xref: /linux/arch/powerpc/kernel/pci-common.c (revision cc4589ebfae6f8dbb5cf880a0a67eedab3416492)
1 /*
2  * Contains common pci routines for ALL ppc platform
3  * (based on pci_32.c and pci_64.c)
4  *
5  * Port for PPC64 David Engebretsen, IBM Corp.
6  * Contains common pci routines for ppc64 platform, pSeries and iSeries brands.
7  *
8  * Copyright (C) 2003 Anton Blanchard <anton@au.ibm.com>, IBM
9  *   Rework, based on alpha PCI code.
10  *
11  * Common pmac/prep/chrp pci routines. -- Cort
12  *
13  * This program is free software; you can redistribute it and/or
14  * modify it under the terms of the GNU General Public License
15  * as published by the Free Software Foundation; either version
16  * 2 of the License, or (at your option) any later version.
17  */
18 
19 #include <linux/kernel.h>
20 #include <linux/pci.h>
21 #include <linux/string.h>
22 #include <linux/init.h>
23 #include <linux/bootmem.h>
24 #include <linux/of_address.h>
25 #include <linux/mm.h>
26 #include <linux/list.h>
27 #include <linux/syscalls.h>
28 #include <linux/irq.h>
29 #include <linux/vmalloc.h>
30 #include <linux/slab.h>
31 
32 #include <asm/processor.h>
33 #include <asm/io.h>
34 #include <asm/prom.h>
35 #include <asm/pci-bridge.h>
36 #include <asm/byteorder.h>
37 #include <asm/machdep.h>
38 #include <asm/ppc-pci.h>
39 #include <asm/firmware.h>
40 #include <asm/eeh.h>
41 
42 static DEFINE_SPINLOCK(hose_spinlock);
43 LIST_HEAD(hose_list);
44 
45 /* XXX kill that some day ... */
46 static int global_phb_number;		/* Global phb counter */
47 
48 /* ISA Memory physical address */
49 resource_size_t isa_mem_base;
50 
51 /* Default PCI flags is 0 on ppc32, modified at boot on ppc64 */
52 unsigned int ppc_pci_flags = 0;
53 
54 
55 static struct dma_map_ops *pci_dma_ops = &dma_direct_ops;
56 
57 void set_pci_dma_ops(struct dma_map_ops *dma_ops)
58 {
59 	pci_dma_ops = dma_ops;
60 }
61 
62 struct dma_map_ops *get_pci_dma_ops(void)
63 {
64 	return pci_dma_ops;
65 }
66 EXPORT_SYMBOL(get_pci_dma_ops);
67 
68 struct pci_controller *pcibios_alloc_controller(struct device_node *dev)
69 {
70 	struct pci_controller *phb;
71 
72 	phb = zalloc_maybe_bootmem(sizeof(struct pci_controller), GFP_KERNEL);
73 	if (phb == NULL)
74 		return NULL;
75 	spin_lock(&hose_spinlock);
76 	phb->global_number = global_phb_number++;
77 	list_add_tail(&phb->list_node, &hose_list);
78 	spin_unlock(&hose_spinlock);
79 	phb->dn = dev;
80 	phb->is_dynamic = mem_init_done;
81 #ifdef CONFIG_PPC64
82 	if (dev) {
83 		int nid = of_node_to_nid(dev);
84 
85 		if (nid < 0 || !node_online(nid))
86 			nid = -1;
87 
88 		PHB_SET_NODE(phb, nid);
89 	}
90 #endif
91 	return phb;
92 }
93 
94 void pcibios_free_controller(struct pci_controller *phb)
95 {
96 	spin_lock(&hose_spinlock);
97 	list_del(&phb->list_node);
98 	spin_unlock(&hose_spinlock);
99 
100 	if (phb->is_dynamic)
101 		kfree(phb);
102 }
103 
104 static resource_size_t pcibios_io_size(const struct pci_controller *hose)
105 {
106 #ifdef CONFIG_PPC64
107 	return hose->pci_io_size;
108 #else
109 	return hose->io_resource.end - hose->io_resource.start + 1;
110 #endif
111 }
112 
113 int pcibios_vaddr_is_ioport(void __iomem *address)
114 {
115 	int ret = 0;
116 	struct pci_controller *hose;
117 	resource_size_t size;
118 
119 	spin_lock(&hose_spinlock);
120 	list_for_each_entry(hose, &hose_list, list_node) {
121 		size = pcibios_io_size(hose);
122 		if (address >= hose->io_base_virt &&
123 		    address < (hose->io_base_virt + size)) {
124 			ret = 1;
125 			break;
126 		}
127 	}
128 	spin_unlock(&hose_spinlock);
129 	return ret;
130 }
131 
132 unsigned long pci_address_to_pio(phys_addr_t address)
133 {
134 	struct pci_controller *hose;
135 	resource_size_t size;
136 	unsigned long ret = ~0;
137 
138 	spin_lock(&hose_spinlock);
139 	list_for_each_entry(hose, &hose_list, list_node) {
140 		size = pcibios_io_size(hose);
141 		if (address >= hose->io_base_phys &&
142 		    address < (hose->io_base_phys + size)) {
143 			unsigned long base =
144 				(unsigned long)hose->io_base_virt - _IO_BASE;
145 			ret = base + (address - hose->io_base_phys);
146 			break;
147 		}
148 	}
149 	spin_unlock(&hose_spinlock);
150 
151 	return ret;
152 }
153 EXPORT_SYMBOL_GPL(pci_address_to_pio);
154 
155 /*
156  * Return the domain number for this bus.
157  */
158 int pci_domain_nr(struct pci_bus *bus)
159 {
160 	struct pci_controller *hose = pci_bus_to_host(bus);
161 
162 	return hose->global_number;
163 }
164 EXPORT_SYMBOL(pci_domain_nr);
165 
166 /* This routine is meant to be used early during boot, when the
167  * PCI bus numbers have not yet been assigned, and you need to
168  * issue PCI config cycles to an OF device.
169  * It could also be used to "fix" RTAS config cycles if you want
170  * to set pci_assign_all_buses to 1 and still use RTAS for PCI
171  * config cycles.
172  */
173 struct pci_controller* pci_find_hose_for_OF_device(struct device_node* node)
174 {
175 	while(node) {
176 		struct pci_controller *hose, *tmp;
177 		list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
178 			if (hose->dn == node)
179 				return hose;
180 		node = node->parent;
181 	}
182 	return NULL;
183 }
184 
185 static ssize_t pci_show_devspec(struct device *dev,
186 		struct device_attribute *attr, char *buf)
187 {
188 	struct pci_dev *pdev;
189 	struct device_node *np;
190 
191 	pdev = to_pci_dev (dev);
192 	np = pci_device_to_OF_node(pdev);
193 	if (np == NULL || np->full_name == NULL)
194 		return 0;
195 	return sprintf(buf, "%s", np->full_name);
196 }
197 static DEVICE_ATTR(devspec, S_IRUGO, pci_show_devspec, NULL);
198 
199 /* Add sysfs properties */
200 int pcibios_add_platform_entries(struct pci_dev *pdev)
201 {
202 	return device_create_file(&pdev->dev, &dev_attr_devspec);
203 }
204 
205 char __devinit *pcibios_setup(char *str)
206 {
207 	return str;
208 }
209 
210 /*
211  * Reads the interrupt pin to determine if interrupt is use by card.
212  * If the interrupt is used, then gets the interrupt line from the
213  * openfirmware and sets it in the pci_dev and pci_config line.
214  */
215 int pci_read_irq_line(struct pci_dev *pci_dev)
216 {
217 	struct of_irq oirq;
218 	unsigned int virq;
219 
220 	/* The current device-tree that iSeries generates from the HV
221 	 * PCI informations doesn't contain proper interrupt routing,
222 	 * and all the fallback would do is print out crap, so we
223 	 * don't attempt to resolve the interrupts here at all, some
224 	 * iSeries specific fixup does it.
225 	 *
226 	 * In the long run, we will hopefully fix the generated device-tree
227 	 * instead.
228 	 */
229 #ifdef CONFIG_PPC_ISERIES
230 	if (firmware_has_feature(FW_FEATURE_ISERIES))
231 		return -1;
232 #endif
233 
234 	pr_debug("PCI: Try to map irq for %s...\n", pci_name(pci_dev));
235 
236 #ifdef DEBUG
237 	memset(&oirq, 0xff, sizeof(oirq));
238 #endif
239 	/* Try to get a mapping from the device-tree */
240 	if (of_irq_map_pci(pci_dev, &oirq)) {
241 		u8 line, pin;
242 
243 		/* If that fails, lets fallback to what is in the config
244 		 * space and map that through the default controller. We
245 		 * also set the type to level low since that's what PCI
246 		 * interrupts are. If your platform does differently, then
247 		 * either provide a proper interrupt tree or don't use this
248 		 * function.
249 		 */
250 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_PIN, &pin))
251 			return -1;
252 		if (pin == 0)
253 			return -1;
254 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_LINE, &line) ||
255 		    line == 0xff || line == 0) {
256 			return -1;
257 		}
258 		pr_debug(" No map ! Using line %d (pin %d) from PCI config\n",
259 			 line, pin);
260 
261 		virq = irq_create_mapping(NULL, line);
262 		if (virq != NO_IRQ)
263 			set_irq_type(virq, IRQ_TYPE_LEVEL_LOW);
264 	} else {
265 		pr_debug(" Got one, spec %d cells (0x%08x 0x%08x...) on %s\n",
266 			 oirq.size, oirq.specifier[0], oirq.specifier[1],
267 			 oirq.controller ? oirq.controller->full_name :
268 			 "<default>");
269 
270 		virq = irq_create_of_mapping(oirq.controller, oirq.specifier,
271 					     oirq.size);
272 	}
273 	if(virq == NO_IRQ) {
274 		pr_debug(" Failed to map !\n");
275 		return -1;
276 	}
277 
278 	pr_debug(" Mapped to linux irq %d\n", virq);
279 
280 	pci_dev->irq = virq;
281 
282 	return 0;
283 }
284 EXPORT_SYMBOL(pci_read_irq_line);
285 
286 /*
287  * Platform support for /proc/bus/pci/X/Y mmap()s,
288  * modelled on the sparc64 implementation by Dave Miller.
289  *  -- paulus.
290  */
291 
292 /*
293  * Adjust vm_pgoff of VMA such that it is the physical page offset
294  * corresponding to the 32-bit pci bus offset for DEV requested by the user.
295  *
296  * Basically, the user finds the base address for his device which he wishes
297  * to mmap.  They read the 32-bit value from the config space base register,
298  * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
299  * offset parameter of mmap on /proc/bus/pci/XXX for that device.
300  *
301  * Returns negative error code on failure, zero on success.
302  */
303 static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
304 					       resource_size_t *offset,
305 					       enum pci_mmap_state mmap_state)
306 {
307 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
308 	unsigned long io_offset = 0;
309 	int i, res_bit;
310 
311 	if (hose == 0)
312 		return NULL;		/* should never happen */
313 
314 	/* If memory, add on the PCI bridge address offset */
315 	if (mmap_state == pci_mmap_mem) {
316 #if 0 /* See comment in pci_resource_to_user() for why this is disabled */
317 		*offset += hose->pci_mem_offset;
318 #endif
319 		res_bit = IORESOURCE_MEM;
320 	} else {
321 		io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
322 		*offset += io_offset;
323 		res_bit = IORESOURCE_IO;
324 	}
325 
326 	/*
327 	 * Check that the offset requested corresponds to one of the
328 	 * resources of the device.
329 	 */
330 	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
331 		struct resource *rp = &dev->resource[i];
332 		int flags = rp->flags;
333 
334 		/* treat ROM as memory (should be already) */
335 		if (i == PCI_ROM_RESOURCE)
336 			flags |= IORESOURCE_MEM;
337 
338 		/* Active and same type? */
339 		if ((flags & res_bit) == 0)
340 			continue;
341 
342 		/* In the range of this resource? */
343 		if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
344 			continue;
345 
346 		/* found it! construct the final physical address */
347 		if (mmap_state == pci_mmap_io)
348 			*offset += hose->io_base_phys - io_offset;
349 		return rp;
350 	}
351 
352 	return NULL;
353 }
354 
355 /*
356  * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
357  * device mapping.
358  */
359 static pgprot_t __pci_mmap_set_pgprot(struct pci_dev *dev, struct resource *rp,
360 				      pgprot_t protection,
361 				      enum pci_mmap_state mmap_state,
362 				      int write_combine)
363 {
364 	unsigned long prot = pgprot_val(protection);
365 
366 	/* Write combine is always 0 on non-memory space mappings. On
367 	 * memory space, if the user didn't pass 1, we check for a
368 	 * "prefetchable" resource. This is a bit hackish, but we use
369 	 * this to workaround the inability of /sysfs to provide a write
370 	 * combine bit
371 	 */
372 	if (mmap_state != pci_mmap_mem)
373 		write_combine = 0;
374 	else if (write_combine == 0) {
375 		if (rp->flags & IORESOURCE_PREFETCH)
376 			write_combine = 1;
377 	}
378 
379 	/* XXX would be nice to have a way to ask for write-through */
380 	if (write_combine)
381 		return pgprot_noncached_wc(prot);
382 	else
383 		return pgprot_noncached(prot);
384 }
385 
386 /*
387  * This one is used by /dev/mem and fbdev who have no clue about the
388  * PCI device, it tries to find the PCI device first and calls the
389  * above routine
390  */
391 pgprot_t pci_phys_mem_access_prot(struct file *file,
392 				  unsigned long pfn,
393 				  unsigned long size,
394 				  pgprot_t prot)
395 {
396 	struct pci_dev *pdev = NULL;
397 	struct resource *found = NULL;
398 	resource_size_t offset = ((resource_size_t)pfn) << PAGE_SHIFT;
399 	int i;
400 
401 	if (page_is_ram(pfn))
402 		return prot;
403 
404 	prot = pgprot_noncached(prot);
405 	for_each_pci_dev(pdev) {
406 		for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
407 			struct resource *rp = &pdev->resource[i];
408 			int flags = rp->flags;
409 
410 			/* Active and same type? */
411 			if ((flags & IORESOURCE_MEM) == 0)
412 				continue;
413 			/* In the range of this resource? */
414 			if (offset < (rp->start & PAGE_MASK) ||
415 			    offset > rp->end)
416 				continue;
417 			found = rp;
418 			break;
419 		}
420 		if (found)
421 			break;
422 	}
423 	if (found) {
424 		if (found->flags & IORESOURCE_PREFETCH)
425 			prot = pgprot_noncached_wc(prot);
426 		pci_dev_put(pdev);
427 	}
428 
429 	pr_debug("PCI: Non-PCI map for %llx, prot: %lx\n",
430 		 (unsigned long long)offset, pgprot_val(prot));
431 
432 	return prot;
433 }
434 
435 
436 /*
437  * Perform the actual remap of the pages for a PCI device mapping, as
438  * appropriate for this architecture.  The region in the process to map
439  * is described by vm_start and vm_end members of VMA, the base physical
440  * address is found in vm_pgoff.
441  * The pci device structure is provided so that architectures may make mapping
442  * decisions on a per-device or per-bus basis.
443  *
444  * Returns a negative error code on failure, zero on success.
445  */
446 int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
447 			enum pci_mmap_state mmap_state, int write_combine)
448 {
449 	resource_size_t offset =
450 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
451 	struct resource *rp;
452 	int ret;
453 
454 	rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
455 	if (rp == NULL)
456 		return -EINVAL;
457 
458 	vma->vm_pgoff = offset >> PAGE_SHIFT;
459 	vma->vm_page_prot = __pci_mmap_set_pgprot(dev, rp,
460 						  vma->vm_page_prot,
461 						  mmap_state, write_combine);
462 
463 	ret = remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
464 			       vma->vm_end - vma->vm_start, vma->vm_page_prot);
465 
466 	return ret;
467 }
468 
469 /* This provides legacy IO read access on a bus */
470 int pci_legacy_read(struct pci_bus *bus, loff_t port, u32 *val, size_t size)
471 {
472 	unsigned long offset;
473 	struct pci_controller *hose = pci_bus_to_host(bus);
474 	struct resource *rp = &hose->io_resource;
475 	void __iomem *addr;
476 
477 	/* Check if port can be supported by that bus. We only check
478 	 * the ranges of the PHB though, not the bus itself as the rules
479 	 * for forwarding legacy cycles down bridges are not our problem
480 	 * here. So if the host bridge supports it, we do it.
481 	 */
482 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
483 	offset += port;
484 
485 	if (!(rp->flags & IORESOURCE_IO))
486 		return -ENXIO;
487 	if (offset < rp->start || (offset + size) > rp->end)
488 		return -ENXIO;
489 	addr = hose->io_base_virt + port;
490 
491 	switch(size) {
492 	case 1:
493 		*((u8 *)val) = in_8(addr);
494 		return 1;
495 	case 2:
496 		if (port & 1)
497 			return -EINVAL;
498 		*((u16 *)val) = in_le16(addr);
499 		return 2;
500 	case 4:
501 		if (port & 3)
502 			return -EINVAL;
503 		*((u32 *)val) = in_le32(addr);
504 		return 4;
505 	}
506 	return -EINVAL;
507 }
508 
509 /* This provides legacy IO write access on a bus */
510 int pci_legacy_write(struct pci_bus *bus, loff_t port, u32 val, size_t size)
511 {
512 	unsigned long offset;
513 	struct pci_controller *hose = pci_bus_to_host(bus);
514 	struct resource *rp = &hose->io_resource;
515 	void __iomem *addr;
516 
517 	/* Check if port can be supported by that bus. We only check
518 	 * the ranges of the PHB though, not the bus itself as the rules
519 	 * for forwarding legacy cycles down bridges are not our problem
520 	 * here. So if the host bridge supports it, we do it.
521 	 */
522 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
523 	offset += port;
524 
525 	if (!(rp->flags & IORESOURCE_IO))
526 		return -ENXIO;
527 	if (offset < rp->start || (offset + size) > rp->end)
528 		return -ENXIO;
529 	addr = hose->io_base_virt + port;
530 
531 	/* WARNING: The generic code is idiotic. It gets passed a pointer
532 	 * to what can be a 1, 2 or 4 byte quantity and always reads that
533 	 * as a u32, which means that we have to correct the location of
534 	 * the data read within those 32 bits for size 1 and 2
535 	 */
536 	switch(size) {
537 	case 1:
538 		out_8(addr, val >> 24);
539 		return 1;
540 	case 2:
541 		if (port & 1)
542 			return -EINVAL;
543 		out_le16(addr, val >> 16);
544 		return 2;
545 	case 4:
546 		if (port & 3)
547 			return -EINVAL;
548 		out_le32(addr, val);
549 		return 4;
550 	}
551 	return -EINVAL;
552 }
553 
554 /* This provides legacy IO or memory mmap access on a bus */
555 int pci_mmap_legacy_page_range(struct pci_bus *bus,
556 			       struct vm_area_struct *vma,
557 			       enum pci_mmap_state mmap_state)
558 {
559 	struct pci_controller *hose = pci_bus_to_host(bus);
560 	resource_size_t offset =
561 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
562 	resource_size_t size = vma->vm_end - vma->vm_start;
563 	struct resource *rp;
564 
565 	pr_debug("pci_mmap_legacy_page_range(%04x:%02x, %s @%llx..%llx)\n",
566 		 pci_domain_nr(bus), bus->number,
567 		 mmap_state == pci_mmap_mem ? "MEM" : "IO",
568 		 (unsigned long long)offset,
569 		 (unsigned long long)(offset + size - 1));
570 
571 	if (mmap_state == pci_mmap_mem) {
572 		/* Hack alert !
573 		 *
574 		 * Because X is lame and can fail starting if it gets an error trying
575 		 * to mmap legacy_mem (instead of just moving on without legacy memory
576 		 * access) we fake it here by giving it anonymous memory, effectively
577 		 * behaving just like /dev/zero
578 		 */
579 		if ((offset + size) > hose->isa_mem_size) {
580 			printk(KERN_DEBUG
581 			       "Process %s (pid:%d) mapped non-existing PCI legacy memory for 0%04x:%02x\n",
582 			       current->comm, current->pid, pci_domain_nr(bus), bus->number);
583 			if (vma->vm_flags & VM_SHARED)
584 				return shmem_zero_setup(vma);
585 			return 0;
586 		}
587 		offset += hose->isa_mem_phys;
588 	} else {
589 		unsigned long io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
590 		unsigned long roffset = offset + io_offset;
591 		rp = &hose->io_resource;
592 		if (!(rp->flags & IORESOURCE_IO))
593 			return -ENXIO;
594 		if (roffset < rp->start || (roffset + size) > rp->end)
595 			return -ENXIO;
596 		offset += hose->io_base_phys;
597 	}
598 	pr_debug(" -> mapping phys %llx\n", (unsigned long long)offset);
599 
600 	vma->vm_pgoff = offset >> PAGE_SHIFT;
601 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
602 	return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
603 			       vma->vm_end - vma->vm_start,
604 			       vma->vm_page_prot);
605 }
606 
607 void pci_resource_to_user(const struct pci_dev *dev, int bar,
608 			  const struct resource *rsrc,
609 			  resource_size_t *start, resource_size_t *end)
610 {
611 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
612 	resource_size_t offset = 0;
613 
614 	if (hose == NULL)
615 		return;
616 
617 	if (rsrc->flags & IORESOURCE_IO)
618 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
619 
620 	/* We pass a fully fixed up address to userland for MMIO instead of
621 	 * a BAR value because X is lame and expects to be able to use that
622 	 * to pass to /dev/mem !
623 	 *
624 	 * That means that we'll have potentially 64 bits values where some
625 	 * userland apps only expect 32 (like X itself since it thinks only
626 	 * Sparc has 64 bits MMIO) but if we don't do that, we break it on
627 	 * 32 bits CHRPs :-(
628 	 *
629 	 * Hopefully, the sysfs insterface is immune to that gunk. Once X
630 	 * has been fixed (and the fix spread enough), we can re-enable the
631 	 * 2 lines below and pass down a BAR value to userland. In that case
632 	 * we'll also have to re-enable the matching code in
633 	 * __pci_mmap_make_offset().
634 	 *
635 	 * BenH.
636 	 */
637 #if 0
638 	else if (rsrc->flags & IORESOURCE_MEM)
639 		offset = hose->pci_mem_offset;
640 #endif
641 
642 	*start = rsrc->start - offset;
643 	*end = rsrc->end - offset;
644 }
645 
646 /**
647  * pci_process_bridge_OF_ranges - Parse PCI bridge resources from device tree
648  * @hose: newly allocated pci_controller to be setup
649  * @dev: device node of the host bridge
650  * @primary: set if primary bus (32 bits only, soon to be deprecated)
651  *
652  * This function will parse the "ranges" property of a PCI host bridge device
653  * node and setup the resource mapping of a pci controller based on its
654  * content.
655  *
656  * Life would be boring if it wasn't for a few issues that we have to deal
657  * with here:
658  *
659  *   - We can only cope with one IO space range and up to 3 Memory space
660  *     ranges. However, some machines (thanks Apple !) tend to split their
661  *     space into lots of small contiguous ranges. So we have to coalesce.
662  *
663  *   - We can only cope with all memory ranges having the same offset
664  *     between CPU addresses and PCI addresses. Unfortunately, some bridges
665  *     are setup for a large 1:1 mapping along with a small "window" which
666  *     maps PCI address 0 to some arbitrary high address of the CPU space in
667  *     order to give access to the ISA memory hole.
668  *     The way out of here that I've chosen for now is to always set the
669  *     offset based on the first resource found, then override it if we
670  *     have a different offset and the previous was set by an ISA hole.
671  *
672  *   - Some busses have IO space not starting at 0, which causes trouble with
673  *     the way we do our IO resource renumbering. The code somewhat deals with
674  *     it for 64 bits but I would expect problems on 32 bits.
675  *
676  *   - Some 32 bits platforms such as 4xx can have physical space larger than
677  *     32 bits so we need to use 64 bits values for the parsing
678  */
679 void __devinit pci_process_bridge_OF_ranges(struct pci_controller *hose,
680 					    struct device_node *dev,
681 					    int primary)
682 {
683 	const u32 *ranges;
684 	int rlen;
685 	int pna = of_n_addr_cells(dev);
686 	int np = pna + 5;
687 	int memno = 0, isa_hole = -1;
688 	u32 pci_space;
689 	unsigned long long pci_addr, cpu_addr, pci_next, cpu_next, size;
690 	unsigned long long isa_mb = 0;
691 	struct resource *res;
692 
693 	printk(KERN_INFO "PCI host bridge %s %s ranges:\n",
694 	       dev->full_name, primary ? "(primary)" : "");
695 
696 	/* Get ranges property */
697 	ranges = of_get_property(dev, "ranges", &rlen);
698 	if (ranges == NULL)
699 		return;
700 
701 	/* Parse it */
702 	while ((rlen -= np * 4) >= 0) {
703 		/* Read next ranges element */
704 		pci_space = ranges[0];
705 		pci_addr = of_read_number(ranges + 1, 2);
706 		cpu_addr = of_translate_address(dev, ranges + 3);
707 		size = of_read_number(ranges + pna + 3, 2);
708 		ranges += np;
709 
710 		/* If we failed translation or got a zero-sized region
711 		 * (some FW try to feed us with non sensical zero sized regions
712 		 * such as power3 which look like some kind of attempt at exposing
713 		 * the VGA memory hole)
714 		 */
715 		if (cpu_addr == OF_BAD_ADDR || size == 0)
716 			continue;
717 
718 		/* Now consume following elements while they are contiguous */
719 		for (; rlen >= np * sizeof(u32);
720 		     ranges += np, rlen -= np * 4) {
721 			if (ranges[0] != pci_space)
722 				break;
723 			pci_next = of_read_number(ranges + 1, 2);
724 			cpu_next = of_translate_address(dev, ranges + 3);
725 			if (pci_next != pci_addr + size ||
726 			    cpu_next != cpu_addr + size)
727 				break;
728 			size += of_read_number(ranges + pna + 3, 2);
729 		}
730 
731 		/* Act based on address space type */
732 		res = NULL;
733 		switch ((pci_space >> 24) & 0x3) {
734 		case 1:		/* PCI IO space */
735 			printk(KERN_INFO
736 			       "  IO 0x%016llx..0x%016llx -> 0x%016llx\n",
737 			       cpu_addr, cpu_addr + size - 1, pci_addr);
738 
739 			/* We support only one IO range */
740 			if (hose->pci_io_size) {
741 				printk(KERN_INFO
742 				       " \\--> Skipped (too many) !\n");
743 				continue;
744 			}
745 #ifdef CONFIG_PPC32
746 			/* On 32 bits, limit I/O space to 16MB */
747 			if (size > 0x01000000)
748 				size = 0x01000000;
749 
750 			/* 32 bits needs to map IOs here */
751 			hose->io_base_virt = ioremap(cpu_addr, size);
752 
753 			/* Expect trouble if pci_addr is not 0 */
754 			if (primary)
755 				isa_io_base =
756 					(unsigned long)hose->io_base_virt;
757 #endif /* CONFIG_PPC32 */
758 			/* pci_io_size and io_base_phys always represent IO
759 			 * space starting at 0 so we factor in pci_addr
760 			 */
761 			hose->pci_io_size = pci_addr + size;
762 			hose->io_base_phys = cpu_addr - pci_addr;
763 
764 			/* Build resource */
765 			res = &hose->io_resource;
766 			res->flags = IORESOURCE_IO;
767 			res->start = pci_addr;
768 			break;
769 		case 2:		/* PCI Memory space */
770 		case 3:		/* PCI 64 bits Memory space */
771 			printk(KERN_INFO
772 			       " MEM 0x%016llx..0x%016llx -> 0x%016llx %s\n",
773 			       cpu_addr, cpu_addr + size - 1, pci_addr,
774 			       (pci_space & 0x40000000) ? "Prefetch" : "");
775 
776 			/* We support only 3 memory ranges */
777 			if (memno >= 3) {
778 				printk(KERN_INFO
779 				       " \\--> Skipped (too many) !\n");
780 				continue;
781 			}
782 			/* Handles ISA memory hole space here */
783 			if (pci_addr == 0) {
784 				isa_mb = cpu_addr;
785 				isa_hole = memno;
786 				if (primary || isa_mem_base == 0)
787 					isa_mem_base = cpu_addr;
788 				hose->isa_mem_phys = cpu_addr;
789 				hose->isa_mem_size = size;
790 			}
791 
792 			/* We get the PCI/Mem offset from the first range or
793 			 * the, current one if the offset came from an ISA
794 			 * hole. If they don't match, bugger.
795 			 */
796 			if (memno == 0 ||
797 			    (isa_hole >= 0 && pci_addr != 0 &&
798 			     hose->pci_mem_offset == isa_mb))
799 				hose->pci_mem_offset = cpu_addr - pci_addr;
800 			else if (pci_addr != 0 &&
801 				 hose->pci_mem_offset != cpu_addr - pci_addr) {
802 				printk(KERN_INFO
803 				       " \\--> Skipped (offset mismatch) !\n");
804 				continue;
805 			}
806 
807 			/* Build resource */
808 			res = &hose->mem_resources[memno++];
809 			res->flags = IORESOURCE_MEM;
810 			if (pci_space & 0x40000000)
811 				res->flags |= IORESOURCE_PREFETCH;
812 			res->start = cpu_addr;
813 			break;
814 		}
815 		if (res != NULL) {
816 			res->name = dev->full_name;
817 			res->end = res->start + size - 1;
818 			res->parent = NULL;
819 			res->sibling = NULL;
820 			res->child = NULL;
821 		}
822 	}
823 
824 	/* If there's an ISA hole and the pci_mem_offset is -not- matching
825 	 * the ISA hole offset, then we need to remove the ISA hole from
826 	 * the resource list for that brige
827 	 */
828 	if (isa_hole >= 0 && hose->pci_mem_offset != isa_mb) {
829 		unsigned int next = isa_hole + 1;
830 		printk(KERN_INFO " Removing ISA hole at 0x%016llx\n", isa_mb);
831 		if (next < memno)
832 			memmove(&hose->mem_resources[isa_hole],
833 				&hose->mem_resources[next],
834 				sizeof(struct resource) * (memno - next));
835 		hose->mem_resources[--memno].flags = 0;
836 	}
837 }
838 
839 /* Decide whether to display the domain number in /proc */
840 int pci_proc_domain(struct pci_bus *bus)
841 {
842 	struct pci_controller *hose = pci_bus_to_host(bus);
843 
844 	if (!(ppc_pci_flags & PPC_PCI_ENABLE_PROC_DOMAINS))
845 		return 0;
846 	if (ppc_pci_flags & PPC_PCI_COMPAT_DOMAIN_0)
847 		return hose->global_number != 0;
848 	return 1;
849 }
850 
851 void pcibios_resource_to_bus(struct pci_dev *dev, struct pci_bus_region *region,
852 			     struct resource *res)
853 {
854 	resource_size_t offset = 0, mask = (resource_size_t)-1;
855 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
856 
857 	if (!hose)
858 		return;
859 	if (res->flags & IORESOURCE_IO) {
860 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
861 		mask = 0xffffffffu;
862 	} else if (res->flags & IORESOURCE_MEM)
863 		offset = hose->pci_mem_offset;
864 
865 	region->start = (res->start - offset) & mask;
866 	region->end = (res->end - offset) & mask;
867 }
868 EXPORT_SYMBOL(pcibios_resource_to_bus);
869 
870 void pcibios_bus_to_resource(struct pci_dev *dev, struct resource *res,
871 			     struct pci_bus_region *region)
872 {
873 	resource_size_t offset = 0, mask = (resource_size_t)-1;
874 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
875 
876 	if (!hose)
877 		return;
878 	if (res->flags & IORESOURCE_IO) {
879 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
880 		mask = 0xffffffffu;
881 	} else if (res->flags & IORESOURCE_MEM)
882 		offset = hose->pci_mem_offset;
883 	res->start = (region->start + offset) & mask;
884 	res->end = (region->end + offset) & mask;
885 }
886 EXPORT_SYMBOL(pcibios_bus_to_resource);
887 
888 /* Fixup a bus resource into a linux resource */
889 static void __devinit fixup_resource(struct resource *res, struct pci_dev *dev)
890 {
891 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
892 	resource_size_t offset = 0, mask = (resource_size_t)-1;
893 
894 	if (res->flags & IORESOURCE_IO) {
895 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
896 		mask = 0xffffffffu;
897 	} else if (res->flags & IORESOURCE_MEM)
898 		offset = hose->pci_mem_offset;
899 
900 	res->start = (res->start + offset) & mask;
901 	res->end = (res->end + offset) & mask;
902 }
903 
904 
905 /* This header fixup will do the resource fixup for all devices as they are
906  * probed, but not for bridge ranges
907  */
908 static void __devinit pcibios_fixup_resources(struct pci_dev *dev)
909 {
910 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
911 	int i;
912 
913 	if (!hose) {
914 		printk(KERN_ERR "No host bridge for PCI dev %s !\n",
915 		       pci_name(dev));
916 		return;
917 	}
918 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
919 		struct resource *res = dev->resource + i;
920 		if (!res->flags)
921 			continue;
922 		/* On platforms that have PPC_PCI_PROBE_ONLY set, we don't
923 		 * consider 0 as an unassigned BAR value. It's technically
924 		 * a valid value, but linux doesn't like it... so when we can
925 		 * re-assign things, we do so, but if we can't, we keep it
926 		 * around and hope for the best...
927 		 */
928 		if (res->start == 0 && !(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
929 			pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] is unassigned\n",
930 				 pci_name(dev), i,
931 				 (unsigned long long)res->start,
932 				 (unsigned long long)res->end,
933 				 (unsigned int)res->flags);
934 			res->end -= res->start;
935 			res->start = 0;
936 			res->flags |= IORESOURCE_UNSET;
937 			continue;
938 		}
939 
940 		pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] fixup...\n",
941 			 pci_name(dev), i,
942 			 (unsigned long long)res->start,\
943 			 (unsigned long long)res->end,
944 			 (unsigned int)res->flags);
945 
946 		fixup_resource(res, dev);
947 
948 		pr_debug("PCI:%s            %016llx-%016llx\n",
949 			 pci_name(dev),
950 			 (unsigned long long)res->start,
951 			 (unsigned long long)res->end);
952 	}
953 
954 	/* Call machine specific resource fixup */
955 	if (ppc_md.pcibios_fixup_resources)
956 		ppc_md.pcibios_fixup_resources(dev);
957 }
958 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_resources);
959 
960 /* This function tries to figure out if a bridge resource has been initialized
961  * by the firmware or not. It doesn't have to be absolutely bullet proof, but
962  * things go more smoothly when it gets it right. It should covers cases such
963  * as Apple "closed" bridge resources and bare-metal pSeries unassigned bridges
964  */
965 static int __devinit pcibios_uninitialized_bridge_resource(struct pci_bus *bus,
966 							   struct resource *res)
967 {
968 	struct pci_controller *hose = pci_bus_to_host(bus);
969 	struct pci_dev *dev = bus->self;
970 	resource_size_t offset;
971 	u16 command;
972 	int i;
973 
974 	/* We don't do anything if PCI_PROBE_ONLY is set */
975 	if (ppc_pci_flags & PPC_PCI_PROBE_ONLY)
976 		return 0;
977 
978 	/* Job is a bit different between memory and IO */
979 	if (res->flags & IORESOURCE_MEM) {
980 		/* If the BAR is non-0 (res != pci_mem_offset) then it's probably been
981 		 * initialized by somebody
982 		 */
983 		if (res->start != hose->pci_mem_offset)
984 			return 0;
985 
986 		/* The BAR is 0, let's check if memory decoding is enabled on
987 		 * the bridge. If not, we consider it unassigned
988 		 */
989 		pci_read_config_word(dev, PCI_COMMAND, &command);
990 		if ((command & PCI_COMMAND_MEMORY) == 0)
991 			return 1;
992 
993 		/* Memory decoding is enabled and the BAR is 0. If any of the bridge
994 		 * resources covers that starting address (0 then it's good enough for
995 		 * us for memory
996 		 */
997 		for (i = 0; i < 3; i++) {
998 			if ((hose->mem_resources[i].flags & IORESOURCE_MEM) &&
999 			    hose->mem_resources[i].start == hose->pci_mem_offset)
1000 				return 0;
1001 		}
1002 
1003 		/* Well, it starts at 0 and we know it will collide so we may as
1004 		 * well consider it as unassigned. That covers the Apple case.
1005 		 */
1006 		return 1;
1007 	} else {
1008 		/* If the BAR is non-0, then we consider it assigned */
1009 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1010 		if (((res->start - offset) & 0xfffffffful) != 0)
1011 			return 0;
1012 
1013 		/* Here, we are a bit different than memory as typically IO space
1014 		 * starting at low addresses -is- valid. What we do instead if that
1015 		 * we consider as unassigned anything that doesn't have IO enabled
1016 		 * in the PCI command register, and that's it.
1017 		 */
1018 		pci_read_config_word(dev, PCI_COMMAND, &command);
1019 		if (command & PCI_COMMAND_IO)
1020 			return 0;
1021 
1022 		/* It's starting at 0 and IO is disabled in the bridge, consider
1023 		 * it unassigned
1024 		 */
1025 		return 1;
1026 	}
1027 }
1028 
1029 /* Fixup resources of a PCI<->PCI bridge */
1030 static void __devinit pcibios_fixup_bridge(struct pci_bus *bus)
1031 {
1032 	struct resource *res;
1033 	int i;
1034 
1035 	struct pci_dev *dev = bus->self;
1036 
1037 	pci_bus_for_each_resource(bus, res, i) {
1038 		if (!res || !res->flags)
1039 			continue;
1040 		if (i >= 3 && bus->self->transparent)
1041 			continue;
1042 
1043 		pr_debug("PCI:%s Bus rsrc %d %016llx-%016llx [%x] fixup...\n",
1044 			 pci_name(dev), i,
1045 			 (unsigned long long)res->start,\
1046 			 (unsigned long long)res->end,
1047 			 (unsigned int)res->flags);
1048 
1049 		/* Perform fixup */
1050 		fixup_resource(res, dev);
1051 
1052 		/* Try to detect uninitialized P2P bridge resources,
1053 		 * and clear them out so they get re-assigned later
1054 		 */
1055 		if (pcibios_uninitialized_bridge_resource(bus, res)) {
1056 			res->flags = 0;
1057 			pr_debug("PCI:%s            (unassigned)\n", pci_name(dev));
1058 		} else {
1059 
1060 			pr_debug("PCI:%s            %016llx-%016llx\n",
1061 				 pci_name(dev),
1062 				 (unsigned long long)res->start,
1063 				 (unsigned long long)res->end);
1064 		}
1065 	}
1066 }
1067 
1068 void __devinit pcibios_setup_bus_self(struct pci_bus *bus)
1069 {
1070 	/* Fix up the bus resources for P2P bridges */
1071 	if (bus->self != NULL)
1072 		pcibios_fixup_bridge(bus);
1073 
1074 	/* Platform specific bus fixups. This is currently only used
1075 	 * by fsl_pci and I'm hoping to get rid of it at some point
1076 	 */
1077 	if (ppc_md.pcibios_fixup_bus)
1078 		ppc_md.pcibios_fixup_bus(bus);
1079 
1080 	/* Setup bus DMA mappings */
1081 	if (ppc_md.pci_dma_bus_setup)
1082 		ppc_md.pci_dma_bus_setup(bus);
1083 }
1084 
1085 void __devinit pcibios_setup_bus_devices(struct pci_bus *bus)
1086 {
1087 	struct pci_dev *dev;
1088 
1089 	pr_debug("PCI: Fixup bus devices %d (%s)\n",
1090 		 bus->number, bus->self ? pci_name(bus->self) : "PHB");
1091 
1092 	list_for_each_entry(dev, &bus->devices, bus_list) {
1093 		struct dev_archdata *sd = &dev->dev.archdata;
1094 
1095 		/* Cardbus can call us to add new devices to a bus, so ignore
1096 		 * those who are already fully discovered
1097 		 */
1098 		if (dev->is_added)
1099 			continue;
1100 
1101 		/* Setup OF node pointer in the device */
1102 		dev->dev.of_node = pci_device_to_OF_node(dev);
1103 
1104 		/* Fixup NUMA node as it may not be setup yet by the generic
1105 		 * code and is needed by the DMA init
1106 		 */
1107 		set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
1108 
1109 		/* Hook up default DMA ops */
1110 		sd->dma_ops = pci_dma_ops;
1111 		set_dma_offset(&dev->dev, PCI_DRAM_OFFSET);
1112 
1113 		/* Additional platform DMA/iommu setup */
1114 		if (ppc_md.pci_dma_dev_setup)
1115 			ppc_md.pci_dma_dev_setup(dev);
1116 
1117 		/* Read default IRQs and fixup if necessary */
1118 		pci_read_irq_line(dev);
1119 		if (ppc_md.pci_irq_fixup)
1120 			ppc_md.pci_irq_fixup(dev);
1121 	}
1122 }
1123 
1124 void __devinit pcibios_fixup_bus(struct pci_bus *bus)
1125 {
1126 	/* When called from the generic PCI probe, read PCI<->PCI bridge
1127 	 * bases. This is -not- called when generating the PCI tree from
1128 	 * the OF device-tree.
1129 	 */
1130 	if (bus->self != NULL)
1131 		pci_read_bridge_bases(bus);
1132 
1133 	/* Now fixup the bus bus */
1134 	pcibios_setup_bus_self(bus);
1135 
1136 	/* Now fixup devices on that bus */
1137 	pcibios_setup_bus_devices(bus);
1138 }
1139 EXPORT_SYMBOL(pcibios_fixup_bus);
1140 
1141 void __devinit pci_fixup_cardbus(struct pci_bus *bus)
1142 {
1143 	/* Now fixup devices on that bus */
1144 	pcibios_setup_bus_devices(bus);
1145 }
1146 
1147 
1148 static int skip_isa_ioresource_align(struct pci_dev *dev)
1149 {
1150 	if ((ppc_pci_flags & PPC_PCI_CAN_SKIP_ISA_ALIGN) &&
1151 	    !(dev->bus->bridge_ctl & PCI_BRIDGE_CTL_ISA))
1152 		return 1;
1153 	return 0;
1154 }
1155 
1156 /*
1157  * We need to avoid collisions with `mirrored' VGA ports
1158  * and other strange ISA hardware, so we always want the
1159  * addresses to be allocated in the 0x000-0x0ff region
1160  * modulo 0x400.
1161  *
1162  * Why? Because some silly external IO cards only decode
1163  * the low 10 bits of the IO address. The 0x00-0xff region
1164  * is reserved for motherboard devices that decode all 16
1165  * bits, so it's ok to allocate at, say, 0x2800-0x28ff,
1166  * but we want to try to avoid allocating at 0x2900-0x2bff
1167  * which might have be mirrored at 0x0100-0x03ff..
1168  */
1169 resource_size_t pcibios_align_resource(void *data, const struct resource *res,
1170 				resource_size_t size, resource_size_t align)
1171 {
1172 	struct pci_dev *dev = data;
1173 	resource_size_t start = res->start;
1174 
1175 	if (res->flags & IORESOURCE_IO) {
1176 		if (skip_isa_ioresource_align(dev))
1177 			return start;
1178 		if (start & 0x300)
1179 			start = (start + 0x3ff) & ~0x3ff;
1180 	}
1181 
1182 	return start;
1183 }
1184 EXPORT_SYMBOL(pcibios_align_resource);
1185 
1186 /*
1187  * Reparent resource children of pr that conflict with res
1188  * under res, and make res replace those children.
1189  */
1190 static int reparent_resources(struct resource *parent,
1191 				     struct resource *res)
1192 {
1193 	struct resource *p, **pp;
1194 	struct resource **firstpp = NULL;
1195 
1196 	for (pp = &parent->child; (p = *pp) != NULL; pp = &p->sibling) {
1197 		if (p->end < res->start)
1198 			continue;
1199 		if (res->end < p->start)
1200 			break;
1201 		if (p->start < res->start || p->end > res->end)
1202 			return -1;	/* not completely contained */
1203 		if (firstpp == NULL)
1204 			firstpp = pp;
1205 	}
1206 	if (firstpp == NULL)
1207 		return -1;	/* didn't find any conflicting entries? */
1208 	res->parent = parent;
1209 	res->child = *firstpp;
1210 	res->sibling = *pp;
1211 	*firstpp = res;
1212 	*pp = NULL;
1213 	for (p = res->child; p != NULL; p = p->sibling) {
1214 		p->parent = res;
1215 		pr_debug("PCI: Reparented %s [%llx..%llx] under %s\n",
1216 			 p->name,
1217 			 (unsigned long long)p->start,
1218 			 (unsigned long long)p->end, res->name);
1219 	}
1220 	return 0;
1221 }
1222 
1223 /*
1224  *  Handle resources of PCI devices.  If the world were perfect, we could
1225  *  just allocate all the resource regions and do nothing more.  It isn't.
1226  *  On the other hand, we cannot just re-allocate all devices, as it would
1227  *  require us to know lots of host bridge internals.  So we attempt to
1228  *  keep as much of the original configuration as possible, but tweak it
1229  *  when it's found to be wrong.
1230  *
1231  *  Known BIOS problems we have to work around:
1232  *	- I/O or memory regions not configured
1233  *	- regions configured, but not enabled in the command register
1234  *	- bogus I/O addresses above 64K used
1235  *	- expansion ROMs left enabled (this may sound harmless, but given
1236  *	  the fact the PCI specs explicitly allow address decoders to be
1237  *	  shared between expansion ROMs and other resource regions, it's
1238  *	  at least dangerous)
1239  *
1240  *  Our solution:
1241  *	(1) Allocate resources for all buses behind PCI-to-PCI bridges.
1242  *	    This gives us fixed barriers on where we can allocate.
1243  *	(2) Allocate resources for all enabled devices.  If there is
1244  *	    a collision, just mark the resource as unallocated. Also
1245  *	    disable expansion ROMs during this step.
1246  *	(3) Try to allocate resources for disabled devices.  If the
1247  *	    resources were assigned correctly, everything goes well,
1248  *	    if they weren't, they won't disturb allocation of other
1249  *	    resources.
1250  *	(4) Assign new addresses to resources which were either
1251  *	    not configured at all or misconfigured.  If explicitly
1252  *	    requested by the user, configure expansion ROM address
1253  *	    as well.
1254  */
1255 
1256 void pcibios_allocate_bus_resources(struct pci_bus *bus)
1257 {
1258 	struct pci_bus *b;
1259 	int i;
1260 	struct resource *res, *pr;
1261 
1262 	pr_debug("PCI: Allocating bus resources for %04x:%02x...\n",
1263 		 pci_domain_nr(bus), bus->number);
1264 
1265 	pci_bus_for_each_resource(bus, res, i) {
1266 		if (!res || !res->flags || res->start > res->end || res->parent)
1267 			continue;
1268 		if (bus->parent == NULL)
1269 			pr = (res->flags & IORESOURCE_IO) ?
1270 				&ioport_resource : &iomem_resource;
1271 		else {
1272 			/* Don't bother with non-root busses when
1273 			 * re-assigning all resources. We clear the
1274 			 * resource flags as if they were colliding
1275 			 * and as such ensure proper re-allocation
1276 			 * later.
1277 			 */
1278 			if (ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)
1279 				goto clear_resource;
1280 			pr = pci_find_parent_resource(bus->self, res);
1281 			if (pr == res) {
1282 				/* this happens when the generic PCI
1283 				 * code (wrongly) decides that this
1284 				 * bridge is transparent  -- paulus
1285 				 */
1286 				continue;
1287 			}
1288 		}
1289 
1290 		pr_debug("PCI: %s (bus %d) bridge rsrc %d: %016llx-%016llx "
1291 			 "[0x%x], parent %p (%s)\n",
1292 			 bus->self ? pci_name(bus->self) : "PHB",
1293 			 bus->number, i,
1294 			 (unsigned long long)res->start,
1295 			 (unsigned long long)res->end,
1296 			 (unsigned int)res->flags,
1297 			 pr, (pr && pr->name) ? pr->name : "nil");
1298 
1299 		if (pr && !(pr->flags & IORESOURCE_UNSET)) {
1300 			if (request_resource(pr, res) == 0)
1301 				continue;
1302 			/*
1303 			 * Must be a conflict with an existing entry.
1304 			 * Move that entry (or entries) under the
1305 			 * bridge resource and try again.
1306 			 */
1307 			if (reparent_resources(pr, res) == 0)
1308 				continue;
1309 		}
1310 		printk(KERN_WARNING "PCI: Cannot allocate resource region "
1311 		       "%d of PCI bridge %d, will remap\n", i, bus->number);
1312 clear_resource:
1313 		res->start = res->end = 0;
1314 		res->flags = 0;
1315 	}
1316 
1317 	list_for_each_entry(b, &bus->children, node)
1318 		pcibios_allocate_bus_resources(b);
1319 }
1320 
1321 static inline void __devinit alloc_resource(struct pci_dev *dev, int idx)
1322 {
1323 	struct resource *pr, *r = &dev->resource[idx];
1324 
1325 	pr_debug("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1326 		 pci_name(dev), idx,
1327 		 (unsigned long long)r->start,
1328 		 (unsigned long long)r->end,
1329 		 (unsigned int)r->flags);
1330 
1331 	pr = pci_find_parent_resource(dev, r);
1332 	if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1333 	    request_resource(pr, r) < 0) {
1334 		printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1335 		       " of device %s, will remap\n", idx, pci_name(dev));
1336 		if (pr)
1337 			pr_debug("PCI:  parent is %p: %016llx-%016llx [%x]\n",
1338 				 pr,
1339 				 (unsigned long long)pr->start,
1340 				 (unsigned long long)pr->end,
1341 				 (unsigned int)pr->flags);
1342 		/* We'll assign a new address later */
1343 		r->flags |= IORESOURCE_UNSET;
1344 		r->end -= r->start;
1345 		r->start = 0;
1346 	}
1347 }
1348 
1349 static void __init pcibios_allocate_resources(int pass)
1350 {
1351 	struct pci_dev *dev = NULL;
1352 	int idx, disabled;
1353 	u16 command;
1354 	struct resource *r;
1355 
1356 	for_each_pci_dev(dev) {
1357 		pci_read_config_word(dev, PCI_COMMAND, &command);
1358 		for (idx = 0; idx <= PCI_ROM_RESOURCE; idx++) {
1359 			r = &dev->resource[idx];
1360 			if (r->parent)		/* Already allocated */
1361 				continue;
1362 			if (!r->flags || (r->flags & IORESOURCE_UNSET))
1363 				continue;	/* Not assigned at all */
1364 			/* We only allocate ROMs on pass 1 just in case they
1365 			 * have been screwed up by firmware
1366 			 */
1367 			if (idx == PCI_ROM_RESOURCE )
1368 				disabled = 1;
1369 			if (r->flags & IORESOURCE_IO)
1370 				disabled = !(command & PCI_COMMAND_IO);
1371 			else
1372 				disabled = !(command & PCI_COMMAND_MEMORY);
1373 			if (pass == disabled)
1374 				alloc_resource(dev, idx);
1375 		}
1376 		if (pass)
1377 			continue;
1378 		r = &dev->resource[PCI_ROM_RESOURCE];
1379 		if (r->flags) {
1380 			/* Turn the ROM off, leave the resource region,
1381 			 * but keep it unregistered.
1382 			 */
1383 			u32 reg;
1384 			pci_read_config_dword(dev, dev->rom_base_reg, &reg);
1385 			if (reg & PCI_ROM_ADDRESS_ENABLE) {
1386 				pr_debug("PCI: Switching off ROM of %s\n",
1387 					 pci_name(dev));
1388 				r->flags &= ~IORESOURCE_ROM_ENABLE;
1389 				pci_write_config_dword(dev, dev->rom_base_reg,
1390 						       reg & ~PCI_ROM_ADDRESS_ENABLE);
1391 			}
1392 		}
1393 	}
1394 }
1395 
1396 static void __init pcibios_reserve_legacy_regions(struct pci_bus *bus)
1397 {
1398 	struct pci_controller *hose = pci_bus_to_host(bus);
1399 	resource_size_t	offset;
1400 	struct resource *res, *pres;
1401 	int i;
1402 
1403 	pr_debug("Reserving legacy ranges for domain %04x\n", pci_domain_nr(bus));
1404 
1405 	/* Check for IO */
1406 	if (!(hose->io_resource.flags & IORESOURCE_IO))
1407 		goto no_io;
1408 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1409 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1410 	BUG_ON(res == NULL);
1411 	res->name = "Legacy IO";
1412 	res->flags = IORESOURCE_IO;
1413 	res->start = offset;
1414 	res->end = (offset + 0xfff) & 0xfffffffful;
1415 	pr_debug("Candidate legacy IO: %pR\n", res);
1416 	if (request_resource(&hose->io_resource, res)) {
1417 		printk(KERN_DEBUG
1418 		       "PCI %04x:%02x Cannot reserve Legacy IO %pR\n",
1419 		       pci_domain_nr(bus), bus->number, res);
1420 		kfree(res);
1421 	}
1422 
1423  no_io:
1424 	/* Check for memory */
1425 	offset = hose->pci_mem_offset;
1426 	pr_debug("hose mem offset: %016llx\n", (unsigned long long)offset);
1427 	for (i = 0; i < 3; i++) {
1428 		pres = &hose->mem_resources[i];
1429 		if (!(pres->flags & IORESOURCE_MEM))
1430 			continue;
1431 		pr_debug("hose mem res: %pR\n", pres);
1432 		if ((pres->start - offset) <= 0xa0000 &&
1433 		    (pres->end - offset) >= 0xbffff)
1434 			break;
1435 	}
1436 	if (i >= 3)
1437 		return;
1438 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1439 	BUG_ON(res == NULL);
1440 	res->name = "Legacy VGA memory";
1441 	res->flags = IORESOURCE_MEM;
1442 	res->start = 0xa0000 + offset;
1443 	res->end = 0xbffff + offset;
1444 	pr_debug("Candidate VGA memory: %pR\n", res);
1445 	if (request_resource(pres, res)) {
1446 		printk(KERN_DEBUG
1447 		       "PCI %04x:%02x Cannot reserve VGA memory %pR\n",
1448 		       pci_domain_nr(bus), bus->number, res);
1449 		kfree(res);
1450 	}
1451 }
1452 
1453 void __init pcibios_resource_survey(void)
1454 {
1455 	struct pci_bus *b;
1456 
1457 	/* Allocate and assign resources. If we re-assign everything, then
1458 	 * we skip the allocate phase
1459 	 */
1460 	list_for_each_entry(b, &pci_root_buses, node)
1461 		pcibios_allocate_bus_resources(b);
1462 
1463 	if (!(ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)) {
1464 		pcibios_allocate_resources(0);
1465 		pcibios_allocate_resources(1);
1466 	}
1467 
1468 	/* Before we start assigning unassigned resource, we try to reserve
1469 	 * the low IO area and the VGA memory area if they intersect the
1470 	 * bus available resources to avoid allocating things on top of them
1471 	 */
1472 	if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1473 		list_for_each_entry(b, &pci_root_buses, node)
1474 			pcibios_reserve_legacy_regions(b);
1475 	}
1476 
1477 	/* Now, if the platform didn't decide to blindly trust the firmware,
1478 	 * we proceed to assigning things that were left unassigned
1479 	 */
1480 	if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1481 		pr_debug("PCI: Assigning unassigned resources...\n");
1482 		pci_assign_unassigned_resources();
1483 	}
1484 
1485 	/* Call machine dependent fixup */
1486 	if (ppc_md.pcibios_fixup)
1487 		ppc_md.pcibios_fixup();
1488 }
1489 
1490 #ifdef CONFIG_HOTPLUG
1491 
1492 /* This is used by the PCI hotplug driver to allocate resource
1493  * of newly plugged busses. We can try to consolidate with the
1494  * rest of the code later, for now, keep it as-is as our main
1495  * resource allocation function doesn't deal with sub-trees yet.
1496  */
1497 void pcibios_claim_one_bus(struct pci_bus *bus)
1498 {
1499 	struct pci_dev *dev;
1500 	struct pci_bus *child_bus;
1501 
1502 	list_for_each_entry(dev, &bus->devices, bus_list) {
1503 		int i;
1504 
1505 		for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1506 			struct resource *r = &dev->resource[i];
1507 
1508 			if (r->parent || !r->start || !r->flags)
1509 				continue;
1510 
1511 			pr_debug("PCI: Claiming %s: "
1512 				 "Resource %d: %016llx..%016llx [%x]\n",
1513 				 pci_name(dev), i,
1514 				 (unsigned long long)r->start,
1515 				 (unsigned long long)r->end,
1516 				 (unsigned int)r->flags);
1517 
1518 			pci_claim_resource(dev, i);
1519 		}
1520 	}
1521 
1522 	list_for_each_entry(child_bus, &bus->children, node)
1523 		pcibios_claim_one_bus(child_bus);
1524 }
1525 
1526 
1527 /* pcibios_finish_adding_to_bus
1528  *
1529  * This is to be called by the hotplug code after devices have been
1530  * added to a bus, this include calling it for a PHB that is just
1531  * being added
1532  */
1533 void pcibios_finish_adding_to_bus(struct pci_bus *bus)
1534 {
1535 	pr_debug("PCI: Finishing adding to hotplug bus %04x:%02x\n",
1536 		 pci_domain_nr(bus), bus->number);
1537 
1538 	/* Allocate bus and devices resources */
1539 	pcibios_allocate_bus_resources(bus);
1540 	pcibios_claim_one_bus(bus);
1541 
1542 	/* Add new devices to global lists.  Register in proc, sysfs. */
1543 	pci_bus_add_devices(bus);
1544 
1545 	/* Fixup EEH */
1546 	eeh_add_device_tree_late(bus);
1547 }
1548 EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
1549 
1550 #endif /* CONFIG_HOTPLUG */
1551 
1552 int pcibios_enable_device(struct pci_dev *dev, int mask)
1553 {
1554 	if (ppc_md.pcibios_enable_device_hook)
1555 		if (ppc_md.pcibios_enable_device_hook(dev))
1556 			return -EINVAL;
1557 
1558 	return pci_enable_resources(dev, mask);
1559 }
1560 
1561 void __devinit pcibios_setup_phb_resources(struct pci_controller *hose)
1562 {
1563 	struct pci_bus *bus = hose->bus;
1564 	struct resource *res;
1565 	int i;
1566 
1567 	/* Hookup PHB IO resource */
1568 	bus->resource[0] = res = &hose->io_resource;
1569 
1570 	if (!res->flags) {
1571 		printk(KERN_WARNING "PCI: I/O resource not set for host"
1572 		       " bridge %s (domain %d)\n",
1573 		       hose->dn->full_name, hose->global_number);
1574 #ifdef CONFIG_PPC32
1575 		/* Workaround for lack of IO resource only on 32-bit */
1576 		res->start = (unsigned long)hose->io_base_virt - isa_io_base;
1577 		res->end = res->start + IO_SPACE_LIMIT;
1578 		res->flags = IORESOURCE_IO;
1579 #endif /* CONFIG_PPC32 */
1580 	}
1581 
1582 	pr_debug("PCI: PHB IO resource    = %016llx-%016llx [%lx]\n",
1583 		 (unsigned long long)res->start,
1584 		 (unsigned long long)res->end,
1585 		 (unsigned long)res->flags);
1586 
1587 	/* Hookup PHB Memory resources */
1588 	for (i = 0; i < 3; ++i) {
1589 		res = &hose->mem_resources[i];
1590 		if (!res->flags) {
1591 			if (i > 0)
1592 				continue;
1593 			printk(KERN_ERR "PCI: Memory resource 0 not set for "
1594 			       "host bridge %s (domain %d)\n",
1595 			       hose->dn->full_name, hose->global_number);
1596 #ifdef CONFIG_PPC32
1597 			/* Workaround for lack of MEM resource only on 32-bit */
1598 			res->start = hose->pci_mem_offset;
1599 			res->end = (resource_size_t)-1LL;
1600 			res->flags = IORESOURCE_MEM;
1601 #endif /* CONFIG_PPC32 */
1602 		}
1603 		bus->resource[i+1] = res;
1604 
1605 		pr_debug("PCI: PHB MEM resource %d = %016llx-%016llx [%lx]\n", i,
1606 			 (unsigned long long)res->start,
1607 			 (unsigned long long)res->end,
1608 			 (unsigned long)res->flags);
1609 	}
1610 
1611 	pr_debug("PCI: PHB MEM offset     = %016llx\n",
1612 		 (unsigned long long)hose->pci_mem_offset);
1613 	pr_debug("PCI: PHB IO  offset     = %08lx\n",
1614 		 (unsigned long)hose->io_base_virt - _IO_BASE);
1615 
1616 }
1617 
1618 /*
1619  * Null PCI config access functions, for the case when we can't
1620  * find a hose.
1621  */
1622 #define NULL_PCI_OP(rw, size, type)					\
1623 static int								\
1624 null_##rw##_config_##size(struct pci_dev *dev, int offset, type val)	\
1625 {									\
1626 	return PCIBIOS_DEVICE_NOT_FOUND;    				\
1627 }
1628 
1629 static int
1630 null_read_config(struct pci_bus *bus, unsigned int devfn, int offset,
1631 		 int len, u32 *val)
1632 {
1633 	return PCIBIOS_DEVICE_NOT_FOUND;
1634 }
1635 
1636 static int
1637 null_write_config(struct pci_bus *bus, unsigned int devfn, int offset,
1638 		  int len, u32 val)
1639 {
1640 	return PCIBIOS_DEVICE_NOT_FOUND;
1641 }
1642 
1643 static struct pci_ops null_pci_ops =
1644 {
1645 	.read = null_read_config,
1646 	.write = null_write_config,
1647 };
1648 
1649 /*
1650  * These functions are used early on before PCI scanning is done
1651  * and all of the pci_dev and pci_bus structures have been created.
1652  */
1653 static struct pci_bus *
1654 fake_pci_bus(struct pci_controller *hose, int busnr)
1655 {
1656 	static struct pci_bus bus;
1657 
1658 	if (hose == 0) {
1659 		printk(KERN_ERR "Can't find hose for PCI bus %d!\n", busnr);
1660 	}
1661 	bus.number = busnr;
1662 	bus.sysdata = hose;
1663 	bus.ops = hose? hose->ops: &null_pci_ops;
1664 	return &bus;
1665 }
1666 
1667 #define EARLY_PCI_OP(rw, size, type)					\
1668 int early_##rw##_config_##size(struct pci_controller *hose, int bus,	\
1669 			       int devfn, int offset, type value)	\
1670 {									\
1671 	return pci_bus_##rw##_config_##size(fake_pci_bus(hose, bus),	\
1672 					    devfn, offset, value);	\
1673 }
1674 
1675 EARLY_PCI_OP(read, byte, u8 *)
1676 EARLY_PCI_OP(read, word, u16 *)
1677 EARLY_PCI_OP(read, dword, u32 *)
1678 EARLY_PCI_OP(write, byte, u8)
1679 EARLY_PCI_OP(write, word, u16)
1680 EARLY_PCI_OP(write, dword, u32)
1681 
1682 extern int pci_bus_find_capability (struct pci_bus *bus, unsigned int devfn, int cap);
1683 int early_find_capability(struct pci_controller *hose, int bus, int devfn,
1684 			  int cap)
1685 {
1686 	return pci_bus_find_capability(fake_pci_bus(hose, bus), devfn, cap);
1687 }
1688 
1689 /**
1690  * pci_scan_phb - Given a pci_controller, setup and scan the PCI bus
1691  * @hose: Pointer to the PCI host controller instance structure
1692  * @sysdata: value to use for sysdata pointer.  ppc32 and ppc64 differ here
1693  *
1694  * Note: the 'data' pointer is a temporary measure.  As 32 and 64 bit
1695  * pci code gets merged, this parameter should become unnecessary because
1696  * both will use the same value.
1697  */
1698 void __devinit pcibios_scan_phb(struct pci_controller *hose, void *sysdata)
1699 {
1700 	struct pci_bus *bus;
1701 	struct device_node *node = hose->dn;
1702 	int mode;
1703 
1704 	pr_debug("PCI: Scanning PHB %s\n",
1705 		 node ? node->full_name : "<NO NAME>");
1706 
1707 	/* Create an empty bus for the toplevel */
1708 	bus = pci_create_bus(hose->parent, hose->first_busno, hose->ops,
1709 			     sysdata);
1710 	if (bus == NULL) {
1711 		pr_err("Failed to create bus for PCI domain %04x\n",
1712 			hose->global_number);
1713 		return;
1714 	}
1715 	bus->secondary = hose->first_busno;
1716 	hose->bus = bus;
1717 
1718 	/* Get some IO space for the new PHB */
1719 	pcibios_setup_phb_io_space(hose);
1720 
1721 	/* Wire up PHB bus resources */
1722 	pcibios_setup_phb_resources(hose);
1723 
1724 	/* Get probe mode and perform scan */
1725 	mode = PCI_PROBE_NORMAL;
1726 	if (node && ppc_md.pci_probe_mode)
1727 		mode = ppc_md.pci_probe_mode(bus);
1728 	pr_debug("    probe mode: %d\n", mode);
1729 	if (mode == PCI_PROBE_DEVTREE) {
1730 		bus->subordinate = hose->last_busno;
1731 		of_scan_bus(node, bus);
1732 	}
1733 
1734 	if (mode == PCI_PROBE_NORMAL)
1735 		hose->last_busno = bus->subordinate = pci_scan_child_bus(bus);
1736 }
1737