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 ([email protected])
   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
  28 #include "pci_sun4v.h"
  29
  30 #define DRIVER_NAME     "pci_sun4v"
  31 #define PFX             DRIVER_NAME ": "
  32
  33 static unsigned long vpci_major;
  34 static unsigned long vpci_minor;
  35
  36 struct vpci_version {
  37         unsigned long major;
  38         unsigned long minor;
  39 };
  40
  41 /* Ordered from largest major to lowest */
  42 static struct vpci_version vpci_versions[] = {
  43         { .major = 2, .minor = 0 },
  44         { .major = 1, .minor = 1 },
  45 };
  46
  47 #define PGLIST_NENTS    (PAGE_SIZE / sizeof(u64))
  48
  49 struct iommu_batch {
  50         struct device   *dev;           /* Device mapping is for.       */
  51         unsigned long   prot;           /* IOMMU page protections       */
  52         unsigned long   entry;          /* Index into IOTSB.            */
  53         u64             *pglist;        /* List of physical pages       */
  54         unsigned long   npages;         /* Number of pages in list.     */
  55 };
  56
  57 static DEFINE_PER_CPU(struct iommu_batch, iommu_batch);
  58 static int iommu_batch_initialized;
  59
  60 /* Interrupts must be disabled.  */
  61 static inline void iommu_batch_start(struct device *dev, unsigned long prot, unsigned long entry)
  62 {
  63         struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
  64
  65         p->dev          = dev;
  66         p->prot         = prot;
  67         p->entry        = entry;
  68         p->npages       = 0;
  69 }
  70
  71 /* Interrupts must be disabled.  */
  72 static long iommu_batch_flush(struct iommu_batch *p)
  73 {
  74         struct pci_pbm_info *pbm = p->dev->archdata.host_controller;
  75         unsigned long devhandle = pbm->devhandle;
  76         unsigned long prot = p->prot;
  77         unsigned long entry = p->entry;
  78         u64 *pglist = p->pglist;
  79         unsigned long npages = p->npages;
  80
  81         /* VPCI maj=1, min=[0,1] only supports read and write */
  82         if (vpci_major < 2)
  83                 prot &= (HV_PCI_MAP_ATTR_READ | HV_PCI_MAP_ATTR_WRITE);
  84
  85         while (npages != 0) {
  86                 long num;
  87
  88                 num = pci_sun4v_iommu_map(devhandle, HV_PCI_TSBID(0, entry),
  89                                           npages, prot, __pa(pglist));
  90                 if (unlikely(num < 0)) {
  91                         if (printk_ratelimit())
  92                                 printk("iommu_batch_flush: IOMMU map of "
  93                                        "[%08lx:%08llx:%lx:%lx:%lx] failed with "
  94                                        "status %ld\n",
  95                                        devhandle, HV_PCI_TSBID(0, entry),
  96                                        npages, prot, __pa(pglist), num);
  97                         return -1;
  98                 }
  99
 100                 entry += num;
 101                 npages -= num;
 102                 pglist += num;
 103         }
 104
 105         p->entry = entry;
 106         p->npages = 0;
 107
 108         return 0;
 109 }
 110
 111 static inline void iommu_batch_new_entry(unsigned long entry)
 112 {
 113         struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
 114
 115         if (p->entry + p->npages == entry)
 116                 return;
 117         if (p->entry != ~0UL)
 118                 iommu_batch_flush(p);
 119         p->entry = entry;
 120 }
 121
 122 /* Interrupts must be disabled.  */
 123 static inline long iommu_batch_add(u64 phys_page)
 124 {
 125         struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
 126
 127         BUG_ON(p->npages >= PGLIST_NENTS);
 128
 129         p->pglist[p->npages++] = phys_page;
 130         if (p->npages == PGLIST_NENTS)
 131                 return iommu_batch_flush(p);
 132
 133         return 0;
 134 }
 135
 136 /* Interrupts must be disabled.  */
 137 static inline long iommu_batch_end(void)
 138 {
 139         struct iommu_batch *p = this_cpu_ptr(&iommu_batch);
 140
 141         BUG_ON(p->npages >= PGLIST_NENTS);
 142
 143         return iommu_batch_flush(p);
 144 }
 145
 146 static void *dma_4v_alloc_coherent(struct device *dev, size_t size,
 147                                    dma_addr_t *dma_addrp, gfp_t gfp,
 148                                    unsigned long attrs)
 149 {
 150         unsigned long flags, order, first_page, npages, n;
 151         unsigned long prot = 0;
 152         struct iommu *iommu;
 153         struct page *page;
 154         void *ret;
 155         long entry;
 156         int nid;
 157
 158         size = IO_PAGE_ALIGN(size);
 159         order = get_order(size);
 160         if (unlikely(order >= MAX_ORDER))
 161                 return NULL;
 162
 163         npages = size >> IO_PAGE_SHIFT;
 164
 165         if (attrs & DMA_ATTR_WEAK_ORDERING)
 166                 prot = HV_PCI_MAP_ATTR_RELAXED_ORDER;
 167
 168         nid = dev->archdata.numa_node;
 169         page = alloc_pages_node(nid, gfp, order);
 170         if (unlikely(!page))
 171                 return NULL;
 172
 173         first_page = (unsigned long) page_address(page);
 174         memset((char *)first_page, 0, PAGE_SIZE << order);
 175
 176         iommu = dev->archdata.iommu;
 177
 178         entry = iommu_tbl_range_alloc(dev, &iommu->tbl, npages, NULL,
 179                                       (unsigned long)(-1), 0);
 180
 181         if (unlikely(entry == IOMMU_ERROR_CODE))
 182                 goto range_alloc_fail;
 183
 184         *dma_addrp = (iommu->tbl.table_map_base + (entry << IO_PAGE_SHIFT));
 185         ret = (void *) first_page;
 186         first_page = __pa(first_page);
 187
 188         local_irq_save(flags);
 189
 190         iommu_batch_start(dev,
 191                           (HV_PCI_MAP_ATTR_READ | prot |
 192                            HV_PCI_MAP_ATTR_WRITE),
 193                           entry);
 194
 195         for (n = 0; n < npages; n++) {
 196                 long err = iommu_batch_add(first_page + (n * PAGE_SIZE));
 197                 if (unlikely(err < 0L))
 198                         goto iommu_map_fail;
 199         }
 200
 201         if (unlikely(iommu_batch_end() < 0L))
 202                 goto iommu_map_fail;
 203
 204         local_irq_restore(flags);
 205
 206         return ret;
 207
 208 iommu_map_fail:
 209         iommu_tbl_range_free(&iommu->tbl, *dma_addrp, npages, IOMMU_ERROR_CODE);
 210
 211 range_alloc_fail:
 212         free_pages(first_page, order);
 213         return NULL;
 214 }
 215
 216 static void dma_4v_iommu_demap(void *demap_arg, unsigned long entry,
 217                                unsigned long npages)
 218 {
 219         u32 devhandle = *(u32 *)demap_arg;
 220         unsigned long num, flags;
 221
 222         local_irq_save(flags);
 223         do {
 224                 num = pci_sun4v_iommu_demap(devhandle,
 225                                             HV_PCI_TSBID(0, entry),
 226                                             npages);
 227
 228                 entry += num;
 229                 npages -= num;
 230         } while (npages != 0);
 231         local_irq_restore(flags);
 232 }
 233
 234 static void dma_4v_free_coherent(struct device *dev, size_t size, void *cpu,
 235                                  dma_addr_t dvma, unsigned long attrs)
 236 {
 237         struct pci_pbm_info *pbm;
 238         struct iommu *iommu;
 239         unsigned long order, npages, entry;
 240         u32 devhandle;
 241
 242         npages = IO_PAGE_ALIGN(size) >> IO_PAGE_SHIFT;
 243         iommu = dev->archdata.iommu;
 244         pbm = dev->archdata.host_controller;
 245         devhandle = pbm->devhandle;
 246         entry = ((dvma - iommu->tbl.table_map_base) >> IO_PAGE_SHIFT);
 247         dma_4v_iommu_demap(&devhandle, entry, npages);
 248         iommu_tbl_range_free(&iommu->tbl, dvma, npages, IOMMU_ERROR_CODE);
 249         order = get_order(size);
 250         if (order < 10)
 251                 free_pages((unsigned long)cpu, order);
 252 }
 253
 254 static dma_addr_t dma_4v_map_page(struct device *dev, struct page *page,
 255                                   unsigned long offset, size_t sz,
 256                                   enum dma_data_direction direction,
 257                                   unsigned long attrs)
 258 {
 259         struct iommu *iommu;
 260         unsigned long flags, npages, oaddr;
 261         unsigned long i, base_paddr;
 262         u32 bus_addr, ret;
 263         unsigned long prot;
 264         long entry;
 265
 266         iommu = dev->archdata.iommu;
 267
 268         if (unlikely(direction == DMA_NONE))
 269                 goto bad;
 270
 271         oaddr = (unsigned long)(page_address(page) + offset);
 272         npages = IO_PAGE_ALIGN(oaddr + sz) - (oaddr & IO_PAGE_MASK);
 273         npages >>= IO_PAGE_SHIFT;
 274
 275         entry = iommu_tbl_range_alloc(dev, &iommu->tbl, npages, NULL,
 276                                       (unsigned long)(-1), 0);
 277
 278         if (unlikely(entry == IOMMU_ERROR_CODE))
 279                 goto bad;
 280
 281         bus_addr = (iommu->tbl.table_map_base + (entry << IO_PAGE_SHIFT));
 282         ret = bus_addr | (oaddr & ~IO_PAGE_MASK);
 283         base_paddr = __pa(oaddr & IO_PAGE_MASK);
 284         prot = HV_PCI_MAP_ATTR_READ;
 285         if (direction != DMA_TO_DEVICE)
 286                 prot |= HV_PCI_MAP_ATTR_WRITE;
 287
 288         if (attrs & DMA_ATTR_WEAK_ORDERING)
 289                 prot |= HV_PCI_MAP_ATTR_RELAXED_ORDER;
 290
 291         local_irq_save(flags);
 292
 293         iommu_batch_start(dev, prot, entry);
 294
 295         for (i = 0; i < npages; i++, base_paddr += IO_PAGE_SIZE) {
 296                 long err = iommu_batch_add(base_paddr);
 297                 if (unlikely(err < 0L))
 298                         goto iommu_map_fail;
 299         }
 300         if (unlikely(iommu_batch_end() < 0L))
 301                 goto iommu_map_fail;
 302
 303         local_irq_restore(flags);
 304
 305         return ret;
 306
 307 bad:
 308         if (printk_ratelimit())
 309                 WARN_ON(1);
 310         return DMA_ERROR_CODE;
 311
 312 iommu_map_fail:
 313         iommu_tbl_range_free(&iommu->tbl, bus_addr, npages, IOMMU_ERROR_CODE);
 314         return DMA_ERROR_CODE;
 315 }
 316
 317 static void dma_4v_unmap_page(struct device *dev, dma_addr_t bus_addr,
 318                               size_t sz, enum dma_data_direction direction,
 319                               unsigned long attrs)
 320 {
 321         struct pci_pbm_info *pbm;
 322         struct iommu *iommu;
 323         unsigned long npages;
 324         long entry;
 325         u32 devhandle;
 326
 327         if (unlikely(direction == DMA_NONE)) {
 328                 if (printk_ratelimit())
 329                         WARN_ON(1);
 330                 return;
 331         }
 332
 333         iommu = dev->archdata.iommu;
 334         pbm = dev->archdata.host_controller;
 335         devhandle = pbm->devhandle;
 336
 337         npages = IO_PAGE_ALIGN(bus_addr + sz) - (bus_addr & IO_PAGE_MASK);
 338         npages >>= IO_PAGE_SHIFT;
 339         bus_addr &= IO_PAGE_MASK;
 340         entry = (bus_addr - iommu->tbl.table_map_base) >> IO_PAGE_SHIFT;
 341         dma_4v_iommu_demap(&devhandle, entry, npages);
 342         iommu_tbl_range_free(&iommu->tbl, bus_addr, npages, IOMMU_ERROR_CODE);
 343 }
 344
 345 static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,
 346                          int nelems, enum dma_data_direction direction,
 347                          unsigned long attrs)
 348 {
 349         struct scatterlist *s, *outs, *segstart;
 350         unsigned long flags, handle, prot;
 351         dma_addr_t dma_next = 0, dma_addr;
 352         unsigned int max_seg_size;
 353         unsigned long seg_boundary_size;
 354         int outcount, incount, i;
 355         struct iommu *iommu;
 356         unsigned long base_shift;
 357         long err;
 358
 359         BUG_ON(direction == DMA_NONE);
 360
 361         iommu = dev->archdata.iommu;
 362         if (nelems == 0 || !iommu)
 363                 return 0;
 364
 365         prot = HV_PCI_MAP_ATTR_READ;
 366         if (direction != DMA_TO_DEVICE)
 367                 prot |= HV_PCI_MAP_ATTR_WRITE;
 368
 369         if (attrs & DMA_ATTR_WEAK_ORDERING)
 370                 prot |= HV_PCI_MAP_ATTR_RELAXED_ORDER;
 371
 372         outs = s = segstart = &sglist[0];
 373         outcount = 1;
 374         incount = nelems;
 375         handle = 0;
 376
 377         /* Init first segment length for backout at failure */
 378         outs->dma_length = 0;
 379
 380         local_irq_save(flags);
 381
 382         iommu_batch_start(dev, prot, ~0UL);
 383
 384         max_seg_size = dma_get_max_seg_size(dev);
 385         seg_boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
 386                                   IO_PAGE_SIZE) >> IO_PAGE_SHIFT;
 387         base_shift = iommu->tbl.table_map_base >> IO_PAGE_SHIFT;
 388         for_each_sg(sglist, s, nelems, i) {
 389                 unsigned long paddr, npages, entry, out_entry = 0, slen;
 390
 391                 slen = s->length;
 392                 /* Sanity check */
 393                 if (slen == 0) {
 394                         dma_next = 0;
 395                         continue;
 396                 }
 397                 /* Allocate iommu entries for that segment */
 398                 paddr = (unsigned long) SG_ENT_PHYS_ADDRESS(s);
 399                 npages = iommu_num_pages(paddr, slen, IO_PAGE_SIZE);
 400                 entry = iommu_tbl_range_alloc(dev, &iommu->tbl, npages,
 401                                               &handle, (unsigned long)(-1), 0);
 402
 403                 /* Handle failure */
 404                 if (unlikely(entry == IOMMU_ERROR_CODE)) {
 405                         if (printk_ratelimit())
 406                                 printk(KERN_INFO "iommu_alloc failed, iommu %p paddr %lx"
 407                                        " npages %lx\n", iommu, paddr, npages);
 408                         goto iommu_map_failed;
 409                 }
 410
 411                 iommu_batch_new_entry(entry);
 412
 413                 /* Convert entry to a dma_addr_t */
 414                 dma_addr = iommu->tbl.table_map_base + (entry << IO_PAGE_SHIFT);
 415                 dma_addr |= (s->offset & ~IO_PAGE_MASK);
 416
 417                 /* Insert into HW table */
 418                 paddr &= IO_PAGE_MASK;
 419                 while (npages--) {
 420                         err = iommu_batch_add(paddr);
 421                         if (unlikely(err < 0L))
 422                                 goto iommu_map_failed;
 423                         paddr += IO_PAGE_SIZE;
 424                 }
 425
 426                 /* If we are in an open segment, try merging */
 427                 if (segstart != s) {
 428                         /* We cannot merge if:
 429                          * - allocated dma_addr isn't contiguous to previous allocation
 430                          */
 431                         if ((dma_addr != dma_next) ||
 432                             (outs->dma_length + s->length > max_seg_size) ||
 433                             (is_span_boundary(out_entry, base_shift,
 434                                               seg_boundary_size, outs, s))) {
 435                                 /* Can't merge: create a new segment */
 436                                 segstart = s;
 437                                 outcount++;
 438                                 outs = sg_next(outs);
 439                         } else {
 440                                 outs->dma_length += s->length;
 441                         }
 442                 }
 443
 444                 if (segstart == s) {
 445                         /* This is a new segment, fill entries */
 446                         outs->dma_address = dma_addr;
 447                         outs->dma_length = slen;
 448                         out_entry = entry;
 449                 }
 450
 451                 /* Calculate next page pointer for contiguous check */
 452                 dma_next = dma_addr + slen;
 453         }
 454
 455         err = iommu_batch_end();
 456
 457         if (unlikely(err < 0L))
 458                 goto iommu_map_failed;
 459
 460         local_irq_restore(flags);
 461
 462         if (outcount < incount) {
 463                 outs = sg_next(outs);
 464                 outs->dma_address = DMA_ERROR_CODE;
 465                 outs->dma_length = 0;
 466         }
 467
 468         return outcount;
 469
 470 iommu_map_failed:
 471         for_each_sg(sglist, s, nelems, i) {
 472                 if (s->dma_length != 0) {
 473                         unsigned long vaddr, npages;
 474
 475                         vaddr = s->dma_address & IO_PAGE_MASK;
 476                         npages = iommu_num_pages(s->dma_address, s->dma_length,
 477                                                  IO_PAGE_SIZE);
 478                         iommu_tbl_range_free(&iommu->tbl, vaddr, npages,
 479                                              IOMMU_ERROR_CODE);
 480                         /* XXX demap? XXX */
 481                         s->dma_address = DMA_ERROR_CODE;
 482                         s->dma_length = 0;
 483                 }
 484                 if (s == outs)
 485                         break;
 486         }
 487         local_irq_restore(flags);
 488
 489         return 0;
 490 }
 491
 492 static void dma_4v_unmap_sg(struct device *dev, struct scatterlist *sglist,
 493                             int nelems, enum dma_data_direction direction,
 494                             unsigned long attrs)
 495 {
 496         struct pci_pbm_info *pbm;
 497         struct scatterlist *sg;
 498         struct iommu *iommu;
 499         unsigned long flags, entry;
 500         u32 devhandle;
 501
 502         BUG_ON(direction == DMA_NONE);
 503
 504         iommu = dev->archdata.iommu;
 505         pbm = dev->archdata.host_controller;
 506         devhandle = pbm->devhandle;
 507
 508         local_irq_save(flags);
 509
 510         sg = sglist;
 511         while (nelems--) {
 512                 dma_addr_t dma_handle = sg->dma_address;
 513                 unsigned int len = sg->dma_length;
 514                 unsigned long npages;
 515                 struct iommu_map_table *tbl = &iommu->tbl;
 516                 unsigned long shift = IO_PAGE_SHIFT;
 517
 518                 if (!len)
 519                         break;
 520                 npages = iommu_num_pages(dma_handle, len, IO_PAGE_SIZE);
 521                 entry = ((dma_handle - tbl->table_map_base) >> shift);
 522                 dma_4v_iommu_demap(&devhandle, entry, npages);
 523                 iommu_tbl_range_free(&iommu->tbl, dma_handle, npages,
 524                                      IOMMU_ERROR_CODE);
 525                 sg = sg_next(sg);
 526         }
 527
 528         local_irq_restore(flags);
 529 }
 530
 531 static struct dma_map_ops sun4v_dma_ops = {
 532         .alloc                          = dma_4v_alloc_coherent,
 533         .free                           = dma_4v_free_coherent,
 534         .map_page                       = dma_4v_map_page,
 535         .unmap_page                     = dma_4v_unmap_page,
 536         .map_sg                         = dma_4v_map_sg,
 537         .unmap_sg                       = dma_4v_unmap_sg,
 538 };
 539
 540 static void pci_sun4v_scan_bus(struct pci_pbm_info *pbm, struct device *parent)
 541 {
 542         struct property *prop;
 543         struct device_node *dp;
 544
 545         dp = pbm->op->dev.of_node;
 546         prop = of_find_property(dp, "66mhz-capable", NULL);
 547         pbm->is_66mhz_capable = (prop != NULL);
 548         pbm->pci_bus = pci_scan_one_pbm(pbm, parent);
 549
 550         /* XXX register error interrupt handlers XXX */
 551 }
 552
 553 static unsigned long probe_existing_entries(struct pci_pbm_info *pbm,
 554                                             struct iommu_map_table *iommu)
 555 {
 556         struct iommu_pool *pool;
 557         unsigned long i, pool_nr, cnt = 0;
 558         u32 devhandle;
 559
 560         devhandle = pbm->devhandle;
 561         for (pool_nr = 0; pool_nr < iommu->nr_pools; pool_nr++) {
 562                 pool = &(iommu->pools[pool_nr]);
 563                 for (i = pool->start; i <= pool->end; i++) {
 564                         unsigned long ret, io_attrs, ra;
 565
 566                         ret = pci_sun4v_iommu_getmap(devhandle,
 567                                                      HV_PCI_TSBID(0, i),
 568                                                      &io_attrs, &ra);
 569                         if (ret == HV_EOK) {
 570                                 if (page_in_phys_avail(ra)) {
 571                                         pci_sun4v_iommu_demap(devhandle,
 572                                                               HV_PCI_TSBID(0,
 573                                                               i), 1);
 574                                 } else {
 575                                         cnt++;
 576                                         __set_bit(i, iommu->map);
 577                                 }
 578                         }
 579                 }
 580         }
 581         return cnt;
 582 }
 583
 584 static int pci_sun4v_iommu_init(struct pci_pbm_info *pbm)
 585 {
 586         static const u32 vdma_default[] = { 0x80000000, 0x80000000 };
 587         struct iommu *iommu = pbm->iommu;
 588         unsigned long num_tsb_entries, sz;
 589         u32 dma_mask, dma_offset;
 590         const u32 *vdma;
 591
 592         vdma = of_get_property(pbm->op->dev.of_node, "virtual-dma", NULL);
 593         if (!vdma)
 594                 vdma = vdma_default;
 595
 596         if ((vdma[0] | vdma[1]) & ~IO_PAGE_MASK) {
 597                 printk(KERN_ERR PFX "Strange virtual-dma[%08x:%08x].\n",
 598                        vdma[0], vdma[1]);
 599                 return -EINVAL;
 600         }
 601
 602         dma_mask = (roundup_pow_of_two(vdma[1]) - 1UL);
 603         num_tsb_entries = vdma[1] / IO_PAGE_SIZE;
 604
 605         dma_offset = vdma[0];
 606
 607         /* Setup initial software IOMMU state. */
 608         spin_lock_init(&iommu->lock);
 609         iommu->ctx_lowest_free = 1;
 610         iommu->tbl.table_map_base = dma_offset;
 611         iommu->dma_addr_mask = dma_mask;
 612
 613         /* Allocate and initialize the free area map.  */
 614         sz = (num_tsb_entries + 7) / 8;
 615         sz = (sz + 7UL) & ~7UL;
 616         iommu->tbl.map = kzalloc(sz, GFP_KERNEL);
 617         if (!iommu->tbl.map) {
 618                 printk(KERN_ERR PFX "Error, kmalloc(arena.map) failed.\n");
 619                 return -ENOMEM;
 620         }
 621         iommu_tbl_pool_init(&iommu->tbl, num_tsb_entries, IO_PAGE_SHIFT,
 622                             NULL, false /* no large_pool */,
 623                             0 /* default npools */,
 624                             false /* want span boundary checking */);
 625         sz = probe_existing_entries(pbm, &iommu->tbl);
 626         if (sz)
 627                 printk("%s: Imported %lu TSB entries from OBP\n",
 628                        pbm->name, sz);
 629
 630         return 0;
 631 }
 632
 633 #ifdef CONFIG_PCI_MSI
 634 struct pci_sun4v_msiq_entry {
 635         u64             version_type;
 636 #define MSIQ_VERSION_MASK               0xffffffff00000000UL
 637 #define MSIQ_VERSION_SHIFT              32
 638 #define MSIQ_TYPE_MASK                  0x00000000000000ffUL
 639 #define MSIQ_TYPE_SHIFT                 0
 640 #define MSIQ_TYPE_NONE                  0x00
 641 #define MSIQ_TYPE_MSG                   0x01
 642 #define MSIQ_TYPE_MSI32                 0x02
 643 #define MSIQ_TYPE_MSI64                 0x03
 644 #define MSIQ_TYPE_INTX                  0x08
 645 #define MSIQ_TYPE_NONE2                 0xff
 646
 647         u64             intx_sysino;
 648         u64             reserved1;
 649         u64             stick;
 650         u64             req_id;  /* bus/device/func */
 651 #define MSIQ_REQID_BUS_MASK             0xff00UL
 652 #define MSIQ_REQID_BUS_SHIFT            8
 653 #define MSIQ_REQID_DEVICE_MASK          0x00f8UL
 654 #define MSIQ_REQID_DEVICE_SHIFT         3
 655 #define MSIQ_REQID_FUNC_MASK            0x0007UL
 656 #define MSIQ_REQID_FUNC_SHIFT           0
 657
 658         u64             msi_address;
 659
 660         /* The format of this value is message type dependent.
 661          * For MSI bits 15:0 are the data from the MSI packet.
 662          * For MSI-X bits 31:0 are the data from the MSI packet.
 663          * For MSG, the message code and message routing code where:
 664          *      bits 39:32 is the bus/device/fn of the msg target-id
 665          *      bits 18:16 is the message routing code
 666          *      bits 7:0 is the message code
 667          * For INTx the low order 2-bits are:
 668          *      00 - INTA
 669          *      01 - INTB
 670          *      10 - INTC
 671          *      11 - INTD
 672          */
 673         u64             msi_data;
 674
 675         u64             reserved2;
 676 };
 677
 678 static int pci_sun4v_get_head(struct pci_pbm_info *pbm, unsigned long msiqid,
 679                               unsigned long *head)
 680 {
 681         unsigned long err, limit;
 682
 683         err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, head);
 684         if (unlikely(err))
 685                 return -ENXIO;
 686
 687         limit = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
 688         if (unlikely(*head >= limit))
 689                 return -EFBIG;
 690
 691         return 0;
 692 }
 693
 694 static int pci_sun4v_dequeue_msi(struct pci_pbm_info *pbm,
 695                                  unsigned long msiqid, unsigned long *head,
 696                                  unsigned long *msi)
 697 {
 698         struct pci_sun4v_msiq_entry *ep;
 699         unsigned long err, type;
 700
 701         /* Note: void pointer arithmetic, 'head' is a byte offset  */
 702         ep = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
 703                                  (pbm->msiq_ent_count *
 704                                   sizeof(struct pci_sun4v_msiq_entry))) +
 705               *head);
 706
 707         if ((ep->version_type & MSIQ_TYPE_MASK) == 0)
 708                 return 0;
 709
 710         type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
 711         if (unlikely(type != MSIQ_TYPE_MSI32 &&
 712                      type != MSIQ_TYPE_MSI64))
 713                 return -EINVAL;
 714
 715         *msi = ep->msi_data;
 716
 717         err = pci_sun4v_msi_setstate(pbm->devhandle,
 718                                      ep->msi_data /* msi_num */,
 719                                      HV_MSISTATE_IDLE);
 720         if (unlikely(err))
 721                 return -ENXIO;
 722
 723         /* Clear the entry.  */
 724         ep->version_type &= ~MSIQ_TYPE_MASK;
 725
 726         (*head) += sizeof(struct pci_sun4v_msiq_entry);
 727         if (*head >=
 728             (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry)))
 729                 *head = 0;
 730
 731         return 1;
 732 }
 733
 734 static int pci_sun4v_set_head(struct pci_pbm_info *pbm, unsigned long msiqid,
 735                               unsigned long head)
 736 {
 737         unsigned long err;
 738
 739         err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
 740         if (unlikely(err))
 741                 return -EINVAL;
 742
 743         return 0;
 744 }
 745
 746 static int pci_sun4v_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid,
 747                                unsigned long msi, int is_msi64)
 748 {
 749         if (pci_sun4v_msi_setmsiq(pbm->devhandle, msi, msiqid,
 750                                   (is_msi64 ?
 751                                    HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
 752                 return -ENXIO;
 753         if (pci_sun4v_msi_setstate(pbm->devhandle, msi, HV_MSISTATE_IDLE))
 754                 return -ENXIO;
 755         if (pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_VALID))
 756                 return -ENXIO;
 757         return 0;
 758 }
 759
 760 static int pci_sun4v_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi)
 761 {
 762         unsigned long err, msiqid;
 763
 764         err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi, &msiqid);
 765         if (err)
 766                 return -ENXIO;
 767
 768         pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_INVALID);
 769
 770         return 0;
 771 }
 772
 773 static int pci_sun4v_msiq_alloc(struct pci_pbm_info *pbm)
 774 {
 775         unsigned long q_size, alloc_size, pages, order;
 776         int i;
 777
 778         q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
 779         alloc_size = (pbm->msiq_num * q_size);
 780         order = get_order(alloc_size);
 781         pages = __get_free_pages(GFP_KERNEL | __GFP_COMP, order);
 782         if (pages == 0UL) {
 783                 printk(KERN_ERR "MSI: Cannot allocate MSI queues (o=%lu).\n",
 784                        order);
 785                 return -ENOMEM;
 786         }
 787         memset((char *)pages, 0, PAGE_SIZE << order);
 788         pbm->msi_queues = (void *) pages;
 789
 790         for (i = 0; i < pbm->msiq_num; i++) {
 791                 unsigned long err, base = __pa(pages + (i * q_size));
 792                 unsigned long ret1, ret2;
 793
 794                 err = pci_sun4v_msiq_conf(pbm->devhandle,
 795                                           pbm->msiq_first + i,
 796                                           base, pbm->msiq_ent_count);
 797                 if (err) {
 798                         printk(KERN_ERR "MSI: msiq register fails (err=%lu)\n",
 799                                err);
 800                         goto h_error;
 801                 }
 802
 803                 err = pci_sun4v_msiq_info(pbm->devhandle,
 804                                           pbm->msiq_first + i,
 805                                           &ret1, &ret2);
 806                 if (err) {
 807                         printk(KERN_ERR "MSI: Cannot read msiq (err=%lu)\n",
 808                                err);
 809                         goto h_error;
 810                 }
 811                 if (ret1 != base || ret2 != pbm->msiq_ent_count) {
 812                         printk(KERN_ERR "MSI: Bogus qconf "
 813                                "expected[%lx:%x] got[%lx:%lx]\n",
 814                                base, pbm->msiq_ent_count,
 815                                ret1, ret2);
 816                         goto h_error;
 817                 }
 818         }
 819
 820         return 0;
 821
 822 h_error:
 823         free_pages(pages, order);
 824         return -EINVAL;
 825 }
 826
 827 static void pci_sun4v_msiq_free(struct pci_pbm_info *pbm)
 828 {
 829         unsigned long q_size, alloc_size, pages, order;
 830         int i;
 831
 832         for (i = 0; i < pbm->msiq_num; i++) {
 833                 unsigned long msiqid = pbm->msiq_first + i;
 834
 835                 (void) pci_sun4v_msiq_conf(pbm->devhandle, msiqid, 0UL, 0);
 836         }
 837
 838         q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
 839         alloc_size = (pbm->msiq_num * q_size);
 840         order = get_order(alloc_size);
 841
 842         pages = (unsigned long) pbm->msi_queues;
 843
 844         free_pages(pages, order);
 845
 846         pbm->msi_queues = NULL;
 847 }
 848
 849 static int pci_sun4v_msiq_build_irq(struct pci_pbm_info *pbm,
 850                                     unsigned long msiqid,
 851                                     unsigned long devino)
 852 {
 853         unsigned int irq = sun4v_build_irq(pbm->devhandle, devino);
 854
 855         if (!irq)
 856                 return -ENOMEM;
 857
 858         if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
 859                 return -EINVAL;
 860         if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
 861                 return -EINVAL;
 862
 863         return irq;
 864 }
 865
 866 static const struct sparc64_msiq_ops pci_sun4v_msiq_ops = {
 867         .get_head       =       pci_sun4v_get_head,
 868         .dequeue_msi    =       pci_sun4v_dequeue_msi,
 869         .set_head       =       pci_sun4v_set_head,
 870         .msi_setup      =       pci_sun4v_msi_setup,
 871         .msi_teardown   =       pci_sun4v_msi_teardown,
 872         .msiq_alloc     =       pci_sun4v_msiq_alloc,
 873         .msiq_free      =       pci_sun4v_msiq_free,
 874         .msiq_build_irq =       pci_sun4v_msiq_build_irq,
 875 };
 876
 877 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
 878 {
 879         sparc64_pbm_msi_init(pbm, &pci_sun4v_msiq_ops);
 880 }
 881 #else /* CONFIG_PCI_MSI */
 882 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
 883 {
 884 }
 885 #endif /* !(CONFIG_PCI_MSI) */
 886
 887 static int pci_sun4v_pbm_init(struct pci_pbm_info *pbm,
 888                               struct platform_device *op, u32 devhandle)
 889 {
 890         struct device_node *dp = op->dev.of_node;
 891         int err;
 892
 893         pbm->numa_node = of_node_to_nid(dp);
 894
 895         pbm->pci_ops = &sun4v_pci_ops;
 896         pbm->config_space_reg_bits = 12;
 897
 898         pbm->index = pci_num_pbms++;
 899
 900         pbm->op = op;
 901
 902         pbm->devhandle = devhandle;
 903
 904         pbm->name = dp->full_name;
 905
 906         printk("%s: SUN4V PCI Bus Module\n", pbm->name);
 907         printk("%s: On NUMA node %d\n", pbm->name, pbm->numa_node);
 908
 909         pci_determine_mem_io_space(pbm);
 910
 911         pci_get_pbm_props(pbm);
 912
 913         err = pci_sun4v_iommu_init(pbm);
 914         if (err)
 915                 return err;
 916
 917         pci_sun4v_msi_init(pbm);
 918
 919         pci_sun4v_scan_bus(pbm, &op->dev);
 920
 921         pbm->next = pci_pbm_root;
 922         pci_pbm_root = pbm;
 923
 924         return 0;
 925 }
 926
 927 static int pci_sun4v_probe(struct platform_device *op)
 928 {
 929         const struct linux_prom64_registers *regs;
 930         static int hvapi_negotiated = 0;
 931         struct pci_pbm_info *pbm;
 932         struct device_node *dp;
 933         struct iommu *iommu;
 934         u32 devhandle;
 935         int i, err = -ENODEV;
 936
 937         dp = op->dev.of_node;
 938
 939         if (!hvapi_negotiated++) {
 940                 for (i = 0; i < ARRAY_SIZE(vpci_versions); i++) {
 941                         vpci_major = vpci_versions[i].major;
 942                         vpci_minor = vpci_versions[i].minor;
 943
 944                         err = sun4v_hvapi_register(HV_GRP_PCI, vpci_major,
 945                                                    &vpci_minor);
 946                         if (!err)
 947                                 break;
 948                 }
 949
 950                 if (err) {
 951                         pr_err(PFX "Could not register hvapi, err=%d\n", err);
 952                         return err;
 953                 }
 954                 pr_info(PFX "Registered hvapi major[%lu] minor[%lu]\n",
 955                         vpci_major, vpci_minor);
 956
 957                 dma_ops = &sun4v_dma_ops;
 958         }
 959
 960         regs = of_get_property(dp, "reg", NULL);
 961         err = -ENODEV;
 962         if (!regs) {
 963                 printk(KERN_ERR PFX "Could not find config registers\n");
 964                 goto out_err;
 965         }
 966         devhandle = (regs->phys_addr >> 32UL) & 0x0fffffff;
 967
 968         err = -ENOMEM;
 969         if (!iommu_batch_initialized) {
 970                 for_each_possible_cpu(i) {
 971                         unsigned long page = get_zeroed_page(GFP_KERNEL);
 972
 973                         if (!page)
 974                                 goto out_err;
 975
 976                         per_cpu(iommu_batch, i).pglist = (u64 *) page;
 977                 }
 978                 iommu_batch_initialized = 1;
 979         }
 980
 981         pbm = kzalloc(sizeof(*pbm), GFP_KERNEL);
 982         if (!pbm) {
 983                 printk(KERN_ERR PFX "Could not allocate pci_pbm_info\n");
 984                 goto out_err;
 985         }
 986
 987         iommu = kzalloc(sizeof(struct iommu), GFP_KERNEL);
 988         if (!iommu) {
 989                 printk(KERN_ERR PFX "Could not allocate pbm iommu\n");
 990                 goto out_free_controller;
 991         }
 992
 993         pbm->iommu = iommu;
 994
 995         err = pci_sun4v_pbm_init(pbm, op, devhandle);
 996         if (err)
 997                 goto out_free_iommu;
 998
 999         dev_set_drvdata(&op->dev, pbm);
1000
1001         return 0;
1002
1003 out_free_iommu:
1004         kfree(pbm->iommu);
1005
1006 out_free_controller:
1007         kfree(pbm);
1008
1009 out_err:
1010         return err;
1011 }
1012
1013 static const struct of_device_id pci_sun4v_match[] = {
1014         {
1015                 .name = "pci",
1016                 .compatible = "SUNW,sun4v-pci",
1017         },
1018         {},
1019 };
1020
1021 static struct platform_driver pci_sun4v_driver = {
1022         .driver = {
1023                 .name = DRIVER_NAME,
1024                 .of_match_table = pci_sun4v_match,
1025         },
1026         .probe          = pci_sun4v_probe,
1027 };
1028
1029 static int __init pci_sun4v_init(void)
1030 {
1031         return platform_driver_register(&pci_sun4v_driver);
1032 }
1033
1034 subsys_initcall(pci_sun4v_init);