arch/sparc/kernel/pci_sun4v.c

   1 /* pci_sun4v.c: SUN4V specific PCI controller support.
   2  *
   3  * Copyright (C) 2006, 2007, 2008 David S. Miller (davem@davemloft.net)
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/types.h>
   8 #include <linux/pci.h>
   9 #include <linux/init.h>
  10 #include <linux/slab.h>
  11 #include <linux/interrupt.h>
  12 #include <linux/percpu.h>
  13 #include <linux/irq.h>
  14 #include <linux/msi.h>
  15 #include <linux/export.h>
  16 #include <linux/log2.h>
  17 #include <linux/of_device.h>
  18 #include <linux/iommu-common.h>
  19
  20 #include <asm/iommu.h>
  21 #include <asm/irq.h>
  22 #include <asm/hypervisor.h>
  23 #include <asm/prom.h>
  24
  25 #include "pci_impl.h"
  26 #include "iommu_common.h"
  27 #include "kernel.h"
  28
  29 #include "pci_sun4v.h"
  30
  31 #define DRIVER_NAME     "pci_sun4v"
  32 #define PFX             DRIVER_NAME ": "
  33
  34 static unsigned long vpci_major;
  35 static unsigned long vpci_minor;
  36
  37 struct vpci_version {
  38         unsigned long major;
  39         unsigned long minor;
  40 };
  41
  42 /* Ordered from largest major to lowest */
  43 static struct vpci_version vpci_versions[] = {
  44         { .major = 2, .minor = 0 },
  45         { .major = 1, .minor = 1 },
  46 };
  47
  48 static unsigned long vatu_major = 1;
  49 static unsigned long vatu_minor = 1;
  50
  51 #define PGLIST_NENTS    (PAGE_SIZE / sizeof(u64))
  52
  53 struct iommu_batch {
  54         struct device   *dev;           /* Device mapping is for.       */
  55         unsigned long   prot;           /* IOMMU page protections       */
  56         unsigned long   entry;          /* Index into IOTSB.            */
  57         u64             *pglist;        /* List of physical pages       */
  58         unsigned long   npages;         /* Number of pages in list.     */
  59 };
  60
  61 static DEFINE_PER_CPU(struct iommu_batch, iommu_batch);
  62 static int iommu_batch_initialized;
  63
  64 /* Interrupts must be disabled.  */
  65 static inline void iommu_batch_start(struct device *dev, unsigned long prot, unsigned long entry)
  66 {
  67         struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
  68
  69         p->dev          = dev;
  70         p->prot         = prot;
  71         p->entry        = entry;
  72         p->npages       = 0;
  73 }
  74
  75 /* Interrupts must be disabled.  */
  76 static long iommu_batch_flush(struct iommu_batch *p, u64 mask)
  77 {
  78         struct pci_pbm_info *pbm = p->dev->archdata.host_controller;
  79         u64 *pglist = p->pglist;
  80         u64 index_count;
  81         unsigned long devhandle = pbm->devhandle;
  82         unsigned long prot = p->prot;
  83         unsigned long entry = p->entry;
  84         unsigned long npages = p->npages;
  85         unsigned long iotsb_num;
  86         unsigned long ret;
  87         long num;
  88
  89         /* VPCI maj=1, min=[0,1] only supports read and write */
  90         if (vpci_major < 2)
  91                 prot &= (HV_PCI_MAP_ATTR_READ | HV_PCI_MAP_ATTR_WRITE);
  92
  93         while (npages != 0) {
  94                 if (mask <= DMA_BIT_MASK(32)) {
  95                         num = pci_sun4v_iommu_map(devhandle,
  96                                                   HV_PCI_TSBID(0, entry),
  97                                                   npages,
  98                                                   prot,
  99                                                   __pa(pglist));
 100                         if (unlikely(num < 0)) {
 101                                 pr_err_ratelimited("%s: IOMMU map of [%08lx:%08llx:%lx:%lx:%lx] failed with status %ld\n",
 102                                                    __func__,
 103                                                    devhandle,
 104                                                    HV_PCI_TSBID(0, entry),
 105                                                    npages, prot, __pa(pglist),
 106                                                    num);
 107                                 return -1;
 108                         }
 109                 } else {
 110                         index_count = HV_PCI_IOTSB_INDEX_COUNT(npages, entry),
 111                         iotsb_num = pbm->iommu->atu->iotsb->iotsb_num;
 112                         ret = pci_sun4v_iotsb_map(devhandle,
 113                                                   iotsb_num,
 114                                                   index_count,
 115                                                   prot,
 116                                                   __pa(pglist),
 117                                                   &num);
 118                         if (unlikely(ret != HV_EOK)) {
 119                                 pr_err_ratelimited("%s: ATU map of [%08lx:%lx:%llx:%lx:%lx] failed with status %ld\n",
 120                                                    __func__,
 121                                                    devhandle, iotsb_num,
 122                                                    index_count, prot,
 123                                                    __pa(pglist), ret);
 124                                 return -1;
 125                         }
 126                 }
 127                 entry += num;
 128                 npages -= num;
 129                 pglist += num;
 130         }
 131
 132         p->entry = entry;
 133         p->npages = 0;
 134
 135         return 0;
 136 }
 137
 138 static inline void iommu_batch_new_entry(unsigned long entry, u64 mask)
 139 {
 140         struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
 141
 142         if (p->entry + p->npages == entry)
 143                 return;
 144         if (p->entry != ~0UL)
 145                 iommu_batch_flush(p, mask);
 146         p->entry = entry;
 147 }
 148
 149 /* Interrupts must be disabled.  */
 150 static inline long iommu_batch_add(u64 phys_page, u64 mask)
 151 {
 152         struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
 153
 154         BUG_ON(p->npages >= PGLIST_NENTS);
 155
 156         p->pglist[p->npages++] = phys_page;
 157         if (p->npages == PGLIST_NENTS)
 158                 return iommu_batch_flush(p, mask);
 159
 160         return 0;
 161 }
 162
 163 /* Interrupts must be disabled.  */
 164 static inline long iommu_batch_end(u64 mask)
 165 {
 166         struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
 167
 168         BUG_ON(p->npages >= PGLIST_NENTS);
 169
 170         return iommu_batch_flush(p, mask);
 171 }
 172
 173 static void *dma_4v_alloc_coherent(struct device *dev, size_t size,
 174                                    dma_addr_t *dma_addrp, gfp_t gfp,
 175                                    unsigned long attrs)
 176 {
 177         u64 mask;
 178         unsigned long flags, order, first_page, npages, n;
 179         unsigned long prot = 0;
 180         struct iommu *iommu;
 181         struct atu *atu;
 182         struct iommu_map_table *tbl;
 183         struct page *page;
 184         void *ret;
 185         long entry;
 186         int nid;
 187
 188         size = IO_PAGE_ALIGN(size);
 189         order = get_order(size);
 190         if (unlikely(order >= MAX_ORDER))
 191                 return NULL;
 192
 193         npages = size >> IO_PAGE_SHIFT;
 194
 195         if (attrs & DMA_ATTR_WEAK_ORDERING)
 196                 prot = HV_PCI_MAP_ATTR_RELAXED_ORDER;
 197
 198         nid = dev->archdata.numa_node;
 199         page = alloc_pages_node(nid, gfp, order);
 200         if (unlikely(!page))
 201                 return NULL;
 202
 203         first_page = (unsigned long) page_address(page);
 204         memset((char *)first_page, 0, PAGE_SIZE << order);
 205
 206         iommu = dev->archdata.iommu;
 207         atu = iommu->atu;
 208
 209         mask = dev->coherent_dma_mask;
 210         if (mask <= DMA_BIT_MASK(32))
 211                 tbl = &iommu->tbl;
 212         else
 213                 tbl = &atu->tbl;
 214
 215         entry = iommu_tbl_range_alloc(dev, tbl, npages, NULL,
 216                                       (unsigned long)(-1), 0);
 217
 218         if (unlikely(entry == IOMMU_ERROR_CODE))
 219                 goto range_alloc_fail;
 220
 221         *dma_addrp = (tbl->table_map_base + (entry << IO_PAGE_SHIFT));
 222         ret = (void *) first_page;
 223         first_page = __pa(first_page);
 224
 225         local_irq_save(flags);
 226
 227         iommu_batch_start(dev,
 228                           (HV_PCI_MAP_ATTR_READ | prot |
 229                            HV_PCI_MAP_ATTR_WRITE),
 230                           entry);
 231
 232         for (n = 0; n < npages; n++) {
 233                 long err = iommu_batch_add(first_page + (n * PAGE_SIZE), mask);
 234                 if (unlikely(err < 0L))
 235                         goto iommu_map_fail;
 236         }
 237
 238         if (unlikely(iommu_batch_end(mask) < 0L))
 239                 goto iommu_map_fail;
 240
 241         local_irq_restore(flags);
 242
 243         return ret;
 244
 245 iommu_map_fail:
 246         local_irq_restore(flags);
 247         iommu_tbl_range_free(tbl, *dma_addrp, npages, IOMMU_ERROR_CODE);
 248
 249 range_alloc_fail:
 250         free_pages(first_page, order);
 251         return NULL;
 252 }
 253
 254 unsigned long dma_4v_iotsb_bind(unsigned long devhandle,
 255                                 unsigned long iotsb_num,
 256                                 struct pci_bus *bus_dev)
 257 {
 258         struct pci_dev *pdev;
 259         unsigned long err;
 260         unsigned int bus;
 261         unsigned int device;
 262         unsigned int fun;
 263
 264         list_for_each_entry(pdev, &bus_dev->devices, bus_list) {
 265                 if (pdev->subordinate) {
 266                         /* No need to bind pci bridge */
 267                         dma_4v_iotsb_bind(devhandle, iotsb_num,
 268                                           pdev->subordinate);
 269                 } else {
 270                         bus = bus_dev->number;
 271                         device = PCI_SLOT(pdev->devfn);
 272                         fun = PCI_FUNC(pdev->devfn);
 273                         err = pci_sun4v_iotsb_bind(devhandle, iotsb_num,
 274                                                    HV_PCI_DEVICE_BUILD(bus,
 275                                                                        device,
 276                                                                        fun));
 277
 278                         /* If bind fails for one device it is going to fail
 279                          * for rest of the devices because we are sharing
 280                          * IOTSB. So in case of failure simply return with
 281                          * error.
 282                          */
 283                         if (err)
 284                                 return err;
 285                 }
 286         }
 287
 288         return 0;
 289 }
 290
 291 static void dma_4v_iommu_demap(struct device *dev, unsigned long devhandle,
 292                                dma_addr_t dvma, unsigned long iotsb_num,
 293                                unsigned long entry, unsigned long npages)
 294 {
 295         unsigned long num, flags;
 296         unsigned long ret;
 297
 298         local_irq_save(flags);
 299         do {
 300                 if (dvma <= DMA_BIT_MASK(32)) {
 301                         num = pci_sun4v_iommu_demap(devhandle,
 302                                                     HV_PCI_TSBID(0, entry),
 303                                                     npages);
 304                 } else {
 305                         ret = pci_sun4v_iotsb_demap(devhandle, iotsb_num,
 306                                                     entry, npages, &num);
 307                         if (unlikely(ret != HV_EOK)) {
 308                                 pr_err_ratelimited("pci_iotsb_demap() failed with error: %ld\n",
 309                                                    ret);
 310                         }
 311                 }
 312                 entry += num;
 313                 npages -= num;
 314         } while (npages != 0);
 315         local_irq_restore(flags);
 316 }
 317
 318 static void dma_4v_free_coherent(struct device *dev, size_t size, void *cpu,
 319                                  dma_addr_t dvma, unsigned long attrs)
 320 {
 321         struct pci_pbm_info *pbm;
 322         struct iommu *iommu;
 323         struct atu *atu;
 324         struct iommu_map_table *tbl;
 325         unsigned long order, npages, entry;
 326         unsigned long iotsb_num;
 327         u32 devhandle;
 328
 329         npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
 330         iommu = dev->archdata.iommu;
 331         pbm = dev->archdata.host_controller;
 332         atu = iommu->atu;
 333         devhandle = pbm->devhandle;
 334
 335         if (dvma <= DMA_BIT_MASK(32)) {
 336                 tbl = &iommu->tbl;
 337                 iotsb_num = 0; /* we don't care for legacy iommu */
 338         } else {
 339                 tbl = &atu->tbl;
 340                 iotsb_num = atu->iotsb->iotsb_num;
 341         }
 342         entry = ((dvma - tbl->table_map_base) >> IO_PAGE_SHIFT);
 343         dma_4v_iommu_demap(dev, devhandle, dvma, iotsb_num, entry, npages);
 344         iommu_tbl_range_free(tbl, dvma, npages, IOMMU_ERROR_CODE);
 345         order = get_order(size);
 346         if (order < 10)
 347                 free_pages((unsigned long)cpu, order);
 348 }
 349
 350 static dma_addr_t dma_4v_map_page(struct device *dev, struct page *page,
 351                                   unsigned long offset, size_t sz,
 352                                   enum dma_data_direction direction,
 353                                   unsigned long attrs)
 354 {
 355         struct iommu *iommu;
 356         struct atu *atu;
 357         struct iommu_map_table *tbl;
 358         u64 mask;
 359         unsigned long flags, npages, oaddr;
 360         unsigned long i, base_paddr;
 361         unsigned long prot;
 362         dma_addr_t bus_addr, ret;
 363         long entry;
 364
 365         iommu = dev->archdata.iommu;
 366         atu = iommu->atu;
 367
 368         if (unlikely(direction == DMA_NONE))
 369                 goto bad;
 370
 371         oaddr = (unsigned long)(page_address(page) + offset);
 372         npages = IO_PAGE_ALIGN(oaddr + sz) - (oaddr & IO_PAGE_MASK);
 373         npages >>= IO_PAGE_SHIFT;
 374
 375         mask = *dev->dma_mask;
 376         if (mask <= DMA_BIT_MASK(32))
 377                 tbl = &iommu->tbl;
 378         else
 379                 tbl = &atu->tbl;
 380
 381         entry = iommu_tbl_range_alloc(dev, tbl, npages, NULL,
 382                                       (unsigned long)(-1), 0);
 383
 384         if (unlikely(entry == IOMMU_ERROR_CODE))
 385                 goto bad;
 386
 387         bus_addr = (tbl->table_map_base + (entry << IO_PAGE_SHIFT));
 388         ret = bus_addr | (oaddr & ~IO_PAGE_MASK);
 389         base_paddr = __pa(oaddr & IO_PAGE_MASK);
 390         prot = HV_PCI_MAP_ATTR_READ;
 391         if (direction != DMA_TO_DEVICE)
 392                 prot |= HV_PCI_MAP_ATTR_WRITE;
 393
 394         if (attrs & DMA_ATTR_WEAK_ORDERING)
 395                 prot |= HV_PCI_MAP_ATTR_RELAXED_ORDER;
 396
 397         local_irq_save(flags);
 398
 399         iommu_batch_start(dev, prot, entry);
 400
 401         for (i = 0; i < npages; i++, base_paddr += IO_PAGE_SIZE) {
 402                 long err = iommu_batch_add(base_paddr, mask);
 403                 if (unlikely(err < 0L))
 404                         goto iommu_map_fail;
 405         }
 406         if (unlikely(iommu_batch_end(mask) < 0L))
 407                 goto iommu_map_fail;
 408
 409         local_irq_restore(flags);
 410
 411         return ret;
 412
 413 bad:
 414         if (printk_ratelimit())
 415                 WARN_ON(1);
 416         return SPARC_MAPPING_ERROR;
 417
 418 iommu_map_fail:
 419         local_irq_restore(flags);
 420         iommu_tbl_range_free(tbl, bus_addr, npages, IOMMU_ERROR_CODE);
 421         return SPARC_MAPPING_ERROR;
 422 }
 423
 424 static void dma_4v_unmap_page(struct device *dev, dma_addr_t bus_addr,
 425                               size_t sz, enum dma_data_direction direction,
 426                               unsigned long attrs)
 427 {
 428         struct pci_pbm_info *pbm;
 429         struct iommu *iommu;
 430         struct atu *atu;
 431         struct iommu_map_table *tbl;
 432         unsigned long npages;
 433         unsigned long iotsb_num;
 434         long entry;
 435         u32 devhandle;
 436
 437         if (unlikely(direction == DMA_NONE)) {
 438                 if (printk_ratelimit())
 439                         WARN_ON(1);
 440                 return;
 441         }
 442
 443         iommu = dev->archdata.iommu;
 444         pbm = dev->archdata.host_controller;
 445         atu = iommu->atu;
 446         devhandle = pbm->devhandle;
 447
 448         npages = IO_PAGE_ALIGN(bus_addr + sz) - (bus_addr & IO_PAGE_MASK);
 449         npages >>= IO_PAGE_SHIFT;
 450         bus_addr &= IO_PAGE_MASK;
 451
 452         if (bus_addr <= DMA_BIT_MASK(32)) {
 453                 iotsb_num = 0; /* we don't care for legacy iommu */
 454                 tbl = &iommu->tbl;
 455         } else {
 456                 iotsb_num = atu->iotsb->iotsb_num;
 457                 tbl = &atu->tbl;
 458         }
 459         entry = (bus_addr - tbl->table_map_base) >> IO_PAGE_SHIFT;
 460         dma_4v_iommu_demap(dev, devhandle, bus_addr, iotsb_num, entry, npages);
 461         iommu_tbl_range_free(tbl, bus_addr, npages, IOMMU_ERROR_CODE);
 462 }
 463
 464 static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,
 465                          int nelems, enum dma_data_direction direction,
 466                          unsigned long attrs)
 467 {
 468         struct scatterlist *s, *outs, *segstart;
 469         unsigned long flags, handle, prot;
 470         dma_addr_t dma_next = 0, dma_addr;
 471         unsigned int max_seg_size;
 472         unsigned long seg_boundary_size;
 473         int outcount, incount, i;
 474         struct iommu *iommu;
 475         struct atu *atu;
 476         struct iommu_map_table *tbl;
 477         u64 mask;
 478         unsigned long base_shift;
 479         long err;
 480
 481         BUG_ON(direction == DMA_NONE);
 482
 483         iommu = dev->archdata.iommu;
 484         if (nelems == 0 || !iommu)
 485                 return 0;
 486         atu = iommu->atu;
 487
 488         prot = HV_PCI_MAP_ATTR_READ;
 489         if (direction != DMA_TO_DEVICE)
 490                 prot |= HV_PCI_MAP_ATTR_WRITE;
 491
 492         if (attrs & DMA_ATTR_WEAK_ORDERING)
 493                 prot |= HV_PCI_MAP_ATTR_RELAXED_ORDER;
 494
 495         outs = s = segstart = &sglist[0];
 496         outcount = 1;
 497         incount = nelems;
 498         handle = 0;
 499
 500         /* Init first segment length for backout at failure */
 501         outs->dma_length = 0;
 502
 503         local_irq_save(flags);
 504
 505         iommu_batch_start(dev, prot, ~0UL);
 506
 507         max_seg_size = dma_get_max_seg_size(dev);
 508         seg_boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
 509                                   IO_PAGE_SIZE) >> IO_PAGE_SHIFT;
 510
 511         mask = *dev->dma_mask;
 512         if (mask <= DMA_BIT_MASK(32))
 513                 tbl = &iommu->tbl;
 514         else
 515                 tbl = &atu->tbl;
 516
 517         base_shift = tbl->table_map_base >> IO_PAGE_SHIFT;
 518
 519         for_each_sg(sglist, s, nelems, i) {
 520                 unsigned long paddr, npages, entry, out_entry = 0, slen;
 521
 522                 slen = s->length;
 523                 /* Sanity check */
 524                 if (slen == 0) {
 525                         dma_next = 0;
 526                         continue;
 527                 }
 528                 /* Allocate iommu entries for that segment */
 529                 paddr = (unsigned long) SG_ENT_PHYS_ADDRESS(s);
 530                 npages = iommu_num_pages(paddr, slen, IO_PAGE_SIZE);
 531                 entry = iommu_tbl_range_alloc(dev, tbl, npages,
 532                                               &handle, (unsigned long)(-1), 0);
 533
 534                 /* Handle failure */
 535                 if (unlikely(entry == IOMMU_ERROR_CODE)) {
 536                         pr_err_ratelimited("iommu_alloc failed, iommu %p paddr %lx npages %lx\n",
 537                                            tbl, paddr, npages);
 538                         goto iommu_map_failed;
 539                 }
 540
 541                 iommu_batch_new_entry(entry, mask);
 542
 543                 /* Convert entry to a dma_addr_t */
 544                 dma_addr = tbl->table_map_base + (entry << IO_PAGE_SHIFT);
 545                 dma_addr |= (s->offset & ~IO_PAGE_MASK);
 546
 547                 /* Insert into HW table */
 548                 paddr &= IO_PAGE_MASK;
 549                 while (npages--) {
 550                         err = iommu_batch_add(paddr, mask);
 551                         if (unlikely(err < 0L))
 552                                 goto iommu_map_failed;
 553                         paddr += IO_PAGE_SIZE;
 554                 }
 555
 556                 /* If we are in an open segment, try merging */
 557                 if (segstart != s) {
 558                         /* We cannot merge if:
 559                          * - allocated dma_addr isn't contiguous to previous allocation
 560                          */
 561                         if ((dma_addr != dma_next) ||
 562                             (outs->dma_length + s->length > max_seg_size) ||
 563                             (is_span_boundary(out_entry, base_shift,
 564                                               seg_boundary_size, outs, s))) {
 565                                 /* Can't merge: create a new segment */
 566                                 segstart = s;
 567                                 outcount++;
 568                                 outs = sg_next(outs);
 569                         } else {
 570                                 outs->dma_length += s->length;
 571                         }
 572                 }
 573
 574                 if (segstart == s) {
 575                         /* This is a new segment, fill entries */
 576                         outs->dma_address = dma_addr;
 577                         outs->dma_length = slen;
 578                         out_entry = entry;
 579                 }
 580
 581                 /* Calculate next page pointer for contiguous check */
 582                 dma_next = dma_addr + slen;
 583         }
 584
 585         err = iommu_batch_end(mask);
 586
 587         if (unlikely(err < 0L))
 588                 goto iommu_map_failed;
 589
 590         local_irq_restore(flags);
 591
 592         if (outcount < incount) {
 593                 outs = sg_next(outs);
 594                 outs->dma_address = SPARC_MAPPING_ERROR;
 595                 outs->dma_length = 0;
 596         }
 597
 598         return outcount;
 599
 600 iommu_map_failed:
 601         for_each_sg(sglist, s, nelems, i) {
 602                 if (s->dma_length != 0) {
 603                         unsigned long vaddr, npages;
 604
 605                         vaddr = s->dma_address & IO_PAGE_MASK;
 606                         npages = iommu_num_pages(s->dma_address, s->dma_length,
 607                                                  IO_PAGE_SIZE);
 608                         iommu_tbl_range_free(tbl, vaddr, npages,
 609                                              IOMMU_ERROR_CODE);
 610                         /* XXX demap? XXX */
 611                         s->dma_address = SPARC_MAPPING_ERROR;
 612                         s->dma_length = 0;
 613                 }
 614                 if (s == outs)
 615                         break;
 616         }
 617         local_irq_restore(flags);
 618
 619         return 0;
 620 }
 621
 622 static void dma_4v_unmap_sg(struct device *dev, struct scatterlist *sglist,
 623                             int nelems, enum dma_data_direction direction,
 624                             unsigned long attrs)
 625 {
 626         struct pci_pbm_info *pbm;
 627         struct scatterlist *sg;
 628         struct iommu *iommu;
 629         struct atu *atu;
 630         unsigned long flags, entry;
 631         unsigned long iotsb_num;
 632         u32 devhandle;
 633
 634         BUG_ON(direction == DMA_NONE);
 635
 636         iommu = dev->archdata.iommu;
 637         pbm = dev->archdata.host_controller;
 638         atu = iommu->atu;
 639         devhandle = pbm->devhandle;
 640
 641         local_irq_save(flags);
 642
 643         sg = sglist;
 644         while (nelems--) {
 645                 dma_addr_t dma_handle = sg->dma_address;
 646                 unsigned int len = sg->dma_length;
 647                 unsigned long npages;
 648                 struct iommu_map_table *tbl;
 649                 unsigned long shift = IO_PAGE_SHIFT;
 650
 651                 if (!len)
 652                         break;
 653                 npages = iommu_num_pages(dma_handle, len, IO_PAGE_SIZE);
 654
 655                 if (dma_handle <= DMA_BIT_MASK(32)) {
 656                         iotsb_num = 0; /* we don't care for legacy iommu */
 657                         tbl = &iommu->tbl;
 658                 } else {
 659                         iotsb_num = atu->iotsb->iotsb_num;
 660                         tbl = &atu->tbl;
 661                 }
 662                 entry = ((dma_handle - tbl->table_map_base) >> shift);
 663                 dma_4v_iommu_demap(dev, devhandle, dma_handle, iotsb_num,
 664                                    entry, npages);
 665                 iommu_tbl_range_free(tbl, dma_handle, npages,
 666                                      IOMMU_ERROR_CODE);
 667                 sg = sg_next(sg);
 668         }
 669
 670         local_irq_restore(flags);
 671 }
 672
 673 static int dma_4v_supported(struct device *dev, u64 device_mask)
 674 {
 675         struct iommu *iommu = dev->archdata.iommu;
 676         u64 dma_addr_mask;
 677
 678         if (device_mask > DMA_BIT_MASK(32) && iommu->atu)
 679                 dma_addr_mask = iommu->atu->dma_addr_mask;
 680         else
 681                 dma_addr_mask = iommu->dma_addr_mask;
 682
 683         if ((device_mask & dma_addr_mask) == dma_addr_mask)
 684                 return 1;
 685         return pci64_dma_supported(to_pci_dev(dev), device_mask);
 686 }
 687
 688 static int dma_4v_mapping_error(struct device *dev, dma_addr_t dma_addr)
 689 {
 690         return dma_addr == SPARC_MAPPING_ERROR;
 691 }
 692
 693 static const struct dma_map_ops sun4v_dma_ops = {
 694         .alloc                          = dma_4v_alloc_coherent,
 695         .free                           = dma_4v_free_coherent,
 696         .map_page                       = dma_4v_map_page,
 697         .unmap_page                     = dma_4v_unmap_page,
 698         .map_sg                         = dma_4v_map_sg,
 699         .unmap_sg                       = dma_4v_unmap_sg,
 700         .dma_supported                  = dma_4v_supported,
 701         .mapping_error                  = dma_4v_mapping_error,
 702 };
 703
 704 static void pci_sun4v_scan_bus(struct pci_pbm_info *pbm, struct device *parent)
 705 {
 706         struct property *prop;
 707         struct device_node *dp;
 708
 709         dp = pbm->op->dev.of_node;
 710         prop = of_find_property(dp, "66mhz-capable", NULL);
 711         pbm->is_66mhz_capable = (prop != NULL);
 712         pbm->pci_bus = pci_scan_one_pbm(pbm, parent);
 713
 714         /* XXX register error interrupt handlers XXX */
 715 }
 716
 717 static unsigned long probe_existing_entries(struct pci_pbm_info *pbm,
 718                                             struct iommu_map_table *iommu)
 719 {
 720         struct iommu_pool *pool;
 721         unsigned long i, pool_nr, cnt = 0;
 722         u32 devhandle;
 723
 724         devhandle = pbm->devhandle;
 725         for (pool_nr = 0; pool_nr < iommu->nr_pools; pool_nr++) {
 726                 pool = &(iommu->pools[pool_nr]);
 727                 for (i = pool->start; i <= pool->end; i++) {
 728                         unsigned long ret, io_attrs, ra;
 729
 730                         ret = pci_sun4v_iommu_getmap(devhandle,
 731                                                      HV_PCI_TSBID(0, i),
 732                                                      &io_attrs, &ra);
 733                         if (ret == HV_EOK) {
 734                                 if (page_in_phys_avail(ra)) {
 735                                         pci_sun4v_iommu_demap(devhandle,
 736                                                               HV_PCI_TSBID(0,
 737                                                               i), 1);
 738                                 } else {
 739                                         cnt++;
 740                                         __set_bit(i, iommu->map);
 741                                 }
 742                         }
 743                 }
 744         }
 745         return cnt;
 746 }
 747
 748 static int pci_sun4v_atu_alloc_iotsb(struct pci_pbm_info *pbm)
 749 {
 750         struct atu *atu = pbm->iommu->atu;
 751         struct atu_iotsb *iotsb;
 752         void *table;
 753         u64 table_size;
 754         u64 iotsb_num;
 755         unsigned long order;
 756         unsigned long err;
 757
 758         iotsb = kzalloc(sizeof(*iotsb), GFP_KERNEL);
 759         if (!iotsb) {
 760                 err = -ENOMEM;
 761                 goto out_err;
 762         }
 763         atu->iotsb = iotsb;
 764
 765         /* calculate size of IOTSB */
 766         table_size = (atu->size / IO_PAGE_SIZE) * 8;
 767         order = get_order(table_size);
 768         table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
 769         if (!table) {
 770                 err = -ENOMEM;
 771                 goto table_failed;
 772         }
 773         iotsb->table = table;
 774         iotsb->ra = __pa(table);
 775         iotsb->dvma_size = atu->size;
 776         iotsb->dvma_base = atu->base;
 777         iotsb->table_size = table_size;
 778         iotsb->page_size = IO_PAGE_SIZE;
 779
 780         /* configure and register IOTSB with HV */
 781         err = pci_sun4v_iotsb_conf(pbm->devhandle,
 782                                    iotsb->ra,
 783                                    iotsb->table_size,
 784                                    iotsb->page_size,
 785                                    iotsb->dvma_base,
 786                                    &iotsb_num);
 787         if (err) {
 788                 pr_err(PFX "pci_iotsb_conf failed error: %ld\n", err);
 789                 goto iotsb_conf_failed;
 790         }
 791         iotsb->iotsb_num = iotsb_num;
 792
 793         err = dma_4v_iotsb_bind(pbm->devhandle, iotsb_num, pbm->pci_bus);
 794         if (err) {
 795                 pr_err(PFX "pci_iotsb_bind failed error: %ld\n", err);
 796                 goto iotsb_conf_failed;
 797         }
 798
 799         return 0;
 800
 801 iotsb_conf_failed:
 802         free_pages((unsigned long)table, order);
 803 table_failed:
 804         kfree(iotsb);
 805 out_err:
 806         return err;
 807 }
 808
 809 static int pci_sun4v_atu_init(struct pci_pbm_info *pbm)
 810 {
 811         struct atu *atu = pbm->iommu->atu;
 812         unsigned long err;
 813         const u64 *ranges;
 814         u64 map_size, num_iotte;
 815         u64 dma_mask;
 816         const u32 *page_size;
 817         int len;
 818
 819         ranges = of_get_property(pbm->op->dev.of_node, "iommu-address-ranges",
 820                                  &len);
 821         if (!ranges) {
 822                 pr_err(PFX "No iommu-address-ranges\n");
 823                 return -EINVAL;
 824         }
 825
 826         page_size = of_get_property(pbm->op->dev.of_node, "iommu-pagesizes",
 827                                     NULL);
 828         if (!page_size) {
 829                 pr_err(PFX "No iommu-pagesizes\n");
 830                 return -EINVAL;
 831         }
 832
 833         /* There are 4 iommu-address-ranges supported. Each range is pair of
 834          * {base, size}. The ranges[0] and ranges[1] are 32bit address space
 835          * while ranges[2] and ranges[3] are 64bit space.  We want to use 64bit
 836          * address ranges to support 64bit addressing. Because 'size' for
 837          * address ranges[2] and ranges[3] are same we can select either of
 838          * ranges[2] or ranges[3] for mapping. However due to 'size' is too
 839          * large for OS to allocate IOTSB we are using fix size 32G
 840          * (ATU_64_SPACE_SIZE) which is more than enough for all PCIe devices
 841          * to share.
 842          */
 843         atu->ranges = (struct atu_ranges *)ranges;
 844         atu->base = atu->ranges[3].base;
 845         atu->size = ATU_64_SPACE_SIZE;
 846
 847         /* Create IOTSB */
 848         err = pci_sun4v_atu_alloc_iotsb(pbm);
 849         if (err) {
 850                 pr_err(PFX "Error creating ATU IOTSB\n");
 851                 return err;
 852         }
 853
 854         /* Create ATU iommu map.
 855          * One bit represents one iotte in IOTSB table.
 856          */
 857         dma_mask = (roundup_pow_of_two(atu->size) - 1UL);
 858         num_iotte = atu->size / IO_PAGE_SIZE;
 859         map_size = num_iotte / 8;
 860         atu->tbl.table_map_base = atu->base;
 861         atu->dma_addr_mask = dma_mask;
 862         atu->tbl.map = kzalloc(map_size, GFP_KERNEL);
 863         if (!atu->tbl.map)
 864                 return -ENOMEM;
 865
 866         iommu_tbl_pool_init(&atu->tbl, num_iotte, IO_PAGE_SHIFT,
 867                             NULL, false /* no large_pool */,
 868                             0 /* default npools */,
 869                             false /* want span boundary checking */);
 870
 871         return 0;
 872 }
 873
 874 static int pci_sun4v_iommu_init(struct pci_pbm_info *pbm)
 875 {
 876         static const u32 vdma_default[] = { 0x80000000, 0x80000000 };
 877         struct iommu *iommu = pbm->iommu;
 878         unsigned long num_tsb_entries, sz;
 879         u32 dma_mask, dma_offset;
 880         const u32 *vdma;
 881
 882         vdma = of_get_property(pbm->op->dev.of_node, "virtual-dma", NULL);
 883         if (!vdma)
 884                 vdma = vdma_default;
 885
 886         if ((vdma[0] | vdma[1]) & ~IO_PAGE_MASK) {
 887                 printk(KERN_ERR PFX "Strange virtual-dma[%08x:%08x].\n",
 888                        vdma[0], vdma[1]);
 889                 return -EINVAL;
 890         }
 891
 892         dma_mask = (roundup_pow_of_two(vdma[1]) - 1UL);
 893         num_tsb_entries = vdma[1] / IO_PAGE_SIZE;
 894
 895         dma_offset = vdma[0];
 896
 897         /* Setup initial software IOMMU state. */
 898         spin_lock_init(&iommu->lock);
 899         iommu->ctx_lowest_free = 1;
 900         iommu->tbl.table_map_base = dma_offset;
 901         iommu->dma_addr_mask = dma_mask;
 902
 903         /* Allocate and initialize the free area map.  */
 904         sz = (num_tsb_entries + 7) / 8;
 905         sz = (sz + 7UL) & ~7UL;
 906         iommu->tbl.map = kzalloc(sz, GFP_KERNEL);
 907         if (!iommu->tbl.map) {
 908                 printk(KERN_ERR PFX "Error, kmalloc(arena.map) failed.\n");
 909                 return -ENOMEM;
 910         }
 911         iommu_tbl_pool_init(&iommu->tbl, num_tsb_entries, IO_PAGE_SHIFT,
 912                             NULL, false /* no large_pool */,
 913                             0 /* default npools */,
 914                             false /* want span boundary checking */);
 915         sz = probe_existing_entries(pbm, &iommu->tbl);
 916         if (sz)
 917                 printk("%s: Imported %lu TSB entries from OBP\n",
 918                        pbm->name, sz);
 919
 920         return 0;
 921 }
 922
 923 #ifdef CONFIG_PCI_MSI
 924 struct pci_sun4v_msiq_entry {
 925         u64             version_type;
 926 #define MSIQ_VERSION_MASK               0xffffffff00000000UL
 927 #define MSIQ_VERSION_SHIFT              32
 928 #define MSIQ_TYPE_MASK                  0x00000000000000ffUL
 929 #define MSIQ_TYPE_SHIFT                 0
 930 #define MSIQ_TYPE_NONE                  0x00
 931 #define MSIQ_TYPE_MSG                   0x01
 932 #define MSIQ_TYPE_MSI32                 0x02
 933 #define MSIQ_TYPE_MSI64                 0x03
 934 #define MSIQ_TYPE_INTX                  0x08
 935 #define MSIQ_TYPE_NONE2                 0xff
 936
 937         u64             intx_sysino;
 938         u64             reserved1;
 939         u64             stick;
 940         u64             req_id;  /* bus/device/func */
 941 #define MSIQ_REQID_BUS_MASK             0xff00UL
 942 #define MSIQ_REQID_BUS_SHIFT            8
 943 #define MSIQ_REQID_DEVICE_MASK          0x00f8UL
 944 #define MSIQ_REQID_DEVICE_SHIFT         3
 945 #define MSIQ_REQID_FUNC_MASK            0x0007UL
 946 #define MSIQ_REQID_FUNC_SHIFT           0
 947
 948         u64             msi_address;
 949
 950         /* The format of this value is message type dependent.
 951          * For MSI bits 15:0 are the data from the MSI packet.
 952          * For MSI-X bits 31:0 are the data from the MSI packet.
 953          * For MSG, the message code and message routing code where:
 954          *      bits 39:32 is the bus/device/fn of the msg target-id
 955          *      bits 18:16 is the message routing code
 956          *      bits 7:0 is the message code
 957          * For INTx the low order 2-bits are:
 958          *      00 - INTA
 959          *      01 - INTB
 960          *      10 - INTC
 961          *      11 - INTD
 962          */
 963         u64             msi_data;
 964
 965         u64             reserved2;
 966 };
 967
 968 static int pci_sun4v_get_head(struct pci_pbm_info *pbm, unsigned long msiqid,
 969                               unsigned long *head)
 970 {
 971         unsigned long err, limit;
 972
 973         err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, head);
 974         if (unlikely(err))
 975                 return -ENXIO;
 976
 977         limit = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
 978         if (unlikely(*head >= limit))
 979                 return -EFBIG;
 980
 981         return 0;
 982 }
 983
 984 static int pci_sun4v_dequeue_msi(struct pci_pbm_info *pbm,
 985                                  unsigned long msiqid, unsigned long *head,
 986                                  unsigned long *msi)
 987 {
 988         struct pci_sun4v_msiq_entry *ep;
 989         unsigned long err, type;
 990
 991         /* Note: void pointer arithmetic, 'head' is a byte offset  */
 992         ep = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
 993                                  (pbm->msiq_ent_count *
 994                                   sizeof(struct pci_sun4v_msiq_entry))) +
 995               *head);
 996
 997         if ((ep->version_type & MSIQ_TYPE_MASK) == 0)
 998                 return 0;
 999
1000         type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
1001         if (unlikely(type != MSIQ_TYPE_MSI32 &&
1002                      type != MSIQ_TYPE_MSI64))
1003                 return -EINVAL;
1004
1005         *msi = ep->msi_data;
1006
1007         err = pci_sun4v_msi_setstate(pbm->devhandle,
1008                                      ep->msi_data /* msi_num */,
1009                                      HV_MSISTATE_IDLE);
1010         if (unlikely(err))
1011                 return -ENXIO;
1012
1013         /* Clear the entry.  */
1014         ep->version_type &= ~MSIQ_TYPE_MASK;
1015
1016         (*head) += sizeof(struct pci_sun4v_msiq_entry);
1017         if (*head >=
1018             (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry)))
1019                 *head = 0;
1020
1021         return 1;
1022 }
1023
1024 static int pci_sun4v_set_head(struct pci_pbm_info *pbm, unsigned long msiqid,
1025                               unsigned long head)
1026 {
1027         unsigned long err;
1028
1029         err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
1030         if (unlikely(err))
1031                 return -EINVAL;
1032
1033         return 0;
1034 }
1035
1036 static int pci_sun4v_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid,
1037                                unsigned long msi, int is_msi64)
1038 {
1039         if (pci_sun4v_msi_setmsiq(pbm->devhandle, msi, msiqid,
1040                                   (is_msi64 ?
1041                                    HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
1042                 return -ENXIO;
1043         if (pci_sun4v_msi_setstate(pbm->devhandle, msi, HV_MSISTATE_IDLE))
1044                 return -ENXIO;
1045         if (pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_VALID))
1046                 return -ENXIO;
1047         return 0;
1048 }
1049
1050 static int pci_sun4v_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi)
1051 {
1052         unsigned long err, msiqid;
1053
1054         err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi, &msiqid);
1055         if (err)
1056                 return -ENXIO;
1057
1058         pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_INVALID);
1059
1060         return 0;
1061 }
1062
1063 static int pci_sun4v_msiq_alloc(struct pci_pbm_info *pbm)
1064 {
1065         unsigned long q_size, alloc_size, pages, order;
1066         int i;
1067
1068         q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
1069         alloc_size = (pbm->msiq_num * q_size);
1070         order = get_order(alloc_size);
1071         pages = __get_free_pages(GFP_KERNEL | __GFP_COMP, order);
1072         if (pages == 0UL) {
1073                 printk(KERN_ERR "MSI: Cannot allocate MSI queues (o=%lu).\n",
1074                        order);
1075                 return -ENOMEM;
1076         }
1077         memset((char *)pages, 0, PAGE_SIZE << order);
1078         pbm->msi_queues = (void *) pages;
1079
1080         for (i = 0; i < pbm->msiq_num; i++) {
1081                 unsigned long err, base = __pa(pages + (i * q_size));
1082                 unsigned long ret1, ret2;
1083
1084                 err = pci_sun4v_msiq_conf(pbm->devhandle,
1085                                           pbm->msiq_first + i,
1086                                           base, pbm->msiq_ent_count);
1087                 if (err) {
1088                         printk(KERN_ERR "MSI: msiq register fails (err=%lu)\n",
1089                                err);
1090                         goto h_error;
1091                 }
1092
1093                 err = pci_sun4v_msiq_info(pbm->devhandle,
1094                                           pbm->msiq_first + i,
1095                                           &ret1, &ret2);
1096                 if (err) {
1097                         printk(KERN_ERR "MSI: Cannot read msiq (err=%lu)\n",
1098                                err);
1099                         goto h_error;
1100                 }
1101                 if (ret1 != base || ret2 != pbm->msiq_ent_count) {
1102                         printk(KERN_ERR "MSI: Bogus qconf "
1103                                "expected[%lx:%x] got[%lx:%lx]\n",
1104                                base, pbm->msiq_ent_count,
1105                                ret1, ret2);
1106                         goto h_error;
1107                 }
1108         }
1109
1110         return 0;
1111
1112 h_error:
1113         free_pages(pages, order);
1114         return -EINVAL;
1115 }
1116
1117 static void pci_sun4v_msiq_free(struct pci_pbm_info *pbm)
1118 {
1119         unsigned long q_size, alloc_size, pages, order;
1120         int i;
1121
1122         for (i = 0; i < pbm->msiq_num; i++) {
1123                 unsigned long msiqid = pbm->msiq_first + i;
1124
1125                 (void) pci_sun4v_msiq_conf(pbm->devhandle, msiqid, 0UL, 0);
1126         }
1127
1128         q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
1129         alloc_size = (pbm->msiq_num * q_size);
1130         order = get_order(alloc_size);
1131
1132         pages = (unsigned long) pbm->msi_queues;
1133
1134         free_pages(pages, order);
1135
1136         pbm->msi_queues = NULL;
1137 }
1138
1139 static int pci_sun4v_msiq_build_irq(struct pci_pbm_info *pbm,
1140                                     unsigned long msiqid,
1141                                     unsigned long devino)
1142 {
1143         unsigned int irq = sun4v_build_irq(pbm->devhandle, devino);
1144
1145         if (!irq)
1146                 return -ENOMEM;
1147
1148         if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
1149                 return -EINVAL;
1150         if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
1151                 return -EINVAL;
1152
1153         return irq;
1154 }
1155
1156 static const struct sparc64_msiq_ops pci_sun4v_msiq_ops = {
1157         .get_head       =       pci_sun4v_get_head,
1158         .dequeue_msi    =       pci_sun4v_dequeue_msi,
1159         .set_head       =       pci_sun4v_set_head,
1160         .msi_setup      =       pci_sun4v_msi_setup,
1161         .msi_teardown   =       pci_sun4v_msi_teardown,
1162         .msiq_alloc     =       pci_sun4v_msiq_alloc,
1163         .msiq_free      =       pci_sun4v_msiq_free,
1164         .msiq_build_irq =       pci_sun4v_msiq_build_irq,
1165 };
1166
1167 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
1168 {
1169         sparc64_pbm_msi_init(pbm, &pci_sun4v_msiq_ops);
1170 }
1171 #else /* CONFIG_PCI_MSI */
1172 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
1173 {
1174 }
1175 #endif /* !(CONFIG_PCI_MSI) */
1176
1177 static int pci_sun4v_pbm_init(struct pci_pbm_info *pbm,
1178                               struct platform_device *op, u32 devhandle)
1179 {
1180         struct device_node *dp = op->dev.of_node;
1181         int err;
1182
1183         pbm->numa_node = of_node_to_nid(dp);
1184
1185         pbm->pci_ops = &sun4v_pci_ops;
1186         pbm->config_space_reg_bits = 12;
1187
1188         pbm->index = pci_num_pbms++;
1189
1190         pbm->op = op;
1191
1192         pbm->devhandle = devhandle;
1193
1194         pbm->name = dp->full_name;
1195
1196         printk("%s: SUN4V PCI Bus Module\n", pbm->name);
1197         printk("%s: On NUMA node %d\n", pbm->name, pbm->numa_node);
1198
1199         pci_determine_mem_io_space(pbm);
1200
1201         pci_get_pbm_props(pbm);
1202
1203         err = pci_sun4v_iommu_init(pbm);
1204         if (err)
1205                 return err;
1206
1207         pci_sun4v_msi_init(pbm);
1208
1209         pci_sun4v_scan_bus(pbm, &op->dev);
1210
1211         /* if atu_init fails its not complete failure.
1212          * we can still continue using legacy iommu.
1213          */
1214         if (pbm->iommu->atu) {
1215                 err = pci_sun4v_atu_init(pbm);
1216                 if (err) {
1217                         kfree(pbm->iommu->atu);
1218                         pbm->iommu->atu = NULL;
1219                         pr_err(PFX "ATU init failed, err=%d\n", err);
1220                 }
1221         }
1222
1223         pbm->next = pci_pbm_root;
1224         pci_pbm_root = pbm;
1225
1226         return 0;
1227 }
1228
1229 static int pci_sun4v_probe(struct platform_device *op)
1230 {
1231         const struct linux_prom64_registers *regs;
1232         static int hvapi_negotiated = 0;
1233         struct pci_pbm_info *pbm;
1234         struct device_node *dp;
1235         struct iommu *iommu;
1236         struct atu *atu;
1237         u32 devhandle;
1238         int i, err = -ENODEV;
1239         static bool hv_atu = true;
1240
1241         dp = op->dev.of_node;
1242
1243         if (!hvapi_negotiated++) {
1244                 for (i = 0; i < ARRAY_SIZE(vpci_versions); i++) {
1245                         vpci_major = vpci_versions[i].major;
1246                         vpci_minor = vpci_versions[i].minor;
1247
1248                         err = sun4v_hvapi_register(HV_GRP_PCI, vpci_major,
1249                                                    &vpci_minor);
1250                         if (!err)
1251                                 break;
1252                 }
1253
1254                 if (err) {
1255                         pr_err(PFX "Could not register hvapi, err=%d\n", err);
1256                         return err;
1257                 }
1258                 pr_info(PFX "Registered hvapi major[%lu] minor[%lu]\n",
1259                         vpci_major, vpci_minor);
1260
1261                 err = sun4v_hvapi_register(HV_GRP_ATU, vatu_major, &vatu_minor);
1262                 if (err) {
1263                         /* don't return an error if we fail to register the
1264                          * ATU group, but ATU hcalls won't be available.
1265                          */
1266                         hv_atu = false;
1267                         pr_err(PFX "Could not register hvapi ATU err=%d\n",
1268                                err);
1269                 } else {
1270                         pr_info(PFX "Registered hvapi ATU major[%lu] minor[%lu]\n",
1271                                 vatu_major, vatu_minor);
1272                 }
1273
1274                 dma_ops = &sun4v_dma_ops;
1275         }
1276
1277         regs = of_get_property(dp, "reg", NULL);
1278         err = -ENODEV;
1279         if (!regs) {
1280                 printk(KERN_ERR PFX "Could not find config registers\n");
1281                 goto out_err;
1282         }
1283         devhandle = (regs->phys_addr >> 32UL) & 0x0fffffff;
1284
1285         err = -ENOMEM;
1286         if (!iommu_batch_initialized) {
1287                 for_each_possible_cpu(i) {
1288                         unsigned long page = get_zeroed_page(GFP_KERNEL);
1289
1290                         if (!page)
1291                                 goto out_err;
1292
1293                         per_cpu(iommu_batch, i).pglist = (u64 *) page;
1294                 }
1295                 iommu_batch_initialized = 1;
1296         }
1297
1298         pbm = kzalloc(sizeof(*pbm), GFP_KERNEL);
1299         if (!pbm) {
1300                 printk(KERN_ERR PFX "Could not allocate pci_pbm_info\n");
1301                 goto out_err;
1302         }
1303
1304         iommu = kzalloc(sizeof(struct iommu), GFP_KERNEL);
1305         if (!iommu) {
1306                 printk(KERN_ERR PFX "Could not allocate pbm iommu\n");
1307                 goto out_free_controller;
1308         }
1309
1310         pbm->iommu = iommu;
1311         iommu->atu = NULL;
1312         if (hv_atu) {
1313                 atu = kzalloc(sizeof(*atu), GFP_KERNEL);
1314                 if (!atu)
1315                         pr_err(PFX "Could not allocate atu\n");
1316                 else
1317                         iommu->atu = atu;
1318         }
1319
1320         err = pci_sun4v_pbm_init(pbm, op, devhandle);
1321         if (err)
1322                 goto out_free_iommu;
1323
1324         dev_set_drvdata(&op->dev, pbm);
1325
1326         return 0;
1327
1328 out_free_iommu:
1329         kfree(iommu->atu);
1330         kfree(pbm->iommu);
1331
1332 out_free_controller:
1333         kfree(pbm);
1334
1335 out_err:
1336         return err;
1337 }
1338
1339 static const struct of_device_id pci_sun4v_match[] = {
1340         {
1341                 .name = "pci",
1342                 .compatible = "SUNW,sun4v-pci",
1343         },
1344         {},
1345 };
1346
1347 static struct platform_driver pci_sun4v_driver = {
1348         .driver = {
1349                 .name = DRIVER_NAME,
1350                 .of_match_table = pci_sun4v_match,
1351         },
1352         .probe          = pci_sun4v_probe,
1353 };
1354
1355 static int __init pci_sun4v_init(void)
1356 {
1357         return platform_driver_register(&pci_sun4v_driver);
1358 }
1359
1360 subsys_initcall(pci_sun4v_init);