xref: /openbmc/linux/arch/powerpc/kernel/pci-common.c (revision 38973ba7903fa0660a31b2bdc50ff711ec8d08c9)
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 
54 static struct dma_map_ops *pci_dma_ops = &dma_direct_ops;
55 
56 void set_pci_dma_ops(struct dma_map_ops *dma_ops)
57 {
58 	pci_dma_ops = dma_ops;
59 }
60 
61 struct dma_map_ops *get_pci_dma_ops(void)
62 {
63 	return pci_dma_ops;
64 }
65 EXPORT_SYMBOL(get_pci_dma_ops);
66 
67 struct pci_controller *pcibios_alloc_controller(struct device_node *dev)
68 {
69 	struct pci_controller *phb;
70 
71 	phb = zalloc_maybe_bootmem(sizeof(struct pci_controller), GFP_KERNEL);
72 	if (phb == NULL)
73 		return NULL;
74 	spin_lock(&hose_spinlock);
75 	phb->global_number = global_phb_number++;
76 	list_add_tail(&phb->list_node, &hose_list);
77 	spin_unlock(&hose_spinlock);
78 	phb->dn = dev;
79 	phb->is_dynamic = mem_init_done;
80 #ifdef CONFIG_PPC64
81 	if (dev) {
82 		int nid = of_node_to_nid(dev);
83 
84 		if (nid < 0 || !node_online(nid))
85 			nid = -1;
86 
87 		PHB_SET_NODE(phb, nid);
88 	}
89 #endif
90 	return phb;
91 }
92 
93 void pcibios_free_controller(struct pci_controller *phb)
94 {
95 	spin_lock(&hose_spinlock);
96 	list_del(&phb->list_node);
97 	spin_unlock(&hose_spinlock);
98 
99 	if (phb->is_dynamic)
100 		kfree(phb);
101 }
102 
103 static resource_size_t pcibios_io_size(const struct pci_controller *hose)
104 {
105 #ifdef CONFIG_PPC64
106 	return hose->pci_io_size;
107 #else
108 	return resource_size(&hose->io_resource);
109 #endif
110 }
111 
112 int pcibios_vaddr_is_ioport(void __iomem *address)
113 {
114 	int ret = 0;
115 	struct pci_controller *hose;
116 	resource_size_t size;
117 
118 	spin_lock(&hose_spinlock);
119 	list_for_each_entry(hose, &hose_list, list_node) {
120 		size = pcibios_io_size(hose);
121 		if (address >= hose->io_base_virt &&
122 		    address < (hose->io_base_virt + size)) {
123 			ret = 1;
124 			break;
125 		}
126 	}
127 	spin_unlock(&hose_spinlock);
128 	return ret;
129 }
130 
131 unsigned long pci_address_to_pio(phys_addr_t address)
132 {
133 	struct pci_controller *hose;
134 	resource_size_t size;
135 	unsigned long ret = ~0;
136 
137 	spin_lock(&hose_spinlock);
138 	list_for_each_entry(hose, &hose_list, list_node) {
139 		size = pcibios_io_size(hose);
140 		if (address >= hose->io_base_phys &&
141 		    address < (hose->io_base_phys + size)) {
142 			unsigned long base =
143 				(unsigned long)hose->io_base_virt - _IO_BASE;
144 			ret = base + (address - hose->io_base_phys);
145 			break;
146 		}
147 	}
148 	spin_unlock(&hose_spinlock);
149 
150 	return ret;
151 }
152 EXPORT_SYMBOL_GPL(pci_address_to_pio);
153 
154 /*
155  * Return the domain number for this bus.
156  */
157 int pci_domain_nr(struct pci_bus *bus)
158 {
159 	struct pci_controller *hose = pci_bus_to_host(bus);
160 
161 	return hose->global_number;
162 }
163 EXPORT_SYMBOL(pci_domain_nr);
164 
165 /* This routine is meant to be used early during boot, when the
166  * PCI bus numbers have not yet been assigned, and you need to
167  * issue PCI config cycles to an OF device.
168  * It could also be used to "fix" RTAS config cycles if you want
169  * to set pci_assign_all_buses to 1 and still use RTAS for PCI
170  * config cycles.
171  */
172 struct pci_controller* pci_find_hose_for_OF_device(struct device_node* node)
173 {
174 	while(node) {
175 		struct pci_controller *hose, *tmp;
176 		list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
177 			if (hose->dn == node)
178 				return hose;
179 		node = node->parent;
180 	}
181 	return NULL;
182 }
183 
184 static ssize_t pci_show_devspec(struct device *dev,
185 		struct device_attribute *attr, char *buf)
186 {
187 	struct pci_dev *pdev;
188 	struct device_node *np;
189 
190 	pdev = to_pci_dev (dev);
191 	np = pci_device_to_OF_node(pdev);
192 	if (np == NULL || np->full_name == NULL)
193 		return 0;
194 	return sprintf(buf, "%s", np->full_name);
195 }
196 static DEVICE_ATTR(devspec, S_IRUGO, pci_show_devspec, NULL);
197 
198 /* Add sysfs properties */
199 int pcibios_add_platform_entries(struct pci_dev *pdev)
200 {
201 	return device_create_file(&pdev->dev, &dev_attr_devspec);
202 }
203 
204 char __devinit *pcibios_setup(char *str)
205 {
206 	return str;
207 }
208 
209 /*
210  * Reads the interrupt pin to determine if interrupt is use by card.
211  * If the interrupt is used, then gets the interrupt line from the
212  * openfirmware and sets it in the pci_dev and pci_config line.
213  */
214 static int pci_read_irq_line(struct pci_dev *pci_dev)
215 {
216 	struct of_irq oirq;
217 	unsigned int virq;
218 
219 	/* The current device-tree that iSeries generates from the HV
220 	 * PCI informations doesn't contain proper interrupt routing,
221 	 * and all the fallback would do is print out crap, so we
222 	 * don't attempt to resolve the interrupts here at all, some
223 	 * iSeries specific fixup does it.
224 	 *
225 	 * In the long run, we will hopefully fix the generated device-tree
226 	 * instead.
227 	 */
228 #ifdef CONFIG_PPC_ISERIES
229 	if (firmware_has_feature(FW_FEATURE_ISERIES))
230 		return -1;
231 #endif
232 
233 	pr_debug("PCI: Try to map irq for %s...\n", pci_name(pci_dev));
234 
235 #ifdef DEBUG
236 	memset(&oirq, 0xff, sizeof(oirq));
237 #endif
238 	/* Try to get a mapping from the device-tree */
239 	if (of_irq_map_pci(pci_dev, &oirq)) {
240 		u8 line, pin;
241 
242 		/* If that fails, lets fallback to what is in the config
243 		 * space and map that through the default controller. We
244 		 * also set the type to level low since that's what PCI
245 		 * interrupts are. If your platform does differently, then
246 		 * either provide a proper interrupt tree or don't use this
247 		 * function.
248 		 */
249 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_PIN, &pin))
250 			return -1;
251 		if (pin == 0)
252 			return -1;
253 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_LINE, &line) ||
254 		    line == 0xff || line == 0) {
255 			return -1;
256 		}
257 		pr_debug(" No map ! Using line %d (pin %d) from PCI config\n",
258 			 line, pin);
259 
260 		virq = irq_create_mapping(NULL, line);
261 		if (virq != NO_IRQ)
262 			irq_set_irq_type(virq, IRQ_TYPE_LEVEL_LOW);
263 	} else {
264 		pr_debug(" Got one, spec %d cells (0x%08x 0x%08x...) on %s\n",
265 			 oirq.size, oirq.specifier[0], oirq.specifier[1],
266 			 oirq.controller ? oirq.controller->full_name :
267 			 "<default>");
268 
269 		virq = irq_create_of_mapping(oirq.controller, oirq.specifier,
270 					     oirq.size);
271 	}
272 	if(virq == NO_IRQ) {
273 		pr_debug(" Failed to map !\n");
274 		return -1;
275 	}
276 
277 	pr_debug(" Mapped to linux irq %d\n", virq);
278 
279 	pci_dev->irq = virq;
280 
281 	return 0;
282 }
283 
284 /*
285  * Platform support for /proc/bus/pci/X/Y mmap()s,
286  * modelled on the sparc64 implementation by Dave Miller.
287  *  -- paulus.
288  */
289 
290 /*
291  * Adjust vm_pgoff of VMA such that it is the physical page offset
292  * corresponding to the 32-bit pci bus offset for DEV requested by the user.
293  *
294  * Basically, the user finds the base address for his device which he wishes
295  * to mmap.  They read the 32-bit value from the config space base register,
296  * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
297  * offset parameter of mmap on /proc/bus/pci/XXX for that device.
298  *
299  * Returns negative error code on failure, zero on success.
300  */
301 static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
302 					       resource_size_t *offset,
303 					       enum pci_mmap_state mmap_state)
304 {
305 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
306 	unsigned long io_offset = 0;
307 	int i, res_bit;
308 
309 	if (hose == 0)
310 		return NULL;		/* should never happen */
311 
312 	/* If memory, add on the PCI bridge address offset */
313 	if (mmap_state == pci_mmap_mem) {
314 #if 0 /* See comment in pci_resource_to_user() for why this is disabled */
315 		*offset += hose->pci_mem_offset;
316 #endif
317 		res_bit = IORESOURCE_MEM;
318 	} else {
319 		io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
320 		*offset += io_offset;
321 		res_bit = IORESOURCE_IO;
322 	}
323 
324 	/*
325 	 * Check that the offset requested corresponds to one of the
326 	 * resources of the device.
327 	 */
328 	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
329 		struct resource *rp = &dev->resource[i];
330 		int flags = rp->flags;
331 
332 		/* treat ROM as memory (should be already) */
333 		if (i == PCI_ROM_RESOURCE)
334 			flags |= IORESOURCE_MEM;
335 
336 		/* Active and same type? */
337 		if ((flags & res_bit) == 0)
338 			continue;
339 
340 		/* In the range of this resource? */
341 		if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
342 			continue;
343 
344 		/* found it! construct the final physical address */
345 		if (mmap_state == pci_mmap_io)
346 			*offset += hose->io_base_phys - io_offset;
347 		return rp;
348 	}
349 
350 	return NULL;
351 }
352 
353 /*
354  * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
355  * device mapping.
356  */
357 static pgprot_t __pci_mmap_set_pgprot(struct pci_dev *dev, struct resource *rp,
358 				      pgprot_t protection,
359 				      enum pci_mmap_state mmap_state,
360 				      int write_combine)
361 {
362 	unsigned long prot = pgprot_val(protection);
363 
364 	/* Write combine is always 0 on non-memory space mappings. On
365 	 * memory space, if the user didn't pass 1, we check for a
366 	 * "prefetchable" resource. This is a bit hackish, but we use
367 	 * this to workaround the inability of /sysfs to provide a write
368 	 * combine bit
369 	 */
370 	if (mmap_state != pci_mmap_mem)
371 		write_combine = 0;
372 	else if (write_combine == 0) {
373 		if (rp->flags & IORESOURCE_PREFETCH)
374 			write_combine = 1;
375 	}
376 
377 	/* XXX would be nice to have a way to ask for write-through */
378 	if (write_combine)
379 		return pgprot_noncached_wc(prot);
380 	else
381 		return pgprot_noncached(prot);
382 }
383 
384 /*
385  * This one is used by /dev/mem and fbdev who have no clue about the
386  * PCI device, it tries to find the PCI device first and calls the
387  * above routine
388  */
389 pgprot_t pci_phys_mem_access_prot(struct file *file,
390 				  unsigned long pfn,
391 				  unsigned long size,
392 				  pgprot_t prot)
393 {
394 	struct pci_dev *pdev = NULL;
395 	struct resource *found = NULL;
396 	resource_size_t offset = ((resource_size_t)pfn) << PAGE_SHIFT;
397 	int i;
398 
399 	if (page_is_ram(pfn))
400 		return prot;
401 
402 	prot = pgprot_noncached(prot);
403 	for_each_pci_dev(pdev) {
404 		for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
405 			struct resource *rp = &pdev->resource[i];
406 			int flags = rp->flags;
407 
408 			/* Active and same type? */
409 			if ((flags & IORESOURCE_MEM) == 0)
410 				continue;
411 			/* In the range of this resource? */
412 			if (offset < (rp->start & PAGE_MASK) ||
413 			    offset > rp->end)
414 				continue;
415 			found = rp;
416 			break;
417 		}
418 		if (found)
419 			break;
420 	}
421 	if (found) {
422 		if (found->flags & IORESOURCE_PREFETCH)
423 			prot = pgprot_noncached_wc(prot);
424 		pci_dev_put(pdev);
425 	}
426 
427 	pr_debug("PCI: Non-PCI map for %llx, prot: %lx\n",
428 		 (unsigned long long)offset, pgprot_val(prot));
429 
430 	return prot;
431 }
432 
433 
434 /*
435  * Perform the actual remap of the pages for a PCI device mapping, as
436  * appropriate for this architecture.  The region in the process to map
437  * is described by vm_start and vm_end members of VMA, the base physical
438  * address is found in vm_pgoff.
439  * The pci device structure is provided so that architectures may make mapping
440  * decisions on a per-device or per-bus basis.
441  *
442  * Returns a negative error code on failure, zero on success.
443  */
444 int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
445 			enum pci_mmap_state mmap_state, int write_combine)
446 {
447 	resource_size_t offset =
448 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
449 	struct resource *rp;
450 	int ret;
451 
452 	rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
453 	if (rp == NULL)
454 		return -EINVAL;
455 
456 	vma->vm_pgoff = offset >> PAGE_SHIFT;
457 	vma->vm_page_prot = __pci_mmap_set_pgprot(dev, rp,
458 						  vma->vm_page_prot,
459 						  mmap_state, write_combine);
460 
461 	ret = remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
462 			       vma->vm_end - vma->vm_start, vma->vm_page_prot);
463 
464 	return ret;
465 }
466 
467 /* This provides legacy IO read access on a bus */
468 int pci_legacy_read(struct pci_bus *bus, loff_t port, u32 *val, size_t size)
469 {
470 	unsigned long offset;
471 	struct pci_controller *hose = pci_bus_to_host(bus);
472 	struct resource *rp = &hose->io_resource;
473 	void __iomem *addr;
474 
475 	/* Check if port can be supported by that bus. We only check
476 	 * the ranges of the PHB though, not the bus itself as the rules
477 	 * for forwarding legacy cycles down bridges are not our problem
478 	 * here. So if the host bridge supports it, we do it.
479 	 */
480 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
481 	offset += port;
482 
483 	if (!(rp->flags & IORESOURCE_IO))
484 		return -ENXIO;
485 	if (offset < rp->start || (offset + size) > rp->end)
486 		return -ENXIO;
487 	addr = hose->io_base_virt + port;
488 
489 	switch(size) {
490 	case 1:
491 		*((u8 *)val) = in_8(addr);
492 		return 1;
493 	case 2:
494 		if (port & 1)
495 			return -EINVAL;
496 		*((u16 *)val) = in_le16(addr);
497 		return 2;
498 	case 4:
499 		if (port & 3)
500 			return -EINVAL;
501 		*((u32 *)val) = in_le32(addr);
502 		return 4;
503 	}
504 	return -EINVAL;
505 }
506 
507 /* This provides legacy IO write access on a bus */
508 int pci_legacy_write(struct pci_bus *bus, loff_t port, u32 val, size_t size)
509 {
510 	unsigned long offset;
511 	struct pci_controller *hose = pci_bus_to_host(bus);
512 	struct resource *rp = &hose->io_resource;
513 	void __iomem *addr;
514 
515 	/* Check if port can be supported by that bus. We only check
516 	 * the ranges of the PHB though, not the bus itself as the rules
517 	 * for forwarding legacy cycles down bridges are not our problem
518 	 * here. So if the host bridge supports it, we do it.
519 	 */
520 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
521 	offset += port;
522 
523 	if (!(rp->flags & IORESOURCE_IO))
524 		return -ENXIO;
525 	if (offset < rp->start || (offset + size) > rp->end)
526 		return -ENXIO;
527 	addr = hose->io_base_virt + port;
528 
529 	/* WARNING: The generic code is idiotic. It gets passed a pointer
530 	 * to what can be a 1, 2 or 4 byte quantity and always reads that
531 	 * as a u32, which means that we have to correct the location of
532 	 * the data read within those 32 bits for size 1 and 2
533 	 */
534 	switch(size) {
535 	case 1:
536 		out_8(addr, val >> 24);
537 		return 1;
538 	case 2:
539 		if (port & 1)
540 			return -EINVAL;
541 		out_le16(addr, val >> 16);
542 		return 2;
543 	case 4:
544 		if (port & 3)
545 			return -EINVAL;
546 		out_le32(addr, val);
547 		return 4;
548 	}
549 	return -EINVAL;
550 }
551 
552 /* This provides legacy IO or memory mmap access on a bus */
553 int pci_mmap_legacy_page_range(struct pci_bus *bus,
554 			       struct vm_area_struct *vma,
555 			       enum pci_mmap_state mmap_state)
556 {
557 	struct pci_controller *hose = pci_bus_to_host(bus);
558 	resource_size_t offset =
559 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
560 	resource_size_t size = vma->vm_end - vma->vm_start;
561 	struct resource *rp;
562 
563 	pr_debug("pci_mmap_legacy_page_range(%04x:%02x, %s @%llx..%llx)\n",
564 		 pci_domain_nr(bus), bus->number,
565 		 mmap_state == pci_mmap_mem ? "MEM" : "IO",
566 		 (unsigned long long)offset,
567 		 (unsigned long long)(offset + size - 1));
568 
569 	if (mmap_state == pci_mmap_mem) {
570 		/* Hack alert !
571 		 *
572 		 * Because X is lame and can fail starting if it gets an error trying
573 		 * to mmap legacy_mem (instead of just moving on without legacy memory
574 		 * access) we fake it here by giving it anonymous memory, effectively
575 		 * behaving just like /dev/zero
576 		 */
577 		if ((offset + size) > hose->isa_mem_size) {
578 			printk(KERN_DEBUG
579 			       "Process %s (pid:%d) mapped non-existing PCI legacy memory for 0%04x:%02x\n",
580 			       current->comm, current->pid, pci_domain_nr(bus), bus->number);
581 			if (vma->vm_flags & VM_SHARED)
582 				return shmem_zero_setup(vma);
583 			return 0;
584 		}
585 		offset += hose->isa_mem_phys;
586 	} else {
587 		unsigned long io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
588 		unsigned long roffset = offset + io_offset;
589 		rp = &hose->io_resource;
590 		if (!(rp->flags & IORESOURCE_IO))
591 			return -ENXIO;
592 		if (roffset < rp->start || (roffset + size) > rp->end)
593 			return -ENXIO;
594 		offset += hose->io_base_phys;
595 	}
596 	pr_debug(" -> mapping phys %llx\n", (unsigned long long)offset);
597 
598 	vma->vm_pgoff = offset >> PAGE_SHIFT;
599 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
600 	return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
601 			       vma->vm_end - vma->vm_start,
602 			       vma->vm_page_prot);
603 }
604 
605 void pci_resource_to_user(const struct pci_dev *dev, int bar,
606 			  const struct resource *rsrc,
607 			  resource_size_t *start, resource_size_t *end)
608 {
609 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
610 	resource_size_t offset = 0;
611 
612 	if (hose == NULL)
613 		return;
614 
615 	if (rsrc->flags & IORESOURCE_IO)
616 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
617 
618 	/* We pass a fully fixed up address to userland for MMIO instead of
619 	 * a BAR value because X is lame and expects to be able to use that
620 	 * to pass to /dev/mem !
621 	 *
622 	 * That means that we'll have potentially 64 bits values where some
623 	 * userland apps only expect 32 (like X itself since it thinks only
624 	 * Sparc has 64 bits MMIO) but if we don't do that, we break it on
625 	 * 32 bits CHRPs :-(
626 	 *
627 	 * Hopefully, the sysfs insterface is immune to that gunk. Once X
628 	 * has been fixed (and the fix spread enough), we can re-enable the
629 	 * 2 lines below and pass down a BAR value to userland. In that case
630 	 * we'll also have to re-enable the matching code in
631 	 * __pci_mmap_make_offset().
632 	 *
633 	 * BenH.
634 	 */
635 #if 0
636 	else if (rsrc->flags & IORESOURCE_MEM)
637 		offset = hose->pci_mem_offset;
638 #endif
639 
640 	*start = rsrc->start - offset;
641 	*end = rsrc->end - offset;
642 }
643 
644 /**
645  * pci_process_bridge_OF_ranges - Parse PCI bridge resources from device tree
646  * @hose: newly allocated pci_controller to be setup
647  * @dev: device node of the host bridge
648  * @primary: set if primary bus (32 bits only, soon to be deprecated)
649  *
650  * This function will parse the "ranges" property of a PCI host bridge device
651  * node and setup the resource mapping of a pci controller based on its
652  * content.
653  *
654  * Life would be boring if it wasn't for a few issues that we have to deal
655  * with here:
656  *
657  *   - We can only cope with one IO space range and up to 3 Memory space
658  *     ranges. However, some machines (thanks Apple !) tend to split their
659  *     space into lots of small contiguous ranges. So we have to coalesce.
660  *
661  *   - We can only cope with all memory ranges having the same offset
662  *     between CPU addresses and PCI addresses. Unfortunately, some bridges
663  *     are setup for a large 1:1 mapping along with a small "window" which
664  *     maps PCI address 0 to some arbitrary high address of the CPU space in
665  *     order to give access to the ISA memory hole.
666  *     The way out of here that I've chosen for now is to always set the
667  *     offset based on the first resource found, then override it if we
668  *     have a different offset and the previous was set by an ISA hole.
669  *
670  *   - Some busses have IO space not starting at 0, which causes trouble with
671  *     the way we do our IO resource renumbering. The code somewhat deals with
672  *     it for 64 bits but I would expect problems on 32 bits.
673  *
674  *   - Some 32 bits platforms such as 4xx can have physical space larger than
675  *     32 bits so we need to use 64 bits values for the parsing
676  */
677 void __devinit pci_process_bridge_OF_ranges(struct pci_controller *hose,
678 					    struct device_node *dev,
679 					    int primary)
680 {
681 	const u32 *ranges;
682 	int rlen;
683 	int pna = of_n_addr_cells(dev);
684 	int np = pna + 5;
685 	int memno = 0, isa_hole = -1;
686 	u32 pci_space;
687 	unsigned long long pci_addr, cpu_addr, pci_next, cpu_next, size;
688 	unsigned long long isa_mb = 0;
689 	struct resource *res;
690 
691 	printk(KERN_INFO "PCI host bridge %s %s ranges:\n",
692 	       dev->full_name, primary ? "(primary)" : "");
693 
694 	/* Get ranges property */
695 	ranges = of_get_property(dev, "ranges", &rlen);
696 	if (ranges == NULL)
697 		return;
698 
699 	/* Parse it */
700 	while ((rlen -= np * 4) >= 0) {
701 		/* Read next ranges element */
702 		pci_space = ranges[0];
703 		pci_addr = of_read_number(ranges + 1, 2);
704 		cpu_addr = of_translate_address(dev, ranges + 3);
705 		size = of_read_number(ranges + pna + 3, 2);
706 		ranges += np;
707 
708 		/* If we failed translation or got a zero-sized region
709 		 * (some FW try to feed us with non sensical zero sized regions
710 		 * such as power3 which look like some kind of attempt at exposing
711 		 * the VGA memory hole)
712 		 */
713 		if (cpu_addr == OF_BAD_ADDR || size == 0)
714 			continue;
715 
716 		/* Now consume following elements while they are contiguous */
717 		for (; rlen >= np * sizeof(u32);
718 		     ranges += np, rlen -= np * 4) {
719 			if (ranges[0] != pci_space)
720 				break;
721 			pci_next = of_read_number(ranges + 1, 2);
722 			cpu_next = of_translate_address(dev, ranges + 3);
723 			if (pci_next != pci_addr + size ||
724 			    cpu_next != cpu_addr + size)
725 				break;
726 			size += of_read_number(ranges + pna + 3, 2);
727 		}
728 
729 		/* Act based on address space type */
730 		res = NULL;
731 		switch ((pci_space >> 24) & 0x3) {
732 		case 1:		/* PCI IO space */
733 			printk(KERN_INFO
734 			       "  IO 0x%016llx..0x%016llx -> 0x%016llx\n",
735 			       cpu_addr, cpu_addr + size - 1, pci_addr);
736 
737 			/* We support only one IO range */
738 			if (hose->pci_io_size) {
739 				printk(KERN_INFO
740 				       " \\--> Skipped (too many) !\n");
741 				continue;
742 			}
743 #ifdef CONFIG_PPC32
744 			/* On 32 bits, limit I/O space to 16MB */
745 			if (size > 0x01000000)
746 				size = 0x01000000;
747 
748 			/* 32 bits needs to map IOs here */
749 			hose->io_base_virt = ioremap(cpu_addr, size);
750 
751 			/* Expect trouble if pci_addr is not 0 */
752 			if (primary)
753 				isa_io_base =
754 					(unsigned long)hose->io_base_virt;
755 #endif /* CONFIG_PPC32 */
756 			/* pci_io_size and io_base_phys always represent IO
757 			 * space starting at 0 so we factor in pci_addr
758 			 */
759 			hose->pci_io_size = pci_addr + size;
760 			hose->io_base_phys = cpu_addr - pci_addr;
761 
762 			/* Build resource */
763 			res = &hose->io_resource;
764 			res->flags = IORESOURCE_IO;
765 			res->start = pci_addr;
766 			break;
767 		case 2:		/* PCI Memory space */
768 		case 3:		/* PCI 64 bits Memory space */
769 			printk(KERN_INFO
770 			       " MEM 0x%016llx..0x%016llx -> 0x%016llx %s\n",
771 			       cpu_addr, cpu_addr + size - 1, pci_addr,
772 			       (pci_space & 0x40000000) ? "Prefetch" : "");
773 
774 			/* We support only 3 memory ranges */
775 			if (memno >= 3) {
776 				printk(KERN_INFO
777 				       " \\--> Skipped (too many) !\n");
778 				continue;
779 			}
780 			/* Handles ISA memory hole space here */
781 			if (pci_addr == 0) {
782 				isa_mb = cpu_addr;
783 				isa_hole = memno;
784 				if (primary || isa_mem_base == 0)
785 					isa_mem_base = cpu_addr;
786 				hose->isa_mem_phys = cpu_addr;
787 				hose->isa_mem_size = size;
788 			}
789 
790 			/* We get the PCI/Mem offset from the first range or
791 			 * the, current one if the offset came from an ISA
792 			 * hole. If they don't match, bugger.
793 			 */
794 			if (memno == 0 ||
795 			    (isa_hole >= 0 && pci_addr != 0 &&
796 			     hose->pci_mem_offset == isa_mb))
797 				hose->pci_mem_offset = cpu_addr - pci_addr;
798 			else if (pci_addr != 0 &&
799 				 hose->pci_mem_offset != cpu_addr - pci_addr) {
800 				printk(KERN_INFO
801 				       " \\--> Skipped (offset mismatch) !\n");
802 				continue;
803 			}
804 
805 			/* Build resource */
806 			res = &hose->mem_resources[memno++];
807 			res->flags = IORESOURCE_MEM;
808 			if (pci_space & 0x40000000)
809 				res->flags |= IORESOURCE_PREFETCH;
810 			res->start = cpu_addr;
811 			break;
812 		}
813 		if (res != NULL) {
814 			res->name = dev->full_name;
815 			res->end = res->start + size - 1;
816 			res->parent = NULL;
817 			res->sibling = NULL;
818 			res->child = NULL;
819 		}
820 	}
821 
822 	/* If there's an ISA hole and the pci_mem_offset is -not- matching
823 	 * the ISA hole offset, then we need to remove the ISA hole from
824 	 * the resource list for that brige
825 	 */
826 	if (isa_hole >= 0 && hose->pci_mem_offset != isa_mb) {
827 		unsigned int next = isa_hole + 1;
828 		printk(KERN_INFO " Removing ISA hole at 0x%016llx\n", isa_mb);
829 		if (next < memno)
830 			memmove(&hose->mem_resources[isa_hole],
831 				&hose->mem_resources[next],
832 				sizeof(struct resource) * (memno - next));
833 		hose->mem_resources[--memno].flags = 0;
834 	}
835 }
836 
837 /* Decide whether to display the domain number in /proc */
838 int pci_proc_domain(struct pci_bus *bus)
839 {
840 	struct pci_controller *hose = pci_bus_to_host(bus);
841 
842 	if (!pci_has_flag(PCI_ENABLE_PROC_DOMAINS))
843 		return 0;
844 	if (pci_has_flag(PCI_COMPAT_DOMAIN_0))
845 		return hose->global_number != 0;
846 	return 1;
847 }
848 
849 /* This header fixup will do the resource fixup for all devices as they are
850  * probed, but not for bridge ranges
851  */
852 static void __devinit pcibios_fixup_resources(struct pci_dev *dev)
853 {
854 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
855 	int i;
856 
857 	if (!hose) {
858 		printk(KERN_ERR "No host bridge for PCI dev %s !\n",
859 		       pci_name(dev));
860 		return;
861 	}
862 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
863 		struct resource *res = dev->resource + i;
864 		if (!res->flags)
865 			continue;
866 
867 		/* If we're going to re-assign everything, we mark all resources
868 		 * as unset (and 0-base them). In addition, we mark BARs starting
869 		 * at 0 as unset as well, except if PCI_PROBE_ONLY is also set
870 		 * since in that case, we don't want to re-assign anything
871 		 */
872 		if (pci_has_flag(PCI_REASSIGN_ALL_RSRC) ||
873 		    (res->start == 0 && !pci_has_flag(PCI_PROBE_ONLY))) {
874 			/* Only print message if not re-assigning */
875 			if (!pci_has_flag(PCI_REASSIGN_ALL_RSRC))
876 				pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] "
877 					 "is unassigned\n",
878 					 pci_name(dev), i,
879 					 (unsigned long long)res->start,
880 					 (unsigned long long)res->end,
881 					 (unsigned int)res->flags);
882 			res->end -= res->start;
883 			res->start = 0;
884 			res->flags |= IORESOURCE_UNSET;
885 			continue;
886 		}
887 
888 		pr_debug("PCI:%s Resource %d %016llx-%016llx [%x]\n",
889 			 pci_name(dev), i,
890 			 (unsigned long long)res->start,\
891 			 (unsigned long long)res->end,
892 			 (unsigned int)res->flags);
893 	}
894 
895 	/* Call machine specific resource fixup */
896 	if (ppc_md.pcibios_fixup_resources)
897 		ppc_md.pcibios_fixup_resources(dev);
898 }
899 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_resources);
900 
901 /* This function tries to figure out if a bridge resource has been initialized
902  * by the firmware or not. It doesn't have to be absolutely bullet proof, but
903  * things go more smoothly when it gets it right. It should covers cases such
904  * as Apple "closed" bridge resources and bare-metal pSeries unassigned bridges
905  */
906 static int __devinit pcibios_uninitialized_bridge_resource(struct pci_bus *bus,
907 							   struct resource *res)
908 {
909 	struct pci_controller *hose = pci_bus_to_host(bus);
910 	struct pci_dev *dev = bus->self;
911 	resource_size_t offset;
912 	u16 command;
913 	int i;
914 
915 	/* We don't do anything if PCI_PROBE_ONLY is set */
916 	if (pci_has_flag(PCI_PROBE_ONLY))
917 		return 0;
918 
919 	/* Job is a bit different between memory and IO */
920 	if (res->flags & IORESOURCE_MEM) {
921 		/* If the BAR is non-0 (res != pci_mem_offset) then it's probably been
922 		 * initialized by somebody
923 		 */
924 		if (res->start != hose->pci_mem_offset)
925 			return 0;
926 
927 		/* The BAR is 0, let's check if memory decoding is enabled on
928 		 * the bridge. If not, we consider it unassigned
929 		 */
930 		pci_read_config_word(dev, PCI_COMMAND, &command);
931 		if ((command & PCI_COMMAND_MEMORY) == 0)
932 			return 1;
933 
934 		/* Memory decoding is enabled and the BAR is 0. If any of the bridge
935 		 * resources covers that starting address (0 then it's good enough for
936 		 * us for memory
937 		 */
938 		for (i = 0; i < 3; i++) {
939 			if ((hose->mem_resources[i].flags & IORESOURCE_MEM) &&
940 			    hose->mem_resources[i].start == hose->pci_mem_offset)
941 				return 0;
942 		}
943 
944 		/* Well, it starts at 0 and we know it will collide so we may as
945 		 * well consider it as unassigned. That covers the Apple case.
946 		 */
947 		return 1;
948 	} else {
949 		/* If the BAR is non-0, then we consider it assigned */
950 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
951 		if (((res->start - offset) & 0xfffffffful) != 0)
952 			return 0;
953 
954 		/* Here, we are a bit different than memory as typically IO space
955 		 * starting at low addresses -is- valid. What we do instead if that
956 		 * we consider as unassigned anything that doesn't have IO enabled
957 		 * in the PCI command register, and that's it.
958 		 */
959 		pci_read_config_word(dev, PCI_COMMAND, &command);
960 		if (command & PCI_COMMAND_IO)
961 			return 0;
962 
963 		/* It's starting at 0 and IO is disabled in the bridge, consider
964 		 * it unassigned
965 		 */
966 		return 1;
967 	}
968 }
969 
970 /* Fixup resources of a PCI<->PCI bridge */
971 static void __devinit pcibios_fixup_bridge(struct pci_bus *bus)
972 {
973 	struct resource *res;
974 	int i;
975 
976 	struct pci_dev *dev = bus->self;
977 
978 	pci_bus_for_each_resource(bus, res, i) {
979 		if (!res || !res->flags)
980 			continue;
981 		if (i >= 3 && bus->self->transparent)
982 			continue;
983 
984 		/* If we are going to re-assign everything, mark the resource
985 		 * as unset and move it down to 0
986 		 */
987 		if (pci_has_flag(PCI_REASSIGN_ALL_RSRC)) {
988 			res->flags |= IORESOURCE_UNSET;
989 			res->end -= res->start;
990 			res->start = 0;
991 			continue;
992 		}
993 
994 		pr_debug("PCI:%s Bus rsrc %d %016llx-%016llx [%x]\n",
995 			 pci_name(dev), i,
996 			 (unsigned long long)res->start,\
997 			 (unsigned long long)res->end,
998 			 (unsigned int)res->flags);
999 
1000 		/* Try to detect uninitialized P2P bridge resources,
1001 		 * and clear them out so they get re-assigned later
1002 		 */
1003 		if (pcibios_uninitialized_bridge_resource(bus, res)) {
1004 			res->flags = 0;
1005 			pr_debug("PCI:%s            (unassigned)\n", pci_name(dev));
1006 		}
1007 	}
1008 }
1009 
1010 void __devinit pcibios_setup_bus_self(struct pci_bus *bus)
1011 {
1012 	/* Fix up the bus resources for P2P bridges */
1013 	if (bus->self != NULL)
1014 		pcibios_fixup_bridge(bus);
1015 
1016 	/* Platform specific bus fixups. This is currently only used
1017 	 * by fsl_pci and I'm hoping to get rid of it at some point
1018 	 */
1019 	if (ppc_md.pcibios_fixup_bus)
1020 		ppc_md.pcibios_fixup_bus(bus);
1021 
1022 	/* Setup bus DMA mappings */
1023 	if (ppc_md.pci_dma_bus_setup)
1024 		ppc_md.pci_dma_bus_setup(bus);
1025 }
1026 
1027 void __devinit pcibios_setup_bus_devices(struct pci_bus *bus)
1028 {
1029 	struct pci_dev *dev;
1030 
1031 	pr_debug("PCI: Fixup bus devices %d (%s)\n",
1032 		 bus->number, bus->self ? pci_name(bus->self) : "PHB");
1033 
1034 	list_for_each_entry(dev, &bus->devices, bus_list) {
1035 		/* Cardbus can call us to add new devices to a bus, so ignore
1036 		 * those who are already fully discovered
1037 		 */
1038 		if (dev->is_added)
1039 			continue;
1040 
1041 		/* Fixup NUMA node as it may not be setup yet by the generic
1042 		 * code and is needed by the DMA init
1043 		 */
1044 		set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
1045 
1046 		/* Hook up default DMA ops */
1047 		set_dma_ops(&dev->dev, pci_dma_ops);
1048 		set_dma_offset(&dev->dev, PCI_DRAM_OFFSET);
1049 
1050 		/* Additional platform DMA/iommu setup */
1051 		if (ppc_md.pci_dma_dev_setup)
1052 			ppc_md.pci_dma_dev_setup(dev);
1053 
1054 		/* Read default IRQs and fixup if necessary */
1055 		pci_read_irq_line(dev);
1056 		if (ppc_md.pci_irq_fixup)
1057 			ppc_md.pci_irq_fixup(dev);
1058 	}
1059 }
1060 
1061 void pcibios_set_master(struct pci_dev *dev)
1062 {
1063 	/* No special bus mastering setup handling */
1064 }
1065 
1066 void __devinit pcibios_fixup_bus(struct pci_bus *bus)
1067 {
1068 	/* When called from the generic PCI probe, read PCI<->PCI bridge
1069 	 * bases. This is -not- called when generating the PCI tree from
1070 	 * the OF device-tree.
1071 	 */
1072 	if (bus->self != NULL)
1073 		pci_read_bridge_bases(bus);
1074 
1075 	/* Now fixup the bus bus */
1076 	pcibios_setup_bus_self(bus);
1077 
1078 	/* Now fixup devices on that bus */
1079 	pcibios_setup_bus_devices(bus);
1080 }
1081 EXPORT_SYMBOL(pcibios_fixup_bus);
1082 
1083 void __devinit pci_fixup_cardbus(struct pci_bus *bus)
1084 {
1085 	/* Now fixup devices on that bus */
1086 	pcibios_setup_bus_devices(bus);
1087 }
1088 
1089 
1090 static int skip_isa_ioresource_align(struct pci_dev *dev)
1091 {
1092 	if (pci_has_flag(PCI_CAN_SKIP_ISA_ALIGN) &&
1093 	    !(dev->bus->bridge_ctl & PCI_BRIDGE_CTL_ISA))
1094 		return 1;
1095 	return 0;
1096 }
1097 
1098 /*
1099  * We need to avoid collisions with `mirrored' VGA ports
1100  * and other strange ISA hardware, so we always want the
1101  * addresses to be allocated in the 0x000-0x0ff region
1102  * modulo 0x400.
1103  *
1104  * Why? Because some silly external IO cards only decode
1105  * the low 10 bits of the IO address. The 0x00-0xff region
1106  * is reserved for motherboard devices that decode all 16
1107  * bits, so it's ok to allocate at, say, 0x2800-0x28ff,
1108  * but we want to try to avoid allocating at 0x2900-0x2bff
1109  * which might have be mirrored at 0x0100-0x03ff..
1110  */
1111 resource_size_t pcibios_align_resource(void *data, const struct resource *res,
1112 				resource_size_t size, resource_size_t align)
1113 {
1114 	struct pci_dev *dev = data;
1115 	resource_size_t start = res->start;
1116 
1117 	if (res->flags & IORESOURCE_IO) {
1118 		if (skip_isa_ioresource_align(dev))
1119 			return start;
1120 		if (start & 0x300)
1121 			start = (start + 0x3ff) & ~0x3ff;
1122 	}
1123 
1124 	return start;
1125 }
1126 EXPORT_SYMBOL(pcibios_align_resource);
1127 
1128 /*
1129  * Reparent resource children of pr that conflict with res
1130  * under res, and make res replace those children.
1131  */
1132 static int reparent_resources(struct resource *parent,
1133 				     struct resource *res)
1134 {
1135 	struct resource *p, **pp;
1136 	struct resource **firstpp = NULL;
1137 
1138 	for (pp = &parent->child; (p = *pp) != NULL; pp = &p->sibling) {
1139 		if (p->end < res->start)
1140 			continue;
1141 		if (res->end < p->start)
1142 			break;
1143 		if (p->start < res->start || p->end > res->end)
1144 			return -1;	/* not completely contained */
1145 		if (firstpp == NULL)
1146 			firstpp = pp;
1147 	}
1148 	if (firstpp == NULL)
1149 		return -1;	/* didn't find any conflicting entries? */
1150 	res->parent = parent;
1151 	res->child = *firstpp;
1152 	res->sibling = *pp;
1153 	*firstpp = res;
1154 	*pp = NULL;
1155 	for (p = res->child; p != NULL; p = p->sibling) {
1156 		p->parent = res;
1157 		pr_debug("PCI: Reparented %s [%llx..%llx] under %s\n",
1158 			 p->name,
1159 			 (unsigned long long)p->start,
1160 			 (unsigned long long)p->end, res->name);
1161 	}
1162 	return 0;
1163 }
1164 
1165 /*
1166  *  Handle resources of PCI devices.  If the world were perfect, we could
1167  *  just allocate all the resource regions and do nothing more.  It isn't.
1168  *  On the other hand, we cannot just re-allocate all devices, as it would
1169  *  require us to know lots of host bridge internals.  So we attempt to
1170  *  keep as much of the original configuration as possible, but tweak it
1171  *  when it's found to be wrong.
1172  *
1173  *  Known BIOS problems we have to work around:
1174  *	- I/O or memory regions not configured
1175  *	- regions configured, but not enabled in the command register
1176  *	- bogus I/O addresses above 64K used
1177  *	- expansion ROMs left enabled (this may sound harmless, but given
1178  *	  the fact the PCI specs explicitly allow address decoders to be
1179  *	  shared between expansion ROMs and other resource regions, it's
1180  *	  at least dangerous)
1181  *
1182  *  Our solution:
1183  *	(1) Allocate resources for all buses behind PCI-to-PCI bridges.
1184  *	    This gives us fixed barriers on where we can allocate.
1185  *	(2) Allocate resources for all enabled devices.  If there is
1186  *	    a collision, just mark the resource as unallocated. Also
1187  *	    disable expansion ROMs during this step.
1188  *	(3) Try to allocate resources for disabled devices.  If the
1189  *	    resources were assigned correctly, everything goes well,
1190  *	    if they weren't, they won't disturb allocation of other
1191  *	    resources.
1192  *	(4) Assign new addresses to resources which were either
1193  *	    not configured at all or misconfigured.  If explicitly
1194  *	    requested by the user, configure expansion ROM address
1195  *	    as well.
1196  */
1197 
1198 void pcibios_allocate_bus_resources(struct pci_bus *bus)
1199 {
1200 	struct pci_bus *b;
1201 	int i;
1202 	struct resource *res, *pr;
1203 
1204 	pr_debug("PCI: Allocating bus resources for %04x:%02x...\n",
1205 		 pci_domain_nr(bus), bus->number);
1206 
1207 	pci_bus_for_each_resource(bus, res, i) {
1208 		if (!res || !res->flags || res->start > res->end || res->parent)
1209 			continue;
1210 
1211 		/* If the resource was left unset at this point, we clear it */
1212 		if (res->flags & IORESOURCE_UNSET)
1213 			goto clear_resource;
1214 
1215 		if (bus->parent == NULL)
1216 			pr = (res->flags & IORESOURCE_IO) ?
1217 				&ioport_resource : &iomem_resource;
1218 		else {
1219 			pr = pci_find_parent_resource(bus->self, res);
1220 			if (pr == res) {
1221 				/* this happens when the generic PCI
1222 				 * code (wrongly) decides that this
1223 				 * bridge is transparent  -- paulus
1224 				 */
1225 				continue;
1226 			}
1227 		}
1228 
1229 		pr_debug("PCI: %s (bus %d) bridge rsrc %d: %016llx-%016llx "
1230 			 "[0x%x], parent %p (%s)\n",
1231 			 bus->self ? pci_name(bus->self) : "PHB",
1232 			 bus->number, i,
1233 			 (unsigned long long)res->start,
1234 			 (unsigned long long)res->end,
1235 			 (unsigned int)res->flags,
1236 			 pr, (pr && pr->name) ? pr->name : "nil");
1237 
1238 		if (pr && !(pr->flags & IORESOURCE_UNSET)) {
1239 			if (request_resource(pr, res) == 0)
1240 				continue;
1241 			/*
1242 			 * Must be a conflict with an existing entry.
1243 			 * Move that entry (or entries) under the
1244 			 * bridge resource and try again.
1245 			 */
1246 			if (reparent_resources(pr, res) == 0)
1247 				continue;
1248 		}
1249 		pr_warning("PCI: Cannot allocate resource region "
1250 			   "%d of PCI bridge %d, will remap\n", i, bus->number);
1251 	clear_resource:
1252 		res->start = res->end = 0;
1253 		res->flags = 0;
1254 	}
1255 
1256 	list_for_each_entry(b, &bus->children, node)
1257 		pcibios_allocate_bus_resources(b);
1258 }
1259 
1260 static inline void __devinit alloc_resource(struct pci_dev *dev, int idx)
1261 {
1262 	struct resource *pr, *r = &dev->resource[idx];
1263 
1264 	pr_debug("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1265 		 pci_name(dev), idx,
1266 		 (unsigned long long)r->start,
1267 		 (unsigned long long)r->end,
1268 		 (unsigned int)r->flags);
1269 
1270 	pr = pci_find_parent_resource(dev, r);
1271 	if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1272 	    request_resource(pr, r) < 0) {
1273 		printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1274 		       " of device %s, will remap\n", idx, pci_name(dev));
1275 		if (pr)
1276 			pr_debug("PCI:  parent is %p: %016llx-%016llx [%x]\n",
1277 				 pr,
1278 				 (unsigned long long)pr->start,
1279 				 (unsigned long long)pr->end,
1280 				 (unsigned int)pr->flags);
1281 		/* We'll assign a new address later */
1282 		r->flags |= IORESOURCE_UNSET;
1283 		r->end -= r->start;
1284 		r->start = 0;
1285 	}
1286 }
1287 
1288 static void __init pcibios_allocate_resources(int pass)
1289 {
1290 	struct pci_dev *dev = NULL;
1291 	int idx, disabled;
1292 	u16 command;
1293 	struct resource *r;
1294 
1295 	for_each_pci_dev(dev) {
1296 		pci_read_config_word(dev, PCI_COMMAND, &command);
1297 		for (idx = 0; idx <= PCI_ROM_RESOURCE; idx++) {
1298 			r = &dev->resource[idx];
1299 			if (r->parent)		/* Already allocated */
1300 				continue;
1301 			if (!r->flags || (r->flags & IORESOURCE_UNSET))
1302 				continue;	/* Not assigned at all */
1303 			/* We only allocate ROMs on pass 1 just in case they
1304 			 * have been screwed up by firmware
1305 			 */
1306 			if (idx == PCI_ROM_RESOURCE )
1307 				disabled = 1;
1308 			if (r->flags & IORESOURCE_IO)
1309 				disabled = !(command & PCI_COMMAND_IO);
1310 			else
1311 				disabled = !(command & PCI_COMMAND_MEMORY);
1312 			if (pass == disabled)
1313 				alloc_resource(dev, idx);
1314 		}
1315 		if (pass)
1316 			continue;
1317 		r = &dev->resource[PCI_ROM_RESOURCE];
1318 		if (r->flags) {
1319 			/* Turn the ROM off, leave the resource region,
1320 			 * but keep it unregistered.
1321 			 */
1322 			u32 reg;
1323 			pci_read_config_dword(dev, dev->rom_base_reg, &reg);
1324 			if (reg & PCI_ROM_ADDRESS_ENABLE) {
1325 				pr_debug("PCI: Switching off ROM of %s\n",
1326 					 pci_name(dev));
1327 				r->flags &= ~IORESOURCE_ROM_ENABLE;
1328 				pci_write_config_dword(dev, dev->rom_base_reg,
1329 						       reg & ~PCI_ROM_ADDRESS_ENABLE);
1330 			}
1331 		}
1332 	}
1333 }
1334 
1335 static void __init pcibios_reserve_legacy_regions(struct pci_bus *bus)
1336 {
1337 	struct pci_controller *hose = pci_bus_to_host(bus);
1338 	resource_size_t	offset;
1339 	struct resource *res, *pres;
1340 	int i;
1341 
1342 	pr_debug("Reserving legacy ranges for domain %04x\n", pci_domain_nr(bus));
1343 
1344 	/* Check for IO */
1345 	if (!(hose->io_resource.flags & IORESOURCE_IO))
1346 		goto no_io;
1347 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1348 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1349 	BUG_ON(res == NULL);
1350 	res->name = "Legacy IO";
1351 	res->flags = IORESOURCE_IO;
1352 	res->start = offset;
1353 	res->end = (offset + 0xfff) & 0xfffffffful;
1354 	pr_debug("Candidate legacy IO: %pR\n", res);
1355 	if (request_resource(&hose->io_resource, res)) {
1356 		printk(KERN_DEBUG
1357 		       "PCI %04x:%02x Cannot reserve Legacy IO %pR\n",
1358 		       pci_domain_nr(bus), bus->number, res);
1359 		kfree(res);
1360 	}
1361 
1362  no_io:
1363 	/* Check for memory */
1364 	offset = hose->pci_mem_offset;
1365 	pr_debug("hose mem offset: %016llx\n", (unsigned long long)offset);
1366 	for (i = 0; i < 3; i++) {
1367 		pres = &hose->mem_resources[i];
1368 		if (!(pres->flags & IORESOURCE_MEM))
1369 			continue;
1370 		pr_debug("hose mem res: %pR\n", pres);
1371 		if ((pres->start - offset) <= 0xa0000 &&
1372 		    (pres->end - offset) >= 0xbffff)
1373 			break;
1374 	}
1375 	if (i >= 3)
1376 		return;
1377 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1378 	BUG_ON(res == NULL);
1379 	res->name = "Legacy VGA memory";
1380 	res->flags = IORESOURCE_MEM;
1381 	res->start = 0xa0000 + offset;
1382 	res->end = 0xbffff + offset;
1383 	pr_debug("Candidate VGA memory: %pR\n", res);
1384 	if (request_resource(pres, res)) {
1385 		printk(KERN_DEBUG
1386 		       "PCI %04x:%02x Cannot reserve VGA memory %pR\n",
1387 		       pci_domain_nr(bus), bus->number, res);
1388 		kfree(res);
1389 	}
1390 }
1391 
1392 void __init pcibios_resource_survey(void)
1393 {
1394 	struct pci_bus *b;
1395 
1396 	/* Allocate and assign resources */
1397 	list_for_each_entry(b, &pci_root_buses, node)
1398 		pcibios_allocate_bus_resources(b);
1399 	pcibios_allocate_resources(0);
1400 	pcibios_allocate_resources(1);
1401 
1402 	/* Before we start assigning unassigned resource, we try to reserve
1403 	 * the low IO area and the VGA memory area if they intersect the
1404 	 * bus available resources to avoid allocating things on top of them
1405 	 */
1406 	if (!pci_has_flag(PCI_PROBE_ONLY)) {
1407 		list_for_each_entry(b, &pci_root_buses, node)
1408 			pcibios_reserve_legacy_regions(b);
1409 	}
1410 
1411 	/* Now, if the platform didn't decide to blindly trust the firmware,
1412 	 * we proceed to assigning things that were left unassigned
1413 	 */
1414 	if (!pci_has_flag(PCI_PROBE_ONLY)) {
1415 		pr_debug("PCI: Assigning unassigned resources...\n");
1416 		pci_assign_unassigned_resources();
1417 	}
1418 
1419 	/* Call machine dependent fixup */
1420 	if (ppc_md.pcibios_fixup)
1421 		ppc_md.pcibios_fixup();
1422 }
1423 
1424 #ifdef CONFIG_HOTPLUG
1425 
1426 /* This is used by the PCI hotplug driver to allocate resource
1427  * of newly plugged busses. We can try to consolidate with the
1428  * rest of the code later, for now, keep it as-is as our main
1429  * resource allocation function doesn't deal with sub-trees yet.
1430  */
1431 void pcibios_claim_one_bus(struct pci_bus *bus)
1432 {
1433 	struct pci_dev *dev;
1434 	struct pci_bus *child_bus;
1435 
1436 	list_for_each_entry(dev, &bus->devices, bus_list) {
1437 		int i;
1438 
1439 		for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1440 			struct resource *r = &dev->resource[i];
1441 
1442 			if (r->parent || !r->start || !r->flags)
1443 				continue;
1444 
1445 			pr_debug("PCI: Claiming %s: "
1446 				 "Resource %d: %016llx..%016llx [%x]\n",
1447 				 pci_name(dev), i,
1448 				 (unsigned long long)r->start,
1449 				 (unsigned long long)r->end,
1450 				 (unsigned int)r->flags);
1451 
1452 			pci_claim_resource(dev, i);
1453 		}
1454 	}
1455 
1456 	list_for_each_entry(child_bus, &bus->children, node)
1457 		pcibios_claim_one_bus(child_bus);
1458 }
1459 
1460 
1461 /* pcibios_finish_adding_to_bus
1462  *
1463  * This is to be called by the hotplug code after devices have been
1464  * added to a bus, this include calling it for a PHB that is just
1465  * being added
1466  */
1467 void pcibios_finish_adding_to_bus(struct pci_bus *bus)
1468 {
1469 	pr_debug("PCI: Finishing adding to hotplug bus %04x:%02x\n",
1470 		 pci_domain_nr(bus), bus->number);
1471 
1472 	/* Allocate bus and devices resources */
1473 	pcibios_allocate_bus_resources(bus);
1474 	pcibios_claim_one_bus(bus);
1475 
1476 	/* Add new devices to global lists.  Register in proc, sysfs. */
1477 	pci_bus_add_devices(bus);
1478 
1479 	/* Fixup EEH */
1480 	eeh_add_device_tree_late(bus);
1481 }
1482 EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
1483 
1484 #endif /* CONFIG_HOTPLUG */
1485 
1486 int pcibios_enable_device(struct pci_dev *dev, int mask)
1487 {
1488 	if (ppc_md.pcibios_enable_device_hook)
1489 		if (ppc_md.pcibios_enable_device_hook(dev))
1490 			return -EINVAL;
1491 
1492 	return pci_enable_resources(dev, mask);
1493 }
1494 
1495 resource_size_t pcibios_io_space_offset(struct pci_controller *hose)
1496 {
1497 	return (unsigned long) hose->io_base_virt - _IO_BASE;
1498 }
1499 
1500 static void __devinit pcibios_setup_phb_resources(struct pci_controller *hose, struct list_head *resources)
1501 {
1502 	struct resource *res;
1503 	int i;
1504 
1505 	/* Hookup PHB IO resource */
1506 	res = &hose->io_resource;
1507 
1508 	if (!res->flags) {
1509 		printk(KERN_WARNING "PCI: I/O resource not set for host"
1510 		       " bridge %s (domain %d)\n",
1511 		       hose->dn->full_name, hose->global_number);
1512 #ifdef CONFIG_PPC32
1513 		/* Workaround for lack of IO resource only on 32-bit */
1514 		res->start = (unsigned long)hose->io_base_virt - isa_io_base;
1515 		res->end = res->start + IO_SPACE_LIMIT;
1516 		res->flags = IORESOURCE_IO;
1517 #endif /* CONFIG_PPC32 */
1518 	}
1519 
1520 	pr_debug("PCI: PHB IO resource    = %016llx-%016llx [%lx]\n",
1521 		 (unsigned long long)res->start,
1522 		 (unsigned long long)res->end,
1523 		 (unsigned long)res->flags);
1524 	pci_add_resource_offset(resources, res, pcibios_io_space_offset(hose));
1525 
1526 	/* Hookup PHB Memory resources */
1527 	for (i = 0; i < 3; ++i) {
1528 		res = &hose->mem_resources[i];
1529 		if (!res->flags) {
1530 			if (i > 0)
1531 				continue;
1532 			printk(KERN_ERR "PCI: Memory resource 0 not set for "
1533 			       "host bridge %s (domain %d)\n",
1534 			       hose->dn->full_name, hose->global_number);
1535 #ifdef CONFIG_PPC32
1536 			/* Workaround for lack of MEM resource only on 32-bit */
1537 			res->start = hose->pci_mem_offset;
1538 			res->end = (resource_size_t)-1LL;
1539 			res->flags = IORESOURCE_MEM;
1540 #endif /* CONFIG_PPC32 */
1541 		}
1542 
1543 		pr_debug("PCI: PHB MEM resource %d = %016llx-%016llx [%lx]\n", i,
1544 			 (unsigned long long)res->start,
1545 			 (unsigned long long)res->end,
1546 			 (unsigned long)res->flags);
1547 		pci_add_resource_offset(resources, res, hose->pci_mem_offset);
1548 	}
1549 
1550 	pr_debug("PCI: PHB MEM offset     = %016llx\n",
1551 		 (unsigned long long)hose->pci_mem_offset);
1552 	pr_debug("PCI: PHB IO  offset     = %08lx\n",
1553 		 (unsigned long)hose->io_base_virt - _IO_BASE);
1554 
1555 }
1556 
1557 /*
1558  * Null PCI config access functions, for the case when we can't
1559  * find a hose.
1560  */
1561 #define NULL_PCI_OP(rw, size, type)					\
1562 static int								\
1563 null_##rw##_config_##size(struct pci_dev *dev, int offset, type val)	\
1564 {									\
1565 	return PCIBIOS_DEVICE_NOT_FOUND;    				\
1566 }
1567 
1568 static int
1569 null_read_config(struct pci_bus *bus, unsigned int devfn, int offset,
1570 		 int len, u32 *val)
1571 {
1572 	return PCIBIOS_DEVICE_NOT_FOUND;
1573 }
1574 
1575 static int
1576 null_write_config(struct pci_bus *bus, unsigned int devfn, int offset,
1577 		  int len, u32 val)
1578 {
1579 	return PCIBIOS_DEVICE_NOT_FOUND;
1580 }
1581 
1582 static struct pci_ops null_pci_ops =
1583 {
1584 	.read = null_read_config,
1585 	.write = null_write_config,
1586 };
1587 
1588 /*
1589  * These functions are used early on before PCI scanning is done
1590  * and all of the pci_dev and pci_bus structures have been created.
1591  */
1592 static struct pci_bus *
1593 fake_pci_bus(struct pci_controller *hose, int busnr)
1594 {
1595 	static struct pci_bus bus;
1596 
1597 	if (hose == 0) {
1598 		printk(KERN_ERR "Can't find hose for PCI bus %d!\n", busnr);
1599 	}
1600 	bus.number = busnr;
1601 	bus.sysdata = hose;
1602 	bus.ops = hose? hose->ops: &null_pci_ops;
1603 	return &bus;
1604 }
1605 
1606 #define EARLY_PCI_OP(rw, size, type)					\
1607 int early_##rw##_config_##size(struct pci_controller *hose, int bus,	\
1608 			       int devfn, int offset, type value)	\
1609 {									\
1610 	return pci_bus_##rw##_config_##size(fake_pci_bus(hose, bus),	\
1611 					    devfn, offset, value);	\
1612 }
1613 
1614 EARLY_PCI_OP(read, byte, u8 *)
1615 EARLY_PCI_OP(read, word, u16 *)
1616 EARLY_PCI_OP(read, dword, u32 *)
1617 EARLY_PCI_OP(write, byte, u8)
1618 EARLY_PCI_OP(write, word, u16)
1619 EARLY_PCI_OP(write, dword, u32)
1620 
1621 extern int pci_bus_find_capability (struct pci_bus *bus, unsigned int devfn, int cap);
1622 int early_find_capability(struct pci_controller *hose, int bus, int devfn,
1623 			  int cap)
1624 {
1625 	return pci_bus_find_capability(fake_pci_bus(hose, bus), devfn, cap);
1626 }
1627 
1628 struct device_node *pcibios_get_phb_of_node(struct pci_bus *bus)
1629 {
1630 	struct pci_controller *hose = bus->sysdata;
1631 
1632 	return of_node_get(hose->dn);
1633 }
1634 
1635 /**
1636  * pci_scan_phb - Given a pci_controller, setup and scan the PCI bus
1637  * @hose: Pointer to the PCI host controller instance structure
1638  */
1639 void __devinit pcibios_scan_phb(struct pci_controller *hose)
1640 {
1641 	LIST_HEAD(resources);
1642 	struct pci_bus *bus;
1643 	struct device_node *node = hose->dn;
1644 	int mode;
1645 
1646 	pr_debug("PCI: Scanning PHB %s\n",
1647 		 node ? node->full_name : "<NO NAME>");
1648 
1649 	/* Get some IO space for the new PHB */
1650 	pcibios_setup_phb_io_space(hose);
1651 
1652 	/* Wire up PHB bus resources */
1653 	pcibios_setup_phb_resources(hose, &resources);
1654 
1655 	/* Create an empty bus for the toplevel */
1656 	bus = pci_create_root_bus(hose->parent, hose->first_busno,
1657 				  hose->ops, hose, &resources);
1658 	if (bus == NULL) {
1659 		pr_err("Failed to create bus for PCI domain %04x\n",
1660 			hose->global_number);
1661 		pci_free_resource_list(&resources);
1662 		return;
1663 	}
1664 	bus->secondary = hose->first_busno;
1665 	hose->bus = bus;
1666 
1667 	/* Get probe mode and perform scan */
1668 	mode = PCI_PROBE_NORMAL;
1669 	if (node && ppc_md.pci_probe_mode)
1670 		mode = ppc_md.pci_probe_mode(bus);
1671 	pr_debug("    probe mode: %d\n", mode);
1672 	if (mode == PCI_PROBE_DEVTREE) {
1673 		bus->subordinate = hose->last_busno;
1674 		of_scan_bus(node, bus);
1675 	}
1676 
1677 	if (mode == PCI_PROBE_NORMAL)
1678 		hose->last_busno = bus->subordinate = pci_scan_child_bus(bus);
1679 
1680 	/* Platform gets a chance to do some global fixups before
1681 	 * we proceed to resource allocation
1682 	 */
1683 	if (ppc_md.pcibios_fixup_phb)
1684 		ppc_md.pcibios_fixup_phb(hose);
1685 
1686 	/* Configure PCI Express settings */
1687 	if (bus && !pci_has_flag(PCI_PROBE_ONLY)) {
1688 		struct pci_bus *child;
1689 		list_for_each_entry(child, &bus->children, node) {
1690 			struct pci_dev *self = child->self;
1691 			if (!self)
1692 				continue;
1693 			pcie_bus_configure_settings(child, self->pcie_mpss);
1694 		}
1695 	}
1696 }
1697 
1698 static void fixup_hide_host_resource_fsl(struct pci_dev *dev)
1699 {
1700 	int i, class = dev->class >> 8;
1701 	/* When configured as agent, programing interface = 1 */
1702 	int prog_if = dev->class & 0xf;
1703 
1704 	if ((class == PCI_CLASS_PROCESSOR_POWERPC ||
1705 	     class == PCI_CLASS_BRIDGE_OTHER) &&
1706 		(dev->hdr_type == PCI_HEADER_TYPE_NORMAL) &&
1707 		(prog_if == 0) &&
1708 		(dev->bus->parent == NULL)) {
1709 		for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
1710 			dev->resource[i].start = 0;
1711 			dev->resource[i].end = 0;
1712 			dev->resource[i].flags = 0;
1713 		}
1714 	}
1715 }
1716 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_MOTOROLA, PCI_ANY_ID, fixup_hide_host_resource_fsl);
1717 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_FREESCALE, PCI_ANY_ID, fixup_hide_host_resource_fsl);
1718