pci device structure when all users of it are finished.
831  * @dev: device that's been disconnected
832  *
833  * Will be called only by the device core when all users of this pci device are
834  * done.
835  */
836 static void pci_release_dev(struct device *dev)
837 {
838 	struct pci_dev *pci_dev;
839 
840 	pci_dev = to_pci_dev(dev);
841 	pci_release_capabilities(pci_dev);
842 	kfree(pci_dev);
843 }
844 
845 /**
846  * pci_cfg_space_size - get the configuration space size of the PCI device.
847  * @dev: PCI device
848  *
849  * Regular PCI devices have 256 bytes, but PCI-X 2 and PCI Express devices
850  * have 4096 bytes.  Even if the device is capable, that doesn't mean we can
851  * access it.  Maybe we don't have a way to generate extended config space
852  * accesses, or the device is behind a reverse Express bridge.  So we try
853  * reading the dword at 0x100 which must either be 0 or a valid extended
854  * capability header.
855  */
856 int pci_cfg_space_size_ext(struct pci_dev *dev)
857 {
858 	u32 status;
859 	int pos = PCI_CFG_SPACE_SIZE;
860 
861 	if (pci_read_config_dword(dev, pos, &status) != PCIBIOS_SUCCESSFUL)
862 		goto fail;
863 	if (status == 0xffffffff)
864 		goto fail;
865 
866 	return PCI_CFG_SPACE_EXP_SIZE;
867 
868  fail:
869 	return PCI_CFG_SPACE_SIZE;
870 }
871 
872 int pci_cfg_space_size(struct pci_dev *dev)
873 {
874 	int pos;
875 	u32 status;
876 	u16 class;
877 
878 	class = dev->class >> 8;
879 	if (class == PCI_CLASS_BRIDGE_HOST)
880 		return pci_cfg_space_size_ext(dev);
881 
882 	pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
883 	if (!pos) {
884 		pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
885 		if (!pos)
886 			goto fail;
887 
888 		pci_read_config_dword(dev, pos + PCI_X_STATUS, &status);
889 		if (!(status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ)))
890 			goto fail;
891 	}
892 
893 	return pci_cfg_space_size_ext(dev);
894 
895  fail:
896 	return PCI_CFG_SPACE_SIZE;
897 }
898 
899 static void pci_release_bus_bridge_dev(struct device *dev)
900 {
901 	kfree(dev);
902 }
903 
904 struct pci_dev *alloc_pci_dev(void)
905 {
906 	struct pci_dev *dev;
907 
908 	dev = kzalloc(sizeof(struct pci_dev), GFP_KERNEL);
909 	if (!dev)
910 		return NULL;
911 
912 	INIT_LIST_HEAD(&dev->bus_list);
913 
914 	return dev;
915 }
916 EXPORT_SYMBOL(alloc_pci_dev);
917 
918 /*
919  * Read the config data for a PCI device, sanity-check it
920  * and fill in the dev structure...
921  */
922 static struct pci_dev *pci_scan_device(struct pci_bus *bus, int devfn)
923 {
924 	struct pci_dev *dev;
925 	u32 l;
926 	int delay = 1;
927 
928 	if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, &l))
929 		return NULL;
930 
931 	/* some broken boards return 0 or ~0 if a slot is empty: */
932 	if (l == 0xffffffff || l == 0x00000000 ||
933 	    l == 0x0000ffff || l == 0xffff0000)
934 		return NULL;
935 
936 	/* Configuration request Retry Status */
937 	while (l == 0xffff0001) {
938 		msleep(delay);
939 		delay *= 2;
940 		if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, &l))
941 			return NULL;
942 		/* Card hasn't responded in 60 seconds?  Must be stuck. */
943 		if (delay > 60 * 1000) {
944 			printk(KERN_WARNING "pci %04x:%02x:%02x.%d: not "
945 					"responding\n", pci_domain_nr(bus),
946 					bus->number, PCI_SLOT(devfn),
947 					PCI_FUNC(devfn));
948 			return NULL;
949 		}
950 	}
951 
952 	dev = alloc_pci_dev();
953 	if (!dev)
954 		return NULL;
955 
956 	dev->bus = bus;
957 	dev->devfn = devfn;
958 	dev->vendor = l & 0xffff;
959 	dev->device = (l >> 16) & 0xffff;
960 
961 	if (pci_setup_device(dev)) {
962 		kfree(dev);
963 		return NULL;
964 	}
965 
966 	return dev;
967 }
968 
969 static void pci_init_capabilities(struct pci_dev *dev)
970 {
971 	/* MSI/MSI-X list */
972 	pci_msi_init_pci_dev(dev);
973 
974 	/* Buffers for saving PCIe and PCI-X capabilities */
975 	pci_allocate_cap_save_buffers(dev);
976 
977 	/* Power Management */
978 	pci_pm_init(dev);
979 	platform_pci_wakeup_init(dev);
980 
981 	/* Vital Product Data */
982 	pci_vpd_pci22_init(dev);
983 
984 	/* Alternative Routing-ID Forwarding */
985 	pci_enable_ari(dev);
986 
987 	/* Single Root I/O Virtualization */
988 	pci_iov_init(dev);
989 }
990 
991 void pci_device_add(struct pci_dev *dev, struct pci_bus *bus)
992 {
993 	device_initialize(&dev->dev);
994 	dev->dev.release = pci_release_dev;
995 	pci_dev_get(dev);
996 
997 	dev->dev.dma_mask = &dev->dma_mask;
998 	dev->dev.dma_parms = &dev->dma_parms;
999 	dev->dev.coherent_dma_mask = 0xffffffffull;
1000 
1001 	pci_set_dma_max_seg_size(dev, 65536);
1002 	pci_set_dma_seg_boundary(dev, 0xffffffff);
1003 
1004 	/* Fix up broken headers */
1005 	pci_fixup_device(pci_fixup_header, dev);
1006 
1007 	/* Initialize various capabilities */
1008 	pci_init_capabilities(dev);
1009 
1010 	/*
1011 	 * Add the device to our list of discovered devices
1012 	 * and the bus list for fixup functions, etc.
1013 	 */
1014 	down_write(&pci_bus_sem);
1015 	list_add_tail(&dev->bus_list, &bus->devices);
1016 	up_write(&pci_bus_sem);
1017 }
1018 
1019 struct pci_dev *__ref pci_scan_single_device(struct pci_bus *bus, int devfn)
1020 {
1021 	struct pci_dev *dev;
1022 
1023 	dev = pci_get_slot(bus, devfn);
1024 	if (dev) {
1025 		pci_dev_put(dev);
1026 		return dev;
1027 	}
1028 
1029 	dev = pci_scan_device(bus, devfn);
1030 	if (!dev)
1031 		return NULL;
1032 
1033 	pci_device_add(dev, bus);
1034 
1035 	return dev;
1036 }
1037 EXPORT_SYMBOL(pci_scan_single_device);
1038 
1039 /**
1040  * pci_scan_slot - scan a PCI slot on a bus for devices.
1041  * @bus: PCI bus to scan
1042  * @devfn: slot number to scan (must have zero function.)
1043  *
1044  * Scan a PCI slot on the specified PCI bus for devices, adding
1045  * discovered devices to the @bus->devices list.  New devices
1046  * will not have is_added set.
1047  *
1048  * Returns the number of new devices found.
1049  */
1050 int pci_scan_slot(struct pci_bus *bus, int devfn)
1051 {
1052 	int fn, nr = 0;
1053 	struct pci_dev *dev;
1054 
1055 	dev = pci_scan_single_device(bus, devfn);
1056 	if (dev && !dev->is_added)	/* new device? */
1057 		nr++;
1058 
1059 	if ((dev && dev->multifunction) ||
1060 	    (!dev && pcibios_scan_all_fns(bus, devfn))) {
1061 		for (fn = 1; fn < 8; fn++) {
1062 			dev = pci_scan_single_device(bus, devfn + fn);
1063 			if (dev) {
1064 				if (!dev->is_added)
1065 					nr++;
1066 				dev->multifunction = 1;
1067 			}
1068 		}
1069 	}
1070 
1071 	/* only one slot has pcie device */
1072 	if (bus->self && nr)
1073 		pcie_aspm_init_link_state(bus->self);
1074 
1075 	return nr;
1076 }
1077 
1078 unsigned int __devinit pci_scan_child_bus(struct pci_bus *bus)
1079 {
1080 	unsigned int devfn, pass, max = bus->secondary;
1081 	struct pci_dev *dev;
1082 
1083 	pr_debug("PCI: Scanning bus %04x:%02x\n", pci_domain_nr(bus), bus->number);
1084 
1085 	/* Go find them, Rover! */
1086 	for (devfn = 0; devfn < 0x100; devfn += 8)
1087 		pci_scan_slot(bus, devfn);
1088 
1089 	/* Reserve buses for SR-IOV capability. */
1090 	max += pci_iov_bus_range(bus);
1091 
1092 	/*
1093 	 * After performing arch-dependent fixup of the bus, look behind
1094 	 * all PCI-to-PCI bridges on this bus.
1095 	 */
1096 	if (!bus->is_added) {
1097 		pr_debug("PCI: Fixups for bus %04x:%02x\n",
1098 			 pci_domain_nr(bus), bus->number);
1099 		pcibios_fixup_bus(bus);
1100 		if (pci_is_root_bus(bus))
1101 			bus->is_added = 1;
1102 	}
1103 
1104 	for (pass=0; pass < 2; pass++)
1105 		list_for_each_entry(dev, &bus->devices, bus_list) {
1106 			if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE ||
1107 			    dev->hdr_type == PCI_HEADER_TYPE_CARDBUS)
1108 				max = pci_scan_bridge(bus, dev, max, pass);
1109 		}
1110 
1111 	/*
1112 	 * We've scanned the bus and so we know all about what's on
1113 	 * the other side of any bridges that may be on this bus plus
1114 	 * any devices.
1115 	 *
1116 	 * Return how far we've got finding sub-buses.
1117 	 */
1118 	pr_debug("PCI: Bus scan for %04x:%02x returning with max=%02x\n",
1119 		pci_domain_nr(bus), bus->number, max);
1120 	return max;
1121 }
1122 
1123 struct pci_bus * pci_create_bus(struct device *parent,
1124 		int bus, struct pci_ops *ops, void *sysdata)
1125 {
1126 	int error;
1127 	struct pci_bus *b;
1128 	struct device *dev;
1129 
1130 	b = pci_alloc_bus();
1131 	if (!b)
1132 		return NULL;
1133 
1134 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1135 	if (!dev){
1136 		kfree(b);
1137 		return NULL;
1138 	}
1139 
1140 	b->sysdata = sysdata;
1141 	b->ops = ops;
1142 
1143 	if (pci_find_bus(pci_domain_nr(b), bus)) {
1144 		/* If we already got to this bus through a different bridge, ignore it */
1145 		pr_debug("PCI: Bus %04x:%02x already known\n", pci_domain_nr(b), bus);
1146 		goto err_out;
1147 	}
1148 
1149 	down_write(&pci_bus_sem);
1150 	list_add_tail(&b->node, &pci_root_buses);
1151 	up_write(&pci_bus_sem);
1152 
1153 	dev->parent = parent;
1154 	dev->release = pci_release_bus_bridge_dev;
1155 	dev_set_name(dev, "pci%04x:%02x", pci_domain_nr(b), bus);
1156 	error = device_register(dev);
1157 	if (error)
1158 		goto dev_reg_err;
1159 	b->bridge = get_device(dev);
1160 
1161 	if (!parent)
1162 		set_dev_node(b->bridge, pcibus_to_node(b));
1163 
1164 	b->dev.class = &pcibus_class;
1165 	b->dev.parent = b->bridge;
1166 	dev_set_name(&b->dev, "%04x:%02x", pci_domain_nr(b), bus);
1167 	error = device_register(&b->dev);
1168 	if (error)
1169 		goto class_dev_reg_err;
1170 	error = device_create_file(&b->dev, &dev_attr_cpuaffinity);
1171 	if (error)
1172 		goto dev_create_file_err;
1173 
1174 	/* Create legacy_io and legacy_mem files for this bus */
1175 	pci_create_legacy_files(b);
1176 
1177 	b->number = b->secondary = bus;
1178 	b->resource[0] = &ioport_resource;
1179 	b->resource[1] = &iomem_resource;
1180 
1181 	return b;
1182 
1183 dev_create_file_err:
1184 	device_unregister(&b->dev);
1185 class_dev_reg_err:
1186 	device_unregister(dev);
1187 dev_reg_err:
1188 	down_write(&pci_bus_sem);
1189 	list_del(&b->node);
1190 	up_write(&pci_bus_sem);
1191 err_out:
1192 	kfree(dev);
1193 	kfree(b);
1194 	return NULL;
1195 }
1196 
1197 struct pci_bus * __devinit pci_scan_bus_parented(struct device *parent,
1198 		int bus, struct pci_ops *ops, void *sysdata)
1199 {
1200 	struct pci_bus *b;
1201 
1202 	b = pci_create_bus(parent, bus, ops, sysdata);
1203 	if (b)
1204 		b->subordinate = pci_scan_child_bus(b);
1205 	return b;
1206 }
1207 EXPORT_SYMBOL(pci_scan_bus_parented);
1208 
1209 #ifdef CONFIG_HOTPLUG
1210 /**
1211  * pci_rescan_bus - scan a PCI bus for devices.
1212  * @bus: PCI bus to scan
1213  *
1214  * Scan a PCI bus and child buses for new devices, adds them,
1215  * and enables them.
1216  *
1217  * Returns the max number of subordinate bus discovered.
1218  */
1219 unsigned int __ref pci_rescan_bus(struct pci_bus *bus)
1220 {
1221 	unsigned int max;
1222 	struct pci_dev *dev;
1223 
1224 	max = pci_scan_child_bus(bus);
1225 
1226 	down_read(&pci_bus_sem);
1227 	list_for_each_entry(dev, &bus->devices, bus_list)
1228 		if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE ||
1229 		    dev->hdr_type == PCI_HEADER_TYPE_CARDBUS)
1230 			if (dev->subordinate)
1231 				pci_bus_size_bridges(dev->subordinate);
1232 	up_read(&pci_bus_sem);
1233 
1234 	pci_bus_assign_resources(bus);
1235 	pci_enable_bridges(bus);
1236 	pci_bus_add_devices(bus);
1237 
1238 	return max;
1239 }
1240 EXPORT_SYMBOL_GPL(pci_rescan_bus);
1241 
1242 EXPORT_SYMBOL(pci_add_new_bus);
1243 EXPORT_SYMBOL(pci_scan_slot);
1244 EXPORT_SYMBOL(pci_scan_bridge);
1245 EXPORT_SYMBOL_GPL(pci_scan_child_bus);
1246 #endif
1247 
1248 static int __init pci_sort_bf_cmp(const struct device *d_a, const struct device *d_b)
1249 {
1250 	const struct pci_dev *a = to_pci_dev(d_a);
1251 	const struct pci_dev *b = to_pci_dev(d_b);
1252 
1253 	if      (pci_domain_nr(a->bus) < pci_domain_nr(b->bus)) return -1;
1254 	else if (pci_domain_nr(a->bus) > pci_domain_nr(b->bus)) return  1;
1255 
1256 	if      (a->bus->number < b->bus->number) return -1;
1257 	else if (a->bus->number > b->bus->number) return  1;
1258 
1259 	if      (a->devfn < b->devfn) return -1;
1260 	else if (a->devfn > b->devfn) return  1;
1261 
1262 	return 0;
1263 }
1264 
1265 void __init pci_sort_breadthfirst(void)
1266 {
1267 	bus_sort_breadthfirst(&pci_bus_type, &pci_sort_bf_cmp);
1268 }
1269