xref: /linux/arch/powerpc/kernel/pci-common.c (revision e9fb13bfec7e017130ddc5c1b5466340470f4900)
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/of_pci.h>
26 #include <linux/mm.h>
27 #include <linux/list.h>
28 #include <linux/syscalls.h>
29 #include <linux/irq.h>
30 #include <linux/vmalloc.h>
31 #include <linux/slab.h>
32 
33 #include <asm/processor.h>
34 #include <asm/io.h>
35 #include <asm/prom.h>
36 #include <asm/pci-bridge.h>
37 #include <asm/byteorder.h>
38 #include <asm/machdep.h>
39 #include <asm/ppc-pci.h>
40 #include <asm/firmware.h>
41 #include <asm/eeh.h>
42 
43 static DEFINE_SPINLOCK(hose_spinlock);
44 LIST_HEAD(hose_list);
45 
46 /* XXX kill that some day ... */
47 static int global_phb_number;		/* Global phb counter */
48 
49 /* ISA Memory physical address */
50 resource_size_t isa_mem_base;
51 
52 /* Default PCI flags is 0 on ppc32, modified at boot on ppc64 */
53 unsigned int ppc_pci_flags = 0;
54 
55 
56 static struct dma_map_ops *pci_dma_ops = &dma_direct_ops;
57 
58 void set_pci_dma_ops(struct dma_map_ops *dma_ops)
59 {
60 	pci_dma_ops = dma_ops;
61 }
62 
63 struct dma_map_ops *get_pci_dma_ops(void)
64 {
65 	return pci_dma_ops;
66 }
67 EXPORT_SYMBOL(get_pci_dma_ops);
68 
69 struct pci_controller *pcibios_alloc_controller(struct device_node *dev)
70 {
71 	struct pci_controller *phb;
72 
73 	phb = zalloc_maybe_bootmem(sizeof(struct pci_controller), GFP_KERNEL);
74 	if (phb == NULL)
75 		return NULL;
76 	spin_lock(&hose_spinlock);
77 	phb->global_number = global_phb_number++;
78 	list_add_tail(&phb->list_node, &hose_list);
79 	spin_unlock(&hose_spinlock);
80 	phb->dn = dev;
81 	phb->is_dynamic = mem_init_done;
82 #ifdef CONFIG_PPC64
83 	if (dev) {
84 		int nid = of_node_to_nid(dev);
85 
86 		if (nid < 0 || !node_online(nid))
87 			nid = -1;
88 
89 		PHB_SET_NODE(phb, nid);
90 	}
91 #endif
92 	return phb;
93 }
94 
95 void pcibios_free_controller(struct pci_controller *phb)
96 {
97 	spin_lock(&hose_spinlock);
98 	list_del(&phb->list_node);
99 	spin_unlock(&hose_spinlock);
100 
101 	if (phb->is_dynamic)
102 		kfree(phb);
103 }
104 
105 static resource_size_t pcibios_io_size(const struct pci_controller *hose)
106 {
107 #ifdef CONFIG_PPC64
108 	return hose->pci_io_size;
109 #else
110 	return hose->io_resource.end - hose->io_resource.start + 1;
111 #endif
112 }
113 
114 int pcibios_vaddr_is_ioport(void __iomem *address)
115 {
116 	int ret = 0;
117 	struct pci_controller *hose;
118 	resource_size_t size;
119 
120 	spin_lock(&hose_spinlock);
121 	list_for_each_entry(hose, &hose_list, list_node) {
122 		size = pcibios_io_size(hose);
123 		if (address >= hose->io_base_virt &&
124 		    address < (hose->io_base_virt + size)) {
125 			ret = 1;
126 			break;
127 		}
128 	}
129 	spin_unlock(&hose_spinlock);
130 	return ret;
131 }
132 
133 unsigned long pci_address_to_pio(phys_addr_t address)
134 {
135 	struct pci_controller *hose;
136 	resource_size_t size;
137 	unsigned long ret = ~0;
138 
139 	spin_lock(&hose_spinlock);
140 	list_for_each_entry(hose, &hose_list, list_node) {
141 		size = pcibios_io_size(hose);
142 		if (address >= hose->io_base_phys &&
143 		    address < (hose->io_base_phys + size)) {
144 			unsigned long base =
145 				(unsigned long)hose->io_base_virt - _IO_BASE;
146 			ret = base + (address - hose->io_base_phys);
147 			break;
148 		}
149 	}
150 	spin_unlock(&hose_spinlock);
151 
152 	return ret;
153 }
154 EXPORT_SYMBOL_GPL(pci_address_to_pio);
155 
156 /*
157  * Return the domain number for this bus.
158  */
159 int pci_domain_nr(struct pci_bus *bus)
160 {
161 	struct pci_controller *hose = pci_bus_to_host(bus);
162 
163 	return hose->global_number;
164 }
165 EXPORT_SYMBOL(pci_domain_nr);
166 
167 /* This routine is meant to be used early during boot, when the
168  * PCI bus numbers have not yet been assigned, and you need to
169  * issue PCI config cycles to an OF device.
170  * It could also be used to "fix" RTAS config cycles if you want
171  * to set pci_assign_all_buses to 1 and still use RTAS for PCI
172  * config cycles.
173  */
174 struct pci_controller* pci_find_hose_for_OF_device(struct device_node* node)
175 {
176 	while(node) {
177 		struct pci_controller *hose, *tmp;
178 		list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
179 			if (hose->dn == node)
180 				return hose;
181 		node = node->parent;
182 	}
183 	return NULL;
184 }
185 
186 static ssize_t pci_show_devspec(struct device *dev,
187 		struct device_attribute *attr, char *buf)
188 {
189 	struct pci_dev *pdev;
190 	struct device_node *np;
191 
192 	pdev = to_pci_dev (dev);
193 	np = pci_device_to_OF_node(pdev);
194 	if (np == NULL || np->full_name == NULL)
195 		return 0;
196 	return sprintf(buf, "%s", np->full_name);
197 }
198 static DEVICE_ATTR(devspec, S_IRUGO, pci_show_devspec, NULL);
199 
200 /* Add sysfs properties */
201 int pcibios_add_platform_entries(struct pci_dev *pdev)
202 {
203 	return device_create_file(&pdev->dev, &dev_attr_devspec);
204 }
205 
206 char __devinit *pcibios_setup(char *str)
207 {
208 	return str;
209 }
210 
211 /*
212  * Reads the interrupt pin to determine if interrupt is use by card.
213  * If the interrupt is used, then gets the interrupt line from the
214  * openfirmware and sets it in the pci_dev and pci_config line.
215  */
216 int pci_read_irq_line(struct pci_dev *pci_dev)
217 {
218 	struct of_irq oirq;
219 	unsigned int virq;
220 
221 	/* The current device-tree that iSeries generates from the HV
222 	 * PCI informations doesn't contain proper interrupt routing,
223 	 * and all the fallback would do is print out crap, so we
224 	 * don't attempt to resolve the interrupts here at all, some
225 	 * iSeries specific fixup does it.
226 	 *
227 	 * In the long run, we will hopefully fix the generated device-tree
228 	 * instead.
229 	 */
230 #ifdef CONFIG_PPC_ISERIES
231 	if (firmware_has_feature(FW_FEATURE_ISERIES))
232 		return -1;
233 #endif
234 
235 	pr_debug("PCI: Try to map irq for %s...\n", pci_name(pci_dev));
236 
237 #ifdef DEBUG
238 	memset(&oirq, 0xff, sizeof(oirq));
239 #endif
240 	/* Try to get a mapping from the device-tree */
241 	if (of_irq_map_pci(pci_dev, &oirq)) {
242 		u8 line, pin;
243 
244 		/* If that fails, lets fallback to what is in the config
245 		 * space and map that through the default controller. We
246 		 * also set the type to level low since that's what PCI
247 		 * interrupts are. If your platform does differently, then
248 		 * either provide a proper interrupt tree or don't use this
249 		 * function.
250 		 */
251 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_PIN, &pin))
252 			return -1;
253 		if (pin == 0)
254 			return -1;
255 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_LINE, &line) ||
256 		    line == 0xff || line == 0) {
257 			return -1;
258 		}
259 		pr_debug(" No map ! Using line %d (pin %d) from PCI config\n",
260 			 line, pin);
261 
262 		virq = irq_create_mapping(NULL, line);
263 		if (virq != NO_IRQ)
264 			irq_set_irq_type(virq, IRQ_TYPE_LEVEL_LOW);
265 	} else {
266 		pr_debug(" Got one, spec %d cells (0x%08x 0x%08x...) on %s\n",
267 			 oirq.size, oirq.specifier[0], oirq.specifier[1],
268 			 oirq.controller ? oirq.controller->full_name :
269 			 "<default>");
270 
271 		virq = irq_create_of_mapping(oirq.controller, oirq.specifier,
272 					     oirq.size);
273 	}
274 	if(virq == NO_IRQ) {
275 		pr_debug(" Failed to map !\n");
276 		return -1;
277 	}
278 
279 	pr_debug(" Mapped to linux irq %d\n", virq);
280 
281 	pci_dev->irq = virq;
282 
283 	return 0;
284 }
285 EXPORT_SYMBOL(pci_read_irq_line);
286 
287 /*
288  * Platform support for /proc/bus/pci/X/Y mmap()s,
289  * modelled on the sparc64 implementation by Dave Miller.
290  *  -- paulus.
291  */
292 
293 /*
294  * Adjust vm_pgoff of VMA such that it is the physical page offset
295  * corresponding to the 32-bit pci bus offset for DEV requested by the user.
296  *
297  * Basically, the user finds the base address for his device which he wishes
298  * to mmap.  They read the 32-bit value from the config space base register,
299  * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
300  * offset parameter of mmap on /proc/bus/pci/XXX for that device.
301  *
302  * Returns negative error code on failure, zero on success.
303  */
304 static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
305 					       resource_size_t *offset,
306 					       enum pci_mmap_state mmap_state)
307 {
308 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
309 	unsigned long io_offset = 0;
310 	int i, res_bit;
311 
312 	if (hose == 0)
313 		return NULL;		/* should never happen */
314 
315 	/* If memory, add on the PCI bridge address offset */
316 	if (mmap_state == pci_mmap_mem) {
317 #if 0 /* See comment in pci_resource_to_user() for why this is disabled */
318 		*offset += hose->pci_mem_offset;
319 #endif
320 		res_bit = IORESOURCE_MEM;
321 	} else {
322 		io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
323 		*offset += io_offset;
324 		res_bit = IORESOURCE_IO;
325 	}
326 
327 	/*
328 	 * Check that the offset requested corresponds to one of the
329 	 * resources of the device.
330 	 */
331 	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
332 		struct resource *rp = &dev->resource[i];
333 		int flags = rp->flags;
334 
335 		/* treat ROM as memory (should be already) */
336 		if (i == PCI_ROM_RESOURCE)
337 			flags |= IORESOURCE_MEM;
338 
339 		/* Active and same type? */
340 		if ((flags & res_bit) == 0)
341 			continue;
342 
343 		/* In the range of this resource? */
344 		if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
345 			continue;
346 
347 		/* found it! construct the final physical address */
348 		if (mmap_state == pci_mmap_io)
349 			*offset += hose->io_base_phys - io_offset;
350 		return rp;
351 	}
352 
353 	return NULL;
354 }
355 
356 /*
357  * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
358  * device mapping.
359  */
360 static pgprot_t __pci_mmap_set_pgprot(struct pci_dev *dev, struct resource *rp,
361 				      pgprot_t protection,
362 				      enum pci_mmap_state mmap_state,
363 				      int write_combine)
364 {
365 	unsigned long prot = pgprot_val(protection);
366 
367 	/* Write combine is always 0 on non-memory space mappings. On
368 	 * memory space, if the user didn't pass 1, we check for a
369 	 * "prefetchable" resource. This is a bit hackish, but we use
370 	 * this to workaround the inability of /sysfs to provide a write
371 	 * combine bit
372 	 */
373 	if (mmap_state != pci_mmap_mem)
374 		write_combine = 0;
375 	else if (write_combine == 0) {
376 		if (rp->flags & IORESOURCE_PREFETCH)
377 			write_combine = 1;
378 	}
379 
380 	/* XXX would be nice to have a way to ask for write-through */
381 	if (write_combine)
382 		return pgprot_noncached_wc(prot);
383 	else
384 		return pgprot_noncached(prot);
385 }
386 
387 /*
388  * This one is used by /dev/mem and fbdev who have no clue about the
389  * PCI device, it tries to find the PCI device first and calls the
390  * above routine
391  */
392 pgprot_t pci_phys_mem_access_prot(struct file *file,
393 				  unsigned long pfn,
394 				  unsigned long size,
395 				  pgprot_t prot)
396 {
397 	struct pci_dev *pdev = NULL;
398 	struct resource *found = NULL;
399 	resource_size_t offset = ((resource_size_t)pfn) << PAGE_SHIFT;
400 	int i;
401 
402 	if (page_is_ram(pfn))
403 		return prot;
404 
405 	prot = pgprot_noncached(prot);
406 	for_each_pci_dev(pdev) {
407 		for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
408 			struct resource *rp = &pdev->resource[i];
409 			int flags = rp->flags;
410 
411 			/* Active and same type? */
412 			if ((flags & IORESOURCE_MEM) == 0)
413 				continue;
414 			/* In the range of this resource? */
415 			if (offset < (rp->start & PAGE_MASK) ||
416 			    offset > rp->end)
417 				continue;
418 			found = rp;
419 			break;
420 		}
421 		if (found)
422 			break;
423 	}
424 	if (found) {
425 		if (found->flags & IORESOURCE_PREFETCH)
426 			prot = pgprot_noncached_wc(prot);
427 		pci_dev_put(pdev);
428 	}
429 
430 	pr_debug("PCI: Non-PCI map for %llx, prot: %lx\n",
431 		 (unsigned long long)offset, pgprot_val(prot));
432 
433 	return prot;
434 }
435 
436 
437 /*
438  * Perform the actual remap of the pages for a PCI device mapping, as
439  * appropriate for this architecture.  The region in the process to map
440  * is described by vm_start and vm_end members of VMA, the base physical
441  * address is found in vm_pgoff.
442  * The pci device structure is provided so that architectures may make mapping
443  * decisions on a per-device or per-bus basis.
444  *
445  * Returns a negative error code on failure, zero on success.
446  */
447 int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
448 			enum pci_mmap_state mmap_state, int write_combine)
449 {
450 	resource_size_t offset =
451 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
452 	struct resource *rp;
453 	int ret;
454 
455 	rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
456 	if (rp == NULL)
457 		return -EINVAL;
458 
459 	vma->vm_pgoff = offset >> PAGE_SHIFT;
460 	vma->vm_page_prot = __pci_mmap_set_pgprot(dev, rp,
461 						  vma->vm_page_prot,
462 						  mmap_state, write_combine);
463 
464 	ret = remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
465 			       vma->vm_end - vma->vm_start, vma->vm_page_prot);
466 
467 	return ret;
468 }
469 
470 /* This provides legacy IO read access on a bus */
471 int pci_legacy_read(struct pci_bus *bus, loff_t port, u32 *val, size_t size)
472 {
473 	unsigned long offset;
474 	struct pci_controller *hose = pci_bus_to_host(bus);
475 	struct resource *rp = &hose->io_resource;
476 	void __iomem *addr;
477 
478 	/* Check if port can be supported by that bus. We only check
479 	 * the ranges of the PHB though, not the bus itself as the rules
480 	 * for forwarding legacy cycles down bridges are not our problem
481 	 * here. So if the host bridge supports it, we do it.
482 	 */
483 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
484 	offset += port;
485 
486 	if (!(rp->flags & IORESOURCE_IO))
487 		return -ENXIO;
488 	if (offset < rp->start || (offset + size) > rp->end)
489 		return -ENXIO;
490 	addr = hose->io_base_virt + port;
491 
492 	switch(size) {
493 	case 1:
494 		*((u8 *)val) = in_8(addr);
495 		return 1;
496 	case 2:
497 		if (port & 1)
498 			return -EINVAL;
499 		*((u16 *)val) = in_le16(addr);
500 		return 2;
501 	case 4:
502 		if (port & 3)
503 			return -EINVAL;
504 		*((u32 *)val) = in_le32(addr);
505 		return 4;
506 	}
507 	return -EINVAL;
508 }
509 
510 /* This provides legacy IO write access on a bus */
511 int pci_legacy_write(struct pci_bus *bus, loff_t port, u32 val, size_t size)
512 {
513 	unsigned long offset;
514 	struct pci_controller *hose = pci_bus_to_host(bus);
515 	struct resource *rp = &hose->io_resource;
516 	void __iomem *addr;
517 
518 	/* Check if port can be supported by that bus. We only check
519 	 * the ranges of the PHB though, not the bus itself as the rules
520 	 * for forwarding legacy cycles down bridges are not our problem
521 	 * here. So if the host bridge supports it, we do it.
522 	 */
523 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
524 	offset += port;
525 
526 	if (!(rp->flags & IORESOURCE_IO))
527 		return -ENXIO;
528 	if (offset < rp->start || (offset + size) > rp->end)
529 		return -ENXIO;
530 	addr = hose->io_base_virt + port;
531 
532 	/* WARNING: The generic code is idiotic. It gets passed a pointer
533 	 * to what can be a 1, 2 or 4 byte quantity and always reads that
534 	 * as a u32, which means that we have to correct the location of
535 	 * the data read within those 32 bits for size 1 and 2
536 	 */
537 	switch(size) {
538 	case 1:
539 		out_8(addr, val >> 24);
540 		return 1;
541 	case 2:
542 		if (port & 1)
543 			return -EINVAL;
544 		out_le16(addr, val >> 16);
545 		return 2;
546 	case 4:
547 		if (port & 3)
548 			return -EINVAL;
549 		out_le32(addr, val);
550 		return 4;
551 	}
552 	return -EINVAL;
553 }
554 
555 /* This provides legacy IO or memory mmap access on a bus */
556 int pci_mmap_legacy_page_range(struct pci_bus *bus,
557 			       struct vm_area_struct *vma,
558 			       enum pci_mmap_state mmap_state)
559 {
560 	struct pci_controller *hose = pci_bus_to_host(bus);
561 	resource_size_t offset =
562 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
563 	resource_size_t size = vma->vm_end - vma->vm_start;
564 	struct resource *rp;
565 
566 	pr_debug("pci_mmap_legacy_page_range(%04x:%02x, %s @%llx..%llx)\n",
567 		 pci_domain_nr(bus), bus->number,
568 		 mmap_state == pci_mmap_mem ? "MEM" : "IO",
569 		 (unsigned long long)offset,
570 		 (unsigned long long)(offset + size - 1));
571 
572 	if (mmap_state == pci_mmap_mem) {
573 		/* Hack alert !
574 		 *
575 		 * Because X is lame and can fail starting if it gets an error trying
576 		 * to mmap legacy_mem (instead of just moving on without legacy memory
577 		 * access) we fake it here by giving it anonymous memory, effectively
578 		 * behaving just like /dev/zero
579 		 */
580 		if ((offset + size) > hose->isa_mem_size) {
581 			printk(KERN_DEBUG
582 			       "Process %s (pid:%d) mapped non-existing PCI legacy memory for 0%04x:%02x\n",
583 			       current->comm, current->pid, pci_domain_nr(bus), bus->number);
584 			if (vma->vm_flags & VM_SHARED)
585 				return shmem_zero_setup(vma);
586 			return 0;
587 		}
588 		offset += hose->isa_mem_phys;
589 	} else {
590 		unsigned long io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
591 		unsigned long roffset = offset + io_offset;
592 		rp = &hose->io_resource;
593 		if (!(rp->flags & IORESOURCE_IO))
594 			return -ENXIO;
595 		if (roffset < rp->start || (roffset + size) > rp->end)
596 			return -ENXIO;
597 		offset += hose->io_base_phys;
598 	}
599 	pr_debug(" -> mapping phys %llx\n", (unsigned long long)offset);
600 
601 	vma->vm_pgoff = offset >> PAGE_SHIFT;
602 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
603 	return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
604 			       vma->vm_end - vma->vm_start,
605 			       vma->vm_page_prot);
606 }
607 
608 void pci_resource_to_user(const struct pci_dev *dev, int bar,
609 			  const struct resource *rsrc,
610 			  resource_size_t *start, resource_size_t *end)
611 {
612 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
613 	resource_size_t offset = 0;
614 
615 	if (hose == NULL)
616 		return;
617 
618 	if (rsrc->flags & IORESOURCE_IO)
619 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
620 
621 	/* We pass a fully fixed up address to userland for MMIO instead of
622 	 * a BAR value because X is lame and expects to be able to use that
623 	 * to pass to /dev/mem !
624 	 *
625 	 * That means that we'll have potentially 64 bits values where some
626 	 * userland apps only expect 32 (like X itself since it thinks only
627 	 * Sparc has 64 bits MMIO) but if we don't do that, we break it on
628 	 * 32 bits CHRPs :-(
629 	 *
630 	 * Hopefully, the sysfs insterface is immune to that gunk. Once X
631 	 * has been fixed (and the fix spread enough), we can re-enable the
632 	 * 2 lines below and pass down a BAR value to userland. In that case
633 	 * we'll also have to re-enable the matching code in
634 	 * __pci_mmap_make_offset().
635 	 *
636 	 * BenH.
637 	 */
638 #if 0
639 	else if (rsrc->flags & IORESOURCE_MEM)
640 		offset = hose->pci_mem_offset;
641 #endif
642 
643 	*start = rsrc->start - offset;
644 	*end = rsrc->end - offset;
645 }
646 
647 /**
648  * pci_process_bridge_OF_ranges - Parse PCI bridge resources from device tree
649  * @hose: newly allocated pci_controller to be setup
650  * @dev: device node of the host bridge
651  * @primary: set if primary bus (32 bits only, soon to be deprecated)
652  *
653  * This function will parse the "ranges" property of a PCI host bridge device
654  * node and setup the resource mapping of a pci controller based on its
655  * content.
656  *
657  * Life would be boring if it wasn't for a few issues that we have to deal
658  * with here:
659  *
660  *   - We can only cope with one IO space range and up to 3 Memory space
661  *     ranges. However, some machines (thanks Apple !) tend to split their
662  *     space into lots of small contiguous ranges. So we have to coalesce.
663  *
664  *   - We can only cope with all memory ranges having the same offset
665  *     between CPU addresses and PCI addresses. Unfortunately, some bridges
666  *     are setup for a large 1:1 mapping along with a small "window" which
667  *     maps PCI address 0 to some arbitrary high address of the CPU space in
668  *     order to give access to the ISA memory hole.
669  *     The way out of here that I've chosen for now is to always set the
670  *     offset based on the first resource found, then override it if we
671  *     have a different offset and the previous was set by an ISA hole.
672  *
673  *   - Some busses have IO space not starting at 0, which causes trouble with
674  *     the way we do our IO resource renumbering. The code somewhat deals with
675  *     it for 64 bits but I would expect problems on 32 bits.
676  *
677  *   - Some 32 bits platforms such as 4xx can have physical space larger than
678  *     32 bits so we need to use 64 bits values for the parsing
679  */
680 void __devinit pci_process_bridge_OF_ranges(struct pci_controller *hose,
681 					    struct device_node *dev,
682 					    int primary)
683 {
684 	const u32 *ranges;
685 	int rlen;
686 	int pna = of_n_addr_cells(dev);
687 	int np = pna + 5;
688 	int memno = 0, isa_hole = -1;
689 	u32 pci_space;
690 	unsigned long long pci_addr, cpu_addr, pci_next, cpu_next, size;
691 	unsigned long long isa_mb = 0;
692 	struct resource *res;
693 
694 	printk(KERN_INFO "PCI host bridge %s %s ranges:\n",
695 	       dev->full_name, primary ? "(primary)" : "");
696 
697 	/* Get ranges property */
698 	ranges = of_get_property(dev, "ranges", &rlen);
699 	if (ranges == NULL)
700 		return;
701 
702 	/* Parse it */
703 	while ((rlen -= np * 4) >= 0) {
704 		/* Read next ranges element */
705 		pci_space = ranges[0];
706 		pci_addr = of_read_number(ranges + 1, 2);
707 		cpu_addr = of_translate_address(dev, ranges + 3);
708 		size = of_read_number(ranges + pna + 3, 2);
709 		ranges += np;
710 
711 		/* If we failed translation or got a zero-sized region
712 		 * (some FW try to feed us with non sensical zero sized regions
713 		 * such as power3 which look like some kind of attempt at exposing
714 		 * the VGA memory hole)
715 		 */
716 		if (cpu_addr == OF_BAD_ADDR || size == 0)
717 			continue;
718 
719 		/* Now consume following elements while they are contiguous */
720 		for (; rlen >= np * sizeof(u32);
721 		     ranges += np, rlen -= np * 4) {
722 			if (ranges[0] != pci_space)
723 				break;
724 			pci_next = of_read_number(ranges + 1, 2);
725 			cpu_next = of_translate_address(dev, ranges + 3);
726 			if (pci_next != pci_addr + size ||
727 			    cpu_next != cpu_addr + size)
728 				break;
729 			size += of_read_number(ranges + pna + 3, 2);
730 		}
731 
732 		/* Act based on address space type */
733 		res = NULL;
734 		switch ((pci_space >> 24) & 0x3) {
735 		case 1:		/* PCI IO space */
736 			printk(KERN_INFO
737 			       "  IO 0x%016llx..0x%016llx -> 0x%016llx\n",
738 			       cpu_addr, cpu_addr + size - 1, pci_addr);
739 
740 			/* We support only one IO range */
741 			if (hose->pci_io_size) {
742 				printk(KERN_INFO
743 				       " \\--> Skipped (too many) !\n");
744 				continue;
745 			}
746 #ifdef CONFIG_PPC32
747 			/* On 32 bits, limit I/O space to 16MB */
748 			if (size > 0x01000000)
749 				size = 0x01000000;
750 
751 			/* 32 bits needs to map IOs here */
752 			hose->io_base_virt = ioremap(cpu_addr, size);
753 
754 			/* Expect trouble if pci_addr is not 0 */
755 			if (primary)
756 				isa_io_base =
757 					(unsigned long)hose->io_base_virt;
758 #endif /* CONFIG_PPC32 */
759 			/* pci_io_size and io_base_phys always represent IO
760 			 * space starting at 0 so we factor in pci_addr
761 			 */
762 			hose->pci_io_size = pci_addr + size;
763 			hose->io_base_phys = cpu_addr - pci_addr;
764 
765 			/* Build resource */
766 			res = &hose->io_resource;
767 			res->flags = IORESOURCE_IO;
768 			res->start = pci_addr;
769 			break;
770 		case 2:		/* PCI Memory space */
771 		case 3:		/* PCI 64 bits Memory space */
772 			printk(KERN_INFO
773 			       " MEM 0x%016llx..0x%016llx -> 0x%016llx %s\n",
774 			       cpu_addr, cpu_addr + size - 1, pci_addr,
775 			       (pci_space & 0x40000000) ? "Prefetch" : "");
776 
777 			/* We support only 3 memory ranges */
778 			if (memno >= 3) {
779 				printk(KERN_INFO
780 				       " \\--> Skipped (too many) !\n");
781 				continue;
782 			}
783 			/* Handles ISA memory hole space here */
784 			if (pci_addr == 0) {
785 				isa_mb = cpu_addr;
786 				isa_hole = memno;
787 				if (primary || isa_mem_base == 0)
788 					isa_mem_base = cpu_addr;
789 				hose->isa_mem_phys = cpu_addr;
790 				hose->isa_mem_size = size;
791 			}
792 
793 			/* We get the PCI/Mem offset from the first range or
794 			 * the, current one if the offset came from an ISA
795 			 * hole. If they don't match, bugger.
796 			 */
797 			if (memno == 0 ||
798 			    (isa_hole >= 0 && pci_addr != 0 &&
799 			     hose->pci_mem_offset == isa_mb))
800 				hose->pci_mem_offset = cpu_addr - pci_addr;
801 			else if (pci_addr != 0 &&
802 				 hose->pci_mem_offset != cpu_addr - pci_addr) {
803 				printk(KERN_INFO
804 				       " \\--> Skipped (offset mismatch) !\n");
805 				continue;
806 			}
807 
808 			/* Build resource */
809 			res = &hose->mem_resources[memno++];
810 			res->flags = IORESOURCE_MEM;
811 			if (pci_space & 0x40000000)
812 				res->flags |= IORESOURCE_PREFETCH;
813 			res->start = cpu_addr;
814 			break;
815 		}
816 		if (res != NULL) {
817 			res->name = dev->full_name;
818 			res->end = res->start + size - 1;
819 			res->parent = NULL;
820 			res->sibling = NULL;
821 			res->child = NULL;
822 		}
823 	}
824 
825 	/* If there's an ISA hole and the pci_mem_offset is -not- matching
826 	 * the ISA hole offset, then we need to remove the ISA hole from
827 	 * the resource list for that brige
828 	 */
829 	if (isa_hole >= 0 && hose->pci_mem_offset != isa_mb) {
830 		unsigned int next = isa_hole + 1;
831 		printk(KERN_INFO " Removing ISA hole at 0x%016llx\n", isa_mb);
832 		if (next < memno)
833 			memmove(&hose->mem_resources[isa_hole],
834 				&hose->mem_resources[next],
835 				sizeof(struct resource) * (memno - next));
836 		hose->mem_resources[--memno].flags = 0;
837 	}
838 }
839 
840 /* Decide whether to display the domain number in /proc */
841 int pci_proc_domain(struct pci_bus *bus)
842 {
843 	struct pci_controller *hose = pci_bus_to_host(bus);
844 
845 	if (!(ppc_pci_flags & PPC_PCI_ENABLE_PROC_DOMAINS))
846 		return 0;
847 	if (ppc_pci_flags & PPC_PCI_COMPAT_DOMAIN_0)
848 		return hose->global_number != 0;
849 	return 1;
850 }
851 
852 void pcibios_resource_to_bus(struct pci_dev *dev, struct pci_bus_region *region,
853 			     struct resource *res)
854 {
855 	resource_size_t offset = 0, mask = (resource_size_t)-1;
856 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
857 
858 	if (!hose)
859 		return;
860 	if (res->flags & IORESOURCE_IO) {
861 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
862 		mask = 0xffffffffu;
863 	} else if (res->flags & IORESOURCE_MEM)
864 		offset = hose->pci_mem_offset;
865 
866 	region->start = (res->start - offset) & mask;
867 	region->end = (res->end - offset) & mask;
868 }
869 EXPORT_SYMBOL(pcibios_resource_to_bus);
870 
871 void pcibios_bus_to_resource(struct pci_dev *dev, struct resource *res,
872 			     struct pci_bus_region *region)
873 {
874 	resource_size_t offset = 0, mask = (resource_size_t)-1;
875 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
876 
877 	if (!hose)
878 		return;
879 	if (res->flags & IORESOURCE_IO) {
880 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
881 		mask = 0xffffffffu;
882 	} else if (res->flags & IORESOURCE_MEM)
883 		offset = hose->pci_mem_offset;
884 	res->start = (region->start + offset) & mask;
885 	res->end = (region->end + offset) & mask;
886 }
887 EXPORT_SYMBOL(pcibios_bus_to_resource);
888 
889 /* Fixup a bus resource into a linux resource */
890 static void __devinit fixup_resource(struct resource *res, struct pci_dev *dev)
891 {
892 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
893 	resource_size_t offset = 0, mask = (resource_size_t)-1;
894 
895 	if (res->flags & IORESOURCE_IO) {
896 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
897 		mask = 0xffffffffu;
898 	} else if (res->flags & IORESOURCE_MEM)
899 		offset = hose->pci_mem_offset;
900 
901 	res->start = (res->start + offset) & mask;
902 	res->end = (res->end + offset) & mask;
903 }
904 
905 
906 /* This header fixup will do the resource fixup for all devices as they are
907  * probed, but not for bridge ranges
908  */
909 static void __devinit pcibios_fixup_resources(struct pci_dev *dev)
910 {
911 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
912 	int i;
913 
914 	if (!hose) {
915 		printk(KERN_ERR "No host bridge for PCI dev %s !\n",
916 		       pci_name(dev));
917 		return;
918 	}
919 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
920 		struct resource *res = dev->resource + i;
921 		if (!res->flags)
922 			continue;
923 		/* On platforms that have PPC_PCI_PROBE_ONLY set, we don't
924 		 * consider 0 as an unassigned BAR value. It's technically
925 		 * a valid value, but linux doesn't like it... so when we can
926 		 * re-assign things, we do so, but if we can't, we keep it
927 		 * around and hope for the best...
928 		 */
929 		if (res->start == 0 && !(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
930 			pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] is unassigned\n",
931 				 pci_name(dev), i,
932 				 (unsigned long long)res->start,
933 				 (unsigned long long)res->end,
934 				 (unsigned int)res->flags);
935 			res->end -= res->start;
936 			res->start = 0;
937 			res->flags |= IORESOURCE_UNSET;
938 			continue;
939 		}
940 
941 		pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] fixup...\n",
942 			 pci_name(dev), i,
943 			 (unsigned long long)res->start,\
944 			 (unsigned long long)res->end,
945 			 (unsigned int)res->flags);
946 
947 		fixup_resource(res, dev);
948 
949 		pr_debug("PCI:%s            %016llx-%016llx\n",
950 			 pci_name(dev),
951 			 (unsigned long long)res->start,
952 			 (unsigned long long)res->end);
953 	}
954 
955 	/* Call machine specific resource fixup */
956 	if (ppc_md.pcibios_fixup_resources)
957 		ppc_md.pcibios_fixup_resources(dev);
958 }
959 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_resources);
960 
961 /* This function tries to figure out if a bridge resource has been initialized
962  * by the firmware or not. It doesn't have to be absolutely bullet proof, but
963  * things go more smoothly when it gets it right. It should covers cases such
964  * as Apple "closed" bridge resources and bare-metal pSeries unassigned bridges
965  */
966 static int __devinit pcibios_uninitialized_bridge_resource(struct pci_bus *bus,
967 							   struct resource *res)
968 {
969 	struct pci_controller *hose = pci_bus_to_host(bus);
970 	struct pci_dev *dev = bus->self;
971 	resource_size_t offset;
972 	u16 command;
973 	int i;
974 
975 	/* We don't do anything if PCI_PROBE_ONLY is set */
976 	if (ppc_pci_flags & PPC_PCI_PROBE_ONLY)
977 		return 0;
978 
979 	/* Job is a bit different between memory and IO */
980 	if (res->flags & IORESOURCE_MEM) {
981 		/* If the BAR is non-0 (res != pci_mem_offset) then it's probably been
982 		 * initialized by somebody
983 		 */
984 		if (res->start != hose->pci_mem_offset)
985 			return 0;
986 
987 		/* The BAR is 0, let's check if memory decoding is enabled on
988 		 * the bridge. If not, we consider it unassigned
989 		 */
990 		pci_read_config_word(dev, PCI_COMMAND, &command);
991 		if ((command & PCI_COMMAND_MEMORY) == 0)
992 			return 1;
993 
994 		/* Memory decoding is enabled and the BAR is 0. If any of the bridge
995 		 * resources covers that starting address (0 then it's good enough for
996 		 * us for memory
997 		 */
998 		for (i = 0; i < 3; i++) {
999 			if ((hose->mem_resources[i].flags & IORESOURCE_MEM) &&
1000 			    hose->mem_resources[i].start == hose->pci_mem_offset)
1001 				return 0;
1002 		}
1003 
1004 		/* Well, it starts at 0 and we know it will collide so we may as
1005 		 * well consider it as unassigned. That covers the Apple case.
1006 		 */
1007 		return 1;
1008 	} else {
1009 		/* If the BAR is non-0, then we consider it assigned */
1010 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1011 		if (((res->start - offset) & 0xfffffffful) != 0)
1012 			return 0;
1013 
1014 		/* Here, we are a bit different than memory as typically IO space
1015 		 * starting at low addresses -is- valid. What we do instead if that
1016 		 * we consider as unassigned anything that doesn't have IO enabled
1017 		 * in the PCI command register, and that's it.
1018 		 */
1019 		pci_read_config_word(dev, PCI_COMMAND, &command);
1020 		if (command & PCI_COMMAND_IO)
1021 			return 0;
1022 
1023 		/* It's starting at 0 and IO is disabled in the bridge, consider
1024 		 * it unassigned
1025 		 */
1026 		return 1;
1027 	}
1028 }
1029 
1030 /* Fixup resources of a PCI<->PCI bridge */
1031 static void __devinit pcibios_fixup_bridge(struct pci_bus *bus)
1032 {
1033 	struct resource *res;
1034 	int i;
1035 
1036 	struct pci_dev *dev = bus->self;
1037 
1038 	pci_bus_for_each_resource(bus, res, i) {
1039 		if (!res || !res->flags)
1040 			continue;
1041 		if (i >= 3 && bus->self->transparent)
1042 			continue;
1043 
1044 		pr_debug("PCI:%s Bus rsrc %d %016llx-%016llx [%x] fixup...\n",
1045 			 pci_name(dev), i,
1046 			 (unsigned long long)res->start,\
1047 			 (unsigned long long)res->end,
1048 			 (unsigned int)res->flags);
1049 
1050 		/* Perform fixup */
1051 		fixup_resource(res, dev);
1052 
1053 		/* Try to detect uninitialized P2P bridge resources,
1054 		 * and clear them out so they get re-assigned later
1055 		 */
1056 		if (pcibios_uninitialized_bridge_resource(bus, res)) {
1057 			res->flags = 0;
1058 			pr_debug("PCI:%s            (unassigned)\n", pci_name(dev));
1059 		} else {
1060 
1061 			pr_debug("PCI:%s            %016llx-%016llx\n",
1062 				 pci_name(dev),
1063 				 (unsigned long long)res->start,
1064 				 (unsigned long long)res->end);
1065 		}
1066 	}
1067 }
1068 
1069 void __devinit pcibios_setup_bus_self(struct pci_bus *bus)
1070 {
1071 	/* Fix up the bus resources for P2P bridges */
1072 	if (bus->self != NULL)
1073 		pcibios_fixup_bridge(bus);
1074 
1075 	/* Platform specific bus fixups. This is currently only used
1076 	 * by fsl_pci and I'm hoping to get rid of it at some point
1077 	 */
1078 	if (ppc_md.pcibios_fixup_bus)
1079 		ppc_md.pcibios_fixup_bus(bus);
1080 
1081 	/* Setup bus DMA mappings */
1082 	if (ppc_md.pci_dma_bus_setup)
1083 		ppc_md.pci_dma_bus_setup(bus);
1084 }
1085 
1086 void __devinit pcibios_setup_bus_devices(struct pci_bus *bus)
1087 {
1088 	struct pci_dev *dev;
1089 
1090 	pr_debug("PCI: Fixup bus devices %d (%s)\n",
1091 		 bus->number, bus->self ? pci_name(bus->self) : "PHB");
1092 
1093 	list_for_each_entry(dev, &bus->devices, bus_list) {
1094 		/* Cardbus can call us to add new devices to a bus, so ignore
1095 		 * those who are already fully discovered
1096 		 */
1097 		if (dev->is_added)
1098 			continue;
1099 
1100 		/* Setup OF node pointer in the device */
1101 		dev->dev.of_node = pci_device_to_OF_node(dev);
1102 
1103 		/* Fixup NUMA node as it may not be setup yet by the generic
1104 		 * code and is needed by the DMA init
1105 		 */
1106 		set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
1107 
1108 		/* Hook up default DMA ops */
1109 		set_dma_ops(&dev->dev, pci_dma_ops);
1110 		set_dma_offset(&dev->dev, PCI_DRAM_OFFSET);
1111 
1112 		/* Additional platform DMA/iommu setup */
1113 		if (ppc_md.pci_dma_dev_setup)
1114 			ppc_md.pci_dma_dev_setup(dev);
1115 
1116 		/* Read default IRQs and fixup if necessary */
1117 		pci_read_irq_line(dev);
1118 		if (ppc_md.pci_irq_fixup)
1119 			ppc_md.pci_irq_fixup(dev);
1120 	}
1121 }
1122 
1123 void __devinit pcibios_fixup_bus(struct pci_bus *bus)
1124 {
1125 	/* When called from the generic PCI probe, read PCI<->PCI bridge
1126 	 * bases. This is -not- called when generating the PCI tree from
1127 	 * the OF device-tree.
1128 	 */
1129 	if (bus->self != NULL)
1130 		pci_read_bridge_bases(bus);
1131 
1132 	/* Now fixup the bus bus */
1133 	pcibios_setup_bus_self(bus);
1134 
1135 	/* Now fixup devices on that bus */
1136 	pcibios_setup_bus_devices(bus);
1137 }
1138 EXPORT_SYMBOL(pcibios_fixup_bus);
1139 
1140 void __devinit pci_fixup_cardbus(struct pci_bus *bus)
1141 {
1142 	/* Now fixup devices on that bus */
1143 	pcibios_setup_bus_devices(bus);
1144 }
1145 
1146 
1147 static int skip_isa_ioresource_align(struct pci_dev *dev)
1148 {
1149 	if ((ppc_pci_flags & PPC_PCI_CAN_SKIP_ISA_ALIGN) &&
1150 	    !(dev->bus->bridge_ctl & PCI_BRIDGE_CTL_ISA))
1151 		return 1;
1152 	return 0;
1153 }
1154 
1155 /*
1156  * We need to avoid collisions with `mirrored' VGA ports
1157  * and other strange ISA hardware, so we always want the
1158  * addresses to be allocated in the 0x000-0x0ff region
1159  * modulo 0x400.
1160  *
1161  * Why? Because some silly external IO cards only decode
1162  * the low 10 bits of the IO address. The 0x00-0xff region
1163  * is reserved for motherboard devices that decode all 16
1164  * bits, so it's ok to allocate at, say, 0x2800-0x28ff,
1165  * but we want to try to avoid allocating at 0x2900-0x2bff
1166  * which might have be mirrored at 0x0100-0x03ff..
1167  */
1168 resource_size_t pcibios_align_resource(void *data, const struct resource *res,
1169 				resource_size_t size, resource_size_t align)
1170 {
1171 	struct pci_dev *dev = data;
1172 	resource_size_t start = res->start;
1173 
1174 	if (res->flags & IORESOURCE_IO) {
1175 		if (skip_isa_ioresource_align(dev))
1176 			return start;
1177 		if (start & 0x300)
1178 			start = (start + 0x3ff) & ~0x3ff;
1179 	}
1180 
1181 	return start;
1182 }
1183 EXPORT_SYMBOL(pcibios_align_resource);
1184 
1185 /*
1186  * Reparent resource children of pr that conflict with res
1187  * under res, and make res replace those children.
1188  */
1189 static int reparent_resources(struct resource *parent,
1190 				     struct resource *res)
1191 {
1192 	struct resource *p, **pp;
1193 	struct resource **firstpp = NULL;
1194 
1195 	for (pp = &parent->child; (p = *pp) != NULL; pp = &p->sibling) {
1196 		if (p->end < res->start)
1197 			continue;
1198 		if (res->end < p->start)
1199 			break;
1200 		if (p->start < res->start || p->end > res->end)
1201 			return -1;	/* not completely contained */
1202 		if (firstpp == NULL)
1203 			firstpp = pp;
1204 	}
1205 	if (firstpp == NULL)
1206 		return -1;	/* didn't find any conflicting entries? */
1207 	res->parent = parent;
1208 	res->child = *firstpp;
1209 	res->sibling = *pp;
1210 	*firstpp = res;
1211 	*pp = NULL;
1212 	for (p = res->child; p != NULL; p = p->sibling) {
1213 		p->parent = res;
1214 		pr_debug("PCI: Reparented %s [%llx..%llx] under %s\n",
1215 			 p->name,
1216 			 (unsigned long long)p->start,
1217 			 (unsigned long long)p->end, res->name);
1218 	}
1219 	return 0;
1220 }
1221 
1222 /*
1223  *  Handle resources of PCI devices.  If the world were perfect, we could
1224  *  just allocate all the resource regions and do nothing more.  It isn't.
1225  *  On the other hand, we cannot just re-allocate all devices, as it would
1226  *  require us to know lots of host bridge internals.  So we attempt to
1227  *  keep as much of the original configuration as possible, but tweak it
1228  *  when it's found to be wrong.
1229  *
1230  *  Known BIOS problems we have to work around:
1231  *	- I/O or memory regions not configured
1232  *	- regions configured, but not enabled in the command register
1233  *	- bogus I/O addresses above 64K used
1234  *	- expansion ROMs left enabled (this may sound harmless, but given
1235  *	  the fact the PCI specs explicitly allow address decoders to be
1236  *	  shared between expansion ROMs and other resource regions, it's
1237  *	  at least dangerous)
1238  *
1239  *  Our solution:
1240  *	(1) Allocate resources for all buses behind PCI-to-PCI bridges.
1241  *	    This gives us fixed barriers on where we can allocate.
1242  *	(2) Allocate resources for all enabled devices.  If there is
1243  *	    a collision, just mark the resource as unallocated. Also
1244  *	    disable expansion ROMs during this step.
1245  *	(3) Try to allocate resources for disabled devices.  If the
1246  *	    resources were assigned correctly, everything goes well,
1247  *	    if they weren't, they won't disturb allocation of other
1248  *	    resources.
1249  *	(4) Assign new addresses to resources which were either
1250  *	    not configured at all or misconfigured.  If explicitly
1251  *	    requested by the user, configure expansion ROM address
1252  *	    as well.
1253  */
1254 
1255 void pcibios_allocate_bus_resources(struct pci_bus *bus)
1256 {
1257 	struct pci_bus *b;
1258 	int i;
1259 	struct resource *res, *pr;
1260 
1261 	pr_debug("PCI: Allocating bus resources for %04x:%02x...\n",
1262 		 pci_domain_nr(bus), bus->number);
1263 
1264 	pci_bus_for_each_resource(bus, res, i) {
1265 		if (!res || !res->flags || res->start > res->end || res->parent)
1266 			continue;
1267 		if (bus->parent == NULL)
1268 			pr = (res->flags & IORESOURCE_IO) ?
1269 				&ioport_resource : &iomem_resource;
1270 		else {
1271 			/* Don't bother with non-root busses when
1272 			 * re-assigning all resources. We clear the
1273 			 * resource flags as if they were colliding
1274 			 * and as such ensure proper re-allocation
1275 			 * later.
1276 			 */
1277 			if (ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)
1278 				goto clear_resource;
1279 			pr = pci_find_parent_resource(bus->self, res);
1280 			if (pr == res) {
1281 				/* this happens when the generic PCI
1282 				 * code (wrongly) decides that this
1283 				 * bridge is transparent  -- paulus
1284 				 */
1285 				continue;
1286 			}
1287 		}
1288 
1289 		pr_debug("PCI: %s (bus %d) bridge rsrc %d: %016llx-%016llx "
1290 			 "[0x%x], parent %p (%s)\n",
1291 			 bus->self ? pci_name(bus->self) : "PHB",
1292 			 bus->number, i,
1293 			 (unsigned long long)res->start,
1294 			 (unsigned long long)res->end,
1295 			 (unsigned int)res->flags,
1296 			 pr, (pr && pr->name) ? pr->name : "nil");
1297 
1298 		if (pr && !(pr->flags & IORESOURCE_UNSET)) {
1299 			if (request_resource(pr, res) == 0)
1300 				continue;
1301 			/*
1302 			 * Must be a conflict with an existing entry.
1303 			 * Move that entry (or entries) under the
1304 			 * bridge resource and try again.
1305 			 */
1306 			if (reparent_resources(pr, res) == 0)
1307 				continue;
1308 		}
1309 		printk(KERN_WARNING "PCI: Cannot allocate resource region "
1310 		       "%d of PCI bridge %d, will remap\n", i, bus->number);
1311 clear_resource:
1312 		res->start = res->end = 0;
1313 		res->flags = 0;
1314 	}
1315 
1316 	list_for_each_entry(b, &bus->children, node)
1317 		pcibios_allocate_bus_resources(b);
1318 }
1319 
1320 static inline void __devinit alloc_resource(struct pci_dev *dev, int idx)
1321 {
1322 	struct resource *pr, *r = &dev->resource[idx];
1323 
1324 	pr_debug("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1325 		 pci_name(dev), idx,
1326 		 (unsigned long long)r->start,
1327 		 (unsigned long long)r->end,
1328 		 (unsigned int)r->flags);
1329 
1330 	pr = pci_find_parent_resource(dev, r);
1331 	if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1332 	    request_resource(pr, r) < 0) {
1333 		printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1334 		       " of device %s, will remap\n", idx, pci_name(dev));
1335 		if (pr)
1336 			pr_debug("PCI:  parent is %p: %016llx-%016llx [%x]\n",
1337 				 pr,
1338 				 (unsigned long long)pr->start,
1339 				 (unsigned long long)pr->end,
1340 				 (unsigned int)pr->flags);
1341 		/* We'll assign a new address later */
1342 		r->flags |= IORESOURCE_UNSET;
1343 		r->end -= r->start;
1344 		r->start = 0;
1345 	}
1346 }
1347 
1348 static void __init pcibios_allocate_resources(int pass)
1349 {
1350 	struct pci_dev *dev = NULL;
1351 	int idx, disabled;
1352 	u16 command;
1353 	struct resource *r;
1354 
1355 	for_each_pci_dev(dev) {
1356 		pci_read_config_word(dev, PCI_COMMAND, &command);
1357 		for (idx = 0; idx <= PCI_ROM_RESOURCE; idx++) {
1358 			r = &dev->resource[idx];
1359 			if (r->parent)		/* Already allocated */
1360 				continue;
1361 			if (!r->flags || (r->flags & IORESOURCE_UNSET))
1362 				continue;	/* Not assigned at all */
1363 			/* We only allocate ROMs on pass 1 just in case they
1364 			 * have been screwed up by firmware
1365 			 */
1366 			if (idx == PCI_ROM_RESOURCE )
1367 				disabled = 1;
1368 			if (r->flags & IORESOURCE_IO)
1369 				disabled = !(command & PCI_COMMAND_IO);
1370 			else
1371 				disabled = !(command & PCI_COMMAND_MEMORY);
1372 			if (pass == disabled)
1373 				alloc_resource(dev, idx);
1374 		}
1375 		if (pass)
1376 			continue;
1377 		r = &dev->resource[PCI_ROM_RESOURCE];
1378 		if (r->flags) {
1379 			/* Turn the ROM off, leave the resource region,
1380 			 * but keep it unregistered.
1381 			 */
1382 			u32 reg;
1383 			pci_read_config_dword(dev, dev->rom_base_reg, &reg);
1384 			if (reg & PCI_ROM_ADDRESS_ENABLE) {
1385 				pr_debug("PCI: Switching off ROM of %s\n",
1386 					 pci_name(dev));
1387 				r->flags &= ~IORESOURCE_ROM_ENABLE;
1388 				pci_write_config_dword(dev, dev->rom_base_reg,
1389 						       reg & ~PCI_ROM_ADDRESS_ENABLE);
1390 			}
1391 		}
1392 	}
1393 }
1394 
1395 static void __init pcibios_reserve_legacy_regions(struct pci_bus *bus)
1396 {
1397 	struct pci_controller *hose = pci_bus_to_host(bus);
1398 	resource_size_t	offset;
1399 	struct resource *res, *pres;
1400 	int i;
1401 
1402 	pr_debug("Reserving legacy ranges for domain %04x\n", pci_domain_nr(bus));
1403 
1404 	/* Check for IO */
1405 	if (!(hose->io_resource.flags & IORESOURCE_IO))
1406 		goto no_io;
1407 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1408 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1409 	BUG_ON(res == NULL);
1410 	res->name = "Legacy IO";
1411 	res->flags = IORESOURCE_IO;
1412 	res->start = offset;
1413 	res->end = (offset + 0xfff) & 0xfffffffful;
1414 	pr_debug("Candidate legacy IO: %pR\n", res);
1415 	if (request_resource(&hose->io_resource, res)) {
1416 		printk(KERN_DEBUG
1417 		       "PCI %04x:%02x Cannot reserve Legacy IO %pR\n",
1418 		       pci_domain_nr(bus), bus->number, res);
1419 		kfree(res);
1420 	}
1421 
1422  no_io:
1423 	/* Check for memory */
1424 	offset = hose->pci_mem_offset;
1425 	pr_debug("hose mem offset: %016llx\n", (unsigned long long)offset);
1426 	for (i = 0; i < 3; i++) {
1427 		pres = &hose->mem_resources[i];
1428 		if (!(pres->flags & IORESOURCE_MEM))
1429 			continue;
1430 		pr_debug("hose mem res: %pR\n", pres);
1431 		if ((pres->start - offset) <= 0xa0000 &&
1432 		    (pres->end - offset) >= 0xbffff)
1433 			break;
1434 	}
1435 	if (i >= 3)
1436 		return;
1437 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1438 	BUG_ON(res == NULL);
1439 	res->name = "Legacy VGA memory";
1440 	res->flags = IORESOURCE_MEM;
1441 	res->start = 0xa0000 + offset;
1442 	res->end = 0xbffff + offset;
1443 	pr_debug("Candidate VGA memory: %pR\n", res);
1444 	if (request_resource(pres, res)) {
1445 		printk(KERN_DEBUG
1446 		       "PCI %04x:%02x Cannot reserve VGA memory %pR\n",
1447 		       pci_domain_nr(bus), bus->number, res);
1448 		kfree(res);
1449 	}
1450 }
1451 
1452 void __init pcibios_resource_survey(void)
1453 {
1454 	struct pci_bus *b;
1455 
1456 	/* Allocate and assign resources. If we re-assign everything, then
1457 	 * we skip the allocate phase
1458 	 */
1459 	list_for_each_entry(b, &pci_root_buses, node)
1460 		pcibios_allocate_bus_resources(b);
1461 
1462 	if (!(ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)) {
1463 		pcibios_allocate_resources(0);
1464 		pcibios_allocate_resources(1);
1465 	}
1466 
1467 	/* Before we start assigning unassigned resource, we try to reserve
1468 	 * the low IO area and the VGA memory area if they intersect the
1469 	 * bus available resources to avoid allocating things on top of them
1470 	 */
1471 	if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1472 		list_for_each_entry(b, &pci_root_buses, node)
1473 			pcibios_reserve_legacy_regions(b);
1474 	}
1475 
1476 	/* Now, if the platform didn't decide to blindly trust the firmware,
1477 	 * we proceed to assigning things that were left unassigned
1478 	 */
1479 	if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1480 		pr_debug("PCI: Assigning unassigned resources...\n");
1481 		pci_assign_unassigned_resources();
1482 	}
1483 
1484 	/* Call machine dependent fixup */
1485 	if (ppc_md.pcibios_fixup)
1486 		ppc_md.pcibios_fixup();
1487 }
1488 
1489 #ifdef CONFIG_HOTPLUG
1490 
1491 /* This is used by the PCI hotplug driver to allocate resource
1492  * of newly plugged busses. We can try to consolidate with the
1493  * rest of the code later, for now, keep it as-is as our main
1494  * resource allocation function doesn't deal with sub-trees yet.
1495  */
1496 void pcibios_claim_one_bus(struct pci_bus *bus)
1497 {
1498 	struct pci_dev *dev;
1499 	struct pci_bus *child_bus;
1500 
1501 	list_for_each_entry(dev, &bus->devices, bus_list) {
1502 		int i;
1503 
1504 		for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1505 			struct resource *r = &dev->resource[i];
1506 
1507 			if (r->parent || !r->start || !r->flags)
1508 				continue;
1509 
1510 			pr_debug("PCI: Claiming %s: "
1511 				 "Resource %d: %016llx..%016llx [%x]\n",
1512 				 pci_name(dev), i,
1513 				 (unsigned long long)r->start,
1514 				 (unsigned long long)r->end,
1515 				 (unsigned int)r->flags);
1516 
1517 			pci_claim_resource(dev, i);
1518 		}
1519 	}
1520 
1521 	list_for_each_entry(child_bus, &bus->children, node)
1522 		pcibios_claim_one_bus(child_bus);
1523 }
1524 
1525 
1526 /* pcibios_finish_adding_to_bus
1527  *
1528  * This is to be called by the hotplug code after devices have been
1529  * added to a bus, this include calling it for a PHB that is just
1530  * being added
1531  */
1532 void pcibios_finish_adding_to_bus(struct pci_bus *bus)
1533 {
1534 	pr_debug("PCI: Finishing adding to hotplug bus %04x:%02x\n",
1535 		 pci_domain_nr(bus), bus->number);
1536 
1537 	/* Allocate bus and devices resources */
1538 	pcibios_allocate_bus_resources(bus);
1539 	pcibios_claim_one_bus(bus);
1540 
1541 	/* Add new devices to global lists.  Register in proc, sysfs. */
1542 	pci_bus_add_devices(bus);
1543 
1544 	/* Fixup EEH */
1545 	eeh_add_device_tree_late(bus);
1546 }
1547 EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
1548 
1549 #endif /* CONFIG_HOTPLUG */
1550 
1551 int pcibios_enable_device(struct pci_dev *dev, int mask)
1552 {
1553 	if (ppc_md.pcibios_enable_device_hook)
1554 		if (ppc_md.pcibios_enable_device_hook(dev))
1555 			return -EINVAL;
1556 
1557 	return pci_enable_resources(dev, mask);
1558 }
1559 
1560 void __devinit pcibios_setup_phb_resources(struct pci_controller *hose)
1561 {
1562 	struct pci_bus *bus = hose->bus;
1563 	struct resource *res;
1564 	int i;
1565 
1566 	/* Hookup PHB IO resource */
1567 	bus->resource[0] = res = &hose->io_resource;
1568 
1569 	if (!res->flags) {
1570 		printk(KERN_WARNING "PCI: I/O resource not set for host"
1571 		       " bridge %s (domain %d)\n",
1572 		       hose->dn->full_name, hose->global_number);
1573 #ifdef CONFIG_PPC32
1574 		/* Workaround for lack of IO resource only on 32-bit */
1575 		res->start = (unsigned long)hose->io_base_virt - isa_io_base;
1576 		res->end = res->start + IO_SPACE_LIMIT;
1577 		res->flags = IORESOURCE_IO;
1578 #endif /* CONFIG_PPC32 */
1579 	}
1580 
1581 	pr_debug("PCI: PHB IO resource    = %016llx-%016llx [%lx]\n",
1582 		 (unsigned long long)res->start,
1583 		 (unsigned long long)res->end,
1584 		 (unsigned long)res->flags);
1585 
1586 	/* Hookup PHB Memory resources */
1587 	for (i = 0; i < 3; ++i) {
1588 		res = &hose->mem_resources[i];
1589 		if (!res->flags) {
1590 			if (i > 0)
1591 				continue;
1592 			printk(KERN_ERR "PCI: Memory resource 0 not set for "
1593 			       "host bridge %s (domain %d)\n",
1594 			       hose->dn->full_name, hose->global_number);
1595 #ifdef CONFIG_PPC32
1596 			/* Workaround for lack of MEM resource only on 32-bit */
1597 			res->start = hose->pci_mem_offset;
1598 			res->end = (resource_size_t)-1LL;
1599 			res->flags = IORESOURCE_MEM;
1600 #endif /* CONFIG_PPC32 */
1601 		}
1602 		bus->resource[i+1] = res;
1603 
1604 		pr_debug("PCI: PHB MEM resource %d = %016llx-%016llx [%lx]\n", i,
1605 			 (unsigned long long)res->start,
1606 			 (unsigned long long)res->end,
1607 			 (unsigned long)res->flags);
1608 	}
1609 
1610 	pr_debug("PCI: PHB MEM offset     = %016llx\n",
1611 		 (unsigned long long)hose->pci_mem_offset);
1612 	pr_debug("PCI: PHB IO  offset     = %08lx\n",
1613 		 (unsigned long)hose->io_base_virt - _IO_BASE);
1614 
1615 }
1616 
1617 /*
1618  * Null PCI config access functions, for the case when we can't
1619  * find a hose.
1620  */
1621 #define NULL_PCI_OP(rw, size, type)					\
1622 static int								\
1623 null_##rw##_config_##size(struct pci_dev *dev, int offset, type val)	\
1624 {									\
1625 	return PCIBIOS_DEVICE_NOT_FOUND;    				\
1626 }
1627 
1628 static int
1629 null_read_config(struct pci_bus *bus, unsigned int devfn, int offset,
1630 		 int len, u32 *val)
1631 {
1632 	return PCIBIOS_DEVICE_NOT_FOUND;
1633 }
1634 
1635 static int
1636 null_write_config(struct pci_bus *bus, unsigned int devfn, int offset,
1637 		  int len, u32 val)
1638 {
1639 	return PCIBIOS_DEVICE_NOT_FOUND;
1640 }
1641 
1642 static struct pci_ops null_pci_ops =
1643 {
1644 	.read = null_read_config,
1645 	.write = null_write_config,
1646 };
1647 
1648 /*
1649  * These functions are used early on before PCI scanning is done
1650  * and all of the pci_dev and pci_bus structures have been created.
1651  */
1652 static struct pci_bus *
1653 fake_pci_bus(struct pci_controller *hose, int busnr)
1654 {
1655 	static struct pci_bus bus;
1656 
1657 	if (hose == 0) {
1658 		printk(KERN_ERR "Can't find hose for PCI bus %d!\n", busnr);
1659 	}
1660 	bus.number = busnr;
1661 	bus.sysdata = hose;
1662 	bus.ops = hose? hose->ops: &null_pci_ops;
1663 	return &bus;
1664 }
1665 
1666 #define EARLY_PCI_OP(rw, size, type)					\
1667 int early_##rw##_config_##size(struct pci_controller *hose, int bus,	\
1668 			       int devfn, int offset, type value)	\
1669 {									\
1670 	return pci_bus_##rw##_config_##size(fake_pci_bus(hose, bus),	\
1671 					    devfn, offset, value);	\
1672 }
1673 
1674 EARLY_PCI_OP(read, byte, u8 *)
1675 EARLY_PCI_OP(read, word, u16 *)
1676 EARLY_PCI_OP(read, dword, u32 *)
1677 EARLY_PCI_OP(write, byte, u8)
1678 EARLY_PCI_OP(write, word, u16)
1679 EARLY_PCI_OP(write, dword, u32)
1680 
1681 extern int pci_bus_find_capability (struct pci_bus *bus, unsigned int devfn, int cap);
1682 int early_find_capability(struct pci_controller *hose, int bus, int devfn,
1683 			  int cap)
1684 {
1685 	return pci_bus_find_capability(fake_pci_bus(hose, bus), devfn, cap);
1686 }
1687 
1688 /**
1689  * pci_scan_phb - Given a pci_controller, setup and scan the PCI bus
1690  * @hose: Pointer to the PCI host controller instance structure
1691  */
1692 void __devinit pcibios_scan_phb(struct pci_controller *hose)
1693 {
1694 	struct pci_bus *bus;
1695 	struct device_node *node = hose->dn;
1696 	int mode;
1697 
1698 	pr_debug("PCI: Scanning PHB %s\n",
1699 		 node ? node->full_name : "<NO NAME>");
1700 
1701 	/* Create an empty bus for the toplevel */
1702 	bus = pci_create_bus(hose->parent, hose->first_busno, hose->ops, hose);
1703 	if (bus == NULL) {
1704 		pr_err("Failed to create bus for PCI domain %04x\n",
1705 			hose->global_number);
1706 		return;
1707 	}
1708 	bus->dev.of_node = of_node_get(node);
1709 	bus->secondary = hose->first_busno;
1710 	hose->bus = bus;
1711 
1712 	/* Get some IO space for the new PHB */
1713 	pcibios_setup_phb_io_space(hose);
1714 
1715 	/* Wire up PHB bus resources */
1716 	pcibios_setup_phb_resources(hose);
1717 
1718 	/* Get probe mode and perform scan */
1719 	mode = PCI_PROBE_NORMAL;
1720 	if (node && ppc_md.pci_probe_mode)
1721 		mode = ppc_md.pci_probe_mode(bus);
1722 	pr_debug("    probe mode: %d\n", mode);
1723 	if (mode == PCI_PROBE_DEVTREE) {
1724 		bus->subordinate = hose->last_busno;
1725 		of_scan_bus(node, bus);
1726 	}
1727 
1728 	if (mode == PCI_PROBE_NORMAL)
1729 		hose->last_busno = bus->subordinate = pci_scan_child_bus(bus);
1730 }
1731