xref: /openbmc/linux/arch/powerpc/kernel/pci-common.c (revision 4fc665b88a79a45bae8bbf3a05563c27c7337c3d)
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 #undef DEBUG
20 
21 #include <linux/kernel.h>
22 #include <linux/pci.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/bootmem.h>
26 #include <linux/mm.h>
27 #include <linux/list.h>
28 #include <linux/syscalls.h>
29 #include <linux/irq.h>
30 #include <linux/vmalloc.h>
31 
32 #include <asm/processor.h>
33 #include <asm/io.h>
34 #include <asm/prom.h>
35 #include <asm/pci-bridge.h>
36 #include <asm/byteorder.h>
37 #include <asm/machdep.h>
38 #include <asm/ppc-pci.h>
39 #include <asm/firmware.h>
40 
41 #ifdef DEBUG
42 #include <asm/udbg.h>
43 #define DBG(fmt...) printk(fmt)
44 #else
45 #define DBG(fmt...)
46 #endif
47 
48 static DEFINE_SPINLOCK(hose_spinlock);
49 
50 /* XXX kill that some day ... */
51 static int global_phb_number;		/* Global phb counter */
52 
53 /* ISA Memory physical address */
54 resource_size_t isa_mem_base;
55 
56 /* Default PCI flags is 0 */
57 unsigned int ppc_pci_flags;
58 
59 static struct dma_mapping_ops *pci_dma_ops;
60 
61 void set_pci_dma_ops(struct dma_mapping_ops *dma_ops)
62 {
63 	pci_dma_ops = dma_ops;
64 }
65 
66 struct dma_mapping_ops *get_pci_dma_ops(void)
67 {
68 	return pci_dma_ops;
69 }
70 EXPORT_SYMBOL(get_pci_dma_ops);
71 
72 int pci_set_dma_mask(struct pci_dev *dev, u64 mask)
73 {
74 	return dma_set_mask(&dev->dev, mask);
75 }
76 
77 int pci_set_consistent_dma_mask(struct pci_dev *dev, u64 mask)
78 {
79 	int rc;
80 
81 	rc = dma_set_mask(&dev->dev, mask);
82 	dev->dev.coherent_dma_mask = dev->dma_mask;
83 
84 	return rc;
85 }
86 
87 struct pci_controller *pcibios_alloc_controller(struct device_node *dev)
88 {
89 	struct pci_controller *phb;
90 
91 	phb = zalloc_maybe_bootmem(sizeof(struct pci_controller), GFP_KERNEL);
92 	if (phb == NULL)
93 		return NULL;
94 	spin_lock(&hose_spinlock);
95 	phb->global_number = global_phb_number++;
96 	list_add_tail(&phb->list_node, &hose_list);
97 	spin_unlock(&hose_spinlock);
98 	phb->dn = dev;
99 	phb->is_dynamic = mem_init_done;
100 #ifdef CONFIG_PPC64
101 	if (dev) {
102 		int nid = of_node_to_nid(dev);
103 
104 		if (nid < 0 || !node_online(nid))
105 			nid = -1;
106 
107 		PHB_SET_NODE(phb, nid);
108 	}
109 #endif
110 	return phb;
111 }
112 
113 void pcibios_free_controller(struct pci_controller *phb)
114 {
115 	spin_lock(&hose_spinlock);
116 	list_del(&phb->list_node);
117 	spin_unlock(&hose_spinlock);
118 
119 	if (phb->is_dynamic)
120 		kfree(phb);
121 }
122 
123 int pcibios_vaddr_is_ioport(void __iomem *address)
124 {
125 	int ret = 0;
126 	struct pci_controller *hose;
127 	unsigned long size;
128 
129 	spin_lock(&hose_spinlock);
130 	list_for_each_entry(hose, &hose_list, list_node) {
131 #ifdef CONFIG_PPC64
132 		size = hose->pci_io_size;
133 #else
134 		size = hose->io_resource.end - hose->io_resource.start + 1;
135 #endif
136 		if (address >= hose->io_base_virt &&
137 		    address < (hose->io_base_virt + size)) {
138 			ret = 1;
139 			break;
140 		}
141 	}
142 	spin_unlock(&hose_spinlock);
143 	return ret;
144 }
145 
146 /*
147  * Return the domain number for this bus.
148  */
149 int pci_domain_nr(struct pci_bus *bus)
150 {
151 	struct pci_controller *hose = pci_bus_to_host(bus);
152 
153 	return hose->global_number;
154 }
155 EXPORT_SYMBOL(pci_domain_nr);
156 
157 #ifdef CONFIG_PPC_OF
158 
159 /* This routine is meant to be used early during boot, when the
160  * PCI bus numbers have not yet been assigned, and you need to
161  * issue PCI config cycles to an OF device.
162  * It could also be used to "fix" RTAS config cycles if you want
163  * to set pci_assign_all_buses to 1 and still use RTAS for PCI
164  * config cycles.
165  */
166 struct pci_controller* pci_find_hose_for_OF_device(struct device_node* node)
167 {
168 	if (!have_of)
169 		return NULL;
170 	while(node) {
171 		struct pci_controller *hose, *tmp;
172 		list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
173 			if (hose->dn == node)
174 				return hose;
175 		node = node->parent;
176 	}
177 	return NULL;
178 }
179 
180 static ssize_t pci_show_devspec(struct device *dev,
181 		struct device_attribute *attr, char *buf)
182 {
183 	struct pci_dev *pdev;
184 	struct device_node *np;
185 
186 	pdev = to_pci_dev (dev);
187 	np = pci_device_to_OF_node(pdev);
188 	if (np == NULL || np->full_name == NULL)
189 		return 0;
190 	return sprintf(buf, "%s", np->full_name);
191 }
192 static DEVICE_ATTR(devspec, S_IRUGO, pci_show_devspec, NULL);
193 #endif /* CONFIG_PPC_OF */
194 
195 /* Add sysfs properties */
196 int pcibios_add_platform_entries(struct pci_dev *pdev)
197 {
198 #ifdef CONFIG_PPC_OF
199 	return device_create_file(&pdev->dev, &dev_attr_devspec);
200 #else
201 	return 0;
202 #endif /* CONFIG_PPC_OF */
203 
204 }
205 
206 char __devinit *pcibios_setup(char *str)
207 {
208 	return str;
209 }
210 
211 void __devinit pcibios_setup_new_device(struct pci_dev *dev)
212 {
213 	struct dev_archdata *sd = &dev->dev.archdata;
214 
215 	sd->of_node = pci_device_to_OF_node(dev);
216 
217 	DBG("PCI: device %s OF node: %s\n", pci_name(dev),
218 	    sd->of_node ? sd->of_node->full_name : "<none>");
219 
220 	sd->dma_ops = pci_dma_ops;
221 #ifdef CONFIG_PPC32
222 	sd->dma_data = (void *)PCI_DRAM_OFFSET;
223 #endif
224 	set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
225 
226 	if (ppc_md.pci_dma_dev_setup)
227 		ppc_md.pci_dma_dev_setup(dev);
228 }
229 EXPORT_SYMBOL(pcibios_setup_new_device);
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 	DBG("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 		DBG(" -> 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 		DBG(" -> got one, spec %d cells (0x%08x 0x%08x...) on %s\n",
287 		    oirq.size, oirq.specifier[0], oirq.specifier[1],
288 		    oirq.controller->full_name);
289 
290 		virq = irq_create_of_mapping(oirq.controller, oirq.specifier,
291 					     oirq.size);
292 	}
293 	if(virq == NO_IRQ) {
294 		DBG(" -> failed to map !\n");
295 		return -1;
296 	}
297 
298 	DBG(" -> mapped to linux irq %d\n", virq);
299 
300 	pci_dev->irq = virq;
301 
302 	return 0;
303 }
304 EXPORT_SYMBOL(pci_read_irq_line);
305 
306 /*
307  * Platform support for /proc/bus/pci/X/Y mmap()s,
308  * modelled on the sparc64 implementation by Dave Miller.
309  *  -- paulus.
310  */
311 
312 /*
313  * Adjust vm_pgoff of VMA such that it is the physical page offset
314  * corresponding to the 32-bit pci bus offset for DEV requested by the user.
315  *
316  * Basically, the user finds the base address for his device which he wishes
317  * to mmap.  They read the 32-bit value from the config space base register,
318  * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
319  * offset parameter of mmap on /proc/bus/pci/XXX for that device.
320  *
321  * Returns negative error code on failure, zero on success.
322  */
323 static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
324 					       resource_size_t *offset,
325 					       enum pci_mmap_state mmap_state)
326 {
327 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
328 	unsigned long io_offset = 0;
329 	int i, res_bit;
330 
331 	if (hose == 0)
332 		return NULL;		/* should never happen */
333 
334 	/* If memory, add on the PCI bridge address offset */
335 	if (mmap_state == pci_mmap_mem) {
336 #if 0 /* See comment in pci_resource_to_user() for why this is disabled */
337 		*offset += hose->pci_mem_offset;
338 #endif
339 		res_bit = IORESOURCE_MEM;
340 	} else {
341 		io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
342 		*offset += io_offset;
343 		res_bit = IORESOURCE_IO;
344 	}
345 
346 	/*
347 	 * Check that the offset requested corresponds to one of the
348 	 * resources of the device.
349 	 */
350 	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
351 		struct resource *rp = &dev->resource[i];
352 		int flags = rp->flags;
353 
354 		/* treat ROM as memory (should be already) */
355 		if (i == PCI_ROM_RESOURCE)
356 			flags |= IORESOURCE_MEM;
357 
358 		/* Active and same type? */
359 		if ((flags & res_bit) == 0)
360 			continue;
361 
362 		/* In the range of this resource? */
363 		if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
364 			continue;
365 
366 		/* found it! construct the final physical address */
367 		if (mmap_state == pci_mmap_io)
368 			*offset += hose->io_base_phys - io_offset;
369 		return rp;
370 	}
371 
372 	return NULL;
373 }
374 
375 /*
376  * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
377  * device mapping.
378  */
379 static pgprot_t __pci_mmap_set_pgprot(struct pci_dev *dev, struct resource *rp,
380 				      pgprot_t protection,
381 				      enum pci_mmap_state mmap_state,
382 				      int write_combine)
383 {
384 	unsigned long prot = pgprot_val(protection);
385 
386 	/* Write combine is always 0 on non-memory space mappings. On
387 	 * memory space, if the user didn't pass 1, we check for a
388 	 * "prefetchable" resource. This is a bit hackish, but we use
389 	 * this to workaround the inability of /sysfs to provide a write
390 	 * combine bit
391 	 */
392 	if (mmap_state != pci_mmap_mem)
393 		write_combine = 0;
394 	else if (write_combine == 0) {
395 		if (rp->flags & IORESOURCE_PREFETCH)
396 			write_combine = 1;
397 	}
398 
399 	/* XXX would be nice to have a way to ask for write-through */
400 	prot |= _PAGE_NO_CACHE;
401 	if (write_combine)
402 		prot &= ~_PAGE_GUARDED;
403 	else
404 		prot |= _PAGE_GUARDED;
405 
406 	return __pgprot(prot);
407 }
408 
409 /*
410  * This one is used by /dev/mem and fbdev who have no clue about the
411  * PCI device, it tries to find the PCI device first and calls the
412  * above routine
413  */
414 pgprot_t pci_phys_mem_access_prot(struct file *file,
415 				  unsigned long pfn,
416 				  unsigned long size,
417 				  pgprot_t protection)
418 {
419 	struct pci_dev *pdev = NULL;
420 	struct resource *found = NULL;
421 	unsigned long prot = pgprot_val(protection);
422 	unsigned long offset = pfn << PAGE_SHIFT;
423 	int i;
424 
425 	if (page_is_ram(pfn))
426 		return __pgprot(prot);
427 
428 	prot |= _PAGE_NO_CACHE | _PAGE_GUARDED;
429 
430 	for_each_pci_dev(pdev) {
431 		for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
432 			struct resource *rp = &pdev->resource[i];
433 			int flags = rp->flags;
434 
435 			/* Active and same type? */
436 			if ((flags & IORESOURCE_MEM) == 0)
437 				continue;
438 			/* In the range of this resource? */
439 			if (offset < (rp->start & PAGE_MASK) ||
440 			    offset > rp->end)
441 				continue;
442 			found = rp;
443 			break;
444 		}
445 		if (found)
446 			break;
447 	}
448 	if (found) {
449 		if (found->flags & IORESOURCE_PREFETCH)
450 			prot &= ~_PAGE_GUARDED;
451 		pci_dev_put(pdev);
452 	}
453 
454 	DBG("non-PCI map for %lx, prot: %lx\n", offset, prot);
455 
456 	return __pgprot(prot);
457 }
458 
459 
460 /*
461  * Perform the actual remap of the pages for a PCI device mapping, as
462  * appropriate for this architecture.  The region in the process to map
463  * is described by vm_start and vm_end members of VMA, the base physical
464  * address is found in vm_pgoff.
465  * The pci device structure is provided so that architectures may make mapping
466  * decisions on a per-device or per-bus basis.
467  *
468  * Returns a negative error code on failure, zero on success.
469  */
470 int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
471 			enum pci_mmap_state mmap_state, int write_combine)
472 {
473 	resource_size_t offset = vma->vm_pgoff << PAGE_SHIFT;
474 	struct resource *rp;
475 	int ret;
476 
477 	rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
478 	if (rp == NULL)
479 		return -EINVAL;
480 
481 	vma->vm_pgoff = offset >> PAGE_SHIFT;
482 	vma->vm_page_prot = __pci_mmap_set_pgprot(dev, rp,
483 						  vma->vm_page_prot,
484 						  mmap_state, write_combine);
485 
486 	ret = remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
487 			       vma->vm_end - vma->vm_start, vma->vm_page_prot);
488 
489 	return ret;
490 }
491 
492 void pci_resource_to_user(const struct pci_dev *dev, int bar,
493 			  const struct resource *rsrc,
494 			  resource_size_t *start, resource_size_t *end)
495 {
496 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
497 	resource_size_t offset = 0;
498 
499 	if (hose == NULL)
500 		return;
501 
502 	if (rsrc->flags & IORESOURCE_IO)
503 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
504 
505 	/* We pass a fully fixed up address to userland for MMIO instead of
506 	 * a BAR value because X is lame and expects to be able to use that
507 	 * to pass to /dev/mem !
508 	 *
509 	 * That means that we'll have potentially 64 bits values where some
510 	 * userland apps only expect 32 (like X itself since it thinks only
511 	 * Sparc has 64 bits MMIO) but if we don't do that, we break it on
512 	 * 32 bits CHRPs :-(
513 	 *
514 	 * Hopefully, the sysfs insterface is immune to that gunk. Once X
515 	 * has been fixed (and the fix spread enough), we can re-enable the
516 	 * 2 lines below and pass down a BAR value to userland. In that case
517 	 * we'll also have to re-enable the matching code in
518 	 * __pci_mmap_make_offset().
519 	 *
520 	 * BenH.
521 	 */
522 #if 0
523 	else if (rsrc->flags & IORESOURCE_MEM)
524 		offset = hose->pci_mem_offset;
525 #endif
526 
527 	*start = rsrc->start - offset;
528 	*end = rsrc->end - offset;
529 }
530 
531 /**
532  * pci_process_bridge_OF_ranges - Parse PCI bridge resources from device tree
533  * @hose: newly allocated pci_controller to be setup
534  * @dev: device node of the host bridge
535  * @primary: set if primary bus (32 bits only, soon to be deprecated)
536  *
537  * This function will parse the "ranges" property of a PCI host bridge device
538  * node and setup the resource mapping of a pci controller based on its
539  * content.
540  *
541  * Life would be boring if it wasn't for a few issues that we have to deal
542  * with here:
543  *
544  *   - We can only cope with one IO space range and up to 3 Memory space
545  *     ranges. However, some machines (thanks Apple !) tend to split their
546  *     space into lots of small contiguous ranges. So we have to coalesce.
547  *
548  *   - We can only cope with all memory ranges having the same offset
549  *     between CPU addresses and PCI addresses. Unfortunately, some bridges
550  *     are setup for a large 1:1 mapping along with a small "window" which
551  *     maps PCI address 0 to some arbitrary high address of the CPU space in
552  *     order to give access to the ISA memory hole.
553  *     The way out of here that I've chosen for now is to always set the
554  *     offset based on the first resource found, then override it if we
555  *     have a different offset and the previous was set by an ISA hole.
556  *
557  *   - Some busses have IO space not starting at 0, which causes trouble with
558  *     the way we do our IO resource renumbering. The code somewhat deals with
559  *     it for 64 bits but I would expect problems on 32 bits.
560  *
561  *   - Some 32 bits platforms such as 4xx can have physical space larger than
562  *     32 bits so we need to use 64 bits values for the parsing
563  */
564 void __devinit pci_process_bridge_OF_ranges(struct pci_controller *hose,
565 					    struct device_node *dev,
566 					    int primary)
567 {
568 	const u32 *ranges;
569 	int rlen;
570 	int pna = of_n_addr_cells(dev);
571 	int np = pna + 5;
572 	int memno = 0, isa_hole = -1;
573 	u32 pci_space;
574 	unsigned long long pci_addr, cpu_addr, pci_next, cpu_next, size;
575 	unsigned long long isa_mb = 0;
576 	struct resource *res;
577 
578 	printk(KERN_INFO "PCI host bridge %s %s ranges:\n",
579 	       dev->full_name, primary ? "(primary)" : "");
580 
581 	/* Get ranges property */
582 	ranges = of_get_property(dev, "ranges", &rlen);
583 	if (ranges == NULL)
584 		return;
585 
586 	/* Parse it */
587 	while ((rlen -= np * 4) >= 0) {
588 		/* Read next ranges element */
589 		pci_space = ranges[0];
590 		pci_addr = of_read_number(ranges + 1, 2);
591 		cpu_addr = of_translate_address(dev, ranges + 3);
592 		size = of_read_number(ranges + pna + 3, 2);
593 		ranges += np;
594 		if (cpu_addr == OF_BAD_ADDR || size == 0)
595 			continue;
596 
597 		/* Now consume following elements while they are contiguous */
598 		for (; rlen >= np * sizeof(u32);
599 		     ranges += np, rlen -= np * 4) {
600 			if (ranges[0] != pci_space)
601 				break;
602 			pci_next = of_read_number(ranges + 1, 2);
603 			cpu_next = of_translate_address(dev, ranges + 3);
604 			if (pci_next != pci_addr + size ||
605 			    cpu_next != cpu_addr + size)
606 				break;
607 			size += of_read_number(ranges + pna + 3, 2);
608 		}
609 
610 		/* Act based on address space type */
611 		res = NULL;
612 		switch ((pci_space >> 24) & 0x3) {
613 		case 1:		/* PCI IO space */
614 			printk(KERN_INFO
615 			       "  IO 0x%016llx..0x%016llx -> 0x%016llx\n",
616 			       cpu_addr, cpu_addr + size - 1, pci_addr);
617 
618 			/* We support only one IO range */
619 			if (hose->pci_io_size) {
620 				printk(KERN_INFO
621 				       " \\--> Skipped (too many) !\n");
622 				continue;
623 			}
624 #ifdef CONFIG_PPC32
625 			/* On 32 bits, limit I/O space to 16MB */
626 			if (size > 0x01000000)
627 				size = 0x01000000;
628 
629 			/* 32 bits needs to map IOs here */
630 			hose->io_base_virt = ioremap(cpu_addr, size);
631 
632 			/* Expect trouble if pci_addr is not 0 */
633 			if (primary)
634 				isa_io_base =
635 					(unsigned long)hose->io_base_virt;
636 #endif /* CONFIG_PPC32 */
637 			/* pci_io_size and io_base_phys always represent IO
638 			 * space starting at 0 so we factor in pci_addr
639 			 */
640 			hose->pci_io_size = pci_addr + size;
641 			hose->io_base_phys = cpu_addr - pci_addr;
642 
643 			/* Build resource */
644 			res = &hose->io_resource;
645 			res->flags = IORESOURCE_IO;
646 			res->start = pci_addr;
647 			break;
648 		case 2:		/* PCI Memory space */
649 		case 3:		/* PCI 64 bits Memory space */
650 			printk(KERN_INFO
651 			       " MEM 0x%016llx..0x%016llx -> 0x%016llx %s\n",
652 			       cpu_addr, cpu_addr + size - 1, pci_addr,
653 			       (pci_space & 0x40000000) ? "Prefetch" : "");
654 
655 			/* We support only 3 memory ranges */
656 			if (memno >= 3) {
657 				printk(KERN_INFO
658 				       " \\--> Skipped (too many) !\n");
659 				continue;
660 			}
661 			/* Handles ISA memory hole space here */
662 			if (pci_addr == 0) {
663 				isa_mb = cpu_addr;
664 				isa_hole = memno;
665 				if (primary || isa_mem_base == 0)
666 					isa_mem_base = cpu_addr;
667 			}
668 
669 			/* We get the PCI/Mem offset from the first range or
670 			 * the, current one if the offset came from an ISA
671 			 * hole. If they don't match, bugger.
672 			 */
673 			if (memno == 0 ||
674 			    (isa_hole >= 0 && pci_addr != 0 &&
675 			     hose->pci_mem_offset == isa_mb))
676 				hose->pci_mem_offset = cpu_addr - pci_addr;
677 			else if (pci_addr != 0 &&
678 				 hose->pci_mem_offset != cpu_addr - pci_addr) {
679 				printk(KERN_INFO
680 				       " \\--> Skipped (offset mismatch) !\n");
681 				continue;
682 			}
683 
684 			/* Build resource */
685 			res = &hose->mem_resources[memno++];
686 			res->flags = IORESOURCE_MEM;
687 			if (pci_space & 0x40000000)
688 				res->flags |= IORESOURCE_PREFETCH;
689 			res->start = cpu_addr;
690 			break;
691 		}
692 		if (res != NULL) {
693 			res->name = dev->full_name;
694 			res->end = res->start + size - 1;
695 			res->parent = NULL;
696 			res->sibling = NULL;
697 			res->child = NULL;
698 		}
699 	}
700 
701 	/* If there's an ISA hole and the pci_mem_offset is -not- matching
702 	 * the ISA hole offset, then we need to remove the ISA hole from
703 	 * the resource list for that brige
704 	 */
705 	if (isa_hole >= 0 && hose->pci_mem_offset != isa_mb) {
706 		unsigned int next = isa_hole + 1;
707 		printk(KERN_INFO " Removing ISA hole at 0x%016llx\n", isa_mb);
708 		if (next < memno)
709 			memmove(&hose->mem_resources[isa_hole],
710 				&hose->mem_resources[next],
711 				sizeof(struct resource) * (memno - next));
712 		hose->mem_resources[--memno].flags = 0;
713 	}
714 }
715 
716 /* Decide whether to display the domain number in /proc */
717 int pci_proc_domain(struct pci_bus *bus)
718 {
719 	struct pci_controller *hose = pci_bus_to_host(bus);
720 #ifdef CONFIG_PPC64
721 	return hose->buid != 0;
722 #else
723 	if (!(ppc_pci_flags & PPC_PCI_ENABLE_PROC_DOMAINS))
724 		return 0;
725 	if (ppc_pci_flags & PPC_PCI_COMPAT_DOMAIN_0)
726 		return hose->global_number != 0;
727 	return 1;
728 #endif
729 }
730 
731 void pcibios_resource_to_bus(struct pci_dev *dev, struct pci_bus_region *region,
732 			     struct resource *res)
733 {
734 	resource_size_t offset = 0, mask = (resource_size_t)-1;
735 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
736 
737 	if (!hose)
738 		return;
739 	if (res->flags & IORESOURCE_IO) {
740 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
741 		mask = 0xffffffffu;
742 	} else if (res->flags & IORESOURCE_MEM)
743 		offset = hose->pci_mem_offset;
744 
745 	region->start = (res->start - offset) & mask;
746 	region->end = (res->end - offset) & mask;
747 }
748 EXPORT_SYMBOL(pcibios_resource_to_bus);
749 
750 void pcibios_bus_to_resource(struct pci_dev *dev, struct resource *res,
751 			     struct pci_bus_region *region)
752 {
753 	resource_size_t offset = 0, mask = (resource_size_t)-1;
754 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
755 
756 	if (!hose)
757 		return;
758 	if (res->flags & IORESOURCE_IO) {
759 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
760 		mask = 0xffffffffu;
761 	} else if (res->flags & IORESOURCE_MEM)
762 		offset = hose->pci_mem_offset;
763 	res->start = (region->start + offset) & mask;
764 	res->end = (region->end + offset) & mask;
765 }
766 EXPORT_SYMBOL(pcibios_bus_to_resource);
767 
768 /* Fixup a bus resource into a linux resource */
769 static void __devinit fixup_resource(struct resource *res, struct pci_dev *dev)
770 {
771 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
772 	resource_size_t offset = 0, mask = (resource_size_t)-1;
773 
774 	if (res->flags & IORESOURCE_IO) {
775 		offset = (unsigned long)hose->io_base_virt - _IO_BASE;
776 		mask = 0xffffffffu;
777 	} else if (res->flags & IORESOURCE_MEM)
778 		offset = hose->pci_mem_offset;
779 
780 	res->start = (res->start + offset) & mask;
781 	res->end = (res->end + offset) & mask;
782 
783 	pr_debug("PCI:%s            %016llx-%016llx\n",
784 		 pci_name(dev),
785 		 (unsigned long long)res->start,
786 		 (unsigned long long)res->end);
787 }
788 
789 
790 /* This header fixup will do the resource fixup for all devices as they are
791  * probed, but not for bridge ranges
792  */
793 static void __devinit pcibios_fixup_resources(struct pci_dev *dev)
794 {
795 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
796 	int i;
797 
798 	if (!hose) {
799 		printk(KERN_ERR "No host bridge for PCI dev %s !\n",
800 		       pci_name(dev));
801 		return;
802 	}
803 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
804 		struct resource *res = dev->resource + i;
805 		if (!res->flags)
806 			continue;
807 		/* On platforms that have PPC_PCI_PROBE_ONLY set, we don't
808 		 * consider 0 as an unassigned BAR value. It's technically
809 		 * a valid value, but linux doesn't like it... so when we can
810 		 * re-assign things, we do so, but if we can't, we keep it
811 		 * around and hope for the best...
812 		 */
813 		if (res->start == 0 && !(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
814 			pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] is unassigned\n",
815 				 pci_name(dev), i,
816 				 (unsigned long long)res->start,
817 				 (unsigned long long)res->end,
818 				 (unsigned int)res->flags);
819 			res->end -= res->start;
820 			res->start = 0;
821 			res->flags |= IORESOURCE_UNSET;
822 			continue;
823 		}
824 
825 		pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] fixup...\n",
826 			 pci_name(dev), i,
827 			 (unsigned long long)res->start,\
828 			 (unsigned long long)res->end,
829 			 (unsigned int)res->flags);
830 
831 		fixup_resource(res, dev);
832 	}
833 
834 	/* Call machine specific resource fixup */
835 	if (ppc_md.pcibios_fixup_resources)
836 		ppc_md.pcibios_fixup_resources(dev);
837 }
838 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_resources);
839 
840 static void __devinit __pcibios_fixup_bus(struct pci_bus *bus)
841 {
842 	struct pci_controller *hose = pci_bus_to_host(bus);
843 	struct pci_dev *dev = bus->self;
844 
845 	pr_debug("PCI: Fixup bus %d (%s)\n", bus->number, dev ? pci_name(dev) : "PHB");
846 
847 	/* Fixup PCI<->PCI bridges. Host bridges are handled separately, for
848 	 * now differently between 32 and 64 bits.
849 	 */
850 	if (dev != NULL) {
851 		struct resource *res;
852 		int i;
853 
854 		for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
855 			if ((res = bus->resource[i]) == NULL)
856 				continue;
857 			if (!res->flags)
858 				continue;
859 			if (i >= 3 && bus->self->transparent)
860 				continue;
861 			/* On PowerMac, Apple leaves bridge windows open over
862 			 * an inaccessible region of memory space (0...fffff)
863 			 * which is somewhat bogus, but that's what they think
864 			 * means disabled...
865 			 *
866 			 * We clear those to force them to be reallocated later
867 			 *
868 			 * We detect such regions by the fact that the base is
869 			 * equal to the pci_mem_offset of the host bridge and
870 			 * their size is smaller than 1M.
871 			 */
872 			if (res->flags & IORESOURCE_MEM &&
873 			    res->start == hose->pci_mem_offset &&
874 			    res->end < 0x100000) {
875 				printk(KERN_INFO
876 				       "PCI: Closing bogus Apple Firmware"
877 				       " region %d on bus 0x%02x\n",
878 				       i, bus->number);
879 				res->flags = 0;
880 				continue;
881 			}
882 
883 			pr_debug("PCI:%s Bus rsrc %d %016llx-%016llx [%x] fixup...\n",
884 				 pci_name(dev), i,
885 				 (unsigned long long)res->start,\
886 				 (unsigned long long)res->end,
887 				 (unsigned int)res->flags);
888 
889 			fixup_resource(res, dev);
890 		}
891 	}
892 
893 	/* Additional setup that is different between 32 and 64 bits for now */
894 	pcibios_do_bus_setup(bus);
895 
896 	/* Platform specific bus fixups */
897 	if (ppc_md.pcibios_fixup_bus)
898 		ppc_md.pcibios_fixup_bus(bus);
899 
900 	/* Read default IRQs and fixup if necessary */
901 	list_for_each_entry(dev, &bus->devices, bus_list) {
902 		pci_read_irq_line(dev);
903 		if (ppc_md.pci_irq_fixup)
904 			ppc_md.pci_irq_fixup(dev);
905 	}
906 }
907 
908 void __devinit pcibios_fixup_bus(struct pci_bus *bus)
909 {
910 	/* When called from the generic PCI probe, read PCI<->PCI bridge
911 	 * bases before proceeding
912 	 */
913 	if (bus->self != NULL)
914 		pci_read_bridge_bases(bus);
915 	__pcibios_fixup_bus(bus);
916 }
917 EXPORT_SYMBOL(pcibios_fixup_bus);
918 
919 /* When building a bus from the OF tree rather than probing, we need a
920  * slightly different version of the fixup which doesn't read the
921  * bridge bases using config space accesses
922  */
923 void __devinit pcibios_fixup_of_probed_bus(struct pci_bus *bus)
924 {
925 	__pcibios_fixup_bus(bus);
926 }
927 
928 static int skip_isa_ioresource_align(struct pci_dev *dev)
929 {
930 	if ((ppc_pci_flags & PPC_PCI_CAN_SKIP_ISA_ALIGN) &&
931 	    !(dev->bus->bridge_ctl & PCI_BRIDGE_CTL_ISA))
932 		return 1;
933 	return 0;
934 }
935 
936 /*
937  * We need to avoid collisions with `mirrored' VGA ports
938  * and other strange ISA hardware, so we always want the
939  * addresses to be allocated in the 0x000-0x0ff region
940  * modulo 0x400.
941  *
942  * Why? Because some silly external IO cards only decode
943  * the low 10 bits of the IO address. The 0x00-0xff region
944  * is reserved for motherboard devices that decode all 16
945  * bits, so it's ok to allocate at, say, 0x2800-0x28ff,
946  * but we want to try to avoid allocating at 0x2900-0x2bff
947  * which might have be mirrored at 0x0100-0x03ff..
948  */
949 void pcibios_align_resource(void *data, struct resource *res,
950 				resource_size_t size, resource_size_t align)
951 {
952 	struct pci_dev *dev = data;
953 
954 	if (res->flags & IORESOURCE_IO) {
955 		resource_size_t start = res->start;
956 
957 		if (skip_isa_ioresource_align(dev))
958 			return;
959 		if (start & 0x300) {
960 			start = (start + 0x3ff) & ~0x3ff;
961 			res->start = start;
962 		}
963 	}
964 }
965 EXPORT_SYMBOL(pcibios_align_resource);
966 
967 /*
968  * Reparent resource children of pr that conflict with res
969  * under res, and make res replace those children.
970  */
971 static int __init reparent_resources(struct resource *parent,
972 				     struct resource *res)
973 {
974 	struct resource *p, **pp;
975 	struct resource **firstpp = NULL;
976 
977 	for (pp = &parent->child; (p = *pp) != NULL; pp = &p->sibling) {
978 		if (p->end < res->start)
979 			continue;
980 		if (res->end < p->start)
981 			break;
982 		if (p->start < res->start || p->end > res->end)
983 			return -1;	/* not completely contained */
984 		if (firstpp == NULL)
985 			firstpp = pp;
986 	}
987 	if (firstpp == NULL)
988 		return -1;	/* didn't find any conflicting entries? */
989 	res->parent = parent;
990 	res->child = *firstpp;
991 	res->sibling = *pp;
992 	*firstpp = res;
993 	*pp = NULL;
994 	for (p = res->child; p != NULL; p = p->sibling) {
995 		p->parent = res;
996 		DBG(KERN_INFO "PCI: reparented %s [%llx..%llx] under %s\n",
997 		    p->name,
998 		    (unsigned long long)p->start,
999 		    (unsigned long long)p->end, res->name);
1000 	}
1001 	return 0;
1002 }
1003 
1004 /*
1005  *  Handle resources of PCI devices.  If the world were perfect, we could
1006  *  just allocate all the resource regions and do nothing more.  It isn't.
1007  *  On the other hand, we cannot just re-allocate all devices, as it would
1008  *  require us to know lots of host bridge internals.  So we attempt to
1009  *  keep as much of the original configuration as possible, but tweak it
1010  *  when it's found to be wrong.
1011  *
1012  *  Known BIOS problems we have to work around:
1013  *	- I/O or memory regions not configured
1014  *	- regions configured, but not enabled in the command register
1015  *	- bogus I/O addresses above 64K used
1016  *	- expansion ROMs left enabled (this may sound harmless, but given
1017  *	  the fact the PCI specs explicitly allow address decoders to be
1018  *	  shared between expansion ROMs and other resource regions, it's
1019  *	  at least dangerous)
1020  *
1021  *  Our solution:
1022  *	(1) Allocate resources for all buses behind PCI-to-PCI bridges.
1023  *	    This gives us fixed barriers on where we can allocate.
1024  *	(2) Allocate resources for all enabled devices.  If there is
1025  *	    a collision, just mark the resource as unallocated. Also
1026  *	    disable expansion ROMs during this step.
1027  *	(3) Try to allocate resources for disabled devices.  If the
1028  *	    resources were assigned correctly, everything goes well,
1029  *	    if they weren't, they won't disturb allocation of other
1030  *	    resources.
1031  *	(4) Assign new addresses to resources which were either
1032  *	    not configured at all or misconfigured.  If explicitly
1033  *	    requested by the user, configure expansion ROM address
1034  *	    as well.
1035  */
1036 
1037 static void __init pcibios_allocate_bus_resources(struct list_head *bus_list)
1038 {
1039 	struct pci_bus *bus;
1040 	int i;
1041 	struct resource *res, *pr;
1042 
1043 	/* Depth-First Search on bus tree */
1044 	list_for_each_entry(bus, bus_list, node) {
1045 		for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
1046 			if ((res = bus->resource[i]) == NULL || !res->flags
1047 			    || res->start > res->end)
1048 				continue;
1049 			if (bus->parent == NULL)
1050 				pr = (res->flags & IORESOURCE_IO) ?
1051 					&ioport_resource : &iomem_resource;
1052 			else {
1053 				/* Don't bother with non-root busses when
1054 				 * re-assigning all resources. We clear the
1055 				 * resource flags as if they were colliding
1056 				 * and as such ensure proper re-allocation
1057 				 * later.
1058 				 */
1059 				if (ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)
1060 					goto clear_resource;
1061 				pr = pci_find_parent_resource(bus->self, res);
1062 				if (pr == res) {
1063 					/* this happens when the generic PCI
1064 					 * code (wrongly) decides that this
1065 					 * bridge is transparent  -- paulus
1066 					 */
1067 					continue;
1068 				}
1069 			}
1070 
1071 			DBG("PCI: %s (bus %d) bridge rsrc %d: %016llx-%016llx "
1072 			    "[0x%x], parent %p (%s)\n",
1073 			    bus->self ? pci_name(bus->self) : "PHB",
1074 			    bus->number, i,
1075 			    (unsigned long long)res->start,
1076 			    (unsigned long long)res->end,
1077 			    (unsigned int)res->flags,
1078 			    pr, (pr && pr->name) ? pr->name : "nil");
1079 
1080 			if (pr && !(pr->flags & IORESOURCE_UNSET)) {
1081 				if (request_resource(pr, res) == 0)
1082 					continue;
1083 				/*
1084 				 * Must be a conflict with an existing entry.
1085 				 * Move that entry (or entries) under the
1086 				 * bridge resource and try again.
1087 				 */
1088 				if (reparent_resources(pr, res) == 0)
1089 					continue;
1090 			}
1091 			printk(KERN_WARNING
1092 			       "PCI: Cannot allocate resource region "
1093 			       "%d of PCI bridge %d, will remap\n",
1094 			       i, bus->number);
1095 clear_resource:
1096 			res->flags = 0;
1097 		}
1098 		pcibios_allocate_bus_resources(&bus->children);
1099 	}
1100 }
1101 
1102 static inline void __devinit alloc_resource(struct pci_dev *dev, int idx)
1103 {
1104 	struct resource *pr, *r = &dev->resource[idx];
1105 
1106 	DBG("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1107 	    pci_name(dev), idx,
1108 	    (unsigned long long)r->start,
1109 	    (unsigned long long)r->end,
1110 	    (unsigned int)r->flags);
1111 
1112 	pr = pci_find_parent_resource(dev, r);
1113 	if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1114 	    request_resource(pr, r) < 0) {
1115 		printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1116 		       " of device %s, will remap\n", idx, pci_name(dev));
1117 		if (pr)
1118 			DBG("PCI:  parent is %p: %016llx-%016llx [%x]\n", pr,
1119 			    (unsigned long long)pr->start,
1120 			    (unsigned long long)pr->end,
1121 			    (unsigned int)pr->flags);
1122 		/* We'll assign a new address later */
1123 		r->flags |= IORESOURCE_UNSET;
1124 		r->end -= r->start;
1125 		r->start = 0;
1126 	}
1127 }
1128 
1129 static void __init pcibios_allocate_resources(int pass)
1130 {
1131 	struct pci_dev *dev = NULL;
1132 	int idx, disabled;
1133 	u16 command;
1134 	struct resource *r;
1135 
1136 	for_each_pci_dev(dev) {
1137 		pci_read_config_word(dev, PCI_COMMAND, &command);
1138 		for (idx = 0; idx < 6; idx++) {
1139 			r = &dev->resource[idx];
1140 			if (r->parent)		/* Already allocated */
1141 				continue;
1142 			if (!r->flags || (r->flags & IORESOURCE_UNSET))
1143 				continue;	/* Not assigned at all */
1144 			if (r->flags & IORESOURCE_IO)
1145 				disabled = !(command & PCI_COMMAND_IO);
1146 			else
1147 				disabled = !(command & PCI_COMMAND_MEMORY);
1148 			if (pass == disabled)
1149 				alloc_resource(dev, idx);
1150 		}
1151 		if (pass)
1152 			continue;
1153 		r = &dev->resource[PCI_ROM_RESOURCE];
1154 		if (r->flags & IORESOURCE_ROM_ENABLE) {
1155 			/* Turn the ROM off, leave the resource region,
1156 			 * but keep it unregistered.
1157 			 */
1158 			u32 reg;
1159 			DBG("PCI: Switching off ROM of %s\n", pci_name(dev));
1160 			r->flags &= ~IORESOURCE_ROM_ENABLE;
1161 			pci_read_config_dword(dev, dev->rom_base_reg, &reg);
1162 			pci_write_config_dword(dev, dev->rom_base_reg,
1163 					       reg & ~PCI_ROM_ADDRESS_ENABLE);
1164 		}
1165 	}
1166 }
1167 
1168 void __init pcibios_resource_survey(void)
1169 {
1170 	/* Allocate and assign resources. If we re-assign everything, then
1171 	 * we skip the allocate phase
1172 	 */
1173 	pcibios_allocate_bus_resources(&pci_root_buses);
1174 
1175 	if (!(ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)) {
1176 		pcibios_allocate_resources(0);
1177 		pcibios_allocate_resources(1);
1178 	}
1179 
1180 	if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1181 		DBG("PCI: Assigning unassigned resouces...\n");
1182 		pci_assign_unassigned_resources();
1183 	}
1184 
1185 	/* Call machine dependent fixup */
1186 	if (ppc_md.pcibios_fixup)
1187 		ppc_md.pcibios_fixup();
1188 }
1189 
1190 #ifdef CONFIG_HOTPLUG
1191 /* This is used by the pSeries hotplug driver to allocate resource
1192  * of newly plugged busses. We can try to consolidate with the
1193  * rest of the code later, for now, keep it as-is
1194  */
1195 void __devinit pcibios_claim_one_bus(struct pci_bus *bus)
1196 {
1197 	struct pci_dev *dev;
1198 	struct pci_bus *child_bus;
1199 
1200 	list_for_each_entry(dev, &bus->devices, bus_list) {
1201 		int i;
1202 
1203 		for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1204 			struct resource *r = &dev->resource[i];
1205 
1206 			if (r->parent || !r->start || !r->flags)
1207 				continue;
1208 			pci_claim_resource(dev, i);
1209 		}
1210 	}
1211 
1212 	list_for_each_entry(child_bus, &bus->children, node)
1213 		pcibios_claim_one_bus(child_bus);
1214 }
1215 EXPORT_SYMBOL_GPL(pcibios_claim_one_bus);
1216 #endif /* CONFIG_HOTPLUG */
1217 
1218 int pcibios_enable_device(struct pci_dev *dev, int mask)
1219 {
1220 	if (ppc_md.pcibios_enable_device_hook)
1221 		if (ppc_md.pcibios_enable_device_hook(dev))
1222 			return -EINVAL;
1223 
1224 	return pci_enable_resources(dev, mask);
1225 }
1226