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 = iommu->dma_addr_mask; 677 678 if (device_mask > DMA_BIT_MASK(32)) { 679 if (iommu->atu) 680 dma_addr_mask = iommu->atu->dma_addr_mask; 681 else 682 return 0; 683 } 684 685 if ((device_mask & dma_addr_mask) == dma_addr_mask) 686 return 1; 687 return pci64_dma_supported(to_pci_dev(dev), device_mask); 688 } 689 690 static int dma_4v_mapping_error(struct device *dev, dma_addr_t dma_addr) 691 { 692 return dma_addr == SPARC_MAPPING_ERROR; 693 } 694 695 static const struct dma_map_ops sun4v_dma_ops = { 696 .alloc = dma_4v_alloc_coherent, 697 .free = dma_4v_free_coherent, 698 .map_page = dma_4v_map_page, 699 .unmap_page = dma_4v_unmap_page, 700 .map_sg = dma_4v_map_sg, 701 .unmap_sg = dma_4v_unmap_sg, 702 .dma_supported = dma_4v_supported, 703 .mapping_error = dma_4v_mapping_error, 704 }; 705 706 static void pci_sun4v_scan_bus(struct pci_pbm_info *pbm, struct device *parent) 707 { 708 struct property *prop; 709 struct device_node *dp; 710 711 dp = pbm->op->dev.of_node; 712 prop = of_find_property(dp, "66mhz-capable", NULL); 713 pbm->is_66mhz_capable = (prop != NULL); 714 pbm->pci_bus = pci_scan_one_pbm(pbm, parent); 715 716 /* XXX register error interrupt handlers XXX */ 717 } 718 719 static unsigned long probe_existing_entries(struct pci_pbm_info *pbm, 720 struct iommu_map_table *iommu) 721 { 722 struct iommu_pool *pool; 723 unsigned long i, pool_nr, cnt = 0; 724 u32 devhandle; 725 726 devhandle = pbm->devhandle; 727 for (pool_nr = 0; pool_nr < iommu->nr_pools; pool_nr++) { 728 pool = &(iommu->pools[pool_nr]); 729 for (i = pool->start; i <= pool->end; i++) { 730 unsigned long ret, io_attrs, ra; 731 732 ret = pci_sun4v_iommu_getmap(devhandle, 733 HV_PCI_TSBID(0, i), 734 &io_attrs, &ra); 735 if (ret == HV_EOK) { 736 if (page_in_phys_avail(ra)) { 737 pci_sun4v_iommu_demap(devhandle, 738 HV_PCI_TSBID(0, 739 i), 1); 740 } else { 741 cnt++; 742 __set_bit(i, iommu->map); 743 } 744 } 745 } 746 } 747 return cnt; 748 } 749 750 static int pci_sun4v_atu_alloc_iotsb(struct pci_pbm_info *pbm) 751 { 752 struct atu *atu = pbm->iommu->atu; 753 struct atu_iotsb *iotsb; 754 void *table; 755 u64 table_size; 756 u64 iotsb_num; 757 unsigned long order; 758 unsigned long err; 759 760 iotsb = kzalloc(sizeof(*iotsb), GFP_KERNEL); 761 if (!iotsb) { 762 err = -ENOMEM; 763 goto out_err; 764 } 765 atu->iotsb = iotsb; 766 767 /* calculate size of IOTSB */ 768 table_size = (atu->size / IO_PAGE_SIZE) * 8; 769 order = get_order(table_size); 770 table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order); 771 if (!table) { 772 err = -ENOMEM; 773 goto table_failed; 774 } 775 iotsb->table = table; 776 iotsb->ra = __pa(table); 777 iotsb->dvma_size = atu->size; 778 iotsb->dvma_base = atu->base; 779 iotsb->table_size = table_size; 780 iotsb->page_size = IO_PAGE_SIZE; 781 782 /* configure and register IOTSB with HV */ 783 err = pci_sun4v_iotsb_conf(pbm->devhandle, 784 iotsb->ra, 785 iotsb->table_size, 786 iotsb->page_size, 787 iotsb->dvma_base, 788 &iotsb_num); 789 if (err) { 790 pr_err(PFX "pci_iotsb_conf failed error: %ld\n", err); 791 goto iotsb_conf_failed; 792 } 793 iotsb->iotsb_num = iotsb_num; 794 795 err = dma_4v_iotsb_bind(pbm->devhandle, iotsb_num, pbm->pci_bus); 796 if (err) { 797 pr_err(PFX "pci_iotsb_bind failed error: %ld\n", err); 798 goto iotsb_conf_failed; 799 } 800 801 return 0; 802 803 iotsb_conf_failed: 804 free_pages((unsigned long)table, order); 805 table_failed: 806 kfree(iotsb); 807 out_err: 808 return err; 809 } 810 811 static int pci_sun4v_atu_init(struct pci_pbm_info *pbm) 812 { 813 struct atu *atu = pbm->iommu->atu; 814 unsigned long err; 815 const u64 *ranges; 816 u64 map_size, num_iotte; 817 u64 dma_mask; 818 const u32 *page_size; 819 int len; 820 821 ranges = of_get_property(pbm->op->dev.of_node, "iommu-address-ranges", 822 &len); 823 if (!ranges) { 824 pr_err(PFX "No iommu-address-ranges\n"); 825 return -EINVAL; 826 } 827 828 page_size = of_get_property(pbm->op->dev.of_node, "iommu-pagesizes", 829 NULL); 830 if (!page_size) { 831 pr_err(PFX "No iommu-pagesizes\n"); 832 return -EINVAL; 833 } 834 835 /* There are 4 iommu-address-ranges supported. Each range is pair of 836 * {base, size}. The ranges[0] and ranges[1] are 32bit address space 837 * while ranges[2] and ranges[3] are 64bit space. We want to use 64bit 838 * address ranges to support 64bit addressing. Because 'size' for 839 * address ranges[2] and ranges[3] are same we can select either of 840 * ranges[2] or ranges[3] for mapping. However due to 'size' is too 841 * large for OS to allocate IOTSB we are using fix size 32G 842 * (ATU_64_SPACE_SIZE) which is more than enough for all PCIe devices 843 * to share. 844 */ 845 atu->ranges = (struct atu_ranges *)ranges; 846 atu->base = atu->ranges[3].base; 847 atu->size = ATU_64_SPACE_SIZE; 848 849 /* Create IOTSB */ 850 err = pci_sun4v_atu_alloc_iotsb(pbm); 851 if (err) { 852 pr_err(PFX "Error creating ATU IOTSB\n"); 853 return err; 854 } 855 856 /* Create ATU iommu map. 857 * One bit represents one iotte in IOTSB table. 858 */ 859 dma_mask = (roundup_pow_of_two(atu->size) - 1UL); 860 num_iotte = atu->size / IO_PAGE_SIZE; 861 map_size = num_iotte / 8; 862 atu->tbl.table_map_base = atu->base; 863 atu->dma_addr_mask = dma_mask; 864 atu->tbl.map = kzalloc(map_size, GFP_KERNEL); 865 if (!atu->tbl.map) 866 return -ENOMEM; 867 868 iommu_tbl_pool_init(&atu->tbl, num_iotte, IO_PAGE_SHIFT, 869 NULL, false /* no large_pool */, 870 0 /* default npools */, 871 false /* want span boundary checking */); 872 873 return 0; 874 } 875 876 static int pci_sun4v_iommu_init(struct pci_pbm_info *pbm) 877 { 878 static const u32 vdma_default[] = { 0x80000000, 0x80000000 }; 879 struct iommu *iommu = pbm->iommu; 880 unsigned long num_tsb_entries, sz; 881 u32 dma_mask, dma_offset; 882 const u32 *vdma; 883 884 vdma = of_get_property(pbm->op->dev.of_node, "virtual-dma", NULL); 885 if (!vdma) 886 vdma = vdma_default; 887 888 if ((vdma[0] | vdma[1]) & ~IO_PAGE_MASK) { 889 printk(KERN_ERR PFX "Strange virtual-dma[%08x:%08x].\n", 890 vdma[0], vdma[1]); 891 return -EINVAL; 892 } 893 894 dma_mask = (roundup_pow_of_two(vdma[1]) - 1UL); 895 num_tsb_entries = vdma[1] / IO_PAGE_SIZE; 896 897 dma_offset = vdma[0]; 898 899 /* Setup initial software IOMMU state. */ 900 spin_lock_init(&iommu->lock); 901 iommu->ctx_lowest_free = 1; 902 iommu->tbl.table_map_base = dma_offset; 903 iommu->dma_addr_mask = dma_mask; 904 905 /* Allocate and initialize the free area map. */ 906 sz = (num_tsb_entries + 7) / 8; 907 sz = (sz + 7UL) & ~7UL; 908 iommu->tbl.map = kzalloc(sz, GFP_KERNEL); 909 if (!iommu->tbl.map) { 910 printk(KERN_ERR PFX "Error, kmalloc(arena.map) failed.\n"); 911 return -ENOMEM; 912 } 913 iommu_tbl_pool_init(&iommu->tbl, num_tsb_entries, IO_PAGE_SHIFT, 914 NULL, false /* no large_pool */, 915 0 /* default npools */, 916 false /* want span boundary checking */); 917 sz = probe_existing_entries(pbm, &iommu->tbl); 918 if (sz) 919 printk("%s: Imported %lu TSB entries from OBP\n", 920 pbm->name, sz); 921 922 return 0; 923 } 924 925 #ifdef CONFIG_PCI_MSI 926 struct pci_sun4v_msiq_entry { 927 u64 version_type; 928 #define MSIQ_VERSION_MASK 0xffffffff00000000UL 929 #define MSIQ_VERSION_SHIFT 32 930 #define MSIQ_TYPE_MASK 0x00000000000000ffUL 931 #define MSIQ_TYPE_SHIFT 0 932 #define MSIQ_TYPE_NONE 0x00 933 #define MSIQ_TYPE_MSG 0x01 934 #define MSIQ_TYPE_MSI32 0x02 935 #define MSIQ_TYPE_MSI64 0x03 936 #define MSIQ_TYPE_INTX 0x08 937 #define MSIQ_TYPE_NONE2 0xff 938 939 u64 intx_sysino; 940 u64 reserved1; 941 u64 stick; 942 u64 req_id; /* bus/device/func */ 943 #define MSIQ_REQID_BUS_MASK 0xff00UL 944 #define MSIQ_REQID_BUS_SHIFT 8 945 #define MSIQ_REQID_DEVICE_MASK 0x00f8UL 946 #define MSIQ_REQID_DEVICE_SHIFT 3 947 #define MSIQ_REQID_FUNC_MASK 0x0007UL 948 #define MSIQ_REQID_FUNC_SHIFT 0 949 950 u64 msi_address; 951 952 /* The format of this value is message type dependent. 953 * For MSI bits 15:0 are the data from the MSI packet. 954 * For MSI-X bits 31:0 are the data from the MSI packet. 955 * For MSG, the message code and message routing code where: 956 * bits 39:32 is the bus/device/fn of the msg target-id 957 * bits 18:16 is the message routing code 958 * bits 7:0 is the message code 959 * For INTx the low order 2-bits are: 960 * 00 - INTA 961 * 01 - INTB 962 * 10 - INTC 963 * 11 - INTD 964 */ 965 u64 msi_data; 966 967 u64 reserved2; 968 }; 969 970 static int pci_sun4v_get_head(struct pci_pbm_info *pbm, unsigned long msiqid, 971 unsigned long *head) 972 { 973 unsigned long err, limit; 974 975 err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, head); 976 if (unlikely(err)) 977 return -ENXIO; 978 979 limit = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry); 980 if (unlikely(*head >= limit)) 981 return -EFBIG; 982 983 return 0; 984 } 985 986 static int pci_sun4v_dequeue_msi(struct pci_pbm_info *pbm, 987 unsigned long msiqid, unsigned long *head, 988 unsigned long *msi) 989 { 990 struct pci_sun4v_msiq_entry *ep; 991 unsigned long err, type; 992 993 /* Note: void pointer arithmetic, 'head' is a byte offset */ 994 ep = (pbm->msi_queues + ((msiqid - pbm->msiq_first) * 995 (pbm->msiq_ent_count * 996 sizeof(struct pci_sun4v_msiq_entry))) + 997 *head); 998 999 if ((ep->version_type & MSIQ_TYPE_MASK) == 0) 1000 return 0; 1001 1002 type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT; 1003 if (unlikely(type != MSIQ_TYPE_MSI32 && 1004 type != MSIQ_TYPE_MSI64)) 1005 return -EINVAL; 1006 1007 *msi = ep->msi_data; 1008 1009 err = pci_sun4v_msi_setstate(pbm->devhandle, 1010 ep->msi_data /* msi_num */, 1011 HV_MSISTATE_IDLE); 1012 if (unlikely(err)) 1013 return -ENXIO; 1014 1015 /* Clear the entry. */ 1016 ep->version_type &= ~MSIQ_TYPE_MASK; 1017 1018 (*head) += sizeof(struct pci_sun4v_msiq_entry); 1019 if (*head >= 1020 (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry))) 1021 *head = 0; 1022 1023 return 1; 1024 } 1025 1026 static int pci_sun4v_set_head(struct pci_pbm_info *pbm, unsigned long msiqid, 1027 unsigned long head) 1028 { 1029 unsigned long err; 1030 1031 err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head); 1032 if (unlikely(err)) 1033 return -EINVAL; 1034 1035 return 0; 1036 } 1037 1038 static int pci_sun4v_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid, 1039 unsigned long msi, int is_msi64) 1040 { 1041 if (pci_sun4v_msi_setmsiq(pbm->devhandle, msi, msiqid, 1042 (is_msi64 ? 1043 HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32))) 1044 return -ENXIO; 1045 if (pci_sun4v_msi_setstate(pbm->devhandle, msi, HV_MSISTATE_IDLE)) 1046 return -ENXIO; 1047 if (pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_VALID)) 1048 return -ENXIO; 1049 return 0; 1050 } 1051 1052 static int pci_sun4v_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi) 1053 { 1054 unsigned long err, msiqid; 1055 1056 err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi, &msiqid); 1057 if (err) 1058 return -ENXIO; 1059 1060 pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_INVALID); 1061 1062 return 0; 1063 } 1064 1065 static int pci_sun4v_msiq_alloc(struct pci_pbm_info *pbm) 1066 { 1067 unsigned long q_size, alloc_size, pages, order; 1068 int i; 1069 1070 q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry); 1071 alloc_size = (pbm->msiq_num * q_size); 1072 order = get_order(alloc_size); 1073 pages = __get_free_pages(GFP_KERNEL | __GFP_COMP, order); 1074 if (pages == 0UL) { 1075 printk(KERN_ERR "MSI: Cannot allocate MSI queues (o=%lu).\n", 1076 order); 1077 return -ENOMEM; 1078 } 1079 memset((char *)pages, 0, PAGE_SIZE << order); 1080 pbm->msi_queues = (void *) pages; 1081 1082 for (i = 0; i < pbm->msiq_num; i++) { 1083 unsigned long err, base = __pa(pages + (i * q_size)); 1084 unsigned long ret1, ret2; 1085 1086 err = pci_sun4v_msiq_conf(pbm->devhandle, 1087 pbm->msiq_first + i, 1088 base, pbm->msiq_ent_count); 1089 if (err) { 1090 printk(KERN_ERR "MSI: msiq register fails (err=%lu)\n", 1091 err); 1092 goto h_error; 1093 } 1094 1095 err = pci_sun4v_msiq_info(pbm->devhandle, 1096 pbm->msiq_first + i, 1097 &ret1, &ret2); 1098 if (err) { 1099 printk(KERN_ERR "MSI: Cannot read msiq (err=%lu)\n", 1100 err); 1101 goto h_error; 1102 } 1103 if (ret1 != base || ret2 != pbm->msiq_ent_count) { 1104 printk(KERN_ERR "MSI: Bogus qconf " 1105 "expected[%lx:%x] got[%lx:%lx]\n", 1106 base, pbm->msiq_ent_count, 1107 ret1, ret2); 1108 goto h_error; 1109 } 1110 } 1111 1112 return 0; 1113 1114 h_error: 1115 free_pages(pages, order); 1116 return -EINVAL; 1117 } 1118 1119 static void pci_sun4v_msiq_free(struct pci_pbm_info *pbm) 1120 { 1121 unsigned long q_size, alloc_size, pages, order; 1122 int i; 1123 1124 for (i = 0; i < pbm->msiq_num; i++) { 1125 unsigned long msiqid = pbm->msiq_first + i; 1126 1127 (void) pci_sun4v_msiq_conf(pbm->devhandle, msiqid, 0UL, 0); 1128 } 1129 1130 q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry); 1131 alloc_size = (pbm->msiq_num * q_size); 1132 order = get_order(alloc_size); 1133 1134 pages = (unsigned long) pbm->msi_queues; 1135 1136 free_pages(pages, order); 1137 1138 pbm->msi_queues = NULL; 1139 } 1140 1141 static int pci_sun4v_msiq_build_irq(struct pci_pbm_info *pbm, 1142 unsigned long msiqid, 1143 unsigned long devino) 1144 { 1145 unsigned int irq = sun4v_build_irq(pbm->devhandle, devino); 1146 1147 if (!irq) 1148 return -ENOMEM; 1149 1150 if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID)) 1151 return -EINVAL; 1152 if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE)) 1153 return -EINVAL; 1154 1155 return irq; 1156 } 1157 1158 static const struct sparc64_msiq_ops pci_sun4v_msiq_ops = { 1159 .get_head = pci_sun4v_get_head, 1160 .dequeue_msi = pci_sun4v_dequeue_msi, 1161 .set_head = pci_sun4v_set_head, 1162 .msi_setup = pci_sun4v_msi_setup, 1163 .msi_teardown = pci_sun4v_msi_teardown, 1164 .msiq_alloc = pci_sun4v_msiq_alloc, 1165 .msiq_free = pci_sun4v_msiq_free, 1166 .msiq_build_irq = pci_sun4v_msiq_build_irq, 1167 }; 1168 1169 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm) 1170 { 1171 sparc64_pbm_msi_init(pbm, &pci_sun4v_msiq_ops); 1172 } 1173 #else /* CONFIG_PCI_MSI */ 1174 static void pci_sun4v_msi_init(struct pci_pbm_info *pbm) 1175 { 1176 } 1177 #endif /* !(CONFIG_PCI_MSI) */ 1178 1179 static int pci_sun4v_pbm_init(struct pci_pbm_info *pbm, 1180 struct platform_device *op, u32 devhandle) 1181 { 1182 struct device_node *dp = op->dev.of_node; 1183 int err; 1184 1185 pbm->numa_node = of_node_to_nid(dp); 1186 1187 pbm->pci_ops = &sun4v_pci_ops; 1188 pbm->config_space_reg_bits = 12; 1189 1190 pbm->index = pci_num_pbms++; 1191 1192 pbm->op = op; 1193 1194 pbm->devhandle = devhandle; 1195 1196 pbm->name = dp->full_name; 1197 1198 printk("%s: SUN4V PCI Bus Module\n", pbm->name); 1199 printk("%s: On NUMA node %d\n", pbm->name, pbm->numa_node); 1200 1201 pci_determine_mem_io_space(pbm); 1202 1203 pci_get_pbm_props(pbm); 1204 1205 err = pci_sun4v_iommu_init(pbm); 1206 if (err) 1207 return err; 1208 1209 pci_sun4v_msi_init(pbm); 1210 1211 pci_sun4v_scan_bus(pbm, &op->dev); 1212 1213 /* if atu_init fails its not complete failure. 1214 * we can still continue using legacy iommu. 1215 */ 1216 if (pbm->iommu->atu) { 1217 err = pci_sun4v_atu_init(pbm); 1218 if (err) { 1219 kfree(pbm->iommu->atu); 1220 pbm->iommu->atu = NULL; 1221 pr_err(PFX "ATU init failed, err=%d\n", err); 1222 } 1223 } 1224 1225 pbm->next = pci_pbm_root; 1226 pci_pbm_root = pbm; 1227 1228 return 0; 1229 } 1230 1231 static int pci_sun4v_probe(struct platform_device *op) 1232 { 1233 const struct linux_prom64_registers *regs; 1234 static int hvapi_negotiated = 0; 1235 struct pci_pbm_info *pbm; 1236 struct device_node *dp; 1237 struct iommu *iommu; 1238 struct atu *atu; 1239 u32 devhandle; 1240 int i, err = -ENODEV; 1241 static bool hv_atu = true; 1242 1243 dp = op->dev.of_node; 1244 1245 if (!hvapi_negotiated++) { 1246 for (i = 0; i < ARRAY_SIZE(vpci_versions); i++) { 1247 vpci_major = vpci_versions[i].major; 1248 vpci_minor = vpci_versions[i].minor; 1249 1250 err = sun4v_hvapi_register(HV_GRP_PCI, vpci_major, 1251 &vpci_minor); 1252 if (!err) 1253 break; 1254 } 1255 1256 if (err) { 1257 pr_err(PFX "Could not register hvapi, err=%d\n", err); 1258 return err; 1259 } 1260 pr_info(PFX "Registered hvapi major[%lu] minor[%lu]\n", 1261 vpci_major, vpci_minor); 1262 1263 err = sun4v_hvapi_register(HV_GRP_ATU, vatu_major, &vatu_minor); 1264 if (err) { 1265 /* don't return an error if we fail to register the 1266 * ATU group, but ATU hcalls won't be available. 1267 */ 1268 hv_atu = false; 1269 pr_err(PFX "Could not register hvapi ATU err=%d\n", 1270 err); 1271 } else { 1272 pr_info(PFX "Registered hvapi ATU major[%lu] minor[%lu]\n", 1273 vatu_major, vatu_minor); 1274 } 1275 1276 dma_ops = &sun4v_dma_ops; 1277 } 1278 1279 regs = of_get_property(dp, "reg", NULL); 1280 err = -ENODEV; 1281 if (!regs) { 1282 printk(KERN_ERR PFX "Could not find config registers\n"); 1283 goto out_err; 1284 } 1285 devhandle = (regs->phys_addr >> 32UL) & 0x0fffffff; 1286 1287 err = -ENOMEM; 1288 if (!iommu_batch_initialized) { 1289 for_each_possible_cpu(i) { 1290 unsigned long page = get_zeroed_page(GFP_KERNEL); 1291 1292 if (!page) 1293 goto out_err; 1294 1295 per_cpu(iommu_batch, i).pglist = (u64 *) page; 1296 } 1297 iommu_batch_initialized = 1; 1298 } 1299 1300 pbm = kzalloc(sizeof(*pbm), GFP_KERNEL); 1301 if (!pbm) { 1302 printk(KERN_ERR PFX "Could not allocate pci_pbm_info\n"); 1303 goto out_err; 1304 } 1305 1306 iommu = kzalloc(sizeof(struct iommu), GFP_KERNEL); 1307 if (!iommu) { 1308 printk(KERN_ERR PFX "Could not allocate pbm iommu\n"); 1309 goto out_free_controller; 1310 } 1311 1312 pbm->iommu = iommu; 1313 iommu->atu = NULL; 1314 if (hv_atu) { 1315 atu = kzalloc(sizeof(*atu), GFP_KERNEL); 1316 if (!atu) 1317 pr_err(PFX "Could not allocate atu\n"); 1318 else 1319 iommu->atu = atu; 1320 } 1321 1322 err = pci_sun4v_pbm_init(pbm, op, devhandle); 1323 if (err) 1324 goto out_free_iommu; 1325 1326 dev_set_drvdata(&op->dev, pbm); 1327 1328 return 0; 1329 1330 out_free_iommu: 1331 kfree(iommu->atu); 1332 kfree(pbm->iommu); 1333 1334 out_free_controller: 1335 kfree(pbm); 1336 1337 out_err: 1338 return err; 1339 } 1340 1341 static const struct of_device_id pci_sun4v_match[] = { 1342 { 1343 .name = "pci", 1344 .compatible = "SUNW,sun4v-pci", 1345 }, 1346 {}, 1347 }; 1348 1349 static struct platform_driver pci_sun4v_driver = { 1350 .driver = { 1351 .name = DRIVER_NAME, 1352 .of_match_table = pci_sun4v_match, 1353 }, 1354 .probe = pci_sun4v_probe, 1355 }; 1356 1357 static int __init pci_sun4v_init(void) 1358 { 1359 return platform_driver_register(&pci_sun4v_driver); 1360 } 1361 1362 subsys_initcall(pci_sun4v_init); 1363