linux/arch/powerpc/kernel/pci-common.c
<<
>>
Prefs
   1/*
   2 * Contains common pci routines for ALL ppc platform
   3 * (based on pci_32.c and pci_64.c)
   4 *
   5 * Port for PPC64 David Engebretsen, IBM Corp.
   6 * Contains common pci routines for ppc64 platform, pSeries and iSeries brands.
   7 *
   8 * Copyright (C) 2003 Anton Blanchard <anton@au.ibm.com>, IBM
   9 *   Rework, based on alpha PCI code.
  10 *
  11 * Common pmac/prep/chrp pci routines. -- Cort
  12 *
  13 * This program is free software; you can redistribute it and/or
  14 * modify it under the terms of the GNU General Public License
  15 * as published by the Free Software Foundation; either version
  16 * 2 of the License, or (at your option) any later version.
  17 */
  18
  19#include <linux/kernel.h>
  20#include <linux/pci.h>
  21#include <linux/string.h>
  22#include <linux/init.h>
  23#include <linux/bootmem.h>
  24#include <linux/mm.h>
  25#include <linux/list.h>
  26#include <linux/syscalls.h>
  27#include <linux/irq.h>
  28#include <linux/vmalloc.h>
  29
  30#include <asm/processor.h>
  31#include <asm/io.h>
  32#include <asm/prom.h>
  33#include <asm/pci-bridge.h>
  34#include <asm/byteorder.h>
  35#include <asm/machdep.h>
  36#include <asm/ppc-pci.h>
  37#include <asm/firmware.h>
  38#include <asm/eeh.h>
  39
  40static DEFINE_SPINLOCK(hose_spinlock);
  41LIST_HEAD(hose_list);
  42
  43/* XXX kill that some day ... */
  44static int global_phb_number;           /* Global phb counter */
  45
  46/* ISA Memory physical address */
  47resource_size_t isa_mem_base;
  48
  49/* Default PCI flags is 0 on ppc32, modified at boot on ppc64 */
  50unsigned int ppc_pci_flags = 0;
  51
  52
  53static struct dma_map_ops *pci_dma_ops = &dma_direct_ops;
  54
  55void set_pci_dma_ops(struct dma_map_ops *dma_ops)
  56{
  57        pci_dma_ops = dma_ops;
  58}
  59
  60struct dma_map_ops *get_pci_dma_ops(void)
  61{
  62        return pci_dma_ops;
  63}
  64EXPORT_SYMBOL(get_pci_dma_ops);
  65
  66int pci_set_dma_mask(struct pci_dev *dev, u64 mask)
  67{
  68        return dma_set_mask(&dev->dev, mask);
  69}
  70
  71int pci_set_consistent_dma_mask(struct pci_dev *dev, u64 mask)
  72{
  73        int rc;
  74
  75        rc = dma_set_mask(&dev->dev, mask);
  76        dev->dev.coherent_dma_mask = dev->dma_mask;
  77
  78        return rc;
  79}
  80
  81struct pci_controller *pcibios_alloc_controller(struct device_node *dev)
  82{
  83        struct pci_controller *phb;
  84
  85        phb = zalloc_maybe_bootmem(sizeof(struct pci_controller), GFP_KERNEL);
  86        if (phb == NULL)
  87                return NULL;
  88        spin_lock(&hose_spinlock);
  89        phb->global_number = global_phb_number++;
  90        list_add_tail(&phb->list_node, &hose_list);
  91        spin_unlock(&hose_spinlock);
  92        phb->dn = dev;
  93        phb->is_dynamic = mem_init_done;
  94#ifdef CONFIG_PPC64
  95        if (dev) {
  96                int nid = of_node_to_nid(dev);
  97
  98                if (nid < 0 || !node_online(nid))
  99                        nid = -1;
 100
 101                PHB_SET_NODE(phb, nid);
 102        }
 103#endif
 104        return phb;
 105}
 106
 107void pcibios_free_controller(struct pci_controller *phb)
 108{
 109        spin_lock(&hose_spinlock);
 110        list_del(&phb->list_node);
 111        spin_unlock(&hose_spinlock);
 112
 113        if (phb->is_dynamic)
 114                kfree(phb);
 115}
 116
 117static resource_size_t pcibios_io_size(const struct pci_controller *hose)
 118{
 119#ifdef CONFIG_PPC64
 120        return hose->pci_io_size;
 121#else
 122        return hose->io_resource.end - hose->io_resource.start + 1;
 123#endif
 124}
 125
 126int pcibios_vaddr_is_ioport(void __iomem *address)
 127{
 128        int ret = 0;
 129        struct pci_controller *hose;
 130        resource_size_t size;
 131
 132        spin_lock(&hose_spinlock);
 133        list_for_each_entry(hose, &hose_list, list_node) {
 134                size = pcibios_io_size(hose);
 135                if (address >= hose->io_base_virt &&
 136                    address < (hose->io_base_virt + size)) {
 137                        ret = 1;
 138                        break;
 139                }
 140        }
 141        spin_unlock(&hose_spinlock);
 142        return ret;
 143}
 144
 145unsigned long pci_address_to_pio(phys_addr_t address)
 146{
 147        struct pci_controller *hose;
 148        resource_size_t size;
 149        unsigned long ret = ~0;
 150
 151        spin_lock(&hose_spinlock);
 152        list_for_each_entry(hose, &hose_list, list_node) {
 153                size = pcibios_io_size(hose);
 154                if (address >= hose->io_base_phys &&
 155                    address < (hose->io_base_phys + size)) {
 156                        unsigned long base =
 157                                (unsigned long)hose->io_base_virt - _IO_BASE;
 158                        ret = base + (address - hose->io_base_phys);
 159                        break;
 160                }
 161        }
 162        spin_unlock(&hose_spinlock);
 163
 164        return ret;
 165}
 166EXPORT_SYMBOL_GPL(pci_address_to_pio);
 167
 168/*
 169 * Return the domain number for this bus.
 170 */
 171int pci_domain_nr(struct pci_bus *bus)
 172{
 173        struct pci_controller *hose = pci_bus_to_host(bus);
 174
 175        return hose->global_number;
 176}
 177EXPORT_SYMBOL(pci_domain_nr);
 178
 179/* This routine is meant to be used early during boot, when the
 180 * PCI bus numbers have not yet been assigned, and you need to
 181 * issue PCI config cycles to an OF device.
 182 * It could also be used to "fix" RTAS config cycles if you want
 183 * to set pci_assign_all_buses to 1 and still use RTAS for PCI
 184 * config cycles.
 185 */
 186struct pci_controller* pci_find_hose_for_OF_device(struct device_node* node)
 187{
 188        while(node) {
 189                struct pci_controller *hose, *tmp;
 190                list_for_each_entry_safe(hose, tmp, &hose_list, list_node)
 191                        if (hose->dn == node)
 192                                return hose;
 193                node = node->parent;
 194        }
 195        return NULL;
 196}
 197
 198static ssize_t pci_show_devspec(struct device *dev,
 199                struct device_attribute *attr, char *buf)
 200{
 201        struct pci_dev *pdev;
 202        struct device_node *np;
 203
 204        pdev = to_pci_dev (dev);
 205        np = pci_device_to_OF_node(pdev);
 206        if (np == NULL || np->full_name == NULL)
 207                return 0;
 208        return sprintf(buf, "%s", np->full_name);
 209}
 210static DEVICE_ATTR(devspec, S_IRUGO, pci_show_devspec, NULL);
 211
 212/* Add sysfs properties */
 213int pcibios_add_platform_entries(struct pci_dev *pdev)
 214{
 215        return device_create_file(&pdev->dev, &dev_attr_devspec);
 216}
 217
 218char __devinit *pcibios_setup(char *str)
 219{
 220        return str;
 221}
 222
 223/*
 224 * Reads the interrupt pin to determine if interrupt is use by card.
 225 * If the interrupt is used, then gets the interrupt line from the
 226 * openfirmware and sets it in the pci_dev and pci_config line.
 227 */
 228int pci_read_irq_line(struct pci_dev *pci_dev)
 229{
 230        struct of_irq oirq;
 231        unsigned int virq;
 232
 233        /* The current device-tree that iSeries generates from the HV
 234         * PCI informations doesn't contain proper interrupt routing,
 235         * and all the fallback would do is print out crap, so we
 236         * don't attempt to resolve the interrupts here at all, some
 237         * iSeries specific fixup does it.
 238         *
 239         * In the long run, we will hopefully fix the generated device-tree
 240         * instead.
 241         */
 242#ifdef CONFIG_PPC_ISERIES
 243        if (firmware_has_feature(FW_FEATURE_ISERIES))
 244                return -1;
 245#endif
 246
 247        pr_debug("PCI: Try to map irq for %s...\n", pci_name(pci_dev));
 248
 249#ifdef DEBUG
 250        memset(&oirq, 0xff, sizeof(oirq));
 251#endif
 252        /* Try to get a mapping from the device-tree */
 253        if (of_irq_map_pci(pci_dev, &oirq)) {
 254                u8 line, pin;
 255
 256                /* If that fails, lets fallback to what is in the config
 257                 * space and map that through the default controller. We
 258                 * also set the type to level low since that's what PCI
 259                 * interrupts are. If your platform does differently, then
 260                 * either provide a proper interrupt tree or don't use this
 261                 * function.
 262                 */
 263                if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_PIN, &pin))
 264                        return -1;
 265                if (pin == 0)
 266                        return -1;
 267                if (pci_read_config_byte(pci_dev, PCI_INTERRUPT_LINE, &line) ||
 268                    line == 0xff || line == 0) {
 269                        return -1;
 270                }
 271                pr_debug(" No map ! Using line %d (pin %d) from PCI config\n",
 272                         line, pin);
 273
 274                virq = irq_create_mapping(NULL, line);
 275                if (virq != NO_IRQ)
 276                        set_irq_type(virq, IRQ_TYPE_LEVEL_LOW);
 277        } else {
 278                pr_debug(" Got one, spec %d cells (0x%08x 0x%08x...) on %s\n",
 279                         oirq.size, oirq.specifier[0], oirq.specifier[1],
 280                         oirq.controller ? oirq.controller->full_name :
 281                         "<default>");
 282
 283                virq = irq_create_of_mapping(oirq.controller, oirq.specifier,
 284                                             oirq.size);
 285        }
 286        if(virq == NO_IRQ) {
 287                pr_debug(" Failed to map !\n");
 288                return -1;
 289        }
 290
 291        pr_debug(" Mapped to linux irq %d\n", virq);
 292
 293        pci_dev->irq = virq;
 294
 295        return 0;
 296}
 297EXPORT_SYMBOL(pci_read_irq_line);
 298
 299/*
 300 * Platform support for /proc/bus/pci/X/Y mmap()s,
 301 * modelled on the sparc64 implementation by Dave Miller.
 302 *  -- paulus.
 303 */
 304
 305/*
 306 * Adjust vm_pgoff of VMA such that it is the physical page offset
 307 * corresponding to the 32-bit pci bus offset for DEV requested by the user.
 308 *
 309 * Basically, the user finds the base address for his device which he wishes
 310 * to mmap.  They read the 32-bit value from the config space base register,
 311 * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
 312 * offset parameter of mmap on /proc/bus/pci/XXX for that device.
 313 *
 314 * Returns negative error code on failure, zero on success.
 315 */
 316static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
 317                                               resource_size_t *offset,
 318                                               enum pci_mmap_state mmap_state)
 319{
 320        struct pci_controller *hose = pci_bus_to_host(dev->bus);
 321        unsigned long io_offset = 0;
 322        int i, res_bit;
 323
 324        if (hose == 0)
 325                return NULL;            /* should never happen */
 326
 327        /* If memory, add on the PCI bridge address offset */
 328        if (mmap_state == pci_mmap_mem) {
 329#if 0 /* See comment in pci_resource_to_user() for why this is disabled */
 330                *offset += hose->pci_mem_offset;
 331#endif
 332                res_bit = IORESOURCE_MEM;
 333        } else {
 334                io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
 335                *offset += io_offset;
 336                res_bit = IORESOURCE_IO;
 337        }
 338
 339        /*
 340         * Check that the offset requested corresponds to one of the
 341         * resources of the device.
 342         */
 343        for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
 344                struct resource *rp = &dev->resource[i];
 345                int flags = rp->flags;
 346
 347                /* treat ROM as memory (should be already) */
 348                if (i == PCI_ROM_RESOURCE)
 349                        flags |= IORESOURCE_MEM;
 350
 351                /* Active and same type? */
 352                if ((flags & res_bit) == 0)
 353                        continue;
 354
 355                /* In the range of this resource? */
 356                if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
 357                        continue;
 358
 359                /* found it! construct the final physical address */
 360                if (mmap_state == pci_mmap_io)
 361                        *offset += hose->io_base_phys - io_offset;
 362                return rp;
 363        }
 364
 365        return NULL;
 366}
 367
 368/*
 369 * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
 370 * device mapping.
 371 */
 372static pgprot_t __pci_mmap_set_pgprot(struct pci_dev *dev, struct resource *rp,
 373                                      pgprot_t protection,
 374                                      enum pci_mmap_state mmap_state,
 375                                      int write_combine)
 376{
 377        unsigned long prot = pgprot_val(protection);
 378
 379        /* Write combine is always 0 on non-memory space mappings. On
 380         * memory space, if the user didn't pass 1, we check for a
 381         * "prefetchable" resource. This is a bit hackish, but we use
 382         * this to workaround the inability of /sysfs to provide a write
 383         * combine bit
 384         */
 385        if (mmap_state != pci_mmap_mem)
 386                write_combine = 0;
 387        else if (write_combine == 0) {
 388                if (rp->flags & IORESOURCE_PREFETCH)
 389                        write_combine = 1;
 390        }
 391
 392        /* XXX would be nice to have a way to ask for write-through */
 393        if (write_combine)
 394                return pgprot_noncached_wc(prot);
 395        else
 396                return pgprot_noncached(prot);
 397}
 398
 399/*
 400 * This one is used by /dev/mem and fbdev who have no clue about the
 401 * PCI device, it tries to find the PCI device first and calls the
 402 * above routine
 403 */
 404pgprot_t pci_phys_mem_access_prot(struct file *file,
 405                                  unsigned long pfn,
 406                                  unsigned long size,
 407                                  pgprot_t prot)
 408{
 409        struct pci_dev *pdev = NULL;
 410        struct resource *found = NULL;
 411        resource_size_t offset = ((resource_size_t)pfn) << PAGE_SHIFT;
 412        int i;
 413
 414        if (page_is_ram(pfn))
 415                return prot;
 416
 417        prot = pgprot_noncached(prot);
 418        for_each_pci_dev(pdev) {
 419                for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
 420                        struct resource *rp = &pdev->resource[i];
 421                        int flags = rp->flags;
 422
 423                        /* Active and same type? */
 424                        if ((flags & IORESOURCE_MEM) == 0)
 425                                continue;
 426                        /* In the range of this resource? */
 427                        if (offset < (rp->start & PAGE_MASK) ||
 428                            offset > rp->end)
 429                                continue;
 430                        found = rp;
 431                        break;
 432                }
 433                if (found)
 434                        break;
 435        }
 436        if (found) {
 437                if (found->flags & IORESOURCE_PREFETCH)
 438                        prot = pgprot_noncached_wc(prot);
 439                pci_dev_put(pdev);
 440        }
 441
 442        pr_debug("PCI: Non-PCI map for %llx, prot: %lx\n",
 443                 (unsigned long long)offset, pgprot_val(prot));
 444
 445        return prot;
 446}
 447
 448
 449/*
 450 * Perform the actual remap of the pages for a PCI device mapping, as
 451 * appropriate for this architecture.  The region in the process to map
 452 * is described by vm_start and vm_end members of VMA, the base physical
 453 * address is found in vm_pgoff.
 454 * The pci device structure is provided so that architectures may make mapping
 455 * decisions on a per-device or per-bus basis.
 456 *
 457 * Returns a negative error code on failure, zero on success.
 458 */
 459int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
 460                        enum pci_mmap_state mmap_state, int write_combine)
 461{
 462        resource_size_t offset =
 463                ((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
 464        struct resource *rp;
 465        int ret;
 466
 467        rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
 468        if (rp == NULL)
 469                return -EINVAL;
 470
 471        vma->vm_pgoff = offset >> PAGE_SHIFT;
 472        vma->vm_page_prot = __pci_mmap_set_pgprot(dev, rp,
 473                                                  vma->vm_page_prot,
 474                                                  mmap_state, write_combine);
 475
 476        ret = remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
 477                               vma->vm_end - vma->vm_start, vma->vm_page_prot);
 478
 479        return ret;
 480}
 481
 482/* This provides legacy IO read access on a bus */
 483int pci_legacy_read(struct pci_bus *bus, loff_t port, u32 *val, size_t size)
 484{
 485        unsigned long offset;
 486        struct pci_controller *hose = pci_bus_to_host(bus);
 487        struct resource *rp = &hose->io_resource;
 488        void __iomem *addr;
 489
 490        /* Check if port can be supported by that bus. We only check
 491         * the ranges of the PHB though, not the bus itself as the rules
 492         * for forwarding legacy cycles down bridges are not our problem
 493         * here. So if the host bridge supports it, we do it.
 494         */
 495        offset = (unsigned long)hose->io_base_virt - _IO_BASE;
 496        offset += port;
 497
 498        if (!(rp->flags & IORESOURCE_IO))
 499                return -ENXIO;
 500        if (offset < rp->start || (offset + size) > rp->end)
 501                return -ENXIO;
 502        addr = hose->io_base_virt + port;
 503
 504        switch(size) {
 505        case 1:
 506                *((u8 *)val) = in_8(addr);
 507                return 1;
 508        case 2:
 509                if (port & 1)
 510                        return -EINVAL;
 511                *((u16 *)val) = in_le16(addr);
 512                return 2;
 513        case 4:
 514                if (port & 3)
 515                        return -EINVAL;
 516                *((u32 *)val) = in_le32(addr);
 517                return 4;
 518        }
 519        return -EINVAL;
 520}
 521
 522/* This provides legacy IO write access on a bus */
 523int pci_legacy_write(struct pci_bus *bus, loff_t port, u32 val, size_t size)
 524{
 525        unsigned long offset;
 526        struct pci_controller *hose = pci_bus_to_host(bus);
 527        struct resource *rp = &hose->io_resource;
 528        void __iomem *addr;
 529
 530        /* Check if port can be supported by that bus. We only check
 531         * the ranges of the PHB though, not the bus itself as the rules
 532         * for forwarding legacy cycles down bridges are not our problem
 533         * here. So if the host bridge supports it, we do it.
 534         */
 535        offset = (unsigned long)hose->io_base_virt - _IO_BASE;
 536        offset += port;
 537
 538        if (!(rp->flags & IORESOURCE_IO))
 539                return -ENXIO;
 540        if (offset < rp->start || (offset + size) > rp->end)
 541                return -ENXIO;
 542        addr = hose->io_base_virt + port;
 543
 544        /* WARNING: The generic code is idiotic. It gets passed a pointer
 545         * to what can be a 1, 2 or 4 byte quantity and always reads that
 546         * as a u32, which means that we have to correct the location of
 547         * the data read within those 32 bits for size 1 and 2
 548         */
 549        switch(size) {
 550        case 1:
 551                out_8(addr, val >> 24);
 552                return 1;
 553        case 2:
 554                if (port & 1)
 555                        return -EINVAL;
 556                out_le16(addr, val >> 16);
 557                return 2;
 558        case 4:
 559                if (port & 3)
 560                        return -EINVAL;
 561                out_le32(addr, val);
 562                return 4;
 563        }
 564        return -EINVAL;
 565}
 566
 567/* This provides legacy IO or memory mmap access on a bus */
 568int pci_mmap_legacy_page_range(struct pci_bus *bus,
 569                               struct vm_area_struct *vma,
 570                               enum pci_mmap_state mmap_state)
 571{
 572        struct pci_controller *hose = pci_bus_to_host(bus);
 573        resource_size_t offset =
 574                ((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
 575        resource_size_t size = vma->vm_end - vma->vm_start;
 576        struct resource *rp;
 577
 578        pr_debug("pci_mmap_legacy_page_range(%04x:%02x, %s @%llx..%llx)\n",
 579                 pci_domain_nr(bus), bus->number,
 580                 mmap_state == pci_mmap_mem ? "MEM" : "IO",
 581                 (unsigned long long)offset,
 582                 (unsigned long long)(offset + size - 1));
 583
 584        if (mmap_state == pci_mmap_mem) {
 585                /* Hack alert !
 586                 *
 587                 * Because X is lame and can fail starting if it gets an error trying
 588                 * to mmap legacy_mem (instead of just moving on without legacy memory
 589                 * access) we fake it here by giving it anonymous memory, effectively
 590                 * behaving just like /dev/zero
 591                 */
 592                if ((offset + size) > hose->isa_mem_size) {
 593                        printk(KERN_DEBUG
 594                               "Process %s (pid:%d) mapped non-existing PCI legacy memory for 0%04x:%02x\n",
 595                               current->comm, current->pid, pci_domain_nr(bus), bus->number);
 596                        if (vma->vm_flags & VM_SHARED)
 597                                return shmem_zero_setup(vma);
 598                        return 0;
 599                }
 600                offset += hose->isa_mem_phys;
 601        } else {
 602                unsigned long io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
 603                unsigned long roffset = offset + io_offset;
 604                rp = &hose->io_resource;
 605                if (!(rp->flags & IORESOURCE_IO))
 606                        return -ENXIO;
 607                if (roffset < rp->start || (roffset + size) > rp->end)
 608                        return -ENXIO;
 609                offset += hose->io_base_phys;
 610        }
 611        pr_debug(" -> mapping phys %llx\n", (unsigned long long)offset);
 612
 613        vma->vm_pgoff = offset >> PAGE_SHIFT;
 614        vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 615        return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
 616                               vma->vm_end - vma->vm_start,
 617                               vma->vm_page_prot);
 618}
 619
 620void pci_resource_to_user(const struct pci_dev *dev, int bar,
 621                          const struct resource *rsrc,
 622                          resource_size_t *start, resource_size_t *end)
 623{
 624        struct pci_controller *hose = pci_bus_to_host(dev->bus);
 625        resource_size_t offset = 0;
 626
 627        if (hose == NULL)
 628                return;
 629
 630        if (rsrc->flags & IORESOURCE_IO)
 631                offset = (unsigned long)hose->io_base_virt - _IO_BASE;
 632
 633        /* We pass a fully fixed up address to userland for MMIO instead of
 634         * a BAR value because X is lame and expects to be able to use that
 635         * to pass to /dev/mem !
 636         *
 637         * That means that we'll have potentially 64 bits values where some
 638         * userland apps only expect 32 (like X itself since it thinks only
 639         * Sparc has 64 bits MMIO) but if we don't do that, we break it on
 640         * 32 bits CHRPs :-(
 641         *
 642         * Hopefully, the sysfs insterface is immune to that gunk. Once X
 643         * has been fixed (and the fix spread enough), we can re-enable the
 644         * 2 lines below and pass down a BAR value to userland. In that case
 645         * we'll also have to re-enable the matching code in
 646         * __pci_mmap_make_offset().
 647         *
 648         * BenH.
 649         */
 650#if 0
 651        else if (rsrc->flags & IORESOURCE_MEM)
 652                offset = hose->pci_mem_offset;
 653#endif
 654
 655        *start = rsrc->start - offset;
 656        *end = rsrc->end - offset;
 657}
 658
 659/**
 660 * pci_process_bridge_OF_ranges - Parse PCI bridge resources from device tree
 661 * @hose: newly allocated pci_controller to be setup
 662 * @dev: device node of the host bridge
 663 * @primary: set if primary bus (32 bits only, soon to be deprecated)
 664 *
 665 * This function will parse the "ranges" property of a PCI host bridge device
 666 * node and setup the resource mapping of a pci controller based on its
 667 * content.
 668 *
 669 * Life would be boring if it wasn't for a few issues that we have to deal
 670 * with here:
 671 *
 672 *   - We can only cope with one IO space range and up to 3 Memory space
 673 *     ranges. However, some machines (thanks Apple !) tend to split their
 674 *     space into lots of small contiguous ranges. So we have to coalesce.
 675 *
 676 *   - We can only cope with all memory ranges having the same offset
 677 *     between CPU addresses and PCI addresses. Unfortunately, some bridges
 678 *     are setup for a large 1:1 mapping along with a small "window" which
 679 *     maps PCI address 0 to some arbitrary high address of the CPU space in
 680 *     order to give access to the ISA memory hole.
 681 *     The way out of here that I've chosen for now is to always set the
 682 *     offset based on the first resource found, then override it if we
 683 *     have a different offset and the previous was set by an ISA hole.
 684 *
 685 *   - Some busses have IO space not starting at 0, which causes trouble with
 686 *     the way we do our IO resource renumbering. The code somewhat deals with
 687 *     it for 64 bits but I would expect problems on 32 bits.
 688 *
 689 *   - Some 32 bits platforms such as 4xx can have physical space larger than
 690 *     32 bits so we need to use 64 bits values for the parsing
 691 */
 692void __devinit pci_process_bridge_OF_ranges(struct pci_controller *hose,
 693                                            struct device_node *dev,
 694                                            int primary)
 695{
 696        const u32 *ranges;
 697        int rlen;
 698        int pna = of_n_addr_cells(dev);
 699        int np = pna + 5;
 700        int memno = 0, isa_hole = -1;
 701        u32 pci_space;
 702        unsigned long long pci_addr, cpu_addr, pci_next, cpu_next, size;
 703        unsigned long long isa_mb = 0;
 704        struct resource *res;
 705
 706        printk(KERN_INFO "PCI host bridge %s %s ranges:\n",
 707               dev->full_name, primary ? "(primary)" : "");
 708
 709        /* Get ranges property */
 710        ranges = of_get_property(dev, "ranges", &rlen);
 711        if (ranges == NULL)
 712                return;
 713
 714        /* Parse it */
 715        while ((rlen -= np * 4) >= 0) {
 716                /* Read next ranges element */
 717                pci_space = ranges[0];
 718                pci_addr = of_read_number(ranges + 1, 2);
 719                cpu_addr = of_translate_address(dev, ranges + 3);
 720                size = of_read_number(ranges + pna + 3, 2);
 721                ranges += np;
 722
 723                /* If we failed translation or got a zero-sized region
 724                 * (some FW try to feed us with non sensical zero sized regions
 725                 * such as power3 which look like some kind of attempt at exposing
 726                 * the VGA memory hole)
 727                 */
 728                if (cpu_addr == OF_BAD_ADDR || size == 0)
 729                        continue;
 730
 731                /* Now consume following elements while they are contiguous */
 732                for (; rlen >= np * sizeof(u32);
 733                     ranges += np, rlen -= np * 4) {
 734                        if (ranges[0] != pci_space)
 735                                break;
 736                        pci_next = of_read_number(ranges + 1, 2);
 737                        cpu_next = of_translate_address(dev, ranges + 3);
 738                        if (pci_next != pci_addr + size ||
 739                            cpu_next != cpu_addr + size)
 740                                break;
 741                        size += of_read_number(ranges + pna + 3, 2);
 742                }
 743
 744                /* Act based on address space type */
 745                res = NULL;
 746                switch ((pci_space >> 24) & 0x3) {
 747                case 1:         /* PCI IO space */
 748                        printk(KERN_INFO
 749                               "  IO 0x%016llx..0x%016llx -> 0x%016llx\n",
 750                               cpu_addr, cpu_addr + size - 1, pci_addr);
 751
 752                        /* We support only one IO range */
 753                        if (hose->pci_io_size) {
 754                                printk(KERN_INFO
 755                                       " \\--> Skipped (too many) !\n");
 756                                continue;
 757                        }
 758#ifdef CONFIG_PPC32
 759                        /* On 32 bits, limit I/O space to 16MB */
 760                        if (size > 0x01000000)
 761                                size = 0x01000000;
 762
 763                        /* 32 bits needs to map IOs here */
 764                        hose->io_base_virt = ioremap(cpu_addr, size);
 765
 766                        /* Expect trouble if pci_addr is not 0 */
 767                        if (primary)
 768                                isa_io_base =
 769                                        (unsigned long)hose->io_base_virt;
 770#endif /* CONFIG_PPC32 */
 771                        /* pci_io_size and io_base_phys always represent IO
 772                         * space starting at 0 so we factor in pci_addr
 773                         */
 774                        hose->pci_io_size = pci_addr + size;
 775                        hose->io_base_phys = cpu_addr - pci_addr;
 776
 777                        /* Build resource */
 778                        res = &hose->io_resource;
 779                        res->flags = IORESOURCE_IO;
 780                        res->start = pci_addr;
 781                        break;
 782                case 2:         /* PCI Memory space */
 783                case 3:         /* PCI 64 bits Memory space */
 784                        printk(KERN_INFO
 785                               " MEM 0x%016llx..0x%016llx -> 0x%016llx %s\n",
 786                               cpu_addr, cpu_addr + size - 1, pci_addr,
 787                               (pci_space & 0x40000000) ? "Prefetch" : "");
 788
 789                        /* We support only 3 memory ranges */
 790                        if (memno >= 3) {
 791                                printk(KERN_INFO
 792                                       " \\--> Skipped (too many) !\n");
 793                                continue;
 794                        }
 795                        /* Handles ISA memory hole space here */
 796                        if (pci_addr == 0) {
 797                                isa_mb = cpu_addr;
 798                                isa_hole = memno;
 799                                if (primary || isa_mem_base == 0)
 800                                        isa_mem_base = cpu_addr;
 801                                hose->isa_mem_phys = cpu_addr;
 802                                hose->isa_mem_size = size;
 803                        }
 804
 805                        /* We get the PCI/Mem offset from the first range or
 806                         * the, current one if the offset came from an ISA
 807                         * hole. If they don't match, bugger.
 808                         */
 809                        if (memno == 0 ||
 810                            (isa_hole >= 0 && pci_addr != 0 &&
 811                             hose->pci_mem_offset == isa_mb))
 812                                hose->pci_mem_offset = cpu_addr - pci_addr;
 813                        else if (pci_addr != 0 &&
 814                                 hose->pci_mem_offset != cpu_addr - pci_addr) {
 815                                printk(KERN_INFO
 816                                       " \\--> Skipped (offset mismatch) !\n");
 817                                continue;
 818                        }
 819
 820                        /* Build resource */
 821                        res = &hose->mem_resources[memno++];
 822                        res->flags = IORESOURCE_MEM;
 823                        if (pci_space & 0x40000000)
 824                                res->flags |= IORESOURCE_PREFETCH;
 825                        res->start = cpu_addr;
 826                        break;
 827                }
 828                if (res != NULL) {
 829                        res->name = dev->full_name;
 830                        res->end = res->start + size - 1;
 831                        res->parent = NULL;
 832                        res->sibling = NULL;
 833                        res->child = NULL;
 834                }
 835        }
 836
 837        /* If there's an ISA hole and the pci_mem_offset is -not- matching
 838         * the ISA hole offset, then we need to remove the ISA hole from
 839         * the resource list for that brige
 840         */
 841        if (isa_hole >= 0 && hose->pci_mem_offset != isa_mb) {
 842                unsigned int next = isa_hole + 1;
 843                printk(KERN_INFO " Removing ISA hole at 0x%016llx\n", isa_mb);
 844                if (next < memno)
 845                        memmove(&hose->mem_resources[isa_hole],
 846                                &hose->mem_resources[next],
 847                                sizeof(struct resource) * (memno - next));
 848                hose->mem_resources[--memno].flags = 0;
 849        }
 850}
 851
 852/* Decide whether to display the domain number in /proc */
 853int pci_proc_domain(struct pci_bus *bus)
 854{
 855        struct pci_controller *hose = pci_bus_to_host(bus);
 856
 857        if (!(ppc_pci_flags & PPC_PCI_ENABLE_PROC_DOMAINS))
 858                return 0;
 859        if (ppc_pci_flags & PPC_PCI_COMPAT_DOMAIN_0)
 860                return hose->global_number != 0;
 861        return 1;
 862}
 863
 864void pcibios_resource_to_bus(struct pci_dev *dev, struct pci_bus_region *region,
 865                             struct resource *res)
 866{
 867        resource_size_t offset = 0, mask = (resource_size_t)-1;
 868        struct pci_controller *hose = pci_bus_to_host(dev->bus);
 869
 870        if (!hose)
 871                return;
 872        if (res->flags & IORESOURCE_IO) {
 873                offset = (unsigned long)hose->io_base_virt - _IO_BASE;
 874                mask = 0xffffffffu;
 875        } else if (res->flags & IORESOURCE_MEM)
 876                offset = hose->pci_mem_offset;
 877
 878        region->start = (res->start - offset) & mask;
 879        region->end = (res->end - offset) & mask;
 880}
 881EXPORT_SYMBOL(pcibios_resource_to_bus);
 882
 883void pcibios_bus_to_resource(struct pci_dev *dev, struct resource *res,
 884                             struct pci_bus_region *region)
 885{
 886        resource_size_t offset = 0, mask = (resource_size_t)-1;
 887        struct pci_controller *hose = pci_bus_to_host(dev->bus);
 888
 889        if (!hose)
 890                return;
 891        if (res->flags & IORESOURCE_IO) {
 892                offset = (unsigned long)hose->io_base_virt - _IO_BASE;
 893                mask = 0xffffffffu;
 894        } else if (res->flags & IORESOURCE_MEM)
 895                offset = hose->pci_mem_offset;
 896        res->start = (region->start + offset) & mask;
 897        res->end = (region->end + offset) & mask;
 898}
 899EXPORT_SYMBOL(pcibios_bus_to_resource);
 900
 901/* Fixup a bus resource into a linux resource */
 902static void __devinit fixup_resource(struct resource *res, struct pci_dev *dev)
 903{
 904        struct pci_controller *hose = pci_bus_to_host(dev->bus);
 905        resource_size_t offset = 0, mask = (resource_size_t)-1;
 906
 907        if (res->flags & IORESOURCE_IO) {
 908                offset = (unsigned long)hose->io_base_virt - _IO_BASE;
 909                mask = 0xffffffffu;
 910        } else if (res->flags & IORESOURCE_MEM)
 911                offset = hose->pci_mem_offset;
 912
 913        res->start = (res->start + offset) & mask;
 914        res->end = (res->end + offset) & mask;
 915}
 916
 917
 918/* This header fixup will do the resource fixup for all devices as they are
 919 * probed, but not for bridge ranges
 920 */
 921static void __devinit pcibios_fixup_resources(struct pci_dev *dev)
 922{
 923        struct pci_controller *hose = pci_bus_to_host(dev->bus);
 924        int i;
 925
 926        if (!hose) {
 927                printk(KERN_ERR "No host bridge for PCI dev %s !\n",
 928                       pci_name(dev));
 929                return;
 930        }
 931        for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
 932                struct resource *res = dev->resource + i;
 933                if (!res->flags)
 934                        continue;
 935                /* On platforms that have PPC_PCI_PROBE_ONLY set, we don't
 936                 * consider 0 as an unassigned BAR value. It's technically
 937                 * a valid value, but linux doesn't like it... so when we can
 938                 * re-assign things, we do so, but if we can't, we keep it
 939                 * around and hope for the best...
 940                 */
 941                if (res->start == 0 && !(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
 942                        pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] is unassigned\n",
 943                                 pci_name(dev), i,
 944                                 (unsigned long long)res->start,
 945                                 (unsigned long long)res->end,
 946                                 (unsigned int)res->flags);
 947                        res->end -= res->start;
 948                        res->start = 0;
 949                        res->flags |= IORESOURCE_UNSET;
 950                        continue;
 951                }
 952
 953                pr_debug("PCI:%s Resource %d %016llx-%016llx [%x] fixup...\n",
 954                         pci_name(dev), i,
 955                         (unsigned long long)res->start,\
 956                         (unsigned long long)res->end,
 957                         (unsigned int)res->flags);
 958
 959                fixup_resource(res, dev);
 960
 961                pr_debug("PCI:%s            %016llx-%016llx\n",
 962                         pci_name(dev),
 963                         (unsigned long long)res->start,
 964                         (unsigned long long)res->end);
 965        }
 966
 967        /* Call machine specific resource fixup */
 968        if (ppc_md.pcibios_fixup_resources)
 969                ppc_md.pcibios_fixup_resources(dev);
 970}
 971DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_resources);
 972
 973/* This function tries to figure out if a bridge resource has been initialized
 974 * by the firmware or not. It doesn't have to be absolutely bullet proof, but
 975 * things go more smoothly when it gets it right. It should covers cases such
 976 * as Apple "closed" bridge resources and bare-metal pSeries unassigned bridges
 977 */
 978static int __devinit pcibios_uninitialized_bridge_resource(struct pci_bus *bus,
 979                                                           struct resource *res)
 980{
 981        struct pci_controller *hose = pci_bus_to_host(bus);
 982        struct pci_dev *dev = bus->self;
 983        resource_size_t offset;
 984        u16 command;
 985        int i;
 986
 987        /* We don't do anything if PCI_PROBE_ONLY is set */
 988        if (ppc_pci_flags & PPC_PCI_PROBE_ONLY)
 989                return 0;
 990
 991        /* Job is a bit different between memory and IO */
 992        if (res->flags & IORESOURCE_MEM) {
 993                /* If the BAR is non-0 (res != pci_mem_offset) then it's probably been
 994                 * initialized by somebody
 995                 */
 996                if (res->start != hose->pci_mem_offset)
 997                        return 0;
 998
 999                /* The BAR is 0, let's check if memory decoding is enabled on
1000                 * the bridge. If not, we consider it unassigned
1001                 */
1002                pci_read_config_word(dev, PCI_COMMAND, &command);
1003                if ((command & PCI_COMMAND_MEMORY) == 0)
1004                        return 1;
1005
1006                /* Memory decoding is enabled and the BAR is 0. If any of the bridge
1007                 * resources covers that starting address (0 then it's good enough for
1008                 * us for memory
1009                 */
1010                for (i = 0; i < 3; i++) {
1011                        if ((hose->mem_resources[i].flags & IORESOURCE_MEM) &&
1012                            hose->mem_resources[i].start == hose->pci_mem_offset)
1013                                return 0;
1014                }
1015
1016                /* Well, it starts at 0 and we know it will collide so we may as
1017                 * well consider it as unassigned. That covers the Apple case.
1018                 */
1019                return 1;
1020        } else {
1021                /* If the BAR is non-0, then we consider it assigned */
1022                offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1023                if (((res->start - offset) & 0xfffffffful) != 0)
1024                        return 0;
1025
1026                /* Here, we are a bit different than memory as typically IO space
1027                 * starting at low addresses -is- valid. What we do instead if that
1028                 * we consider as unassigned anything that doesn't have IO enabled
1029                 * in the PCI command register, and that's it.
1030                 */
1031                pci_read_config_word(dev, PCI_COMMAND, &command);
1032                if (command & PCI_COMMAND_IO)
1033                        return 0;
1034
1035                /* It's starting at 0 and IO is disabled in the bridge, consider
1036                 * it unassigned
1037                 */
1038                return 1;
1039        }
1040}
1041
1042/* Fixup resources of a PCI<->PCI bridge */
1043static void __devinit pcibios_fixup_bridge(struct pci_bus *bus)
1044{
1045        struct resource *res;
1046        int i;
1047
1048        struct pci_dev *dev = bus->self;
1049
1050        for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
1051                if ((res = bus->resource[i]) == NULL)
1052                        continue;
1053                if (!res->flags)
1054                        continue;
1055                if (i >= 3 && bus->self->transparent)
1056                        continue;
1057
1058                pr_debug("PCI:%s Bus rsrc %d %016llx-%016llx [%x] fixup...\n",
1059                         pci_name(dev), i,
1060                         (unsigned long long)res->start,\
1061                         (unsigned long long)res->end,
1062                         (unsigned int)res->flags);
1063
1064                /* Perform fixup */
1065                fixup_resource(res, dev);
1066
1067                /* Try to detect uninitialized P2P bridge resources,
1068                 * and clear them out so they get re-assigned later
1069                 */
1070                if (pcibios_uninitialized_bridge_resource(bus, res)) {
1071                        res->flags = 0;
1072                        pr_debug("PCI:%s            (unassigned)\n", pci_name(dev));
1073                } else {
1074
1075                        pr_debug("PCI:%s            %016llx-%016llx\n",
1076                                 pci_name(dev),
1077                                 (unsigned long long)res->start,
1078                                 (unsigned long long)res->end);
1079                }
1080        }
1081}
1082
1083void __devinit pcibios_setup_bus_self(struct pci_bus *bus)
1084{
1085        /* Fix up the bus resources for P2P bridges */
1086        if (bus->self != NULL)
1087                pcibios_fixup_bridge(bus);
1088
1089        /* Platform specific bus fixups. This is currently only used
1090         * by fsl_pci and I'm hoping to get rid of it at some point
1091         */
1092        if (ppc_md.pcibios_fixup_bus)
1093                ppc_md.pcibios_fixup_bus(bus);
1094
1095        /* Setup bus DMA mappings */
1096        if (ppc_md.pci_dma_bus_setup)
1097                ppc_md.pci_dma_bus_setup(bus);
1098}
1099
1100void __devinit pcibios_setup_bus_devices(struct pci_bus *bus)
1101{
1102        struct pci_dev *dev;
1103
1104        pr_debug("PCI: Fixup bus devices %d (%s)\n",
1105                 bus->number, bus->self ? pci_name(bus->self) : "PHB");
1106
1107        list_for_each_entry(dev, &bus->devices, bus_list) {
1108                struct dev_archdata *sd = &dev->dev.archdata;
1109
1110                /* Setup OF node pointer in archdata */
1111                sd->of_node = pci_device_to_OF_node(dev);
1112
1113                /* Fixup NUMA node as it may not be setup yet by the generic
1114                 * code and is needed by the DMA init
1115                 */
1116                set_dev_node(&dev->dev, pcibus_to_node(dev->bus));
1117
1118                /* Hook up default DMA ops */
1119                sd->dma_ops = pci_dma_ops;
1120                set_dma_offset(&dev->dev, PCI_DRAM_OFFSET);
1121
1122                /* Additional platform DMA/iommu setup */
1123                if (ppc_md.pci_dma_dev_setup)
1124                        ppc_md.pci_dma_dev_setup(dev);
1125
1126                /* Read default IRQs and fixup if necessary */
1127                pci_read_irq_line(dev);
1128                if (ppc_md.pci_irq_fixup)
1129                        ppc_md.pci_irq_fixup(dev);
1130        }
1131}
1132
1133void __devinit pcibios_fixup_bus(struct pci_bus *bus)
1134{
1135        /* When called from the generic PCI probe, read PCI<->PCI bridge
1136         * bases. This is -not- called when generating the PCI tree from
1137         * the OF device-tree.
1138         */
1139        if (bus->self != NULL)
1140                pci_read_bridge_bases(bus);
1141
1142        /* Now fixup the bus bus */
1143        pcibios_setup_bus_self(bus);
1144
1145        /* Now fixup devices on that bus */
1146        pcibios_setup_bus_devices(bus);
1147}
1148EXPORT_SYMBOL(pcibios_fixup_bus);
1149
1150static int skip_isa_ioresource_align(struct pci_dev *dev)
1151{
1152        if ((ppc_pci_flags & PPC_PCI_CAN_SKIP_ISA_ALIGN) &&
1153            !(dev->bus->bridge_ctl & PCI_BRIDGE_CTL_ISA))
1154                return 1;
1155        return 0;
1156}
1157
1158/*
1159 * We need to avoid collisions with `mirrored' VGA ports
1160 * and other strange ISA hardware, so we always want the
1161 * addresses to be allocated in the 0x000-0x0ff region
1162 * modulo 0x400.
1163 *
1164 * Why? Because some silly external IO cards only decode
1165 * the low 10 bits of the IO address. The 0x00-0xff region
1166 * is reserved for motherboard devices that decode all 16
1167 * bits, so it's ok to allocate at, say, 0x2800-0x28ff,
1168 * but we want to try to avoid allocating at 0x2900-0x2bff
1169 * which might have be mirrored at 0x0100-0x03ff..
1170 */
1171void pcibios_align_resource(void *data, struct resource *res,
1172                                resource_size_t size, resource_size_t align)
1173{
1174        struct pci_dev *dev = data;
1175
1176        if (res->flags & IORESOURCE_IO) {
1177                resource_size_t start = res->start;
1178
1179                if (skip_isa_ioresource_align(dev))
1180                        return;
1181                if (start & 0x300) {
1182                        start = (start + 0x3ff) & ~0x3ff;
1183                        res->start = start;
1184                }
1185        }
1186}
1187EXPORT_SYMBOL(pcibios_align_resource);
1188
1189/*
1190 * Reparent resource children of pr that conflict with res
1191 * under res, and make res replace those children.
1192 */
1193static int reparent_resources(struct resource *parent,
1194                                     struct resource *res)
1195{
1196        struct resource *p, **pp;
1197        struct resource **firstpp = NULL;
1198
1199        for (pp = &parent->child; (p = *pp) != NULL; pp = &p->sibling) {
1200                if (p->end < res->start)
1201                        continue;
1202                if (res->end < p->start)
1203                        break;
1204                if (p->start < res->start || p->end > res->end)
1205                        return -1;      /* not completely contained */
1206                if (firstpp == NULL)
1207                        firstpp = pp;
1208        }
1209        if (firstpp == NULL)
1210                return -1;      /* didn't find any conflicting entries? */
1211        res->parent = parent;
1212        res->child = *firstpp;
1213        res->sibling = *pp;
1214        *firstpp = res;
1215        *pp = NULL;
1216        for (p = res->child; p != NULL; p = p->sibling) {
1217                p->parent = res;
1218                pr_debug("PCI: Reparented %s [%llx..%llx] under %s\n",
1219                         p->name,
1220                         (unsigned long long)p->start,
1221                         (unsigned long long)p->end, res->name);
1222        }
1223        return 0;
1224}
1225
1226/*
1227 *  Handle resources of PCI devices.  If the world were perfect, we could
1228 *  just allocate all the resource regions and do nothing more.  It isn't.
1229 *  On the other hand, we cannot just re-allocate all devices, as it would
1230 *  require us to know lots of host bridge internals.  So we attempt to
1231 *  keep as much of the original configuration as possible, but tweak it
1232 *  when it's found to be wrong.
1233 *
1234 *  Known BIOS problems we have to work around:
1235 *      - I/O or memory regions not configured
1236 *      - regions configured, but not enabled in the command register
1237 *      - bogus I/O addresses above 64K used
1238 *      - expansion ROMs left enabled (this may sound harmless, but given
1239 *        the fact the PCI specs explicitly allow address decoders to be
1240 *        shared between expansion ROMs and other resource regions, it's
1241 *        at least dangerous)
1242 *
1243 *  Our solution:
1244 *      (1) Allocate resources for all buses behind PCI-to-PCI bridges.
1245 *          This gives us fixed barriers on where we can allocate.
1246 *      (2) Allocate resources for all enabled devices.  If there is
1247 *          a collision, just mark the resource as unallocated. Also
1248 *          disable expansion ROMs during this step.
1249 *      (3) Try to allocate resources for disabled devices.  If the
1250 *          resources were assigned correctly, everything goes well,
1251 *          if they weren't, they won't disturb allocation of other
1252 *          resources.
1253 *      (4) Assign new addresses to resources which were either
1254 *          not configured at all or misconfigured.  If explicitly
1255 *          requested by the user, configure expansion ROM address
1256 *          as well.
1257 */
1258
1259void pcibios_allocate_bus_resources(struct pci_bus *bus)
1260{
1261        struct pci_bus *b;
1262        int i;
1263        struct resource *res, *pr;
1264
1265        pr_debug("PCI: Allocating bus resources for %04x:%02x...\n",
1266                 pci_domain_nr(bus), bus->number);
1267
1268        for (i = 0; i < PCI_BUS_NUM_RESOURCES; ++i) {
1269                if ((res = bus->resource[i]) == NULL || !res->flags
1270                    || res->start > res->end || res->parent)
1271                        continue;
1272                if (bus->parent == NULL)
1273                        pr = (res->flags & IORESOURCE_IO) ?
1274                                &ioport_resource : &iomem_resource;
1275                else {
1276                        /* Don't bother with non-root busses when
1277                         * re-assigning all resources. We clear the
1278                         * resource flags as if they were colliding
1279                         * and as such ensure proper re-allocation
1280                         * later.
1281                         */
1282                        if (ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)
1283                                goto clear_resource;
1284                        pr = pci_find_parent_resource(bus->self, res);
1285                        if (pr == res) {
1286                                /* this happens when the generic PCI
1287                                 * code (wrongly) decides that this
1288                                 * bridge is transparent  -- paulus
1289                                 */
1290                                continue;
1291                        }
1292                }
1293
1294                pr_debug("PCI: %s (bus %d) bridge rsrc %d: %016llx-%016llx "
1295                         "[0x%x], parent %p (%s)\n",
1296                         bus->self ? pci_name(bus->self) : "PHB",
1297                         bus->number, i,
1298                         (unsigned long long)res->start,
1299                         (unsigned long long)res->end,
1300                         (unsigned int)res->flags,
1301                         pr, (pr && pr->name) ? pr->name : "nil");
1302
1303                if (pr && !(pr->flags & IORESOURCE_UNSET)) {
1304                        if (request_resource(pr, res) == 0)
1305                                continue;
1306                        /*
1307                         * Must be a conflict with an existing entry.
1308                         * Move that entry (or entries) under the
1309                         * bridge resource and try again.
1310                         */
1311                        if (reparent_resources(pr, res) == 0)
1312                                continue;
1313                }
1314                printk(KERN_WARNING "PCI: Cannot allocate resource region "
1315                       "%d of PCI bridge %d, will remap\n", i, bus->number);
1316clear_resource:
1317                res->flags = 0;
1318        }
1319
1320        list_for_each_entry(b, &bus->children, node)
1321                pcibios_allocate_bus_resources(b);
1322}
1323
1324static inline void __devinit alloc_resource(struct pci_dev *dev, int idx)
1325{
1326        struct resource *pr, *r = &dev->resource[idx];
1327
1328        pr_debug("PCI: Allocating %s: Resource %d: %016llx..%016llx [%x]\n",
1329                 pci_name(dev), idx,
1330                 (unsigned long long)r->start,
1331                 (unsigned long long)r->end,
1332                 (unsigned int)r->flags);
1333
1334        pr = pci_find_parent_resource(dev, r);
1335        if (!pr || (pr->flags & IORESOURCE_UNSET) ||
1336            request_resource(pr, r) < 0) {
1337                printk(KERN_WARNING "PCI: Cannot allocate resource region %d"
1338                       " of device %s, will remap\n", idx, pci_name(dev));
1339                if (pr)
1340                        pr_debug("PCI:  parent is %p: %016llx-%016llx [%x]\n",
1341                                 pr,
1342                                 (unsigned long long)pr->start,
1343                                 (unsigned long long)pr->end,
1344                                 (unsigned int)pr->flags);
1345                /* We'll assign a new address later */
1346                r->flags |= IORESOURCE_UNSET;
1347                r->end -= r->start;
1348                r->start = 0;
1349        }
1350}
1351
1352static void __init pcibios_allocate_resources(int pass)
1353{
1354        struct pci_dev *dev = NULL;
1355        int idx, disabled;
1356        u16 command;
1357        struct resource *r;
1358
1359        for_each_pci_dev(dev) {
1360                pci_read_config_word(dev, PCI_COMMAND, &command);
1361                for (idx = 0; idx <= PCI_ROM_RESOURCE; idx++) {
1362                        r = &dev->resource[idx];
1363                        if (r->parent)          /* Already allocated */
1364                                continue;
1365                        if (!r->flags || (r->flags & IORESOURCE_UNSET))
1366                                continue;       /* Not assigned at all */
1367                        /* We only allocate ROMs on pass 1 just in case they
1368                         * have been screwed up by firmware
1369                         */
1370                        if (idx == PCI_ROM_RESOURCE )
1371                                disabled = 1;
1372                        if (r->flags & IORESOURCE_IO)
1373                                disabled = !(command & PCI_COMMAND_IO);
1374                        else
1375                                disabled = !(command & PCI_COMMAND_MEMORY);
1376                        if (pass == disabled)
1377                                alloc_resource(dev, idx);
1378                }
1379                if (pass)
1380                        continue;
1381                r = &dev->resource[PCI_ROM_RESOURCE];
1382                if (r->flags) {
1383                        /* Turn the ROM off, leave the resource region,
1384                         * but keep it unregistered.
1385                         */
1386                        u32 reg;
1387                        pci_read_config_dword(dev, dev->rom_base_reg, &reg);
1388                        if (reg & PCI_ROM_ADDRESS_ENABLE) {
1389                                pr_debug("PCI: Switching off ROM of %s\n",
1390                                         pci_name(dev));
1391                                r->flags &= ~IORESOURCE_ROM_ENABLE;
1392                                pci_write_config_dword(dev, dev->rom_base_reg,
1393                                                       reg & ~PCI_ROM_ADDRESS_ENABLE);
1394                        }
1395                }
1396        }
1397}
1398
1399static void __init pcibios_reserve_legacy_regions(struct pci_bus *bus)
1400{
1401        struct pci_controller *hose = pci_bus_to_host(bus);
1402        resource_size_t offset;
1403        struct resource *res, *pres;
1404        int i;
1405
1406        pr_debug("Reserving legacy ranges for domain %04x\n", pci_domain_nr(bus));
1407
1408        /* Check for IO */
1409        if (!(hose->io_resource.flags & IORESOURCE_IO))
1410                goto no_io;
1411        offset = (unsigned long)hose->io_base_virt - _IO_BASE;
1412        res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1413        BUG_ON(res == NULL);
1414        res->name = "Legacy IO";
1415        res->flags = IORESOURCE_IO;
1416        res->start = offset;
1417        res->end = (offset + 0xfff) & 0xfffffffful;
1418        pr_debug("Candidate legacy IO: %pR\n", res);
1419        if (request_resource(&hose->io_resource, res)) {
1420                printk(KERN_DEBUG
1421                       "PCI %04x:%02x Cannot reserve Legacy IO %pR\n",
1422                       pci_domain_nr(bus), bus->number, res);
1423                kfree(res);
1424        }
1425
1426 no_io:
1427        /* Check for memory */
1428        offset = hose->pci_mem_offset;
1429        pr_debug("hose mem offset: %016llx\n", (unsigned long long)offset);
1430        for (i = 0; i < 3; i++) {
1431                pres = &hose->mem_resources[i];
1432                if (!(pres->flags & IORESOURCE_MEM))
1433                        continue;
1434                pr_debug("hose mem res: %pR\n", pres);
1435                if ((pres->start - offset) <= 0xa0000 &&
1436                    (pres->end - offset) >= 0xbffff)
1437                        break;
1438        }
1439        if (i >= 3)
1440                return;
1441        res = kzalloc(sizeof(struct resource), GFP_KERNEL);
1442        BUG_ON(res == NULL);
1443        res->name = "Legacy VGA memory";
1444        res->flags = IORESOURCE_MEM;
1445        res->start = 0xa0000 + offset;
1446        res->end = 0xbffff + offset;
1447        pr_debug("Candidate VGA memory: %pR\n", res);
1448        if (request_resource(pres, res)) {
1449                printk(KERN_DEBUG
1450                       "PCI %04x:%02x Cannot reserve VGA memory %pR\n",
1451                       pci_domain_nr(bus), bus->number, res);
1452                kfree(res);
1453        }
1454}
1455
1456void __init pcibios_resource_survey(void)
1457{
1458        struct pci_bus *b;
1459
1460        /* Allocate and assign resources. If we re-assign everything, then
1461         * we skip the allocate phase
1462         */
1463        list_for_each_entry(b, &pci_root_buses, node)
1464                pcibios_allocate_bus_resources(b);
1465
1466        if (!(ppc_pci_flags & PPC_PCI_REASSIGN_ALL_RSRC)) {
1467                pcibios_allocate_resources(0);
1468                pcibios_allocate_resources(1);
1469        }
1470
1471        /* Before we start assigning unassigned resource, we try to reserve
1472         * the low IO area and the VGA memory area if they intersect the
1473         * bus available resources to avoid allocating things on top of them
1474         */
1475        if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1476                list_for_each_entry(b, &pci_root_buses, node)
1477                        pcibios_reserve_legacy_regions(b);
1478        }
1479
1480        /* Now, if the platform didn't decide to blindly trust the firmware,
1481         * we proceed to assigning things that were left unassigned
1482         */
1483        if (!(ppc_pci_flags & PPC_PCI_PROBE_ONLY)) {
1484                pr_debug("PCI: Assigning unassigned resources...\n");
1485                pci_assign_unassigned_resources();
1486        }
1487
1488        /* Call machine dependent fixup */
1489        if (ppc_md.pcibios_fixup)
1490                ppc_md.pcibios_fixup();
1491}
1492
1493#ifdef CONFIG_HOTPLUG
1494
1495/* This is used by the PCI hotplug driver to allocate resource
1496 * of newly plugged busses. We can try to consolidate with the
1497 * rest of the code later, for now, keep it as-is as our main
1498 * resource allocation function doesn't deal with sub-trees yet.
1499 */
1500void pcibios_claim_one_bus(struct pci_bus *bus)
1501{
1502        struct pci_dev *dev;
1503        struct pci_bus *child_bus;
1504
1505        list_for_each_entry(dev, &bus->devices, bus_list) {
1506                int i;
1507
1508                for (i = 0; i < PCI_NUM_RESOURCES; i++) {
1509                        struct resource *r = &dev->resource[i];
1510
1511                        if (r->parent || !r->start || !r->flags)
1512                                continue;
1513
1514                        pr_debug("PCI: Claiming %s: "
1515                                 "Resource %d: %016llx..%016llx [%x]\n",
1516                                 pci_name(dev), i,
1517                                 (unsigned long long)r->start,
1518                                 (unsigned long long)r->end,
1519                                 (unsigned int)r->flags);
1520
1521                        pci_claim_resource(dev, i);
1522                }
1523        }
1524
1525        list_for_each_entry(child_bus, &bus->children, node)
1526                pcibios_claim_one_bus(child_bus);
1527}
1528
1529
1530/* pcibios_finish_adding_to_bus
1531 *
1532 * This is to be called by the hotplug code after devices have been
1533 * added to a bus, this include calling it for a PHB that is just
1534 * being added
1535 */
1536void pcibios_finish_adding_to_bus(struct pci_bus *bus)
1537{
1538        pr_debug("PCI: Finishing adding to hotplug bus %04x:%02x\n",
1539                 pci_domain_nr(bus), bus->number);
1540
1541        /* Allocate bus and devices resources */
1542        pcibios_allocate_bus_resources(bus);
1543        pcibios_claim_one_bus(bus);
1544
1545        /* Add new devices to global lists.  Register in proc, sysfs. */
1546        pci_bus_add_devices(bus);
1547
1548        /* Fixup EEH */
1549        eeh_add_device_tree_late(bus);
1550}
1551EXPORT_SYMBOL_GPL(pcibios_finish_adding_to_bus);
1552
1553#endif /* CONFIG_HOTPLUG */
1554
1555int pcibios_enable_device(struct pci_dev *dev, int mask)
1556{
1557        if (ppc_md.pcibios_enable_device_hook)
1558                if (ppc_md.pcibios_enable_device_hook(dev))
1559                        return -EINVAL;
1560
1561        return pci_enable_resources(dev, mask);
1562}
1563
1564void __devinit pcibios_setup_phb_resources(struct pci_controller *hose)
1565{
1566        struct pci_bus *bus = hose->bus;
1567        struct resource *res;
1568        int i;
1569
1570        /* Hookup PHB IO resource */
1571        bus->resource[0] = res = &hose->io_resource;
1572
1573        if (!res->flags) {
1574                printk(KERN_WARNING "PCI: I/O resource not set for host"
1575                       " bridge %s (domain %d)\n",
1576                       hose->dn->full_name, hose->global_number);
1577#ifdef CONFIG_PPC32
1578                /* Workaround for lack of IO resource only on 32-bit */
1579                res->start = (unsigned long)hose->io_base_virt - isa_io_base;
1580                res->end = res->start + IO_SPACE_LIMIT;
1581                res->flags = IORESOURCE_IO;
1582#endif /* CONFIG_PPC32 */
1583        }
1584
1585        pr_debug("PCI: PHB IO resource    = %016llx-%016llx [%lx]\n",
1586                 (unsigned long long)res->start,
1587                 (unsigned long long)res->end,
1588                 (unsigned long)res->flags);
1589
1590        /* Hookup PHB Memory resources */
1591        for (i = 0; i < 3; ++i) {
1592                res = &hose->mem_resources[i];
1593                if (!res->flags) {
1594                        if (i > 0)
1595                                continue;
1596                        printk(KERN_ERR "PCI: Memory resource 0 not set for "
1597                               "host bridge %s (domain %d)\n",
1598                               hose->dn->full_name, hose->global_number);
1599#ifdef CONFIG_PPC32
1600                        /* Workaround for lack of MEM resource only on 32-bit */
1601                        res->start = hose->pci_mem_offset;
1602                        res->end = (resource_size_t)-1LL;
1603                        res->flags = IORESOURCE_MEM;
1604#endif /* CONFIG_PPC32 */
1605                }
1606                bus->resource[i+1] = res;
1607
1608                pr_debug("PCI: PHB MEM resource %d = %016llx-%016llx [%lx]\n", i,
1609                         (unsigned long long)res->start,
1610                         (unsigned long long)res->end,
1611                         (unsigned long)res->flags);
1612        }
1613
1614        pr_debug("PCI: PHB MEM offset     = %016llx\n",
1615                 (unsigned long long)hose->pci_mem_offset);
1616        pr_debug("PCI: PHB IO  offset     = %08lx\n",
1617                 (unsigned long)hose->io_base_virt - _IO_BASE);
1618
1619}
1620
1621/*
1622 * Null PCI config access functions, for the case when we can't
1623 * find a hose.
1624 */
1625#define NULL_PCI_OP(rw, size, type)                                     \
1626static int                                                              \
1627null_##rw##_config_##size(struct pci_dev *dev, int offset, type val)    \
1628{                                                                       \
1629        return PCIBIOS_DEVICE_NOT_FOUND;                                \
1630}
1631
1632static int
1633null_read_config(struct pci_bus *bus, unsigned int devfn, int offset,
1634                 int len, u32 *val)
1635{
1636        return PCIBIOS_DEVICE_NOT_FOUND;
1637}
1638
1639static int
1640null_write_config(struct pci_bus *bus, unsigned int devfn, int offset,
1641                  int len, u32 val)
1642{
1643        return PCIBIOS_DEVICE_NOT_FOUND;
1644}
1645
1646static struct pci_ops null_pci_ops =
1647{
1648        .read = null_read_config,
1649        .write = null_write_config,
1650};
1651
1652/*
1653 * These functions are used early on before PCI scanning is done
1654 * and all of the pci_dev and pci_bus structures have been created.
1655 */
1656static struct pci_bus *
1657fake_pci_bus(struct pci_controller *hose, int busnr)
1658{
1659        static struct pci_bus bus;
1660
1661        if (hose == 0) {
1662                printk(KERN_ERR "Can't find hose for PCI bus %d!\n", busnr);
1663        }
1664        bus.number = busnr;
1665        bus.sysdata = hose;
1666        bus.ops = hose? hose->ops: &null_pci_ops;
1667        return &bus;
1668}
1669
1670#define EARLY_PCI_OP(rw, size, type)                                    \
1671int early_##rw##_config_##size(struct pci_controller *hose, int bus,    \
1672                               int devfn, int offset, type value)       \
1673{                                                                       \
1674        return pci_bus_##rw##_config_##size(fake_pci_bus(hose, bus),    \
1675                                            devfn, offset, value);      \
1676}
1677
1678EARLY_PCI_OP(read, byte, u8 *)
1679EARLY_PCI_OP(read, word, u16 *)
1680EARLY_PCI_OP(read, dword, u32 *)
1681EARLY_PCI_OP(write, byte, u8)
1682EARLY_PCI_OP(write, word, u16)
1683EARLY_PCI_OP(write, dword, u32)
1684
1685extern int pci_bus_find_capability (struct pci_bus *bus, unsigned int devfn, int cap);
1686int early_find_capability(struct pci_controller *hose, int bus, int devfn,
1687                          int cap)
1688{
1689        return pci_bus_find_capability(fake_pci_bus(hose, bus), devfn, cap);
1690}
1691
1692/**
1693 * pci_scan_phb - Given a pci_controller, setup and scan the PCI bus
1694 * @hose: Pointer to the PCI host controller instance structure
1695 * @sysdata: value to use for sysdata pointer.  ppc32 and ppc64 differ here
1696 *
1697 * Note: the 'data' pointer is a temporary measure.  As 32 and 64 bit
1698 * pci code gets merged, this parameter should become unnecessary because
1699 * both will use the same value.
1700 */
1701void __devinit pcibios_scan_phb(struct pci_controller *hose, void *sysdata)
1702{
1703        struct pci_bus *bus;
1704        struct device_node *node = hose->dn;
1705        int mode;
1706
1707        pr_debug("PCI: Scanning PHB %s\n",
1708                 node ? node->full_name : "<NO NAME>");
1709
1710        /* Create an empty bus for the toplevel */
1711        bus = pci_create_bus(hose->parent, hose->first_busno, hose->ops,
1712                             sysdata);
1713        if (bus == NULL) {
1714                pr_err("Failed to create bus for PCI domain %04x\n",
1715                        hose->global_number);
1716                return;
1717        }
1718        bus->secondary = hose->first_busno;
1719        hose->bus = bus;
1720
1721        /* Get some IO space for the new PHB */
1722        pcibios_setup_phb_io_space(hose);
1723
1724        /* Wire up PHB bus resources */
1725        pcibios_setup_phb_resources(hose);
1726
1727        /* Get probe mode and perform scan */
1728        mode = PCI_PROBE_NORMAL;
1729        if (node && ppc_md.pci_probe_mode)
1730                mode = ppc_md.pci_probe_mode(bus);
1731        pr_debug("    probe mode: %d\n", mode);
1732        if (mode == PCI_PROBE_DEVTREE) {
1733                bus->subordinate = hose->last_busno;
1734                of_scan_bus(node, bus);
1735        }
1736
1737        if (mode == PCI_PROBE_NORMAL)
1738                hose->last_busno = bus->subordinate = pci_scan_child_bus(bus);
1739}
1740