xref: /linux/arch/powerpc/kernel/pci-common.c (revision c39b9fd728d8173ecda993524089fbc38211a17f)
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 #include <linux/vgaarb.h>
34 
35 #include <asm/processor.h>
36 #include <asm/io.h>
37 #include <asm/prom.h>
38 #include <asm/pci-bridge.h>
39 #include <asm/byteorder.h>
40 #include <asm/machdep.h>
41 #include <asm/ppc-pci.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 /*
104  * The function is used to return the minimal alignment
105  * for memory or I/O windows of the associated P2P bridge.
106  * By default, 4KiB alignment for I/O windows and 1MiB for
107  * memory windows.
108  */
109 resource_size_t pcibios_window_alignment(struct pci_bus *bus,
110 					 unsigned long type)
111 {
112 	if (ppc_md.pcibios_window_alignment)
113 		return ppc_md.pcibios_window_alignment(bus, type);
114 
115 	/*
116 	 * PCI core will figure out the default
117 	 * alignment: 4KiB for I/O and 1MiB for
118 	 * memory window.
119 	 */
120 	return 1;
121 }
122 
123 static resource_size_t pcibios_io_size(const struct pci_controller *hose)
124 {
125 #ifdef CONFIG_PPC64
126 	return hose->pci_io_size;
127 #else
128 	return resource_size(&hose->io_resource);
129 #endif
130 }
131 
132 int pcibios_vaddr_is_ioport(void __iomem *address)
133 {
134 	int ret = 0;
135 	struct pci_controller *hose;
136 	resource_size_t size;
137 
138 	spin_lock(&hose_spinlock);
139 	list_for_each_entry(hose, &hose_list, list_node) {
140 		size = pcibios_io_size(hose);
141 		if (address >= hose->io_base_virt &&
142 		    address < (hose->io_base_virt + size)) {
143 			ret = 1;
144 			break;
145 		}
146 	}
147 	spin_unlock(&hose_spinlock);
148 	return ret;
149 }
150 
151 unsigned long pci_address_to_pio(phys_addr_t address)
152 {
153 	struct pci_controller *hose;
154 	resource_size_t size;
155 	unsigned long ret = ~0;
156 
157 	spin_lock(&hose_spinlock);
158 	list_for_each_entry(hose, &hose_list, list_node) {
159 		size = pcibios_io_size(hose);
160 		if (address >= hose->io_base_phys &&
161 		    address < (hose->io_base_phys + size)) {
162 			unsigned long base =
163 				(unsigned long)hose->io_base_virt - _IO_BASE;
164 			ret = base + (address - hose->io_base_phys);
165 			break;
166 		}
167 	}
168 	spin_unlock(&hose_spinlock);
169 
170 	return ret;
171 }
172 EXPORT_SYMBOL_GPL(pci_address_to_pio);
173 
174 /*
175  * Return the domain number for this bus.
176  */
177 int pci_domain_nr(struct pci_bus *bus)
178 {
179 	struct pci_controller *hose = pci_bus_to_host(bus);
180 
181 	return hose->global_number;
182 }
183 EXPORT_SYMBOL(pci_domain_nr);
184 
185 /* This routine is meant to be used early during boot, when the
186  * PCI bus numbers have not yet been assigned, and you need to
187  * issue PCI config cycles to an OF device.
188  * It could also be used to "fix" RTAS config cycles if you want
189  * to set pci_assign_all_buses to 1 and still use RTAS for PCI
190  * config cycles.
191  */
192 struct pci_controller* pci_find_hose_for_OF_device(struct device_node* node)
193 {
194 	while(node) {
195 		struct pci_controller *hose, *tmp;
196 		list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
197 			if (hose->dn == node)
198 				return hose;
199 		node = node->parent;
200 	}
201 	return NULL;
202 }
203 
204 static ssize_t pci_show_devspec(struct device *dev,
205 		struct device_attribute *attr, char *buf)
206 {
207 	struct pci_dev *pdev;
208 	struct device_node *np;
209 
210 	pdev = to_pci_dev (dev);
211 	np = pci_device_to_OF_node(pdev);
212 	if (np == NULL || np->full_name == NULL)
213 		return 0;
214 	return sprintf(buf, "%s", np->full_name);
215 }
216 static DEVICE_ATTR(devspec, S_IRUGO, pci_show_devspec, NULL);
217 
218 /* Add sysfs properties */
219 int pcibios_add_platform_entries(struct pci_dev *pdev)
220 {
221 	return device_create_file(&pdev->dev, &dev_attr_devspec);
222 }
223 
224 /*
225  * Reads the interrupt pin to determine if interrupt is use by card.
226  * If the interrupt is used, then gets the interrupt line from the
227  * openfirmware and sets it in the pci_dev and pci_config line.
228  */
229 static int pci_read_irq_line(struct pci_dev *pci_dev)
230 {
231 	struct of_irq oirq;
232 	unsigned int virq;
233 
234 	pr_debug("PCI: Try to map irq for %s...\n", pci_name(pci_dev));
235 
236 #ifdef DEBUG
237 	memset(&oirq, 0xff, sizeof(oirq));
238 #endif
239 	/* Try to get a mapping from the device-tree */
240 	if (of_irq_map_pci(pci_dev, &oirq)) {
241 		u8 line, pin;
242 
243 		/* If that fails, lets fallback to what is in the config
244 		 * space and map that through the default controller. We
245 		 * also set the type to level low since that's what PCI
246 		 * interrupts are. If your platform does differently, then
247 		 * either provide a proper interrupt tree or don't use this
248 		 * function.
249 		 */
250 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_PIN, &pin))
251 			return -1;
252 		if (pin == 0)
253 			return -1;
254 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_LINE, &line) ||
255 		    line == 0xff || line == 0) {
256 			return -1;
257 		}
258 		pr_debug(" No map ! Using line %d (pin %d) from PCI config\n",
259 			 line, pin);
260 
261 		virq = irq_create_mapping(NULL, line);
262 		if (virq != NO_IRQ)
263 			irq_set_irq_type(virq, IRQ_TYPE_LEVEL_LOW);
264 	} else {
265 		pr_debug(" Got one, spec %d cells (0x%08x 0x%08x...) on %s\n",
266 			 oirq.size, oirq.specifier[0], oirq.specifier[1],
267 			 of_node_full_name(oirq.controller));
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 pci_process_bridge_OF_ranges(struct pci_controller *hose,
678 				  struct device_node *dev, int primary)
679 {
680 	const u32 *ranges;
681 	int rlen;
682 	int pna = of_n_addr_cells(dev);
683 	int np = pna + 5;
684 	int memno = 0, isa_hole = -1;
685 	u32 pci_space;
686 	unsigned long long pci_addr, cpu_addr, pci_next, cpu_next, size;
687 	unsigned long long isa_mb = 0;
688 	struct resource *res;
689 
690 	printk(KERN_INFO "PCI host bridge %s %s ranges:\n",
691 	       dev->full_name, primary ? "(primary)" : "");
692 
693 	/* Get ranges property */
694 	ranges = of_get_property(dev, "ranges", &rlen);
695 	if (ranges == NULL)
696 		return;
697 
698 	/* Parse it */
699 	while ((rlen -= np * 4) >= 0) {
700 		/* Read next ranges element */
701 		pci_space = ranges[0];
702 		pci_addr = of_read_number(ranges + 1, 2);
703 		cpu_addr = of_translate_address(dev, ranges + 3);
704 		size = of_read_number(ranges + pna + 3, 2);
705 		ranges += np;
706 
707 		/* If we failed translation or got a zero-sized region
708 		 * (some FW try to feed us with non sensical zero sized regions
709 		 * such as power3 which look like some kind of attempt at exposing
710 		 * the VGA memory hole)
711 		 */
712 		if (cpu_addr == OF_BAD_ADDR || size == 0)
713 			continue;
714 
715 		/* Now consume following elements while they are contiguous */
716 		for (; rlen >= np * sizeof(u32);
717 		     ranges += np, rlen -= np * 4) {
718 			if (ranges[0] != pci_space)
719 				break;
720 			pci_next = of_read_number(ranges + 1, 2);
721 			cpu_next = of_translate_address(dev, ranges + 3);
722 			if (pci_next != pci_addr + size ||
723 			    cpu_next != cpu_addr + size)
724 				break;
725 			size += of_read_number(ranges + pna + 3, 2);
726 		}
727 
728 		/* Act based on address space type */
729 		res = NULL;
730 		switch ((pci_space >> 24) & 0x3) {
731 		case 1:		/* PCI IO space */
732 			printk(KERN_INFO
733 			       "  IO 0x%016llx..0x%016llx -> 0x%016llx\n",
734 			       cpu_addr, cpu_addr + size - 1, pci_addr);
735 
736 			/* We support only one IO range */
737 			if (hose->pci_io_size) {
738 				printk(KERN_INFO
739 				       " \\--> Skipped (too many) !\n");
740 				continue;
741 			}
742 #ifdef CONFIG_PPC32
743 			/* On 32 bits, limit I/O space to 16MB */
744 			if (size > 0x01000000)
745 				size = 0x01000000;
746 
747 			/* 32 bits needs to map IOs here */
748 			hose->io_base_virt = ioremap(cpu_addr, size);
749 
750 			/* Expect trouble if pci_addr is not 0 */
751 			if (primary)
752 				isa_io_base =
753 					(unsigned long)hose->io_base_virt;
754 #endif /* CONFIG_PPC32 */
755 			/* pci_io_size and io_base_phys always represent IO
756 			 * space starting at 0 so we factor in pci_addr
757 			 */
758 			hose->pci_io_size = pci_addr + size;
759 			hose->io_base_phys = cpu_addr - pci_addr;
760 
761 			/* Build resource */
762 			res = &hose->io_resource;
763 			res->flags = IORESOURCE_IO;
764 			res->start = pci_addr;
765 			break;
766 		case 2:		/* PCI Memory space */
767 		case 3:		/* PCI 64 bits Memory space */
768 			printk(KERN_INFO
769 			       " MEM 0x%016llx..0x%016llx -> 0x%016llx %s\n",
770 			       cpu_addr, cpu_addr + size - 1, pci_addr,
771 			       (pci_space & 0x40000000) ? "Prefetch" : "");
772 
773 			/* We support only 3 memory ranges */
774 			if (memno >= 3) {
775 				printk(KERN_INFO
776 				       " \\--> Skipped (too many) !\n");
777 				continue;
778 			}
779 			/* Handles ISA memory hole space here */
780 			if (pci_addr == 0) {
781 				isa_mb = cpu_addr;
782 				isa_hole = memno;
783 				if (primary || isa_mem_base == 0)
784 					isa_mem_base = cpu_addr;
785 				hose->isa_mem_phys = cpu_addr;
786 				hose->isa_mem_size = size;
787 			}
788 
789 			/* We get the PCI/Mem offset from the first range or
790 			 * the, current one if the offset came from an ISA
791 			 * hole. If they don't match, bugger.
792 			 */
793 			if (memno == 0 ||
794 			    (isa_hole >= 0 && pci_addr != 0 &&
795 			     hose->pci_mem_offset == isa_mb))
796 				hose->pci_mem_offset = cpu_addr - pci_addr;
797 			else if (pci_addr != 0 &&
798 				 hose->pci_mem_offset != cpu_addr - pci_addr) {
799 				printk(KERN_INFO
800 				       " \\--> Skipped (offset mismatch) !\n");
801 				continue;
802 			}
803 
804 			/* Build resource */
805 			res = &hose->mem_resources[memno++];
806 			res->flags = IORESOURCE_MEM;
807 			if (pci_space & 0x40000000)
808 				res->flags |= IORESOURCE_PREFETCH;
809 			res->start = cpu_addr;
810 			break;
811 		}
812 		if (res != NULL) {
813 			res->name = dev->full_name;
814 			res->end = res->start + size - 1;
815 			res->parent = NULL;
816 			res->sibling = NULL;
817 			res->child = NULL;
818 		}
819 	}
820 
821 	/* If there's an ISA hole and the pci_mem_offset is -not- matching
822 	 * the ISA hole offset, then we need to remove the ISA hole from
823 	 * the resource list for that brige
824 	 */
825 	if (isa_hole >= 0 && hose->pci_mem_offset != isa_mb) {
826 		unsigned int next = isa_hole + 1;
827 		printk(KERN_INFO " Removing ISA hole at 0x%016llx\n", isa_mb);
828 		if (next < memno)
829 			memmove(&hose->mem_resources[isa_hole],
830 				&hose->mem_resources[next],
831 				sizeof(struct resource) * (memno - next));
832 		hose->mem_resources[--memno].flags = 0;
833 	}
834 }
835 
836 /* Decide whether to display the domain number in /proc */
837 int pci_proc_domain(struct pci_bus *bus)
838 {
839 	struct pci_controller *hose = pci_bus_to_host(bus);
840 
841 	if (!pci_has_flag(PCI_ENABLE_PROC_DOMAINS))
842 		return 0;
843 	if (pci_has_flag(PCI_COMPAT_DOMAIN_0))
844 		return hose->global_number != 0;
845 	return 1;
846 }
847 
848 /* This header fixup will do the resource fixup for all devices as they are
849  * probed, but not for bridge ranges
850  */
851 static void pcibios_fixup_resources(struct pci_dev *dev)
852 {
853 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
854 	int i;
855 
856 	if (!hose) {
857 		printk(KERN_ERR "No host bridge for PCI dev %s !\n",
858 		       pci_name(dev));
859 		return;
860 	}
861 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
862 		struct resource *res = dev->resource + i;
863 		if (!res->flags)
864 			continue;
865 
866 		/* If we're going to re-assign everything, we mark all resources
867 		 * as unset (and 0-base them). In addition, we mark BARs starting
868 		 * at 0 as unset as well, except if PCI_PROBE_ONLY is also set
869 		 * since in that case, we don't want to re-assign anything
870 		 */
871 		if (pci_has_flag(PCI_REASSIGN_ALL_RSRC) ||
872 		    (res->start == 0 && !pci_has_flag(PCI_PROBE_ONLY))) {
873 			/* Only print message if not re-assigning */
874 			if (!pci_has_flag(PCI_REASSIGN_ALL_RSRC))
875 				pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] "
876 					 "is unassigned\n",
877 					 pci_name(dev), i,
878 					 (unsigned long long)res->start,
879 					 (unsigned long long)res->end,
880 					 (unsigned int)res->flags);
881 			res->end -= res->start;
882 			res->start = 0;
883 			res->flags |= IORESOURCE_UNSET;
884 			continue;
885 		}
886 
887 		pr_debug("PCI:%s Resource %d %016llx-%016llx [%x]\n",
888 			 pci_name(dev), i,
889 			 (unsigned long long)res->start,\
890 			 (unsigned long long)res->end,
891 			 (unsigned int)res->flags);
892 	}
893 
894 	/* Call machine specific resource fixup */
895 	if (ppc_md.pcibios_fixup_resources)
896 		ppc_md.pcibios_fixup_resources(dev);
897 }
898 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_resources);
899 
900 /* This function tries to figure out if a bridge resource has been initialized
901  * by the firmware or not. It doesn't have to be absolutely bullet proof, but
902  * things go more smoothly when it gets it right. It should covers cases such
903  * as Apple "closed" bridge resources and bare-metal pSeries unassigned bridges
904  */
905 static int pcibios_uninitialized_bridge_resource(struct pci_bus *bus,
906 						 struct resource *res)
907 {
908 	struct pci_controller *hose = pci_bus_to_host(bus);
909 	struct pci_dev *dev = bus->self;
910 	resource_size_t offset;
911 	u16 command;
912 	int i;
913 
914 	/* We don't do anything if PCI_PROBE_ONLY is set */
915 	if (pci_has_flag(PCI_PROBE_ONLY))
916 		return 0;
917 
918 	/* Job is a bit different between memory and IO */
919 	if (res->flags & IORESOURCE_MEM) {
920 		/* If the BAR is non-0 (res != pci_mem_offset) then it's probably been
921 		 * initialized by somebody
922 		 */
923 		if (res->start != hose->pci_mem_offset)
924 			return 0;
925 
926 		/* The BAR is 0, let's check if memory decoding is enabled on
927 		 * the bridge. If not, we consider it unassigned
928 		 */
929 		pci_read_config_word(dev, PCI_COMMAND, &command);
930 		if ((command & PCI_COMMAND_MEMORY) == 0)
931 			return 1;
932 
933 		/* Memory decoding is enabled and the BAR is 0. If any of the bridge
934 		 * resources covers that starting address (0 then it's good enough for
935 		 * us for memory
936 		 */
937 		for (i = 0; i < 3; i++) {
938 			if ((hose->mem_resources[i].flags & IORESOURCE_MEM) &&
939 			    hose->mem_resources[i].start == hose->pci_mem_offset)
940 				return 0;
941 		}
942 
943 		/* Well, it starts at 0 and we know it will collide so we may as
944 		 * well consider it as unassigned. That covers the Apple case.
945 		 */
946 		return 1;
947 	} else {
948 		/* If the BAR is non-0, then we consider it assigned */
949 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
950 		if (((res->start - offset) & 0xfffffffful) != 0)
951 			return 0;
952 
953 		/* Here, we are a bit different than memory as typically IO space
954 		 * starting at low addresses -is- valid. What we do instead if that
955 		 * we consider as unassigned anything that doesn't have IO enabled
956 		 * in the PCI command register, and that's it.
957 		 */
958 		pci_read_config_word(dev, PCI_COMMAND, &command);
959 		if (command & PCI_COMMAND_IO)
960 			return 0;
961 
962 		/* It's starting at 0 and IO is disabled in the bridge, consider
963 		 * it unassigned
964 		 */
965 		return 1;
966 	}
967 }
968 
969 /* Fixup resources of a PCI<->PCI bridge */
970 static void pcibios_fixup_bridge(struct pci_bus *bus)
971 {
972 	struct resource *res;
973 	int i;
974 
975 	struct pci_dev *dev = bus->self;
976 
977 	pci_bus_for_each_resource(bus, res, i) {
978 		if (!res || !res->flags)
979 			continue;
980 		if (i >= 3 && bus->self->transparent)
981 			continue;
982 
983 		/* If we're going to reassign everything, we can
984 		 * shrink the P2P resource to have size as being
985 		 * of 0 in order to save space.
986 		 */
987 		if (pci_has_flag(PCI_REASSIGN_ALL_RSRC)) {
988 			res->flags |= IORESOURCE_UNSET;
989 			res->start = 0;
990 			res->end = -1;
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 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 pcibios_setup_device(struct pci_dev *dev)
1028 {
1029 	/* Fixup NUMA node as it may not be setup yet by the generic
1030 	 * code and is needed by the DMA init
1031 	 */
1032 	set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
1033 
1034 	/* Hook up default DMA ops */
1035 	set_dma_ops(&dev->dev, pci_dma_ops);
1036 	set_dma_offset(&dev->dev, PCI_DRAM_OFFSET);
1037 
1038 	/* Additional platform DMA/iommu setup */
1039 	if (ppc_md.pci_dma_dev_setup)
1040 		ppc_md.pci_dma_dev_setup(dev);
1041 
1042 	/* Read default IRQs and fixup if necessary */
1043 	pci_read_irq_line(dev);
1044 	if (ppc_md.pci_irq_fixup)
1045 		ppc_md.pci_irq_fixup(dev);
1046 }
1047 
1048 void pcibios_setup_bus_devices(struct pci_bus *bus)
1049 {
1050 	struct pci_dev *dev;
1051 
1052 	pr_debug("PCI: Fixup bus devices %d (%s)\n",
1053 		 bus->number, bus->self ? pci_name(bus->self) : "PHB");
1054 
1055 	list_for_each_entry(dev, &bus->devices, bus_list) {
1056 		/* Cardbus can call us to add new devices to a bus, so ignore
1057 		 * those who are already fully discovered
1058 		 */
1059 		if (dev->is_added)
1060 			continue;
1061 
1062 		pcibios_setup_device(dev);
1063 	}
1064 }
1065 
1066 void pcibios_set_master(struct pci_dev *dev)
1067 {
1068 	/* No special bus mastering setup handling */
1069 }
1070 
1071 void pcibios_fixup_bus(struct pci_bus *bus)
1072 {
1073 	/* When called from the generic PCI probe, read PCI<->PCI bridge
1074 	 * bases. This is -not- called when generating the PCI tree from
1075 	 * the OF device-tree.
1076 	 */
1077 	if (bus->self != NULL)
1078 		pci_read_bridge_bases(bus);
1079 
1080 	/* Now fixup the bus bus */
1081 	pcibios_setup_bus_self(bus);
1082 
1083 	/* Now fixup devices on that bus */
1084 	pcibios_setup_bus_devices(bus);
1085 }
1086 EXPORT_SYMBOL(pcibios_fixup_bus);
1087 
1088 void pci_fixup_cardbus(struct pci_bus *bus)
1089 {
1090 	/* Now fixup devices on that bus */
1091 	pcibios_setup_bus_devices(bus);
1092 }
1093 
1094 
1095 static int skip_isa_ioresource_align(struct pci_dev *dev)
1096 {
1097 	if (pci_has_flag(PCI_CAN_SKIP_ISA_ALIGN) &&
1098 	    !(dev->bus->bridge_ctl & PCI_BRIDGE_CTL_ISA))
1099 		return 1;
1100 	return 0;
1101 }
1102 
1103 /*
1104  * We need to avoid collisions with `mirrored' VGA ports
1105  * and other strange ISA hardware, so we always want the
1106  * addresses to be allocated in the 0x000-0x0ff region
1107  * modulo 0x400.
1108  *
1109  * Why? Because some silly external IO cards only decode
1110  * the low 10 bits of the IO address. The 0x00-0xff region
1111  * is reserved for motherboard devices that decode all 16
1112  * bits, so it's ok to allocate at, say, 0x2800-0x28ff,
1113  * but we want to try to avoid allocating at 0x2900-0x2bff
1114  * which might have be mirrored at 0x0100-0x03ff..
1115  */
1116 resource_size_t pcibios_align_resource(void *data, const struct resource *res,
1117 				resource_size_t size, resource_size_t align)
1118 {
1119 	struct pci_dev *dev = data;
1120 	resource_size_t start = res->start;
1121 
1122 	if (res->flags & IORESOURCE_IO) {
1123 		if (skip_isa_ioresource_align(dev))
1124 			return start;
1125 		if (start & 0x300)
1126 			start = (start + 0x3ff) & ~0x3ff;
1127 	}
1128 
1129 	return start;
1130 }
1131 EXPORT_SYMBOL(pcibios_align_resource);
1132 
1133 /*
1134  * Reparent resource children of pr that conflict with res
1135  * under res, and make res replace those children.
1136  */
1137 static int reparent_resources(struct resource *parent,
1138 				     struct resource *res)
1139 {
1140 	struct resource *p, **pp;
1141 	struct resource **firstpp = NULL;
1142 
1143 	for (pp = &parent->child; (p = *pp) != NULL; pp = &p->sibling) {
1144 		if (p->end < res->start)
1145 			continue;
1146 		if (res->end < p->start)
1147 			break;
1148 		if (p->start < res->start || p->end > res->end)
1149 			return -1;	/* not completely contained */
1150 		if (firstpp == NULL)
1151 			firstpp = pp;
1152 	}
1153 	if (firstpp == NULL)
1154 		return -1;	/* didn't find any conflicting entries? */
1155 	res->parent = parent;
1156 	res->child = *firstpp;
1157 	res->sibling = *pp;
1158 	*firstpp = res;
1159 	*pp = NULL;
1160 	for (p = res->child; p != NULL; p = p->sibling) {
1161 		p->parent = res;
1162 		pr_debug("PCI: Reparented %s [%llx..%llx] under %s\n",
1163 			 p->name,
1164 			 (unsigned long long)p->start,
1165 			 (unsigned long long)p->end, res->name);
1166 	}
1167 	return 0;
1168 }
1169 
1170 /*
1171  *  Handle resources of PCI devices.  If the world were perfect, we could
1172  *  just allocate all the resource regions and do nothing more.  It isn't.
1173  *  On the other hand, we cannot just re-allocate all devices, as it would
1174  *  require us to know lots of host bridge internals.  So we attempt to
1175  *  keep as much of the original configuration as possible, but tweak it
1176  *  when it's found to be wrong.
1177  *
1178  *  Known BIOS problems we have to work around:
1179  *	- I/O or memory regions not configured
1180  *	- regions configured, but not enabled in the command register
1181  *	- bogus I/O addresses above 64K used
1182  *	- expansion ROMs left enabled (this may sound harmless, but given
1183  *	  the fact the PCI specs explicitly allow address decoders to be
1184  *	  shared between expansion ROMs and other resource regions, it's
1185  *	  at least dangerous)
1186  *
1187  *  Our solution:
1188  *	(1) Allocate resources for all buses behind PCI-to-PCI bridges.
1189  *	    This gives us fixed barriers on where we can allocate.
1190  *	(2) Allocate resources for all enabled devices.  If there is
1191  *	    a collision, just mark the resource as unallocated. Also
1192  *	    disable expansion ROMs during this step.
1193  *	(3) Try to allocate resources for disabled devices.  If the
1194  *	    resources were assigned correctly, everything goes well,
1195  *	    if they weren't, they won't disturb allocation of other
1196  *	    resources.
1197  *	(4) Assign new addresses to resources which were either
1198  *	    not configured at all or misconfigured.  If explicitly
1199  *	    requested by the user, configure expansion ROM address
1200  *	    as well.
1201  */
1202 
1203 void pcibios_allocate_bus_resources(struct pci_bus *bus)
1204 {
1205 	struct pci_bus *b;
1206 	int i;
1207 	struct resource *res, *pr;
1208 
1209 	pr_debug("PCI: Allocating bus resources for %04x:%02x...\n",
1210 		 pci_domain_nr(bus), bus->number);
1211 
1212 	pci_bus_for_each_resource(bus, res, i) {
1213 		if (!res || !res->flags || res->start > res->end || res->parent)
1214 			continue;
1215 
1216 		/* If the resource was left unset at this point, we clear it */
1217 		if (res->flags & IORESOURCE_UNSET)
1218 			goto clear_resource;
1219 
1220 		if (bus->parent == NULL)
1221 			pr = (res->flags & IORESOURCE_IO) ?
1222 				&ioport_resource : &iomem_resource;
1223 		else {
1224 			pr = pci_find_parent_resource(bus->self, res);
1225 			if (pr == res) {
1226 				/* this happens when the generic PCI
1227 				 * code (wrongly) decides that this
1228 				 * bridge is transparent  -- paulus
1229 				 */
1230 				continue;
1231 			}
1232 		}
1233 
1234 		pr_debug("PCI: %s (bus %d) bridge rsrc %d: %016llx-%016llx "
1235 			 "[0x%x], parent %p (%s)\n",
1236 			 bus->self ? pci_name(bus->self) : "PHB",
1237 			 bus->number, i,
1238 			 (unsigned long long)res->start,
1239 			 (unsigned long long)res->end,
1240 			 (unsigned int)res->flags,
1241 			 pr, (pr && pr->name) ? pr->name : "nil");
1242 
1243 		if (pr && !(pr->flags & IORESOURCE_UNSET)) {
1244 			if (request_resource(pr, res) == 0)
1245 				continue;
1246 			/*
1247 			 * Must be a conflict with an existing entry.
1248 			 * Move that entry (or entries) under the
1249 			 * bridge resource and try again.
1250 			 */
1251 			if (reparent_resources(pr, res) == 0)
1252 				continue;
1253 		}
1254 		pr_warning("PCI: Cannot allocate resource region "
1255 			   "%d of PCI bridge %d, will remap\n", i, bus->number);
1256 	clear_resource:
1257 		/* The resource might be figured out when doing
1258 		 * reassignment based on the resources required
1259 		 * by the downstream PCI devices. Here we set
1260 		 * the size of the resource to be 0 in order to
1261 		 * save more space.
1262 		 */
1263 		res->start = 0;
1264 		res->end = -1;
1265 		res->flags = 0;
1266 	}
1267 
1268 	list_for_each_entry(b, &bus->children, node)
1269 		pcibios_allocate_bus_resources(b);
1270 }
1271 
1272 static inline void alloc_resource(struct pci_dev *dev, int idx)
1273 {
1274 	struct resource *pr, *r = &dev->resource[idx];
1275 
1276 	pr_debug("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1277 		 pci_name(dev), idx,
1278 		 (unsigned long long)r->start,
1279 		 (unsigned long long)r->end,
1280 		 (unsigned int)r->flags);
1281 
1282 	pr = pci_find_parent_resource(dev, r);
1283 	if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1284 	    request_resource(pr, r) < 0) {
1285 		printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1286 		       " of device %s, will remap\n", idx, pci_name(dev));
1287 		if (pr)
1288 			pr_debug("PCI:  parent is %p: %016llx-%016llx [%x]\n",
1289 				 pr,
1290 				 (unsigned long long)pr->start,
1291 				 (unsigned long long)pr->end,
1292 				 (unsigned int)pr->flags);
1293 		/* We'll assign a new address later */
1294 		r->flags |= IORESOURCE_UNSET;
1295 		r->end -= r->start;
1296 		r->start = 0;
1297 	}
1298 }
1299 
1300 static void __init pcibios_allocate_resources(int pass)
1301 {
1302 	struct pci_dev *dev = NULL;
1303 	int idx, disabled;
1304 	u16 command;
1305 	struct resource *r;
1306 
1307 	for_each_pci_dev(dev) {
1308 		pci_read_config_word(dev, PCI_COMMAND, &command);
1309 		for (idx = 0; idx <= PCI_ROM_RESOURCE; idx++) {
1310 			r = &dev->resource[idx];
1311 			if (r->parent)		/* Already allocated */
1312 				continue;
1313 			if (!r->flags || (r->flags & IORESOURCE_UNSET))
1314 				continue;	/* Not assigned at all */
1315 			/* We only allocate ROMs on pass 1 just in case they
1316 			 * have been screwed up by firmware
1317 			 */
1318 			if (idx == PCI_ROM_RESOURCE )
1319 				disabled = 1;
1320 			if (r->flags & IORESOURCE_IO)
1321 				disabled = !(command & PCI_COMMAND_IO);
1322 			else
1323 				disabled = !(command & PCI_COMMAND_MEMORY);
1324 			if (pass == disabled)
1325 				alloc_resource(dev, idx);
1326 		}
1327 		if (pass)
1328 			continue;
1329 		r = &dev->resource[PCI_ROM_RESOURCE];
1330 		if (r->flags) {
1331 			/* Turn the ROM off, leave the resource region,
1332 			 * but keep it unregistered.
1333 			 */
1334 			u32 reg;
1335 			pci_read_config_dword(dev, dev->rom_base_reg, &reg);
1336 			if (reg & PCI_ROM_ADDRESS_ENABLE) {
1337 				pr_debug("PCI: Switching off ROM of %s\n",
1338 					 pci_name(dev));
1339 				r->flags &= ~IORESOURCE_ROM_ENABLE;
1340 				pci_write_config_dword(dev, dev->rom_base_reg,
1341 						       reg & ~PCI_ROM_ADDRESS_ENABLE);
1342 			}
1343 		}
1344 	}
1345 }
1346 
1347 static void __init pcibios_reserve_legacy_regions(struct pci_bus *bus)
1348 {
1349 	struct pci_controller *hose = pci_bus_to_host(bus);
1350 	resource_size_t	offset;
1351 	struct resource *res, *pres;
1352 	int i;
1353 
1354 	pr_debug("Reserving legacy ranges for domain %04x\n", pci_domain_nr(bus));
1355 
1356 	/* Check for IO */
1357 	if (!(hose->io_resource.flags & IORESOURCE_IO))
1358 		goto no_io;
1359 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1360 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1361 	BUG_ON(res == NULL);
1362 	res->name = "Legacy IO";
1363 	res->flags = IORESOURCE_IO;
1364 	res->start = offset;
1365 	res->end = (offset + 0xfff) & 0xfffffffful;
1366 	pr_debug("Candidate legacy IO: %pR\n", res);
1367 	if (request_resource(&hose->io_resource, res)) {
1368 		printk(KERN_DEBUG
1369 		       "PCI %04x:%02x Cannot reserve Legacy IO %pR\n",
1370 		       pci_domain_nr(bus), bus->number, res);
1371 		kfree(res);
1372 	}
1373 
1374  no_io:
1375 	/* Check for memory */
1376 	offset = hose->pci_mem_offset;
1377 	pr_debug("hose mem offset: %016llx\n", (unsigned long long)offset);
1378 	for (i = 0; i < 3; i++) {
1379 		pres = &hose->mem_resources[i];
1380 		if (!(pres->flags & IORESOURCE_MEM))
1381 			continue;
1382 		pr_debug("hose mem res: %pR\n", pres);
1383 		if ((pres->start - offset) <= 0xa0000 &&
1384 		    (pres->end - offset) >= 0xbffff)
1385 			break;
1386 	}
1387 	if (i >= 3)
1388 		return;
1389 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1390 	BUG_ON(res == NULL);
1391 	res->name = "Legacy VGA memory";
1392 	res->flags = IORESOURCE_MEM;
1393 	res->start = 0xa0000 + offset;
1394 	res->end = 0xbffff + offset;
1395 	pr_debug("Candidate VGA memory: %pR\n", res);
1396 	if (request_resource(pres, res)) {
1397 		printk(KERN_DEBUG
1398 		       "PCI %04x:%02x Cannot reserve VGA memory %pR\n",
1399 		       pci_domain_nr(bus), bus->number, res);
1400 		kfree(res);
1401 	}
1402 }
1403 
1404 void __init pcibios_resource_survey(void)
1405 {
1406 	struct pci_bus *b;
1407 
1408 	/* Allocate and assign resources */
1409 	list_for_each_entry(b, &pci_root_buses, node)
1410 		pcibios_allocate_bus_resources(b);
1411 	pcibios_allocate_resources(0);
1412 	pcibios_allocate_resources(1);
1413 
1414 	/* Before we start assigning unassigned resource, we try to reserve
1415 	 * the low IO area and the VGA memory area if they intersect the
1416 	 * bus available resources to avoid allocating things on top of them
1417 	 */
1418 	if (!pci_has_flag(PCI_PROBE_ONLY)) {
1419 		list_for_each_entry(b, &pci_root_buses, node)
1420 			pcibios_reserve_legacy_regions(b);
1421 	}
1422 
1423 	/* Now, if the platform didn't decide to blindly trust the firmware,
1424 	 * we proceed to assigning things that were left unassigned
1425 	 */
1426 	if (!pci_has_flag(PCI_PROBE_ONLY)) {
1427 		pr_debug("PCI: Assigning unassigned resources...\n");
1428 		pci_assign_unassigned_resources();
1429 	}
1430 
1431 	/* Call machine dependent fixup */
1432 	if (ppc_md.pcibios_fixup)
1433 		ppc_md.pcibios_fixup();
1434 }
1435 
1436 /* This is used by the PCI hotplug driver to allocate resource
1437  * of newly plugged busses. We can try to consolidate with the
1438  * rest of the code later, for now, keep it as-is as our main
1439  * resource allocation function doesn't deal with sub-trees yet.
1440  */
1441 void pcibios_claim_one_bus(struct pci_bus *bus)
1442 {
1443 	struct pci_dev *dev;
1444 	struct pci_bus *child_bus;
1445 
1446 	list_for_each_entry(dev, &bus->devices, bus_list) {
1447 		int i;
1448 
1449 		for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1450 			struct resource *r = &dev->resource[i];
1451 
1452 			if (r->parent || !r->start || !r->flags)
1453 				continue;
1454 
1455 			pr_debug("PCI: Claiming %s: "
1456 				 "Resource %d: %016llx..%016llx [%x]\n",
1457 				 pci_name(dev), i,
1458 				 (unsigned long long)r->start,
1459 				 (unsigned long long)r->end,
1460 				 (unsigned int)r->flags);
1461 
1462 			pci_claim_resource(dev, i);
1463 		}
1464 	}
1465 
1466 	list_for_each_entry(child_bus, &bus->children, node)
1467 		pcibios_claim_one_bus(child_bus);
1468 }
1469 
1470 
1471 /* pcibios_finish_adding_to_bus
1472  *
1473  * This is to be called by the hotplug code after devices have been
1474  * added to a bus, this include calling it for a PHB that is just
1475  * being added
1476  */
1477 void pcibios_finish_adding_to_bus(struct pci_bus *bus)
1478 {
1479 	pr_debug("PCI: Finishing adding to hotplug bus %04x:%02x\n",
1480 		 pci_domain_nr(bus), bus->number);
1481 
1482 	/* Allocate bus and devices resources */
1483 	pcibios_allocate_bus_resources(bus);
1484 	pcibios_claim_one_bus(bus);
1485 
1486 	/* Fixup EEH */
1487 	eeh_add_device_tree_late(bus);
1488 
1489 	/* Add new devices to global lists.  Register in proc, sysfs. */
1490 	pci_bus_add_devices(bus);
1491 
1492 	/* sysfs files should only be added after devices are added */
1493 	eeh_add_sysfs_files(bus);
1494 }
1495 EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
1496 
1497 int pcibios_enable_device(struct pci_dev *dev, int mask)
1498 {
1499 	if (ppc_md.pcibios_enable_device_hook)
1500 		if (ppc_md.pcibios_enable_device_hook(dev))
1501 			return -EINVAL;
1502 
1503 	/* avoid pcie irq fix up impact on cardbus */
1504 	if (dev->hdr_type != PCI_HEADER_TYPE_CARDBUS)
1505 		pcibios_setup_device(dev);
1506 
1507 	return pci_enable_resources(dev, mask);
1508 }
1509 
1510 resource_size_t pcibios_io_space_offset(struct pci_controller *hose)
1511 {
1512 	return (unsigned long) hose->io_base_virt - _IO_BASE;
1513 }
1514 
1515 static void pcibios_setup_phb_resources(struct pci_controller *hose,
1516 					struct list_head *resources)
1517 {
1518 	struct resource *res;
1519 	int i;
1520 
1521 	/* Hookup PHB IO resource */
1522 	res = &hose->io_resource;
1523 
1524 	if (!res->flags) {
1525 		printk(KERN_WARNING "PCI: I/O resource not set for host"
1526 		       " bridge %s (domain %d)\n",
1527 		       hose->dn->full_name, hose->global_number);
1528 #ifdef CONFIG_PPC32
1529 		/* Workaround for lack of IO resource only on 32-bit */
1530 		res->start = (unsigned long)hose->io_base_virt - isa_io_base;
1531 		res->end = res->start + IO_SPACE_LIMIT;
1532 		res->flags = IORESOURCE_IO;
1533 #endif /* CONFIG_PPC32 */
1534 	}
1535 
1536 	pr_debug("PCI: PHB IO resource    = %016llx-%016llx [%lx]\n",
1537 		 (unsigned long long)res->start,
1538 		 (unsigned long long)res->end,
1539 		 (unsigned long)res->flags);
1540 	pci_add_resource_offset(resources, res, pcibios_io_space_offset(hose));
1541 
1542 	/* Hookup PHB Memory resources */
1543 	for (i = 0; i < 3; ++i) {
1544 		res = &hose->mem_resources[i];
1545 		if (!res->flags) {
1546 			if (i > 0)
1547 				continue;
1548 			printk(KERN_ERR "PCI: Memory resource 0 not set for "
1549 			       "host bridge %s (domain %d)\n",
1550 			       hose->dn->full_name, hose->global_number);
1551 #ifdef CONFIG_PPC32
1552 			/* Workaround for lack of MEM resource only on 32-bit */
1553 			res->start = hose->pci_mem_offset;
1554 			res->end = (resource_size_t)-1LL;
1555 			res->flags = IORESOURCE_MEM;
1556 #endif /* CONFIG_PPC32 */
1557 		}
1558 
1559 		pr_debug("PCI: PHB MEM resource %d = %016llx-%016llx [%lx]\n", i,
1560 			 (unsigned long long)res->start,
1561 			 (unsigned long long)res->end,
1562 			 (unsigned long)res->flags);
1563 		pci_add_resource_offset(resources, res, hose->pci_mem_offset);
1564 	}
1565 
1566 	pr_debug("PCI: PHB MEM offset     = %016llx\n",
1567 		 (unsigned long long)hose->pci_mem_offset);
1568 	pr_debug("PCI: PHB IO  offset     = %08lx\n",
1569 		 (unsigned long)hose->io_base_virt - _IO_BASE);
1570 
1571 }
1572 
1573 /*
1574  * Null PCI config access functions, for the case when we can't
1575  * find a hose.
1576  */
1577 #define NULL_PCI_OP(rw, size, type)					\
1578 static int								\
1579 null_##rw##_config_##size(struct pci_dev *dev, int offset, type val)	\
1580 {									\
1581 	return PCIBIOS_DEVICE_NOT_FOUND;    				\
1582 }
1583 
1584 static int
1585 null_read_config(struct pci_bus *bus, unsigned int devfn, int offset,
1586 		 int len, u32 *val)
1587 {
1588 	return PCIBIOS_DEVICE_NOT_FOUND;
1589 }
1590 
1591 static int
1592 null_write_config(struct pci_bus *bus, unsigned int devfn, int offset,
1593 		  int len, u32 val)
1594 {
1595 	return PCIBIOS_DEVICE_NOT_FOUND;
1596 }
1597 
1598 static struct pci_ops null_pci_ops =
1599 {
1600 	.read = null_read_config,
1601 	.write = null_write_config,
1602 };
1603 
1604 /*
1605  * These functions are used early on before PCI scanning is done
1606  * and all of the pci_dev and pci_bus structures have been created.
1607  */
1608 static struct pci_bus *
1609 fake_pci_bus(struct pci_controller *hose, int busnr)
1610 {
1611 	static struct pci_bus bus;
1612 
1613 	if (hose == 0) {
1614 		printk(KERN_ERR "Can't find hose for PCI bus %d!\n", busnr);
1615 	}
1616 	bus.number = busnr;
1617 	bus.sysdata = hose;
1618 	bus.ops = hose? hose->ops: &null_pci_ops;
1619 	return &bus;
1620 }
1621 
1622 #define EARLY_PCI_OP(rw, size, type)					\
1623 int early_##rw##_config_##size(struct pci_controller *hose, int bus,	\
1624 			       int devfn, int offset, type value)	\
1625 {									\
1626 	return pci_bus_##rw##_config_##size(fake_pci_bus(hose, bus),	\
1627 					    devfn, offset, value);	\
1628 }
1629 
1630 EARLY_PCI_OP(read, byte, u8 *)
1631 EARLY_PCI_OP(read, word, u16 *)
1632 EARLY_PCI_OP(read, dword, u32 *)
1633 EARLY_PCI_OP(write, byte, u8)
1634 EARLY_PCI_OP(write, word, u16)
1635 EARLY_PCI_OP(write, dword, u32)
1636 
1637 extern int pci_bus_find_capability (struct pci_bus *bus, unsigned int devfn, int cap);
1638 int early_find_capability(struct pci_controller *hose, int bus, int devfn,
1639 			  int cap)
1640 {
1641 	return pci_bus_find_capability(fake_pci_bus(hose, bus), devfn, cap);
1642 }
1643 
1644 struct device_node *pcibios_get_phb_of_node(struct pci_bus *bus)
1645 {
1646 	struct pci_controller *hose = bus->sysdata;
1647 
1648 	return of_node_get(hose->dn);
1649 }
1650 
1651 /**
1652  * pci_scan_phb - Given a pci_controller, setup and scan the PCI bus
1653  * @hose: Pointer to the PCI host controller instance structure
1654  */
1655 void pcibios_scan_phb(struct pci_controller *hose)
1656 {
1657 	LIST_HEAD(resources);
1658 	struct pci_bus *bus;
1659 	struct device_node *node = hose->dn;
1660 	int mode;
1661 
1662 	pr_debug("PCI: Scanning PHB %s\n", of_node_full_name(node));
1663 
1664 	/* Get some IO space for the new PHB */
1665 	pcibios_setup_phb_io_space(hose);
1666 
1667 	/* Wire up PHB bus resources */
1668 	pcibios_setup_phb_resources(hose, &resources);
1669 
1670 	hose->busn.start = hose->first_busno;
1671 	hose->busn.end	 = hose->last_busno;
1672 	hose->busn.flags = IORESOURCE_BUS;
1673 	pci_add_resource(&resources, &hose->busn);
1674 
1675 	/* Create an empty bus for the toplevel */
1676 	bus = pci_create_root_bus(hose->parent, hose->first_busno,
1677 				  hose->ops, hose, &resources);
1678 	if (bus == NULL) {
1679 		pr_err("Failed to create bus for PCI domain %04x\n",
1680 			hose->global_number);
1681 		pci_free_resource_list(&resources);
1682 		return;
1683 	}
1684 	hose->bus = bus;
1685 
1686 	/* Get probe mode and perform scan */
1687 	mode = PCI_PROBE_NORMAL;
1688 	if (node && ppc_md.pci_probe_mode)
1689 		mode = ppc_md.pci_probe_mode(bus);
1690 	pr_debug("    probe mode: %d\n", mode);
1691 	if (mode == PCI_PROBE_DEVTREE)
1692 		of_scan_bus(node, bus);
1693 
1694 	if (mode == PCI_PROBE_NORMAL) {
1695 		pci_bus_update_busn_res_end(bus, 255);
1696 		hose->last_busno = pci_scan_child_bus(bus);
1697 		pci_bus_update_busn_res_end(bus, hose->last_busno);
1698 	}
1699 
1700 	/* Platform gets a chance to do some global fixups before
1701 	 * we proceed to resource allocation
1702 	 */
1703 	if (ppc_md.pcibios_fixup_phb)
1704 		ppc_md.pcibios_fixup_phb(hose);
1705 
1706 	/* Configure PCI Express settings */
1707 	if (bus && !pci_has_flag(PCI_PROBE_ONLY)) {
1708 		struct pci_bus *child;
1709 		list_for_each_entry(child, &bus->children, node) {
1710 			struct pci_dev *self = child->self;
1711 			if (!self)
1712 				continue;
1713 			pcie_bus_configure_settings(child, self->pcie_mpss);
1714 		}
1715 	}
1716 }
1717 
1718 static void fixup_hide_host_resource_fsl(struct pci_dev *dev)
1719 {
1720 	int i, class = dev->class >> 8;
1721 	/* When configured as agent, programing interface = 1 */
1722 	int prog_if = dev->class & 0xf;
1723 
1724 	if ((class == PCI_CLASS_PROCESSOR_POWERPC ||
1725 	     class == PCI_CLASS_BRIDGE_OTHER) &&
1726 		(dev->hdr_type == PCI_HEADER_TYPE_NORMAL) &&
1727 		(prog_if == 0) &&
1728 		(dev->bus->parent == NULL)) {
1729 		for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
1730 			dev->resource[i].start = 0;
1731 			dev->resource[i].end = 0;
1732 			dev->resource[i].flags = 0;
1733 		}
1734 	}
1735 }
1736 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_MOTOROLA, PCI_ANY_ID, fixup_hide_host_resource_fsl);
1737 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_FREESCALE, PCI_ANY_ID, fixup_hide_host_resource_fsl);
1738 
1739 static void fixup_vga(struct pci_dev *pdev)
1740 {
1741 	u16 cmd;
1742 
1743 	pci_read_config_word(pdev, PCI_COMMAND, &cmd);
1744 	if ((cmd & (PCI_COMMAND_IO | PCI_COMMAND_MEMORY)) || !vga_default_device())
1745 		vga_set_default_device(pdev);
1746 
1747 }
1748 DECLARE_PCI_FIXUP_CLASS_FINAL(PCI_ANY_ID, PCI_ANY_ID,
1749 			      PCI_CLASS_DISPLAY_VGA, 8, fixup_vga);
1750