2 * Contains common pci routines for ALL ppc platform
3 * (based on pci_32.c and pci_64.c)
5 * Port for PPC64 David Engebretsen, IBM Corp.
6 * Contains common pci routines for ppc64 platform, pSeries and iSeries brands.
8 * Copyright (C) 2003 Anton Blanchard <anton@au.ibm.com>, IBM
9 * Rework, based on alpha PCI code.
11 * Common pmac/prep/chrp pci routines. -- Cort
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.
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>
25 #include <linux/list.h>
26 #include <linux/syscalls.h>
27 #include <linux/irq.h>
28 #include <linux/vmalloc.h>
30 #include <asm/processor.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>
40 static DEFINE_SPINLOCK(hose_spinlock);
42 /* XXX kill that some day ... */
43 static int global_phb_number; /* Global phb counter */
45 /* ISA Memory physical address */
46 resource_size_t isa_mem_base;
48 /* Default PCI flags is 0 on ppc32, modified at boot on ppc64 */
49 unsigned int ppc_pci_flags = 0;
52 static struct dma_mapping_ops *pci_dma_ops;
54 void set_pci_dma_ops(struct dma_mapping_ops *dma_ops)
56 pci_dma_ops = dma_ops;
59 struct dma_mapping_ops *get_pci_dma_ops(void)
63 EXPORT_SYMBOL(get_pci_dma_ops);
65 int pci_set_dma_mask(struct pci_dev *dev, u64 mask)
67 return dma_set_mask(&dev->dev, mask);
70 int pci_set_consistent_dma_mask(struct pci_dev *dev, u64 mask)
74 rc = dma_set_mask(&dev->dev, mask);
75 dev->dev.coherent_dma_mask = dev->dma_mask;
80 struct pci_controller *pcibios_alloc_controller(struct device_node *dev)
82 struct pci_controller *phb;
84 phb = zalloc_maybe_bootmem(sizeof(struct pci_controller), GFP_KERNEL);
87 spin_lock(&hose_spinlock);
88 phb->global_number = global_phb_number++;
89 list_add_tail(&phb->list_node, &hose_list);
90 spin_unlock(&hose_spinlock);
92 phb->is_dynamic = mem_init_done;
95 int nid = of_node_to_nid(dev);
97 if (nid < 0 || !node_online(nid))
100 PHB_SET_NODE(phb, nid);
106 void pcibios_free_controller(struct pci_controller *phb)
108 spin_lock(&hose_spinlock);
109 list_del(&phb->list_node);
110 spin_unlock(&hose_spinlock);
116 int pcibios_vaddr_is_ioport(void __iomem *address)
119 struct pci_controller *hose;
122 spin_lock(&hose_spinlock);
123 list_for_each_entry(hose, &hose_list, list_node) {
125 size = hose->pci_io_size;
127 size = hose->io_resource.end - hose->io_resource.start + 1;
129 if (address >= hose->io_base_virt &&
130 address < (hose->io_base_virt + size)) {
135 spin_unlock(&hose_spinlock);
140 * Return the domain number for this bus.
142 int pci_domain_nr(struct pci_bus *bus)
144 struct pci_controller *hose = pci_bus_to_host(bus);
146 return hose->global_number;
148 EXPORT_SYMBOL(pci_domain_nr);
152 /* This routine is meant to be used early during boot, when the
153 * PCI bus numbers have not yet been assigned, and you need to
154 * issue PCI config cycles to an OF device.
155 * It could also be used to "fix" RTAS config cycles if you want
156 * to set pci_assign_all_buses to 1 and still use RTAS for PCI
159 struct pci_controller* pci_find_hose_for_OF_device(struct device_node* node)
162 struct pci_controller *hose, *tmp;
163 list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
164 if (hose->dn == node)
171 static ssize_t pci_show_devspec(struct device *dev,
172 struct device_attribute *attr, char *buf)
174 struct pci_dev *pdev;
175 struct device_node *np;
177 pdev = to_pci_dev (dev);
178 np = pci_device_to_OF_node(pdev);
179 if (np == NULL || np->full_name == NULL)
181 return sprintf(buf, "%s", np->full_name);
183 static DEVICE_ATTR(devspec, S_IRUGO, pci_show_devspec, NULL);
184 #endif /* CONFIG_PPC_OF */
186 /* Add sysfs properties */
187 int pcibios_add_platform_entries(struct pci_dev *pdev)
190 return device_create_file(&pdev->dev, &dev_attr_devspec);
193 #endif /* CONFIG_PPC_OF */
197 char __devinit *pcibios_setup(char *str)
203 * Reads the interrupt pin to determine if interrupt is use by card.
204 * If the interrupt is used, then gets the interrupt line from the
205 * openfirmware and sets it in the pci_dev and pci_config line.
207 int pci_read_irq_line(struct pci_dev *pci_dev)
212 /* The current device-tree that iSeries generates from the HV
213 * PCI informations doesn't contain proper interrupt routing,
214 * and all the fallback would do is print out crap, so we
215 * don't attempt to resolve the interrupts here at all, some
216 * iSeries specific fixup does it.
218 * In the long run, we will hopefully fix the generated device-tree
221 #ifdef CONFIG_PPC_ISERIES
222 if (firmware_has_feature(FW_FEATURE_ISERIES))
226 pr_debug("PCI: Try to map irq for %s...\n", pci_name(pci_dev));
229 memset(&oirq, 0xff, sizeof(oirq));
231 /* Try to get a mapping from the device-tree */
232 if (of_irq_map_pci(pci_dev, &oirq)) {
235 /* If that fails, lets fallback to what is in the config
236 * space and map that through the default controller. We
237 * also set the type to level low since that's what PCI
238 * interrupts are. If your platform does differently, then
239 * either provide a proper interrupt tree or don't use this
242 if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_PIN, &pin))
246 if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_LINE, &line) ||
247 line == 0xff || line == 0) {
250 pr_debug(" No map ! Using line %d (pin %d) from PCI config\n",
253 virq = irq_create_mapping(NULL, line);
255 set_irq_type(virq, IRQ_TYPE_LEVEL_LOW);
257 pr_debug(" Got one, spec %d cells (0x%08x 0x%08x...) on %s\n",
258 oirq.size, oirq.specifier[0], oirq.specifier[1],
259 oirq.controller ? oirq.controller->full_name :
262 virq = irq_create_of_mapping(oirq.controller, oirq.specifier,
266 pr_debug(" Failed to map !\n");
270 pr_debug(" Mapped to linux irq %d\n", virq);
276 EXPORT_SYMBOL(pci_read_irq_line);
279 * Platform support for /proc/bus/pci/X/Y mmap()s,
280 * modelled on the sparc64 implementation by Dave Miller.
285 * Adjust vm_pgoff of VMA such that it is the physical page offset
286 * corresponding to the 32-bit pci bus offset for DEV requested by the user.
288 * Basically, the user finds the base address for his device which he wishes
289 * to mmap. They read the 32-bit value from the config space base register,
290 * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
291 * offset parameter of mmap on /proc/bus/pci/XXX for that device.
293 * Returns negative error code on failure, zero on success.
295 static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
296 resource_size_t *offset,
297 enum pci_mmap_state mmap_state)
299 struct pci_controller *hose = pci_bus_to_host(dev->bus);
300 unsigned long io_offset = 0;
304 return NULL; /* should never happen */
306 /* If memory, add on the PCI bridge address offset */
307 if (mmap_state == pci_mmap_mem) {
308 #if 0 /* See comment in pci_resource_to_user() for why this is disabled */
309 *offset += hose->pci_mem_offset;
311 res_bit = IORESOURCE_MEM;
313 io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
314 *offset += io_offset;
315 res_bit = IORESOURCE_IO;
319 * Check that the offset requested corresponds to one of the
320 * resources of the device.
322 for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
323 struct resource *rp = &dev->resource[i];
324 int flags = rp->flags;
326 /* treat ROM as memory (should be already) */
327 if (i == PCI_ROM_RESOURCE)
328 flags |= IORESOURCE_MEM;
330 /* Active and same type? */
331 if ((flags & res_bit) == 0)
334 /* In the range of this resource? */
335 if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
338 /* found it! construct the final physical address */
339 if (mmap_state == pci_mmap_io)
340 *offset += hose->io_base_phys - io_offset;
348 * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
351 static pgprot_t __pci_mmap_set_pgprot(struct pci_dev *dev, struct resource *rp,
353 enum pci_mmap_state mmap_state,
356 unsigned long prot = pgprot_val(protection);
358 /* Write combine is always 0 on non-memory space mappings. On
359 * memory space, if the user didn't pass 1, we check for a
360 * "prefetchable" resource. This is a bit hackish, but we use
361 * this to workaround the inability of /sysfs to provide a write
364 if (mmap_state != pci_mmap_mem)
366 else if (write_combine == 0) {
367 if (rp->flags & IORESOURCE_PREFETCH)
371 /* XXX would be nice to have a way to ask for write-through */
373 return pgprot_noncached_wc(prot);
375 return pgprot_noncached(prot);
379 * This one is used by /dev/mem and fbdev who have no clue about the
380 * PCI device, it tries to find the PCI device first and calls the
383 pgprot_t pci_phys_mem_access_prot(struct file *file,
388 struct pci_dev *pdev = NULL;
389 struct resource *found = NULL;
390 resource_size_t offset = ((resource_size_t)pfn) << PAGE_SHIFT;
393 if (page_is_ram(pfn))
396 prot = pgprot_noncached(prot);
397 for_each_pci_dev(pdev) {
398 for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
399 struct resource *rp = &pdev->resource[i];
400 int flags = rp->flags;
402 /* Active and same type? */
403 if ((flags & IORESOURCE_MEM) == 0)
405 /* In the range of this resource? */
406 if (offset < (rp->start & PAGE_MASK) ||
416 if (found->flags & IORESOURCE_PREFETCH)
417 prot = pgprot_noncached_wc(prot);
421 pr_debug("PCI: Non-PCI map for %llx, prot: %lx\n",
422 (unsigned long long)offset, pgprot_val(prot));
429 * Perform the actual remap of the pages for a PCI device mapping, as
430 * appropriate for this architecture. The region in the process to map
431 * is described by vm_start and vm_end members of VMA, the base physical
432 * address is found in vm_pgoff.
433 * The pci device structure is provided so that architectures may make mapping
434 * decisions on a per-device or per-bus basis.
436 * Returns a negative error code on failure, zero on success.
438 int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
439 enum pci_mmap_state mmap_state, int write_combine)
441 resource_size_t offset =
442 ((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
446 rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
450 vma->vm_pgoff = offset >> PAGE_SHIFT;
451 vma->vm_page_prot = __pci_mmap_set_pgprot(dev, rp,
453 mmap_state, write_combine);
455 ret = remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
456 vma->vm_end - vma->vm_start, vma->vm_page_prot);
461 /* This provides legacy IO read access on a bus */
462 int pci_legacy_read(struct pci_bus *bus, loff_t port, u32 *val, size_t size)
464 unsigned long offset;
465 struct pci_controller *hose = pci_bus_to_host(bus);
466 struct resource *rp = &hose->io_resource;
469 /* Check if port can be supported by that bus. We only check
470 * the ranges of the PHB though, not the bus itself as the rules
471 * for forwarding legacy cycles down bridges are not our problem
472 * here. So if the host bridge supports it, we do it.
474 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
477 if (!(rp->flags & IORESOURCE_IO))
479 if (offset < rp->start || (offset + size) > rp->end)
481 addr = hose->io_base_virt + port;
485 *((u8 *)val) = in_8(addr);
490 *((u16 *)val) = in_le16(addr);
495 *((u32 *)val) = in_le32(addr);
501 /* This provides legacy IO write access on a bus */
502 int pci_legacy_write(struct pci_bus *bus, loff_t port, u32 val, size_t size)
504 unsigned long offset;
505 struct pci_controller *hose = pci_bus_to_host(bus);
506 struct resource *rp = &hose->io_resource;
509 /* Check if port can be supported by that bus. We only check
510 * the ranges of the PHB though, not the bus itself as the rules
511 * for forwarding legacy cycles down bridges are not our problem
512 * here. So if the host bridge supports it, we do it.
514 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
517 if (!(rp->flags & IORESOURCE_IO))
519 if (offset < rp->start || (offset + size) > rp->end)
521 addr = hose->io_base_virt + port;
523 /* WARNING: The generic code is idiotic. It gets passed a pointer
524 * to what can be a 1, 2 or 4 byte quantity and always reads that
525 * as a u32, which means that we have to correct the location of
526 * the data read within those 32 bits for size 1 and 2
530 out_8(addr, val >> 24);
535 out_le16(addr, val >> 16);
546 /* This provides legacy IO or memory mmap access on a bus */
547 int pci_mmap_legacy_page_range(struct pci_bus *bus,
548 struct vm_area_struct *vma,
549 enum pci_mmap_state mmap_state)
551 struct pci_controller *hose = pci_bus_to_host(bus);
552 resource_size_t offset =
553 ((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
554 resource_size_t size = vma->vm_end - vma->vm_start;
557 pr_debug("pci_mmap_legacy_page_range(%04x:%02x, %s @%llx..%llx)\n",
558 pci_domain_nr(bus), bus->number,
559 mmap_state == pci_mmap_mem ? "MEM" : "IO",
560 (unsigned long long)offset,
561 (unsigned long long)(offset + size - 1));
563 if (mmap_state == pci_mmap_mem) {
564 if ((offset + size) > hose->isa_mem_size)
566 offset += hose->isa_mem_phys;
568 unsigned long io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
569 unsigned long roffset = offset + io_offset;
570 rp = &hose->io_resource;
571 if (!(rp->flags & IORESOURCE_IO))
573 if (roffset < rp->start || (roffset + size) > rp->end)
575 offset += hose->io_base_phys;
577 pr_debug(" -> mapping phys %llx\n", (unsigned long long)offset);
579 vma->vm_pgoff = offset >> PAGE_SHIFT;
580 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
581 return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
582 vma->vm_end - vma->vm_start,
586 void pci_resource_to_user(const struct pci_dev *dev, int bar,
587 const struct resource *rsrc,
588 resource_size_t *start, resource_size_t *end)
590 struct pci_controller *hose = pci_bus_to_host(dev->bus);
591 resource_size_t offset = 0;
596 if (rsrc->flags & IORESOURCE_IO)
597 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
599 /* We pass a fully fixed up address to userland for MMIO instead of
600 * a BAR value because X is lame and expects to be able to use that
601 * to pass to /dev/mem !
603 * That means that we'll have potentially 64 bits values where some
604 * userland apps only expect 32 (like X itself since it thinks only
605 * Sparc has 64 bits MMIO) but if we don't do that, we break it on
608 * Hopefully, the sysfs insterface is immune to that gunk. Once X
609 * has been fixed (and the fix spread enough), we can re-enable the
610 * 2 lines below and pass down a BAR value to userland. In that case
611 * we'll also have to re-enable the matching code in
612 * __pci_mmap_make_offset().
617 else if (rsrc->flags & IORESOURCE_MEM)
618 offset = hose->pci_mem_offset;
621 *start = rsrc->start - offset;
622 *end = rsrc->end - offset;
626 * pci_process_bridge_OF_ranges - Parse PCI bridge resources from device tree
627 * @hose: newly allocated pci_controller to be setup
628 * @dev: device node of the host bridge
629 * @primary: set if primary bus (32 bits only, soon to be deprecated)
631 * This function will parse the "ranges" property of a PCI host bridge device
632 * node and setup the resource mapping of a pci controller based on its
635 * Life would be boring if it wasn't for a few issues that we have to deal
638 * - We can only cope with one IO space range and up to 3 Memory space
639 * ranges. However, some machines (thanks Apple !) tend to split their
640 * space into lots of small contiguous ranges. So we have to coalesce.
642 * - We can only cope with all memory ranges having the same offset
643 * between CPU addresses and PCI addresses. Unfortunately, some bridges
644 * are setup for a large 1:1 mapping along with a small "window" which
645 * maps PCI address 0 to some arbitrary high address of the CPU space in
646 * order to give access to the ISA memory hole.
647 * The way out of here that I've chosen for now is to always set the
648 * offset based on the first resource found, then override it if we
649 * have a different offset and the previous was set by an ISA hole.
651 * - Some busses have IO space not starting at 0, which causes trouble with
652 * the way we do our IO resource renumbering. The code somewhat deals with
653 * it for 64 bits but I would expect problems on 32 bits.
655 * - Some 32 bits platforms such as 4xx can have physical space larger than
656 * 32 bits so we need to use 64 bits values for the parsing
658 void __devinit pci_process_bridge_OF_ranges(struct pci_controller *hose,
659 struct device_node *dev,
664 int pna = of_n_addr_cells(dev);
666 int memno = 0, isa_hole = -1;
668 unsigned long long pci_addr, cpu_addr, pci_next, cpu_next, size;
669 unsigned long long isa_mb = 0;
670 struct resource *res;
672 printk(KERN_INFO "PCI host bridge %s %s ranges:\n",
673 dev->full_name, primary ? "(primary)" : "");
675 /* Get ranges property */
676 ranges = of_get_property(dev, "ranges", &rlen);
681 while ((rlen -= np * 4) >= 0) {
682 /* Read next ranges element */
683 pci_space = ranges[0];
684 pci_addr = of_read_number(ranges + 1, 2);
685 cpu_addr = of_translate_address(dev, ranges + 3);
686 size = of_read_number(ranges + pna + 3, 2);
689 /* If we failed translation or got a zero-sized region
690 * (some FW try to feed us with non sensical zero sized regions
691 * such as power3 which look like some kind of attempt at exposing
692 * the VGA memory hole)
694 if (cpu_addr == OF_BAD_ADDR || size == 0)
697 /* Now consume following elements while they are contiguous */
698 for (; rlen >= np * sizeof(u32);
699 ranges += np, rlen -= np * 4) {
700 if (ranges[0] != pci_space)
702 pci_next = of_read_number(ranges + 1, 2);
703 cpu_next = of_translate_address(dev, ranges + 3);
704 if (pci_next != pci_addr + size ||
705 cpu_next != cpu_addr + size)
707 size += of_read_number(ranges + pna + 3, 2);
710 /* Act based on address space type */
712 switch ((pci_space >> 24) & 0x3) {
713 case 1: /* PCI IO space */
715 " IO 0x%016llx..0x%016llx -> 0x%016llx\n",
716 cpu_addr, cpu_addr + size - 1, pci_addr);
718 /* We support only one IO range */
719 if (hose->pci_io_size) {
721 " \\--> Skipped (too many) !\n");
725 /* On 32 bits, limit I/O space to 16MB */
726 if (size > 0x01000000)
729 /* 32 bits needs to map IOs here */
730 hose->io_base_virt = ioremap(cpu_addr, size);
732 /* Expect trouble if pci_addr is not 0 */
735 (unsigned long)hose->io_base_virt;
736 #endif /* CONFIG_PPC32 */
737 /* pci_io_size and io_base_phys always represent IO
738 * space starting at 0 so we factor in pci_addr
740 hose->pci_io_size = pci_addr + size;
741 hose->io_base_phys = cpu_addr - pci_addr;
744 res = &hose->io_resource;
745 res->flags = IORESOURCE_IO;
746 res->start = pci_addr;
748 case 2: /* PCI Memory space */
749 case 3: /* PCI 64 bits Memory space */
751 " MEM 0x%016llx..0x%016llx -> 0x%016llx %s\n",
752 cpu_addr, cpu_addr + size - 1, pci_addr,
753 (pci_space & 0x40000000) ? "Prefetch" : "");
755 /* We support only 3 memory ranges */
758 " \\--> Skipped (too many) !\n");
761 /* Handles ISA memory hole space here */
765 if (primary || isa_mem_base == 0)
766 isa_mem_base = cpu_addr;
767 hose->isa_mem_phys = cpu_addr;
768 hose->isa_mem_size = size;
771 /* We get the PCI/Mem offset from the first range or
772 * the, current one if the offset came from an ISA
773 * hole. If they don't match, bugger.
776 (isa_hole >= 0 && pci_addr != 0 &&
777 hose->pci_mem_offset == isa_mb))
778 hose->pci_mem_offset = cpu_addr - pci_addr;
779 else if (pci_addr != 0 &&
780 hose->pci_mem_offset != cpu_addr - pci_addr) {
782 " \\--> Skipped (offset mismatch) !\n");
787 res = &hose->mem_resources[memno++];
788 res->flags = IORESOURCE_MEM;
789 if (pci_space & 0x40000000)
790 res->flags |= IORESOURCE_PREFETCH;
791 res->start = cpu_addr;
795 res->name = dev->full_name;
796 res->end = res->start + size - 1;
803 /* If there's an ISA hole and the pci_mem_offset is -not- matching
804 * the ISA hole offset, then we need to remove the ISA hole from
805 * the resource list for that brige
807 if (isa_hole >= 0 && hose->pci_mem_offset != isa_mb) {
808 unsigned int next = isa_hole + 1;
809 printk(KERN_INFO " Removing ISA hole at 0x%016llx\n", isa_mb);
811 memmove(&hose->mem_resources[isa_hole],
812 &hose->mem_resources[next],
813 sizeof(struct resource) * (memno - next));
814 hose->mem_resources[--memno].flags = 0;
818 /* Decide whether to display the domain number in /proc */
819 int pci_proc_domain(struct pci_bus *bus)
821 struct pci_controller *hose = pci_bus_to_host(bus);
823 if (!(ppc_pci_flags & PPC_PCI_ENABLE_PROC_DOMAINS))
825 if (ppc_pci_flags & PPC_PCI_COMPAT_DOMAIN_0)
826 return hose->global_number != 0;
830 void pcibios_resource_to_bus(struct pci_dev *dev, struct pci_bus_region *region,
831 struct resource *res)
833 resource_size_t offset = 0, mask = (resource_size_t)-1;
834 struct pci_controller *hose = pci_bus_to_host(dev->bus);
838 if (res->flags & IORESOURCE_IO) {
839 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
841 } else if (res->flags & IORESOURCE_MEM)
842 offset = hose->pci_mem_offset;
844 region->start = (res->start - offset) & mask;
845 region->end = (res->end - offset) & mask;
847 EXPORT_SYMBOL(pcibios_resource_to_bus);
849 void pcibios_bus_to_resource(struct pci_dev *dev, struct resource *res,
850 struct pci_bus_region *region)
852 resource_size_t offset = 0, mask = (resource_size_t)-1;
853 struct pci_controller *hose = pci_bus_to_host(dev->bus);
857 if (res->flags & IORESOURCE_IO) {
858 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
860 } else if (res->flags & IORESOURCE_MEM)
861 offset = hose->pci_mem_offset;
862 res->start = (region->start + offset) & mask;
863 res->end = (region->end + offset) & mask;
865 EXPORT_SYMBOL(pcibios_bus_to_resource);
867 /* Fixup a bus resource into a linux resource */
868 static void __devinit fixup_resource(struct resource *res, struct pci_dev *dev)
870 struct pci_controller *hose = pci_bus_to_host(dev->bus);
871 resource_size_t offset = 0, mask = (resource_size_t)-1;
873 if (res->flags & IORESOURCE_IO) {
874 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
876 } else if (res->flags & IORESOURCE_MEM)
877 offset = hose->pci_mem_offset;
879 res->start = (res->start + offset) & mask;
880 res->end = (res->end + offset) & mask;
884 /* This header fixup will do the resource fixup for all devices as they are
885 * probed, but not for bridge ranges
887 static void __devinit pcibios_fixup_resources(struct pci_dev *dev)
889 struct pci_controller *hose = pci_bus_to_host(dev->bus);
893 printk(KERN_ERR "No host bridge for PCI dev %s !\n",
897 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
898 struct resource *res = dev->resource + i;
901 /* On platforms that have PPC_PCI_PROBE_ONLY set, we don't
902 * consider 0 as an unassigned BAR value. It's technically
903 * a valid value, but linux doesn't like it... so when we can
904 * re-assign things, we do so, but if we can't, we keep it
905 * around and hope for the best...
907 if (res->start == 0 && !(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
908 pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] is unassigned\n",
910 (unsigned long long)res->start,
911 (unsigned long long)res->end,
912 (unsigned int)res->flags);
913 res->end -= res->start;
915 res->flags |= IORESOURCE_UNSET;
919 pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] fixup...\n",
921 (unsigned long long)res->start,\
922 (unsigned long long)res->end,
923 (unsigned int)res->flags);
925 fixup_resource(res, dev);
927 pr_debug("PCI:%s %016llx-%016llx\n",
929 (unsigned long long)res->start,
930 (unsigned long long)res->end);
933 /* Call machine specific resource fixup */
934 if (ppc_md.pcibios_fixup_resources)
935 ppc_md.pcibios_fixup_resources(dev);
937 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_resources);
939 /* This function tries to figure out if a bridge resource has been initialized
940 * by the firmware or not. It doesn't have to be absolutely bullet proof, but
941 * things go more smoothly when it gets it right. It should covers cases such
942 * as Apple "closed" bridge resources and bare-metal pSeries unassigned bridges
944 static int __devinit pcibios_uninitialized_bridge_resource(struct pci_bus *bus,
945 struct resource *res)
947 struct pci_controller *hose = pci_bus_to_host(bus);
948 struct pci_dev *dev = bus->self;
949 resource_size_t offset;
953 /* We don't do anything if PCI_PROBE_ONLY is set */
954 if (ppc_pci_flags & PPC_PCI_PROBE_ONLY)
957 /* Job is a bit different between memory and IO */
958 if (res->flags & IORESOURCE_MEM) {
959 /* If the BAR is non-0 (res != pci_mem_offset) then it's probably been
960 * initialized by somebody
962 if (res->start != hose->pci_mem_offset)
965 /* The BAR is 0, let's check if memory decoding is enabled on
966 * the bridge. If not, we consider it unassigned
968 pci_read_config_word(dev, PCI_COMMAND, &command);
969 if ((command & PCI_COMMAND_MEMORY) == 0)
972 /* Memory decoding is enabled and the BAR is 0. If any of the bridge
973 * resources covers that starting address (0 then it's good enough for
976 for (i = 0; i < 3; i++) {
977 if ((hose->mem_resources[i].flags & IORESOURCE_MEM) &&
978 hose->mem_resources[i].start == hose->pci_mem_offset)
982 /* Well, it starts at 0 and we know it will collide so we may as
983 * well consider it as unassigned. That covers the Apple case.
987 /* If the BAR is non-0, then we consider it assigned */
988 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
989 if (((res->start - offset) & 0xfffffffful) != 0)
992 /* Here, we are a bit different than memory as typically IO space
993 * starting at low addresses -is- valid. What we do instead if that
994 * we consider as unassigned anything that doesn't have IO enabled
995 * in the PCI command register, and that's it.
997 pci_read_config_word(dev, PCI_COMMAND, &command);
998 if (command & PCI_COMMAND_IO)
1001 /* It's starting at 0 and IO is disabled in the bridge, consider
1008 /* Fixup resources of a PCI<->PCI bridge */
1009 static void __devinit pcibios_fixup_bridge(struct pci_bus *bus)
1011 struct resource *res;
1014 struct pci_dev *dev = bus->self;
1016 for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
1017 if ((res = bus->resource[i]) == NULL)
1021 if (i >= 3 && bus->self->transparent)
1024 pr_debug("PCI:%s Bus rsrc %d %016llx-%016llx [%x] fixup...\n",
1026 (unsigned long long)res->start,\
1027 (unsigned long long)res->end,
1028 (unsigned int)res->flags);
1031 fixup_resource(res, dev);
1033 /* Try to detect uninitialized P2P bridge resources,
1034 * and clear them out so they get re-assigned later
1036 if (pcibios_uninitialized_bridge_resource(bus, res)) {
1038 pr_debug("PCI:%s (unassigned)\n", pci_name(dev));
1041 pr_debug("PCI:%s %016llx-%016llx\n",
1043 (unsigned long long)res->start,
1044 (unsigned long long)res->end);
1049 void __devinit pcibios_setup_bus_self(struct pci_bus *bus)
1051 /* Fix up the bus resources for P2P bridges */
1052 if (bus->self != NULL)
1053 pcibios_fixup_bridge(bus);
1055 /* Platform specific bus fixups. This is currently only used
1056 * by fsl_pci and I'm hoping to get rid of it at some point
1058 if (ppc_md.pcibios_fixup_bus)
1059 ppc_md.pcibios_fixup_bus(bus);
1061 /* Setup bus DMA mappings */
1062 if (ppc_md.pci_dma_bus_setup)
1063 ppc_md.pci_dma_bus_setup(bus);
1066 void __devinit pcibios_setup_bus_devices(struct pci_bus *bus)
1068 struct pci_dev *dev;
1070 pr_debug("PCI: Fixup bus devices %d (%s)\n",
1071 bus->number, bus->self ? pci_name(bus->self) : "PHB");
1073 list_for_each_entry(dev, &bus->devices, bus_list) {
1074 struct dev_archdata *sd = &dev->dev.archdata;
1076 /* Setup OF node pointer in archdata */
1077 sd->of_node = pci_device_to_OF_node(dev);
1079 /* Fixup NUMA node as it may not be setup yet by the generic
1080 * code and is needed by the DMA init
1082 set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
1084 /* Hook up default DMA ops */
1085 sd->dma_ops = pci_dma_ops;
1086 sd->dma_data = (void *)PCI_DRAM_OFFSET;
1088 /* Additional platform DMA/iommu setup */
1089 if (ppc_md.pci_dma_dev_setup)
1090 ppc_md.pci_dma_dev_setup(dev);
1092 /* Read default IRQs and fixup if necessary */
1093 pci_read_irq_line(dev);
1094 if (ppc_md.pci_irq_fixup)
1095 ppc_md.pci_irq_fixup(dev);
1099 void __devinit pcibios_fixup_bus(struct pci_bus *bus)
1101 /* When called from the generic PCI probe, read PCI<->PCI bridge
1102 * bases. This is -not- called when generating the PCI tree from
1103 * the OF device-tree.
1105 if (bus->self != NULL)
1106 pci_read_bridge_bases(bus);
1108 /* Now fixup the bus bus */
1109 pcibios_setup_bus_self(bus);
1111 /* Now fixup devices on that bus */
1112 pcibios_setup_bus_devices(bus);
1114 EXPORT_SYMBOL(pcibios_fixup_bus);
1116 static int skip_isa_ioresource_align(struct pci_dev *dev)
1118 if ((ppc_pci_flags & PPC_PCI_CAN_SKIP_ISA_ALIGN) &&
1119 !(dev->bus->bridge_ctl & PCI_BRIDGE_CTL_ISA))
1125 * We need to avoid collisions with `mirrored' VGA ports
1126 * and other strange ISA hardware, so we always want the
1127 * addresses to be allocated in the 0x000-0x0ff region
1130 * Why? Because some silly external IO cards only decode
1131 * the low 10 bits of the IO address. The 0x00-0xff region
1132 * is reserved for motherboard devices that decode all 16
1133 * bits, so it's ok to allocate at, say, 0x2800-0x28ff,
1134 * but we want to try to avoid allocating at 0x2900-0x2bff
1135 * which might have be mirrored at 0x0100-0x03ff..
1137 void pcibios_align_resource(void *data, struct resource *res,
1138 resource_size_t size, resource_size_t align)
1140 struct pci_dev *dev = data;
1142 if (res->flags & IORESOURCE_IO) {
1143 resource_size_t start = res->start;
1145 if (skip_isa_ioresource_align(dev))
1147 if (start & 0x300) {
1148 start = (start + 0x3ff) & ~0x3ff;
1153 EXPORT_SYMBOL(pcibios_align_resource);
1156 * Reparent resource children of pr that conflict with res
1157 * under res, and make res replace those children.
1159 static int __init reparent_resources(struct resource *parent,
1160 struct resource *res)
1162 struct resource *p, **pp;
1163 struct resource **firstpp = NULL;
1165 for (pp = &parent->child; (p = *pp) != NULL; pp = &p->sibling) {
1166 if (p->end < res->start)
1168 if (res->end < p->start)
1170 if (p->start < res->start || p->end > res->end)
1171 return -1; /* not completely contained */
1172 if (firstpp == NULL)
1175 if (firstpp == NULL)
1176 return -1; /* didn't find any conflicting entries? */
1177 res->parent = parent;
1178 res->child = *firstpp;
1182 for (p = res->child; p != NULL; p = p->sibling) {
1184 pr_debug("PCI: Reparented %s [%llx..%llx] under %s\n",
1186 (unsigned long long)p->start,
1187 (unsigned long long)p->end, res->name);
1193 * Handle resources of PCI devices. If the world were perfect, we could
1194 * just allocate all the resource regions and do nothing more. It isn't.
1195 * On the other hand, we cannot just re-allocate all devices, as it would
1196 * require us to know lots of host bridge internals. So we attempt to
1197 * keep as much of the original configuration as possible, but tweak it
1198 * when it's found to be wrong.
1200 * Known BIOS problems we have to work around:
1201 * - I/O or memory regions not configured
1202 * - regions configured, but not enabled in the command register
1203 * - bogus I/O addresses above 64K used
1204 * - expansion ROMs left enabled (this may sound harmless, but given
1205 * the fact the PCI specs explicitly allow address decoders to be
1206 * shared between expansion ROMs and other resource regions, it's
1207 * at least dangerous)
1210 * (1) Allocate resources for all buses behind PCI-to-PCI bridges.
1211 * This gives us fixed barriers on where we can allocate.
1212 * (2) Allocate resources for all enabled devices. If there is
1213 * a collision, just mark the resource as unallocated. Also
1214 * disable expansion ROMs during this step.
1215 * (3) Try to allocate resources for disabled devices. If the
1216 * resources were assigned correctly, everything goes well,
1217 * if they weren't, they won't disturb allocation of other
1219 * (4) Assign new addresses to resources which were either
1220 * not configured at all or misconfigured. If explicitly
1221 * requested by the user, configure expansion ROM address
1225 void pcibios_allocate_bus_resources(struct pci_bus *bus)
1229 struct resource *res, *pr;
1231 pr_debug("PCI: Allocating bus resources for %04x:%02x...\n",
1232 pci_domain_nr(bus), bus->number);
1234 for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
1235 if ((res = bus->resource[i]) == NULL || !res->flags
1236 || res->start > res->end || res->parent)
1238 if (bus->parent == NULL)
1239 pr = (res->flags & IORESOURCE_IO) ?
1240 &ioport_resource : &iomem_resource;
1242 /* Don't bother with non-root busses when
1243 * re-assigning all resources. We clear the
1244 * resource flags as if they were colliding
1245 * and as such ensure proper re-allocation
1248 if (ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)
1249 goto clear_resource;
1250 pr = pci_find_parent_resource(bus->self, res);
1252 /* this happens when the generic PCI
1253 * code (wrongly) decides that this
1254 * bridge is transparent -- paulus
1260 pr_debug("PCI: %s (bus %d) bridge rsrc %d: %016llx-%016llx "
1261 "[0x%x], parent %p (%s)\n",
1262 bus->self ? pci_name(bus->self) : "PHB",
1264 (unsigned long long)res->start,
1265 (unsigned long long)res->end,
1266 (unsigned int)res->flags,
1267 pr, (pr && pr->name) ? pr->name : "nil");
1269 if (pr && !(pr->flags & IORESOURCE_UNSET)) {
1270 if (request_resource(pr, res) == 0)
1273 * Must be a conflict with an existing entry.
1274 * Move that entry (or entries) under the
1275 * bridge resource and try again.
1277 if (reparent_resources(pr, res) == 0)
1280 printk(KERN_WARNING "PCI: Cannot allocate resource region "
1281 "%d of PCI bridge %d, will remap\n", i, bus->number);
1286 list_for_each_entry(b, &bus->children, node)
1287 pcibios_allocate_bus_resources(b);
1290 static inline void __devinit alloc_resource(struct pci_dev *dev, int idx)
1292 struct resource *pr, *r = &dev->resource[idx];
1294 pr_debug("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1296 (unsigned long long)r->start,
1297 (unsigned long long)r->end,
1298 (unsigned int)r->flags);
1300 pr = pci_find_parent_resource(dev, r);
1301 if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1302 request_resource(pr, r) < 0) {
1303 printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1304 " of device %s, will remap\n", idx, pci_name(dev));
1306 pr_debug("PCI: parent is %p: %016llx-%016llx [%x]\n",
1308 (unsigned long long)pr->start,
1309 (unsigned long long)pr->end,
1310 (unsigned int)pr->flags);
1311 /* We'll assign a new address later */
1312 r->flags |= IORESOURCE_UNSET;
1318 static void __init pcibios_allocate_resources(int pass)
1320 struct pci_dev *dev = NULL;
1325 for_each_pci_dev(dev) {
1326 pci_read_config_word(dev, PCI_COMMAND, &command);
1327 for (idx = 0; idx < 6; idx++) {
1328 r = &dev->resource[idx];
1329 if (r->parent) /* Already allocated */
1331 if (!r->flags || (r->flags & IORESOURCE_UNSET))
1332 continue; /* Not assigned at all */
1333 if (r->flags & IORESOURCE_IO)
1334 disabled = !(command & PCI_COMMAND_IO);
1336 disabled = !(command & PCI_COMMAND_MEMORY);
1337 if (pass == disabled)
1338 alloc_resource(dev, idx);
1342 r = &dev->resource[PCI_ROM_RESOURCE];
1343 if (r->flags & IORESOURCE_ROM_ENABLE) {
1344 /* Turn the ROM off, leave the resource region,
1345 * but keep it unregistered.
1348 pr_debug("PCI: Switching off ROM of %s\n",
1350 r->flags &= ~IORESOURCE_ROM_ENABLE;
1351 pci_read_config_dword(dev, dev->rom_base_reg, ®);
1352 pci_write_config_dword(dev, dev->rom_base_reg,
1353 reg & ~PCI_ROM_ADDRESS_ENABLE);
1358 static void __init pcibios_reserve_legacy_regions(struct pci_bus *bus)
1360 struct pci_controller *hose = pci_bus_to_host(bus);
1361 resource_size_t offset;
1362 struct resource *res, *pres;
1365 pr_debug("Reserving legacy ranges for domain %04x\n", pci_domain_nr(bus));
1368 if (!(hose->io_resource.flags & IORESOURCE_IO))
1370 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1371 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1372 BUG_ON(res == NULL);
1373 res->name = "Legacy IO";
1374 res->flags = IORESOURCE_IO;
1375 res->start = offset;
1376 res->end = (offset + 0xfff) & 0xfffffffful;
1377 pr_debug("Candidate legacy IO: %pR\n", res);
1378 if (request_resource(&hose->io_resource, res)) {
1380 "PCI %04x:%02x Cannot reserve Legacy IO %pR\n",
1381 pci_domain_nr(bus), bus->number, res);
1386 /* Check for memory */
1387 offset = hose->pci_mem_offset;
1388 pr_debug("hose mem offset: %016llx\n", (unsigned long long)offset);
1389 for (i = 0; i < 3; i++) {
1390 pres = &hose->mem_resources[i];
1391 if (!(pres->flags & IORESOURCE_MEM))
1393 pr_debug("hose mem res: %pR\n", pres);
1394 if ((pres->start - offset) <= 0xa0000 &&
1395 (pres->end - offset) >= 0xbffff)
1400 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1401 BUG_ON(res == NULL);
1402 res->name = "Legacy VGA memory";
1403 res->flags = IORESOURCE_MEM;
1404 res->start = 0xa0000 + offset;
1405 res->end = 0xbffff + offset;
1406 pr_debug("Candidate VGA memory: %pR\n", res);
1407 if (request_resource(pres, res)) {
1409 "PCI %04x:%02x Cannot reserve VGA memory %pR\n",
1410 pci_domain_nr(bus), bus->number, res);
1415 void __init pcibios_resource_survey(void)
1419 /* Allocate and assign resources. If we re-assign everything, then
1420 * we skip the allocate phase
1422 list_for_each_entry(b, &pci_root_buses, node)
1423 pcibios_allocate_bus_resources(b);
1425 if (!(ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)) {
1426 pcibios_allocate_resources(0);
1427 pcibios_allocate_resources(1);
1430 /* Before we start assigning unassigned resource, we try to reserve
1431 * the low IO area and the VGA memory area if they intersect the
1432 * bus available resources to avoid allocating things on top of them
1434 if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1435 list_for_each_entry(b, &pci_root_buses, node)
1436 pcibios_reserve_legacy_regions(b);
1439 /* Now, if the platform didn't decide to blindly trust the firmware,
1440 * we proceed to assigning things that were left unassigned
1442 if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1443 pr_debug("PCI: Assigning unassigned resouces...\n");
1444 pci_assign_unassigned_resources();
1447 /* Call machine dependent fixup */
1448 if (ppc_md.pcibios_fixup)
1449 ppc_md.pcibios_fixup();
1452 #ifdef CONFIG_HOTPLUG
1454 /* This is used by the PCI hotplug driver to allocate resource
1455 * of newly plugged busses. We can try to consolidate with the
1456 * rest of the code later, for now, keep it as-is as our main
1457 * resource allocation function doesn't deal with sub-trees yet.
1459 void __devinit pcibios_claim_one_bus(struct pci_bus *bus)
1461 struct pci_dev *dev;
1462 struct pci_bus *child_bus;
1464 list_for_each_entry(dev, &bus->devices, bus_list) {
1467 for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1468 struct resource *r = &dev->resource[i];
1470 if (r->parent || !r->start || !r->flags)
1473 pr_debug("PCI: Claiming %s: "
1474 "Resource %d: %016llx..%016llx [%x]\n",
1476 (unsigned long long)r->start,
1477 (unsigned long long)r->end,
1478 (unsigned int)r->flags);
1480 pci_claim_resource(dev, i);
1484 list_for_each_entry(child_bus, &bus->children, node)
1485 pcibios_claim_one_bus(child_bus);
1487 EXPORT_SYMBOL_GPL(pcibios_claim_one_bus);
1490 /* pcibios_finish_adding_to_bus
1492 * This is to be called by the hotplug code after devices have been
1493 * added to a bus, this include calling it for a PHB that is just
1496 void pcibios_finish_adding_to_bus(struct pci_bus *bus)
1498 pr_debug("PCI: Finishing adding to hotplug bus %04x:%02x\n",
1499 pci_domain_nr(bus), bus->number);
1501 /* Allocate bus and devices resources */
1502 pcibios_allocate_bus_resources(bus);
1503 pcibios_claim_one_bus(bus);
1505 /* Add new devices to global lists. Register in proc, sysfs. */
1506 pci_bus_add_devices(bus);
1509 eeh_add_device_tree_late(bus);
1511 EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
1513 #endif /* CONFIG_HOTPLUG */
1515 int pcibios_enable_device(struct pci_dev *dev, int mask)
1517 if (ppc_md.pcibios_enable_device_hook)
1518 if (ppc_md.pcibios_enable_device_hook(dev))
1521 return pci_enable_resources(dev, mask);
1524 void __devinit pcibios_setup_phb_resources(struct pci_controller *hose)
1526 struct pci_bus *bus = hose->bus;
1527 struct resource *res;
1530 /* Hookup PHB IO resource */
1531 bus->resource[0] = res = &hose->io_resource;
1534 printk(KERN_WARNING "PCI: I/O resource not set for host"
1535 " bridge %s (domain %d)\n",
1536 hose->dn->full_name, hose->global_number);
1538 /* Workaround for lack of IO resource only on 32-bit */
1539 res->start = (unsigned long)hose->io_base_virt - isa_io_base;
1540 res->end = res->start + IO_SPACE_LIMIT;
1541 res->flags = IORESOURCE_IO;
1542 #endif /* CONFIG_PPC32 */
1545 pr_debug("PCI: PHB IO resource = %016llx-%016llx [%lx]\n",
1546 (unsigned long long)res->start,
1547 (unsigned long long)res->end,
1548 (unsigned long)res->flags);
1550 /* Hookup PHB Memory resources */
1551 for (i = 0; i < 3; ++i) {
1552 res = &hose->mem_resources[i];
1556 printk(KERN_ERR "PCI: Memory resource 0 not set for "
1557 "host bridge %s (domain %d)\n",
1558 hose->dn->full_name, hose->global_number);
1560 /* Workaround for lack of MEM resource only on 32-bit */
1561 res->start = hose->pci_mem_offset;
1562 res->end = (resource_size_t)-1LL;
1563 res->flags = IORESOURCE_MEM;
1564 #endif /* CONFIG_PPC32 */
1566 bus->resource[i+1] = res;
1568 pr_debug("PCI: PHB MEM resource %d = %016llx-%016llx [%lx]\n", i,
1569 (unsigned long long)res->start,
1570 (unsigned long long)res->end,
1571 (unsigned long)res->flags);
1574 pr_debug("PCI: PHB MEM offset = %016llx\n",
1575 (unsigned long long)hose->pci_mem_offset);
1576 pr_debug("PCI: PHB IO offset = %08lx\n",
1577 (unsigned long)hose->io_base_virt - _IO_BASE);