xref: /linux/drivers/pci/probe.c (revision 643d1f7fe3aa12c8bdea6fa5b4ba874ff6dd601d)
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/aspm.h>
13 #include "pci.h"
14 
15 #define CARDBUS_LATENCY_TIMER	176	/* secondary latency timer */
16 #define CARDBUS_RESERVE_BUSNR	3
17 #define PCI_CFG_SPACE_SIZE	256
18 #define PCI_CFG_SPACE_EXP_SIZE	4096
19 
20 /* Ugh.  Need to stop exporting this to modules. */
21 LIST_HEAD(pci_root_buses);
22 EXPORT_SYMBOL(pci_root_buses);
23 
24 LIST_HEAD(pci_devices);
25 
26 /*
27  * Some device drivers need know if pci is initiated.
28  * Basically, we think pci is not initiated when there
29  * is no device in list of pci_devices.
30  */
31 int no_pci_devices(void)
32 {
33 	return list_empty(&pci_devices);
34 }
35 
36 EXPORT_SYMBOL(no_pci_devices);
37 
38 #ifdef HAVE_PCI_LEGACY
39 /**
40  * pci_create_legacy_files - create legacy I/O port and memory files
41  * @b: bus to create files under
42  *
43  * Some platforms allow access to legacy I/O port and ISA memory space on
44  * a per-bus basis.  This routine creates the files and ties them into
45  * their associated read, write and mmap files from pci-sysfs.c
46  */
47 static void pci_create_legacy_files(struct pci_bus *b)
48 {
49 	b->legacy_io = kzalloc(sizeof(struct bin_attribute) * 2,
50 			       GFP_ATOMIC);
51 	if (b->legacy_io) {
52 		b->legacy_io->attr.name = "legacy_io";
53 		b->legacy_io->size = 0xffff;
54 		b->legacy_io->attr.mode = S_IRUSR | S_IWUSR;
55 		b->legacy_io->read = pci_read_legacy_io;
56 		b->legacy_io->write = pci_write_legacy_io;
57 		device_create_bin_file(&b->dev, b->legacy_io);
58 
59 		/* Allocated above after the legacy_io struct */
60 		b->legacy_mem = b->legacy_io + 1;
61 		b->legacy_mem->attr.name = "legacy_mem";
62 		b->legacy_mem->size = 1024*1024;
63 		b->legacy_mem->attr.mode = S_IRUSR | S_IWUSR;
64 		b->legacy_mem->mmap = pci_mmap_legacy_mem;
65 		device_create_bin_file(&b->dev, b->legacy_mem);
66 	}
67 }
68 
69 void pci_remove_legacy_files(struct pci_bus *b)
70 {
71 	if (b->legacy_io) {
72 		device_remove_bin_file(&b->dev, b->legacy_io);
73 		device_remove_bin_file(&b->dev, b->legacy_mem);
74 		kfree(b->legacy_io); /* both are allocated here */
75 	}
76 }
77 #else /* !HAVE_PCI_LEGACY */
78 static inline void pci_create_legacy_files(struct pci_bus *bus) { return; }
79 void pci_remove_legacy_files(struct pci_bus *bus) { return; }
80 #endif /* HAVE_PCI_LEGACY */
81 
82 /*
83  * PCI Bus Class Devices
84  */
85 static ssize_t pci_bus_show_cpuaffinity(struct device *dev,
86 					struct device_attribute *attr,
87 					char *buf)
88 {
89 	int ret;
90 	cpumask_t cpumask;
91 
92 	cpumask = pcibus_to_cpumask(to_pci_bus(dev));
93 	ret = cpumask_scnprintf(buf, PAGE_SIZE, cpumask);
94 	if (ret < PAGE_SIZE)
95 		buf[ret++] = '\n';
96 	return ret;
97 }
98 DEVICE_ATTR(cpuaffinity, S_IRUGO, pci_bus_show_cpuaffinity, NULL);
99 
100 /*
101  * PCI Bus Class
102  */
103 static void release_pcibus_dev(struct device *dev)
104 {
105 	struct pci_bus *pci_bus = to_pci_bus(dev);
106 
107 	if (pci_bus->bridge)
108 		put_device(pci_bus->bridge);
109 	kfree(pci_bus);
110 }
111 
112 static struct class pcibus_class = {
113 	.name		= "pci_bus",
114 	.dev_release	= &release_pcibus_dev,
115 };
116 
117 static int __init pcibus_class_init(void)
118 {
119 	return class_register(&pcibus_class);
120 }
121 postcore_initcall(pcibus_class_init);
122 
123 /*
124  * Translate the low bits of the PCI base
125  * to the resource type
126  */
127 static inline unsigned int pci_calc_resource_flags(unsigned int flags)
128 {
129 	if (flags & PCI_BASE_ADDRESS_SPACE_IO)
130 		return IORESOURCE_IO;
131 
132 	if (flags & PCI_BASE_ADDRESS_MEM_PREFETCH)
133 		return IORESOURCE_MEM | IORESOURCE_PREFETCH;
134 
135 	return IORESOURCE_MEM;
136 }
137 
138 /*
139  * Find the extent of a PCI decode..
140  */
141 static u32 pci_size(u32 base, u32 maxbase, u32 mask)
142 {
143 	u32 size = mask & maxbase;	/* Find the significant bits */
144 	if (!size)
145 		return 0;
146 
147 	/* Get the lowest of them to find the decode size, and
148 	   from that the extent.  */
149 	size = (size & ~(size-1)) - 1;
150 
151 	/* base == maxbase can be valid only if the BAR has
152 	   already been programmed with all 1s.  */
153 	if (base == maxbase && ((base | size) & mask) != mask)
154 		return 0;
155 
156 	return size;
157 }
158 
159 static u64 pci_size64(u64 base, u64 maxbase, u64 mask)
160 {
161 	u64 size = mask & maxbase;	/* Find the significant bits */
162 	if (!size)
163 		return 0;
164 
165 	/* Get the lowest of them to find the decode size, and
166 	   from that the extent.  */
167 	size = (size & ~(size-1)) - 1;
168 
169 	/* base == maxbase can be valid only if the BAR has
170 	   already been programmed with all 1s.  */
171 	if (base == maxbase && ((base | size) & mask) != mask)
172 		return 0;
173 
174 	return size;
175 }
176 
177 static inline int is_64bit_memory(u32 mask)
178 {
179 	if ((mask & (PCI_BASE_ADDRESS_SPACE|PCI_BASE_ADDRESS_MEM_TYPE_MASK)) ==
180 	    (PCI_BASE_ADDRESS_SPACE_MEMORY|PCI_BASE_ADDRESS_MEM_TYPE_64))
181 		return 1;
182 	return 0;
183 }
184 
185 static void pci_read_bases(struct pci_dev *dev, unsigned int howmany, int rom)
186 {
187 	unsigned int pos, reg, next;
188 	u32 l, sz;
189 	struct resource *res;
190 
191 	for(pos=0; pos<howmany; pos = next) {
192 		u64 l64;
193 		u64 sz64;
194 		u32 raw_sz;
195 
196 		next = pos+1;
197 		res = &dev->resource[pos];
198 		res->name = pci_name(dev);
199 		reg = PCI_BASE_ADDRESS_0 + (pos << 2);
200 		pci_read_config_dword(dev, reg, &l);
201 		pci_write_config_dword(dev, reg, ~0);
202 		pci_read_config_dword(dev, reg, &sz);
203 		pci_write_config_dword(dev, reg, l);
204 		if (!sz || sz == 0xffffffff)
205 			continue;
206 		if (l == 0xffffffff)
207 			l = 0;
208 		raw_sz = sz;
209 		if ((l & PCI_BASE_ADDRESS_SPACE) ==
210 				PCI_BASE_ADDRESS_SPACE_MEMORY) {
211 			sz = pci_size(l, sz, (u32)PCI_BASE_ADDRESS_MEM_MASK);
212 			/*
213 			 * For 64bit prefetchable memory sz could be 0, if the
214 			 * real size is bigger than 4G, so we need to check
215 			 * szhi for that.
216 			 */
217 			if (!is_64bit_memory(l) && !sz)
218 				continue;
219 			res->start = l & PCI_BASE_ADDRESS_MEM_MASK;
220 			res->flags |= l & ~PCI_BASE_ADDRESS_MEM_MASK;
221 		} else {
222 			sz = pci_size(l, sz, PCI_BASE_ADDRESS_IO_MASK & 0xffff);
223 			if (!sz)
224 				continue;
225 			res->start = l & PCI_BASE_ADDRESS_IO_MASK;
226 			res->flags |= l & ~PCI_BASE_ADDRESS_IO_MASK;
227 		}
228 		res->end = res->start + (unsigned long) sz;
229 		res->flags |= pci_calc_resource_flags(l);
230 		if (is_64bit_memory(l)) {
231 			u32 szhi, lhi;
232 
233 			pci_read_config_dword(dev, reg+4, &lhi);
234 			pci_write_config_dword(dev, reg+4, ~0);
235 			pci_read_config_dword(dev, reg+4, &szhi);
236 			pci_write_config_dword(dev, reg+4, lhi);
237 			sz64 = ((u64)szhi << 32) | raw_sz;
238 			l64 = ((u64)lhi << 32) | l;
239 			sz64 = pci_size64(l64, sz64, PCI_BASE_ADDRESS_MEM_MASK);
240 			next++;
241 #if BITS_PER_LONG == 64
242 			if (!sz64) {
243 				res->start = 0;
244 				res->end = 0;
245 				res->flags = 0;
246 				continue;
247 			}
248 			res->start = l64 & PCI_BASE_ADDRESS_MEM_MASK;
249 			res->end = res->start + sz64;
250 #else
251 			if (sz64 > 0x100000000ULL) {
252 				printk(KERN_ERR "PCI: Unable to handle 64-bit "
253 					"BAR for device %s\n", pci_name(dev));
254 				res->start = 0;
255 				res->flags = 0;
256 			} else if (lhi) {
257 				/* 64-bit wide address, treat as disabled */
258 				pci_write_config_dword(dev, reg,
259 					l & ~(u32)PCI_BASE_ADDRESS_MEM_MASK);
260 				pci_write_config_dword(dev, reg+4, 0);
261 				res->start = 0;
262 				res->end = sz;
263 			}
264 #endif
265 		}
266 	}
267 	if (rom) {
268 		dev->rom_base_reg = rom;
269 		res = &dev->resource[PCI_ROM_RESOURCE];
270 		res->name = pci_name(dev);
271 		pci_read_config_dword(dev, rom, &l);
272 		pci_write_config_dword(dev, rom, ~PCI_ROM_ADDRESS_ENABLE);
273 		pci_read_config_dword(dev, rom, &sz);
274 		pci_write_config_dword(dev, rom, l);
275 		if (l == 0xffffffff)
276 			l = 0;
277 		if (sz && sz != 0xffffffff) {
278 			sz = pci_size(l, sz, (u32)PCI_ROM_ADDRESS_MASK);
279 			if (sz) {
280 				res->flags = (l & IORESOURCE_ROM_ENABLE) |
281 				  IORESOURCE_MEM | IORESOURCE_PREFETCH |
282 				  IORESOURCE_READONLY | IORESOURCE_CACHEABLE;
283 				res->start = l & PCI_ROM_ADDRESS_MASK;
284 				res->end = res->start + (unsigned long) sz;
285 			}
286 		}
287 	}
288 }
289 
290 void pci_read_bridge_bases(struct pci_bus *child)
291 {
292 	struct pci_dev *dev = child->self;
293 	u8 io_base_lo, io_limit_lo;
294 	u16 mem_base_lo, mem_limit_lo;
295 	unsigned long base, limit;
296 	struct resource *res;
297 	int i;
298 
299 	if (!dev)		/* It's a host bus, nothing to read */
300 		return;
301 
302 	if (dev->transparent) {
303 		printk(KERN_INFO "PCI: Transparent bridge - %s\n", pci_name(dev));
304 		for(i = 3; i < PCI_BUS_NUM_RESOURCES; i++)
305 			child->resource[i] = child->parent->resource[i - 3];
306 	}
307 
308 	for(i=0; i<3; i++)
309 		child->resource[i] = &dev->resource[PCI_BRIDGE_RESOURCES+i];
310 
311 	res = child->resource[0];
312 	pci_read_config_byte(dev, PCI_IO_BASE, &io_base_lo);
313 	pci_read_config_byte(dev, PCI_IO_LIMIT, &io_limit_lo);
314 	base = (io_base_lo & PCI_IO_RANGE_MASK) << 8;
315 	limit = (io_limit_lo & PCI_IO_RANGE_MASK) << 8;
316 
317 	if ((io_base_lo & PCI_IO_RANGE_TYPE_MASK) == PCI_IO_RANGE_TYPE_32) {
318 		u16 io_base_hi, io_limit_hi;
319 		pci_read_config_word(dev, PCI_IO_BASE_UPPER16, &io_base_hi);
320 		pci_read_config_word(dev, PCI_IO_LIMIT_UPPER16, &io_limit_hi);
321 		base |= (io_base_hi << 16);
322 		limit |= (io_limit_hi << 16);
323 	}
324 
325 	if (base <= limit) {
326 		res->flags = (io_base_lo & PCI_IO_RANGE_TYPE_MASK) | IORESOURCE_IO;
327 		if (!res->start)
328 			res->start = base;
329 		if (!res->end)
330 			res->end = limit + 0xfff;
331 	}
332 
333 	res = child->resource[1];
334 	pci_read_config_word(dev, PCI_MEMORY_BASE, &mem_base_lo);
335 	pci_read_config_word(dev, PCI_MEMORY_LIMIT, &mem_limit_lo);
336 	base = (mem_base_lo & PCI_MEMORY_RANGE_MASK) << 16;
337 	limit = (mem_limit_lo & PCI_MEMORY_RANGE_MASK) << 16;
338 	if (base <= limit) {
339 		res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM;
340 		res->start = base;
341 		res->end = limit + 0xfffff;
342 	}
343 
344 	res = child->resource[2];
345 	pci_read_config_word(dev, PCI_PREF_MEMORY_BASE, &mem_base_lo);
346 	pci_read_config_word(dev, PCI_PREF_MEMORY_LIMIT, &mem_limit_lo);
347 	base = (mem_base_lo & PCI_PREF_RANGE_MASK) << 16;
348 	limit = (mem_limit_lo & PCI_PREF_RANGE_MASK) << 16;
349 
350 	if ((mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) {
351 		u32 mem_base_hi, mem_limit_hi;
352 		pci_read_config_dword(dev, PCI_PREF_BASE_UPPER32, &mem_base_hi);
353 		pci_read_config_dword(dev, PCI_PREF_LIMIT_UPPER32, &mem_limit_hi);
354 
355 		/*
356 		 * Some bridges set the base > limit by default, and some
357 		 * (broken) BIOSes do not initialize them.  If we find
358 		 * this, just assume they are not being used.
359 		 */
360 		if (mem_base_hi <= mem_limit_hi) {
361 #if BITS_PER_LONG == 64
362 			base |= ((long) mem_base_hi) << 32;
363 			limit |= ((long) mem_limit_hi) << 32;
364 #else
365 			if (mem_base_hi || mem_limit_hi) {
366 				printk(KERN_ERR "PCI: Unable to handle 64-bit address space for bridge %s\n", pci_name(dev));
367 				return;
368 			}
369 #endif
370 		}
371 	}
372 	if (base <= limit) {
373 		res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM | IORESOURCE_PREFETCH;
374 		res->start = base;
375 		res->end = limit + 0xfffff;
376 	}
377 }
378 
379 static struct pci_bus * pci_alloc_bus(void)
380 {
381 	struct pci_bus *b;
382 
383 	b = kzalloc(sizeof(*b), GFP_KERNEL);
384 	if (b) {
385 		INIT_LIST_HEAD(&b->node);
386 		INIT_LIST_HEAD(&b->children);
387 		INIT_LIST_HEAD(&b->devices);
388 	}
389 	return b;
390 }
391 
392 static struct pci_bus * __devinit
393 pci_alloc_child_bus(struct pci_bus *parent, struct pci_dev *bridge, int busnr)
394 {
395 	struct pci_bus *child;
396 	int i;
397 
398 	/*
399 	 * Allocate a new bus, and inherit stuff from the parent..
400 	 */
401 	child = pci_alloc_bus();
402 	if (!child)
403 		return NULL;
404 
405 	child->self = bridge;
406 	child->parent = parent;
407 	child->ops = parent->ops;
408 	child->sysdata = parent->sysdata;
409 	child->bus_flags = parent->bus_flags;
410 	child->bridge = get_device(&bridge->dev);
411 
412 	/* initialize some portions of the bus device, but don't register it
413 	 * now as the parent is not properly set up yet.  This device will get
414 	 * registered later in pci_bus_add_devices()
415 	 */
416 	child->dev.class = &pcibus_class;
417 	sprintf(child->dev.bus_id, "%04x:%02x", pci_domain_nr(child), busnr);
418 
419 	/*
420 	 * Set up the primary, secondary and subordinate
421 	 * bus numbers.
422 	 */
423 	child->number = child->secondary = busnr;
424 	child->primary = parent->secondary;
425 	child->subordinate = 0xff;
426 
427 	/* Set up default resource pointers and names.. */
428 	for (i = 0; i < 4; 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 *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 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 
483 	pci_read_config_dword(dev, PCI_PRIMARY_BUS, &buses);
484 
485 	pr_debug("PCI: Scanning behind PCI bridge %s, config %06x, pass %d\n",
486 		 pci_name(dev), buses & 0xffffff, pass);
487 
488 	/* Disable MasterAbortMode during probing to avoid reporting
489 	   of bus errors (in some architectures) */
490 	pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl);
491 	pci_write_config_word(dev, PCI_BRIDGE_CONTROL,
492 			      bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT);
493 
494 	if ((buses & 0xffff00) && !pcibios_assign_all_busses() && !is_cardbus) {
495 		unsigned int cmax, busnr;
496 		/*
497 		 * Bus already configured by firmware, process it in the first
498 		 * pass and just note the configuration.
499 		 */
500 		if (pass)
501 			goto out;
502 		busnr = (buses >> 8) & 0xFF;
503 
504 		/*
505 		 * If we already got to this bus through a different bridge,
506 		 * ignore it.  This can happen with the i450NX chipset.
507 		 */
508 		if (pci_find_bus(pci_domain_nr(bus), busnr)) {
509 			printk(KERN_INFO "PCI: Bus %04x:%02x already known\n",
510 					pci_domain_nr(bus), busnr);
511 			goto out;
512 		}
513 
514 		child = pci_add_new_bus(bus, dev, busnr);
515 		if (!child)
516 			goto out;
517 		child->primary = buses & 0xFF;
518 		child->subordinate = (buses >> 16) & 0xFF;
519 		child->bridge_ctl = bctl;
520 
521 		cmax = pci_scan_child_bus(child);
522 		if (cmax > max)
523 			max = cmax;
524 		if (child->subordinate > max)
525 			max = child->subordinate;
526 	} else {
527 		/*
528 		 * We need to assign a number to this bus which we always
529 		 * do in the second pass.
530 		 */
531 		if (!pass) {
532 			if (pcibios_assign_all_busses())
533 				/* Temporarily disable forwarding of the
534 				   configuration cycles on all bridges in
535 				   this bus segment to avoid possible
536 				   conflicts in the second pass between two
537 				   bridges programmed with overlapping
538 				   bus ranges. */
539 				pci_write_config_dword(dev, PCI_PRIMARY_BUS,
540 						       buses & ~0xffffff);
541 			goto out;
542 		}
543 
544 		/* Clear errors */
545 		pci_write_config_word(dev, PCI_STATUS, 0xffff);
546 
547 		/* Prevent assigning a bus number that already exists.
548 		 * This can happen when a bridge is hot-plugged */
549 		if (pci_find_bus(pci_domain_nr(bus), max+1))
550 			goto out;
551 		child = pci_add_new_bus(bus, dev, ++max);
552 		buses = (buses & 0xff000000)
553 		      | ((unsigned int)(child->primary)     <<  0)
554 		      | ((unsigned int)(child->secondary)   <<  8)
555 		      | ((unsigned int)(child->subordinate) << 16);
556 
557 		/*
558 		 * yenta.c forces a secondary latency timer of 176.
559 		 * Copy that behaviour here.
560 		 */
561 		if (is_cardbus) {
562 			buses &= ~0xff000000;
563 			buses |= CARDBUS_LATENCY_TIMER << 24;
564 		}
565 
566 		/*
567 		 * We need to blast all three values with a single write.
568 		 */
569 		pci_write_config_dword(dev, PCI_PRIMARY_BUS, buses);
570 
571 		if (!is_cardbus) {
572 			child->bridge_ctl = bctl;
573 			/*
574 			 * Adjust subordinate busnr in parent buses.
575 			 * We do this before scanning for children because
576 			 * some devices may not be detected if the bios
577 			 * was lazy.
578 			 */
579 			pci_fixup_parent_subordinate_busnr(child, max);
580 			/* Now we can scan all subordinate buses... */
581 			max = pci_scan_child_bus(child);
582 			/*
583 			 * now fix it up again since we have found
584 			 * the real value of max.
585 			 */
586 			pci_fixup_parent_subordinate_busnr(child, max);
587 		} else {
588 			/*
589 			 * For CardBus bridges, we leave 4 bus numbers
590 			 * as cards with a PCI-to-PCI bridge can be
591 			 * inserted later.
592 			 */
593 			for (i=0; i<CARDBUS_RESERVE_BUSNR; i++) {
594 				struct pci_bus *parent = bus;
595 				if (pci_find_bus(pci_domain_nr(bus),
596 							max+i+1))
597 					break;
598 				while (parent->parent) {
599 					if ((!pcibios_assign_all_busses()) &&
600 					    (parent->subordinate > max) &&
601 					    (parent->subordinate <= max+i)) {
602 						j = 1;
603 					}
604 					parent = parent->parent;
605 				}
606 				if (j) {
607 					/*
608 					 * Often, there are two cardbus bridges
609 					 * -- try to leave one valid bus number
610 					 * for each one.
611 					 */
612 					i /= 2;
613 					break;
614 				}
615 			}
616 			max += i;
617 			pci_fixup_parent_subordinate_busnr(child, max);
618 		}
619 		/*
620 		 * Set the subordinate bus number to its real value.
621 		 */
622 		child->subordinate = max;
623 		pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, max);
624 	}
625 
626 	sprintf(child->name, (is_cardbus ? "PCI CardBus #%02x" : "PCI Bus #%02x"), child->number);
627 
628 	/* Has only triggered on CardBus, fixup is in yenta_socket */
629 	while (bus->parent) {
630 		if ((child->subordinate > bus->subordinate) ||
631 		    (child->number > bus->subordinate) ||
632 		    (child->number < bus->number) ||
633 		    (child->subordinate < bus->number)) {
634 			pr_debug("PCI: Bus #%02x (-#%02x) is %s "
635 				"hidden behind%s bridge #%02x (-#%02x)\n",
636 				child->number, child->subordinate,
637 				(bus->number > child->subordinate &&
638 				 bus->subordinate < child->number) ?
639 					"wholly" : "partially",
640 				bus->self->transparent ? " transparent" : "",
641 				bus->number, bus->subordinate);
642 		}
643 		bus = bus->parent;
644 	}
645 
646 out:
647 	pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl);
648 
649 	return max;
650 }
651 
652 /*
653  * Read interrupt line and base address registers.
654  * The architecture-dependent code can tweak these, of course.
655  */
656 static void pci_read_irq(struct pci_dev *dev)
657 {
658 	unsigned char irq;
659 
660 	pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &irq);
661 	dev->pin = irq;
662 	if (irq)
663 		pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq);
664 	dev->irq = irq;
665 }
666 
667 #define LEGACY_IO_RESOURCE	(IORESOURCE_IO | IORESOURCE_PCI_FIXED)
668 
669 /**
670  * pci_setup_device - fill in class and map information of a device
671  * @dev: the device structure to fill
672  *
673  * Initialize the device structure with information about the device's
674  * vendor,class,memory and IO-space addresses,IRQ lines etc.
675  * Called at initialisation of the PCI subsystem and by CardBus services.
676  * Returns 0 on success and -1 if unknown type of device (not normal, bridge
677  * or CardBus).
678  */
679 static int pci_setup_device(struct pci_dev * dev)
680 {
681 	u32 class;
682 
683 	sprintf(pci_name(dev), "%04x:%02x:%02x.%d", pci_domain_nr(dev->bus),
684 		dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn));
685 
686 	pci_read_config_dword(dev, PCI_CLASS_REVISION, &class);
687 	dev->revision = class & 0xff;
688 	class >>= 8;				    /* upper 3 bytes */
689 	dev->class = class;
690 	class >>= 8;
691 
692 	pr_debug("PCI: Found %s [%04x/%04x] %06x %02x\n", pci_name(dev),
693 		 dev->vendor, dev->device, class, dev->hdr_type);
694 
695 	/* "Unknown power state" */
696 	dev->current_state = PCI_UNKNOWN;
697 
698 	/* Early fixups, before probing the BARs */
699 	pci_fixup_device(pci_fixup_early, dev);
700 	class = dev->class >> 8;
701 
702 	switch (dev->hdr_type) {		    /* header type */
703 	case PCI_HEADER_TYPE_NORMAL:		    /* standard header */
704 		if (class == PCI_CLASS_BRIDGE_PCI)
705 			goto bad;
706 		pci_read_irq(dev);
707 		pci_read_bases(dev, 6, PCI_ROM_ADDRESS);
708 		pci_read_config_word(dev, PCI_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
709 		pci_read_config_word(dev, PCI_SUBSYSTEM_ID, &dev->subsystem_device);
710 
711 		/*
712 		 *	Do the ugly legacy mode stuff here rather than broken chip
713 		 *	quirk code. Legacy mode ATA controllers have fixed
714 		 *	addresses. These are not always echoed in BAR0-3, and
715 		 *	BAR0-3 in a few cases contain junk!
716 		 */
717 		if (class == PCI_CLASS_STORAGE_IDE) {
718 			u8 progif;
719 			pci_read_config_byte(dev, PCI_CLASS_PROG, &progif);
720 			if ((progif & 1) == 0) {
721 				dev->resource[0].start = 0x1F0;
722 				dev->resource[0].end = 0x1F7;
723 				dev->resource[0].flags = LEGACY_IO_RESOURCE;
724 				dev->resource[1].start = 0x3F6;
725 				dev->resource[1].end = 0x3F6;
726 				dev->resource[1].flags = LEGACY_IO_RESOURCE;
727 			}
728 			if ((progif & 4) == 0) {
729 				dev->resource[2].start = 0x170;
730 				dev->resource[2].end = 0x177;
731 				dev->resource[2].flags = LEGACY_IO_RESOURCE;
732 				dev->resource[3].start = 0x376;
733 				dev->resource[3].end = 0x376;
734 				dev->resource[3].flags = LEGACY_IO_RESOURCE;
735 			}
736 		}
737 		break;
738 
739 	case PCI_HEADER_TYPE_BRIDGE:		    /* bridge header */
740 		if (class != PCI_CLASS_BRIDGE_PCI)
741 			goto bad;
742 		/* The PCI-to-PCI bridge spec requires that subtractive
743 		   decoding (i.e. transparent) bridge must have programming
744 		   interface code of 0x01. */
745 		pci_read_irq(dev);
746 		dev->transparent = ((dev->class & 0xff) == 1);
747 		pci_read_bases(dev, 2, PCI_ROM_ADDRESS1);
748 		break;
749 
750 	case PCI_HEADER_TYPE_CARDBUS:		    /* CardBus bridge header */
751 		if (class != PCI_CLASS_BRIDGE_CARDBUS)
752 			goto bad;
753 		pci_read_irq(dev);
754 		pci_read_bases(dev, 1, 0);
755 		pci_read_config_word(dev, PCI_CB_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
756 		pci_read_config_word(dev, PCI_CB_SUBSYSTEM_ID, &dev->subsystem_device);
757 		break;
758 
759 	default:				    /* unknown header */
760 		printk(KERN_ERR "PCI: device %s has unknown header type %02x, ignoring.\n",
761 			pci_name(dev), dev->hdr_type);
762 		return -1;
763 
764 	bad:
765 		printk(KERN_ERR "PCI: %s: class %x doesn't match header type %02x. Ignoring class.\n",
766 		       pci_name(dev), class, dev->hdr_type);
767 		dev->class = PCI_CLASS_NOT_DEFINED;
768 	}
769 
770 	/* We found a fine healthy device, go go go... */
771 	return 0;
772 }
773 
774 /**
775  * pci_release_dev - free a pci device structure when all users of it are finished.
776  * @dev: device that's been disconnected
777  *
778  * Will be called only by the device core when all users of this pci device are
779  * done.
780  */
781 static void pci_release_dev(struct device *dev)
782 {
783 	struct pci_dev *pci_dev;
784 
785 	pci_dev = to_pci_dev(dev);
786 	kfree(pci_dev);
787 }
788 
789 static void set_pcie_port_type(struct pci_dev *pdev)
790 {
791 	int pos;
792 	u16 reg16;
793 
794 	pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
795 	if (!pos)
796 		return;
797 	pdev->is_pcie = 1;
798 	pci_read_config_word(pdev, pos + PCI_EXP_FLAGS, &reg16);
799 	pdev->pcie_type = (reg16 & PCI_EXP_FLAGS_TYPE) >> 4;
800 }
801 
802 /**
803  * pci_cfg_space_size - get the configuration space size of the PCI device.
804  * @dev: PCI device
805  *
806  * Regular PCI devices have 256 bytes, but PCI-X 2 and PCI Express devices
807  * have 4096 bytes.  Even if the device is capable, that doesn't mean we can
808  * access it.  Maybe we don't have a way to generate extended config space
809  * accesses, or the device is behind a reverse Express bridge.  So we try
810  * reading the dword at 0x100 which must either be 0 or a valid extended
811  * capability header.
812  */
813 int pci_cfg_space_size(struct pci_dev *dev)
814 {
815 	int pos;
816 	u32 status;
817 
818 	pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
819 	if (!pos) {
820 		pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
821 		if (!pos)
822 			goto fail;
823 
824 		pci_read_config_dword(dev, pos + PCI_X_STATUS, &status);
825 		if (!(status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ)))
826 			goto fail;
827 	}
828 
829 	if (pci_read_config_dword(dev, 256, &status) != PCIBIOS_SUCCESSFUL)
830 		goto fail;
831 	if (status == 0xffffffff)
832 		goto fail;
833 
834 	return PCI_CFG_SPACE_EXP_SIZE;
835 
836  fail:
837 	return PCI_CFG_SPACE_SIZE;
838 }
839 
840 static void pci_release_bus_bridge_dev(struct device *dev)
841 {
842 	kfree(dev);
843 }
844 
845 struct pci_dev *alloc_pci_dev(void)
846 {
847 	struct pci_dev *dev;
848 
849 	dev = kzalloc(sizeof(struct pci_dev), GFP_KERNEL);
850 	if (!dev)
851 		return NULL;
852 
853 	INIT_LIST_HEAD(&dev->global_list);
854 	INIT_LIST_HEAD(&dev->bus_list);
855 
856 	pci_msi_init_pci_dev(dev);
857 
858 	return dev;
859 }
860 EXPORT_SYMBOL(alloc_pci_dev);
861 
862 /*
863  * Read the config data for a PCI device, sanity-check it
864  * and fill in the dev structure...
865  */
866 static struct pci_dev * __devinit
867 pci_scan_device(struct pci_bus *bus, int devfn)
868 {
869 	struct pci_dev *dev;
870 	u32 l;
871 	u8 hdr_type;
872 	int delay = 1;
873 
874 	if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, &l))
875 		return NULL;
876 
877 	/* some broken boards return 0 or ~0 if a slot is empty: */
878 	if (l == 0xffffffff || l == 0x00000000 ||
879 	    l == 0x0000ffff || l == 0xffff0000)
880 		return NULL;
881 
882 	/* Configuration request Retry Status */
883 	while (l == 0xffff0001) {
884 		msleep(delay);
885 		delay *= 2;
886 		if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, &l))
887 			return NULL;
888 		/* Card hasn't responded in 60 seconds?  Must be stuck. */
889 		if (delay > 60 * 1000) {
890 			printk(KERN_WARNING "Device %04x:%02x:%02x.%d not "
891 					"responding\n", pci_domain_nr(bus),
892 					bus->number, PCI_SLOT(devfn),
893 					PCI_FUNC(devfn));
894 			return NULL;
895 		}
896 	}
897 
898 	if (pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type))
899 		return NULL;
900 
901 	dev = alloc_pci_dev();
902 	if (!dev)
903 		return NULL;
904 
905 	dev->bus = bus;
906 	dev->sysdata = bus->sysdata;
907 	dev->dev.parent = bus->bridge;
908 	dev->dev.bus = &pci_bus_type;
909 	dev->devfn = devfn;
910 	dev->hdr_type = hdr_type & 0x7f;
911 	dev->multifunction = !!(hdr_type & 0x80);
912 	dev->vendor = l & 0xffff;
913 	dev->device = (l >> 16) & 0xffff;
914 	dev->cfg_size = pci_cfg_space_size(dev);
915 	dev->error_state = pci_channel_io_normal;
916 	set_pcie_port_type(dev);
917 
918 	/* Assume 32-bit PCI; let 64-bit PCI cards (which are far rarer)
919 	   set this higher, assuming the system even supports it.  */
920 	dev->dma_mask = 0xffffffff;
921 	if (pci_setup_device(dev) < 0) {
922 		kfree(dev);
923 		return NULL;
924 	}
925 
926 	return dev;
927 }
928 
929 void pci_device_add(struct pci_dev *dev, struct pci_bus *bus)
930 {
931 	device_initialize(&dev->dev);
932 	dev->dev.release = pci_release_dev;
933 	pci_dev_get(dev);
934 
935 	set_dev_node(&dev->dev, pcibus_to_node(bus));
936 	dev->dev.dma_mask = &dev->dma_mask;
937 	dev->dev.coherent_dma_mask = 0xffffffffull;
938 
939 	/* Fix up broken headers */
940 	pci_fixup_device(pci_fixup_header, dev);
941 
942 	/*
943 	 * Add the device to our list of discovered devices
944 	 * and the bus list for fixup functions, etc.
945 	 */
946 	INIT_LIST_HEAD(&dev->global_list);
947 	down_write(&pci_bus_sem);
948 	list_add_tail(&dev->bus_list, &bus->devices);
949 	up_write(&pci_bus_sem);
950 }
951 
952 struct pci_dev *pci_scan_single_device(struct pci_bus *bus, int devfn)
953 {
954 	struct pci_dev *dev;
955 
956 	dev = pci_scan_device(bus, devfn);
957 	if (!dev)
958 		return NULL;
959 
960 	pci_device_add(dev, bus);
961 
962 	return dev;
963 }
964 EXPORT_SYMBOL(pci_scan_single_device);
965 
966 /**
967  * pci_scan_slot - scan a PCI slot on a bus for devices.
968  * @bus: PCI bus to scan
969  * @devfn: slot number to scan (must have zero function.)
970  *
971  * Scan a PCI slot on the specified PCI bus for devices, adding
972  * discovered devices to the @bus->devices list.  New devices
973  * will have an empty dev->global_list head.
974  */
975 int pci_scan_slot(struct pci_bus *bus, int devfn)
976 {
977 	int func, nr = 0;
978 	int scan_all_fns;
979 
980 	scan_all_fns = pcibios_scan_all_fns(bus, devfn);
981 
982 	for (func = 0; func < 8; func++, devfn++) {
983 		struct pci_dev *dev;
984 
985 		dev = pci_scan_single_device(bus, devfn);
986 		if (dev) {
987 			nr++;
988 
989 			/*
990 		 	 * If this is a single function device,
991 		 	 * don't scan past the first function.
992 		 	 */
993 			if (!dev->multifunction) {
994 				if (func > 0) {
995 					dev->multifunction = 1;
996 				} else {
997  					break;
998 				}
999 			}
1000 		} else {
1001 			if (func == 0 && !scan_all_fns)
1002 				break;
1003 		}
1004 	}
1005 
1006 	if (bus->self)
1007 		pcie_aspm_init_link_state(bus->self);
1008 
1009 	return nr;
1010 }
1011 
1012 unsigned int pci_scan_child_bus(struct pci_bus *bus)
1013 {
1014 	unsigned int devfn, pass, max = bus->secondary;
1015 	struct pci_dev *dev;
1016 
1017 	pr_debug("PCI: Scanning bus %04x:%02x\n", pci_domain_nr(bus), bus->number);
1018 
1019 	/* Go find them, Rover! */
1020 	for (devfn = 0; devfn < 0x100; devfn += 8)
1021 		pci_scan_slot(bus, devfn);
1022 
1023 	/*
1024 	 * After performing arch-dependent fixup of the bus, look behind
1025 	 * all PCI-to-PCI bridges on this bus.
1026 	 */
1027 	pr_debug("PCI: Fixups for bus %04x:%02x\n", pci_domain_nr(bus), bus->number);
1028 	pcibios_fixup_bus(bus);
1029 	for (pass=0; pass < 2; pass++)
1030 		list_for_each_entry(dev, &bus->devices, bus_list) {
1031 			if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE ||
1032 			    dev->hdr_type == PCI_HEADER_TYPE_CARDBUS)
1033 				max = pci_scan_bridge(bus, dev, max, pass);
1034 		}
1035 
1036 	/*
1037 	 * We've scanned the bus and so we know all about what's on
1038 	 * the other side of any bridges that may be on this bus plus
1039 	 * any devices.
1040 	 *
1041 	 * Return how far we've got finding sub-buses.
1042 	 */
1043 	pr_debug("PCI: Bus scan for %04x:%02x returning with max=%02x\n",
1044 		pci_domain_nr(bus), bus->number, max);
1045 	return max;
1046 }
1047 
1048 unsigned int __devinit pci_do_scan_bus(struct pci_bus *bus)
1049 {
1050 	unsigned int max;
1051 
1052 	max = pci_scan_child_bus(bus);
1053 
1054 	/*
1055 	 * Make the discovered devices available.
1056 	 */
1057 	pci_bus_add_devices(bus);
1058 
1059 	return max;
1060 }
1061 
1062 struct pci_bus * pci_create_bus(struct device *parent,
1063 		int bus, struct pci_ops *ops, void *sysdata)
1064 {
1065 	int error;
1066 	struct pci_bus *b;
1067 	struct device *dev;
1068 
1069 	b = pci_alloc_bus();
1070 	if (!b)
1071 		return NULL;
1072 
1073 	dev = kmalloc(sizeof(*dev), GFP_KERNEL);
1074 	if (!dev){
1075 		kfree(b);
1076 		return NULL;
1077 	}
1078 
1079 	b->sysdata = sysdata;
1080 	b->ops = ops;
1081 
1082 	if (pci_find_bus(pci_domain_nr(b), bus)) {
1083 		/* If we already got to this bus through a different bridge, ignore it */
1084 		pr_debug("PCI: Bus %04x:%02x already known\n", pci_domain_nr(b), bus);
1085 		goto err_out;
1086 	}
1087 
1088 	down_write(&pci_bus_sem);
1089 	list_add_tail(&b->node, &pci_root_buses);
1090 	up_write(&pci_bus_sem);
1091 
1092 	memset(dev, 0, sizeof(*dev));
1093 	dev->parent = parent;
1094 	dev->release = pci_release_bus_bridge_dev;
1095 	sprintf(dev->bus_id, "pci%04x:%02x", pci_domain_nr(b), bus);
1096 	error = device_register(dev);
1097 	if (error)
1098 		goto dev_reg_err;
1099 	b->bridge = get_device(dev);
1100 
1101 	b->dev.class = &pcibus_class;
1102 	b->dev.parent = b->bridge;
1103 	sprintf(b->dev.bus_id, "%04x:%02x", pci_domain_nr(b), bus);
1104 	error = device_register(&b->dev);
1105 	if (error)
1106 		goto class_dev_reg_err;
1107 	error = device_create_file(&b->dev, &dev_attr_cpuaffinity);
1108 	if (error)
1109 		goto dev_create_file_err;
1110 
1111 	/* Create legacy_io and legacy_mem files for this bus */
1112 	pci_create_legacy_files(b);
1113 
1114 	b->number = b->secondary = bus;
1115 	b->resource[0] = &ioport_resource;
1116 	b->resource[1] = &iomem_resource;
1117 
1118 	return b;
1119 
1120 dev_create_file_err:
1121 	device_unregister(&b->dev);
1122 class_dev_reg_err:
1123 	device_unregister(dev);
1124 dev_reg_err:
1125 	down_write(&pci_bus_sem);
1126 	list_del(&b->node);
1127 	up_write(&pci_bus_sem);
1128 err_out:
1129 	kfree(dev);
1130 	kfree(b);
1131 	return NULL;
1132 }
1133 
1134 struct pci_bus *pci_scan_bus_parented(struct device *parent,
1135 		int bus, struct pci_ops *ops, void *sysdata)
1136 {
1137 	struct pci_bus *b;
1138 
1139 	b = pci_create_bus(parent, bus, ops, sysdata);
1140 	if (b)
1141 		b->subordinate = pci_scan_child_bus(b);
1142 	return b;
1143 }
1144 EXPORT_SYMBOL(pci_scan_bus_parented);
1145 
1146 #ifdef CONFIG_HOTPLUG
1147 EXPORT_SYMBOL(pci_add_new_bus);
1148 EXPORT_SYMBOL(pci_do_scan_bus);
1149 EXPORT_SYMBOL(pci_scan_slot);
1150 EXPORT_SYMBOL(pci_scan_bridge);
1151 EXPORT_SYMBOL_GPL(pci_scan_child_bus);
1152 #endif
1153 
1154 static int __init pci_sort_bf_cmp(const struct pci_dev *a, const struct pci_dev *b)
1155 {
1156 	if      (pci_domain_nr(a->bus) < pci_domain_nr(b->bus)) return -1;
1157 	else if (pci_domain_nr(a->bus) > pci_domain_nr(b->bus)) return  1;
1158 
1159 	if      (a->bus->number < b->bus->number) return -1;
1160 	else if (a->bus->number > b->bus->number) return  1;
1161 
1162 	if      (a->devfn < b->devfn) return -1;
1163 	else if (a->devfn > b->devfn) return  1;
1164 
1165 	return 0;
1166 }
1167 
1168 /*
1169  * Yes, this forcably breaks the klist abstraction temporarily.  It
1170  * just wants to sort the klist, not change reference counts and
1171  * take/drop locks rapidly in the process.  It does all this while
1172  * holding the lock for the list, so objects can't otherwise be
1173  * added/removed while we're swizzling.
1174  */
1175 static void __init pci_insertion_sort_klist(struct pci_dev *a, struct list_head *list)
1176 {
1177 	struct list_head *pos;
1178 	struct klist_node *n;
1179 	struct device *dev;
1180 	struct pci_dev *b;
1181 
1182 	list_for_each(pos, list) {
1183 		n = container_of(pos, struct klist_node, n_node);
1184 		dev = container_of(n, struct device, knode_bus);
1185 		b = to_pci_dev(dev);
1186 		if (pci_sort_bf_cmp(a, b) <= 0) {
1187 			list_move_tail(&a->dev.knode_bus.n_node, &b->dev.knode_bus.n_node);
1188 			return;
1189 		}
1190 	}
1191 	list_move_tail(&a->dev.knode_bus.n_node, list);
1192 }
1193 
1194 static void __init pci_sort_breadthfirst_klist(void)
1195 {
1196 	LIST_HEAD(sorted_devices);
1197 	struct list_head *pos, *tmp;
1198 	struct klist_node *n;
1199 	struct device *dev;
1200 	struct pci_dev *pdev;
1201 	struct klist *device_klist;
1202 
1203 	device_klist = bus_get_device_klist(&pci_bus_type);
1204 
1205 	spin_lock(&device_klist->k_lock);
1206 	list_for_each_safe(pos, tmp, &device_klist->k_list) {
1207 		n = container_of(pos, struct klist_node, n_node);
1208 		dev = container_of(n, struct device, knode_bus);
1209 		pdev = to_pci_dev(dev);
1210 		pci_insertion_sort_klist(pdev, &sorted_devices);
1211 	}
1212 	list_splice(&sorted_devices, &device_klist->k_list);
1213 	spin_unlock(&device_klist->k_lock);
1214 }
1215 
1216 static void __init pci_insertion_sort_devices(struct pci_dev *a, struct list_head *list)
1217 {
1218 	struct pci_dev *b;
1219 
1220 	list_for_each_entry(b, list, global_list) {
1221 		if (pci_sort_bf_cmp(a, b) <= 0) {
1222 			list_move_tail(&a->global_list, &b->global_list);
1223 			return;
1224 		}
1225 	}
1226 	list_move_tail(&a->global_list, list);
1227 }
1228 
1229 static void __init pci_sort_breadthfirst_devices(void)
1230 {
1231 	LIST_HEAD(sorted_devices);
1232 	struct pci_dev *dev, *tmp;
1233 
1234 	down_write(&pci_bus_sem);
1235 	list_for_each_entry_safe(dev, tmp, &pci_devices, global_list) {
1236 		pci_insertion_sort_devices(dev, &sorted_devices);
1237 	}
1238 	list_splice(&sorted_devices, &pci_devices);
1239 	up_write(&pci_bus_sem);
1240 }
1241 
1242 void __init pci_sort_breadthfirst(void)
1243 {
1244 	pci_sort_breadthfirst_devices();
1245 	pci_sort_breadthfirst_klist();
1246 }
1247 
1248