xref: /openbmc/linux/drivers/pci/probe.c (revision f42b3800)
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 "pci.h"
13 
14 #define CARDBUS_LATENCY_TIMER	176	/* secondary latency timer */
15 #define CARDBUS_RESERVE_BUSNR	3
16 #define PCI_CFG_SPACE_SIZE	256
17 #define PCI_CFG_SPACE_EXP_SIZE	4096
18 
19 /* Ugh.  Need to stop exporting this to modules. */
20 LIST_HEAD(pci_root_buses);
21 EXPORT_SYMBOL(pci_root_buses);
22 
23 LIST_HEAD(pci_devices);
24 
25 /*
26  * Some device drivers need know if pci is initiated.
27  * Basically, we think pci is not initiated when there
28  * is no device in list of pci_devices.
29  */
30 int no_pci_devices(void)
31 {
32 	return list_empty(&pci_devices);
33 }
34 
35 EXPORT_SYMBOL(no_pci_devices);
36 
37 #ifdef HAVE_PCI_LEGACY
38 /**
39  * pci_create_legacy_files - create legacy I/O port and memory files
40  * @b: bus to create files under
41  *
42  * Some platforms allow access to legacy I/O port and ISA memory space on
43  * a per-bus basis.  This routine creates the files and ties them into
44  * their associated read, write and mmap files from pci-sysfs.c
45  */
46 static void pci_create_legacy_files(struct pci_bus *b)
47 {
48 	b->legacy_io = kzalloc(sizeof(struct bin_attribute) * 2,
49 			       GFP_ATOMIC);
50 	if (b->legacy_io) {
51 		b->legacy_io->attr.name = "legacy_io";
52 		b->legacy_io->size = 0xffff;
53 		b->legacy_io->attr.mode = S_IRUSR | S_IWUSR;
54 		b->legacy_io->read = pci_read_legacy_io;
55 		b->legacy_io->write = pci_write_legacy_io;
56 		device_create_bin_file(&b->dev, b->legacy_io);
57 
58 		/* Allocated above after the legacy_io struct */
59 		b->legacy_mem = b->legacy_io + 1;
60 		b->legacy_mem->attr.name = "legacy_mem";
61 		b->legacy_mem->size = 1024*1024;
62 		b->legacy_mem->attr.mode = S_IRUSR | S_IWUSR;
63 		b->legacy_mem->mmap = pci_mmap_legacy_mem;
64 		device_create_bin_file(&b->dev, b->legacy_mem);
65 	}
66 }
67 
68 void pci_remove_legacy_files(struct pci_bus *b)
69 {
70 	if (b->legacy_io) {
71 		device_remove_bin_file(&b->dev, b->legacy_io);
72 		device_remove_bin_file(&b->dev, b->legacy_mem);
73 		kfree(b->legacy_io); /* both are allocated here */
74 	}
75 }
76 #else /* !HAVE_PCI_LEGACY */
77 static inline void pci_create_legacy_files(struct pci_bus *bus) { return; }
78 void pci_remove_legacy_files(struct pci_bus *bus) { return; }
79 #endif /* HAVE_PCI_LEGACY */
80 
81 /*
82  * PCI Bus Class Devices
83  */
84 static ssize_t pci_bus_show_cpuaffinity(struct device *dev,
85 					struct device_attribute *attr,
86 					char *buf)
87 {
88 	int ret;
89 	cpumask_t cpumask;
90 
91 	cpumask = pcibus_to_cpumask(to_pci_bus(dev));
92 	ret = cpumask_scnprintf(buf, PAGE_SIZE, cpumask);
93 	if (ret < PAGE_SIZE)
94 		buf[ret++] = '\n';
95 	return ret;
96 }
97 DEVICE_ATTR(cpuaffinity, S_IRUGO, pci_bus_show_cpuaffinity, NULL);
98 
99 /*
100  * PCI Bus Class
101  */
102 static void release_pcibus_dev(struct device *dev)
103 {
104 	struct pci_bus *pci_bus = to_pci_bus(dev);
105 
106 	if (pci_bus->bridge)
107 		put_device(pci_bus->bridge);
108 	kfree(pci_bus);
109 }
110 
111 static struct class pcibus_class = {
112 	.name		= "pci_bus",
113 	.dev_release	= &release_pcibus_dev,
114 };
115 
116 static int __init pcibus_class_init(void)
117 {
118 	return class_register(&pcibus_class);
119 }
120 postcore_initcall(pcibus_class_init);
121 
122 /*
123  * Translate the low bits of the PCI base
124  * to the resource type
125  */
126 static inline unsigned int pci_calc_resource_flags(unsigned int flags)
127 {
128 	if (flags & PCI_BASE_ADDRESS_SPACE_IO)
129 		return IORESOURCE_IO;
130 
131 	if (flags & PCI_BASE_ADDRESS_MEM_PREFETCH)
132 		return IORESOURCE_MEM | IORESOURCE_PREFETCH;
133 
134 	return IORESOURCE_MEM;
135 }
136 
137 /*
138  * Find the extent of a PCI decode..
139  */
140 static u32 pci_size(u32 base, u32 maxbase, u32 mask)
141 {
142 	u32 size = mask & maxbase;	/* Find the significant bits */
143 	if (!size)
144 		return 0;
145 
146 	/* Get the lowest of them to find the decode size, and
147 	   from that the extent.  */
148 	size = (size & ~(size-1)) - 1;
149 
150 	/* base == maxbase can be valid only if the BAR has
151 	   already been programmed with all 1s.  */
152 	if (base == maxbase && ((base | size) & mask) != mask)
153 		return 0;
154 
155 	return size;
156 }
157 
158 static u64 pci_size64(u64 base, u64 maxbase, u64 mask)
159 {
160 	u64 size = mask & maxbase;	/* Find the significant bits */
161 	if (!size)
162 		return 0;
163 
164 	/* Get the lowest of them to find the decode size, and
165 	   from that the extent.  */
166 	size = (size & ~(size-1)) - 1;
167 
168 	/* base == maxbase can be valid only if the BAR has
169 	   already been programmed with all 1s.  */
170 	if (base == maxbase && ((base | size) & mask) != mask)
171 		return 0;
172 
173 	return size;
174 }
175 
176 static inline int is_64bit_memory(u32 mask)
177 {
178 	if ((mask & (PCI_BASE_ADDRESS_SPACE|PCI_BASE_ADDRESS_MEM_TYPE_MASK)) ==
179 	    (PCI_BASE_ADDRESS_SPACE_MEMORY|PCI_BASE_ADDRESS_MEM_TYPE_64))
180 		return 1;
181 	return 0;
182 }
183 
184 static void pci_read_bases(struct pci_dev *dev, unsigned int howmany, int rom)
185 {
186 	unsigned int pos, reg, next;
187 	u32 l, sz;
188 	struct resource *res;
189 
190 	for(pos=0; pos<howmany; pos = next) {
191 		u64 l64;
192 		u64 sz64;
193 		u32 raw_sz;
194 
195 		next = pos+1;
196 		res = &dev->resource[pos];
197 		res->name = pci_name(dev);
198 		reg = PCI_BASE_ADDRESS_0 + (pos << 2);
199 		pci_read_config_dword(dev, reg, &l);
200 		pci_write_config_dword(dev, reg, ~0);
201 		pci_read_config_dword(dev, reg, &sz);
202 		pci_write_config_dword(dev, reg, l);
203 		if (!sz || sz == 0xffffffff)
204 			continue;
205 		if (l == 0xffffffff)
206 			l = 0;
207 		raw_sz = sz;
208 		if ((l & PCI_BASE_ADDRESS_SPACE) ==
209 				PCI_BASE_ADDRESS_SPACE_MEMORY) {
210 			sz = pci_size(l, sz, (u32)PCI_BASE_ADDRESS_MEM_MASK);
211 			/*
212 			 * For 64bit prefetchable memory sz could be 0, if the
213 			 * real size is bigger than 4G, so we need to check
214 			 * szhi for that.
215 			 */
216 			if (!is_64bit_memory(l) && !sz)
217 				continue;
218 			res->start = l & PCI_BASE_ADDRESS_MEM_MASK;
219 			res->flags |= l & ~PCI_BASE_ADDRESS_MEM_MASK;
220 		} else {
221 			sz = pci_size(l, sz, PCI_BASE_ADDRESS_IO_MASK & 0xffff);
222 			if (!sz)
223 				continue;
224 			res->start = l & PCI_BASE_ADDRESS_IO_MASK;
225 			res->flags |= l & ~PCI_BASE_ADDRESS_IO_MASK;
226 		}
227 		res->end = res->start + (unsigned long) sz;
228 		res->flags |= pci_calc_resource_flags(l);
229 		if (is_64bit_memory(l)) {
230 			u32 szhi, lhi;
231 
232 			pci_read_config_dword(dev, reg+4, &lhi);
233 			pci_write_config_dword(dev, reg+4, ~0);
234 			pci_read_config_dword(dev, reg+4, &szhi);
235 			pci_write_config_dword(dev, reg+4, lhi);
236 			sz64 = ((u64)szhi << 32) | raw_sz;
237 			l64 = ((u64)lhi << 32) | l;
238 			sz64 = pci_size64(l64, sz64, PCI_BASE_ADDRESS_MEM_MASK);
239 			next++;
240 #if BITS_PER_LONG == 64
241 			if (!sz64) {
242 				res->start = 0;
243 				res->end = 0;
244 				res->flags = 0;
245 				continue;
246 			}
247 			res->start = l64 & PCI_BASE_ADDRESS_MEM_MASK;
248 			res->end = res->start + sz64;
249 #else
250 			if (sz64 > 0x100000000ULL) {
251 				printk(KERN_ERR "PCI: Unable to handle 64-bit "
252 					"BAR for device %s\n", pci_name(dev));
253 				res->start = 0;
254 				res->flags = 0;
255 			} else if (lhi) {
256 				/* 64-bit wide address, treat as disabled */
257 				pci_write_config_dword(dev, reg,
258 					l & ~(u32)PCI_BASE_ADDRESS_MEM_MASK);
259 				pci_write_config_dword(dev, reg+4, 0);
260 				res->start = 0;
261 				res->end = sz;
262 			}
263 #endif
264 		}
265 	}
266 	if (rom) {
267 		dev->rom_base_reg = rom;
268 		res = &dev->resource[PCI_ROM_RESOURCE];
269 		res->name = pci_name(dev);
270 		pci_read_config_dword(dev, rom, &l);
271 		pci_write_config_dword(dev, rom, ~PCI_ROM_ADDRESS_ENABLE);
272 		pci_read_config_dword(dev, rom, &sz);
273 		pci_write_config_dword(dev, rom, l);
274 		if (l == 0xffffffff)
275 			l = 0;
276 		if (sz && sz != 0xffffffff) {
277 			sz = pci_size(l, sz, (u32)PCI_ROM_ADDRESS_MASK);
278 			if (sz) {
279 				res->flags = (l & IORESOURCE_ROM_ENABLE) |
280 				  IORESOURCE_MEM | IORESOURCE_PREFETCH |
281 				  IORESOURCE_READONLY | IORESOURCE_CACHEABLE;
282 				res->start = l & PCI_ROM_ADDRESS_MASK;
283 				res->end = res->start + (unsigned long) sz;
284 			}
285 		}
286 	}
287 }
288 
289 void __devinit pci_read_bridge_bases(struct pci_bus *child)
290 {
291 	struct pci_dev *dev = child->self;
292 	u8 io_base_lo, io_limit_lo;
293 	u16 mem_base_lo, mem_limit_lo;
294 	unsigned long base, limit;
295 	struct resource *res;
296 	int i;
297 
298 	if (!dev)		/* It's a host bus, nothing to read */
299 		return;
300 
301 	if (dev->transparent) {
302 		printk(KERN_INFO "PCI: Transparent bridge - %s\n", pci_name(dev));
303 		for(i = 3; i < PCI_BUS_NUM_RESOURCES; i++)
304 			child->resource[i] = child->parent->resource[i - 3];
305 	}
306 
307 	for(i=0; i<3; i++)
308 		child->resource[i] = &dev->resource[PCI_BRIDGE_RESOURCES+i];
309 
310 	res = child->resource[0];
311 	pci_read_config_byte(dev, PCI_IO_BASE, &io_base_lo);
312 	pci_read_config_byte(dev, PCI_IO_LIMIT, &io_limit_lo);
313 	base = (io_base_lo & PCI_IO_RANGE_MASK) << 8;
314 	limit = (io_limit_lo & PCI_IO_RANGE_MASK) << 8;
315 
316 	if ((io_base_lo & PCI_IO_RANGE_TYPE_MASK) == PCI_IO_RANGE_TYPE_32) {
317 		u16 io_base_hi, io_limit_hi;
318 		pci_read_config_word(dev, PCI_IO_BASE_UPPER16, &io_base_hi);
319 		pci_read_config_word(dev, PCI_IO_LIMIT_UPPER16, &io_limit_hi);
320 		base |= (io_base_hi << 16);
321 		limit |= (io_limit_hi << 16);
322 	}
323 
324 	if (base <= limit) {
325 		res->flags = (io_base_lo & PCI_IO_RANGE_TYPE_MASK) | IORESOURCE_IO;
326 		if (!res->start)
327 			res->start = base;
328 		if (!res->end)
329 			res->end = limit + 0xfff;
330 	}
331 
332 	res = child->resource[1];
333 	pci_read_config_word(dev, PCI_MEMORY_BASE, &mem_base_lo);
334 	pci_read_config_word(dev, PCI_MEMORY_LIMIT, &mem_limit_lo);
335 	base = (mem_base_lo & PCI_MEMORY_RANGE_MASK) << 16;
336 	limit = (mem_limit_lo & PCI_MEMORY_RANGE_MASK) << 16;
337 	if (base <= limit) {
338 		res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM;
339 		res->start = base;
340 		res->end = limit + 0xfffff;
341 	}
342 
343 	res = child->resource[2];
344 	pci_read_config_word(dev, PCI_PREF_MEMORY_BASE, &mem_base_lo);
345 	pci_read_config_word(dev, PCI_PREF_MEMORY_LIMIT, &mem_limit_lo);
346 	base = (mem_base_lo & PCI_PREF_RANGE_MASK) << 16;
347 	limit = (mem_limit_lo & PCI_PREF_RANGE_MASK) << 16;
348 
349 	if ((mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) {
350 		u32 mem_base_hi, mem_limit_hi;
351 		pci_read_config_dword(dev, PCI_PREF_BASE_UPPER32, &mem_base_hi);
352 		pci_read_config_dword(dev, PCI_PREF_LIMIT_UPPER32, &mem_limit_hi);
353 
354 		/*
355 		 * Some bridges set the base > limit by default, and some
356 		 * (broken) BIOSes do not initialize them.  If we find
357 		 * this, just assume they are not being used.
358 		 */
359 		if (mem_base_hi <= mem_limit_hi) {
360 #if BITS_PER_LONG == 64
361 			base |= ((long) mem_base_hi) << 32;
362 			limit |= ((long) mem_limit_hi) << 32;
363 #else
364 			if (mem_base_hi || mem_limit_hi) {
365 				printk(KERN_ERR "PCI: Unable to handle 64-bit address space for bridge %s\n", pci_name(dev));
366 				return;
367 			}
368 #endif
369 		}
370 	}
371 	if (base <= limit) {
372 		res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM | IORESOURCE_PREFETCH;
373 		res->start = base;
374 		res->end = limit + 0xfffff;
375 	}
376 }
377 
378 static struct pci_bus * pci_alloc_bus(void)
379 {
380 	struct pci_bus *b;
381 
382 	b = kzalloc(sizeof(*b), GFP_KERNEL);
383 	if (b) {
384 		INIT_LIST_HEAD(&b->node);
385 		INIT_LIST_HEAD(&b->children);
386 		INIT_LIST_HEAD(&b->devices);
387 	}
388 	return b;
389 }
390 
391 static struct pci_bus * __devinit
392 pci_alloc_child_bus(struct pci_bus *parent, struct pci_dev *bridge, int busnr)
393 {
394 	struct pci_bus *child;
395 	int i;
396 
397 	/*
398 	 * Allocate a new bus, and inherit stuff from the parent..
399 	 */
400 	child = pci_alloc_bus();
401 	if (!child)
402 		return NULL;
403 
404 	child->self = bridge;
405 	child->parent = parent;
406 	child->ops = parent->ops;
407 	child->sysdata = parent->sysdata;
408 	child->bus_flags = parent->bus_flags;
409 	child->bridge = get_device(&bridge->dev);
410 
411 	/* initialize some portions of the bus device, but don't register it
412 	 * now as the parent is not properly set up yet.  This device will get
413 	 * registered later in pci_bus_add_devices()
414 	 */
415 	child->dev.class = &pcibus_class;
416 	sprintf(child->dev.bus_id, "%04x:%02x", pci_domain_nr(child), busnr);
417 
418 	/*
419 	 * Set up the primary, secondary and subordinate
420 	 * bus numbers.
421 	 */
422 	child->number = child->secondary = busnr;
423 	child->primary = parent->secondary;
424 	child->subordinate = 0xff;
425 
426 	/* Set up default resource pointers and names.. */
427 	for (i = 0; i < 4; i++) {
428 		child->resource[i] = &bridge->resource[PCI_BRIDGE_RESOURCES+i];
429 		child->resource[i]->name = child->name;
430 	}
431 	bridge->subordinate = child;
432 
433 	return child;
434 }
435 
436 struct pci_bus *__ref pci_add_new_bus(struct pci_bus *parent, struct pci_dev *dev, int busnr)
437 {
438 	struct pci_bus *child;
439 
440 	child = pci_alloc_child_bus(parent, dev, busnr);
441 	if (child) {
442 		down_write(&pci_bus_sem);
443 		list_add_tail(&child->node, &parent->children);
444 		up_write(&pci_bus_sem);
445 	}
446 	return child;
447 }
448 
449 static void pci_fixup_parent_subordinate_busnr(struct pci_bus *child, int max)
450 {
451 	struct pci_bus *parent = child->parent;
452 
453 	/* Attempts to fix that up are really dangerous unless
454 	   we're going to re-assign all bus numbers. */
455 	if (!pcibios_assign_all_busses())
456 		return;
457 
458 	while (parent->parent && parent->subordinate < max) {
459 		parent->subordinate = max;
460 		pci_write_config_byte(parent->self, PCI_SUBORDINATE_BUS, max);
461 		parent = parent->parent;
462 	}
463 }
464 
465 /*
466  * If it's a bridge, configure it and scan the bus behind it.
467  * For CardBus bridges, we don't scan behind as the devices will
468  * be handled by the bridge driver itself.
469  *
470  * We need to process bridges in two passes -- first we scan those
471  * already configured by the BIOS and after we are done with all of
472  * them, we proceed to assigning numbers to the remaining buses in
473  * order to avoid overlaps between old and new bus numbers.
474  */
475 int __devinit pci_scan_bridge(struct pci_bus *bus, struct pci_dev *dev, int max, int pass)
476 {
477 	struct pci_bus *child;
478 	int is_cardbus = (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS);
479 	u32 buses, i, j = 0;
480 	u16 bctl;
481 
482 	pci_read_config_dword(dev, PCI_PRIMARY_BUS, &buses);
483 
484 	pr_debug("PCI: Scanning behind PCI bridge %s, config %06x, pass %d\n",
485 		 pci_name(dev), buses & 0xffffff, pass);
486 
487 	/* Disable MasterAbortMode during probing to avoid reporting
488 	   of bus errors (in some architectures) */
489 	pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl);
490 	pci_write_config_word(dev, PCI_BRIDGE_CONTROL,
491 			      bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT);
492 
493 	if ((buses & 0xffff00) && !pcibios_assign_all_busses() && !is_cardbus) {
494 		unsigned int cmax, busnr;
495 		/*
496 		 * Bus already configured by firmware, process it in the first
497 		 * pass and just note the configuration.
498 		 */
499 		if (pass)
500 			goto out;
501 		busnr = (buses >> 8) & 0xFF;
502 
503 		/*
504 		 * If we already got to this bus through a different bridge,
505 		 * ignore it.  This can happen with the i450NX chipset.
506 		 */
507 		if (pci_find_bus(pci_domain_nr(bus), busnr)) {
508 			printk(KERN_INFO "PCI: Bus %04x:%02x already known\n",
509 					pci_domain_nr(bus), busnr);
510 			goto out;
511 		}
512 
513 		child = pci_add_new_bus(bus, dev, busnr);
514 		if (!child)
515 			goto out;
516 		child->primary = buses & 0xFF;
517 		child->subordinate = (buses >> 16) & 0xFF;
518 		child->bridge_ctl = bctl;
519 
520 		cmax = pci_scan_child_bus(child);
521 		if (cmax > max)
522 			max = cmax;
523 		if (child->subordinate > max)
524 			max = child->subordinate;
525 	} else {
526 		/*
527 		 * We need to assign a number to this bus which we always
528 		 * do in the second pass.
529 		 */
530 		if (!pass) {
531 			if (pcibios_assign_all_busses())
532 				/* Temporarily disable forwarding of the
533 				   configuration cycles on all bridges in
534 				   this bus segment to avoid possible
535 				   conflicts in the second pass between two
536 				   bridges programmed with overlapping
537 				   bus ranges. */
538 				pci_write_config_dword(dev, PCI_PRIMARY_BUS,
539 						       buses & ~0xffffff);
540 			goto out;
541 		}
542 
543 		/* Clear errors */
544 		pci_write_config_word(dev, PCI_STATUS, 0xffff);
545 
546 		/* Prevent assigning a bus number that already exists.
547 		 * This can happen when a bridge is hot-plugged */
548 		if (pci_find_bus(pci_domain_nr(bus), max+1))
549 			goto out;
550 		child = pci_add_new_bus(bus, dev, ++max);
551 		buses = (buses & 0xff000000)
552 		      | ((unsigned int)(child->primary)     <<  0)
553 		      | ((unsigned int)(child->secondary)   <<  8)
554 		      | ((unsigned int)(child->subordinate) << 16);
555 
556 		/*
557 		 * yenta.c forces a secondary latency timer of 176.
558 		 * Copy that behaviour here.
559 		 */
560 		if (is_cardbus) {
561 			buses &= ~0xff000000;
562 			buses |= CARDBUS_LATENCY_TIMER << 24;
563 		}
564 
565 		/*
566 		 * We need to blast all three values with a single write.
567 		 */
568 		pci_write_config_dword(dev, PCI_PRIMARY_BUS, buses);
569 
570 		if (!is_cardbus) {
571 			child->bridge_ctl = bctl;
572 			/*
573 			 * Adjust subordinate busnr in parent buses.
574 			 * We do this before scanning for children because
575 			 * some devices may not be detected if the bios
576 			 * was lazy.
577 			 */
578 			pci_fixup_parent_subordinate_busnr(child, max);
579 			/* Now we can scan all subordinate buses... */
580 			max = pci_scan_child_bus(child);
581 			/*
582 			 * now fix it up again since we have found
583 			 * the real value of max.
584 			 */
585 			pci_fixup_parent_subordinate_busnr(child, max);
586 		} else {
587 			/*
588 			 * For CardBus bridges, we leave 4 bus numbers
589 			 * as cards with a PCI-to-PCI bridge can be
590 			 * inserted later.
591 			 */
592 			for (i=0; i<CARDBUS_RESERVE_BUSNR; i++) {
593 				struct pci_bus *parent = bus;
594 				if (pci_find_bus(pci_domain_nr(bus),
595 							max+i+1))
596 					break;
597 				while (parent->parent) {
598 					if ((!pcibios_assign_all_busses()) &&
599 					    (parent->subordinate > max) &&
600 					    (parent->subordinate <= max+i)) {
601 						j = 1;
602 					}
603 					parent = parent->parent;
604 				}
605 				if (j) {
606 					/*
607 					 * Often, there are two cardbus bridges
608 					 * -- try to leave one valid bus number
609 					 * for each one.
610 					 */
611 					i /= 2;
612 					break;
613 				}
614 			}
615 			max += i;
616 			pci_fixup_parent_subordinate_busnr(child, max);
617 		}
618 		/*
619 		 * Set the subordinate bus number to its real value.
620 		 */
621 		child->subordinate = max;
622 		pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, max);
623 	}
624 
625 	sprintf(child->name, (is_cardbus ? "PCI CardBus #%02x" : "PCI Bus #%02x"), child->number);
626 
627 	/* Has only triggered on CardBus, fixup is in yenta_socket */
628 	while (bus->parent) {
629 		if ((child->subordinate > bus->subordinate) ||
630 		    (child->number > bus->subordinate) ||
631 		    (child->number < bus->number) ||
632 		    (child->subordinate < bus->number)) {
633 			pr_debug("PCI: Bus #%02x (-#%02x) is %s "
634 				"hidden behind%s bridge #%02x (-#%02x)\n",
635 				child->number, child->subordinate,
636 				(bus->number > child->subordinate &&
637 				 bus->subordinate < child->number) ?
638 					"wholly" : "partially",
639 				bus->self->transparent ? " transparent" : "",
640 				bus->number, bus->subordinate);
641 		}
642 		bus = bus->parent;
643 	}
644 
645 out:
646 	pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl);
647 
648 	return max;
649 }
650 
651 /*
652  * Read interrupt line and base address registers.
653  * The architecture-dependent code can tweak these, of course.
654  */
655 static void pci_read_irq(struct pci_dev *dev)
656 {
657 	unsigned char irq;
658 
659 	pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &irq);
660 	dev->pin = irq;
661 	if (irq)
662 		pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq);
663 	dev->irq = irq;
664 }
665 
666 #define LEGACY_IO_RESOURCE	(IORESOURCE_IO | IORESOURCE_PCI_FIXED)
667 
668 /**
669  * pci_setup_device - fill in class and map information of a device
670  * @dev: the device structure to fill
671  *
672  * Initialize the device structure with information about the device's
673  * vendor,class,memory and IO-space addresses,IRQ lines etc.
674  * Called at initialisation of the PCI subsystem and by CardBus services.
675  * Returns 0 on success and -1 if unknown type of device (not normal, bridge
676  * or CardBus).
677  */
678 static int pci_setup_device(struct pci_dev * dev)
679 {
680 	u32 class;
681 
682 	sprintf(pci_name(dev), "%04x:%02x:%02x.%d", pci_domain_nr(dev->bus),
683 		dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn));
684 
685 	pci_read_config_dword(dev, PCI_CLASS_REVISION, &class);
686 	dev->revision = class & 0xff;
687 	class >>= 8;				    /* upper 3 bytes */
688 	dev->class = class;
689 	class >>= 8;
690 
691 	pr_debug("PCI: Found %s [%04x/%04x] %06x %02x\n", pci_name(dev),
692 		 dev->vendor, dev->device, class, dev->hdr_type);
693 
694 	/* "Unknown power state" */
695 	dev->current_state = PCI_UNKNOWN;
696 
697 	/* Early fixups, before probing the BARs */
698 	pci_fixup_device(pci_fixup_early, dev);
699 	class = dev->class >> 8;
700 
701 	switch (dev->hdr_type) {		    /* header type */
702 	case PCI_HEADER_TYPE_NORMAL:		    /* standard header */
703 		if (class == PCI_CLASS_BRIDGE_PCI)
704 			goto bad;
705 		pci_read_irq(dev);
706 		pci_read_bases(dev, 6, PCI_ROM_ADDRESS);
707 		pci_read_config_word(dev, PCI_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
708 		pci_read_config_word(dev, PCI_SUBSYSTEM_ID, &dev->subsystem_device);
709 
710 		/*
711 		 *	Do the ugly legacy mode stuff here rather than broken chip
712 		 *	quirk code. Legacy mode ATA controllers have fixed
713 		 *	addresses. These are not always echoed in BAR0-3, and
714 		 *	BAR0-3 in a few cases contain junk!
715 		 */
716 		if (class == PCI_CLASS_STORAGE_IDE) {
717 			u8 progif;
718 			pci_read_config_byte(dev, PCI_CLASS_PROG, &progif);
719 			if ((progif & 1) == 0) {
720 				dev->resource[0].start = 0x1F0;
721 				dev->resource[0].end = 0x1F7;
722 				dev->resource[0].flags = LEGACY_IO_RESOURCE;
723 				dev->resource[1].start = 0x3F6;
724 				dev->resource[1].end = 0x3F6;
725 				dev->resource[1].flags = LEGACY_IO_RESOURCE;
726 			}
727 			if ((progif & 4) == 0) {
728 				dev->resource[2].start = 0x170;
729 				dev->resource[2].end = 0x177;
730 				dev->resource[2].flags = LEGACY_IO_RESOURCE;
731 				dev->resource[3].start = 0x376;
732 				dev->resource[3].end = 0x376;
733 				dev->resource[3].flags = LEGACY_IO_RESOURCE;
734 			}
735 		}
736 		break;
737 
738 	case PCI_HEADER_TYPE_BRIDGE:		    /* bridge header */
739 		if (class != PCI_CLASS_BRIDGE_PCI)
740 			goto bad;
741 		/* The PCI-to-PCI bridge spec requires that subtractive
742 		   decoding (i.e. transparent) bridge must have programming
743 		   interface code of 0x01. */
744 		pci_read_irq(dev);
745 		dev->transparent = ((dev->class & 0xff) == 1);
746 		pci_read_bases(dev, 2, PCI_ROM_ADDRESS1);
747 		break;
748 
749 	case PCI_HEADER_TYPE_CARDBUS:		    /* CardBus bridge header */
750 		if (class != PCI_CLASS_BRIDGE_CARDBUS)
751 			goto bad;
752 		pci_read_irq(dev);
753 		pci_read_bases(dev, 1, 0);
754 		pci_read_config_word(dev, PCI_CB_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
755 		pci_read_config_word(dev, PCI_CB_SUBSYSTEM_ID, &dev->subsystem_device);
756 		break;
757 
758 	default:				    /* unknown header */
759 		printk(KERN_ERR "PCI: device %s has unknown header type %02x, ignoring.\n",
760 			pci_name(dev), dev->hdr_type);
761 		return -1;
762 
763 	bad:
764 		printk(KERN_ERR "PCI: %s: class %x doesn't match header type %02x. Ignoring class.\n",
765 		       pci_name(dev), class, dev->hdr_type);
766 		dev->class = PCI_CLASS_NOT_DEFINED;
767 	}
768 
769 	/* We found a fine healthy device, go go go... */
770 	return 0;
771 }
772 
773 /**
774  * pci_release_dev - free a pci device structure when all users of it are finished.
775  * @dev: device that's been disconnected
776  *
777  * Will be called only by the device core when all users of this pci device are
778  * done.
779  */
780 static void pci_release_dev(struct device *dev)
781 {
782 	struct pci_dev *pci_dev;
783 
784 	pci_dev = to_pci_dev(dev);
785 	kfree(pci_dev);
786 }
787 
788 static void set_pcie_port_type(struct pci_dev *pdev)
789 {
790 	int pos;
791 	u16 reg16;
792 
793 	pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
794 	if (!pos)
795 		return;
796 	pdev->is_pcie = 1;
797 	pci_read_config_word(pdev, pos + PCI_EXP_FLAGS, &reg16);
798 	pdev->pcie_type = (reg16 & PCI_EXP_FLAGS_TYPE) >> 4;
799 }
800 
801 /**
802  * pci_cfg_space_size - get the configuration space size of the PCI device.
803  * @dev: PCI device
804  *
805  * Regular PCI devices have 256 bytes, but PCI-X 2 and PCI Express devices
806  * have 4096 bytes.  Even if the device is capable, that doesn't mean we can
807  * access it.  Maybe we don't have a way to generate extended config space
808  * accesses, or the device is behind a reverse Express bridge.  So we try
809  * reading the dword at 0x100 which must either be 0 or a valid extended
810  * capability header.
811  */
812 int pci_cfg_space_size(struct pci_dev *dev)
813 {
814 	int pos;
815 	u32 status;
816 
817 	pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
818 	if (!pos) {
819 		pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
820 		if (!pos)
821 			goto fail;
822 
823 		pci_read_config_dword(dev, pos + PCI_X_STATUS, &status);
824 		if (!(status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ)))
825 			goto fail;
826 	}
827 
828 	if (pci_read_config_dword(dev, 256, &status) != PCIBIOS_SUCCESSFUL)
829 		goto fail;
830 	if (status == 0xffffffff)
831 		goto fail;
832 
833 	return PCI_CFG_SPACE_EXP_SIZE;
834 
835  fail:
836 	return PCI_CFG_SPACE_SIZE;
837 }
838 
839 static void pci_release_bus_bridge_dev(struct device *dev)
840 {
841 	kfree(dev);
842 }
843 
844 struct pci_dev *alloc_pci_dev(void)
845 {
846 	struct pci_dev *dev;
847 
848 	dev = kzalloc(sizeof(struct pci_dev), GFP_KERNEL);
849 	if (!dev)
850 		return NULL;
851 
852 	INIT_LIST_HEAD(&dev->global_list);
853 	INIT_LIST_HEAD(&dev->bus_list);
854 
855 	pci_msi_init_pci_dev(dev);
856 
857 	return dev;
858 }
859 EXPORT_SYMBOL(alloc_pci_dev);
860 
861 /*
862  * Read the config data for a PCI device, sanity-check it
863  * and fill in the dev structure...
864  */
865 static struct pci_dev * __devinit
866 pci_scan_device(struct pci_bus *bus, int devfn)
867 {
868 	struct pci_dev *dev;
869 	u32 l;
870 	u8 hdr_type;
871 	int delay = 1;
872 
873 	if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, &l))
874 		return NULL;
875 
876 	/* some broken boards return 0 or ~0 if a slot is empty: */
877 	if (l == 0xffffffff || l == 0x00000000 ||
878 	    l == 0x0000ffff || l == 0xffff0000)
879 		return NULL;
880 
881 	/* Configuration request Retry Status */
882 	while (l == 0xffff0001) {
883 		msleep(delay);
884 		delay *= 2;
885 		if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, &l))
886 			return NULL;
887 		/* Card hasn't responded in 60 seconds?  Must be stuck. */
888 		if (delay > 60 * 1000) {
889 			printk(KERN_WARNING "Device %04x:%02x:%02x.%d not "
890 					"responding\n", pci_domain_nr(bus),
891 					bus->number, PCI_SLOT(devfn),
892 					PCI_FUNC(devfn));
893 			return NULL;
894 		}
895 	}
896 
897 	if (pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type))
898 		return NULL;
899 
900 	dev = alloc_pci_dev();
901 	if (!dev)
902 		return NULL;
903 
904 	dev->bus = bus;
905 	dev->sysdata = bus->sysdata;
906 	dev->dev.parent = bus->bridge;
907 	dev->dev.bus = &pci_bus_type;
908 	dev->devfn = devfn;
909 	dev->hdr_type = hdr_type & 0x7f;
910 	dev->multifunction = !!(hdr_type & 0x80);
911 	dev->vendor = l & 0xffff;
912 	dev->device = (l >> 16) & 0xffff;
913 	dev->cfg_size = pci_cfg_space_size(dev);
914 	dev->error_state = pci_channel_io_normal;
915 	set_pcie_port_type(dev);
916 
917 	/* Assume 32-bit PCI; let 64-bit PCI cards (which are far rarer)
918 	   set this higher, assuming the system even supports it.  */
919 	dev->dma_mask = 0xffffffff;
920 	if (pci_setup_device(dev) < 0) {
921 		kfree(dev);
922 		return NULL;
923 	}
924 
925 	return dev;
926 }
927 
928 void pci_device_add(struct pci_dev *dev, struct pci_bus *bus)
929 {
930 	device_initialize(&dev->dev);
931 	dev->dev.release = pci_release_dev;
932 	pci_dev_get(dev);
933 
934 	set_dev_node(&dev->dev, pcibus_to_node(bus));
935 	dev->dev.dma_mask = &dev->dma_mask;
936 	dev->dev.dma_parms = &dev->dma_parms;
937 	dev->dev.coherent_dma_mask = 0xffffffffull;
938 
939 	pci_set_dma_max_seg_size(dev, 65536);
940 	pci_set_dma_seg_boundary(dev, 0xffffffff);
941 
942 	/* Fix up broken headers */
943 	pci_fixup_device(pci_fixup_header, dev);
944 
945 	/*
946 	 * Add the device to our list of discovered devices
947 	 * and the bus list for fixup functions, etc.
948 	 */
949 	INIT_LIST_HEAD(&dev->global_list);
950 	down_write(&pci_bus_sem);
951 	list_add_tail(&dev->bus_list, &bus->devices);
952 	up_write(&pci_bus_sem);
953 }
954 
955 struct pci_dev *__ref pci_scan_single_device(struct pci_bus *bus, int devfn)
956 {
957 	struct pci_dev *dev;
958 
959 	dev = pci_scan_device(bus, devfn);
960 	if (!dev)
961 		return NULL;
962 
963 	pci_device_add(dev, bus);
964 
965 	return dev;
966 }
967 EXPORT_SYMBOL(pci_scan_single_device);
968 
969 /**
970  * pci_scan_slot - scan a PCI slot on a bus for devices.
971  * @bus: PCI bus to scan
972  * @devfn: slot number to scan (must have zero function.)
973  *
974  * Scan a PCI slot on the specified PCI bus for devices, adding
975  * discovered devices to the @bus->devices list.  New devices
976  * will have an empty dev->global_list head.
977  */
978 int pci_scan_slot(struct pci_bus *bus, int devfn)
979 {
980 	int func, nr = 0;
981 	int scan_all_fns;
982 
983 	scan_all_fns = pcibios_scan_all_fns(bus, devfn);
984 
985 	for (func = 0; func < 8; func++, devfn++) {
986 		struct pci_dev *dev;
987 
988 		dev = pci_scan_single_device(bus, devfn);
989 		if (dev) {
990 			nr++;
991 
992 			/*
993 		 	 * If this is a single function device,
994 		 	 * don't scan past the first function.
995 		 	 */
996 			if (!dev->multifunction) {
997 				if (func > 0) {
998 					dev->multifunction = 1;
999 				} else {
1000  					break;
1001 				}
1002 			}
1003 		} else {
1004 			if (func == 0 && !scan_all_fns)
1005 				break;
1006 		}
1007 	}
1008 	return nr;
1009 }
1010 
1011 unsigned int __devinit pci_scan_child_bus(struct pci_bus *bus)
1012 {
1013 	unsigned int devfn, pass, max = bus->secondary;
1014 	struct pci_dev *dev;
1015 
1016 	pr_debug("PCI: Scanning bus %04x:%02x\n", pci_domain_nr(bus), bus->number);
1017 
1018 	/* Go find them, Rover! */
1019 	for (devfn = 0; devfn < 0x100; devfn += 8)
1020 		pci_scan_slot(bus, devfn);
1021 
1022 	/*
1023 	 * After performing arch-dependent fixup of the bus, look behind
1024 	 * all PCI-to-PCI bridges on this bus.
1025 	 */
1026 	pr_debug("PCI: Fixups for bus %04x:%02x\n", pci_domain_nr(bus), bus->number);
1027 	pcibios_fixup_bus(bus);
1028 	for (pass=0; pass < 2; pass++)
1029 		list_for_each_entry(dev, &bus->devices, bus_list) {
1030 			if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE ||
1031 			    dev->hdr_type == PCI_HEADER_TYPE_CARDBUS)
1032 				max = pci_scan_bridge(bus, dev, max, pass);
1033 		}
1034 
1035 	/*
1036 	 * We've scanned the bus and so we know all about what's on
1037 	 * the other side of any bridges that may be on this bus plus
1038 	 * any devices.
1039 	 *
1040 	 * Return how far we've got finding sub-buses.
1041 	 */
1042 	pr_debug("PCI: Bus scan for %04x:%02x returning with max=%02x\n",
1043 		pci_domain_nr(bus), bus->number, max);
1044 	return max;
1045 }
1046 
1047 struct pci_bus * pci_create_bus(struct device *parent,
1048 		int bus, struct pci_ops *ops, void *sysdata)
1049 {
1050 	int error;
1051 	struct pci_bus *b;
1052 	struct device *dev;
1053 
1054 	b = pci_alloc_bus();
1055 	if (!b)
1056 		return NULL;
1057 
1058 	dev = kmalloc(sizeof(*dev), GFP_KERNEL);
1059 	if (!dev){
1060 		kfree(b);
1061 		return NULL;
1062 	}
1063 
1064 	b->sysdata = sysdata;
1065 	b->ops = ops;
1066 
1067 	if (pci_find_bus(pci_domain_nr(b), bus)) {
1068 		/* If we already got to this bus through a different bridge, ignore it */
1069 		pr_debug("PCI: Bus %04x:%02x already known\n", pci_domain_nr(b), bus);
1070 		goto err_out;
1071 	}
1072 
1073 	down_write(&pci_bus_sem);
1074 	list_add_tail(&b->node, &pci_root_buses);
1075 	up_write(&pci_bus_sem);
1076 
1077 	memset(dev, 0, sizeof(*dev));
1078 	dev->parent = parent;
1079 	dev->release = pci_release_bus_bridge_dev;
1080 	sprintf(dev->bus_id, "pci%04x:%02x", pci_domain_nr(b), bus);
1081 	error = device_register(dev);
1082 	if (error)
1083 		goto dev_reg_err;
1084 	b->bridge = get_device(dev);
1085 
1086 	b->dev.class = &pcibus_class;
1087 	b->dev.parent = b->bridge;
1088 	sprintf(b->dev.bus_id, "%04x:%02x", pci_domain_nr(b), bus);
1089 	error = device_register(&b->dev);
1090 	if (error)
1091 		goto class_dev_reg_err;
1092 	error = device_create_file(&b->dev, &dev_attr_cpuaffinity);
1093 	if (error)
1094 		goto dev_create_file_err;
1095 
1096 	/* Create legacy_io and legacy_mem files for this bus */
1097 	pci_create_legacy_files(b);
1098 
1099 	b->number = b->secondary = bus;
1100 	b->resource[0] = &ioport_resource;
1101 	b->resource[1] = &iomem_resource;
1102 
1103 	return b;
1104 
1105 dev_create_file_err:
1106 	device_unregister(&b->dev);
1107 class_dev_reg_err:
1108 	device_unregister(dev);
1109 dev_reg_err:
1110 	down_write(&pci_bus_sem);
1111 	list_del(&b->node);
1112 	up_write(&pci_bus_sem);
1113 err_out:
1114 	kfree(dev);
1115 	kfree(b);
1116 	return NULL;
1117 }
1118 
1119 struct pci_bus * __devinit pci_scan_bus_parented(struct device *parent,
1120 		int bus, struct pci_ops *ops, void *sysdata)
1121 {
1122 	struct pci_bus *b;
1123 
1124 	b = pci_create_bus(parent, bus, ops, sysdata);
1125 	if (b)
1126 		b->subordinate = pci_scan_child_bus(b);
1127 	return b;
1128 }
1129 EXPORT_SYMBOL(pci_scan_bus_parented);
1130 
1131 #ifdef CONFIG_HOTPLUG
1132 EXPORT_SYMBOL(pci_add_new_bus);
1133 EXPORT_SYMBOL(pci_scan_slot);
1134 EXPORT_SYMBOL(pci_scan_bridge);
1135 EXPORT_SYMBOL_GPL(pci_scan_child_bus);
1136 #endif
1137 
1138 static int __init pci_sort_bf_cmp(const struct pci_dev *a, const struct pci_dev *b)
1139 {
1140 	if      (pci_domain_nr(a->bus) < pci_domain_nr(b->bus)) return -1;
1141 	else if (pci_domain_nr(a->bus) > pci_domain_nr(b->bus)) return  1;
1142 
1143 	if      (a->bus->number < b->bus->number) return -1;
1144 	else if (a->bus->number > b->bus->number) return  1;
1145 
1146 	if      (a->devfn < b->devfn) return -1;
1147 	else if (a->devfn > b->devfn) return  1;
1148 
1149 	return 0;
1150 }
1151 
1152 /*
1153  * Yes, this forcably breaks the klist abstraction temporarily.  It
1154  * just wants to sort the klist, not change reference counts and
1155  * take/drop locks rapidly in the process.  It does all this while
1156  * holding the lock for the list, so objects can't otherwise be
1157  * added/removed while we're swizzling.
1158  */
1159 static void __init pci_insertion_sort_klist(struct pci_dev *a, struct list_head *list)
1160 {
1161 	struct list_head *pos;
1162 	struct klist_node *n;
1163 	struct device *dev;
1164 	struct pci_dev *b;
1165 
1166 	list_for_each(pos, list) {
1167 		n = container_of(pos, struct klist_node, n_node);
1168 		dev = container_of(n, struct device, knode_bus);
1169 		b = to_pci_dev(dev);
1170 		if (pci_sort_bf_cmp(a, b) <= 0) {
1171 			list_move_tail(&a->dev.knode_bus.n_node, &b->dev.knode_bus.n_node);
1172 			return;
1173 		}
1174 	}
1175 	list_move_tail(&a->dev.knode_bus.n_node, list);
1176 }
1177 
1178 static void __init pci_sort_breadthfirst_klist(void)
1179 {
1180 	LIST_HEAD(sorted_devices);
1181 	struct list_head *pos, *tmp;
1182 	struct klist_node *n;
1183 	struct device *dev;
1184 	struct pci_dev *pdev;
1185 	struct klist *device_klist;
1186 
1187 	device_klist = bus_get_device_klist(&pci_bus_type);
1188 
1189 	spin_lock(&device_klist->k_lock);
1190 	list_for_each_safe(pos, tmp, &device_klist->k_list) {
1191 		n = container_of(pos, struct klist_node, n_node);
1192 		dev = container_of(n, struct device, knode_bus);
1193 		pdev = to_pci_dev(dev);
1194 		pci_insertion_sort_klist(pdev, &sorted_devices);
1195 	}
1196 	list_splice(&sorted_devices, &device_klist->k_list);
1197 	spin_unlock(&device_klist->k_lock);
1198 }
1199 
1200 static void __init pci_insertion_sort_devices(struct pci_dev *a, struct list_head *list)
1201 {
1202 	struct pci_dev *b;
1203 
1204 	list_for_each_entry(b, list, global_list) {
1205 		if (pci_sort_bf_cmp(a, b) <= 0) {
1206 			list_move_tail(&a->global_list, &b->global_list);
1207 			return;
1208 		}
1209 	}
1210 	list_move_tail(&a->global_list, list);
1211 }
1212 
1213 static void __init pci_sort_breadthfirst_devices(void)
1214 {
1215 	LIST_HEAD(sorted_devices);
1216 	struct pci_dev *dev, *tmp;
1217 
1218 	down_write(&pci_bus_sem);
1219 	list_for_each_entry_safe(dev, tmp, &pci_devices, global_list) {
1220 		pci_insertion_sort_devices(dev, &sorted_devices);
1221 	}
1222 	list_splice(&sorted_devices, &pci_devices);
1223 	up_write(&pci_bus_sem);
1224 }
1225 
1226 void __init pci_sort_breadthfirst(void)
1227 {
1228 	pci_sort_breadthfirst_devices();
1229 	pci_sort_breadthfirst_klist();
1230 }
1231 
1232