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) {
566 * Because X is lame and can fail starting if it gets an error trying
567 * to mmap legacy_mem (instead of just moving on without legacy memory
568 * access) we fake it here by giving it anonymous memory, effectively
569 * behaving just like /dev/zero
571 if ((offset + size) > hose->isa_mem_size) {
573 "Process %s (pid:%d) mapped non-existing PCI legacy memory for 0%04x:%02x\n",
574 current->comm, current->pid, pci_domain_nr(bus), bus->number);
575 if (vma->vm_flags & VM_SHARED)
576 return shmem_zero_setup(vma);
579 offset += hose->isa_mem_phys;
581 unsigned long io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
582 unsigned long roffset = offset + io_offset;
583 rp = &hose->io_resource;
584 if (!(rp->flags & IORESOURCE_IO))
586 if (roffset < rp->start || (roffset + size) > rp->end)
588 offset += hose->io_base_phys;
590 pr_debug(" -> mapping phys %llx\n", (unsigned long long)offset);
592 vma->vm_pgoff = offset >> PAGE_SHIFT;
593 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
594 return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
595 vma->vm_end - vma->vm_start,
599 void pci_resource_to_user(const struct pci_dev *dev, int bar,
600 const struct resource *rsrc,
601 resource_size_t *start, resource_size_t *end)
603 struct pci_controller *hose = pci_bus_to_host(dev->bus);
604 resource_size_t offset = 0;
609 if (rsrc->flags & IORESOURCE_IO)
610 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
612 /* We pass a fully fixed up address to userland for MMIO instead of
613 * a BAR value because X is lame and expects to be able to use that
614 * to pass to /dev/mem !
616 * That means that we'll have potentially 64 bits values where some
617 * userland apps only expect 32 (like X itself since it thinks only
618 * Sparc has 64 bits MMIO) but if we don't do that, we break it on
621 * Hopefully, the sysfs insterface is immune to that gunk. Once X
622 * has been fixed (and the fix spread enough), we can re-enable the
623 * 2 lines below and pass down a BAR value to userland. In that case
624 * we'll also have to re-enable the matching code in
625 * __pci_mmap_make_offset().
630 else if (rsrc->flags & IORESOURCE_MEM)
631 offset = hose->pci_mem_offset;
634 *start = rsrc->start - offset;
635 *end = rsrc->end - offset;
639 * pci_process_bridge_OF_ranges - Parse PCI bridge resources from device tree
640 * @hose: newly allocated pci_controller to be setup
641 * @dev: device node of the host bridge
642 * @primary: set if primary bus (32 bits only, soon to be deprecated)
644 * This function will parse the "ranges" property of a PCI host bridge device
645 * node and setup the resource mapping of a pci controller based on its
648 * Life would be boring if it wasn't for a few issues that we have to deal
651 * - We can only cope with one IO space range and up to 3 Memory space
652 * ranges. However, some machines (thanks Apple !) tend to split their
653 * space into lots of small contiguous ranges. So we have to coalesce.
655 * - We can only cope with all memory ranges having the same offset
656 * between CPU addresses and PCI addresses. Unfortunately, some bridges
657 * are setup for a large 1:1 mapping along with a small "window" which
658 * maps PCI address 0 to some arbitrary high address of the CPU space in
659 * order to give access to the ISA memory hole.
660 * The way out of here that I've chosen for now is to always set the
661 * offset based on the first resource found, then override it if we
662 * have a different offset and the previous was set by an ISA hole.
664 * - Some busses have IO space not starting at 0, which causes trouble with
665 * the way we do our IO resource renumbering. The code somewhat deals with
666 * it for 64 bits but I would expect problems on 32 bits.
668 * - Some 32 bits platforms such as 4xx can have physical space larger than
669 * 32 bits so we need to use 64 bits values for the parsing
671 void __devinit pci_process_bridge_OF_ranges(struct pci_controller *hose,
672 struct device_node *dev,
677 int pna = of_n_addr_cells(dev);
679 int memno = 0, isa_hole = -1;
681 unsigned long long pci_addr, cpu_addr, pci_next, cpu_next, size;
682 unsigned long long isa_mb = 0;
683 struct resource *res;
685 printk(KERN_INFO "PCI host bridge %s %s ranges:\n",
686 dev->full_name, primary ? "(primary)" : "");
688 /* Get ranges property */
689 ranges = of_get_property(dev, "ranges", &rlen);
694 while ((rlen -= np * 4) >= 0) {
695 /* Read next ranges element */
696 pci_space = ranges[0];
697 pci_addr = of_read_number(ranges + 1, 2);
698 cpu_addr = of_translate_address(dev, ranges + 3);
699 size = of_read_number(ranges + pna + 3, 2);
702 /* If we failed translation or got a zero-sized region
703 * (some FW try to feed us with non sensical zero sized regions
704 * such as power3 which look like some kind of attempt at exposing
705 * the VGA memory hole)
707 if (cpu_addr == OF_BAD_ADDR || size == 0)
710 /* Now consume following elements while they are contiguous */
711 for (; rlen >= np * sizeof(u32);
712 ranges += np, rlen -= np * 4) {
713 if (ranges[0] != pci_space)
715 pci_next = of_read_number(ranges + 1, 2);
716 cpu_next = of_translate_address(dev, ranges + 3);
717 if (pci_next != pci_addr + size ||
718 cpu_next != cpu_addr + size)
720 size += of_read_number(ranges + pna + 3, 2);
723 /* Act based on address space type */
725 switch ((pci_space >> 24) & 0x3) {
726 case 1: /* PCI IO space */
728 " IO 0x%016llx..0x%016llx -> 0x%016llx\n",
729 cpu_addr, cpu_addr + size - 1, pci_addr);
731 /* We support only one IO range */
732 if (hose->pci_io_size) {
734 " \\--> Skipped (too many) !\n");
738 /* On 32 bits, limit I/O space to 16MB */
739 if (size > 0x01000000)
742 /* 32 bits needs to map IOs here */
743 hose->io_base_virt = ioremap(cpu_addr, size);
745 /* Expect trouble if pci_addr is not 0 */
748 (unsigned long)hose->io_base_virt;
749 #endif /* CONFIG_PPC32 */
750 /* pci_io_size and io_base_phys always represent IO
751 * space starting at 0 so we factor in pci_addr
753 hose->pci_io_size = pci_addr + size;
754 hose->io_base_phys = cpu_addr - pci_addr;
757 res = &hose->io_resource;
758 res->flags = IORESOURCE_IO;
759 res->start = pci_addr;
761 case 2: /* PCI Memory space */
762 case 3: /* PCI 64 bits Memory space */
764 " MEM 0x%016llx..0x%016llx -> 0x%016llx %s\n",
765 cpu_addr, cpu_addr + size - 1, pci_addr,
766 (pci_space & 0x40000000) ? "Prefetch" : "");
768 /* We support only 3 memory ranges */
771 " \\--> Skipped (too many) !\n");
774 /* Handles ISA memory hole space here */
778 if (primary || isa_mem_base == 0)
779 isa_mem_base = cpu_addr;
780 hose->isa_mem_phys = cpu_addr;
781 hose->isa_mem_size = size;
784 /* We get the PCI/Mem offset from the first range or
785 * the, current one if the offset came from an ISA
786 * hole. If they don't match, bugger.
789 (isa_hole >= 0 && pci_addr != 0 &&
790 hose->pci_mem_offset == isa_mb))
791 hose->pci_mem_offset = cpu_addr - pci_addr;
792 else if (pci_addr != 0 &&
793 hose->pci_mem_offset != cpu_addr - pci_addr) {
795 " \\--> Skipped (offset mismatch) !\n");
800 res = &hose->mem_resources[memno++];
801 res->flags = IORESOURCE_MEM;
802 if (pci_space & 0x40000000)
803 res->flags |= IORESOURCE_PREFETCH;
804 res->start = cpu_addr;
808 res->name = dev->full_name;
809 res->end = res->start + size - 1;
816 /* If there's an ISA hole and the pci_mem_offset is -not- matching
817 * the ISA hole offset, then we need to remove the ISA hole from
818 * the resource list for that brige
820 if (isa_hole >= 0 && hose->pci_mem_offset != isa_mb) {
821 unsigned int next = isa_hole + 1;
822 printk(KERN_INFO " Removing ISA hole at 0x%016llx\n", isa_mb);
824 memmove(&hose->mem_resources[isa_hole],
825 &hose->mem_resources[next],
826 sizeof(struct resource) * (memno - next));
827 hose->mem_resources[--memno].flags = 0;
831 /* Decide whether to display the domain number in /proc */
832 int pci_proc_domain(struct pci_bus *bus)
834 struct pci_controller *hose = pci_bus_to_host(bus);
836 if (!(ppc_pci_flags & PPC_PCI_ENABLE_PROC_DOMAINS))
838 if (ppc_pci_flags & PPC_PCI_COMPAT_DOMAIN_0)
839 return hose->global_number != 0;
843 void pcibios_resource_to_bus(struct pci_dev *dev, struct pci_bus_region *region,
844 struct resource *res)
846 resource_size_t offset = 0, mask = (resource_size_t)-1;
847 struct pci_controller *hose = pci_bus_to_host(dev->bus);
851 if (res->flags & IORESOURCE_IO) {
852 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
854 } else if (res->flags & IORESOURCE_MEM)
855 offset = hose->pci_mem_offset;
857 region->start = (res->start - offset) & mask;
858 region->end = (res->end - offset) & mask;
860 EXPORT_SYMBOL(pcibios_resource_to_bus);
862 void pcibios_bus_to_resource(struct pci_dev *dev, struct resource *res,
863 struct pci_bus_region *region)
865 resource_size_t offset = 0, mask = (resource_size_t)-1;
866 struct pci_controller *hose = pci_bus_to_host(dev->bus);
870 if (res->flags & IORESOURCE_IO) {
871 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
873 } else if (res->flags & IORESOURCE_MEM)
874 offset = hose->pci_mem_offset;
875 res->start = (region->start + offset) & mask;
876 res->end = (region->end + offset) & mask;
878 EXPORT_SYMBOL(pcibios_bus_to_resource);
880 /* Fixup a bus resource into a linux resource */
881 static void __devinit fixup_resource(struct resource *res, struct pci_dev *dev)
883 struct pci_controller *hose = pci_bus_to_host(dev->bus);
884 resource_size_t offset = 0, mask = (resource_size_t)-1;
886 if (res->flags & IORESOURCE_IO) {
887 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
889 } else if (res->flags & IORESOURCE_MEM)
890 offset = hose->pci_mem_offset;
892 res->start = (res->start + offset) & mask;
893 res->end = (res->end + offset) & mask;
897 /* This header fixup will do the resource fixup for all devices as they are
898 * probed, but not for bridge ranges
900 static void __devinit pcibios_fixup_resources(struct pci_dev *dev)
902 struct pci_controller *hose = pci_bus_to_host(dev->bus);
906 printk(KERN_ERR "No host bridge for PCI dev %s !\n",
910 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
911 struct resource *res = dev->resource + i;
914 /* On platforms that have PPC_PCI_PROBE_ONLY set, we don't
915 * consider 0 as an unassigned BAR value. It's technically
916 * a valid value, but linux doesn't like it... so when we can
917 * re-assign things, we do so, but if we can't, we keep it
918 * around and hope for the best...
920 if (res->start == 0 && !(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
921 pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] is unassigned\n",
923 (unsigned long long)res->start,
924 (unsigned long long)res->end,
925 (unsigned int)res->flags);
926 res->end -= res->start;
928 res->flags |= IORESOURCE_UNSET;
932 pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] fixup...\n",
934 (unsigned long long)res->start,\
935 (unsigned long long)res->end,
936 (unsigned int)res->flags);
938 fixup_resource(res, dev);
940 pr_debug("PCI:%s %016llx-%016llx\n",
942 (unsigned long long)res->start,
943 (unsigned long long)res->end);
946 /* Call machine specific resource fixup */
947 if (ppc_md.pcibios_fixup_resources)
948 ppc_md.pcibios_fixup_resources(dev);
950 DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_resources);
952 /* This function tries to figure out if a bridge resource has been initialized
953 * by the firmware or not. It doesn't have to be absolutely bullet proof, but
954 * things go more smoothly when it gets it right. It should covers cases such
955 * as Apple "closed" bridge resources and bare-metal pSeries unassigned bridges
957 static int __devinit pcibios_uninitialized_bridge_resource(struct pci_bus *bus,
958 struct resource *res)
960 struct pci_controller *hose = pci_bus_to_host(bus);
961 struct pci_dev *dev = bus->self;
962 resource_size_t offset;
966 /* We don't do anything if PCI_PROBE_ONLY is set */
967 if (ppc_pci_flags & PPC_PCI_PROBE_ONLY)
970 /* Job is a bit different between memory and IO */
971 if (res->flags & IORESOURCE_MEM) {
972 /* If the BAR is non-0 (res != pci_mem_offset) then it's probably been
973 * initialized by somebody
975 if (res->start != hose->pci_mem_offset)
978 /* The BAR is 0, let's check if memory decoding is enabled on
979 * the bridge. If not, we consider it unassigned
981 pci_read_config_word(dev, PCI_COMMAND, &command);
982 if ((command & PCI_COMMAND_MEMORY) == 0)
985 /* Memory decoding is enabled and the BAR is 0. If any of the bridge
986 * resources covers that starting address (0 then it's good enough for
989 for (i = 0; i < 3; i++) {
990 if ((hose->mem_resources[i].flags & IORESOURCE_MEM) &&
991 hose->mem_resources[i].start == hose->pci_mem_offset)
995 /* Well, it starts at 0 and we know it will collide so we may as
996 * well consider it as unassigned. That covers the Apple case.
1000 /* If the BAR is non-0, then we consider it assigned */
1001 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1002 if (((res->start - offset) & 0xfffffffful) != 0)
1005 /* Here, we are a bit different than memory as typically IO space
1006 * starting at low addresses -is- valid. What we do instead if that
1007 * we consider as unassigned anything that doesn't have IO enabled
1008 * in the PCI command register, and that's it.
1010 pci_read_config_word(dev, PCI_COMMAND, &command);
1011 if (command & PCI_COMMAND_IO)
1014 /* It's starting at 0 and IO is disabled in the bridge, consider
1021 /* Fixup resources of a PCI<->PCI bridge */
1022 static void __devinit pcibios_fixup_bridge(struct pci_bus *bus)
1024 struct resource *res;
1027 struct pci_dev *dev = bus->self;
1029 for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
1030 if ((res = bus->resource[i]) == NULL)
1034 if (i >= 3 && bus->self->transparent)
1037 pr_debug("PCI:%s Bus rsrc %d %016llx-%016llx [%x] fixup...\n",
1039 (unsigned long long)res->start,\
1040 (unsigned long long)res->end,
1041 (unsigned int)res->flags);
1044 fixup_resource(res, dev);
1046 /* Try to detect uninitialized P2P bridge resources,
1047 * and clear them out so they get re-assigned later
1049 if (pcibios_uninitialized_bridge_resource(bus, res)) {
1051 pr_debug("PCI:%s (unassigned)\n", pci_name(dev));
1054 pr_debug("PCI:%s %016llx-%016llx\n",
1056 (unsigned long long)res->start,
1057 (unsigned long long)res->end);
1062 void __devinit pcibios_setup_bus_self(struct pci_bus *bus)
1064 /* Fix up the bus resources for P2P bridges */
1065 if (bus->self != NULL)
1066 pcibios_fixup_bridge(bus);
1068 /* Platform specific bus fixups. This is currently only used
1069 * by fsl_pci and I'm hoping to get rid of it at some point
1071 if (ppc_md.pcibios_fixup_bus)
1072 ppc_md.pcibios_fixup_bus(bus);
1074 /* Setup bus DMA mappings */
1075 if (ppc_md.pci_dma_bus_setup)
1076 ppc_md.pci_dma_bus_setup(bus);
1079 void __devinit pcibios_setup_bus_devices(struct pci_bus *bus)
1081 struct pci_dev *dev;
1083 pr_debug("PCI: Fixup bus devices %d (%s)\n",
1084 bus->number, bus->self ? pci_name(bus->self) : "PHB");
1086 list_for_each_entry(dev, &bus->devices, bus_list) {
1087 struct dev_archdata *sd = &dev->dev.archdata;
1089 /* Setup OF node pointer in archdata */
1090 sd->of_node = pci_device_to_OF_node(dev);
1092 /* Fixup NUMA node as it may not be setup yet by the generic
1093 * code and is needed by the DMA init
1095 set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
1097 /* Hook up default DMA ops */
1098 sd->dma_ops = pci_dma_ops;
1099 sd->dma_data = (void *)PCI_DRAM_OFFSET;
1101 /* Additional platform DMA/iommu setup */
1102 if (ppc_md.pci_dma_dev_setup)
1103 ppc_md.pci_dma_dev_setup(dev);
1105 /* Read default IRQs and fixup if necessary */
1106 pci_read_irq_line(dev);
1107 if (ppc_md.pci_irq_fixup)
1108 ppc_md.pci_irq_fixup(dev);
1112 void __devinit pcibios_fixup_bus(struct pci_bus *bus)
1114 /* When called from the generic PCI probe, read PCI<->PCI bridge
1115 * bases. This is -not- called when generating the PCI tree from
1116 * the OF device-tree.
1118 if (bus->self != NULL)
1119 pci_read_bridge_bases(bus);
1121 /* Now fixup the bus bus */
1122 pcibios_setup_bus_self(bus);
1124 /* Now fixup devices on that bus */
1125 pcibios_setup_bus_devices(bus);
1127 EXPORT_SYMBOL(pcibios_fixup_bus);
1129 static int skip_isa_ioresource_align(struct pci_dev *dev)
1131 if ((ppc_pci_flags & PPC_PCI_CAN_SKIP_ISA_ALIGN) &&
1132 !(dev->bus->bridge_ctl & PCI_BRIDGE_CTL_ISA))
1138 * We need to avoid collisions with `mirrored' VGA ports
1139 * and other strange ISA hardware, so we always want the
1140 * addresses to be allocated in the 0x000-0x0ff region
1143 * Why? Because some silly external IO cards only decode
1144 * the low 10 bits of the IO address. The 0x00-0xff region
1145 * is reserved for motherboard devices that decode all 16
1146 * bits, so it's ok to allocate at, say, 0x2800-0x28ff,
1147 * but we want to try to avoid allocating at 0x2900-0x2bff
1148 * which might have be mirrored at 0x0100-0x03ff..
1150 void pcibios_align_resource(void *data, struct resource *res,
1151 resource_size_t size, resource_size_t align)
1153 struct pci_dev *dev = data;
1155 if (res->flags & IORESOURCE_IO) {
1156 resource_size_t start = res->start;
1158 if (skip_isa_ioresource_align(dev))
1160 if (start & 0x300) {
1161 start = (start + 0x3ff) & ~0x3ff;
1166 EXPORT_SYMBOL(pcibios_align_resource);
1169 * Reparent resource children of pr that conflict with res
1170 * under res, and make res replace those children.
1172 static int __init reparent_resources(struct resource *parent,
1173 struct resource *res)
1175 struct resource *p, **pp;
1176 struct resource **firstpp = NULL;
1178 for (pp = &parent->child; (p = *pp) != NULL; pp = &p->sibling) {
1179 if (p->end < res->start)
1181 if (res->end < p->start)
1183 if (p->start < res->start || p->end > res->end)
1184 return -1; /* not completely contained */
1185 if (firstpp == NULL)
1188 if (firstpp == NULL)
1189 return -1; /* didn't find any conflicting entries? */
1190 res->parent = parent;
1191 res->child = *firstpp;
1195 for (p = res->child; p != NULL; p = p->sibling) {
1197 pr_debug("PCI: Reparented %s [%llx..%llx] under %s\n",
1199 (unsigned long long)p->start,
1200 (unsigned long long)p->end, res->name);
1206 * Handle resources of PCI devices. If the world were perfect, we could
1207 * just allocate all the resource regions and do nothing more. It isn't.
1208 * On the other hand, we cannot just re-allocate all devices, as it would
1209 * require us to know lots of host bridge internals. So we attempt to
1210 * keep as much of the original configuration as possible, but tweak it
1211 * when it's found to be wrong.
1213 * Known BIOS problems we have to work around:
1214 * - I/O or memory regions not configured
1215 * - regions configured, but not enabled in the command register
1216 * - bogus I/O addresses above 64K used
1217 * - expansion ROMs left enabled (this may sound harmless, but given
1218 * the fact the PCI specs explicitly allow address decoders to be
1219 * shared between expansion ROMs and other resource regions, it's
1220 * at least dangerous)
1223 * (1) Allocate resources for all buses behind PCI-to-PCI bridges.
1224 * This gives us fixed barriers on where we can allocate.
1225 * (2) Allocate resources for all enabled devices. If there is
1226 * a collision, just mark the resource as unallocated. Also
1227 * disable expansion ROMs during this step.
1228 * (3) Try to allocate resources for disabled devices. If the
1229 * resources were assigned correctly, everything goes well,
1230 * if they weren't, they won't disturb allocation of other
1232 * (4) Assign new addresses to resources which were either
1233 * not configured at all or misconfigured. If explicitly
1234 * requested by the user, configure expansion ROM address
1238 void pcibios_allocate_bus_resources(struct pci_bus *bus)
1242 struct resource *res, *pr;
1244 pr_debug("PCI: Allocating bus resources for %04x:%02x...\n",
1245 pci_domain_nr(bus), bus->number);
1247 for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
1248 if ((res = bus->resource[i]) == NULL || !res->flags
1249 || res->start > res->end || res->parent)
1251 if (bus->parent == NULL)
1252 pr = (res->flags & IORESOURCE_IO) ?
1253 &ioport_resource : &iomem_resource;
1255 /* Don't bother with non-root busses when
1256 * re-assigning all resources. We clear the
1257 * resource flags as if they were colliding
1258 * and as such ensure proper re-allocation
1261 if (ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)
1262 goto clear_resource;
1263 pr = pci_find_parent_resource(bus->self, res);
1265 /* this happens when the generic PCI
1266 * code (wrongly) decides that this
1267 * bridge is transparent -- paulus
1273 pr_debug("PCI: %s (bus %d) bridge rsrc %d: %016llx-%016llx "
1274 "[0x%x], parent %p (%s)\n",
1275 bus->self ? pci_name(bus->self) : "PHB",
1277 (unsigned long long)res->start,
1278 (unsigned long long)res->end,
1279 (unsigned int)res->flags,
1280 pr, (pr && pr->name) ? pr->name : "nil");
1282 if (pr && !(pr->flags & IORESOURCE_UNSET)) {
1283 if (request_resource(pr, res) == 0)
1286 * Must be a conflict with an existing entry.
1287 * Move that entry (or entries) under the
1288 * bridge resource and try again.
1290 if (reparent_resources(pr, res) == 0)
1293 printk(KERN_WARNING "PCI: Cannot allocate resource region "
1294 "%d of PCI bridge %d, will remap\n", i, bus->number);
1299 list_for_each_entry(b, &bus->children, node)
1300 pcibios_allocate_bus_resources(b);
1303 static inline void __devinit alloc_resource(struct pci_dev *dev, int idx)
1305 struct resource *pr, *r = &dev->resource[idx];
1307 pr_debug("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1309 (unsigned long long)r->start,
1310 (unsigned long long)r->end,
1311 (unsigned int)r->flags);
1313 pr = pci_find_parent_resource(dev, r);
1314 if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1315 request_resource(pr, r) < 0) {
1316 printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1317 " of device %s, will remap\n", idx, pci_name(dev));
1319 pr_debug("PCI: parent is %p: %016llx-%016llx [%x]\n",
1321 (unsigned long long)pr->start,
1322 (unsigned long long)pr->end,
1323 (unsigned int)pr->flags);
1324 /* We'll assign a new address later */
1325 r->flags |= IORESOURCE_UNSET;
1331 static void __init pcibios_allocate_resources(int pass)
1333 struct pci_dev *dev = NULL;
1338 for_each_pci_dev(dev) {
1339 pci_read_config_word(dev, PCI_COMMAND, &command);
1340 for (idx = 0; idx < 6; idx++) {
1341 r = &dev->resource[idx];
1342 if (r->parent) /* Already allocated */
1344 if (!r->flags || (r->flags & IORESOURCE_UNSET))
1345 continue; /* Not assigned at all */
1346 if (r->flags & IORESOURCE_IO)
1347 disabled = !(command & PCI_COMMAND_IO);
1349 disabled = !(command & PCI_COMMAND_MEMORY);
1350 if (pass == disabled)
1351 alloc_resource(dev, idx);
1355 r = &dev->resource[PCI_ROM_RESOURCE];
1356 if (r->flags & IORESOURCE_ROM_ENABLE) {
1357 /* Turn the ROM off, leave the resource region,
1358 * but keep it unregistered.
1361 pr_debug("PCI: Switching off ROM of %s\n",
1363 r->flags &= ~IORESOURCE_ROM_ENABLE;
1364 pci_read_config_dword(dev, dev->rom_base_reg, ®);
1365 pci_write_config_dword(dev, dev->rom_base_reg,
1366 reg & ~PCI_ROM_ADDRESS_ENABLE);
1371 static void __init pcibios_reserve_legacy_regions(struct pci_bus *bus)
1373 struct pci_controller *hose = pci_bus_to_host(bus);
1374 resource_size_t offset;
1375 struct resource *res, *pres;
1378 pr_debug("Reserving legacy ranges for domain %04x\n", pci_domain_nr(bus));
1381 if (!(hose->io_resource.flags & IORESOURCE_IO))
1383 offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1384 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1385 BUG_ON(res == NULL);
1386 res->name = "Legacy IO";
1387 res->flags = IORESOURCE_IO;
1388 res->start = offset;
1389 res->end = (offset + 0xfff) & 0xfffffffful;
1390 pr_debug("Candidate legacy IO: %pR\n", res);
1391 if (request_resource(&hose->io_resource, res)) {
1393 "PCI %04x:%02x Cannot reserve Legacy IO %pR\n",
1394 pci_domain_nr(bus), bus->number, res);
1399 /* Check for memory */
1400 offset = hose->pci_mem_offset;
1401 pr_debug("hose mem offset: %016llx\n", (unsigned long long)offset);
1402 for (i = 0; i < 3; i++) {
1403 pres = &hose->mem_resources[i];
1404 if (!(pres->flags & IORESOURCE_MEM))
1406 pr_debug("hose mem res: %pR\n", pres);
1407 if ((pres->start - offset) <= 0xa0000 &&
1408 (pres->end - offset) >= 0xbffff)
1413 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1414 BUG_ON(res == NULL);
1415 res->name = "Legacy VGA memory";
1416 res->flags = IORESOURCE_MEM;
1417 res->start = 0xa0000 + offset;
1418 res->end = 0xbffff + offset;
1419 pr_debug("Candidate VGA memory: %pR\n", res);
1420 if (request_resource(pres, res)) {
1422 "PCI %04x:%02x Cannot reserve VGA memory %pR\n",
1423 pci_domain_nr(bus), bus->number, res);
1428 void __init pcibios_resource_survey(void)
1432 /* Allocate and assign resources. If we re-assign everything, then
1433 * we skip the allocate phase
1435 list_for_each_entry(b, &pci_root_buses, node)
1436 pcibios_allocate_bus_resources(b);
1438 if (!(ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)) {
1439 pcibios_allocate_resources(0);
1440 pcibios_allocate_resources(1);
1443 /* Before we start assigning unassigned resource, we try to reserve
1444 * the low IO area and the VGA memory area if they intersect the
1445 * bus available resources to avoid allocating things on top of them
1447 if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1448 list_for_each_entry(b, &pci_root_buses, node)
1449 pcibios_reserve_legacy_regions(b);
1452 /* Now, if the platform didn't decide to blindly trust the firmware,
1453 * we proceed to assigning things that were left unassigned
1455 if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1456 pr_debug("PCI: Assigning unassigned resouces...\n");
1457 pci_assign_unassigned_resources();
1460 /* Call machine dependent fixup */
1461 if (ppc_md.pcibios_fixup)
1462 ppc_md.pcibios_fixup();
1465 #ifdef CONFIG_HOTPLUG
1467 /* This is used by the PCI hotplug driver to allocate resource
1468 * of newly plugged busses. We can try to consolidate with the
1469 * rest of the code later, for now, keep it as-is as our main
1470 * resource allocation function doesn't deal with sub-trees yet.
1472 void __devinit pcibios_claim_one_bus(struct pci_bus *bus)
1474 struct pci_dev *dev;
1475 struct pci_bus *child_bus;
1477 list_for_each_entry(dev, &bus->devices, bus_list) {
1480 for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1481 struct resource *r = &dev->resource[i];
1483 if (r->parent || !r->start || !r->flags)
1486 pr_debug("PCI: Claiming %s: "
1487 "Resource %d: %016llx..%016llx [%x]\n",
1489 (unsigned long long)r->start,
1490 (unsigned long long)r->end,
1491 (unsigned int)r->flags);
1493 pci_claim_resource(dev, i);
1497 list_for_each_entry(child_bus, &bus->children, node)
1498 pcibios_claim_one_bus(child_bus);
1500 EXPORT_SYMBOL_GPL(pcibios_claim_one_bus);
1503 /* pcibios_finish_adding_to_bus
1505 * This is to be called by the hotplug code after devices have been
1506 * added to a bus, this include calling it for a PHB that is just
1509 void pcibios_finish_adding_to_bus(struct pci_bus *bus)
1511 pr_debug("PCI: Finishing adding to hotplug bus %04x:%02x\n",
1512 pci_domain_nr(bus), bus->number);
1514 /* Allocate bus and devices resources */
1515 pcibios_allocate_bus_resources(bus);
1516 pcibios_claim_one_bus(bus);
1518 /* Add new devices to global lists. Register in proc, sysfs. */
1519 pci_bus_add_devices(bus);
1522 eeh_add_device_tree_late(bus);
1524 EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
1526 #endif /* CONFIG_HOTPLUG */
1528 int pcibios_enable_device(struct pci_dev *dev, int mask)
1530 if (ppc_md.pcibios_enable_device_hook)
1531 if (ppc_md.pcibios_enable_device_hook(dev))
1534 return pci_enable_resources(dev, mask);
1537 void __devinit pcibios_setup_phb_resources(struct pci_controller *hose)
1539 struct pci_bus *bus = hose->bus;
1540 struct resource *res;
1543 /* Hookup PHB IO resource */
1544 bus->resource[0] = res = &hose->io_resource;
1547 printk(KERN_WARNING "PCI: I/O resource not set for host"
1548 " bridge %s (domain %d)\n",
1549 hose->dn->full_name, hose->global_number);
1551 /* Workaround for lack of IO resource only on 32-bit */
1552 res->start = (unsigned long)hose->io_base_virt - isa_io_base;
1553 res->end = res->start + IO_SPACE_LIMIT;
1554 res->flags = IORESOURCE_IO;
1555 #endif /* CONFIG_PPC32 */
1558 pr_debug("PCI: PHB IO resource = %016llx-%016llx [%lx]\n",
1559 (unsigned long long)res->start,
1560 (unsigned long long)res->end,
1561 (unsigned long)res->flags);
1563 /* Hookup PHB Memory resources */
1564 for (i = 0; i < 3; ++i) {
1565 res = &hose->mem_resources[i];
1569 printk(KERN_ERR "PCI: Memory resource 0 not set for "
1570 "host bridge %s (domain %d)\n",
1571 hose->dn->full_name, hose->global_number);
1573 /* Workaround for lack of MEM resource only on 32-bit */
1574 res->start = hose->pci_mem_offset;
1575 res->end = (resource_size_t)-1LL;
1576 res->flags = IORESOURCE_MEM;
1577 #endif /* CONFIG_PPC32 */
1579 bus->resource[i+1] = res;
1581 pr_debug("PCI: PHB MEM resource %d = %016llx-%016llx [%lx]\n", i,
1582 (unsigned long long)res->start,
1583 (unsigned long long)res->end,
1584 (unsigned long)res->flags);
1587 pr_debug("PCI: PHB MEM offset = %016llx\n",
1588 (unsigned long long)hose->pci_mem_offset);
1589 pr_debug("PCI: PHB IO offset = %08lx\n",
1590 (unsigned long)hose->io_base_virt - _IO_BASE);