xref: /linux/arch/powerpc/kernel/pci-common.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
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/export.h>
25 #include <linux/of_address.h>
26 #include <linux/of_pci.h>
27 #include <linux/mm.h>
28 #include <linux/list.h>
29 #include <linux/syscalls.h>
30 #include <linux/irq.h>
31 #include <linux/vmalloc.h>
32 #include <linux/slab.h>
33 
34 #include <asm/processor.h>
35 #include <asm/io.h>
36 #include <asm/prom.h>
37 #include <asm/pci-bridge.h>
38 #include <asm/byteorder.h>
39 #include <asm/machdep.h>
40 #include <asm/ppc-pci.h>
41 #include <asm/firmware.h>
42 #include <asm/eeh.h>
43 
44 static DEFINE_SPINLOCK(hose_spinlock);
45 LIST_HEAD(hose_list);
46 
47 /* XXX kill that some day ... */
48 static int global_phb_number;		/* Global phb counter */
49 
50 /* ISA Memory physical address */
51 resource_size_t isa_mem_base;
52 
53 /* Default PCI flags is 0 on ppc32, modified at boot on ppc64 */
54 unsigned int pci_flags = 0;
55 
56 
57 static struct dma_map_ops *pci_dma_ops = &dma_direct_ops;
58 
59 void set_pci_dma_ops(struct dma_map_ops *dma_ops)
60 {
61 	pci_dma_ops = dma_ops;
62 }
63 
64 struct dma_map_ops *get_pci_dma_ops(void)
65 {
66 	return pci_dma_ops;
67 }
68 EXPORT_SYMBOL(get_pci_dma_ops);
69 
70 struct pci_controller *pcibios_alloc_controller(struct device_node *dev)
71 {
72 	struct pci_controller *phb;
73 
74 	phb = zalloc_maybe_bootmem(sizeof(struct pci_controller), GFP_KERNEL);
75 	if (phb == NULL)
76 		return NULL;
77 	spin_lock(&hose_spinlock);
78 	phb->global_number = global_phb_number++;
79 	list_add_tail(&phb->list_node, &hose_list);
80 	spin_unlock(&hose_spinlock);
81 	phb->dn = dev;
82 	phb->is_dynamic = mem_init_done;
83 #ifdef CONFIG_PPC64
84 	if (dev) {
85 		int nid = of_node_to_nid(dev);
86 
87 		if (nid < 0 || !node_online(nid))
88 			nid = -1;
89 
90 		PHB_SET_NODE(phb, nid);
91 	}
92 #endif
93 	return phb;
94 }
95 
96 void pcibios_free_controller(struct pci_controller *phb)
97 {
98 	spin_lock(&hose_spinlock);
99 	list_del(&phb->list_node);
100 	spin_unlock(&hose_spinlock);
101 
102 	if (phb->is_dynamic)
103 		kfree(phb);
104 }
105 
106 static resource_size_t pcibios_io_size(const struct pci_controller *hose)
107 {
108 #ifdef CONFIG_PPC64
109 	return hose->pci_io_size;
110 #else
111 	return resource_size(&hose->io_resource);
112 #endif
113 }
114 
115 int pcibios_vaddr_is_ioport(void __iomem *address)
116 {
117 	int ret = 0;
118 	struct pci_controller *hose;
119 	resource_size_t size;
120 
121 	spin_lock(&hose_spinlock);
122 	list_for_each_entry(hose, &hose_list, list_node) {
123 		size = pcibios_io_size(hose);
124 		if (address >= hose->io_base_virt &&
125 		    address < (hose->io_base_virt + size)) {
126 			ret = 1;
127 			break;
128 		}
129 	}
130 	spin_unlock(&hose_spinlock);
131 	return ret;
132 }
133 
134 unsigned long pci_address_to_pio(phys_addr_t address)
135 {
136 	struct pci_controller *hose;
137 	resource_size_t size;
138 	unsigned long ret = ~0;
139 
140 	spin_lock(&hose_spinlock);
141 	list_for_each_entry(hose, &hose_list, list_node) {
142 		size = pcibios_io_size(hose);
143 		if (address >= hose->io_base_phys &&
144 		    address < (hose->io_base_phys + size)) {
145 			unsigned long base =
146 				(unsigned long)hose->io_base_virt - _IO_BASE;
147 			ret = base + (address - hose->io_base_phys);
148 			break;
149 		}
150 	}
151 	spin_unlock(&hose_spinlock);
152 
153 	return ret;
154 }
155 EXPORT_SYMBOL_GPL(pci_address_to_pio);
156 
157 /*
158  * Return the domain number for this bus.
159  */
160 int pci_domain_nr(struct pci_bus *bus)
161 {
162 	struct pci_controller *hose = pci_bus_to_host(bus);
163 
164 	return hose->global_number;
165 }
166 EXPORT_SYMBOL(pci_domain_nr);
167 
168 /* This routine is meant to be used early during boot, when the
169  * PCI bus numbers have not yet been assigned, and you need to
170  * issue PCI config cycles to an OF device.
171  * It could also be used to "fix" RTAS config cycles if you want
172  * to set pci_assign_all_buses to 1 and still use RTAS for PCI
173  * config cycles.
174  */
175 struct pci_controller* pci_find_hose_for_OF_device(struct device_node* node)
176 {
177 	while(node) {
178 		struct pci_controller *hose, *tmp;
179 		list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
180 			if (hose->dn == node)
181 				return hose;
182 		node = node->parent;
183 	}
184 	return NULL;
185 }
186 
187 static ssize_t pci_show_devspec(struct device *dev,
188 		struct device_attribute *attr, char *buf)
189 {
190 	struct pci_dev *pdev;
191 	struct device_node *np;
192 
193 	pdev = to_pci_dev (dev);
194 	np = pci_device_to_OF_node(pdev);
195 	if (np == NULL || np->full_name == NULL)
196 		return 0;
197 	return sprintf(buf, "%s", np->full_name);
198 }
199 static DEVICE_ATTR(devspec, S_IRUGO, pci_show_devspec, NULL);
200 
201 /* Add sysfs properties */
202 int pcibios_add_platform_entries(struct pci_dev *pdev)
203 {
204 	return device_create_file(&pdev->dev, &dev_attr_devspec);
205 }
206 
207 char __devinit *pcibios_setup(char *str)
208 {
209 	return str;
210 }
211 
212 /*
213  * Reads the interrupt pin to determine if interrupt is use by card.
214  * If the interrupt is used, then gets the interrupt line from the
215  * openfirmware and sets it in the pci_dev and pci_config line.
216  */
217 int pci_read_irq_line(struct pci_dev *pci_dev)
218 {
219 	struct of_irq oirq;
220 	unsigned int virq;
221 
222 	/* The current device-tree that iSeries generates from the HV
223 	 * PCI informations doesn't contain proper interrupt routing,
224 	 * and all the fallback would do is print out crap, so we
225 	 * don't attempt to resolve the interrupts here at all, some
226 	 * iSeries specific fixup does it.
227 	 *
228 	 * In the long run, we will hopefully fix the generated device-tree
229 	 * instead.
230 	 */
231 #ifdef CONFIG_PPC_ISERIES
232 	if (firmware_has_feature(FW_FEATURE_ISERIES))
233 		return -1;
234 #endif
235 
236 	pr_debug("PCI: Try to map irq for %s...\n", pci_name(pci_dev));
237 
238 #ifdef DEBUG
239 	memset(&oirq, 0xff, sizeof(oirq));
240 #endif
241 	/* Try to get a mapping from the device-tree */
242 	if (of_irq_map_pci(pci_dev, &oirq)) {
243 		u8 line, pin;
244 
245 		/* If that fails, lets fallback to what is in the config
246 		 * space and map that through the default controller. We
247 		 * also set the type to level low since that's what PCI
248 		 * interrupts are. If your platform does differently, then
249 		 * either provide a proper interrupt tree or don't use this
250 		 * function.
251 		 */
252 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_PIN, &pin))
253 			return -1;
254 		if (pin == 0)
255 			return -1;
256 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_LINE, &line) ||
257 		    line == 0xff || line == 0) {
258 			return -1;
259 		}
260 		pr_debug(" No map ! Using line %d (pin %d) from PCI config\n",
261 			 line, pin);
262 
263 		virq = irq_create_mapping(NULL, line);
264 		if (virq != NO_IRQ)
265 			irq_set_irq_type(virq, IRQ_TYPE_LEVEL_LOW);
266 	} else {
267 		pr_debug(" Got one, spec %d cells (0x%08x 0x%08x...) on %s\n",
268 			 oirq.size, oirq.specifier[0], oirq.specifier[1],
269 			 oirq.controller ? oirq.controller->full_name :
270 			 "<default>");
271 
272 		virq = irq_create_of_mapping(oirq.controller, oirq.specifier,
273 					     oirq.size);
274 	}
275 	if(virq == NO_IRQ) {
276 		pr_debug(" Failed to map !\n");
277 		return -1;
278 	}
279 
280 	pr_debug(" Mapped to linux irq %d\n", virq);
281 
282 	pci_dev->irq = virq;
283 
284 	return 0;
285 }
286 EXPORT_SYMBOL(pci_read_irq_line);
287 
288 /*
289  * Platform support for /proc/bus/pci/X/Y mmap()s,
290  * modelled on the sparc64 implementation by Dave Miller.
291  *  -- paulus.
292  */
293 
294 /*
295  * Adjust vm_pgoff of VMA such that it is the physical page offset
296  * corresponding to the 32-bit pci bus offset for DEV requested by the user.
297  *
298  * Basically, the user finds the base address for his device which he wishes
299  * to mmap.  They read the 32-bit value from the config space base register,
300  * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
301  * offset parameter of mmap on /proc/bus/pci/XXX for that device.
302  *
303  * Returns negative error code on failure, zero on success.
304  */
305 static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
306 					       resource_size_t *offset,
307 					       enum pci_mmap_state mmap_state)
308 {
309 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
310 	unsigned long io_offset = 0;
311 	int i, res_bit;
312 
313 	if (hose == 0)
314 		return NULL;		/* should never happen */
315 
316 	/* If memory, add on the PCI bridge address offset */
317 	if (mmap_state == pci_mmap_mem) {
318 #if 0 /* See comment in pci_resource_to_user() for why this is disabled */
319 		*offset += hose->pci_mem_offset;
320 #endif
321 		res_bit = IORESOURCE_MEM;
322 	} else {
323 		io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
324 		*offset += io_offset;
325 		res_bit = IORESOURCE_IO;
326 	}
327 
328 	/*
329 	 * Check that the offset requested corresponds to one of the
330 	 * resources of the device.
331 	 */
332 	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
333 		struct resource *rp = &dev->resource[i];
334 		int flags = rp->flags;
335 
336 		/* treat ROM as memory (should be already) */
337 		if (i == PCI_ROM_RESOURCE)
338 			flags |= IORESOURCE_MEM;
339 
340 		/* Active and same type? */
341 		if ((flags & res_bit) == 0)
342 			continue;
343 
344 		/* In the range of this resource? */
345 		if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
346 			continue;
347 
348 		/* found it! construct the final physical address */
349 		if (mmap_state == pci_mmap_io)
350 			*offset += hose->io_base_phys - io_offset;
351 		return rp;
352 	}
353 
354 	return NULL;
355 }
356 
357 /*
358  * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
359  * device mapping.
360  */
361 static pgprot_t __pci_mmap_set_pgprot(struct pci_dev *dev, struct resource *rp,
362 				      pgprot_t protection,
363 				      enum pci_mmap_state mmap_state,
364 				      int write_combine)
365 {
366 	unsigned long prot = pgprot_val(protection);
367 
368 	/* Write combine is always 0 on non-memory space mappings. On
369 	 * memory space, if the user didn't pass 1, we check for a
370 	 * "prefetchable" resource. This is a bit hackish, but we use
371 	 * this to workaround the inability of /sysfs to provide a write
372 	 * combine bit
373 	 */
374 	if (mmap_state != pci_mmap_mem)
375 		write_combine = 0;
376 	else if (write_combine == 0) {
377 		if (rp->flags & IORESOURCE_PREFETCH)
378 			write_combine = 1;
379 	}
380 
381 	/* XXX would be nice to have a way to ask for write-through */
382 	if (write_combine)
383 		return pgprot_noncached_wc(prot);
384 	else
385 		return pgprot_noncached(prot);
386 }
387 
388 /*
389  * This one is used by /dev/mem and fbdev who have no clue about the
390  * PCI device, it tries to find the PCI device first and calls the
391  * above routine
392  */
393 pgprot_t pci_phys_mem_access_prot(struct file *file,
394 				  unsigned long pfn,
395 				  unsigned long size,
396 				  pgprot_t prot)
397 {
398 	struct pci_dev *pdev = NULL;
399 	struct resource *found = NULL;
400 	resource_size_t offset = ((resource_size_t)pfn) << PAGE_SHIFT;
401 	int i;
402 
403 	if (page_is_ram(pfn))
404 		return prot;
405 
406 	prot = pgprot_noncached(prot);
407 	for_each_pci_dev(pdev) {
408 		for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
409 			struct resource *rp = &pdev->resource[i];
410 			int flags = rp->flags;
411 
412 			/* Active and same type? */
413 			if ((flags & IORESOURCE_MEM) == 0)
414 				continue;
415 			/* In the range of this resource? */
416 			if (offset < (rp->start & PAGE_MASK) ||
417 			    offset > rp->end)
418 				continue;
419 			found = rp;
420 			break;
421 		}
422 		if (found)
423 			break;
424 	}
425 	if (found) {
426 		if (found->flags & IORESOURCE_PREFETCH)
427 			prot = pgprot_noncached_wc(prot);
428 		pci_dev_put(pdev);
429 	}
430 
431 	pr_debug("PCI: Non-PCI map for %llx, prot: %lx\n",
432 		 (unsigned long long)offset, pgprot_val(prot));
433 
434 	return prot;
435 }
436 
437 
438 /*
439  * Perform the actual remap of the pages for a PCI device mapping, as
440  * appropriate for this architecture.  The region in the process to map
441  * is described by vm_start and vm_end members of VMA, the base physical
442  * address is found in vm_pgoff.
443  * The pci device structure is provided so that architectures may make mapping
444  * decisions on a per-device or per-bus basis.
445  *
446  * Returns a negative error code on failure, zero on success.
447  */
448 int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
449 			enum pci_mmap_state mmap_state, int write_combine)
450 {
451 	resource_size_t offset =
452 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
453 	struct resource *rp;
454 	int ret;
455 
456 	rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
457 	if (rp == NULL)
458 		return -EINVAL;
459 
460 	vma->vm_pgoff = offset >> PAGE_SHIFT;
461 	vma->vm_page_prot = __pci_mmap_set_pgprot(dev, rp,
462 						  vma->vm_page_prot,
463 						  mmap_state, write_combine);
464 
465 	ret = remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
466 			       vma->vm_end - vma->vm_start, vma->vm_page_prot);
467 
468 	return ret;
469 }
470 
471 /* This provides legacy IO read access on a bus */
472 int pci_legacy_read(struct pci_bus *bus, loff_t port, u32 *val, size_t size)
473 {
474 	unsigned long offset;
475 	struct pci_controller *hose = pci_bus_to_host(bus);
476 	struct resource *rp = &hose->io_resource;
477 	void __iomem *addr;
478 
479 	/* Check if port can be supported by that bus. We only check
480 	 * the ranges of the PHB though, not the bus itself as the rules
481 	 * for forwarding legacy cycles down bridges are not our problem
482 	 * here. So if the host bridge supports it, we do it.
483 	 */
484 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
485 	offset += port;
486 
487 	if (!(rp->flags & IORESOURCE_IO))
488 		return -ENXIO;
489 	if (offset < rp->start || (offset + size) > rp->end)
490 		return -ENXIO;
491 	addr = hose->io_base_virt + port;
492 
493 	switch(size) {
494 	case 1:
495 		*((u8 *)val) = in_8(addr);
496 		return 1;
497 	case 2:
498 		if (port & 1)
499 			return -EINVAL;
500 		*((u16 *)val) = in_le16(addr);
501 		return 2;
502 	case 4:
503 		if (port & 3)
504 			return -EINVAL;
505 		*((u32 *)val) = in_le32(addr);
506 		return 4;
507 	}
508 	return -EINVAL;
509 }
510 
511 /* This provides legacy IO write access on a bus */
512 int pci_legacy_write(struct pci_bus *bus, loff_t port, u32 val, size_t size)
513 {
514 	unsigned long offset;
515 	struct pci_controller *hose = pci_bus_to_host(bus);
516 	struct resource *rp = &hose->io_resource;
517 	void __iomem *addr;
518 
519 	/* Check if port can be supported by that bus. We only check
520 	 * the ranges of the PHB though, not the bus itself as the rules
521 	 * for forwarding legacy cycles down bridges are not our problem
522 	 * here. So if the host bridge supports it, we do it.
523 	 */
524 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
525 	offset += port;
526 
527 	if (!(rp->flags & IORESOURCE_IO))
528 		return -ENXIO;
529 	if (offset < rp->start || (offset + size) > rp->end)
530 		return -ENXIO;
531 	addr = hose->io_base_virt + port;
532 
533 	/* WARNING: The generic code is idiotic. It gets passed a pointer
534 	 * to what can be a 1, 2 or 4 byte quantity and always reads that
535 	 * as a u32, which means that we have to correct the location of
536 	 * the data read within those 32 bits for size 1 and 2
537 	 */
538 	switch(size) {
539 	case 1:
540 		out_8(addr, val >> 24);
541 		return 1;
542 	case 2:
543 		if (port & 1)
544 			return -EINVAL;
545 		out_le16(addr, val >> 16);
546 		return 2;
547 	case 4:
548 		if (port & 3)
549 			return -EINVAL;
550 		out_le32(addr, val);
551 		return 4;
552 	}
553 	return -EINVAL;
554 }
555 
556 /* This provides legacy IO or memory mmap access on a bus */
557 int pci_mmap_legacy_page_range(struct pci_bus *bus,
558 			       struct vm_area_struct *vma,
559 			       enum pci_mmap_state mmap_state)
560 {
561 	struct pci_controller *hose = pci_bus_to_host(bus);
562 	resource_size_t offset =
563 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
564 	resource_size_t size = vma->vm_end - vma->vm_start;
565 	struct resource *rp;
566 
567 	pr_debug("pci_mmap_legacy_page_range(%04x:%02x, %s @%llx..%llx)\n",
568 		 pci_domain_nr(bus), bus->number,
569 		 mmap_state == pci_mmap_mem ? "MEM" : "IO",
570 		 (unsigned long long)offset,
571 		 (unsigned long long)(offset + size - 1));
572 
573 	if (mmap_state == pci_mmap_mem) {
574 		/* Hack alert !
575 		 *
576 		 * Because X is lame and can fail starting if it gets an error trying
577 		 * to mmap legacy_mem (instead of just moving on without legacy memory
578 		 * access) we fake it here by giving it anonymous memory, effectively
579 		 * behaving just like /dev/zero
580 		 */
581 		if ((offset + size) > hose->isa_mem_size) {
582 			printk(KERN_DEBUG
583 			       "Process %s (pid:%d) mapped non-existing PCI legacy memory for 0%04x:%02x\n",
584 			       current->comm, current->pid, pci_domain_nr(bus), bus->number);
585 			if (vma->vm_flags & VM_SHARED)
586 				return shmem_zero_setup(vma);
587 			return 0;
588 		}
589 		offset += hose->isa_mem_phys;
590 	} else {
591 		unsigned long io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
592 		unsigned long roffset = offset + io_offset;
593 		rp = &hose->io_resource;
594 		if (!(rp->flags & IORESOURCE_IO))
595 			return -ENXIO;
596 		if (roffset < rp->start || (roffset + size) > rp->end)
597 			return -ENXIO;
598 		offset += hose->io_base_phys;
599 	}
600 	pr_debug(" -> mapping phys %llx\n", (unsigned long long)offset);
601 
602 	vma->vm_pgoff = offset >> PAGE_SHIFT;
603 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
604 	return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
605 			       vma->vm_end - vma->vm_start,
606 			       vma->vm_page_prot);
607 }
608 
609 void pci_resource_to_user(const struct pci_dev *dev, int bar,
610 			  const struct resource *rsrc,
611 			  resource_size_t *start, resource_size_t *end)
612 {
613 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
614 	resource_size_t offset = 0;
615 
616 	if (hose == NULL)
617 		return;
618 
619 	if (rsrc->flags & IORESOURCE_IO)
620 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
621 
622 	/* We pass a fully fixed up address to userland for MMIO instead of
623 	 * a BAR value because X is lame and expects to be able to use that
624 	 * to pass to /dev/mem !
625 	 *
626 	 * That means that we'll have potentially 64 bits values where some
627 	 * userland apps only expect 32 (like X itself since it thinks only
628 	 * Sparc has 64 bits MMIO) but if we don't do that, we break it on
629 	 * 32 bits CHRPs :-(
630 	 *
631 	 * Hopefully, the sysfs insterface is immune to that gunk. Once X
632 	 * has been fixed (and the fix spread enough), we can re-enable the
633 	 * 2 lines below and pass down a BAR value to userland. In that case
634 	 * we'll also have to re-enable the matching code in
635 	 * __pci_mmap_make_offset().
636 	 *
637 	 * BenH.
638 	 */
639 #if 0
640 	else if (rsrc->flags & IORESOURCE_MEM)
641 		offset = hose->pci_mem_offset;
642 #endif
643 
644 	*start = rsrc->start - offset;
645 	*end = rsrc->end - offset;
646 }
647 
648 /**
649  * pci_process_bridge_OF_ranges - Parse PCI bridge resources from device tree
650  * @hose: newly allocated pci_controller to be setup
651  * @dev: device node of the host bridge
652  * @primary: set if primary bus (32 bits only, soon to be deprecated)
653  *
654  * This function will parse the "ranges" property of a PCI host bridge device
655  * node and setup the resource mapping of a pci controller based on its
656  * content.
657  *
658  * Life would be boring if it wasn't for a few issues that we have to deal
659  * with here:
660  *
661  *   - We can only cope with one IO space range and up to 3 Memory space
662  *     ranges. However, some machines (thanks Apple !) tend to split their
663  *     space into lots of small contiguous ranges. So we have to coalesce.
664  *
665  *   - We can only cope with all memory ranges having the same offset
666  *     between CPU addresses and PCI addresses. Unfortunately, some bridges
667  *     are setup for a large 1:1 mapping along with a small "window" which
668  *     maps PCI address 0 to some arbitrary high address of the CPU space in
669  *     order to give access to the ISA memory hole.
670  *     The way out of here that I've chosen for now is to always set the
671  *     offset based on the first resource found, then override it if we
672  *     have a different offset and the previous was set by an ISA hole.
673  *
674  *   - Some busses have IO space not starting at 0, which causes trouble with
675  *     the way we do our IO resource renumbering. The code somewhat deals with
676  *     it for 64 bits but I would expect problems on 32 bits.
677  *
678  *   - Some 32 bits platforms such as 4xx can have physical space larger than
679  *     32 bits so we need to use 64 bits values for the parsing
680  */
681 void __devinit pci_process_bridge_OF_ranges(struct pci_controller *hose,
682 					    struct device_node *dev,
683 					    int primary)
684 {
685 	const u32 *ranges;
686 	int rlen;
687 	int pna = of_n_addr_cells(dev);
688 	int np = pna + 5;
689 	int memno = 0, isa_hole = -1;
690 	u32 pci_space;
691 	unsigned long long pci_addr, cpu_addr, pci_next, cpu_next, size;
692 	unsigned long long isa_mb = 0;
693 	struct resource *res;
694 
695 	printk(KERN_INFO "PCI host bridge %s %s ranges:\n",
696 	       dev->full_name, primary ? "(primary)" : "");
697 
698 	/* Get ranges property */
699 	ranges = of_get_property(dev, "ranges", &rlen);
700 	if (ranges == NULL)
701 		return;
702 
703 	/* Parse it */
704 	while ((rlen -= np * 4) >= 0) {
705 		/* Read next ranges element */
706 		pci_space = ranges[0];
707 		pci_addr = of_read_number(ranges + 1, 2);
708 		cpu_addr = of_translate_address(dev, ranges + 3);
709 		size = of_read_number(ranges + pna + 3, 2);
710 		ranges += np;
711 
712 		/* If we failed translation or got a zero-sized region
713 		 * (some FW try to feed us with non sensical zero sized regions
714 		 * such as power3 which look like some kind of attempt at exposing
715 		 * the VGA memory hole)
716 		 */
717 		if (cpu_addr == OF_BAD_ADDR || size == 0)
718 			continue;
719 
720 		/* Now consume following elements while they are contiguous */
721 		for (; rlen >= np * sizeof(u32);
722 		     ranges += np, rlen -= np * 4) {
723 			if (ranges[0] != pci_space)
724 				break;
725 			pci_next = of_read_number(ranges + 1, 2);
726 			cpu_next = of_translate_address(dev, ranges + 3);
727 			if (pci_next != pci_addr + size ||
728 			    cpu_next != cpu_addr + size)
729 				break;
730 			size += of_read_number(ranges + pna + 3, 2);
731 		}
732 
733 		/* Act based on address space type */
734 		res = NULL;
735 		switch ((pci_space >> 24) & 0x3) {
736 		case 1:		/* PCI IO space */
737 			printk(KERN_INFO
738 			       "  IO 0x%016llx..0x%016llx -> 0x%016llx\n",
739 			       cpu_addr, cpu_addr + size - 1, pci_addr);
740 
741 			/* We support only one IO range */
742 			if (hose->pci_io_size) {
743 				printk(KERN_INFO
744 				       " \\--> Skipped (too many) !\n");
745 				continue;
746 			}
747 #ifdef CONFIG_PPC32
748 			/* On 32 bits, limit I/O space to 16MB */
749 			if (size > 0x01000000)
750 				size = 0x01000000;
751 
752 			/* 32 bits needs to map IOs here */
753 			hose->io_base_virt = ioremap(cpu_addr, size);
754 
755 			/* Expect trouble if pci_addr is not 0 */
756 			if (primary)
757 				isa_io_base =
758 					(unsigned long)hose->io_base_virt;
759 #endif /* CONFIG_PPC32 */
760 			/* pci_io_size and io_base_phys always represent IO
761 			 * space starting at 0 so we factor in pci_addr
762 			 */
763 			hose->pci_io_size = pci_addr + size;
764 			hose->io_base_phys = cpu_addr - pci_addr;
765 
766 			/* Build resource */
767 			res = &hose->io_resource;
768 			res->flags = IORESOURCE_IO;
769 			res->start = pci_addr;
770 			break;
771 		case 2:		/* PCI Memory space */
772 		case 3:		/* PCI 64 bits Memory space */
773 			printk(KERN_INFO
774 			       " MEM 0x%016llx..0x%016llx -> 0x%016llx %s\n",
775 			       cpu_addr, cpu_addr + size - 1, pci_addr,
776 			       (pci_space & 0x40000000) ? "Prefetch" : "");
777 
778 			/* We support only 3 memory ranges */
779 			if (memno >= 3) {
780 				printk(KERN_INFO
781 				       " \\--> Skipped (too many) !\n");
782 				continue;
783 			}
784 			/* Handles ISA memory hole space here */
785 			if (pci_addr == 0) {
786 				isa_mb = cpu_addr;
787 				isa_hole = memno;
788 				if (primary || isa_mem_base == 0)
789 					isa_mem_base = cpu_addr;
790 				hose->isa_mem_phys = cpu_addr;
791 				hose->isa_mem_size = size;
792 			}
793 
794 			/* We get the PCI/Mem offset from the first range or
795 			 * the, current one if the offset came from an ISA
796 			 * hole. If they don't match, bugger.
797 			 */
798 			if (memno == 0 ||
799 			    (isa_hole >= 0 && pci_addr != 0 &&
800 			     hose->pci_mem_offset == isa_mb))
801 				hose->pci_mem_offset = cpu_addr - pci_addr;
802 			else if (pci_addr != 0 &&
803 				 hose->pci_mem_offset != cpu_addr - pci_addr) {
804 				printk(KERN_INFO
805 				       " \\--> Skipped (offset mismatch) !\n");
806 				continue;
807 			}
808 
809 			/* Build resource */
810 			res = &hose->mem_resources[memno++];
811 			res->flags = IORESOURCE_MEM;
812 			if (pci_space & 0x40000000)
813 				res->flags |= IORESOURCE_PREFETCH;
814 			res->start = cpu_addr;
815 			break;
816 		}
817 		if (res != NULL) {
818 			res->name = dev->full_name;
819 			res->end = res->start + size - 1;
820 			res->parent = NULL;
821 			res->sibling = NULL;
822 			res->child = NULL;
823 		}
824 	}
825 
826 	/* If there's an ISA hole and the pci_mem_offset is -not- matching
827 	 * the ISA hole offset, then we need to remove the ISA hole from
828 	 * the resource list for that brige
829 	 */
830 	if (isa_hole >= 0 && hose->pci_mem_offset != isa_mb) {
831 		unsigned int next = isa_hole + 1;
832 		printk(KERN_INFO " Removing ISA hole at 0x%016llx\n", isa_mb);
833 		if (next < memno)
834 			memmove(&hose->mem_resources[isa_hole],
835 				&hose->mem_resources[next],
836 				sizeof(struct resource) * (memno - next));
837 		hose->mem_resources[--memno].flags = 0;
838 	}
839 }
840 
841 /* Decide whether to display the domain number in /proc */
842 int pci_proc_domain(struct pci_bus *bus)
843 {
844 	struct pci_controller *hose = pci_bus_to_host(bus);
845 
846 	if (!pci_has_flag(PCI_ENABLE_PROC_DOMAINS))
847 		return 0;
848 	if (pci_has_flag(PCI_COMPAT_DOMAIN_0))
849 		return hose->global_number != 0;
850 	return 1;
851 }
852 
853 void pcibios_resource_to_bus(struct pci_dev *dev, struct pci_bus_region *region,
854 			     struct resource *res)
855 {
856 	resource_size_t offset = 0, mask = (resource_size_t)-1;
857 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
858 
859 	if (!hose)
860 		return;
861 	if (res->flags & IORESOURCE_IO) {
862 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
863 		mask = 0xffffffffu;
864 	} else if (res->flags & IORESOURCE_MEM)
865 		offset = hose->pci_mem_offset;
866 
867 	region->start = (res->start - offset) & mask;
868 	region->end = (res->end - offset) & mask;
869 }
870 EXPORT_SYMBOL(pcibios_resource_to_bus);
871 
872 void pcibios_bus_to_resource(struct pci_dev *dev, struct resource *res,
873 			     struct pci_bus_region *region)
874 {
875 	resource_size_t offset = 0, mask = (resource_size_t)-1;
876 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
877 
878 	if (!hose)
879 		return;
880 	if (res->flags & IORESOURCE_IO) {
881 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
882 		mask = 0xffffffffu;
883 	} else if (res->flags & IORESOURCE_MEM)
884 		offset = hose->pci_mem_offset;
885 	res->start = (region->start + offset) & mask;
886 	res->end = (region->end + offset) & mask;
887 }
888 EXPORT_SYMBOL(pcibios_bus_to_resource);
889 
890 /* Fixup a bus resource into a linux resource */
891 static void __devinit fixup_resource(struct resource *res, struct pci_dev *dev)
892 {
893 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
894 	resource_size_t offset = 0, mask = (resource_size_t)-1;
895 
896 	if (res->flags & IORESOURCE_IO) {
897 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
898 		mask = 0xffffffffu;
899 	} else if (res->flags & IORESOURCE_MEM)
900 		offset = hose->pci_mem_offset;
901 
902 	res->start = (res->start + offset) & mask;
903 	res->end = (res->end + offset) & mask;
904 }
905 
906 
907 /* This header fixup will do the resource fixup for all devices as they are
908  * probed, but not for bridge ranges
909  */
910 static void __devinit pcibios_fixup_resources(struct pci_dev *dev)
911 {
912 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
913 	int i;
914 
915 	if (!hose) {
916 		printk(KERN_ERR "No host bridge for PCI dev %s !\n",
917 		       pci_name(dev));
918 		return;
919 	}
920 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
921 		struct resource *res = dev->resource + i;
922 		if (!res->flags)
923 			continue;
924 		/* On platforms that have PCI_PROBE_ONLY set, we don't
925 		 * consider 0 as an unassigned BAR value. It's technically
926 		 * a valid value, but linux doesn't like it... so when we can
927 		 * re-assign things, we do so, but if we can't, we keep it
928 		 * around and hope for the best...
929 		 */
930 		if (res->start == 0 && !pci_has_flag(PCI_PROBE_ONLY)) {
931 			pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] is unassigned\n",
932 				 pci_name(dev), i,
933 				 (unsigned long long)res->start,
934 				 (unsigned long long)res->end,
935 				 (unsigned int)res->flags);
936 			res->end -= res->start;
937 			res->start = 0;
938 			res->flags |= IORESOURCE_UNSET;
939 			continue;
940 		}
941 
942 		pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] fixup...\n",
943 			 pci_name(dev), i,
944 			 (unsigned long long)res->start,\
945 			 (unsigned long long)res->end,
946 			 (unsigned int)res->flags);
947 
948 		fixup_resource(res, dev);
949 
950 		pr_debug("PCI:%s            %016llx-%016llx\n",
951 			 pci_name(dev),
952 			 (unsigned long long)res->start,
953 			 (unsigned long long)res->end);
954 	}
955 
956 	/* Call machine specific resource fixup */
957 	if (ppc_md.pcibios_fixup_resources)
958 		ppc_md.pcibios_fixup_resources(dev);
959 }
960 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_resources);
961 
962 /* This function tries to figure out if a bridge resource has been initialized
963  * by the firmware or not. It doesn't have to be absolutely bullet proof, but
964  * things go more smoothly when it gets it right. It should covers cases such
965  * as Apple "closed" bridge resources and bare-metal pSeries unassigned bridges
966  */
967 static int __devinit pcibios_uninitialized_bridge_resource(struct pci_bus *bus,
968 							   struct resource *res)
969 {
970 	struct pci_controller *hose = pci_bus_to_host(bus);
971 	struct pci_dev *dev = bus->self;
972 	resource_size_t offset;
973 	u16 command;
974 	int i;
975 
976 	/* We don't do anything if PCI_PROBE_ONLY is set */
977 	if (pci_has_flag(PCI_PROBE_ONLY))
978 		return 0;
979 
980 	/* Job is a bit different between memory and IO */
981 	if (res->flags & IORESOURCE_MEM) {
982 		/* If the BAR is non-0 (res != pci_mem_offset) then it's probably been
983 		 * initialized by somebody
984 		 */
985 		if (res->start != hose->pci_mem_offset)
986 			return 0;
987 
988 		/* The BAR is 0, let's check if memory decoding is enabled on
989 		 * the bridge. If not, we consider it unassigned
990 		 */
991 		pci_read_config_word(dev, PCI_COMMAND, &command);
992 		if ((command & PCI_COMMAND_MEMORY) == 0)
993 			return 1;
994 
995 		/* Memory decoding is enabled and the BAR is 0. If any of the bridge
996 		 * resources covers that starting address (0 then it's good enough for
997 		 * us for memory
998 		 */
999 		for (i = 0; i < 3; i++) {
1000 			if ((hose->mem_resources[i].flags & IORESOURCE_MEM) &&
1001 			    hose->mem_resources[i].start == hose->pci_mem_offset)
1002 				return 0;
1003 		}
1004 
1005 		/* Well, it starts at 0 and we know it will collide so we may as
1006 		 * well consider it as unassigned. That covers the Apple case.
1007 		 */
1008 		return 1;
1009 	} else {
1010 		/* If the BAR is non-0, then we consider it assigned */
1011 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1012 		if (((res->start - offset) & 0xfffffffful) != 0)
1013 			return 0;
1014 
1015 		/* Here, we are a bit different than memory as typically IO space
1016 		 * starting at low addresses -is- valid. What we do instead if that
1017 		 * we consider as unassigned anything that doesn't have IO enabled
1018 		 * in the PCI command register, and that's it.
1019 		 */
1020 		pci_read_config_word(dev, PCI_COMMAND, &command);
1021 		if (command & PCI_COMMAND_IO)
1022 			return 0;
1023 
1024 		/* It's starting at 0 and IO is disabled in the bridge, consider
1025 		 * it unassigned
1026 		 */
1027 		return 1;
1028 	}
1029 }
1030 
1031 /* Fixup resources of a PCI<->PCI bridge */
1032 static void __devinit pcibios_fixup_bridge(struct pci_bus *bus)
1033 {
1034 	struct resource *res;
1035 	int i;
1036 
1037 	struct pci_dev *dev = bus->self;
1038 
1039 	pci_bus_for_each_resource(bus, res, i) {
1040 		if (!res || !res->flags)
1041 			continue;
1042 		if (i >= 3 && bus->self->transparent)
1043 			continue;
1044 
1045 		pr_debug("PCI:%s Bus rsrc %d %016llx-%016llx [%x] fixup...\n",
1046 			 pci_name(dev), i,
1047 			 (unsigned long long)res->start,\
1048 			 (unsigned long long)res->end,
1049 			 (unsigned int)res->flags);
1050 
1051 		/* Perform fixup */
1052 		fixup_resource(res, dev);
1053 
1054 		/* Try to detect uninitialized P2P bridge resources,
1055 		 * and clear them out so they get re-assigned later
1056 		 */
1057 		if (pcibios_uninitialized_bridge_resource(bus, res)) {
1058 			res->flags = 0;
1059 			pr_debug("PCI:%s            (unassigned)\n", pci_name(dev));
1060 		} else {
1061 
1062 			pr_debug("PCI:%s            %016llx-%016llx\n",
1063 				 pci_name(dev),
1064 				 (unsigned long long)res->start,
1065 				 (unsigned long long)res->end);
1066 		}
1067 	}
1068 }
1069 
1070 void __devinit pcibios_setup_bus_self(struct pci_bus *bus)
1071 {
1072 	/* Fix up the bus resources for P2P bridges */
1073 	if (bus->self != NULL)
1074 		pcibios_fixup_bridge(bus);
1075 
1076 	/* Platform specific bus fixups. This is currently only used
1077 	 * by fsl_pci and I'm hoping to get rid of it at some point
1078 	 */
1079 	if (ppc_md.pcibios_fixup_bus)
1080 		ppc_md.pcibios_fixup_bus(bus);
1081 
1082 	/* Setup bus DMA mappings */
1083 	if (ppc_md.pci_dma_bus_setup)
1084 		ppc_md.pci_dma_bus_setup(bus);
1085 }
1086 
1087 void __devinit pcibios_setup_bus_devices(struct pci_bus *bus)
1088 {
1089 	struct pci_dev *dev;
1090 
1091 	pr_debug("PCI: Fixup bus devices %d (%s)\n",
1092 		 bus->number, bus->self ? pci_name(bus->self) : "PHB");
1093 
1094 	list_for_each_entry(dev, &bus->devices, bus_list) {
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 		/* Fixup NUMA node as it may not be setup yet by the generic
1102 		 * code and is needed by the DMA init
1103 		 */
1104 		set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
1105 
1106 		/* Hook up default DMA ops */
1107 		set_dma_ops(&dev->dev, pci_dma_ops);
1108 		set_dma_offset(&dev->dev, PCI_DRAM_OFFSET);
1109 
1110 		/* Additional platform DMA/iommu setup */
1111 		if (ppc_md.pci_dma_dev_setup)
1112 			ppc_md.pci_dma_dev_setup(dev);
1113 
1114 		/* Read default IRQs and fixup if necessary */
1115 		pci_read_irq_line(dev);
1116 		if (ppc_md.pci_irq_fixup)
1117 			ppc_md.pci_irq_fixup(dev);
1118 	}
1119 }
1120 
1121 void __devinit pcibios_fixup_bus(struct pci_bus *bus)
1122 {
1123 	/* When called from the generic PCI probe, read PCI<->PCI bridge
1124 	 * bases. This is -not- called when generating the PCI tree from
1125 	 * the OF device-tree.
1126 	 */
1127 	if (bus->self != NULL)
1128 		pci_read_bridge_bases(bus);
1129 
1130 	/* Now fixup the bus bus */
1131 	pcibios_setup_bus_self(bus);
1132 
1133 	/* Now fixup devices on that bus */
1134 	pcibios_setup_bus_devices(bus);
1135 }
1136 EXPORT_SYMBOL(pcibios_fixup_bus);
1137 
1138 void __devinit pci_fixup_cardbus(struct pci_bus *bus)
1139 {
1140 	/* Now fixup devices on that bus */
1141 	pcibios_setup_bus_devices(bus);
1142 }
1143 
1144 
1145 static int skip_isa_ioresource_align(struct pci_dev *dev)
1146 {
1147 	if (pci_has_flag(PCI_CAN_SKIP_ISA_ALIGN) &&
1148 	    !(dev->bus->bridge_ctl & PCI_BRIDGE_CTL_ISA))
1149 		return 1;
1150 	return 0;
1151 }
1152 
1153 /*
1154  * We need to avoid collisions with `mirrored' VGA ports
1155  * and other strange ISA hardware, so we always want the
1156  * addresses to be allocated in the 0x000-0x0ff region
1157  * modulo 0x400.
1158  *
1159  * Why? Because some silly external IO cards only decode
1160  * the low 10 bits of the IO address. The 0x00-0xff region
1161  * is reserved for motherboard devices that decode all 16
1162  * bits, so it's ok to allocate at, say, 0x2800-0x28ff,
1163  * but we want to try to avoid allocating at 0x2900-0x2bff
1164  * which might have be mirrored at 0x0100-0x03ff..
1165  */
1166 resource_size_t pcibios_align_resource(void *data, const struct resource *res,
1167 				resource_size_t size, resource_size_t align)
1168 {
1169 	struct pci_dev *dev = data;
1170 	resource_size_t start = res->start;
1171 
1172 	if (res->flags & IORESOURCE_IO) {
1173 		if (skip_isa_ioresource_align(dev))
1174 			return start;
1175 		if (start & 0x300)
1176 			start = (start + 0x3ff) & ~0x3ff;
1177 	}
1178 
1179 	return start;
1180 }
1181 EXPORT_SYMBOL(pcibios_align_resource);
1182 
1183 /*
1184  * Reparent resource children of pr that conflict with res
1185  * under res, and make res replace those children.
1186  */
1187 static int reparent_resources(struct resource *parent,
1188 				     struct resource *res)
1189 {
1190 	struct resource *p, **pp;
1191 	struct resource **firstpp = NULL;
1192 
1193 	for (pp = &parent->child; (p = *pp) != NULL; pp = &p->sibling) {
1194 		if (p->end < res->start)
1195 			continue;
1196 		if (res->end < p->start)
1197 			break;
1198 		if (p->start < res->start || p->end > res->end)
1199 			return -1;	/* not completely contained */
1200 		if (firstpp == NULL)
1201 			firstpp = pp;
1202 	}
1203 	if (firstpp == NULL)
1204 		return -1;	/* didn't find any conflicting entries? */
1205 	res->parent = parent;
1206 	res->child = *firstpp;
1207 	res->sibling = *pp;
1208 	*firstpp = res;
1209 	*pp = NULL;
1210 	for (p = res->child; p != NULL; p = p->sibling) {
1211 		p->parent = res;
1212 		pr_debug("PCI: Reparented %s [%llx..%llx] under %s\n",
1213 			 p->name,
1214 			 (unsigned long long)p->start,
1215 			 (unsigned long long)p->end, res->name);
1216 	}
1217 	return 0;
1218 }
1219 
1220 /*
1221  *  Handle resources of PCI devices.  If the world were perfect, we could
1222  *  just allocate all the resource regions and do nothing more.  It isn't.
1223  *  On the other hand, we cannot just re-allocate all devices, as it would
1224  *  require us to know lots of host bridge internals.  So we attempt to
1225  *  keep as much of the original configuration as possible, but tweak it
1226  *  when it's found to be wrong.
1227  *
1228  *  Known BIOS problems we have to work around:
1229  *	- I/O or memory regions not configured
1230  *	- regions configured, but not enabled in the command register
1231  *	- bogus I/O addresses above 64K used
1232  *	- expansion ROMs left enabled (this may sound harmless, but given
1233  *	  the fact the PCI specs explicitly allow address decoders to be
1234  *	  shared between expansion ROMs and other resource regions, it's
1235  *	  at least dangerous)
1236  *
1237  *  Our solution:
1238  *	(1) Allocate resources for all buses behind PCI-to-PCI bridges.
1239  *	    This gives us fixed barriers on where we can allocate.
1240  *	(2) Allocate resources for all enabled devices.  If there is
1241  *	    a collision, just mark the resource as unallocated. Also
1242  *	    disable expansion ROMs during this step.
1243  *	(3) Try to allocate resources for disabled devices.  If the
1244  *	    resources were assigned correctly, everything goes well,
1245  *	    if they weren't, they won't disturb allocation of other
1246  *	    resources.
1247  *	(4) Assign new addresses to resources which were either
1248  *	    not configured at all or misconfigured.  If explicitly
1249  *	    requested by the user, configure expansion ROM address
1250  *	    as well.
1251  */
1252 
1253 void pcibios_allocate_bus_resources(struct pci_bus *bus)
1254 {
1255 	struct pci_bus *b;
1256 	int i;
1257 	struct resource *res, *pr;
1258 
1259 	pr_debug("PCI: Allocating bus resources for %04x:%02x...\n",
1260 		 pci_domain_nr(bus), bus->number);
1261 
1262 	pci_bus_for_each_resource(bus, res, i) {
1263 		if (!res || !res->flags || res->start > res->end || res->parent)
1264 			continue;
1265 		if (bus->parent == NULL)
1266 			pr = (res->flags & IORESOURCE_IO) ?
1267 				&ioport_resource : &iomem_resource;
1268 		else {
1269 			/* Don't bother with non-root busses when
1270 			 * re-assigning all resources. We clear the
1271 			 * resource flags as if they were colliding
1272 			 * and as such ensure proper re-allocation
1273 			 * later.
1274 			 */
1275 			if (pci_has_flag(PCI_REASSIGN_ALL_RSRC))
1276 				goto clear_resource;
1277 			pr = pci_find_parent_resource(bus->self, res);
1278 			if (pr == res) {
1279 				/* this happens when the generic PCI
1280 				 * code (wrongly) decides that this
1281 				 * bridge is transparent  -- paulus
1282 				 */
1283 				continue;
1284 			}
1285 		}
1286 
1287 		pr_debug("PCI: %s (bus %d) bridge rsrc %d: %016llx-%016llx "
1288 			 "[0x%x], parent %p (%s)\n",
1289 			 bus->self ? pci_name(bus->self) : "PHB",
1290 			 bus->number, i,
1291 			 (unsigned long long)res->start,
1292 			 (unsigned long long)res->end,
1293 			 (unsigned int)res->flags,
1294 			 pr, (pr && pr->name) ? pr->name : "nil");
1295 
1296 		if (pr && !(pr->flags & IORESOURCE_UNSET)) {
1297 			if (request_resource(pr, res) == 0)
1298 				continue;
1299 			/*
1300 			 * Must be a conflict with an existing entry.
1301 			 * Move that entry (or entries) under the
1302 			 * bridge resource and try again.
1303 			 */
1304 			if (reparent_resources(pr, res) == 0)
1305 				continue;
1306 		}
1307 		printk(KERN_WARNING "PCI: Cannot allocate resource region "
1308 		       "%d of PCI bridge %d, will remap\n", i, bus->number);
1309 clear_resource:
1310 		res->start = res->end = 0;
1311 		res->flags = 0;
1312 	}
1313 
1314 	list_for_each_entry(b, &bus->children, node)
1315 		pcibios_allocate_bus_resources(b);
1316 }
1317 
1318 static inline void __devinit alloc_resource(struct pci_dev *dev, int idx)
1319 {
1320 	struct resource *pr, *r = &dev->resource[idx];
1321 
1322 	pr_debug("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1323 		 pci_name(dev), idx,
1324 		 (unsigned long long)r->start,
1325 		 (unsigned long long)r->end,
1326 		 (unsigned int)r->flags);
1327 
1328 	pr = pci_find_parent_resource(dev, r);
1329 	if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1330 	    request_resource(pr, r) < 0) {
1331 		printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1332 		       " of device %s, will remap\n", idx, pci_name(dev));
1333 		if (pr)
1334 			pr_debug("PCI:  parent is %p: %016llx-%016llx [%x]\n",
1335 				 pr,
1336 				 (unsigned long long)pr->start,
1337 				 (unsigned long long)pr->end,
1338 				 (unsigned int)pr->flags);
1339 		/* We'll assign a new address later */
1340 		r->flags |= IORESOURCE_UNSET;
1341 		r->end -= r->start;
1342 		r->start = 0;
1343 	}
1344 }
1345 
1346 static void __init pcibios_allocate_resources(int pass)
1347 {
1348 	struct pci_dev *dev = NULL;
1349 	int idx, disabled;
1350 	u16 command;
1351 	struct resource *r;
1352 
1353 	for_each_pci_dev(dev) {
1354 		pci_read_config_word(dev, PCI_COMMAND, &command);
1355 		for (idx = 0; idx <= PCI_ROM_RESOURCE; idx++) {
1356 			r = &dev->resource[idx];
1357 			if (r->parent)		/* Already allocated */
1358 				continue;
1359 			if (!r->flags || (r->flags & IORESOURCE_UNSET))
1360 				continue;	/* Not assigned at all */
1361 			/* We only allocate ROMs on pass 1 just in case they
1362 			 * have been screwed up by firmware
1363 			 */
1364 			if (idx == PCI_ROM_RESOURCE )
1365 				disabled = 1;
1366 			if (r->flags & IORESOURCE_IO)
1367 				disabled = !(command & PCI_COMMAND_IO);
1368 			else
1369 				disabled = !(command & PCI_COMMAND_MEMORY);
1370 			if (pass == disabled)
1371 				alloc_resource(dev, idx);
1372 		}
1373 		if (pass)
1374 			continue;
1375 		r = &dev->resource[PCI_ROM_RESOURCE];
1376 		if (r->flags) {
1377 			/* Turn the ROM off, leave the resource region,
1378 			 * but keep it unregistered.
1379 			 */
1380 			u32 reg;
1381 			pci_read_config_dword(dev, dev->rom_base_reg, &reg);
1382 			if (reg & PCI_ROM_ADDRESS_ENABLE) {
1383 				pr_debug("PCI: Switching off ROM of %s\n",
1384 					 pci_name(dev));
1385 				r->flags &= ~IORESOURCE_ROM_ENABLE;
1386 				pci_write_config_dword(dev, dev->rom_base_reg,
1387 						       reg & ~PCI_ROM_ADDRESS_ENABLE);
1388 			}
1389 		}
1390 	}
1391 }
1392 
1393 static void __init pcibios_reserve_legacy_regions(struct pci_bus *bus)
1394 {
1395 	struct pci_controller *hose = pci_bus_to_host(bus);
1396 	resource_size_t	offset;
1397 	struct resource *res, *pres;
1398 	int i;
1399 
1400 	pr_debug("Reserving legacy ranges for domain %04x\n", pci_domain_nr(bus));
1401 
1402 	/* Check for IO */
1403 	if (!(hose->io_resource.flags & IORESOURCE_IO))
1404 		goto no_io;
1405 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1406 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1407 	BUG_ON(res == NULL);
1408 	res->name = "Legacy IO";
1409 	res->flags = IORESOURCE_IO;
1410 	res->start = offset;
1411 	res->end = (offset + 0xfff) & 0xfffffffful;
1412 	pr_debug("Candidate legacy IO: %pR\n", res);
1413 	if (request_resource(&hose->io_resource, res)) {
1414 		printk(KERN_DEBUG
1415 		       "PCI %04x:%02x Cannot reserve Legacy IO %pR\n",
1416 		       pci_domain_nr(bus), bus->number, res);
1417 		kfree(res);
1418 	}
1419 
1420  no_io:
1421 	/* Check for memory */
1422 	offset = hose->pci_mem_offset;
1423 	pr_debug("hose mem offset: %016llx\n", (unsigned long long)offset);
1424 	for (i = 0; i < 3; i++) {
1425 		pres = &hose->mem_resources[i];
1426 		if (!(pres->flags & IORESOURCE_MEM))
1427 			continue;
1428 		pr_debug("hose mem res: %pR\n", pres);
1429 		if ((pres->start - offset) <= 0xa0000 &&
1430 		    (pres->end - offset) >= 0xbffff)
1431 			break;
1432 	}
1433 	if (i >= 3)
1434 		return;
1435 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1436 	BUG_ON(res == NULL);
1437 	res->name = "Legacy VGA memory";
1438 	res->flags = IORESOURCE_MEM;
1439 	res->start = 0xa0000 + offset;
1440 	res->end = 0xbffff + offset;
1441 	pr_debug("Candidate VGA memory: %pR\n", res);
1442 	if (request_resource(pres, res)) {
1443 		printk(KERN_DEBUG
1444 		       "PCI %04x:%02x Cannot reserve VGA memory %pR\n",
1445 		       pci_domain_nr(bus), bus->number, res);
1446 		kfree(res);
1447 	}
1448 }
1449 
1450 void __init pcibios_resource_survey(void)
1451 {
1452 	struct pci_bus *b;
1453 
1454 	/* Allocate and assign resources. If we re-assign everything, then
1455 	 * we skip the allocate phase
1456 	 */
1457 	list_for_each_entry(b, &pci_root_buses, node)
1458 		pcibios_allocate_bus_resources(b);
1459 
1460 	if (!pci_has_flag(PCI_REASSIGN_ALL_RSRC)) {
1461 		pcibios_allocate_resources(0);
1462 		pcibios_allocate_resources(1);
1463 	}
1464 
1465 	/* Before we start assigning unassigned resource, we try to reserve
1466 	 * the low IO area and the VGA memory area if they intersect the
1467 	 * bus available resources to avoid allocating things on top of them
1468 	 */
1469 	if (!pci_has_flag(PCI_PROBE_ONLY)) {
1470 		list_for_each_entry(b, &pci_root_buses, node)
1471 			pcibios_reserve_legacy_regions(b);
1472 	}
1473 
1474 	/* Now, if the platform didn't decide to blindly trust the firmware,
1475 	 * we proceed to assigning things that were left unassigned
1476 	 */
1477 	if (!pci_has_flag(PCI_PROBE_ONLY)) {
1478 		pr_debug("PCI: Assigning unassigned resources...\n");
1479 		pci_assign_unassigned_resources();
1480 	}
1481 
1482 	/* Call machine dependent fixup */
1483 	if (ppc_md.pcibios_fixup)
1484 		ppc_md.pcibios_fixup();
1485 }
1486 
1487 #ifdef CONFIG_HOTPLUG
1488 
1489 /* This is used by the PCI hotplug driver to allocate resource
1490  * of newly plugged busses. We can try to consolidate with the
1491  * rest of the code later, for now, keep it as-is as our main
1492  * resource allocation function doesn't deal with sub-trees yet.
1493  */
1494 void pcibios_claim_one_bus(struct pci_bus *bus)
1495 {
1496 	struct pci_dev *dev;
1497 	struct pci_bus *child_bus;
1498 
1499 	list_for_each_entry(dev, &bus->devices, bus_list) {
1500 		int i;
1501 
1502 		for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1503 			struct resource *r = &dev->resource[i];
1504 
1505 			if (r->parent || !r->start || !r->flags)
1506 				continue;
1507 
1508 			pr_debug("PCI: Claiming %s: "
1509 				 "Resource %d: %016llx..%016llx [%x]\n",
1510 				 pci_name(dev), i,
1511 				 (unsigned long long)r->start,
1512 				 (unsigned long long)r->end,
1513 				 (unsigned int)r->flags);
1514 
1515 			pci_claim_resource(dev, i);
1516 		}
1517 	}
1518 
1519 	list_for_each_entry(child_bus, &bus->children, node)
1520 		pcibios_claim_one_bus(child_bus);
1521 }
1522 
1523 
1524 /* pcibios_finish_adding_to_bus
1525  *
1526  * This is to be called by the hotplug code after devices have been
1527  * added to a bus, this include calling it for a PHB that is just
1528  * being added
1529  */
1530 void pcibios_finish_adding_to_bus(struct pci_bus *bus)
1531 {
1532 	pr_debug("PCI: Finishing adding to hotplug bus %04x:%02x\n",
1533 		 pci_domain_nr(bus), bus->number);
1534 
1535 	/* Allocate bus and devices resources */
1536 	pcibios_allocate_bus_resources(bus);
1537 	pcibios_claim_one_bus(bus);
1538 
1539 	/* Add new devices to global lists.  Register in proc, sysfs. */
1540 	pci_bus_add_devices(bus);
1541 
1542 	/* Fixup EEH */
1543 	eeh_add_device_tree_late(bus);
1544 }
1545 EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
1546 
1547 #endif /* CONFIG_HOTPLUG */
1548 
1549 int pcibios_enable_device(struct pci_dev *dev, int mask)
1550 {
1551 	if (ppc_md.pcibios_enable_device_hook)
1552 		if (ppc_md.pcibios_enable_device_hook(dev))
1553 			return -EINVAL;
1554 
1555 	return pci_enable_resources(dev, mask);
1556 }
1557 
1558 void __devinit pcibios_setup_phb_resources(struct pci_controller *hose)
1559 {
1560 	struct pci_bus *bus = hose->bus;
1561 	struct resource *res;
1562 	int i;
1563 
1564 	/* Hookup PHB IO resource */
1565 	bus->resource[0] = res = &hose->io_resource;
1566 
1567 	if (!res->flags) {
1568 		printk(KERN_WARNING "PCI: I/O resource not set for host"
1569 		       " bridge %s (domain %d)\n",
1570 		       hose->dn->full_name, hose->global_number);
1571 #ifdef CONFIG_PPC32
1572 		/* Workaround for lack of IO resource only on 32-bit */
1573 		res->start = (unsigned long)hose->io_base_virt - isa_io_base;
1574 		res->end = res->start + IO_SPACE_LIMIT;
1575 		res->flags = IORESOURCE_IO;
1576 #endif /* CONFIG_PPC32 */
1577 	}
1578 
1579 	pr_debug("PCI: PHB IO resource    = %016llx-%016llx [%lx]\n",
1580 		 (unsigned long long)res->start,
1581 		 (unsigned long long)res->end,
1582 		 (unsigned long)res->flags);
1583 
1584 	/* Hookup PHB Memory resources */
1585 	for (i = 0; i < 3; ++i) {
1586 		res = &hose->mem_resources[i];
1587 		if (!res->flags) {
1588 			if (i > 0)
1589 				continue;
1590 			printk(KERN_ERR "PCI: Memory resource 0 not set for "
1591 			       "host bridge %s (domain %d)\n",
1592 			       hose->dn->full_name, hose->global_number);
1593 #ifdef CONFIG_PPC32
1594 			/* Workaround for lack of MEM resource only on 32-bit */
1595 			res->start = hose->pci_mem_offset;
1596 			res->end = (resource_size_t)-1LL;
1597 			res->flags = IORESOURCE_MEM;
1598 #endif /* CONFIG_PPC32 */
1599 		}
1600 		bus->resource[i+1] = res;
1601 
1602 		pr_debug("PCI: PHB MEM resource %d = %016llx-%016llx [%lx]\n", i,
1603 			 (unsigned long long)res->start,
1604 			 (unsigned long long)res->end,
1605 			 (unsigned long)res->flags);
1606 	}
1607 
1608 	pr_debug("PCI: PHB MEM offset     = %016llx\n",
1609 		 (unsigned long long)hose->pci_mem_offset);
1610 	pr_debug("PCI: PHB IO  offset     = %08lx\n",
1611 		 (unsigned long)hose->io_base_virt - _IO_BASE);
1612 
1613 }
1614 
1615 /*
1616  * Null PCI config access functions, for the case when we can't
1617  * find a hose.
1618  */
1619 #define NULL_PCI_OP(rw, size, type)					\
1620 static int								\
1621 null_##rw##_config_##size(struct pci_dev *dev, int offset, type val)	\
1622 {									\
1623 	return PCIBIOS_DEVICE_NOT_FOUND;    				\
1624 }
1625 
1626 static int
1627 null_read_config(struct pci_bus *bus, unsigned int devfn, int offset,
1628 		 int len, u32 *val)
1629 {
1630 	return PCIBIOS_DEVICE_NOT_FOUND;
1631 }
1632 
1633 static int
1634 null_write_config(struct pci_bus *bus, unsigned int devfn, int offset,
1635 		  int len, u32 val)
1636 {
1637 	return PCIBIOS_DEVICE_NOT_FOUND;
1638 }
1639 
1640 static struct pci_ops null_pci_ops =
1641 {
1642 	.read = null_read_config,
1643 	.write = null_write_config,
1644 };
1645 
1646 /*
1647  * These functions are used early on before PCI scanning is done
1648  * and all of the pci_dev and pci_bus structures have been created.
1649  */
1650 static struct pci_bus *
1651 fake_pci_bus(struct pci_controller *hose, int busnr)
1652 {
1653 	static struct pci_bus bus;
1654 
1655 	if (hose == 0) {
1656 		printk(KERN_ERR "Can't find hose for PCI bus %d!\n", busnr);
1657 	}
1658 	bus.number = busnr;
1659 	bus.sysdata = hose;
1660 	bus.ops = hose? hose->ops: &null_pci_ops;
1661 	return &bus;
1662 }
1663 
1664 #define EARLY_PCI_OP(rw, size, type)					\
1665 int early_##rw##_config_##size(struct pci_controller *hose, int bus,	\
1666 			       int devfn, int offset, type value)	\
1667 {									\
1668 	return pci_bus_##rw##_config_##size(fake_pci_bus(hose, bus),	\
1669 					    devfn, offset, value);	\
1670 }
1671 
1672 EARLY_PCI_OP(read, byte, u8 *)
1673 EARLY_PCI_OP(read, word, u16 *)
1674 EARLY_PCI_OP(read, dword, u32 *)
1675 EARLY_PCI_OP(write, byte, u8)
1676 EARLY_PCI_OP(write, word, u16)
1677 EARLY_PCI_OP(write, dword, u32)
1678 
1679 extern int pci_bus_find_capability (struct pci_bus *bus, unsigned int devfn, int cap);
1680 int early_find_capability(struct pci_controller *hose, int bus, int devfn,
1681 			  int cap)
1682 {
1683 	return pci_bus_find_capability(fake_pci_bus(hose, bus), devfn, cap);
1684 }
1685 
1686 struct device_node *pcibios_get_phb_of_node(struct pci_bus *bus)
1687 {
1688 	struct pci_controller *hose = bus->sysdata;
1689 
1690 	return of_node_get(hose->dn);
1691 }
1692 
1693 /**
1694  * pci_scan_phb - Given a pci_controller, setup and scan the PCI bus
1695  * @hose: Pointer to the PCI host controller instance structure
1696  */
1697 void __devinit pcibios_scan_phb(struct pci_controller *hose)
1698 {
1699 	struct pci_bus *bus;
1700 	struct device_node *node = hose->dn;
1701 	int mode;
1702 
1703 	pr_debug("PCI: Scanning PHB %s\n",
1704 		 node ? node->full_name : "<NO NAME>");
1705 
1706 	/* Create an empty bus for the toplevel */
1707 	bus = pci_create_bus(hose->parent, hose->first_busno, hose->ops, hose);
1708 	if (bus == NULL) {
1709 		pr_err("Failed to create bus for PCI domain %04x\n",
1710 			hose->global_number);
1711 		return;
1712 	}
1713 	bus->secondary = hose->first_busno;
1714 	hose->bus = bus;
1715 
1716 	/* Get some IO space for the new PHB */
1717 	pcibios_setup_phb_io_space(hose);
1718 
1719 	/* Wire up PHB bus resources */
1720 	pcibios_setup_phb_resources(hose);
1721 
1722 	/* Get probe mode and perform scan */
1723 	mode = PCI_PROBE_NORMAL;
1724 	if (node && ppc_md.pci_probe_mode)
1725 		mode = ppc_md.pci_probe_mode(bus);
1726 	pr_debug("    probe mode: %d\n", mode);
1727 	if (mode == PCI_PROBE_DEVTREE) {
1728 		bus->subordinate = hose->last_busno;
1729 		of_scan_bus(node, bus);
1730 	}
1731 
1732 	if (mode == PCI_PROBE_NORMAL)
1733 		hose->last_busno = bus->subordinate = pci_scan_child_bus(bus);
1734 
1735 	/* Configure PCI Express settings */
1736 	if (bus && !pci_has_flag(PCI_PROBE_ONLY)) {
1737 		struct pci_bus *child;
1738 		list_for_each_entry(child, &bus->children, node) {
1739 			struct pci_dev *self = child->self;
1740 			if (!self)
1741 				continue;
1742 			pcie_bus_configure_settings(child, self->pcie_mpss);
1743 		}
1744 	}
1745 }
1746 
1747 static void fixup_hide_host_resource_fsl(struct pci_dev *dev)
1748 {
1749 	int i, class = dev->class >> 8;
1750 
1751 	if ((class == PCI_CLASS_PROCESSOR_POWERPC ||
1752 	     class == PCI_CLASS_BRIDGE_OTHER) &&
1753 		(dev->hdr_type == PCI_HEADER_TYPE_NORMAL) &&
1754 		(dev->bus->parent == NULL)) {
1755 		for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
1756 			dev->resource[i].start = 0;
1757 			dev->resource[i].end = 0;
1758 			dev->resource[i].flags = 0;
1759 		}
1760 	}
1761 }
1762 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_MOTOROLA, PCI_ANY_ID, fixup_hide_host_resource_fsl);
1763 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_FREESCALE, PCI_ANY_ID, fixup_hide_host_resource_fsl);
1764