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. */
20LIST_HEAD(pci_root_buses);
21EXPORT_SYMBOL(pci_root_buses);
22
23LIST_HEAD(pci_devices);
24
25#ifdef HAVE_PCI_LEGACY
26/**
27 * pci_create_legacy_files - create legacy I/O port and memory files
28 * @b: bus to create files under
29 *
30 * Some platforms allow access to legacy I/O port and ISA memory space on
31 * a per-bus basis.  This routine creates the files and ties them into
32 * their associated read, write and mmap files from pci-sysfs.c
33 */
34static void pci_create_legacy_files(struct pci_bus *b)
35{
36	b->legacy_io = kzalloc(sizeof(struct bin_attribute) * 2,
37			       GFP_ATOMIC);
38	if (b->legacy_io) {
39		b->legacy_io->attr.name = "legacy_io";
40		b->legacy_io->size = 0xffff;
41		b->legacy_io->attr.mode = S_IRUSR | S_IWUSR;
42		b->legacy_io->attr.owner = THIS_MODULE;
43		b->legacy_io->read = pci_read_legacy_io;
44		b->legacy_io->write = pci_write_legacy_io;
45		class_device_create_bin_file(&b->class_dev, b->legacy_io);
46
47		/* Allocated above after the legacy_io struct */
48		b->legacy_mem = b->legacy_io + 1;
49		b->legacy_mem->attr.name = "legacy_mem";
50		b->legacy_mem->size = 1024*1024;
51		b->legacy_mem->attr.mode = S_IRUSR | S_IWUSR;
52		b->legacy_mem->attr.owner = THIS_MODULE;
53		b->legacy_mem->mmap = pci_mmap_legacy_mem;
54		class_device_create_bin_file(&b->class_dev, b->legacy_mem);
55	}
56}
57
58void pci_remove_legacy_files(struct pci_bus *b)
59{
60	if (b->legacy_io) {
61		class_device_remove_bin_file(&b->class_dev, b->legacy_io);
62		class_device_remove_bin_file(&b->class_dev, b->legacy_mem);
63		kfree(b->legacy_io); /* both are allocated here */
64	}
65}
66#else /* !HAVE_PCI_LEGACY */
67static inline void pci_create_legacy_files(struct pci_bus *bus) { return; }
68void pci_remove_legacy_files(struct pci_bus *bus) { return; }
69#endif /* HAVE_PCI_LEGACY */
70
71/*
72 * PCI Bus Class Devices
73 */
74static ssize_t pci_bus_show_cpuaffinity(struct class_device *class_dev,
75					char *buf)
76{
77	int ret;
78	cpumask_t cpumask;
79
80	cpumask = pcibus_to_cpumask(to_pci_bus(class_dev));
81	ret = cpumask_scnprintf(buf, PAGE_SIZE, cpumask);
82	if (ret < PAGE_SIZE)
83		buf[ret++] = '\n';
84	return ret;
85}
86CLASS_DEVICE_ATTR(cpuaffinity, S_IRUGO, pci_bus_show_cpuaffinity, NULL);
87
88/*
89 * PCI Bus Class
90 */
91static void release_pcibus_dev(struct class_device *class_dev)
92{
93	struct pci_bus *pci_bus = to_pci_bus(class_dev);
94
95	if (pci_bus->bridge)
96		put_device(pci_bus->bridge);
97	kfree(pci_bus);
98}
99
100static struct class pcibus_class = {
101	.name		= "pci_bus",
102	.release	= &release_pcibus_dev,
103};
104
105static int __init pcibus_class_init(void)
106{
107	return class_register(&pcibus_class);
108}
109postcore_initcall(pcibus_class_init);
110
111/*
112 * Translate the low bits of the PCI base
113 * to the resource type
114 */
115static inline unsigned int pci_calc_resource_flags(unsigned int flags)
116{
117	if (flags & PCI_BASE_ADDRESS_SPACE_IO)
118		return IORESOURCE_IO;
119
120	if (flags & PCI_BASE_ADDRESS_MEM_PREFETCH)
121		return IORESOURCE_MEM | IORESOURCE_PREFETCH;
122
123	return IORESOURCE_MEM;
124}
125
126/*
127 * Find the extent of a PCI decode..
128 */
129static u32 pci_size(u32 base, u32 maxbase, u32 mask)
130{
131	u32 size = mask & maxbase;	/* Find the significant bits */
132	if (!size)
133		return 0;
134
135	/* Get the lowest of them to find the decode size, and
136	   from that the extent.  */
137	size = (size & ~(size-1)) - 1;
138
139	/* base == maxbase can be valid only if the BAR has
140	   already been programmed with all 1s.  */
141	if (base == maxbase && ((base | size) & mask) != mask)
142		return 0;
143
144	return size;
145}
146
147static u64 pci_size64(u64 base, u64 maxbase, u64 mask)
148{
149	u64 size = mask & maxbase;	/* Find the significant bits */
150	if (!size)
151		return 0;
152
153	/* Get the lowest of them to find the decode size, and
154	   from that the extent.  */
155	size = (size & ~(size-1)) - 1;
156
157	/* base == maxbase can be valid only if the BAR has
158	   already been programmed with all 1s.  */
159	if (base == maxbase && ((base | size) & mask) != mask)
160		return 0;
161
162	return size;
163}
164
165static inline int is_64bit_memory(u32 mask)
166{
167	if ((mask & (PCI_BASE_ADDRESS_SPACE|PCI_BASE_ADDRESS_MEM_TYPE_MASK)) ==
168	    (PCI_BASE_ADDRESS_SPACE_MEMORY|PCI_BASE_ADDRESS_MEM_TYPE_64))
169		return 1;
170	return 0;
171}
172
173static void pci_read_bases(struct pci_dev *dev, unsigned int howmany, int rom)
174{
175	unsigned int pos, reg, next;
176	u32 l, sz;
177	struct resource *res;
178
179	for(pos=0; pos<howmany; pos = next) {
180		u64 l64;
181		u64 sz64;
182		u32 raw_sz;
183
184		next = pos+1;
185		res = &dev->resource[pos];
186		res->name = pci_name(dev);
187		reg = PCI_BASE_ADDRESS_0 + (pos << 2);
188		pci_read_config_dword(dev, reg, &l);
189		pci_write_config_dword(dev, reg, ~0);
190		pci_read_config_dword(dev, reg, &sz);
191		pci_write_config_dword(dev, reg, l);
192		if (!sz || sz == 0xffffffff)
193			continue;
194		if (l == 0xffffffff)
195			l = 0;
196		raw_sz = sz;
197		if ((l & PCI_BASE_ADDRESS_SPACE) ==
198				PCI_BASE_ADDRESS_SPACE_MEMORY) {
199			sz = pci_size(l, sz, (u32)PCI_BASE_ADDRESS_MEM_MASK);
200			/*
201			 * For 64bit prefetchable memory sz could be 0, if the
202			 * real size is bigger than 4G, so we need to check
203			 * szhi for that.
204			 */
205			if (!is_64bit_memory(l) && !sz)
206				continue;
207			res->start = l & PCI_BASE_ADDRESS_MEM_MASK;
208			res->flags |= l & ~PCI_BASE_ADDRESS_MEM_MASK;
209		} else {
210			sz = pci_size(l, sz, PCI_BASE_ADDRESS_IO_MASK & 0xffff);
211			if (!sz)
212				continue;
213			res->start = l & PCI_BASE_ADDRESS_IO_MASK;
214			res->flags |= l & ~PCI_BASE_ADDRESS_IO_MASK;
215		}
216		res->end = res->start + (unsigned long) sz;
217		res->flags |= pci_calc_resource_flags(l);
218		if (is_64bit_memory(l)) {
219			u32 szhi, lhi;
220
221			pci_read_config_dword(dev, reg+4, &lhi);
222			pci_write_config_dword(dev, reg+4, ~0);
223			pci_read_config_dword(dev, reg+4, &szhi);
224			pci_write_config_dword(dev, reg+4, lhi);
225			sz64 = ((u64)szhi << 32) | raw_sz;
226			l64 = ((u64)lhi << 32) | l;
227			sz64 = pci_size64(l64, sz64, PCI_BASE_ADDRESS_MEM_MASK);
228			next++;
229#if BITS_PER_LONG == 64
230			if (!sz64) {
231				res->start = 0;
232				res->end = 0;
233				res->flags = 0;
234				continue;
235			}
236			res->start = l64 & PCI_BASE_ADDRESS_MEM_MASK;
237			res->end = res->start + sz64;
238#else
239			if (sz64 > 0x100000000ULL) {
240				printk(KERN_ERR "PCI: Unable to handle 64-bit "
241					"BAR for device %s\n", pci_name(dev));
242				res->start = 0;
243				res->flags = 0;
244			} else if (lhi) {
245				/* 64-bit wide address, treat as disabled */
246				pci_write_config_dword(dev, reg,
247					l & ~(u32)PCI_BASE_ADDRESS_MEM_MASK);
248				pci_write_config_dword(dev, reg+4, 0);
249				res->start = 0;
250				res->end = sz;
251			}
252#endif
253		}
254	}
255	if (rom) {
256		dev->rom_base_reg = rom;
257		res = &dev->resource[PCI_ROM_RESOURCE];
258		res->name = pci_name(dev);
259		pci_read_config_dword(dev, rom, &l);
260		pci_write_config_dword(dev, rom, ~PCI_ROM_ADDRESS_ENABLE);
261		pci_read_config_dword(dev, rom, &sz);
262		pci_write_config_dword(dev, rom, l);
263		if (l == 0xffffffff)
264			l = 0;
265		if (sz && sz != 0xffffffff) {
266			sz = pci_size(l, sz, (u32)PCI_ROM_ADDRESS_MASK);
267			if (sz) {
268				res->flags = (l & IORESOURCE_ROM_ENABLE) |
269				  IORESOURCE_MEM | IORESOURCE_PREFETCH |
270				  IORESOURCE_READONLY | IORESOURCE_CACHEABLE;
271				res->start = l & PCI_ROM_ADDRESS_MASK;
272				res->end = res->start + (unsigned long) sz;
273			}
274		}
275	}
276}
277
278void __devinit pci_read_bridge_bases(struct pci_bus *child)
279{
280	struct pci_dev *dev = child->self;
281	u8 io_base_lo, io_limit_lo;
282	u16 mem_base_lo, mem_limit_lo;
283	unsigned long base, limit;
284	struct resource *res;
285	int i;
286
287	if (!dev)		/* It's a host bus, nothing to read */
288		return;
289
290	if (dev->transparent) {
291		printk(KERN_INFO "PCI: Transparent bridge - %s\n", pci_name(dev));
292		for(i = 3; i < PCI_BUS_NUM_RESOURCES; i++)
293			child->resource[i] = child->parent->resource[i - 3];
294	}
295
296	for(i=0; i<3; i++)
297		child->resource[i] = &dev->resource[PCI_BRIDGE_RESOURCES+i];
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	}
320
321	res = child->resource[1];
322	pci_read_config_word(dev, PCI_MEMORY_BASE, &mem_base_lo);
323	pci_read_config_word(dev, PCI_MEMORY_LIMIT, &mem_limit_lo);
324	base = (mem_base_lo & PCI_MEMORY_RANGE_MASK) << 16;
325	limit = (mem_limit_lo & PCI_MEMORY_RANGE_MASK) << 16;
326	if (base <= limit) {
327		res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM;
328		res->start = base;
329		res->end = limit + 0xfffff;
330	}
331
332	res = child->resource[2];
333	pci_read_config_word(dev, PCI_PREF_MEMORY_BASE, &mem_base_lo);
334	pci_read_config_word(dev, PCI_PREF_MEMORY_LIMIT, &mem_limit_lo);
335	base = (mem_base_lo & PCI_PREF_RANGE_MASK) << 16;
336	limit = (mem_limit_lo & PCI_PREF_RANGE_MASK) << 16;
337
338	if ((mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) {
339		u32 mem_base_hi, mem_limit_hi;
340		pci_read_config_dword(dev, PCI_PREF_BASE_UPPER32, &mem_base_hi);
341		pci_read_config_dword(dev, PCI_PREF_LIMIT_UPPER32, &mem_limit_hi);
342
343		/*
344		 * Some bridges set the base > limit by default, and some
345		 * (broken) BIOSes do not initialize them.  If we find
346		 * this, just assume they are not being used.
347		 */
348		if (mem_base_hi <= mem_limit_hi) {
349#if BITS_PER_LONG == 64
350			base |= ((long) mem_base_hi) << 32;
351			limit |= ((long) mem_limit_hi) << 32;
352#else
353			if (mem_base_hi || mem_limit_hi) {
354				printk(KERN_ERR "PCI: Unable to handle 64-bit address space for bridge %s\n", pci_name(dev));
355				return;
356			}
357#endif
358		}
359	}
360	if (base <= limit) {
361		res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM | IORESOURCE_PREFETCH;
362		res->start = base;
363		res->end = limit + 0xfffff;
364	}
365}
366
367static struct pci_bus * pci_alloc_bus(void)
368{
369	struct pci_bus *b;
370
371	b = kzalloc(sizeof(*b), GFP_KERNEL);
372	if (b) {
373		INIT_LIST_HEAD(&b->node);
374		INIT_LIST_HEAD(&b->children);
375		INIT_LIST_HEAD(&b->devices);
376	}
377	return b;
378}
379
380static struct pci_bus * __devinit
381pci_alloc_child_bus(struct pci_bus *parent, struct pci_dev *bridge, int busnr)
382{
383	struct pci_bus *child;
384	int i;
385	int retval;
386
387	/*
388	 * Allocate a new bus, and inherit stuff from the parent..
389	 */
390	child = pci_alloc_bus();
391	if (!child)
392		return NULL;
393
394	child->self = bridge;
395	child->parent = parent;
396	child->ops = parent->ops;
397	child->sysdata = parent->sysdata;
398	child->bus_flags = parent->bus_flags;
399	child->bridge = get_device(&bridge->dev);
400
401	child->class_dev.class = &pcibus_class;
402	sprintf(child->class_dev.class_id, "%04x:%02x", pci_domain_nr(child), busnr);
403	retval = class_device_register(&child->class_dev);
404	if (retval)
405		goto error_register;
406	retval = class_device_create_file(&child->class_dev,
407					  &class_device_attr_cpuaffinity);
408	if (retval)
409		goto error_file_create;
410
411	/*
412	 * Set up the primary, secondary and subordinate
413	 * bus numbers.
414	 */
415	child->number = child->secondary = busnr;
416	child->primary = parent->secondary;
417	child->subordinate = 0xff;
418
419	/* Set up default resource pointers and names.. */
420	for (i = 0; i < 4; i++) {
421		child->resource[i] = &bridge->resource[PCI_BRIDGE_RESOURCES+i];
422		child->resource[i]->name = child->name;
423	}
424	bridge->subordinate = child;
425
426	return child;
427
428error_file_create:
429	class_device_unregister(&child->class_dev);
430error_register:
431	kfree(child);
432	return NULL;
433}
434
435struct pci_bus *pci_add_new_bus(struct pci_bus *parent, struct pci_dev *dev, int busnr)
436{
437	struct pci_bus *child;
438
439	child = pci_alloc_child_bus(parent, dev, busnr);
440	if (child) {
441		down_write(&pci_bus_sem);
442		list_add_tail(&child->node, &parent->children);
443		up_write(&pci_bus_sem);
444	}
445	return child;
446}
447
448static void pci_enable_crs(struct pci_dev *dev)
449{
450	u16 cap, rpctl;
451	int rpcap = pci_find_capability(dev, PCI_CAP_ID_EXP);
452	if (!rpcap)
453		return;
454
455	pci_read_config_word(dev, rpcap + PCI_CAP_FLAGS, &cap);
456	if (((cap & PCI_EXP_FLAGS_TYPE) >> 4) != PCI_EXP_TYPE_ROOT_PORT)
457		return;
458
459	pci_read_config_word(dev, rpcap + PCI_EXP_RTCTL, &rpctl);
460	rpctl |= PCI_EXP_RTCTL_CRSSVE;
461	pci_write_config_word(dev, rpcap + PCI_EXP_RTCTL, rpctl);
462}
463
464static void pci_fixup_parent_subordinate_busnr(struct pci_bus *child, int max)
465{
466	struct pci_bus *parent = child->parent;
467
468	/* Attempts to fix that up are really dangerous unless
469	   we're going to re-assign all bus numbers. */
470	if (!pcibios_assign_all_busses())
471		return;
472
473	while (parent->parent && parent->subordinate < max) {
474		parent->subordinate = max;
475		pci_write_config_byte(parent->self, PCI_SUBORDINATE_BUS, max);
476		parent = parent->parent;
477	}
478}
479
480unsigned int pci_scan_child_bus(struct pci_bus *bus);
481
482/*
483 * If it's a bridge, configure it and scan the bus behind it.
484 * For CardBus bridges, we don't scan behind as the devices will
485 * be handled by the bridge driver itself.
486 *
487 * We need to process bridges in two passes -- first we scan those
488 * already configured by the BIOS and after we are done with all of
489 * them, we proceed to assigning numbers to the remaining buses in
490 * order to avoid overlaps between old and new bus numbers.
491 */
492int pci_scan_bridge(struct pci_bus *bus, struct pci_dev * dev, int max, int pass)
493{
494	struct pci_bus *child;
495	int is_cardbus = (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS);
496	u32 buses, i, j = 0;
497	u16 bctl;
498
499	pci_read_config_dword(dev, PCI_PRIMARY_BUS, &buses);
500
501	pr_debug("PCI: Scanning behind PCI bridge %s, config %06x, pass %d\n",
502		 pci_name(dev), buses & 0xffffff, pass);
503
504	/* Disable MasterAbortMode during probing to avoid reporting
505	   of bus errors (in some architectures) */
506	pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl);
507	pci_write_config_word(dev, PCI_BRIDGE_CONTROL,
508			      bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT);
509
510	pci_enable_crs(dev);
511
512	if ((buses & 0xffff00) && !pcibios_assign_all_busses() && !is_cardbus) {
513		unsigned int cmax, busnr;
514		/*
515		 * Bus already configured by firmware, process it in the first
516		 * pass and just note the configuration.
517		 */
518		if (pass)
519			goto out;
520		busnr = (buses >> 8) & 0xFF;
521
522		/*
523		 * If we already got to this bus through a different bridge,
524		 * ignore it.  This can happen with the i450NX chipset.
525		 */
526		if (pci_find_bus(pci_domain_nr(bus), busnr)) {
527			printk(KERN_INFO "PCI: Bus %04x:%02x already known\n",
528					pci_domain_nr(bus), busnr);
529			goto out;
530		}
531
532		child = pci_add_new_bus(bus, dev, busnr);
533		if (!child)
534			goto out;
535		child->primary = buses & 0xFF;
536		child->subordinate = (buses >> 16) & 0xFF;
537		child->bridge_ctl = bctl;
538
539		cmax = pci_scan_child_bus(child);
540		if (cmax > max)
541			max = cmax;
542		if (child->subordinate > max)
543			max = child->subordinate;
544	} else {
545		/*
546		 * We need to assign a number to this bus which we always
547		 * do in the second pass.
548		 */
549		if (!pass) {
550			if (pcibios_assign_all_busses())
551				/* Temporarily disable forwarding of the
552				   configuration cycles on all bridges in
553				   this bus segment to avoid possible
554				   conflicts in the second pass between two
555				   bridges programmed with overlapping
556				   bus ranges. */
557				pci_write_config_dword(dev, PCI_PRIMARY_BUS,
558						       buses & ~0xffffff);
559			goto out;
560		}
561
562		/* Clear errors */
563		pci_write_config_word(dev, PCI_STATUS, 0xffff);
564
565		/* Prevent assigning a bus number that already exists.
566		 * This can happen when a bridge is hot-plugged */
567		if (pci_find_bus(pci_domain_nr(bus), max+1))
568			goto out;
569		child = pci_add_new_bus(bus, dev, ++max);
570		buses = (buses & 0xff000000)
571		      | ((unsigned int)(child->primary)     <<  0)
572		      | ((unsigned int)(child->secondary)   <<  8)
573		      | ((unsigned int)(child->subordinate) << 16);
574
575		/*
576		 * yenta.c forces a secondary latency timer of 176.
577		 * Copy that behaviour here.
578		 */
579		if (is_cardbus) {
580			buses &= ~0xff000000;
581			buses |= CARDBUS_LATENCY_TIMER << 24;
582		}
583
584		/*
585		 * We need to blast all three values with a single write.
586		 */
587		pci_write_config_dword(dev, PCI_PRIMARY_BUS, buses);
588
589		if (!is_cardbus) {
590			child->bridge_ctl = bctl | PCI_BRIDGE_CTL_NO_ISA;
591			/*
592			 * Adjust subordinate busnr in parent buses.
593			 * We do this before scanning for children because
594			 * some devices may not be detected if the bios
595			 * was lazy.
596			 */
597			pci_fixup_parent_subordinate_busnr(child, max);
598			/* Now we can scan all subordinate buses... */
599			max = pci_scan_child_bus(child);
600			/*
601			 * now fix it up again since we have found
602			 * the real value of max.
603			 */
604			pci_fixup_parent_subordinate_busnr(child, max);
605		} else {
606			/*
607			 * For CardBus bridges, we leave 4 bus numbers
608			 * as cards with a PCI-to-PCI bridge can be
609			 * inserted later.
610			 */
611			for (i=0; i<CARDBUS_RESERVE_BUSNR; i++) {
612				struct pci_bus *parent = bus;
613				if (pci_find_bus(pci_domain_nr(bus),
614							max+i+1))
615					break;
616				while (parent->parent) {
617					if ((!pcibios_assign_all_busses()) &&
618					    (parent->subordinate > max) &&
619					    (parent->subordinate <= max+i)) {
620						j = 1;
621					}
622					parent = parent->parent;
623				}
624				if (j) {
625					/*
626					 * Often, there are two cardbus bridges
627					 * -- try to leave one valid bus number
628					 * for each one.
629					 */
630					i /= 2;
631					break;
632				}
633			}
634			max += i;
635			pci_fixup_parent_subordinate_busnr(child, max);
636		}
637		/*
638		 * Set the subordinate bus number to its real value.
639		 */
640		child->subordinate = max;
641		pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, max);
642	}
643
644	sprintf(child->name, (is_cardbus ? "PCI CardBus #%02x" : "PCI Bus #%02x"), child->number);
645
646	while (bus->parent) {
647		if ((child->subordinate > bus->subordinate) ||
648		    (child->number > bus->subordinate) ||
649		    (child->number < bus->number) ||
650		    (child->subordinate < bus->number)) {
651			printk(KERN_WARNING "PCI: Bus #%02x (-#%02x) is "
652			       "hidden behind%s bridge #%02x (-#%02x)%s\n",
653			       child->number, child->subordinate,
654			       bus->self->transparent ? " transparent" : " ",
655			       bus->number, bus->subordinate,
656			       pcibios_assign_all_busses() ? " " :
657			       " (try 'pci=assign-busses')");
658			printk(KERN_WARNING "Please report the result to "
659			       "linux-kernel to fix this permanently\n");
660		}
661		bus = bus->parent;
662	}
663
664out:
665	pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl);
666
667	return max;
668}
669
670/*
671 * Read interrupt line and base address registers.
672 * The architecture-dependent code can tweak these, of course.
673 */
674static void pci_read_irq(struct pci_dev *dev)
675{
676	unsigned char irq;
677
678	pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &irq);
679	dev->pin = irq;
680	if (irq)
681		pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq);
682	dev->irq = irq;
683}
684
685#define LEGACY_IO_RESOURCE	(IORESOURCE_IO | IORESOURCE_PCI_FIXED)
686
687/**
688 * pci_setup_device - fill in class and map information of a device
689 * @dev: the device structure to fill
690 *
691 * Initialize the device structure with information about the device's
692 * vendor,class,memory and IO-space addresses,IRQ lines etc.
693 * Called at initialisation of the PCI subsystem and by CardBus services.
694 * Returns 0 on success and -1 if unknown type of device (not normal, bridge
695 * or CardBus).
696 */
697static int pci_setup_device(struct pci_dev * dev)
698{
699	u32 class;
700
701	sprintf(pci_name(dev), "%04x:%02x:%02x.%d", pci_domain_nr(dev->bus),
702		dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn));
703
704	pci_read_config_dword(dev, PCI_CLASS_REVISION, &class);
705	class >>= 8;				    /* upper 3 bytes */
706	dev->class = class;
707	class >>= 8;
708
709	pr_debug("PCI: Found %s [%04x/%04x] %06x %02x\n", pci_name(dev),
710		 dev->vendor, dev->device, class, dev->hdr_type);
711
712	/* "Unknown power state" */
713	dev->current_state = PCI_UNKNOWN;
714
715	/* Early fixups, before probing the BARs */
716	pci_fixup_device(pci_fixup_early, dev);
717	class = dev->class >> 8;
718
719	switch (dev->hdr_type) {		    /* header type */
720	case PCI_HEADER_TYPE_NORMAL:		    /* standard header */
721		if (class == PCI_CLASS_BRIDGE_PCI)
722			goto bad;
723		pci_read_irq(dev);
724		pci_read_bases(dev, 6, PCI_ROM_ADDRESS);
725		pci_read_config_word(dev, PCI_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
726		pci_read_config_word(dev, PCI_SUBSYSTEM_ID, &dev->subsystem_device);
727
728		/*
729		 *	Do the ugly legacy mode stuff here rather than broken chip
730		 *	quirk code. Legacy mode ATA controllers have fixed
731		 *	addresses. These are not always echoed in BAR0-3, and
732		 *	BAR0-3 in a few cases contain junk!
733		 */
734		if (class == PCI_CLASS_STORAGE_IDE) {
735			u8 progif;
736			pci_read_config_byte(dev, PCI_CLASS_PROG, &progif);
737			if ((progif & 1) == 0) {
738				dev->resource[0].start = 0x1F0;
739				dev->resource[0].end = 0x1F7;
740				dev->resource[0].flags = LEGACY_IO_RESOURCE;
741				dev->resource[1].start = 0x3F6;
742				dev->resource[1].end = 0x3F6;
743				dev->resource[1].flags = LEGACY_IO_RESOURCE;
744			}
745			if ((progif & 4) == 0) {
746				dev->resource[2].start = 0x170;
747				dev->resource[2].end = 0x177;
748				dev->resource[2].flags = LEGACY_IO_RESOURCE;
749				dev->resource[3].start = 0x376;
750				dev->resource[3].end = 0x376;
751				dev->resource[3].flags = LEGACY_IO_RESOURCE;
752			}
753		}
754		break;
755
756	case PCI_HEADER_TYPE_BRIDGE:		    /* bridge header */
757		if (class != PCI_CLASS_BRIDGE_PCI)
758			goto bad;
759		/* The PCI-to-PCI bridge spec requires that subtractive
760		   decoding (i.e. transparent) bridge must have programming
761		   interface code of 0x01. */
762		pci_read_irq(dev);
763		dev->transparent = ((dev->class & 0xff) == 1);
764		pci_read_bases(dev, 2, PCI_ROM_ADDRESS1);
765		break;
766
767	case PCI_HEADER_TYPE_CARDBUS:		    /* CardBus bridge header */
768		if (class != PCI_CLASS_BRIDGE_CARDBUS)
769			goto bad;
770		pci_read_irq(dev);
771		pci_read_bases(dev, 1, 0);
772		pci_read_config_word(dev, PCI_CB_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
773		pci_read_config_word(dev, PCI_CB_SUBSYSTEM_ID, &dev->subsystem_device);
774		break;
775
776	default:				    /* unknown header */
777		printk(KERN_ERR "PCI: device %s has unknown header type %02x, ignoring.\n",
778			pci_name(dev), dev->hdr_type);
779		return -1;
780
781	bad:
782		printk(KERN_ERR "PCI: %s: class %x doesn't match header type %02x. Ignoring class.\n",
783		       pci_name(dev), class, dev->hdr_type);
784		dev->class = PCI_CLASS_NOT_DEFINED;
785	}
786
787	/* We found a fine healthy device, go go go... */
788	return 0;
789}
790
791/**
792 * pci_release_dev - free a pci device structure when all users of it are finished.
793 * @dev: device that's been disconnected
794 *
795 * Will be called only by the device core when all users of this pci device are
796 * done.
797 */
798static void pci_release_dev(struct device *dev)
799{
800	struct pci_dev *pci_dev;
801
802	pci_dev = to_pci_dev(dev);
803	kfree(pci_dev);
804}
805
806/**
807 * pci_cfg_space_size - get the configuration space size of the PCI device.
808 * @dev: PCI device
809 *
810 * Regular PCI devices have 256 bytes, but PCI-X 2 and PCI Express devices
811 * have 4096 bytes.  Even if the device is capable, that doesn't mean we can
812 * access it.  Maybe we don't have a way to generate extended config space
813 * accesses, or the device is behind a reverse Express bridge.  So we try
814 * reading the dword at 0x100 which must either be 0 or a valid extended
815 * capability header.
816 */
817int pci_cfg_space_size(struct pci_dev *dev)
818{
819	int pos;
820	u32 status;
821
822	pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
823	if (!pos) {
824		pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
825		if (!pos)
826			goto fail;
827
828		pci_read_config_dword(dev, pos + PCI_X_STATUS, &status);
829		if (!(status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ)))
830			goto fail;
831	}
832
833	if (pci_read_config_dword(dev, 256, &status) != PCIBIOS_SUCCESSFUL)
834		goto fail;
835	if (status == 0xffffffff)
836		goto fail;
837
838	return PCI_CFG_SPACE_EXP_SIZE;
839
840 fail:
841	return PCI_CFG_SPACE_SIZE;
842}
843
844static void pci_release_bus_bridge_dev(struct device *dev)
845{
846	kfree(dev);
847}
848
849struct pci_dev *alloc_pci_dev(void)
850{
851	struct pci_dev *dev;
852
853	dev = kzalloc(sizeof(struct pci_dev), GFP_KERNEL);
854	if (!dev)
855		return NULL;
856
857	INIT_LIST_HEAD(&dev->global_list);
858	INIT_LIST_HEAD(&dev->bus_list);
859
860	pci_msi_init_pci_dev(dev);
861
862	return dev;
863}
864EXPORT_SYMBOL(alloc_pci_dev);
865
866/*
867 * Read the config data for a PCI device, sanity-check it
868 * and fill in the dev structure...
869 */
870static struct pci_dev * __devinit
871pci_scan_device(struct pci_bus *bus, int devfn)
872{
873	struct pci_dev *dev;
874	u32 l;
875	u8 hdr_type;
876	int delay = 1;
877
878	if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, &l))
879		return NULL;
880
881	/* some broken boards return 0 or ~0 if a slot is empty: */
882	if (l == 0xffffffff || l == 0x00000000 ||
883	    l == 0x0000ffff || l == 0xffff0000)
884		return NULL;
885
886	/* Configuration request Retry Status */
887	while (l == 0xffff0001) {
888		msleep(delay);
889		delay *= 2;
890		if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, &l))
891			return NULL;
892		/* Card hasn't responded in 60 seconds?  Must be stuck. */
893		if (delay > 60 * 1000) {
894			printk(KERN_WARNING "Device %04x:%02x:%02x.%d not "
895					"responding\n", pci_domain_nr(bus),
896					bus->number, PCI_SLOT(devfn),
897					PCI_FUNC(devfn));
898			return NULL;
899		}
900	}
901
902	if (pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type))
903		return NULL;
904
905	dev = alloc_pci_dev();
906	if (!dev)
907		return NULL;
908
909	dev->bus = bus;
910	dev->sysdata = bus->sysdata;
911	dev->dev.parent = bus->bridge;
912	dev->dev.bus = &pci_bus_type;
913	dev->devfn = devfn;
914	dev->hdr_type = hdr_type & 0x7f;
915	dev->multifunction = !!(hdr_type & 0x80);
916	dev->vendor = l & 0xffff;
917	dev->device = (l >> 16) & 0xffff;
918	dev->cfg_size = pci_cfg_space_size(dev);
919	dev->error_state = pci_channel_io_normal;
920
921	/* Assume 32-bit PCI; let 64-bit PCI cards (which are far rarer)
922	   set this higher, assuming the system even supports it.  */
923	dev->dma_mask = 0xffffffff;
924	if (pci_setup_device(dev) < 0) {
925		kfree(dev);
926		return NULL;
927	}
928
929	return dev;
930}
931
932void pci_device_add(struct pci_dev *dev, struct pci_bus *bus)
933{
934	device_initialize(&dev->dev);
935	dev->dev.release = pci_release_dev;
936	pci_dev_get(dev);
937
938	set_dev_node(&dev->dev, pcibus_to_node(bus));
939	dev->dev.dma_mask = &dev->dma_mask;
940	dev->dev.coherent_dma_mask = 0xffffffffull;
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
955struct pci_dev *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
968/**
969 * pci_scan_slot - scan a PCI slot on a bus for devices.
970 * @bus: PCI bus to scan
971 * @devfn: slot number to scan (must have zero function.)
972 *
973 * Scan a PCI slot on the specified PCI bus for devices, adding
974 * discovered devices to the @bus->devices list.  New devices
975 * will have an empty dev->global_list head.
976 */
977int pci_scan_slot(struct pci_bus *bus, int devfn)
978{
979	int func, nr = 0;
980	int scan_all_fns;
981
982	scan_all_fns = pcibios_scan_all_fns(bus, devfn);
983
984	for (func = 0; func < 8; func++, devfn++) {
985		struct pci_dev *dev;
986
987		dev = pci_scan_single_device(bus, devfn);
988		if (dev) {
989			nr++;
990
991			/*
992		 	 * If this is a single function device,
993		 	 * don't scan past the first function.
994		 	 */
995			if (!dev->multifunction) {
996				if (func > 0) {
997					dev->multifunction = 1;
998				} else {
999 					break;
1000				}
1001			}
1002		} else {
1003			if (func == 0 && !scan_all_fns)
1004				break;
1005		}
1006	}
1007	return nr;
1008}
1009
1010unsigned int pci_scan_child_bus(struct pci_bus *bus)
1011{
1012	unsigned int devfn, pass, max = bus->secondary;
1013	struct pci_dev *dev;
1014
1015	pr_debug("PCI: Scanning bus %04x:%02x\n", pci_domain_nr(bus), bus->number);
1016
1017	/* Go find them, Rover! */
1018	for (devfn = 0; devfn < 0x100; devfn += 8)
1019		pci_scan_slot(bus, devfn);
1020
1021	/*
1022	 * After performing arch-dependent fixup of the bus, look behind
1023	 * all PCI-to-PCI bridges on this bus.
1024	 */
1025	pr_debug("PCI: Fixups for bus %04x:%02x\n", pci_domain_nr(bus), bus->number);
1026	pcibios_fixup_bus(bus);
1027	for (pass=0; pass < 2; pass++)
1028		list_for_each_entry(dev, &bus->devices, bus_list) {
1029			if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE ||
1030			    dev->hdr_type == PCI_HEADER_TYPE_CARDBUS)
1031				max = pci_scan_bridge(bus, dev, max, pass);
1032		}
1033
1034	/*
1035	 * We've scanned the bus and so we know all about what's on
1036	 * the other side of any bridges that may be on this bus plus
1037	 * any devices.
1038	 *
1039	 * Return how far we've got finding sub-buses.
1040	 */
1041	pr_debug("PCI: Bus scan for %04x:%02x returning with max=%02x\n",
1042		pci_domain_nr(bus), bus->number, max);
1043	return max;
1044}
1045
1046unsigned int __devinit pci_do_scan_bus(struct pci_bus *bus)
1047{
1048	unsigned int max;
1049
1050	max = pci_scan_child_bus(bus);
1051
1052	/*
1053	 * Make the discovered devices available.
1054	 */
1055	pci_bus_add_devices(bus);
1056
1057	return max;
1058}
1059
1060struct pci_bus * pci_create_bus(struct device *parent,
1061		int bus, struct pci_ops *ops, void *sysdata)
1062{
1063	int error;
1064	struct pci_bus *b;
1065	struct device *dev;
1066
1067	b = pci_alloc_bus();
1068	if (!b)
1069		return NULL;
1070
1071	dev = kmalloc(sizeof(*dev), GFP_KERNEL);
1072	if (!dev){
1073		kfree(b);
1074		return NULL;
1075	}
1076
1077	b->sysdata = sysdata;
1078	b->ops = ops;
1079
1080	if (pci_find_bus(pci_domain_nr(b), bus)) {
1081		/* If we already got to this bus through a different bridge, ignore it */
1082		pr_debug("PCI: Bus %04x:%02x already known\n", pci_domain_nr(b), bus);
1083		goto err_out;
1084	}
1085
1086	down_write(&pci_bus_sem);
1087	list_add_tail(&b->node, &pci_root_buses);
1088	up_write(&pci_bus_sem);
1089
1090	memset(dev, 0, sizeof(*dev));
1091	dev->parent = parent;
1092	dev->release = pci_release_bus_bridge_dev;
1093	sprintf(dev->bus_id, "pci%04x:%02x", pci_domain_nr(b), bus);
1094	error = device_register(dev);
1095	if (error)
1096		goto dev_reg_err;
1097	b->bridge = get_device(dev);
1098
1099	b->class_dev.class = &pcibus_class;
1100	sprintf(b->class_dev.class_id, "%04x:%02x", pci_domain_nr(b), bus);
1101	error = class_device_register(&b->class_dev);
1102	if (error)
1103		goto class_dev_reg_err;
1104	error = class_device_create_file(&b->class_dev, &class_device_attr_cpuaffinity);
1105	if (error)
1106		goto class_dev_create_file_err;
1107
1108	/* Create legacy_io and legacy_mem files for this bus */
1109	pci_create_legacy_files(b);
1110
1111	error = sysfs_create_link(&b->class_dev.kobj, &b->bridge->kobj, "bridge");
1112	if (error)
1113		goto sys_create_link_err;
1114
1115	b->number = b->secondary = bus;
1116	b->resource[0] = &ioport_resource;
1117	b->resource[1] = &iomem_resource;
1118
1119	return b;
1120
1121sys_create_link_err:
1122	class_device_remove_file(&b->class_dev, &class_device_attr_cpuaffinity);
1123class_dev_create_file_err:
1124	class_device_unregister(&b->class_dev);
1125class_dev_reg_err:
1126	device_unregister(dev);
1127dev_reg_err:
1128	down_write(&pci_bus_sem);
1129	list_del(&b->node);
1130	up_write(&pci_bus_sem);
1131err_out:
1132	kfree(dev);
1133	kfree(b);
1134	return NULL;
1135}
1136EXPORT_SYMBOL_GPL(pci_create_bus);
1137
1138struct pci_bus *pci_scan_bus_parented(struct device *parent,
1139		int bus, struct pci_ops *ops, void *sysdata)
1140{
1141	struct pci_bus *b;
1142
1143	b = pci_create_bus(parent, bus, ops, sysdata);
1144	if (b)
1145		b->subordinate = pci_scan_child_bus(b);
1146	return b;
1147}
1148EXPORT_SYMBOL(pci_scan_bus_parented);
1149
1150#ifdef CONFIG_HOTPLUG
1151EXPORT_SYMBOL(pci_add_new_bus);
1152EXPORT_SYMBOL(pci_do_scan_bus);
1153EXPORT_SYMBOL(pci_scan_slot);
1154EXPORT_SYMBOL(pci_scan_bridge);
1155EXPORT_SYMBOL(pci_scan_single_device);
1156EXPORT_SYMBOL_GPL(pci_scan_child_bus);
1157#endif
1158
1159static int __init pci_sort_bf_cmp(const struct pci_dev *a, const struct pci_dev *b)
1160{
1161	if      (pci_domain_nr(a->bus) < pci_domain_nr(b->bus)) return -1;
1162	else if (pci_domain_nr(a->bus) > pci_domain_nr(b->bus)) return  1;
1163
1164	if      (a->bus->number < b->bus->number) return -1;
1165	else if (a->bus->number > b->bus->number) return  1;
1166
1167	if      (a->devfn < b->devfn) return -1;
1168	else if (a->devfn > b->devfn) return  1;
1169
1170	return 0;
1171}
1172
1173/*
1174 * Yes, this forcably breaks the klist abstraction temporarily.  It
1175 * just wants to sort the klist, not change reference counts and
1176 * take/drop locks rapidly in the process.  It does all this while
1177 * holding the lock for the list, so objects can't otherwise be
1178 * added/removed while we're swizzling.
1179 */
1180static void __init pci_insertion_sort_klist(struct pci_dev *a, struct list_head *list)
1181{
1182	struct list_head *pos;
1183	struct klist_node *n;
1184	struct device *dev;
1185	struct pci_dev *b;
1186
1187	list_for_each(pos, list) {
1188		n = container_of(pos, struct klist_node, n_node);
1189		dev = container_of(n, struct device, knode_bus);
1190		b = to_pci_dev(dev);
1191		if (pci_sort_bf_cmp(a, b) <= 0) {
1192			list_move_tail(&a->dev.knode_bus.n_node, &b->dev.knode_bus.n_node);
1193			return;
1194		}
1195	}
1196	list_move_tail(&a->dev.knode_bus.n_node, list);
1197}
1198
1199static void __init pci_sort_breadthfirst_klist(void)
1200{
1201	LIST_HEAD(sorted_devices);
1202	struct list_head *pos, *tmp;
1203	struct klist_node *n;
1204	struct device *dev;
1205	struct pci_dev *pdev;
1206
1207	spin_lock(&pci_bus_type.klist_devices.k_lock);
1208	list_for_each_safe(pos, tmp, &pci_bus_type.klist_devices.k_list) {
1209		n = container_of(pos, struct klist_node, n_node);
1210		dev = container_of(n, struct device, knode_bus);
1211		pdev = to_pci_dev(dev);
1212		pci_insertion_sort_klist(pdev, &sorted_devices);
1213	}
1214	list_splice(&sorted_devices, &pci_bus_type.klist_devices.k_list);
1215	spin_unlock(&pci_bus_type.klist_devices.k_lock);
1216}
1217
1218static void __init pci_insertion_sort_devices(struct pci_dev *a, struct list_head *list)
1219{
1220	struct pci_dev *b;
1221
1222	list_for_each_entry(b, list, global_list) {
1223		if (pci_sort_bf_cmp(a, b) <= 0) {
1224			list_move_tail(&a->global_list, &b->global_list);
1225			return;
1226		}
1227	}
1228	list_move_tail(&a->global_list, list);
1229}
1230
1231static void __init pci_sort_breadthfirst_devices(void)
1232{
1233	LIST_HEAD(sorted_devices);
1234	struct pci_dev *dev, *tmp;
1235
1236	down_write(&pci_bus_sem);
1237	list_for_each_entry_safe(dev, tmp, &pci_devices, global_list) {
1238		pci_insertion_sort_devices(dev, &sorted_devices);
1239	}
1240	list_splice(&sorted_devices, &pci_devices);
1241	up_write(&pci_bus_sem);
1242}
1243
1244void __init pci_sort_breadthfirst(void)
1245{
1246	pci_sort_breadthfirst_devices();
1247	pci_sort_breadthfirst_klist();
1248}
1249