xref: /linux/arch/powerpc/kernel/pci-common.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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/eeh.h>
42 
43 static DEFINE_SPINLOCK(hose_spinlock);
44 LIST_HEAD(hose_list);
45 
46 /* XXX kill that some day ... */
47 static int global_phb_number;		/* Global phb counter */
48 
49 /* ISA Memory physical address */
50 resource_size_t isa_mem_base;
51 
52 
53 static struct dma_map_ops *pci_dma_ops = &dma_direct_ops;
54 
55 void set_pci_dma_ops(struct dma_map_ops *dma_ops)
56 {
57 	pci_dma_ops = dma_ops;
58 }
59 
60 struct dma_map_ops *get_pci_dma_ops(void)
61 {
62 	return pci_dma_ops;
63 }
64 EXPORT_SYMBOL(get_pci_dma_ops);
65 
66 struct pci_controller *pcibios_alloc_controller(struct device_node *dev)
67 {
68 	struct pci_controller *phb;
69 
70 	phb = zalloc_maybe_bootmem(sizeof(struct pci_controller), GFP_KERNEL);
71 	if (phb == NULL)
72 		return NULL;
73 	spin_lock(&hose_spinlock);
74 	phb->global_number = global_phb_number++;
75 	list_add_tail(&phb->list_node, &hose_list);
76 	spin_unlock(&hose_spinlock);
77 	phb->dn = dev;
78 	phb->is_dynamic = mem_init_done;
79 #ifdef CONFIG_PPC64
80 	if (dev) {
81 		int nid = of_node_to_nid(dev);
82 
83 		if (nid < 0 || !node_online(nid))
84 			nid = -1;
85 
86 		PHB_SET_NODE(phb, nid);
87 	}
88 #endif
89 	return phb;
90 }
91 
92 void pcibios_free_controller(struct pci_controller *phb)
93 {
94 	spin_lock(&hose_spinlock);
95 	list_del(&phb->list_node);
96 	spin_unlock(&hose_spinlock);
97 
98 	if (phb->is_dynamic)
99 		kfree(phb);
100 }
101 
102 /*
103  * The function is used to return the minimal alignment
104  * for memory or I/O windows of the associated P2P bridge.
105  * By default, 4KiB alignment for I/O windows and 1MiB for
106  * memory windows.
107  */
108 resource_size_t pcibios_window_alignment(struct pci_bus *bus,
109 					 unsigned long type)
110 {
111 	if (ppc_md.pcibios_window_alignment)
112 		return ppc_md.pcibios_window_alignment(bus, type);
113 
114 	/*
115 	 * PCI core will figure out the default
116 	 * alignment: 4KiB for I/O and 1MiB for
117 	 * memory window.
118 	 */
119 	return 1;
120 }
121 
122 static resource_size_t pcibios_io_size(const struct pci_controller *hose)
123 {
124 #ifdef CONFIG_PPC64
125 	return hose->pci_io_size;
126 #else
127 	return resource_size(&hose->io_resource);
128 #endif
129 }
130 
131 int pcibios_vaddr_is_ioport(void __iomem *address)
132 {
133 	int ret = 0;
134 	struct pci_controller *hose;
135 	resource_size_t size;
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_virt &&
141 		    address < (hose->io_base_virt + size)) {
142 			ret = 1;
143 			break;
144 		}
145 	}
146 	spin_unlock(&hose_spinlock);
147 	return ret;
148 }
149 
150 unsigned long pci_address_to_pio(phys_addr_t address)
151 {
152 	struct pci_controller *hose;
153 	resource_size_t size;
154 	unsigned long ret = ~0;
155 
156 	spin_lock(&hose_spinlock);
157 	list_for_each_entry(hose, &hose_list, list_node) {
158 		size = pcibios_io_size(hose);
159 		if (address >= hose->io_base_phys &&
160 		    address < (hose->io_base_phys + size)) {
161 			unsigned long base =
162 				(unsigned long)hose->io_base_virt - _IO_BASE;
163 			ret = base + (address - hose->io_base_phys);
164 			break;
165 		}
166 	}
167 	spin_unlock(&hose_spinlock);
168 
169 	return ret;
170 }
171 EXPORT_SYMBOL_GPL(pci_address_to_pio);
172 
173 /*
174  * Return the domain number for this bus.
175  */
176 int pci_domain_nr(struct pci_bus *bus)
177 {
178 	struct pci_controller *hose = pci_bus_to_host(bus);
179 
180 	return hose->global_number;
181 }
182 EXPORT_SYMBOL(pci_domain_nr);
183 
184 /* This routine is meant to be used early during boot, when the
185  * PCI bus numbers have not yet been assigned, and you need to
186  * issue PCI config cycles to an OF device.
187  * It could also be used to "fix" RTAS config cycles if you want
188  * to set pci_assign_all_buses to 1 and still use RTAS for PCI
189  * config cycles.
190  */
191 struct pci_controller* pci_find_hose_for_OF_device(struct device_node* node)
192 {
193 	while(node) {
194 		struct pci_controller *hose, *tmp;
195 		list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
196 			if (hose->dn == node)
197 				return hose;
198 		node = node->parent;
199 	}
200 	return NULL;
201 }
202 
203 static ssize_t pci_show_devspec(struct device *dev,
204 		struct device_attribute *attr, char *buf)
205 {
206 	struct pci_dev *pdev;
207 	struct device_node *np;
208 
209 	pdev = to_pci_dev (dev);
210 	np = pci_device_to_OF_node(pdev);
211 	if (np == NULL || np->full_name == NULL)
212 		return 0;
213 	return sprintf(buf, "%s", np->full_name);
214 }
215 static DEVICE_ATTR(devspec, S_IRUGO, pci_show_devspec, NULL);
216 
217 /* Add sysfs properties */
218 int pcibios_add_platform_entries(struct pci_dev *pdev)
219 {
220 	return device_create_file(&pdev->dev, &dev_attr_devspec);
221 }
222 
223 /*
224  * Reads the interrupt pin to determine if interrupt is use by card.
225  * If the interrupt is used, then gets the interrupt line from the
226  * openfirmware and sets it in the pci_dev and pci_config line.
227  */
228 static int pci_read_irq_line(struct pci_dev *pci_dev)
229 {
230 	struct of_irq oirq;
231 	unsigned int virq;
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 			 of_node_full_name(oirq.controller));
267 
268 		virq = irq_create_of_mapping(oirq.controller, oirq.specifier,
269 					     oirq.size);
270 	}
271 	if(virq == NO_IRQ) {
272 		pr_debug(" Failed to map !\n");
273 		return -1;
274 	}
275 
276 	pr_debug(" Mapped to linux irq %d\n", virq);
277 
278 	pci_dev->irq = virq;
279 
280 	return 0;
281 }
282 
283 /*
284  * Platform support for /proc/bus/pci/X/Y mmap()s,
285  * modelled on the sparc64 implementation by Dave Miller.
286  *  -- paulus.
287  */
288 
289 /*
290  * Adjust vm_pgoff of VMA such that it is the physical page offset
291  * corresponding to the 32-bit pci bus offset for DEV requested by the user.
292  *
293  * Basically, the user finds the base address for his device which he wishes
294  * to mmap.  They read the 32-bit value from the config space base register,
295  * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
296  * offset parameter of mmap on /proc/bus/pci/XXX for that device.
297  *
298  * Returns negative error code on failure, zero on success.
299  */
300 static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
301 					       resource_size_t *offset,
302 					       enum pci_mmap_state mmap_state)
303 {
304 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
305 	unsigned long io_offset = 0;
306 	int i, res_bit;
307 
308 	if (hose == 0)
309 		return NULL;		/* should never happen */
310 
311 	/* If memory, add on the PCI bridge address offset */
312 	if (mmap_state == pci_mmap_mem) {
313 #if 0 /* See comment in pci_resource_to_user() for why this is disabled */
314 		*offset += hose->pci_mem_offset;
315 #endif
316 		res_bit = IORESOURCE_MEM;
317 	} else {
318 		io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
319 		*offset += io_offset;
320 		res_bit = IORESOURCE_IO;
321 	}
322 
323 	/*
324 	 * Check that the offset requested corresponds to one of the
325 	 * resources of the device.
326 	 */
327 	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
328 		struct resource *rp = &dev->resource[i];
329 		int flags = rp->flags;
330 
331 		/* treat ROM as memory (should be already) */
332 		if (i == PCI_ROM_RESOURCE)
333 			flags |= IORESOURCE_MEM;
334 
335 		/* Active and same type? */
336 		if ((flags & res_bit) == 0)
337 			continue;
338 
339 		/* In the range of this resource? */
340 		if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
341 			continue;
342 
343 		/* found it! construct the final physical address */
344 		if (mmap_state == pci_mmap_io)
345 			*offset += hose->io_base_phys - io_offset;
346 		return rp;
347 	}
348 
349 	return NULL;
350 }
351 
352 /*
353  * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
354  * device mapping.
355  */
356 static pgprot_t __pci_mmap_set_pgprot(struct pci_dev *dev, struct resource *rp,
357 				      pgprot_t protection,
358 				      enum pci_mmap_state mmap_state,
359 				      int write_combine)
360 {
361 	unsigned long prot = pgprot_val(protection);
362 
363 	/* Write combine is always 0 on non-memory space mappings. On
364 	 * memory space, if the user didn't pass 1, we check for a
365 	 * "prefetchable" resource. This is a bit hackish, but we use
366 	 * this to workaround the inability of /sysfs to provide a write
367 	 * combine bit
368 	 */
369 	if (mmap_state != pci_mmap_mem)
370 		write_combine = 0;
371 	else if (write_combine == 0) {
372 		if (rp->flags & IORESOURCE_PREFETCH)
373 			write_combine = 1;
374 	}
375 
376 	/* XXX would be nice to have a way to ask for write-through */
377 	if (write_combine)
378 		return pgprot_noncached_wc(prot);
379 	else
380 		return pgprot_noncached(prot);
381 }
382 
383 /*
384  * This one is used by /dev/mem and fbdev who have no clue about the
385  * PCI device, it tries to find the PCI device first and calls the
386  * above routine
387  */
388 pgprot_t pci_phys_mem_access_prot(struct file *file,
389 				  unsigned long pfn,
390 				  unsigned long size,
391 				  pgprot_t prot)
392 {
393 	struct pci_dev *pdev = NULL;
394 	struct resource *found = NULL;
395 	resource_size_t offset = ((resource_size_t)pfn) << PAGE_SHIFT;
396 	int i;
397 
398 	if (page_is_ram(pfn))
399 		return prot;
400 
401 	prot = pgprot_noncached(prot);
402 	for_each_pci_dev(pdev) {
403 		for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
404 			struct resource *rp = &pdev->resource[i];
405 			int flags = rp->flags;
406 
407 			/* Active and same type? */
408 			if ((flags & IORESOURCE_MEM) == 0)
409 				continue;
410 			/* In the range of this resource? */
411 			if (offset < (rp->start & PAGE_MASK) ||
412 			    offset > rp->end)
413 				continue;
414 			found = rp;
415 			break;
416 		}
417 		if (found)
418 			break;
419 	}
420 	if (found) {
421 		if (found->flags & IORESOURCE_PREFETCH)
422 			prot = pgprot_noncached_wc(prot);
423 		pci_dev_put(pdev);
424 	}
425 
426 	pr_debug("PCI: Non-PCI map for %llx, prot: %lx\n",
427 		 (unsigned long long)offset, pgprot_val(prot));
428 
429 	return prot;
430 }
431 
432 
433 /*
434  * Perform the actual remap of the pages for a PCI device mapping, as
435  * appropriate for this architecture.  The region in the process to map
436  * is described by vm_start and vm_end members of VMA, the base physical
437  * address is found in vm_pgoff.
438  * The pci device structure is provided so that architectures may make mapping
439  * decisions on a per-device or per-bus basis.
440  *
441  * Returns a negative error code on failure, zero on success.
442  */
443 int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
444 			enum pci_mmap_state mmap_state, int write_combine)
445 {
446 	resource_size_t offset =
447 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
448 	struct resource *rp;
449 	int ret;
450 
451 	rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
452 	if (rp == NULL)
453 		return -EINVAL;
454 
455 	vma->vm_pgoff = offset >> PAGE_SHIFT;
456 	vma->vm_page_prot = __pci_mmap_set_pgprot(dev, rp,
457 						  vma->vm_page_prot,
458 						  mmap_state, write_combine);
459 
460 	ret = remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
461 			       vma->vm_end - vma->vm_start, vma->vm_page_prot);
462 
463 	return ret;
464 }
465 
466 /* This provides legacy IO read access on a bus */
467 int pci_legacy_read(struct pci_bus *bus, loff_t port, u32 *val, size_t size)
468 {
469 	unsigned long offset;
470 	struct pci_controller *hose = pci_bus_to_host(bus);
471 	struct resource *rp = &hose->io_resource;
472 	void __iomem *addr;
473 
474 	/* Check if port can be supported by that bus. We only check
475 	 * the ranges of the PHB though, not the bus itself as the rules
476 	 * for forwarding legacy cycles down bridges are not our problem
477 	 * here. So if the host bridge supports it, we do it.
478 	 */
479 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
480 	offset += port;
481 
482 	if (!(rp->flags & IORESOURCE_IO))
483 		return -ENXIO;
484 	if (offset < rp->start || (offset + size) > rp->end)
485 		return -ENXIO;
486 	addr = hose->io_base_virt + port;
487 
488 	switch(size) {
489 	case 1:
490 		*((u8 *)val) = in_8(addr);
491 		return 1;
492 	case 2:
493 		if (port & 1)
494 			return -EINVAL;
495 		*((u16 *)val) = in_le16(addr);
496 		return 2;
497 	case 4:
498 		if (port & 3)
499 			return -EINVAL;
500 		*((u32 *)val) = in_le32(addr);
501 		return 4;
502 	}
503 	return -EINVAL;
504 }
505 
506 /* This provides legacy IO write access on a bus */
507 int pci_legacy_write(struct pci_bus *bus, loff_t port, u32 val, size_t size)
508 {
509 	unsigned long offset;
510 	struct pci_controller *hose = pci_bus_to_host(bus);
511 	struct resource *rp = &hose->io_resource;
512 	void __iomem *addr;
513 
514 	/* Check if port can be supported by that bus. We only check
515 	 * the ranges of the PHB though, not the bus itself as the rules
516 	 * for forwarding legacy cycles down bridges are not our problem
517 	 * here. So if the host bridge supports it, we do it.
518 	 */
519 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
520 	offset += port;
521 
522 	if (!(rp->flags & IORESOURCE_IO))
523 		return -ENXIO;
524 	if (offset < rp->start || (offset + size) > rp->end)
525 		return -ENXIO;
526 	addr = hose->io_base_virt + port;
527 
528 	/* WARNING: The generic code is idiotic. It gets passed a pointer
529 	 * to what can be a 1, 2 or 4 byte quantity and always reads that
530 	 * as a u32, which means that we have to correct the location of
531 	 * the data read within those 32 bits for size 1 and 2
532 	 */
533 	switch(size) {
534 	case 1:
535 		out_8(addr, val >> 24);
536 		return 1;
537 	case 2:
538 		if (port & 1)
539 			return -EINVAL;
540 		out_le16(addr, val >> 16);
541 		return 2;
542 	case 4:
543 		if (port & 3)
544 			return -EINVAL;
545 		out_le32(addr, val);
546 		return 4;
547 	}
548 	return -EINVAL;
549 }
550 
551 /* This provides legacy IO or memory mmap access on a bus */
552 int pci_mmap_legacy_page_range(struct pci_bus *bus,
553 			       struct vm_area_struct *vma,
554 			       enum pci_mmap_state mmap_state)
555 {
556 	struct pci_controller *hose = pci_bus_to_host(bus);
557 	resource_size_t offset =
558 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
559 	resource_size_t size = vma->vm_end - vma->vm_start;
560 	struct resource *rp;
561 
562 	pr_debug("pci_mmap_legacy_page_range(%04x:%02x, %s @%llx..%llx)\n",
563 		 pci_domain_nr(bus), bus->number,
564 		 mmap_state == pci_mmap_mem ? "MEM" : "IO",
565 		 (unsigned long long)offset,
566 		 (unsigned long long)(offset + size - 1));
567 
568 	if (mmap_state == pci_mmap_mem) {
569 		/* Hack alert !
570 		 *
571 		 * Because X is lame and can fail starting if it gets an error trying
572 		 * to mmap legacy_mem (instead of just moving on without legacy memory
573 		 * access) we fake it here by giving it anonymous memory, effectively
574 		 * behaving just like /dev/zero
575 		 */
576 		if ((offset + size) > hose->isa_mem_size) {
577 			printk(KERN_DEBUG
578 			       "Process %s (pid:%d) mapped non-existing PCI legacy memory for 0%04x:%02x\n",
579 			       current->comm, current->pid, pci_domain_nr(bus), bus->number);
580 			if (vma->vm_flags & VM_SHARED)
581 				return shmem_zero_setup(vma);
582 			return 0;
583 		}
584 		offset += hose->isa_mem_phys;
585 	} else {
586 		unsigned long io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
587 		unsigned long roffset = offset + io_offset;
588 		rp = &hose->io_resource;
589 		if (!(rp->flags & IORESOURCE_IO))
590 			return -ENXIO;
591 		if (roffset < rp->start || (roffset + size) > rp->end)
592 			return -ENXIO;
593 		offset += hose->io_base_phys;
594 	}
595 	pr_debug(" -> mapping phys %llx\n", (unsigned long long)offset);
596 
597 	vma->vm_pgoff = offset >> PAGE_SHIFT;
598 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
599 	return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
600 			       vma->vm_end - vma->vm_start,
601 			       vma->vm_page_prot);
602 }
603 
604 void pci_resource_to_user(const struct pci_dev *dev, int bar,
605 			  const struct resource *rsrc,
606 			  resource_size_t *start, resource_size_t *end)
607 {
608 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
609 	resource_size_t offset = 0;
610 
611 	if (hose == NULL)
612 		return;
613 
614 	if (rsrc->flags & IORESOURCE_IO)
615 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
616 
617 	/* We pass a fully fixed up address to userland for MMIO instead of
618 	 * a BAR value because X is lame and expects to be able to use that
619 	 * to pass to /dev/mem !
620 	 *
621 	 * That means that we'll have potentially 64 bits values where some
622 	 * userland apps only expect 32 (like X itself since it thinks only
623 	 * Sparc has 64 bits MMIO) but if we don't do that, we break it on
624 	 * 32 bits CHRPs :-(
625 	 *
626 	 * Hopefully, the sysfs insterface is immune to that gunk. Once X
627 	 * has been fixed (and the fix spread enough), we can re-enable the
628 	 * 2 lines below and pass down a BAR value to userland. In that case
629 	 * we'll also have to re-enable the matching code in
630 	 * __pci_mmap_make_offset().
631 	 *
632 	 * BenH.
633 	 */
634 #if 0
635 	else if (rsrc->flags & IORESOURCE_MEM)
636 		offset = hose->pci_mem_offset;
637 #endif
638 
639 	*start = rsrc->start - offset;
640 	*end = rsrc->end - offset;
641 }
642 
643 /**
644  * pci_process_bridge_OF_ranges - Parse PCI bridge resources from device tree
645  * @hose: newly allocated pci_controller to be setup
646  * @dev: device node of the host bridge
647  * @primary: set if primary bus (32 bits only, soon to be deprecated)
648  *
649  * This function will parse the "ranges" property of a PCI host bridge device
650  * node and setup the resource mapping of a pci controller based on its
651  * content.
652  *
653  * Life would be boring if it wasn't for a few issues that we have to deal
654  * with here:
655  *
656  *   - We can only cope with one IO space range and up to 3 Memory space
657  *     ranges. However, some machines (thanks Apple !) tend to split their
658  *     space into lots of small contiguous ranges. So we have to coalesce.
659  *
660  *   - We can only cope with all memory ranges having the same offset
661  *     between CPU addresses and PCI addresses. Unfortunately, some bridges
662  *     are setup for a large 1:1 mapping along with a small "window" which
663  *     maps PCI address 0 to some arbitrary high address of the CPU space in
664  *     order to give access to the ISA memory hole.
665  *     The way out of here that I've chosen for now is to always set the
666  *     offset based on the first resource found, then override it if we
667  *     have a different offset and the previous was set by an ISA hole.
668  *
669  *   - Some busses have IO space not starting at 0, which causes trouble with
670  *     the way we do our IO resource renumbering. The code somewhat deals with
671  *     it for 64 bits but I would expect problems on 32 bits.
672  *
673  *   - Some 32 bits platforms such as 4xx can have physical space larger than
674  *     32 bits so we need to use 64 bits values for the parsing
675  */
676 void __devinit pci_process_bridge_OF_ranges(struct pci_controller *hose,
677 					    struct device_node *dev,
678 					    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 __devinit 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 __devinit 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 __devinit 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 __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 		/* The resource might be figured out when doing
1253 		 * reassignment based on the resources required
1254 		 * by the downstream PCI devices. Here we set
1255 		 * the size of the resource to be 0 in order to
1256 		 * save more space.
1257 		 */
1258 		res->start = 0;
1259 		res->end = -1;
1260 		res->flags = 0;
1261 	}
1262 
1263 	list_for_each_entry(b, &bus->children, node)
1264 		pcibios_allocate_bus_resources(b);
1265 }
1266 
1267 static inline void __devinit alloc_resource(struct pci_dev *dev, int idx)
1268 {
1269 	struct resource *pr, *r = &dev->resource[idx];
1270 
1271 	pr_debug("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1272 		 pci_name(dev), idx,
1273 		 (unsigned long long)r->start,
1274 		 (unsigned long long)r->end,
1275 		 (unsigned int)r->flags);
1276 
1277 	pr = pci_find_parent_resource(dev, r);
1278 	if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1279 	    request_resource(pr, r) < 0) {
1280 		printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1281 		       " of device %s, will remap\n", idx, pci_name(dev));
1282 		if (pr)
1283 			pr_debug("PCI:  parent is %p: %016llx-%016llx [%x]\n",
1284 				 pr,
1285 				 (unsigned long long)pr->start,
1286 				 (unsigned long long)pr->end,
1287 				 (unsigned int)pr->flags);
1288 		/* We'll assign a new address later */
1289 		r->flags |= IORESOURCE_UNSET;
1290 		r->end -= r->start;
1291 		r->start = 0;
1292 	}
1293 }
1294 
1295 static void __init pcibios_allocate_resources(int pass)
1296 {
1297 	struct pci_dev *dev = NULL;
1298 	int idx, disabled;
1299 	u16 command;
1300 	struct resource *r;
1301 
1302 	for_each_pci_dev(dev) {
1303 		pci_read_config_word(dev, PCI_COMMAND, &command);
1304 		for (idx = 0; idx <= PCI_ROM_RESOURCE; idx++) {
1305 			r = &dev->resource[idx];
1306 			if (r->parent)		/* Already allocated */
1307 				continue;
1308 			if (!r->flags || (r->flags & IORESOURCE_UNSET))
1309 				continue;	/* Not assigned at all */
1310 			/* We only allocate ROMs on pass 1 just in case they
1311 			 * have been screwed up by firmware
1312 			 */
1313 			if (idx == PCI_ROM_RESOURCE )
1314 				disabled = 1;
1315 			if (r->flags & IORESOURCE_IO)
1316 				disabled = !(command & PCI_COMMAND_IO);
1317 			else
1318 				disabled = !(command & PCI_COMMAND_MEMORY);
1319 			if (pass == disabled)
1320 				alloc_resource(dev, idx);
1321 		}
1322 		if (pass)
1323 			continue;
1324 		r = &dev->resource[PCI_ROM_RESOURCE];
1325 		if (r->flags) {
1326 			/* Turn the ROM off, leave the resource region,
1327 			 * but keep it unregistered.
1328 			 */
1329 			u32 reg;
1330 			pci_read_config_dword(dev, dev->rom_base_reg, &reg);
1331 			if (reg & PCI_ROM_ADDRESS_ENABLE) {
1332 				pr_debug("PCI: Switching off ROM of %s\n",
1333 					 pci_name(dev));
1334 				r->flags &= ~IORESOURCE_ROM_ENABLE;
1335 				pci_write_config_dword(dev, dev->rom_base_reg,
1336 						       reg & ~PCI_ROM_ADDRESS_ENABLE);
1337 			}
1338 		}
1339 	}
1340 }
1341 
1342 static void __init pcibios_reserve_legacy_regions(struct pci_bus *bus)
1343 {
1344 	struct pci_controller *hose = pci_bus_to_host(bus);
1345 	resource_size_t	offset;
1346 	struct resource *res, *pres;
1347 	int i;
1348 
1349 	pr_debug("Reserving legacy ranges for domain %04x\n", pci_domain_nr(bus));
1350 
1351 	/* Check for IO */
1352 	if (!(hose->io_resource.flags & IORESOURCE_IO))
1353 		goto no_io;
1354 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1355 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1356 	BUG_ON(res == NULL);
1357 	res->name = "Legacy IO";
1358 	res->flags = IORESOURCE_IO;
1359 	res->start = offset;
1360 	res->end = (offset + 0xfff) & 0xfffffffful;
1361 	pr_debug("Candidate legacy IO: %pR\n", res);
1362 	if (request_resource(&hose->io_resource, res)) {
1363 		printk(KERN_DEBUG
1364 		       "PCI %04x:%02x Cannot reserve Legacy IO %pR\n",
1365 		       pci_domain_nr(bus), bus->number, res);
1366 		kfree(res);
1367 	}
1368 
1369  no_io:
1370 	/* Check for memory */
1371 	offset = hose->pci_mem_offset;
1372 	pr_debug("hose mem offset: %016llx\n", (unsigned long long)offset);
1373 	for (i = 0; i < 3; i++) {
1374 		pres = &hose->mem_resources[i];
1375 		if (!(pres->flags & IORESOURCE_MEM))
1376 			continue;
1377 		pr_debug("hose mem res: %pR\n", pres);
1378 		if ((pres->start - offset) <= 0xa0000 &&
1379 		    (pres->end - offset) >= 0xbffff)
1380 			break;
1381 	}
1382 	if (i >= 3)
1383 		return;
1384 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1385 	BUG_ON(res == NULL);
1386 	res->name = "Legacy VGA memory";
1387 	res->flags = IORESOURCE_MEM;
1388 	res->start = 0xa0000 + offset;
1389 	res->end = 0xbffff + offset;
1390 	pr_debug("Candidate VGA memory: %pR\n", res);
1391 	if (request_resource(pres, res)) {
1392 		printk(KERN_DEBUG
1393 		       "PCI %04x:%02x Cannot reserve VGA memory %pR\n",
1394 		       pci_domain_nr(bus), bus->number, res);
1395 		kfree(res);
1396 	}
1397 }
1398 
1399 void __init pcibios_resource_survey(void)
1400 {
1401 	struct pci_bus *b;
1402 
1403 	/* Allocate and assign resources */
1404 	list_for_each_entry(b, &pci_root_buses, node)
1405 		pcibios_allocate_bus_resources(b);
1406 	pcibios_allocate_resources(0);
1407 	pcibios_allocate_resources(1);
1408 
1409 	/* Before we start assigning unassigned resource, we try to reserve
1410 	 * the low IO area and the VGA memory area if they intersect the
1411 	 * bus available resources to avoid allocating things on top of them
1412 	 */
1413 	if (!pci_has_flag(PCI_PROBE_ONLY)) {
1414 		list_for_each_entry(b, &pci_root_buses, node)
1415 			pcibios_reserve_legacy_regions(b);
1416 	}
1417 
1418 	/* Now, if the platform didn't decide to blindly trust the firmware,
1419 	 * we proceed to assigning things that were left unassigned
1420 	 */
1421 	if (!pci_has_flag(PCI_PROBE_ONLY)) {
1422 		pr_debug("PCI: Assigning unassigned resources...\n");
1423 		pci_assign_unassigned_resources();
1424 	}
1425 
1426 	/* Call machine dependent fixup */
1427 	if (ppc_md.pcibios_fixup)
1428 		ppc_md.pcibios_fixup();
1429 }
1430 
1431 /* This is used by the PCI hotplug driver to allocate resource
1432  * of newly plugged busses. We can try to consolidate with the
1433  * rest of the code later, for now, keep it as-is as our main
1434  * resource allocation function doesn't deal with sub-trees yet.
1435  */
1436 void pcibios_claim_one_bus(struct pci_bus *bus)
1437 {
1438 	struct pci_dev *dev;
1439 	struct pci_bus *child_bus;
1440 
1441 	list_for_each_entry(dev, &bus->devices, bus_list) {
1442 		int i;
1443 
1444 		for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1445 			struct resource *r = &dev->resource[i];
1446 
1447 			if (r->parent || !r->start || !r->flags)
1448 				continue;
1449 
1450 			pr_debug("PCI: Claiming %s: "
1451 				 "Resource %d: %016llx..%016llx [%x]\n",
1452 				 pci_name(dev), i,
1453 				 (unsigned long long)r->start,
1454 				 (unsigned long long)r->end,
1455 				 (unsigned int)r->flags);
1456 
1457 			pci_claim_resource(dev, i);
1458 		}
1459 	}
1460 
1461 	list_for_each_entry(child_bus, &bus->children, node)
1462 		pcibios_claim_one_bus(child_bus);
1463 }
1464 
1465 
1466 /* pcibios_finish_adding_to_bus
1467  *
1468  * This is to be called by the hotplug code after devices have been
1469  * added to a bus, this include calling it for a PHB that is just
1470  * being added
1471  */
1472 void pcibios_finish_adding_to_bus(struct pci_bus *bus)
1473 {
1474 	pr_debug("PCI: Finishing adding to hotplug bus %04x:%02x\n",
1475 		 pci_domain_nr(bus), bus->number);
1476 
1477 	/* Allocate bus and devices resources */
1478 	pcibios_allocate_bus_resources(bus);
1479 	pcibios_claim_one_bus(bus);
1480 
1481 	/* Add new devices to global lists.  Register in proc, sysfs. */
1482 	pci_bus_add_devices(bus);
1483 
1484 	/* Fixup EEH */
1485 	eeh_add_device_tree_late(bus);
1486 }
1487 EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
1488 
1489 int pcibios_enable_device(struct pci_dev *dev, int mask)
1490 {
1491 	if (ppc_md.pcibios_enable_device_hook)
1492 		if (ppc_md.pcibios_enable_device_hook(dev))
1493 			return -EINVAL;
1494 
1495 	return pci_enable_resources(dev, mask);
1496 }
1497 
1498 resource_size_t pcibios_io_space_offset(struct pci_controller *hose)
1499 {
1500 	return (unsigned long) hose->io_base_virt - _IO_BASE;
1501 }
1502 
1503 static void __devinit pcibios_setup_phb_resources(struct pci_controller *hose, struct list_head *resources)
1504 {
1505 	struct resource *res;
1506 	int i;
1507 
1508 	/* Hookup PHB IO resource */
1509 	res = &hose->io_resource;
1510 
1511 	if (!res->flags) {
1512 		printk(KERN_WARNING "PCI: I/O resource not set for host"
1513 		       " bridge %s (domain %d)\n",
1514 		       hose->dn->full_name, hose->global_number);
1515 #ifdef CONFIG_PPC32
1516 		/* Workaround for lack of IO resource only on 32-bit */
1517 		res->start = (unsigned long)hose->io_base_virt - isa_io_base;
1518 		res->end = res->start + IO_SPACE_LIMIT;
1519 		res->flags = IORESOURCE_IO;
1520 #endif /* CONFIG_PPC32 */
1521 	}
1522 
1523 	pr_debug("PCI: PHB IO resource    = %016llx-%016llx [%lx]\n",
1524 		 (unsigned long long)res->start,
1525 		 (unsigned long long)res->end,
1526 		 (unsigned long)res->flags);
1527 	pci_add_resource_offset(resources, res, pcibios_io_space_offset(hose));
1528 
1529 	/* Hookup PHB Memory resources */
1530 	for (i = 0; i < 3; ++i) {
1531 		res = &hose->mem_resources[i];
1532 		if (!res->flags) {
1533 			if (i > 0)
1534 				continue;
1535 			printk(KERN_ERR "PCI: Memory resource 0 not set for "
1536 			       "host bridge %s (domain %d)\n",
1537 			       hose->dn->full_name, hose->global_number);
1538 #ifdef CONFIG_PPC32
1539 			/* Workaround for lack of MEM resource only on 32-bit */
1540 			res->start = hose->pci_mem_offset;
1541 			res->end = (resource_size_t)-1LL;
1542 			res->flags = IORESOURCE_MEM;
1543 #endif /* CONFIG_PPC32 */
1544 		}
1545 
1546 		pr_debug("PCI: PHB MEM resource %d = %016llx-%016llx [%lx]\n", i,
1547 			 (unsigned long long)res->start,
1548 			 (unsigned long long)res->end,
1549 			 (unsigned long)res->flags);
1550 		pci_add_resource_offset(resources, res, hose->pci_mem_offset);
1551 	}
1552 
1553 	pr_debug("PCI: PHB MEM offset     = %016llx\n",
1554 		 (unsigned long long)hose->pci_mem_offset);
1555 	pr_debug("PCI: PHB IO  offset     = %08lx\n",
1556 		 (unsigned long)hose->io_base_virt - _IO_BASE);
1557 
1558 }
1559 
1560 /*
1561  * Null PCI config access functions, for the case when we can't
1562  * find a hose.
1563  */
1564 #define NULL_PCI_OP(rw, size, type)					\
1565 static int								\
1566 null_##rw##_config_##size(struct pci_dev *dev, int offset, type val)	\
1567 {									\
1568 	return PCIBIOS_DEVICE_NOT_FOUND;    				\
1569 }
1570 
1571 static int
1572 null_read_config(struct pci_bus *bus, unsigned int devfn, int offset,
1573 		 int len, u32 *val)
1574 {
1575 	return PCIBIOS_DEVICE_NOT_FOUND;
1576 }
1577 
1578 static int
1579 null_write_config(struct pci_bus *bus, unsigned int devfn, int offset,
1580 		  int len, u32 val)
1581 {
1582 	return PCIBIOS_DEVICE_NOT_FOUND;
1583 }
1584 
1585 static struct pci_ops null_pci_ops =
1586 {
1587 	.read = null_read_config,
1588 	.write = null_write_config,
1589 };
1590 
1591 /*
1592  * These functions are used early on before PCI scanning is done
1593  * and all of the pci_dev and pci_bus structures have been created.
1594  */
1595 static struct pci_bus *
1596 fake_pci_bus(struct pci_controller *hose, int busnr)
1597 {
1598 	static struct pci_bus bus;
1599 
1600 	if (hose == 0) {
1601 		printk(KERN_ERR "Can't find hose for PCI bus %d!\n", busnr);
1602 	}
1603 	bus.number = busnr;
1604 	bus.sysdata = hose;
1605 	bus.ops = hose? hose->ops: &null_pci_ops;
1606 	return &bus;
1607 }
1608 
1609 #define EARLY_PCI_OP(rw, size, type)					\
1610 int early_##rw##_config_##size(struct pci_controller *hose, int bus,	\
1611 			       int devfn, int offset, type value)	\
1612 {									\
1613 	return pci_bus_##rw##_config_##size(fake_pci_bus(hose, bus),	\
1614 					    devfn, offset, value);	\
1615 }
1616 
1617 EARLY_PCI_OP(read, byte, u8 *)
1618 EARLY_PCI_OP(read, word, u16 *)
1619 EARLY_PCI_OP(read, dword, u32 *)
1620 EARLY_PCI_OP(write, byte, u8)
1621 EARLY_PCI_OP(write, word, u16)
1622 EARLY_PCI_OP(write, dword, u32)
1623 
1624 extern int pci_bus_find_capability (struct pci_bus *bus, unsigned int devfn, int cap);
1625 int early_find_capability(struct pci_controller *hose, int bus, int devfn,
1626 			  int cap)
1627 {
1628 	return pci_bus_find_capability(fake_pci_bus(hose, bus), devfn, cap);
1629 }
1630 
1631 struct device_node *pcibios_get_phb_of_node(struct pci_bus *bus)
1632 {
1633 	struct pci_controller *hose = bus->sysdata;
1634 
1635 	return of_node_get(hose->dn);
1636 }
1637 
1638 /**
1639  * pci_scan_phb - Given a pci_controller, setup and scan the PCI bus
1640  * @hose: Pointer to the PCI host controller instance structure
1641  */
1642 void __devinit pcibios_scan_phb(struct pci_controller *hose)
1643 {
1644 	LIST_HEAD(resources);
1645 	struct pci_bus *bus;
1646 	struct device_node *node = hose->dn;
1647 	int mode;
1648 
1649 	pr_debug("PCI: Scanning PHB %s\n", of_node_full_name(node));
1650 
1651 	/* Get some IO space for the new PHB */
1652 	pcibios_setup_phb_io_space(hose);
1653 
1654 	/* Wire up PHB bus resources */
1655 	pcibios_setup_phb_resources(hose, &resources);
1656 
1657 	hose->busn.start = hose->first_busno;
1658 	hose->busn.end	 = hose->last_busno;
1659 	hose->busn.flags = IORESOURCE_BUS;
1660 	pci_add_resource(&resources, &hose->busn);
1661 
1662 	/* Create an empty bus for the toplevel */
1663 	bus = pci_create_root_bus(hose->parent, hose->first_busno,
1664 				  hose->ops, hose, &resources);
1665 	if (bus == NULL) {
1666 		pr_err("Failed to create bus for PCI domain %04x\n",
1667 			hose->global_number);
1668 		pci_free_resource_list(&resources);
1669 		return;
1670 	}
1671 	hose->bus = bus;
1672 
1673 	/* Get probe mode and perform scan */
1674 	mode = PCI_PROBE_NORMAL;
1675 	if (node && ppc_md.pci_probe_mode)
1676 		mode = ppc_md.pci_probe_mode(bus);
1677 	pr_debug("    probe mode: %d\n", mode);
1678 	if (mode == PCI_PROBE_DEVTREE)
1679 		of_scan_bus(node, bus);
1680 
1681 	if (mode == PCI_PROBE_NORMAL) {
1682 		pci_bus_update_busn_res_end(bus, 255);
1683 		hose->last_busno = pci_scan_child_bus(bus);
1684 		pci_bus_update_busn_res_end(bus, hose->last_busno);
1685 	}
1686 
1687 	/* Platform gets a chance to do some global fixups before
1688 	 * we proceed to resource allocation
1689 	 */
1690 	if (ppc_md.pcibios_fixup_phb)
1691 		ppc_md.pcibios_fixup_phb(hose);
1692 
1693 	/* Configure PCI Express settings */
1694 	if (bus && !pci_has_flag(PCI_PROBE_ONLY)) {
1695 		struct pci_bus *child;
1696 		list_for_each_entry(child, &bus->children, node) {
1697 			struct pci_dev *self = child->self;
1698 			if (!self)
1699 				continue;
1700 			pcie_bus_configure_settings(child, self->pcie_mpss);
1701 		}
1702 	}
1703 }
1704 
1705 static void fixup_hide_host_resource_fsl(struct pci_dev *dev)
1706 {
1707 	int i, class = dev->class >> 8;
1708 	/* When configured as agent, programing interface = 1 */
1709 	int prog_if = dev->class & 0xf;
1710 
1711 	if ((class == PCI_CLASS_PROCESSOR_POWERPC ||
1712 	     class == PCI_CLASS_BRIDGE_OTHER) &&
1713 		(dev->hdr_type == PCI_HEADER_TYPE_NORMAL) &&
1714 		(prog_if == 0) &&
1715 		(dev->bus->parent == NULL)) {
1716 		for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
1717 			dev->resource[i].start = 0;
1718 			dev->resource[i].end = 0;
1719 			dev->resource[i].flags = 0;
1720 		}
1721 	}
1722 }
1723 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_MOTOROLA, PCI_ANY_ID, fixup_hide_host_resource_fsl);
1724 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_FREESCALE, PCI_ANY_ID, fixup_hide_host_resource_fsl);
1725