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.
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>
27 #include <linux/list.h>
28 #include <linux/syscalls.h>
29 #include <linux/irq.h>
30 #include <linux/vmalloc.h>
32 #include <asm/processor.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>
42 static DEFINE_SPINLOCK(hose_spinlock);
44 /* XXX kill that some day ... */
45 static int global_phb_number; /* Global phb counter */
47 /* ISA Memory physical address */
48 resource_size_t isa_mem_base;
50 /* Default PCI flags is 0 on ppc32, modified at boot on ppc64 */
51 unsigned int ppc_pci_flags = 0;
54 static struct dma_mapping_ops *pci_dma_ops;
56 void set_pci_dma_ops(struct dma_mapping_ops *dma_ops)
58 pci_dma_ops = dma_ops;
61 struct dma_mapping_ops *get_pci_dma_ops(void)
65 EXPORT_SYMBOL(get_pci_dma_ops);
67 int pci_set_dma_mask(struct pci_dev *dev, u64 mask)
69 return dma_set_mask(&dev->dev, mask);
72 int pci_set_consistent_dma_mask(struct pci_dev *dev, u64 mask)
76 rc = dma_set_mask(&dev->dev, mask);
77 dev->dev.coherent_dma_mask = dev->dma_mask;
82 struct pci_controller *pcibios_alloc_controller(struct device_node *dev)
84 struct pci_controller *phb;
86 phb = zalloc_maybe_bootmem(sizeof(struct pci_controller), GFP_KERNEL);
89 spin_lock(&hose_spinlock);
90 phb->global_number = global_phb_number++;
91 list_add_tail(&phb->list_node, &hose_list);
92 spin_unlock(&hose_spinlock);
94 phb->is_dynamic = mem_init_done;
97 int nid = of_node_to_nid(dev);
99 if (nid < 0 || !node_online(nid))
102 PHB_SET_NODE(phb, nid);
108 void pcibios_free_controller(struct pci_controller *phb)
110 spin_lock(&hose_spinlock);
111 list_del(&phb->list_node);
112 spin_unlock(&hose_spinlock);
118 int pcibios_vaddr_is_ioport(void __iomem *address)
121 struct pci_controller *hose;
124 spin_lock(&hose_spinlock);
125 list_for_each_entry(hose, &hose_list, list_node) {
127 size = hose->pci_io_size;
129 size = hose->io_resource.end - hose->io_resource.start + 1;
131 if (address >= hose->io_base_virt &&
132 address < (hose->io_base_virt + size)) {
137 spin_unlock(&hose_spinlock);
142 * Return the domain number for this bus.
144 int pci_domain_nr(struct pci_bus *bus)
146 struct pci_controller *hose = pci_bus_to_host(bus);
148 return hose->global_number;
150 EXPORT_SYMBOL(pci_domain_nr);
154 /* This routine is meant to be used early during boot, when the
155 * PCI bus numbers have not yet been assigned, and you need to
156 * issue PCI config cycles to an OF device.
157 * It could also be used to "fix" RTAS config cycles if you want
158 * to set pci_assign_all_buses to 1 and still use RTAS for PCI
161 struct pci_controller* pci_find_hose_for_OF_device(struct device_node* node)
164 struct pci_controller *hose, *tmp;
165 list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
166 if (hose->dn == node)
173 static ssize_t pci_show_devspec(struct device *dev,
174 struct device_attribute *attr, char *buf)
176 struct pci_dev *pdev;
177 struct device_node *np;
179 pdev = to_pci_dev (dev);
180 np = pci_device_to_OF_node(pdev);
181 if (np == NULL || np->full_name == NULL)
183 return sprintf(buf, "%s", np->full_name);
185 static DEVICE_ATTR(devspec, S_IRUGO, pci_show_devspec, NULL);
186 #endif /* CONFIG_PPC_OF */
188 /* Add sysfs properties */
189 int pcibios_add_platform_entries(struct pci_dev *pdev)
192 return device_create_file(&pdev->dev, &dev_attr_devspec);
195 #endif /* CONFIG_PPC_OF */
199 char __devinit *pcibios_setup(char *str)
205 * Reads the interrupt pin to determine if interrupt is use by card.
206 * If the interrupt is used, then gets the interrupt line from the
207 * openfirmware and sets it in the pci_dev and pci_config line.
209 int pci_read_irq_line(struct pci_dev *pci_dev)
214 /* The current device-tree that iSeries generates from the HV
215 * PCI informations doesn't contain proper interrupt routing,
216 * and all the fallback would do is print out crap, so we
217 * don't attempt to resolve the interrupts here at all, some
218 * iSeries specific fixup does it.
220 * In the long run, we will hopefully fix the generated device-tree
223 #ifdef CONFIG_PPC_ISERIES
224 if (firmware_has_feature(FW_FEATURE_ISERIES))
228 pr_debug("PCI: Try to map irq for %s...\n", pci_name(pci_dev));
231 memset(&oirq, 0xff, sizeof(oirq));
233 /* Try to get a mapping from the device-tree */
234 if (of_irq_map_pci(pci_dev, &oirq)) {
237 /* If that fails, lets fallback to what is in the config
238 * space and map that through the default controller. We
239 * also set the type to level low since that's what PCI
240 * interrupts are. If your platform does differently, then
241 * either provide a proper interrupt tree or don't use this
244 if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_PIN, &pin))
248 if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_LINE, &line) ||
249 line == 0xff || line == 0) {
252 pr_debug(" No map ! Using line %d (pin %d) from PCI config\n",
255 virq = irq_create_mapping(NULL, line);
257 set_irq_type(virq, IRQ_TYPE_LEVEL_LOW);
259 pr_debug(" Got one, spec %d cells (0x%08x 0x%08x...) on %s\n",
260 oirq.size, oirq.specifier[0], oirq.specifier[1],
261 oirq.controller->full_name);
263 virq = irq_create_of_mapping(oirq.controller, oirq.specifier,
267 pr_debug(" Failed to map !\n");
271 pr_debug(" Mapped to linux irq %d\n", virq);
277 EXPORT_SYMBOL(pci_read_irq_line);
280 * Platform support for /proc/bus/pci/X/Y mmap()s,
281 * modelled on the sparc64 implementation by Dave Miller.
286 * Adjust vm_pgoff of VMA such that it is the physical page offset
287 * corresponding to the 32-bit pci bus offset for DEV requested by the user.
289 * Basically, the user finds the base address for his device which he wishes
290 * to mmap. They read the 32-bit value from the config space base register,
291 * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
292 * offset parameter of mmap on /proc/bus/pci/XXX for that device.
294 * Returns negative error code on failure, zero on success.
296 static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
297 resource_size_t *offset,
298 enum pci_mmap_state mmap_state)
300 struct pci_controller *hose = pci_bus_to_host(dev->bus);
301 unsigned long io_offset = 0;
305 return NULL; /* should never happen */
307 /* If memory, add on the PCI bridge address offset */
308 if (mmap_state == pci_mmap_mem) {
309 #if 0 /* See comment in pci_resource_to_user() for why this is disabled */
310 *offset += hose->pci_mem_offset;
312 res_bit = IORESOURCE_MEM;
314 io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
315 *offset += io_offset;
316 res_bit = IORESOURCE_IO;
320 * Check that the offset requested corresponds to one of the
321 * resources of the device.
323 for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
324 struct resource *rp = &dev->resource[i];
325 int flags = rp->flags;
327 /* treat ROM as memory (should be already) */
328 if (i == PCI_ROM_RESOURCE)
329 flags |= IORESOURCE_MEM;
331 /* Active and same type? */
332 if ((flags & res_bit) == 0)
335 /* In the range of this resource? */
336 if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
339 /* found it! construct the final physical address */
340 if (mmap_state == pci_mmap_io)
341 *offset += hose->io_base_phys - io_offset;
349 * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
352 static pgprot_t __pci_mmap_set_pgprot(struct pci_dev *dev, struct resource *rp,
354 enum pci_mmap_state mmap_state,
357 unsigned long prot = pgprot_val(protection);
359 /* Write combine is always 0 on non-memory space mappings. On
360 * memory space, if the user didn't pass 1, we check for a
361 * "prefetchable" resource. This is a bit hackish, but we use
362 * this to workaround the inability of /sysfs to provide a write
365 if (mmap_state != pci_mmap_mem)
367 else if (write_combine == 0) {
368 if (rp->flags & IORESOURCE_PREFETCH)
372 /* XXX would be nice to have a way to ask for write-through */
374 return pgprot_noncached_wc(prot);
376 return pgprot_noncached(prot);
380 * This one is used by /dev/mem and fbdev who have no clue about the
381 * PCI device, it tries to find the PCI device first and calls the
384 pgprot_t pci_phys_mem_access_prot(struct file *file,
389 struct pci_dev *pdev = NULL;
390 struct resource *found = NULL;
391 resource_size_t offset = ((resource_size_t)pfn) << PAGE_SHIFT;
394 if (page_is_ram(pfn))
397 prot = pgprot_noncached(prot);
398 for_each_pci_dev(pdev) {
399 for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
400 struct resource *rp = &pdev->resource[i];
401 int flags = rp->flags;
403 /* Active and same type? */
404 if ((flags & IORESOURCE_MEM) == 0)
406 /* In the range of this resource? */
407 if (offset < (rp->start & PAGE_MASK) ||
417 if (found->flags & IORESOURCE_PREFETCH)
418 prot = pgprot_noncached_wc(prot);
422 pr_debug("PCI: Non-PCI map for %llx, prot: %lx\n",
423 (unsigned long long)offset, pgprot_val(prot));
430 * Perform the actual remap of the pages for a PCI device mapping, as
431 * appropriate for this architecture. The region in the process to map
432 * is described by vm_start and vm_end members of VMA, the base physical
433 * address is found in vm_pgoff.
434 * The pci device structure is provided so that architectures may make mapping
435 * decisions on a per-device or per-bus basis.
437 * Returns a negative error code on failure, zero on success.
439 int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
440 enum pci_mmap_state mmap_state, int write_combine)
442 resource_size_t offset =
443 ((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
447 rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
451 vma->vm_pgoff = offset >> PAGE_SHIFT;
452 vma->vm_page_prot = __pci_mmap_set_pgprot(dev, rp,
454 mmap_state, write_combine);
456 ret = remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
457 vma->vm_end - vma->vm_start, vma->vm_page_prot);
462 /* This provides legacy IO read access on a bus */
463 int pci_legacy_read(struct pci_bus *bus, loff_t port, u32 *val, size_t size)
465 unsigned long offset;
466 struct pci_controller *hose = pci_bus_to_host(bus);
467 struct resource *rp = &hose->io_resource;
470 /* Check if port can be supported by that bus. We only check
471 * the ranges of the PHB though, not the bus itself as the rules
472 * for forwarding legacy cycles down bridges are not our problem
473 * here. So if the host bridge supports it, we do it.
475 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
478 if (!(rp->flags & IORESOURCE_IO))
480 if (offset < rp->start || (offset + size) > rp->end)
482 addr = hose->io_base_virt + port;
486 *((u8 *)val) = in_8(addr);
491 *((u16 *)val) = in_le16(addr);
496 *((u32 *)val) = in_le32(addr);
502 /* This provides legacy IO write access on a bus */
503 int pci_legacy_write(struct pci_bus *bus, loff_t port, u32 val, size_t size)
505 unsigned long offset;
506 struct pci_controller *hose = pci_bus_to_host(bus);
507 struct resource *rp = &hose->io_resource;
510 /* Check if port can be supported by that bus. We only check
511 * the ranges of the PHB though, not the bus itself as the rules
512 * for forwarding legacy cycles down bridges are not our problem
513 * here. So if the host bridge supports it, we do it.
515 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
518 if (!(rp->flags & IORESOURCE_IO))
520 if (offset < rp->start || (offset + size) > rp->end)
522 addr = hose->io_base_virt + port;
524 /* WARNING: The generic code is idiotic. It gets passed a pointer
525 * to what can be a 1, 2 or 4 byte quantity and always reads that
526 * as a u32, which means that we have to correct the location of
527 * the data read within those 32 bits for size 1 and 2
531 out_8(addr, val >> 24);
536 out_le16(addr, val >> 16);
547 /* This provides legacy IO or memory mmap access on a bus */
548 int pci_mmap_legacy_page_range(struct pci_bus *bus,
549 struct vm_area_struct *vma,
550 enum pci_mmap_state mmap_state)
552 struct pci_controller *hose = pci_bus_to_host(bus);
553 resource_size_t offset =
554 ((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
555 resource_size_t size = vma->vm_end - vma->vm_start;
558 pr_debug("pci_mmap_legacy_page_range(%04x:%02x, %s @%llx..%llx)\n",
559 pci_domain_nr(bus), bus->number,
560 mmap_state == pci_mmap_mem ? "MEM" : "IO",
561 (unsigned long long)offset,
562 (unsigned long long)(offset + size - 1));
564 if (mmap_state == pci_mmap_mem) {
565 if ((offset + size) > hose->isa_mem_size)
567 offset += hose->isa_mem_phys;
569 unsigned long io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
570 unsigned long roffset = offset + io_offset;
571 rp = &hose->io_resource;
572 if (!(rp->flags & IORESOURCE_IO))
574 if (roffset < rp->start || (roffset + size) > rp->end)
576 offset += hose->io_base_phys;
578 pr_debug(" -> mapping phys %llx\n", (unsigned long long)offset);
580 vma->vm_pgoff = offset >> PAGE_SHIFT;
581 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
582 return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
583 vma->vm_end - vma->vm_start,
587 void pci_resource_to_user(const struct pci_dev *dev, int bar,
588 const struct resource *rsrc,
589 resource_size_t *start, resource_size_t *end)
591 struct pci_controller *hose = pci_bus_to_host(dev->bus);
592 resource_size_t offset = 0;
597 if (rsrc->flags & IORESOURCE_IO)
598 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
600 /* We pass a fully fixed up address to userland for MMIO instead of
601 * a BAR value because X is lame and expects to be able to use that
602 * to pass to /dev/mem !
604 * That means that we'll have potentially 64 bits values where some
605 * userland apps only expect 32 (like X itself since it thinks only
606 * Sparc has 64 bits MMIO) but if we don't do that, we break it on
609 * Hopefully, the sysfs insterface is immune to that gunk. Once X
610 * has been fixed (and the fix spread enough), we can re-enable the
611 * 2 lines below and pass down a BAR value to userland. In that case
612 * we'll also have to re-enable the matching code in
613 * __pci_mmap_make_offset().
618 else if (rsrc->flags & IORESOURCE_MEM)
619 offset = hose->pci_mem_offset;
622 *start = rsrc->start - offset;
623 *end = rsrc->end - offset;
627 * pci_process_bridge_OF_ranges - Parse PCI bridge resources from device tree
628 * @hose: newly allocated pci_controller to be setup
629 * @dev: device node of the host bridge
630 * @primary: set if primary bus (32 bits only, soon to be deprecated)
632 * This function will parse the "ranges" property of a PCI host bridge device
633 * node and setup the resource mapping of a pci controller based on its
636 * Life would be boring if it wasn't for a few issues that we have to deal
639 * - We can only cope with one IO space range and up to 3 Memory space
640 * ranges. However, some machines (thanks Apple !) tend to split their
641 * space into lots of small contiguous ranges. So we have to coalesce.
643 * - We can only cope with all memory ranges having the same offset
644 * between CPU addresses and PCI addresses. Unfortunately, some bridges
645 * are setup for a large 1:1 mapping along with a small "window" which
646 * maps PCI address 0 to some arbitrary high address of the CPU space in
647 * order to give access to the ISA memory hole.
648 * The way out of here that I've chosen for now is to always set the
649 * offset based on the first resource found, then override it if we
650 * have a different offset and the previous was set by an ISA hole.
652 * - Some busses have IO space not starting at 0, which causes trouble with
653 * the way we do our IO resource renumbering. The code somewhat deals with
654 * it for 64 bits but I would expect problems on 32 bits.
656 * - Some 32 bits platforms such as 4xx can have physical space larger than
657 * 32 bits so we need to use 64 bits values for the parsing
659 void __devinit pci_process_bridge_OF_ranges(struct pci_controller *hose,
660 struct device_node *dev,
665 int pna = of_n_addr_cells(dev);
667 int memno = 0, isa_hole = -1;
669 unsigned long long pci_addr, cpu_addr, pci_next, cpu_next, size;
670 unsigned long long isa_mb = 0;
671 struct resource *res;
673 printk(KERN_INFO "PCI host bridge %s %s ranges:\n",
674 dev->full_name, primary ? "(primary)" : "");
676 /* Get ranges property */
677 ranges = of_get_property(dev, "ranges", &rlen);
682 while ((rlen -= np * 4) >= 0) {
683 /* Read next ranges element */
684 pci_space = ranges[0];
685 pci_addr = of_read_number(ranges + 1, 2);
686 cpu_addr = of_translate_address(dev, ranges + 3);
687 size = of_read_number(ranges + pna + 3, 2);
690 /* If we failed translation or got a zero-sized region
691 * (some FW try to feed us with non sensical zero sized regions
692 * such as power3 which look like some kind of attempt at exposing
693 * the VGA memory hole)
695 if (cpu_addr == OF_BAD_ADDR || size == 0)
698 /* Now consume following elements while they are contiguous */
699 for (; rlen >= np * sizeof(u32);
700 ranges += np, rlen -= np * 4) {
701 if (ranges[0] != pci_space)
703 pci_next = of_read_number(ranges + 1, 2);
704 cpu_next = of_translate_address(dev, ranges + 3);
705 if (pci_next != pci_addr + size ||
706 cpu_next != cpu_addr + size)
708 size += of_read_number(ranges + pna + 3, 2);
711 /* Act based on address space type */
713 switch ((pci_space >> 24) & 0x3) {
714 case 1: /* PCI IO space */
716 " IO 0x%016llx..0x%016llx -> 0x%016llx\n",
717 cpu_addr, cpu_addr + size - 1, pci_addr);
719 /* We support only one IO range */
720 if (hose->pci_io_size) {
722 " \\--> Skipped (too many) !\n");
726 /* On 32 bits, limit I/O space to 16MB */
727 if (size > 0x01000000)
730 /* 32 bits needs to map IOs here */
731 hose->io_base_virt = ioremap(cpu_addr, size);
733 /* Expect trouble if pci_addr is not 0 */
736 (unsigned long)hose->io_base_virt;
737 #endif /* CONFIG_PPC32 */
738 /* pci_io_size and io_base_phys always represent IO
739 * space starting at 0 so we factor in pci_addr
741 hose->pci_io_size = pci_addr + size;
742 hose->io_base_phys = cpu_addr - pci_addr;
745 res = &hose->io_resource;
746 res->flags = IORESOURCE_IO;
747 res->start = pci_addr;
749 case 2: /* PCI Memory space */
750 case 3: /* PCI 64 bits Memory space */
752 " MEM 0x%016llx..0x%016llx -> 0x%016llx %s\n",
753 cpu_addr, cpu_addr + size - 1, pci_addr,
754 (pci_space & 0x40000000) ? "Prefetch" : "");
756 /* We support only 3 memory ranges */
759 " \\--> Skipped (too many) !\n");
762 /* Handles ISA memory hole space here */
766 if (primary || isa_mem_base == 0)
767 isa_mem_base = cpu_addr;
768 hose->isa_mem_phys = cpu_addr;
769 hose->isa_mem_size = size;
772 /* We get the PCI/Mem offset from the first range or
773 * the, current one if the offset came from an ISA
774 * hole. If they don't match, bugger.
777 (isa_hole >= 0 && pci_addr != 0 &&
778 hose->pci_mem_offset == isa_mb))
779 hose->pci_mem_offset = cpu_addr - pci_addr;
780 else if (pci_addr != 0 &&
781 hose->pci_mem_offset != cpu_addr - pci_addr) {
783 " \\--> Skipped (offset mismatch) !\n");
788 res = &hose->mem_resources[memno++];
789 res->flags = IORESOURCE_MEM;
790 if (pci_space & 0x40000000)
791 res->flags |= IORESOURCE_PREFETCH;
792 res->start = cpu_addr;
796 res->name = dev->full_name;
797 res->end = res->start + size - 1;
804 /* If there's an ISA hole and the pci_mem_offset is -not- matching
805 * the ISA hole offset, then we need to remove the ISA hole from
806 * the resource list for that brige
808 if (isa_hole >= 0 && hose->pci_mem_offset != isa_mb) {
809 unsigned int next = isa_hole + 1;
810 printk(KERN_INFO " Removing ISA hole at 0x%016llx\n", isa_mb);
812 memmove(&hose->mem_resources[isa_hole],
813 &hose->mem_resources[next],
814 sizeof(struct resource) * (memno - next));
815 hose->mem_resources[--memno].flags = 0;
819 /* Decide whether to display the domain number in /proc */
820 int pci_proc_domain(struct pci_bus *bus)
822 struct pci_controller *hose = pci_bus_to_host(bus);
824 if (!(ppc_pci_flags & PPC_PCI_ENABLE_PROC_DOMAINS))
826 if (ppc_pci_flags & PPC_PCI_COMPAT_DOMAIN_0)
827 return hose->global_number != 0;
831 void pcibios_resource_to_bus(struct pci_dev *dev, struct pci_bus_region *region,
832 struct resource *res)
834 resource_size_t offset = 0, mask = (resource_size_t)-1;
835 struct pci_controller *hose = pci_bus_to_host(dev->bus);
839 if (res->flags & IORESOURCE_IO) {
840 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
842 } else if (res->flags & IORESOURCE_MEM)
843 offset = hose->pci_mem_offset;
845 region->start = (res->start - offset) & mask;
846 region->end = (res->end - offset) & mask;
848 EXPORT_SYMBOL(pcibios_resource_to_bus);
850 void pcibios_bus_to_resource(struct pci_dev *dev, struct resource *res,
851 struct pci_bus_region *region)
853 resource_size_t offset = 0, mask = (resource_size_t)-1;
854 struct pci_controller *hose = pci_bus_to_host(dev->bus);
858 if (res->flags & IORESOURCE_IO) {
859 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
861 } else if (res->flags & IORESOURCE_MEM)
862 offset = hose->pci_mem_offset;
863 res->start = (region->start + offset) & mask;
864 res->end = (region->end + offset) & mask;
866 EXPORT_SYMBOL(pcibios_bus_to_resource);
868 /* Fixup a bus resource into a linux resource */
869 static void __devinit fixup_resource(struct resource *res, struct pci_dev *dev)
871 struct pci_controller *hose = pci_bus_to_host(dev->bus);
872 resource_size_t offset = 0, mask = (resource_size_t)-1;
874 if (res->flags & IORESOURCE_IO) {
875 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
877 } else if (res->flags & IORESOURCE_MEM)
878 offset = hose->pci_mem_offset;
880 res->start = (res->start + offset) & mask;
881 res->end = (res->end + offset) & mask;
885 /* This header fixup will do the resource fixup for all devices as they are
886 * probed, but not for bridge ranges
888 static void __devinit pcibios_fixup_resources(struct pci_dev *dev)
890 struct pci_controller *hose = pci_bus_to_host(dev->bus);
894 printk(KERN_ERR "No host bridge for PCI dev %s !\n",
898 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
899 struct resource *res = dev->resource + i;
902 /* On platforms that have PPC_PCI_PROBE_ONLY set, we don't
903 * consider 0 as an unassigned BAR value. It's technically
904 * a valid value, but linux doesn't like it... so when we can
905 * re-assign things, we do so, but if we can't, we keep it
906 * around and hope for the best...
908 if (res->start == 0 && !(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
909 pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] is unassigned\n",
911 (unsigned long long)res->start,
912 (unsigned long long)res->end,
913 (unsigned int)res->flags);
914 res->end -= res->start;
916 res->flags |= IORESOURCE_UNSET;
920 pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] fixup...\n",
922 (unsigned long long)res->start,\
923 (unsigned long long)res->end,
924 (unsigned int)res->flags);
926 fixup_resource(res, dev);
928 pr_debug("PCI:%s %016llx-%016llx\n",
930 (unsigned long long)res->start,
931 (unsigned long long)res->end);
934 /* Call machine specific resource fixup */
935 if (ppc_md.pcibios_fixup_resources)
936 ppc_md.pcibios_fixup_resources(dev);
938 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_resources);
940 /* This function tries to figure out if a bridge resource has been initialized
941 * by the firmware or not. It doesn't have to be absolutely bullet proof, but
942 * things go more smoothly when it gets it right. It should covers cases such
943 * as Apple "closed" bridge resources and bare-metal pSeries unassigned bridges
945 static int __devinit pcibios_uninitialized_bridge_resource(struct pci_bus *bus,
946 struct resource *res)
948 struct pci_controller *hose = pci_bus_to_host(bus);
949 struct pci_dev *dev = bus->self;
950 resource_size_t offset;
954 /* We don't do anything if PCI_PROBE_ONLY is set */
955 if (ppc_pci_flags & PPC_PCI_PROBE_ONLY)
958 /* Job is a bit different between memory and IO */
959 if (res->flags & IORESOURCE_MEM) {
960 /* If the BAR is non-0 (res != pci_mem_offset) then it's probably been
961 * initialized by somebody
963 if (res->start != hose->pci_mem_offset)
966 /* The BAR is 0, let's check if memory decoding is enabled on
967 * the bridge. If not, we consider it unassigned
969 pci_read_config_word(dev, PCI_COMMAND, &command);
970 if ((command & PCI_COMMAND_MEMORY) == 0)
973 /* Memory decoding is enabled and the BAR is 0. If any of the bridge
974 * resources covers that starting address (0 then it's good enough for
977 for (i = 0; i < 3; i++) {
978 if ((hose->mem_resources[i].flags & IORESOURCE_MEM) &&
979 hose->mem_resources[i].start == hose->pci_mem_offset)
983 /* Well, it starts at 0 and we know it will collide so we may as
984 * well consider it as unassigned. That covers the Apple case.
988 /* If the BAR is non-0, then we consider it assigned */
989 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
990 if (((res->start - offset) & 0xfffffffful) != 0)
993 /* Here, we are a bit different than memory as typically IO space
994 * starting at low addresses -is- valid. What we do instead if that
995 * we consider as unassigned anything that doesn't have IO enabled
996 * in the PCI command register, and that's it.
998 pci_read_config_word(dev, PCI_COMMAND, &command);
999 if (command & PCI_COMMAND_IO)
1002 /* It's starting at 0 and IO is disabled in the bridge, consider
1009 /* Fixup resources of a PCI<->PCI bridge */
1010 static void __devinit pcibios_fixup_bridge(struct pci_bus *bus)
1012 struct resource *res;
1015 struct pci_dev *dev = bus->self;
1017 for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
1018 if ((res = bus->resource[i]) == NULL)
1022 if (i >= 3 && bus->self->transparent)
1025 pr_debug("PCI:%s Bus rsrc %d %016llx-%016llx [%x] fixup...\n",
1027 (unsigned long long)res->start,\
1028 (unsigned long long)res->end,
1029 (unsigned int)res->flags);
1032 fixup_resource(res, dev);
1034 /* Try to detect uninitialized P2P bridge resources,
1035 * and clear them out so they get re-assigned later
1037 if (pcibios_uninitialized_bridge_resource(bus, res)) {
1039 pr_debug("PCI:%s (unassigned)\n", pci_name(dev));
1042 pr_debug("PCI:%s %016llx-%016llx\n",
1044 (unsigned long long)res->start,
1045 (unsigned long long)res->end);
1050 void __devinit pcibios_setup_bus_self(struct pci_bus *bus)
1052 /* Fix up the bus resources for P2P bridges */
1053 if (bus->self != NULL)
1054 pcibios_fixup_bridge(bus);
1056 /* Platform specific bus fixups. This is currently only used
1057 * by fsl_pci and I'm hoping to get rid of it at some point
1059 if (ppc_md.pcibios_fixup_bus)
1060 ppc_md.pcibios_fixup_bus(bus);
1062 /* Setup bus DMA mappings */
1063 if (ppc_md.pci_dma_bus_setup)
1064 ppc_md.pci_dma_bus_setup(bus);
1067 void __devinit pcibios_setup_bus_devices(struct pci_bus *bus)
1069 struct pci_dev *dev;
1071 pr_debug("PCI: Fixup bus devices %d (%s)\n",
1072 bus->number, bus->self ? pci_name(bus->self) : "PHB");
1074 list_for_each_entry(dev, &bus->devices, bus_list) {
1075 struct dev_archdata *sd = &dev->dev.archdata;
1077 /* Setup OF node pointer in archdata */
1078 sd->of_node = pci_device_to_OF_node(dev);
1080 /* Fixup NUMA node as it may not be setup yet by the generic
1081 * code and is needed by the DMA init
1083 set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
1085 /* Hook up default DMA ops */
1086 sd->dma_ops = pci_dma_ops;
1087 sd->dma_data = (void *)PCI_DRAM_OFFSET;
1089 /* Additional platform DMA/iommu setup */
1090 if (ppc_md.pci_dma_dev_setup)
1091 ppc_md.pci_dma_dev_setup(dev);
1093 /* Read default IRQs and fixup if necessary */
1094 pci_read_irq_line(dev);
1095 if (ppc_md.pci_irq_fixup)
1096 ppc_md.pci_irq_fixup(dev);
1100 void __devinit pcibios_fixup_bus(struct pci_bus *bus)
1102 /* When called from the generic PCI probe, read PCI<->PCI bridge
1103 * bases. This is -not- called when generating the PCI tree from
1104 * the OF device-tree.
1106 if (bus->self != NULL)
1107 pci_read_bridge_bases(bus);
1109 /* Now fixup the bus bus */
1110 pcibios_setup_bus_self(bus);
1112 /* Now fixup devices on that bus */
1113 pcibios_setup_bus_devices(bus);
1115 EXPORT_SYMBOL(pcibios_fixup_bus);
1117 static int skip_isa_ioresource_align(struct pci_dev *dev)
1119 if ((ppc_pci_flags & PPC_PCI_CAN_SKIP_ISA_ALIGN) &&
1120 !(dev->bus->bridge_ctl & PCI_BRIDGE_CTL_ISA))
1126 * We need to avoid collisions with `mirrored' VGA ports
1127 * and other strange ISA hardware, so we always want the
1128 * addresses to be allocated in the 0x000-0x0ff region
1131 * Why? Because some silly external IO cards only decode
1132 * the low 10 bits of the IO address. The 0x00-0xff region
1133 * is reserved for motherboard devices that decode all 16
1134 * bits, so it's ok to allocate at, say, 0x2800-0x28ff,
1135 * but we want to try to avoid allocating at 0x2900-0x2bff
1136 * which might have be mirrored at 0x0100-0x03ff..
1138 void pcibios_align_resource(void *data, struct resource *res,
1139 resource_size_t size, resource_size_t align)
1141 struct pci_dev *dev = data;
1143 if (res->flags & IORESOURCE_IO) {
1144 resource_size_t start = res->start;
1146 if (skip_isa_ioresource_align(dev))
1148 if (start & 0x300) {
1149 start = (start + 0x3ff) & ~0x3ff;
1154 EXPORT_SYMBOL(pcibios_align_resource);
1157 * Reparent resource children of pr that conflict with res
1158 * under res, and make res replace those children.
1160 static int __init reparent_resources(struct resource *parent,
1161 struct resource *res)
1163 struct resource *p, **pp;
1164 struct resource **firstpp = NULL;
1166 for (pp = &parent->child; (p = *pp) != NULL; pp = &p->sibling) {
1167 if (p->end < res->start)
1169 if (res->end < p->start)
1171 if (p->start < res->start || p->end > res->end)
1172 return -1; /* not completely contained */
1173 if (firstpp == NULL)
1176 if (firstpp == NULL)
1177 return -1; /* didn't find any conflicting entries? */
1178 res->parent = parent;
1179 res->child = *firstpp;
1183 for (p = res->child; p != NULL; p = p->sibling) {
1185 pr_debug("PCI: Reparented %s [%llx..%llx] under %s\n",
1187 (unsigned long long)p->start,
1188 (unsigned long long)p->end, res->name);
1194 * Handle resources of PCI devices. If the world were perfect, we could
1195 * just allocate all the resource regions and do nothing more. It isn't.
1196 * On the other hand, we cannot just re-allocate all devices, as it would
1197 * require us to know lots of host bridge internals. So we attempt to
1198 * keep as much of the original configuration as possible, but tweak it
1199 * when it's found to be wrong.
1201 * Known BIOS problems we have to work around:
1202 * - I/O or memory regions not configured
1203 * - regions configured, but not enabled in the command register
1204 * - bogus I/O addresses above 64K used
1205 * - expansion ROMs left enabled (this may sound harmless, but given
1206 * the fact the PCI specs explicitly allow address decoders to be
1207 * shared between expansion ROMs and other resource regions, it's
1208 * at least dangerous)
1211 * (1) Allocate resources for all buses behind PCI-to-PCI bridges.
1212 * This gives us fixed barriers on where we can allocate.
1213 * (2) Allocate resources for all enabled devices. If there is
1214 * a collision, just mark the resource as unallocated. Also
1215 * disable expansion ROMs during this step.
1216 * (3) Try to allocate resources for disabled devices. If the
1217 * resources were assigned correctly, everything goes well,
1218 * if they weren't, they won't disturb allocation of other
1220 * (4) Assign new addresses to resources which were either
1221 * not configured at all or misconfigured. If explicitly
1222 * requested by the user, configure expansion ROM address
1226 void pcibios_allocate_bus_resources(struct pci_bus *bus)
1230 struct resource *res, *pr;
1232 pr_debug("PCI: Allocating bus resources for %04x:%02x...\n",
1233 pci_domain_nr(bus), bus->number);
1235 for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
1236 if ((res = bus->resource[i]) == NULL || !res->flags
1237 || res->start > res->end || res->parent)
1239 if (bus->parent == NULL)
1240 pr = (res->flags & IORESOURCE_IO) ?
1241 &ioport_resource : &iomem_resource;
1243 /* Don't bother with non-root busses when
1244 * re-assigning all resources. We clear the
1245 * resource flags as if they were colliding
1246 * and as such ensure proper re-allocation
1249 if (ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)
1250 goto clear_resource;
1251 pr = pci_find_parent_resource(bus->self, res);
1253 /* this happens when the generic PCI
1254 * code (wrongly) decides that this
1255 * bridge is transparent -- paulus
1261 pr_debug("PCI: %s (bus %d) bridge rsrc %d: %016llx-%016llx "
1262 "[0x%x], parent %p (%s)\n",
1263 bus->self ? pci_name(bus->self) : "PHB",
1265 (unsigned long long)res->start,
1266 (unsigned long long)res->end,
1267 (unsigned int)res->flags,
1268 pr, (pr && pr->name) ? pr->name : "nil");
1270 if (pr && !(pr->flags & IORESOURCE_UNSET)) {
1271 if (request_resource(pr, res) == 0)
1274 * Must be a conflict with an existing entry.
1275 * Move that entry (or entries) under the
1276 * bridge resource and try again.
1278 if (reparent_resources(pr, res) == 0)
1281 printk(KERN_WARNING "PCI: Cannot allocate resource region "
1282 "%d of PCI bridge %d, will remap\n", i, bus->number);
1287 list_for_each_entry(b, &bus->children, node)
1288 pcibios_allocate_bus_resources(b);
1291 static inline void __devinit alloc_resource(struct pci_dev *dev, int idx)
1293 struct resource *pr, *r = &dev->resource[idx];
1295 pr_debug("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1297 (unsigned long long)r->start,
1298 (unsigned long long)r->end,
1299 (unsigned int)r->flags);
1301 pr = pci_find_parent_resource(dev, r);
1302 if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1303 request_resource(pr, r) < 0) {
1304 printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1305 " of device %s, will remap\n", idx, pci_name(dev));
1307 pr_debug("PCI: parent is %p: %016llx-%016llx [%x]\n",
1309 (unsigned long long)pr->start,
1310 (unsigned long long)pr->end,
1311 (unsigned int)pr->flags);
1312 /* We'll assign a new address later */
1313 r->flags |= IORESOURCE_UNSET;
1319 static void __init pcibios_allocate_resources(int pass)
1321 struct pci_dev *dev = NULL;
1326 for_each_pci_dev(dev) {
1327 pci_read_config_word(dev, PCI_COMMAND, &command);
1328 for (idx = 0; idx < 6; idx++) {
1329 r = &dev->resource[idx];
1330 if (r->parent) /* Already allocated */
1332 if (!r->flags || (r->flags & IORESOURCE_UNSET))
1333 continue; /* Not assigned at all */
1334 if (r->flags & IORESOURCE_IO)
1335 disabled = !(command & PCI_COMMAND_IO);
1337 disabled = !(command & PCI_COMMAND_MEMORY);
1338 if (pass == disabled)
1339 alloc_resource(dev, idx);
1343 r = &dev->resource[PCI_ROM_RESOURCE];
1344 if (r->flags & IORESOURCE_ROM_ENABLE) {
1345 /* Turn the ROM off, leave the resource region,
1346 * but keep it unregistered.
1349 pr_debug("PCI: Switching off ROM of %s\n",
1351 r->flags &= ~IORESOURCE_ROM_ENABLE;
1352 pci_read_config_dword(dev, dev->rom_base_reg, ®);
1353 pci_write_config_dword(dev, dev->rom_base_reg,
1354 reg & ~PCI_ROM_ADDRESS_ENABLE);
1359 static void __init pcibios_reserve_legacy_regions(struct pci_bus *bus)
1361 struct pci_controller *hose = pci_bus_to_host(bus);
1362 resource_size_t offset;
1363 struct resource *res, *pres;
1366 pr_debug("Reserving legacy ranges for domain %04x\n", pci_domain_nr(bus));
1369 if (!(hose->io_resource.flags & IORESOURCE_IO))
1371 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1372 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1373 BUG_ON(res == NULL);
1374 res->name = "Legacy IO";
1375 res->flags = IORESOURCE_IO;
1376 res->start = offset;
1377 res->end = (offset + 0xfff) & 0xfffffffful;
1378 pr_debug("Candidate legacy IO: %pR\n", res);
1379 if (request_resource(&hose->io_resource, res)) {
1381 "PCI %04x:%02x Cannot reserve Legacy IO %pR\n",
1382 pci_domain_nr(bus), bus->number, res);
1387 /* Check for memory */
1388 offset = hose->pci_mem_offset;
1389 pr_debug("hose mem offset: %016llx\n", (unsigned long long)offset);
1390 for (i = 0; i < 3; i++) {
1391 pres = &hose->mem_resources[i];
1392 if (!(pres->flags & IORESOURCE_MEM))
1394 pr_debug("hose mem res: %pR\n", pres);
1395 if ((pres->start - offset) <= 0xa0000 &&
1396 (pres->end - offset) >= 0xbffff)
1401 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1402 BUG_ON(res == NULL);
1403 res->name = "Legacy VGA memory";
1404 res->flags = IORESOURCE_MEM;
1405 res->start = 0xa0000 + offset;
1406 res->end = 0xbffff + offset;
1407 pr_debug("Candidate VGA memory: %pR\n", res);
1408 if (request_resource(pres, res)) {
1410 "PCI %04x:%02x Cannot reserve VGA memory %pR\n",
1411 pci_domain_nr(bus), bus->number, res);
1416 void __init pcibios_resource_survey(void)
1420 /* Allocate and assign resources. If we re-assign everything, then
1421 * we skip the allocate phase
1423 list_for_each_entry(b, &pci_root_buses, node)
1424 pcibios_allocate_bus_resources(b);
1426 if (!(ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)) {
1427 pcibios_allocate_resources(0);
1428 pcibios_allocate_resources(1);
1431 /* Before we start assigning unassigned resource, we try to reserve
1432 * the low IO area and the VGA memory area if they intersect the
1433 * bus available resources to avoid allocating things on top of them
1435 if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1436 list_for_each_entry(b, &pci_root_buses, node)
1437 pcibios_reserve_legacy_regions(b);
1440 /* Now, if the platform didn't decide to blindly trust the firmware,
1441 * we proceed to assigning things that were left unassigned
1443 if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1444 pr_debug("PCI: Assigning unassigned resouces...\n");
1445 pci_assign_unassigned_resources();
1448 /* Call machine dependent fixup */
1449 if (ppc_md.pcibios_fixup)
1450 ppc_md.pcibios_fixup();
1453 #ifdef CONFIG_HOTPLUG
1455 /* This is used by the PCI hotplug driver to allocate resource
1456 * of newly plugged busses. We can try to consolidate with the
1457 * rest of the code later, for now, keep it as-is as our main
1458 * resource allocation function doesn't deal with sub-trees yet.
1460 void __devinit pcibios_claim_one_bus(struct pci_bus *bus)
1462 struct pci_dev *dev;
1463 struct pci_bus *child_bus;
1465 list_for_each_entry(dev, &bus->devices, bus_list) {
1468 for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1469 struct resource *r = &dev->resource[i];
1471 if (r->parent || !r->start || !r->flags)
1474 pr_debug("PCI: Claiming %s: "
1475 "Resource %d: %016llx..%016llx [%x]\n",
1477 (unsigned long long)r->start,
1478 (unsigned long long)r->end,
1479 (unsigned int)r->flags);
1481 pci_claim_resource(dev, i);
1485 list_for_each_entry(child_bus, &bus->children, node)
1486 pcibios_claim_one_bus(child_bus);
1488 EXPORT_SYMBOL_GPL(pcibios_claim_one_bus);
1491 /* pcibios_finish_adding_to_bus
1493 * This is to be called by the hotplug code after devices have been
1494 * added to a bus, this include calling it for a PHB that is just
1497 void pcibios_finish_adding_to_bus(struct pci_bus *bus)
1499 pr_debug("PCI: Finishing adding to hotplug bus %04x:%02x\n",
1500 pci_domain_nr(bus), bus->number);
1502 /* Allocate bus and devices resources */
1503 pcibios_allocate_bus_resources(bus);
1504 pcibios_claim_one_bus(bus);
1506 /* Add new devices to global lists. Register in proc, sysfs. */
1507 pci_bus_add_devices(bus);
1510 eeh_add_device_tree_late(bus);
1512 EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
1514 #endif /* CONFIG_HOTPLUG */
1516 int pcibios_enable_device(struct pci_dev *dev, int mask)
1518 if (ppc_md.pcibios_enable_device_hook)
1519 if (ppc_md.pcibios_enable_device_hook(dev))
1522 return pci_enable_resources(dev, mask);
1525 void __devinit pcibios_setup_phb_resources(struct pci_controller *hose)
1527 struct pci_bus *bus = hose->bus;
1528 struct resource *res;
1531 /* Hookup PHB IO resource */
1532 bus->resource[0] = res = &hose->io_resource;
1535 printk(KERN_WARNING "PCI: I/O resource not set for host"
1536 " bridge %s (domain %d)\n",
1537 hose->dn->full_name, hose->global_number);
1539 /* Workaround for lack of IO resource only on 32-bit */
1540 res->start = (unsigned long)hose->io_base_virt - isa_io_base;
1541 res->end = res->start + IO_SPACE_LIMIT;
1542 res->flags = IORESOURCE_IO;
1543 #endif /* CONFIG_PPC32 */
1546 pr_debug("PCI: PHB IO resource = %016llx-%016llx [%lx]\n",
1547 (unsigned long long)res->start,
1548 (unsigned long long)res->end,
1549 (unsigned long)res->flags);
1551 /* Hookup PHB Memory resources */
1552 for (i = 0; i < 3; ++i) {
1553 res = &hose->mem_resources[i];
1557 printk(KERN_ERR "PCI: Memory resource 0 not set for "
1558 "host bridge %s (domain %d)\n",
1559 hose->dn->full_name, hose->global_number);
1561 /* Workaround for lack of MEM resource only on 32-bit */
1562 res->start = hose->pci_mem_offset;
1563 res->end = (resource_size_t)-1LL;
1564 res->flags = IORESOURCE_MEM;
1565 #endif /* CONFIG_PPC32 */
1567 bus->resource[i+1] = res;
1569 pr_debug("PCI: PHB MEM resource %d = %016llx-%016llx [%lx]\n", i,
1570 (unsigned long long)res->start,
1571 (unsigned long long)res->end,
1572 (unsigned long)res->flags);
1575 pr_debug("PCI: PHB MEM offset = %016llx\n",
1576 (unsigned long long)hose->pci_mem_offset);
1577 pr_debug("PCI: PHB IO offset = %08lx\n",
1578 (unsigned long)hose->io_base_virt - _IO_BASE);