xref: /openbmc/linux/arch/powerpc/kernel/pci-common.c (revision 82ced6fd)
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/mm.h>
25 #include <linux/list.h>
26 #include <linux/syscalls.h>
27 #include <linux/irq.h>
28 #include <linux/vmalloc.h>
29 
30 #include <asm/processor.h>
31 #include <asm/io.h>
32 #include <asm/prom.h>
33 #include <asm/pci-bridge.h>
34 #include <asm/byteorder.h>
35 #include <asm/machdep.h>
36 #include <asm/ppc-pci.h>
37 #include <asm/firmware.h>
38 #include <asm/eeh.h>
39 
40 static DEFINE_SPINLOCK(hose_spinlock);
41 LIST_HEAD(hose_list);
42 
43 /* XXX kill that some day ... */
44 static int global_phb_number;		/* Global phb counter */
45 
46 /* ISA Memory physical address */
47 resource_size_t isa_mem_base;
48 
49 /* Default PCI flags is 0 on ppc32, modified at boot on ppc64 */
50 unsigned int ppc_pci_flags = 0;
51 
52 
53 static struct dma_mapping_ops *pci_dma_ops = &dma_direct_ops;
54 
55 void set_pci_dma_ops(struct dma_mapping_ops *dma_ops)
56 {
57 	pci_dma_ops = dma_ops;
58 }
59 
60 struct dma_mapping_ops *get_pci_dma_ops(void)
61 {
62 	return pci_dma_ops;
63 }
64 EXPORT_SYMBOL(get_pci_dma_ops);
65 
66 int pci_set_dma_mask(struct pci_dev *dev, u64 mask)
67 {
68 	return dma_set_mask(&dev->dev, mask);
69 }
70 
71 int pci_set_consistent_dma_mask(struct pci_dev *dev, u64 mask)
72 {
73 	int rc;
74 
75 	rc = dma_set_mask(&dev->dev, mask);
76 	dev->dev.coherent_dma_mask = dev->dma_mask;
77 
78 	return rc;
79 }
80 
81 struct pci_controller *pcibios_alloc_controller(struct device_node *dev)
82 {
83 	struct pci_controller *phb;
84 
85 	phb = zalloc_maybe_bootmem(sizeof(struct pci_controller), GFP_KERNEL);
86 	if (phb == NULL)
87 		return NULL;
88 	spin_lock(&hose_spinlock);
89 	phb->global_number = global_phb_number++;
90 	list_add_tail(&phb->list_node, &hose_list);
91 	spin_unlock(&hose_spinlock);
92 	phb->dn = dev;
93 	phb->is_dynamic = mem_init_done;
94 #ifdef CONFIG_PPC64
95 	if (dev) {
96 		int nid = of_node_to_nid(dev);
97 
98 		if (nid < 0 || !node_online(nid))
99 			nid = -1;
100 
101 		PHB_SET_NODE(phb, nid);
102 	}
103 #endif
104 	return phb;
105 }
106 
107 void pcibios_free_controller(struct pci_controller *phb)
108 {
109 	spin_lock(&hose_spinlock);
110 	list_del(&phb->list_node);
111 	spin_unlock(&hose_spinlock);
112 
113 	if (phb->is_dynamic)
114 		kfree(phb);
115 }
116 
117 static resource_size_t pcibios_io_size(const struct pci_controller *hose)
118 {
119 #ifdef CONFIG_PPC64
120 	return hose->pci_io_size;
121 #else
122 	return hose->io_resource.end - hose->io_resource.start + 1;
123 #endif
124 }
125 
126 int pcibios_vaddr_is_ioport(void __iomem *address)
127 {
128 	int ret = 0;
129 	struct pci_controller *hose;
130 	resource_size_t size;
131 
132 	spin_lock(&hose_spinlock);
133 	list_for_each_entry(hose, &hose_list, list_node) {
134 		size = pcibios_io_size(hose);
135 		if (address >= hose->io_base_virt &&
136 		    address < (hose->io_base_virt + size)) {
137 			ret = 1;
138 			break;
139 		}
140 	}
141 	spin_unlock(&hose_spinlock);
142 	return ret;
143 }
144 
145 unsigned long pci_address_to_pio(phys_addr_t address)
146 {
147 	struct pci_controller *hose;
148 	resource_size_t size;
149 	unsigned long ret = ~0;
150 
151 	spin_lock(&hose_spinlock);
152 	list_for_each_entry(hose, &hose_list, list_node) {
153 		size = pcibios_io_size(hose);
154 		if (address >= hose->io_base_phys &&
155 		    address < (hose->io_base_phys + size)) {
156 			unsigned long base =
157 				(unsigned long)hose->io_base_virt - _IO_BASE;
158 			ret = base + (address - hose->io_base_phys);
159 			break;
160 		}
161 	}
162 	spin_unlock(&hose_spinlock);
163 
164 	return ret;
165 }
166 EXPORT_SYMBOL_GPL(pci_address_to_pio);
167 
168 /*
169  * Return the domain number for this bus.
170  */
171 int pci_domain_nr(struct pci_bus *bus)
172 {
173 	struct pci_controller *hose = pci_bus_to_host(bus);
174 
175 	return hose->global_number;
176 }
177 EXPORT_SYMBOL(pci_domain_nr);
178 
179 #ifdef CONFIG_PPC_OF
180 
181 /* This routine is meant to be used early during boot, when the
182  * PCI bus numbers have not yet been assigned, and you need to
183  * issue PCI config cycles to an OF device.
184  * It could also be used to "fix" RTAS config cycles if you want
185  * to set pci_assign_all_buses to 1 and still use RTAS for PCI
186  * config cycles.
187  */
188 struct pci_controller* pci_find_hose_for_OF_device(struct device_node* node)
189 {
190 	while(node) {
191 		struct pci_controller *hose, *tmp;
192 		list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
193 			if (hose->dn == node)
194 				return hose;
195 		node = node->parent;
196 	}
197 	return NULL;
198 }
199 
200 static ssize_t pci_show_devspec(struct device *dev,
201 		struct device_attribute *attr, char *buf)
202 {
203 	struct pci_dev *pdev;
204 	struct device_node *np;
205 
206 	pdev = to_pci_dev (dev);
207 	np = pci_device_to_OF_node(pdev);
208 	if (np == NULL || np->full_name == NULL)
209 		return 0;
210 	return sprintf(buf, "%s", np->full_name);
211 }
212 static DEVICE_ATTR(devspec, S_IRUGO, pci_show_devspec, NULL);
213 #endif /* CONFIG_PPC_OF */
214 
215 /* Add sysfs properties */
216 int pcibios_add_platform_entries(struct pci_dev *pdev)
217 {
218 #ifdef CONFIG_PPC_OF
219 	return device_create_file(&pdev->dev, &dev_attr_devspec);
220 #else
221 	return 0;
222 #endif /* CONFIG_PPC_OF */
223 
224 }
225 
226 char __devinit *pcibios_setup(char *str)
227 {
228 	return str;
229 }
230 
231 /*
232  * Reads the interrupt pin to determine if interrupt is use by card.
233  * If the interrupt is used, then gets the interrupt line from the
234  * openfirmware and sets it in the pci_dev and pci_config line.
235  */
236 int pci_read_irq_line(struct pci_dev *pci_dev)
237 {
238 	struct of_irq oirq;
239 	unsigned int virq;
240 
241 	/* The current device-tree that iSeries generates from the HV
242 	 * PCI informations doesn't contain proper interrupt routing,
243 	 * and all the fallback would do is print out crap, so we
244 	 * don't attempt to resolve the interrupts here at all, some
245 	 * iSeries specific fixup does it.
246 	 *
247 	 * In the long run, we will hopefully fix the generated device-tree
248 	 * instead.
249 	 */
250 #ifdef CONFIG_PPC_ISERIES
251 	if (firmware_has_feature(FW_FEATURE_ISERIES))
252 		return -1;
253 #endif
254 
255 	pr_debug("PCI: Try to map irq for %s...\n", pci_name(pci_dev));
256 
257 #ifdef DEBUG
258 	memset(&oirq, 0xff, sizeof(oirq));
259 #endif
260 	/* Try to get a mapping from the device-tree */
261 	if (of_irq_map_pci(pci_dev, &oirq)) {
262 		u8 line, pin;
263 
264 		/* If that fails, lets fallback to what is in the config
265 		 * space and map that through the default controller. We
266 		 * also set the type to level low since that's what PCI
267 		 * interrupts are. If your platform does differently, then
268 		 * either provide a proper interrupt tree or don't use this
269 		 * function.
270 		 */
271 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_PIN, &pin))
272 			return -1;
273 		if (pin == 0)
274 			return -1;
275 		if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_LINE, &line) ||
276 		    line == 0xff || line == 0) {
277 			return -1;
278 		}
279 		pr_debug(" No map ! Using line %d (pin %d) from PCI config\n",
280 			 line, pin);
281 
282 		virq = irq_create_mapping(NULL, line);
283 		if (virq != NO_IRQ)
284 			set_irq_type(virq, IRQ_TYPE_LEVEL_LOW);
285 	} else {
286 		pr_debug(" Got one, spec %d cells (0x%08x 0x%08x...) on %s\n",
287 			 oirq.size, oirq.specifier[0], oirq.specifier[1],
288 			 oirq.controller ? oirq.controller->full_name :
289 			 "<default>");
290 
291 		virq = irq_create_of_mapping(oirq.controller, oirq.specifier,
292 					     oirq.size);
293 	}
294 	if(virq == NO_IRQ) {
295 		pr_debug(" Failed to map !\n");
296 		return -1;
297 	}
298 
299 	pr_debug(" Mapped to linux irq %d\n", virq);
300 
301 	pci_dev->irq = virq;
302 
303 	return 0;
304 }
305 EXPORT_SYMBOL(pci_read_irq_line);
306 
307 /*
308  * Platform support for /proc/bus/pci/X/Y mmap()s,
309  * modelled on the sparc64 implementation by Dave Miller.
310  *  -- paulus.
311  */
312 
313 /*
314  * Adjust vm_pgoff of VMA such that it is the physical page offset
315  * corresponding to the 32-bit pci bus offset for DEV requested by the user.
316  *
317  * Basically, the user finds the base address for his device which he wishes
318  * to mmap.  They read the 32-bit value from the config space base register,
319  * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
320  * offset parameter of mmap on /proc/bus/pci/XXX for that device.
321  *
322  * Returns negative error code on failure, zero on success.
323  */
324 static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
325 					       resource_size_t *offset,
326 					       enum pci_mmap_state mmap_state)
327 {
328 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
329 	unsigned long io_offset = 0;
330 	int i, res_bit;
331 
332 	if (hose == 0)
333 		return NULL;		/* should never happen */
334 
335 	/* If memory, add on the PCI bridge address offset */
336 	if (mmap_state == pci_mmap_mem) {
337 #if 0 /* See comment in pci_resource_to_user() for why this is disabled */
338 		*offset += hose->pci_mem_offset;
339 #endif
340 		res_bit = IORESOURCE_MEM;
341 	} else {
342 		io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
343 		*offset += io_offset;
344 		res_bit = IORESOURCE_IO;
345 	}
346 
347 	/*
348 	 * Check that the offset requested corresponds to one of the
349 	 * resources of the device.
350 	 */
351 	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
352 		struct resource *rp = &dev->resource[i];
353 		int flags = rp->flags;
354 
355 		/* treat ROM as memory (should be already) */
356 		if (i == PCI_ROM_RESOURCE)
357 			flags |= IORESOURCE_MEM;
358 
359 		/* Active and same type? */
360 		if ((flags & res_bit) == 0)
361 			continue;
362 
363 		/* In the range of this resource? */
364 		if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
365 			continue;
366 
367 		/* found it! construct the final physical address */
368 		if (mmap_state == pci_mmap_io)
369 			*offset += hose->io_base_phys - io_offset;
370 		return rp;
371 	}
372 
373 	return NULL;
374 }
375 
376 /*
377  * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
378  * device mapping.
379  */
380 static pgprot_t __pci_mmap_set_pgprot(struct pci_dev *dev, struct resource *rp,
381 				      pgprot_t protection,
382 				      enum pci_mmap_state mmap_state,
383 				      int write_combine)
384 {
385 	unsigned long prot = pgprot_val(protection);
386 
387 	/* Write combine is always 0 on non-memory space mappings. On
388 	 * memory space, if the user didn't pass 1, we check for a
389 	 * "prefetchable" resource. This is a bit hackish, but we use
390 	 * this to workaround the inability of /sysfs to provide a write
391 	 * combine bit
392 	 */
393 	if (mmap_state != pci_mmap_mem)
394 		write_combine = 0;
395 	else if (write_combine == 0) {
396 		if (rp->flags & IORESOURCE_PREFETCH)
397 			write_combine = 1;
398 	}
399 
400 	/* XXX would be nice to have a way to ask for write-through */
401 	if (write_combine)
402 		return pgprot_noncached_wc(prot);
403 	else
404 		return pgprot_noncached(prot);
405 }
406 
407 /*
408  * This one is used by /dev/mem and fbdev who have no clue about the
409  * PCI device, it tries to find the PCI device first and calls the
410  * above routine
411  */
412 pgprot_t pci_phys_mem_access_prot(struct file *file,
413 				  unsigned long pfn,
414 				  unsigned long size,
415 				  pgprot_t prot)
416 {
417 	struct pci_dev *pdev = NULL;
418 	struct resource *found = NULL;
419 	resource_size_t offset = ((resource_size_t)pfn) << PAGE_SHIFT;
420 	int i;
421 
422 	if (page_is_ram(pfn))
423 		return prot;
424 
425 	prot = pgprot_noncached(prot);
426 	for_each_pci_dev(pdev) {
427 		for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
428 			struct resource *rp = &pdev->resource[i];
429 			int flags = rp->flags;
430 
431 			/* Active and same type? */
432 			if ((flags & IORESOURCE_MEM) == 0)
433 				continue;
434 			/* In the range of this resource? */
435 			if (offset < (rp->start & PAGE_MASK) ||
436 			    offset > rp->end)
437 				continue;
438 			found = rp;
439 			break;
440 		}
441 		if (found)
442 			break;
443 	}
444 	if (found) {
445 		if (found->flags & IORESOURCE_PREFETCH)
446 			prot = pgprot_noncached_wc(prot);
447 		pci_dev_put(pdev);
448 	}
449 
450 	pr_debug("PCI: Non-PCI map for %llx, prot: %lx\n",
451 		 (unsigned long long)offset, pgprot_val(prot));
452 
453 	return prot;
454 }
455 
456 
457 /*
458  * Perform the actual remap of the pages for a PCI device mapping, as
459  * appropriate for this architecture.  The region in the process to map
460  * is described by vm_start and vm_end members of VMA, the base physical
461  * address is found in vm_pgoff.
462  * The pci device structure is provided so that architectures may make mapping
463  * decisions on a per-device or per-bus basis.
464  *
465  * Returns a negative error code on failure, zero on success.
466  */
467 int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
468 			enum pci_mmap_state mmap_state, int write_combine)
469 {
470 	resource_size_t offset =
471 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
472 	struct resource *rp;
473 	int ret;
474 
475 	rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
476 	if (rp == NULL)
477 		return -EINVAL;
478 
479 	vma->vm_pgoff = offset >> PAGE_SHIFT;
480 	vma->vm_page_prot = __pci_mmap_set_pgprot(dev, rp,
481 						  vma->vm_page_prot,
482 						  mmap_state, write_combine);
483 
484 	ret = remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
485 			       vma->vm_end - vma->vm_start, vma->vm_page_prot);
486 
487 	return ret;
488 }
489 
490 /* This provides legacy IO read access on a bus */
491 int pci_legacy_read(struct pci_bus *bus, loff_t port, u32 *val, size_t size)
492 {
493 	unsigned long offset;
494 	struct pci_controller *hose = pci_bus_to_host(bus);
495 	struct resource *rp = &hose->io_resource;
496 	void __iomem *addr;
497 
498 	/* Check if port can be supported by that bus. We only check
499 	 * the ranges of the PHB though, not the bus itself as the rules
500 	 * for forwarding legacy cycles down bridges are not our problem
501 	 * here. So if the host bridge supports it, we do it.
502 	 */
503 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
504 	offset += port;
505 
506 	if (!(rp->flags & IORESOURCE_IO))
507 		return -ENXIO;
508 	if (offset < rp->start || (offset + size) > rp->end)
509 		return -ENXIO;
510 	addr = hose->io_base_virt + port;
511 
512 	switch(size) {
513 	case 1:
514 		*((u8 *)val) = in_8(addr);
515 		return 1;
516 	case 2:
517 		if (port & 1)
518 			return -EINVAL;
519 		*((u16 *)val) = in_le16(addr);
520 		return 2;
521 	case 4:
522 		if (port & 3)
523 			return -EINVAL;
524 		*((u32 *)val) = in_le32(addr);
525 		return 4;
526 	}
527 	return -EINVAL;
528 }
529 
530 /* This provides legacy IO write access on a bus */
531 int pci_legacy_write(struct pci_bus *bus, loff_t port, u32 val, size_t size)
532 {
533 	unsigned long offset;
534 	struct pci_controller *hose = pci_bus_to_host(bus);
535 	struct resource *rp = &hose->io_resource;
536 	void __iomem *addr;
537 
538 	/* Check if port can be supported by that bus. We only check
539 	 * the ranges of the PHB though, not the bus itself as the rules
540 	 * for forwarding legacy cycles down bridges are not our problem
541 	 * here. So if the host bridge supports it, we do it.
542 	 */
543 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
544 	offset += port;
545 
546 	if (!(rp->flags & IORESOURCE_IO))
547 		return -ENXIO;
548 	if (offset < rp->start || (offset + size) > rp->end)
549 		return -ENXIO;
550 	addr = hose->io_base_virt + port;
551 
552 	/* WARNING: The generic code is idiotic. It gets passed a pointer
553 	 * to what can be a 1, 2 or 4 byte quantity and always reads that
554 	 * as a u32, which means that we have to correct the location of
555 	 * the data read within those 32 bits for size 1 and 2
556 	 */
557 	switch(size) {
558 	case 1:
559 		out_8(addr, val >> 24);
560 		return 1;
561 	case 2:
562 		if (port & 1)
563 			return -EINVAL;
564 		out_le16(addr, val >> 16);
565 		return 2;
566 	case 4:
567 		if (port & 3)
568 			return -EINVAL;
569 		out_le32(addr, val);
570 		return 4;
571 	}
572 	return -EINVAL;
573 }
574 
575 /* This provides legacy IO or memory mmap access on a bus */
576 int pci_mmap_legacy_page_range(struct pci_bus *bus,
577 			       struct vm_area_struct *vma,
578 			       enum pci_mmap_state mmap_state)
579 {
580 	struct pci_controller *hose = pci_bus_to_host(bus);
581 	resource_size_t offset =
582 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
583 	resource_size_t size = vma->vm_end - vma->vm_start;
584 	struct resource *rp;
585 
586 	pr_debug("pci_mmap_legacy_page_range(%04x:%02x, %s @%llx..%llx)\n",
587 		 pci_domain_nr(bus), bus->number,
588 		 mmap_state == pci_mmap_mem ? "MEM" : "IO",
589 		 (unsigned long long)offset,
590 		 (unsigned long long)(offset + size - 1));
591 
592 	if (mmap_state == pci_mmap_mem) {
593 		/* Hack alert !
594 		 *
595 		 * Because X is lame and can fail starting if it gets an error trying
596 		 * to mmap legacy_mem (instead of just moving on without legacy memory
597 		 * access) we fake it here by giving it anonymous memory, effectively
598 		 * behaving just like /dev/zero
599 		 */
600 		if ((offset + size) > hose->isa_mem_size) {
601 			printk(KERN_DEBUG
602 			       "Process %s (pid:%d) mapped non-existing PCI legacy memory for 0%04x:%02x\n",
603 			       current->comm, current->pid, pci_domain_nr(bus), bus->number);
604 			if (vma->vm_flags & VM_SHARED)
605 				return shmem_zero_setup(vma);
606 			return 0;
607 		}
608 		offset += hose->isa_mem_phys;
609 	} else {
610 		unsigned long io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
611 		unsigned long roffset = offset + io_offset;
612 		rp = &hose->io_resource;
613 		if (!(rp->flags & IORESOURCE_IO))
614 			return -ENXIO;
615 		if (roffset < rp->start || (roffset + size) > rp->end)
616 			return -ENXIO;
617 		offset += hose->io_base_phys;
618 	}
619 	pr_debug(" -> mapping phys %llx\n", (unsigned long long)offset);
620 
621 	vma->vm_pgoff = offset >> PAGE_SHIFT;
622 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
623 	return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
624 			       vma->vm_end - vma->vm_start,
625 			       vma->vm_page_prot);
626 }
627 
628 void pci_resource_to_user(const struct pci_dev *dev, int bar,
629 			  const struct resource *rsrc,
630 			  resource_size_t *start, resource_size_t *end)
631 {
632 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
633 	resource_size_t offset = 0;
634 
635 	if (hose == NULL)
636 		return;
637 
638 	if (rsrc->flags & IORESOURCE_IO)
639 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
640 
641 	/* We pass a fully fixed up address to userland for MMIO instead of
642 	 * a BAR value because X is lame and expects to be able to use that
643 	 * to pass to /dev/mem !
644 	 *
645 	 * That means that we'll have potentially 64 bits values where some
646 	 * userland apps only expect 32 (like X itself since it thinks only
647 	 * Sparc has 64 bits MMIO) but if we don't do that, we break it on
648 	 * 32 bits CHRPs :-(
649 	 *
650 	 * Hopefully, the sysfs insterface is immune to that gunk. Once X
651 	 * has been fixed (and the fix spread enough), we can re-enable the
652 	 * 2 lines below and pass down a BAR value to userland. In that case
653 	 * we'll also have to re-enable the matching code in
654 	 * __pci_mmap_make_offset().
655 	 *
656 	 * BenH.
657 	 */
658 #if 0
659 	else if (rsrc->flags & IORESOURCE_MEM)
660 		offset = hose->pci_mem_offset;
661 #endif
662 
663 	*start = rsrc->start - offset;
664 	*end = rsrc->end - offset;
665 }
666 
667 /**
668  * pci_process_bridge_OF_ranges - Parse PCI bridge resources from device tree
669  * @hose: newly allocated pci_controller to be setup
670  * @dev: device node of the host bridge
671  * @primary: set if primary bus (32 bits only, soon to be deprecated)
672  *
673  * This function will parse the "ranges" property of a PCI host bridge device
674  * node and setup the resource mapping of a pci controller based on its
675  * content.
676  *
677  * Life would be boring if it wasn't for a few issues that we have to deal
678  * with here:
679  *
680  *   - We can only cope with one IO space range and up to 3 Memory space
681  *     ranges. However, some machines (thanks Apple !) tend to split their
682  *     space into lots of small contiguous ranges. So we have to coalesce.
683  *
684  *   - We can only cope with all memory ranges having the same offset
685  *     between CPU addresses and PCI addresses. Unfortunately, some bridges
686  *     are setup for a large 1:1 mapping along with a small "window" which
687  *     maps PCI address 0 to some arbitrary high address of the CPU space in
688  *     order to give access to the ISA memory hole.
689  *     The way out of here that I've chosen for now is to always set the
690  *     offset based on the first resource found, then override it if we
691  *     have a different offset and the previous was set by an ISA hole.
692  *
693  *   - Some busses have IO space not starting at 0, which causes trouble with
694  *     the way we do our IO resource renumbering. The code somewhat deals with
695  *     it for 64 bits but I would expect problems on 32 bits.
696  *
697  *   - Some 32 bits platforms such as 4xx can have physical space larger than
698  *     32 bits so we need to use 64 bits values for the parsing
699  */
700 void __devinit pci_process_bridge_OF_ranges(struct pci_controller *hose,
701 					    struct device_node *dev,
702 					    int primary)
703 {
704 	const u32 *ranges;
705 	int rlen;
706 	int pna = of_n_addr_cells(dev);
707 	int np = pna + 5;
708 	int memno = 0, isa_hole = -1;
709 	u32 pci_space;
710 	unsigned long long pci_addr, cpu_addr, pci_next, cpu_next, size;
711 	unsigned long long isa_mb = 0;
712 	struct resource *res;
713 
714 	printk(KERN_INFO "PCI host bridge %s %s ranges:\n",
715 	       dev->full_name, primary ? "(primary)" : "");
716 
717 	/* Get ranges property */
718 	ranges = of_get_property(dev, "ranges", &rlen);
719 	if (ranges == NULL)
720 		return;
721 
722 	/* Parse it */
723 	while ((rlen -= np * 4) >= 0) {
724 		/* Read next ranges element */
725 		pci_space = ranges[0];
726 		pci_addr = of_read_number(ranges + 1, 2);
727 		cpu_addr = of_translate_address(dev, ranges + 3);
728 		size = of_read_number(ranges + pna + 3, 2);
729 		ranges += np;
730 
731 		/* If we failed translation or got a zero-sized region
732 		 * (some FW try to feed us with non sensical zero sized regions
733 		 * such as power3 which look like some kind of attempt at exposing
734 		 * the VGA memory hole)
735 		 */
736 		if (cpu_addr == OF_BAD_ADDR || size == 0)
737 			continue;
738 
739 		/* Now consume following elements while they are contiguous */
740 		for (; rlen >= np * sizeof(u32);
741 		     ranges += np, rlen -= np * 4) {
742 			if (ranges[0] != pci_space)
743 				break;
744 			pci_next = of_read_number(ranges + 1, 2);
745 			cpu_next = of_translate_address(dev, ranges + 3);
746 			if (pci_next != pci_addr + size ||
747 			    cpu_next != cpu_addr + size)
748 				break;
749 			size += of_read_number(ranges + pna + 3, 2);
750 		}
751 
752 		/* Act based on address space type */
753 		res = NULL;
754 		switch ((pci_space >> 24) & 0x3) {
755 		case 1:		/* PCI IO space */
756 			printk(KERN_INFO
757 			       "  IO 0x%016llx..0x%016llx -> 0x%016llx\n",
758 			       cpu_addr, cpu_addr + size - 1, pci_addr);
759 
760 			/* We support only one IO range */
761 			if (hose->pci_io_size) {
762 				printk(KERN_INFO
763 				       " \\--> Skipped (too many) !\n");
764 				continue;
765 			}
766 #ifdef CONFIG_PPC32
767 			/* On 32 bits, limit I/O space to 16MB */
768 			if (size > 0x01000000)
769 				size = 0x01000000;
770 
771 			/* 32 bits needs to map IOs here */
772 			hose->io_base_virt = ioremap(cpu_addr, size);
773 
774 			/* Expect trouble if pci_addr is not 0 */
775 			if (primary)
776 				isa_io_base =
777 					(unsigned long)hose->io_base_virt;
778 #endif /* CONFIG_PPC32 */
779 			/* pci_io_size and io_base_phys always represent IO
780 			 * space starting at 0 so we factor in pci_addr
781 			 */
782 			hose->pci_io_size = pci_addr + size;
783 			hose->io_base_phys = cpu_addr - pci_addr;
784 
785 			/* Build resource */
786 			res = &hose->io_resource;
787 			res->flags = IORESOURCE_IO;
788 			res->start = pci_addr;
789 			break;
790 		case 2:		/* PCI Memory space */
791 		case 3:		/* PCI 64 bits Memory space */
792 			printk(KERN_INFO
793 			       " MEM 0x%016llx..0x%016llx -> 0x%016llx %s\n",
794 			       cpu_addr, cpu_addr + size - 1, pci_addr,
795 			       (pci_space & 0x40000000) ? "Prefetch" : "");
796 
797 			/* We support only 3 memory ranges */
798 			if (memno >= 3) {
799 				printk(KERN_INFO
800 				       " \\--> Skipped (too many) !\n");
801 				continue;
802 			}
803 			/* Handles ISA memory hole space here */
804 			if (pci_addr == 0) {
805 				isa_mb = cpu_addr;
806 				isa_hole = memno;
807 				if (primary || isa_mem_base == 0)
808 					isa_mem_base = cpu_addr;
809 				hose->isa_mem_phys = cpu_addr;
810 				hose->isa_mem_size = size;
811 			}
812 
813 			/* We get the PCI/Mem offset from the first range or
814 			 * the, current one if the offset came from an ISA
815 			 * hole. If they don't match, bugger.
816 			 */
817 			if (memno == 0 ||
818 			    (isa_hole >= 0 && pci_addr != 0 &&
819 			     hose->pci_mem_offset == isa_mb))
820 				hose->pci_mem_offset = cpu_addr - pci_addr;
821 			else if (pci_addr != 0 &&
822 				 hose->pci_mem_offset != cpu_addr - pci_addr) {
823 				printk(KERN_INFO
824 				       " \\--> Skipped (offset mismatch) !\n");
825 				continue;
826 			}
827 
828 			/* Build resource */
829 			res = &hose->mem_resources[memno++];
830 			res->flags = IORESOURCE_MEM;
831 			if (pci_space & 0x40000000)
832 				res->flags |= IORESOURCE_PREFETCH;
833 			res->start = cpu_addr;
834 			break;
835 		}
836 		if (res != NULL) {
837 			res->name = dev->full_name;
838 			res->end = res->start + size - 1;
839 			res->parent = NULL;
840 			res->sibling = NULL;
841 			res->child = NULL;
842 		}
843 	}
844 
845 	/* If there's an ISA hole and the pci_mem_offset is -not- matching
846 	 * the ISA hole offset, then we need to remove the ISA hole from
847 	 * the resource list for that brige
848 	 */
849 	if (isa_hole >= 0 && hose->pci_mem_offset != isa_mb) {
850 		unsigned int next = isa_hole + 1;
851 		printk(KERN_INFO " Removing ISA hole at 0x%016llx\n", isa_mb);
852 		if (next < memno)
853 			memmove(&hose->mem_resources[isa_hole],
854 				&hose->mem_resources[next],
855 				sizeof(struct resource) * (memno - next));
856 		hose->mem_resources[--memno].flags = 0;
857 	}
858 }
859 
860 /* Decide whether to display the domain number in /proc */
861 int pci_proc_domain(struct pci_bus *bus)
862 {
863 	struct pci_controller *hose = pci_bus_to_host(bus);
864 
865 	if (!(ppc_pci_flags & PPC_PCI_ENABLE_PROC_DOMAINS))
866 		return 0;
867 	if (ppc_pci_flags & PPC_PCI_COMPAT_DOMAIN_0)
868 		return hose->global_number != 0;
869 	return 1;
870 }
871 
872 void pcibios_resource_to_bus(struct pci_dev *dev, struct pci_bus_region *region,
873 			     struct resource *res)
874 {
875 	resource_size_t offset = 0, mask = (resource_size_t)-1;
876 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
877 
878 	if (!hose)
879 		return;
880 	if (res->flags & IORESOURCE_IO) {
881 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
882 		mask = 0xffffffffu;
883 	} else if (res->flags & IORESOURCE_MEM)
884 		offset = hose->pci_mem_offset;
885 
886 	region->start = (res->start - offset) & mask;
887 	region->end = (res->end - offset) & mask;
888 }
889 EXPORT_SYMBOL(pcibios_resource_to_bus);
890 
891 void pcibios_bus_to_resource(struct pci_dev *dev, struct resource *res,
892 			     struct pci_bus_region *region)
893 {
894 	resource_size_t offset = 0, mask = (resource_size_t)-1;
895 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
896 
897 	if (!hose)
898 		return;
899 	if (res->flags & IORESOURCE_IO) {
900 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
901 		mask = 0xffffffffu;
902 	} else if (res->flags & IORESOURCE_MEM)
903 		offset = hose->pci_mem_offset;
904 	res->start = (region->start + offset) & mask;
905 	res->end = (region->end + offset) & mask;
906 }
907 EXPORT_SYMBOL(pcibios_bus_to_resource);
908 
909 /* Fixup a bus resource into a linux resource */
910 static void __devinit fixup_resource(struct resource *res, struct pci_dev *dev)
911 {
912 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
913 	resource_size_t offset = 0, mask = (resource_size_t)-1;
914 
915 	if (res->flags & IORESOURCE_IO) {
916 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
917 		mask = 0xffffffffu;
918 	} else if (res->flags & IORESOURCE_MEM)
919 		offset = hose->pci_mem_offset;
920 
921 	res->start = (res->start + offset) & mask;
922 	res->end = (res->end + offset) & mask;
923 }
924 
925 
926 /* This header fixup will do the resource fixup for all devices as they are
927  * probed, but not for bridge ranges
928  */
929 static void __devinit pcibios_fixup_resources(struct pci_dev *dev)
930 {
931 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
932 	int i;
933 
934 	if (!hose) {
935 		printk(KERN_ERR "No host bridge for PCI dev %s !\n",
936 		       pci_name(dev));
937 		return;
938 	}
939 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
940 		struct resource *res = dev->resource + i;
941 		if (!res->flags)
942 			continue;
943 		/* On platforms that have PPC_PCI_PROBE_ONLY set, we don't
944 		 * consider 0 as an unassigned BAR value. It's technically
945 		 * a valid value, but linux doesn't like it... so when we can
946 		 * re-assign things, we do so, but if we can't, we keep it
947 		 * around and hope for the best...
948 		 */
949 		if (res->start == 0 && !(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
950 			pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] is unassigned\n",
951 				 pci_name(dev), i,
952 				 (unsigned long long)res->start,
953 				 (unsigned long long)res->end,
954 				 (unsigned int)res->flags);
955 			res->end -= res->start;
956 			res->start = 0;
957 			res->flags |= IORESOURCE_UNSET;
958 			continue;
959 		}
960 
961 		pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] fixup...\n",
962 			 pci_name(dev), i,
963 			 (unsigned long long)res->start,\
964 			 (unsigned long long)res->end,
965 			 (unsigned int)res->flags);
966 
967 		fixup_resource(res, dev);
968 
969 		pr_debug("PCI:%s            %016llx-%016llx\n",
970 			 pci_name(dev),
971 			 (unsigned long long)res->start,
972 			 (unsigned long long)res->end);
973 	}
974 
975 	/* Call machine specific resource fixup */
976 	if (ppc_md.pcibios_fixup_resources)
977 		ppc_md.pcibios_fixup_resources(dev);
978 }
979 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_resources);
980 
981 /* This function tries to figure out if a bridge resource has been initialized
982  * by the firmware or not. It doesn't have to be absolutely bullet proof, but
983  * things go more smoothly when it gets it right. It should covers cases such
984  * as Apple "closed" bridge resources and bare-metal pSeries unassigned bridges
985  */
986 static int __devinit pcibios_uninitialized_bridge_resource(struct pci_bus *bus,
987 							   struct resource *res)
988 {
989 	struct pci_controller *hose = pci_bus_to_host(bus);
990 	struct pci_dev *dev = bus->self;
991 	resource_size_t offset;
992 	u16 command;
993 	int i;
994 
995 	/* We don't do anything if PCI_PROBE_ONLY is set */
996 	if (ppc_pci_flags & PPC_PCI_PROBE_ONLY)
997 		return 0;
998 
999 	/* Job is a bit different between memory and IO */
1000 	if (res->flags & IORESOURCE_MEM) {
1001 		/* If the BAR is non-0 (res != pci_mem_offset) then it's probably been
1002 		 * initialized by somebody
1003 		 */
1004 		if (res->start != hose->pci_mem_offset)
1005 			return 0;
1006 
1007 		/* The BAR is 0, let's check if memory decoding is enabled on
1008 		 * the bridge. If not, we consider it unassigned
1009 		 */
1010 		pci_read_config_word(dev, PCI_COMMAND, &command);
1011 		if ((command & PCI_COMMAND_MEMORY) == 0)
1012 			return 1;
1013 
1014 		/* Memory decoding is enabled and the BAR is 0. If any of the bridge
1015 		 * resources covers that starting address (0 then it's good enough for
1016 		 * us for memory
1017 		 */
1018 		for (i = 0; i < 3; i++) {
1019 			if ((hose->mem_resources[i].flags & IORESOURCE_MEM) &&
1020 			    hose->mem_resources[i].start == hose->pci_mem_offset)
1021 				return 0;
1022 		}
1023 
1024 		/* Well, it starts at 0 and we know it will collide so we may as
1025 		 * well consider it as unassigned. That covers the Apple case.
1026 		 */
1027 		return 1;
1028 	} else {
1029 		/* If the BAR is non-0, then we consider it assigned */
1030 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1031 		if (((res->start - offset) & 0xfffffffful) != 0)
1032 			return 0;
1033 
1034 		/* Here, we are a bit different than memory as typically IO space
1035 		 * starting at low addresses -is- valid. What we do instead if that
1036 		 * we consider as unassigned anything that doesn't have IO enabled
1037 		 * in the PCI command register, and that's it.
1038 		 */
1039 		pci_read_config_word(dev, PCI_COMMAND, &command);
1040 		if (command & PCI_COMMAND_IO)
1041 			return 0;
1042 
1043 		/* It's starting at 0 and IO is disabled in the bridge, consider
1044 		 * it unassigned
1045 		 */
1046 		return 1;
1047 	}
1048 }
1049 
1050 /* Fixup resources of a PCI<->PCI bridge */
1051 static void __devinit pcibios_fixup_bridge(struct pci_bus *bus)
1052 {
1053 	struct resource *res;
1054 	int i;
1055 
1056 	struct pci_dev *dev = bus->self;
1057 
1058 	for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
1059 		if ((res = bus->resource[i]) == NULL)
1060 			continue;
1061 		if (!res->flags)
1062 			continue;
1063 		if (i >= 3 && bus->self->transparent)
1064 			continue;
1065 
1066 		pr_debug("PCI:%s Bus rsrc %d %016llx-%016llx [%x] fixup...\n",
1067 			 pci_name(dev), i,
1068 			 (unsigned long long)res->start,\
1069 			 (unsigned long long)res->end,
1070 			 (unsigned int)res->flags);
1071 
1072 		/* Perform fixup */
1073 		fixup_resource(res, dev);
1074 
1075 		/* Try to detect uninitialized P2P bridge resources,
1076 		 * and clear them out so they get re-assigned later
1077 		 */
1078 		if (pcibios_uninitialized_bridge_resource(bus, res)) {
1079 			res->flags = 0;
1080 			pr_debug("PCI:%s            (unassigned)\n", pci_name(dev));
1081 		} else {
1082 
1083 			pr_debug("PCI:%s            %016llx-%016llx\n",
1084 				 pci_name(dev),
1085 				 (unsigned long long)res->start,
1086 				 (unsigned long long)res->end);
1087 		}
1088 	}
1089 }
1090 
1091 void __devinit pcibios_setup_bus_self(struct pci_bus *bus)
1092 {
1093 	/* Fix up the bus resources for P2P bridges */
1094 	if (bus->self != NULL)
1095 		pcibios_fixup_bridge(bus);
1096 
1097 	/* Platform specific bus fixups. This is currently only used
1098 	 * by fsl_pci and I'm hoping to get rid of it at some point
1099 	 */
1100 	if (ppc_md.pcibios_fixup_bus)
1101 		ppc_md.pcibios_fixup_bus(bus);
1102 
1103 	/* Setup bus DMA mappings */
1104 	if (ppc_md.pci_dma_bus_setup)
1105 		ppc_md.pci_dma_bus_setup(bus);
1106 }
1107 
1108 void __devinit pcibios_setup_bus_devices(struct pci_bus *bus)
1109 {
1110 	struct pci_dev *dev;
1111 
1112 	pr_debug("PCI: Fixup bus devices %d (%s)\n",
1113 		 bus->number, bus->self ? pci_name(bus->self) : "PHB");
1114 
1115 	list_for_each_entry(dev, &bus->devices, bus_list) {
1116 		struct dev_archdata *sd = &dev->dev.archdata;
1117 
1118 		/* Setup OF node pointer in archdata */
1119 		sd->of_node = pci_device_to_OF_node(dev);
1120 
1121 		/* Fixup NUMA node as it may not be setup yet by the generic
1122 		 * code and is needed by the DMA init
1123 		 */
1124 		set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
1125 
1126 		/* Hook up default DMA ops */
1127 		sd->dma_ops = pci_dma_ops;
1128 		sd->dma_data = (void *)PCI_DRAM_OFFSET;
1129 
1130 		/* Additional platform DMA/iommu setup */
1131 		if (ppc_md.pci_dma_dev_setup)
1132 			ppc_md.pci_dma_dev_setup(dev);
1133 
1134 		/* Read default IRQs and fixup if necessary */
1135 		pci_read_irq_line(dev);
1136 		if (ppc_md.pci_irq_fixup)
1137 			ppc_md.pci_irq_fixup(dev);
1138 	}
1139 }
1140 
1141 void __devinit pcibios_fixup_bus(struct pci_bus *bus)
1142 {
1143 	/* When called from the generic PCI probe, read PCI<->PCI bridge
1144 	 * bases. This is -not- called when generating the PCI tree from
1145 	 * the OF device-tree.
1146 	 */
1147 	if (bus->self != NULL)
1148 		pci_read_bridge_bases(bus);
1149 
1150 	/* Now fixup the bus bus */
1151 	pcibios_setup_bus_self(bus);
1152 
1153 	/* Now fixup devices on that bus */
1154 	pcibios_setup_bus_devices(bus);
1155 }
1156 EXPORT_SYMBOL(pcibios_fixup_bus);
1157 
1158 static int skip_isa_ioresource_align(struct pci_dev *dev)
1159 {
1160 	if ((ppc_pci_flags & PPC_PCI_CAN_SKIP_ISA_ALIGN) &&
1161 	    !(dev->bus->bridge_ctl & PCI_BRIDGE_CTL_ISA))
1162 		return 1;
1163 	return 0;
1164 }
1165 
1166 /*
1167  * We need to avoid collisions with `mirrored' VGA ports
1168  * and other strange ISA hardware, so we always want the
1169  * addresses to be allocated in the 0x000-0x0ff region
1170  * modulo 0x400.
1171  *
1172  * Why? Because some silly external IO cards only decode
1173  * the low 10 bits of the IO address. The 0x00-0xff region
1174  * is reserved for motherboard devices that decode all 16
1175  * bits, so it's ok to allocate at, say, 0x2800-0x28ff,
1176  * but we want to try to avoid allocating at 0x2900-0x2bff
1177  * which might have be mirrored at 0x0100-0x03ff..
1178  */
1179 void pcibios_align_resource(void *data, struct resource *res,
1180 				resource_size_t size, resource_size_t align)
1181 {
1182 	struct pci_dev *dev = data;
1183 
1184 	if (res->flags & IORESOURCE_IO) {
1185 		resource_size_t start = res->start;
1186 
1187 		if (skip_isa_ioresource_align(dev))
1188 			return;
1189 		if (start & 0x300) {
1190 			start = (start + 0x3ff) & ~0x3ff;
1191 			res->start = start;
1192 		}
1193 	}
1194 }
1195 EXPORT_SYMBOL(pcibios_align_resource);
1196 
1197 /*
1198  * Reparent resource children of pr that conflict with res
1199  * under res, and make res replace those children.
1200  */
1201 static int __init reparent_resources(struct resource *parent,
1202 				     struct resource *res)
1203 {
1204 	struct resource *p, **pp;
1205 	struct resource **firstpp = NULL;
1206 
1207 	for (pp = &parent->child; (p = *pp) != NULL; pp = &p->sibling) {
1208 		if (p->end < res->start)
1209 			continue;
1210 		if (res->end < p->start)
1211 			break;
1212 		if (p->start < res->start || p->end > res->end)
1213 			return -1;	/* not completely contained */
1214 		if (firstpp == NULL)
1215 			firstpp = pp;
1216 	}
1217 	if (firstpp == NULL)
1218 		return -1;	/* didn't find any conflicting entries? */
1219 	res->parent = parent;
1220 	res->child = *firstpp;
1221 	res->sibling = *pp;
1222 	*firstpp = res;
1223 	*pp = NULL;
1224 	for (p = res->child; p != NULL; p = p->sibling) {
1225 		p->parent = res;
1226 		pr_debug("PCI: Reparented %s [%llx..%llx] under %s\n",
1227 			 p->name,
1228 			 (unsigned long long)p->start,
1229 			 (unsigned long long)p->end, res->name);
1230 	}
1231 	return 0;
1232 }
1233 
1234 /*
1235  *  Handle resources of PCI devices.  If the world were perfect, we could
1236  *  just allocate all the resource regions and do nothing more.  It isn't.
1237  *  On the other hand, we cannot just re-allocate all devices, as it would
1238  *  require us to know lots of host bridge internals.  So we attempt to
1239  *  keep as much of the original configuration as possible, but tweak it
1240  *  when it's found to be wrong.
1241  *
1242  *  Known BIOS problems we have to work around:
1243  *	- I/O or memory regions not configured
1244  *	- regions configured, but not enabled in the command register
1245  *	- bogus I/O addresses above 64K used
1246  *	- expansion ROMs left enabled (this may sound harmless, but given
1247  *	  the fact the PCI specs explicitly allow address decoders to be
1248  *	  shared between expansion ROMs and other resource regions, it's
1249  *	  at least dangerous)
1250  *
1251  *  Our solution:
1252  *	(1) Allocate resources for all buses behind PCI-to-PCI bridges.
1253  *	    This gives us fixed barriers on where we can allocate.
1254  *	(2) Allocate resources for all enabled devices.  If there is
1255  *	    a collision, just mark the resource as unallocated. Also
1256  *	    disable expansion ROMs during this step.
1257  *	(3) Try to allocate resources for disabled devices.  If the
1258  *	    resources were assigned correctly, everything goes well,
1259  *	    if they weren't, they won't disturb allocation of other
1260  *	    resources.
1261  *	(4) Assign new addresses to resources which were either
1262  *	    not configured at all or misconfigured.  If explicitly
1263  *	    requested by the user, configure expansion ROM address
1264  *	    as well.
1265  */
1266 
1267 void pcibios_allocate_bus_resources(struct pci_bus *bus)
1268 {
1269 	struct pci_bus *b;
1270 	int i;
1271 	struct resource *res, *pr;
1272 
1273 	pr_debug("PCI: Allocating bus resources for %04x:%02x...\n",
1274 		 pci_domain_nr(bus), bus->number);
1275 
1276 	for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
1277 		if ((res = bus->resource[i]) == NULL || !res->flags
1278 		    || res->start > res->end || res->parent)
1279 			continue;
1280 		if (bus->parent == NULL)
1281 			pr = (res->flags & IORESOURCE_IO) ?
1282 				&ioport_resource : &iomem_resource;
1283 		else {
1284 			/* Don't bother with non-root busses when
1285 			 * re-assigning all resources. We clear the
1286 			 * resource flags as if they were colliding
1287 			 * and as such ensure proper re-allocation
1288 			 * later.
1289 			 */
1290 			if (ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)
1291 				goto clear_resource;
1292 			pr = pci_find_parent_resource(bus->self, res);
1293 			if (pr == res) {
1294 				/* this happens when the generic PCI
1295 				 * code (wrongly) decides that this
1296 				 * bridge is transparent  -- paulus
1297 				 */
1298 				continue;
1299 			}
1300 		}
1301 
1302 		pr_debug("PCI: %s (bus %d) bridge rsrc %d: %016llx-%016llx "
1303 			 "[0x%x], parent %p (%s)\n",
1304 			 bus->self ? pci_name(bus->self) : "PHB",
1305 			 bus->number, i,
1306 			 (unsigned long long)res->start,
1307 			 (unsigned long long)res->end,
1308 			 (unsigned int)res->flags,
1309 			 pr, (pr && pr->name) ? pr->name : "nil");
1310 
1311 		if (pr && !(pr->flags & IORESOURCE_UNSET)) {
1312 			if (request_resource(pr, res) == 0)
1313 				continue;
1314 			/*
1315 			 * Must be a conflict with an existing entry.
1316 			 * Move that entry (or entries) under the
1317 			 * bridge resource and try again.
1318 			 */
1319 			if (reparent_resources(pr, res) == 0)
1320 				continue;
1321 		}
1322 		printk(KERN_WARNING "PCI: Cannot allocate resource region "
1323 		       "%d of PCI bridge %d, will remap\n", i, bus->number);
1324 clear_resource:
1325 		res->flags = 0;
1326 	}
1327 
1328 	list_for_each_entry(b, &bus->children, node)
1329 		pcibios_allocate_bus_resources(b);
1330 }
1331 
1332 static inline void __devinit alloc_resource(struct pci_dev *dev, int idx)
1333 {
1334 	struct resource *pr, *r = &dev->resource[idx];
1335 
1336 	pr_debug("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1337 		 pci_name(dev), idx,
1338 		 (unsigned long long)r->start,
1339 		 (unsigned long long)r->end,
1340 		 (unsigned int)r->flags);
1341 
1342 	pr = pci_find_parent_resource(dev, r);
1343 	if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1344 	    request_resource(pr, r) < 0) {
1345 		printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1346 		       " of device %s, will remap\n", idx, pci_name(dev));
1347 		if (pr)
1348 			pr_debug("PCI:  parent is %p: %016llx-%016llx [%x]\n",
1349 				 pr,
1350 				 (unsigned long long)pr->start,
1351 				 (unsigned long long)pr->end,
1352 				 (unsigned int)pr->flags);
1353 		/* We'll assign a new address later */
1354 		r->flags |= IORESOURCE_UNSET;
1355 		r->end -= r->start;
1356 		r->start = 0;
1357 	}
1358 }
1359 
1360 static void __init pcibios_allocate_resources(int pass)
1361 {
1362 	struct pci_dev *dev = NULL;
1363 	int idx, disabled;
1364 	u16 command;
1365 	struct resource *r;
1366 
1367 	for_each_pci_dev(dev) {
1368 		pci_read_config_word(dev, PCI_COMMAND, &command);
1369 		for (idx = 0; idx <= PCI_ROM_RESOURCE; idx++) {
1370 			r = &dev->resource[idx];
1371 			if (r->parent)		/* Already allocated */
1372 				continue;
1373 			if (!r->flags || (r->flags & IORESOURCE_UNSET))
1374 				continue;	/* Not assigned at all */
1375 			/* We only allocate ROMs on pass 1 just in case they
1376 			 * have been screwed up by firmware
1377 			 */
1378 			if (idx == PCI_ROM_RESOURCE )
1379 				disabled = 1;
1380 			if (r->flags & IORESOURCE_IO)
1381 				disabled = !(command & PCI_COMMAND_IO);
1382 			else
1383 				disabled = !(command & PCI_COMMAND_MEMORY);
1384 			if (pass == disabled)
1385 				alloc_resource(dev, idx);
1386 		}
1387 		if (pass)
1388 			continue;
1389 		r = &dev->resource[PCI_ROM_RESOURCE];
1390 		if (r->flags) {
1391 			/* Turn the ROM off, leave the resource region,
1392 			 * but keep it unregistered.
1393 			 */
1394 			u32 reg;
1395 			pci_read_config_dword(dev, dev->rom_base_reg, &reg);
1396 			if (reg & PCI_ROM_ADDRESS_ENABLE) {
1397 				pr_debug("PCI: Switching off ROM of %s\n",
1398 					 pci_name(dev));
1399 				r->flags &= ~IORESOURCE_ROM_ENABLE;
1400 				pci_write_config_dword(dev, dev->rom_base_reg,
1401 						       reg & ~PCI_ROM_ADDRESS_ENABLE);
1402 			}
1403 		}
1404 	}
1405 }
1406 
1407 static void __init pcibios_reserve_legacy_regions(struct pci_bus *bus)
1408 {
1409 	struct pci_controller *hose = pci_bus_to_host(bus);
1410 	resource_size_t	offset;
1411 	struct resource *res, *pres;
1412 	int i;
1413 
1414 	pr_debug("Reserving legacy ranges for domain %04x\n", pci_domain_nr(bus));
1415 
1416 	/* Check for IO */
1417 	if (!(hose->io_resource.flags & IORESOURCE_IO))
1418 		goto no_io;
1419 	offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1420 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1421 	BUG_ON(res == NULL);
1422 	res->name = "Legacy IO";
1423 	res->flags = IORESOURCE_IO;
1424 	res->start = offset;
1425 	res->end = (offset + 0xfff) & 0xfffffffful;
1426 	pr_debug("Candidate legacy IO: %pR\n", res);
1427 	if (request_resource(&hose->io_resource, res)) {
1428 		printk(KERN_DEBUG
1429 		       "PCI %04x:%02x Cannot reserve Legacy IO %pR\n",
1430 		       pci_domain_nr(bus), bus->number, res);
1431 		kfree(res);
1432 	}
1433 
1434  no_io:
1435 	/* Check for memory */
1436 	offset = hose->pci_mem_offset;
1437 	pr_debug("hose mem offset: %016llx\n", (unsigned long long)offset);
1438 	for (i = 0; i < 3; i++) {
1439 		pres = &hose->mem_resources[i];
1440 		if (!(pres->flags & IORESOURCE_MEM))
1441 			continue;
1442 		pr_debug("hose mem res: %pR\n", pres);
1443 		if ((pres->start - offset) <= 0xa0000 &&
1444 		    (pres->end - offset) >= 0xbffff)
1445 			break;
1446 	}
1447 	if (i >= 3)
1448 		return;
1449 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1450 	BUG_ON(res == NULL);
1451 	res->name = "Legacy VGA memory";
1452 	res->flags = IORESOURCE_MEM;
1453 	res->start = 0xa0000 + offset;
1454 	res->end = 0xbffff + offset;
1455 	pr_debug("Candidate VGA memory: %pR\n", res);
1456 	if (request_resource(pres, res)) {
1457 		printk(KERN_DEBUG
1458 		       "PCI %04x:%02x Cannot reserve VGA memory %pR\n",
1459 		       pci_domain_nr(bus), bus->number, res);
1460 		kfree(res);
1461 	}
1462 }
1463 
1464 void __init pcibios_resource_survey(void)
1465 {
1466 	struct pci_bus *b;
1467 
1468 	/* Allocate and assign resources. If we re-assign everything, then
1469 	 * we skip the allocate phase
1470 	 */
1471 	list_for_each_entry(b, &pci_root_buses, node)
1472 		pcibios_allocate_bus_resources(b);
1473 
1474 	if (!(ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)) {
1475 		pcibios_allocate_resources(0);
1476 		pcibios_allocate_resources(1);
1477 	}
1478 
1479 	/* Before we start assigning unassigned resource, we try to reserve
1480 	 * the low IO area and the VGA memory area if they intersect the
1481 	 * bus available resources to avoid allocating things on top of them
1482 	 */
1483 	if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1484 		list_for_each_entry(b, &pci_root_buses, node)
1485 			pcibios_reserve_legacy_regions(b);
1486 	}
1487 
1488 	/* Now, if the platform didn't decide to blindly trust the firmware,
1489 	 * we proceed to assigning things that were left unassigned
1490 	 */
1491 	if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1492 		pr_debug("PCI: Assigning unassigned resources...\n");
1493 		pci_assign_unassigned_resources();
1494 	}
1495 
1496 	/* Call machine dependent fixup */
1497 	if (ppc_md.pcibios_fixup)
1498 		ppc_md.pcibios_fixup();
1499 }
1500 
1501 #ifdef CONFIG_HOTPLUG
1502 
1503 /* This is used by the PCI hotplug driver to allocate resource
1504  * of newly plugged busses. We can try to consolidate with the
1505  * rest of the code later, for now, keep it as-is as our main
1506  * resource allocation function doesn't deal with sub-trees yet.
1507  */
1508 void __devinit pcibios_claim_one_bus(struct pci_bus *bus)
1509 {
1510 	struct pci_dev *dev;
1511 	struct pci_bus *child_bus;
1512 
1513 	list_for_each_entry(dev, &bus->devices, bus_list) {
1514 		int i;
1515 
1516 		for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1517 			struct resource *r = &dev->resource[i];
1518 
1519 			if (r->parent || !r->start || !r->flags)
1520 				continue;
1521 
1522 			pr_debug("PCI: Claiming %s: "
1523 				 "Resource %d: %016llx..%016llx [%x]\n",
1524 				 pci_name(dev), i,
1525 				 (unsigned long long)r->start,
1526 				 (unsigned long long)r->end,
1527 				 (unsigned int)r->flags);
1528 
1529 			pci_claim_resource(dev, i);
1530 		}
1531 	}
1532 
1533 	list_for_each_entry(child_bus, &bus->children, node)
1534 		pcibios_claim_one_bus(child_bus);
1535 }
1536 EXPORT_SYMBOL_GPL(pcibios_claim_one_bus);
1537 
1538 
1539 /* pcibios_finish_adding_to_bus
1540  *
1541  * This is to be called by the hotplug code after devices have been
1542  * added to a bus, this include calling it for a PHB that is just
1543  * being added
1544  */
1545 void pcibios_finish_adding_to_bus(struct pci_bus *bus)
1546 {
1547 	pr_debug("PCI: Finishing adding to hotplug bus %04x:%02x\n",
1548 		 pci_domain_nr(bus), bus->number);
1549 
1550 	/* Allocate bus and devices resources */
1551 	pcibios_allocate_bus_resources(bus);
1552 	pcibios_claim_one_bus(bus);
1553 
1554 	/* Add new devices to global lists.  Register in proc, sysfs. */
1555 	pci_bus_add_devices(bus);
1556 
1557 	/* Fixup EEH */
1558 	eeh_add_device_tree_late(bus);
1559 }
1560 EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
1561 
1562 #endif /* CONFIG_HOTPLUG */
1563 
1564 int pcibios_enable_device(struct pci_dev *dev, int mask)
1565 {
1566 	if (ppc_md.pcibios_enable_device_hook)
1567 		if (ppc_md.pcibios_enable_device_hook(dev))
1568 			return -EINVAL;
1569 
1570 	return pci_enable_resources(dev, mask);
1571 }
1572 
1573 void __devinit pcibios_setup_phb_resources(struct pci_controller *hose)
1574 {
1575 	struct pci_bus *bus = hose->bus;
1576 	struct resource *res;
1577 	int i;
1578 
1579 	/* Hookup PHB IO resource */
1580 	bus->resource[0] = res = &hose->io_resource;
1581 
1582 	if (!res->flags) {
1583 		printk(KERN_WARNING "PCI: I/O resource not set for host"
1584 		       " bridge %s (domain %d)\n",
1585 		       hose->dn->full_name, hose->global_number);
1586 #ifdef CONFIG_PPC32
1587 		/* Workaround for lack of IO resource only on 32-bit */
1588 		res->start = (unsigned long)hose->io_base_virt - isa_io_base;
1589 		res->end = res->start + IO_SPACE_LIMIT;
1590 		res->flags = IORESOURCE_IO;
1591 #endif /* CONFIG_PPC32 */
1592 	}
1593 
1594 	pr_debug("PCI: PHB IO resource    = %016llx-%016llx [%lx]\n",
1595 		 (unsigned long long)res->start,
1596 		 (unsigned long long)res->end,
1597 		 (unsigned long)res->flags);
1598 
1599 	/* Hookup PHB Memory resources */
1600 	for (i = 0; i < 3; ++i) {
1601 		res = &hose->mem_resources[i];
1602 		if (!res->flags) {
1603 			if (i > 0)
1604 				continue;
1605 			printk(KERN_ERR "PCI: Memory resource 0 not set for "
1606 			       "host bridge %s (domain %d)\n",
1607 			       hose->dn->full_name, hose->global_number);
1608 #ifdef CONFIG_PPC32
1609 			/* Workaround for lack of MEM resource only on 32-bit */
1610 			res->start = hose->pci_mem_offset;
1611 			res->end = (resource_size_t)-1LL;
1612 			res->flags = IORESOURCE_MEM;
1613 #endif /* CONFIG_PPC32 */
1614 		}
1615 		bus->resource[i+1] = res;
1616 
1617 		pr_debug("PCI: PHB MEM resource %d = %016llx-%016llx [%lx]\n", i,
1618 			 (unsigned long long)res->start,
1619 			 (unsigned long long)res->end,
1620 			 (unsigned long)res->flags);
1621 	}
1622 
1623 	pr_debug("PCI: PHB MEM offset     = %016llx\n",
1624 		 (unsigned long long)hose->pci_mem_offset);
1625 	pr_debug("PCI: PHB IO  offset     = %08lx\n",
1626 		 (unsigned long)hose->io_base_virt - _IO_BASE);
1627 
1628 }
1629 
1630