Rev 5367 | Rev 6296 | Go to most recent revision | Show entire file | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 5367 | Rev 6084 | ||
---|---|---|---|
Line 250... | Line 250... | ||
250 | 250 | ||
251 | static inline int use_cpu_reloc(struct drm_i915_gem_object *obj) |
251 | static inline int use_cpu_reloc(struct drm_i915_gem_object *obj) |
252 | { |
252 | { |
253 | return (HAS_LLC(obj->base.dev) || |
253 | return (HAS_LLC(obj->base.dev) || |
254 | obj->base.write_domain == I915_GEM_DOMAIN_CPU || |
- | |
255 | !obj->map_and_fenceable || |
254 | obj->base.write_domain == I915_GEM_DOMAIN_CPU || |
256 | obj->cache_level != I915_CACHE_NONE); |
255 | obj->cache_level != I915_CACHE_NONE); |
Line 257... | Line 256... | ||
257 | } |
256 | } |
258 | 257 | ||
Line 318... | Line 317... | ||
318 | MapPage(dev_priv->gtt.mappable,dev_priv->gtt.mappable_base + |
317 | MapPage(dev_priv->gtt.mappable,dev_priv->gtt.mappable_base + |
319 | (offset & PAGE_MASK), PG_SW); |
318 | (offset & PAGE_MASK), PG_SW); |
320 | reloc_page = dev_priv->gtt.mappable; |
319 | reloc_page = dev_priv->gtt.mappable; |
321 | iowrite32(lower_32_bits(delta), reloc_page + offset_in_page(offset)); |
320 | iowrite32(lower_32_bits(delta), reloc_page + offset_in_page(offset)); |
Line -... | Line 321... | ||
- | 321 | ||
Line -... | Line 322... | ||
- | 322 | // io_mapping_unmap_atomic(reloc_page); |
|
- | 323 | ||
- | 324 | return 0; |
|
- | 325 | } |
|
- | 326 | ||
- | 327 | static void |
|
- | 328 | clflush_write32(void *addr, uint32_t value) |
|
- | 329 | { |
|
- | 330 | /* This is not a fast path, so KISS. */ |
|
- | 331 | drm_clflush_virt_range(addr, sizeof(uint32_t)); |
|
- | 332 | *(uint32_t *)addr = value; |
|
- | 333 | drm_clflush_virt_range(addr, sizeof(uint32_t)); |
|
- | 334 | } |
|
- | 335 | ||
- | 336 | static int |
|
- | 337 | relocate_entry_clflush(struct drm_i915_gem_object *obj, |
|
- | 338 | struct drm_i915_gem_relocation_entry *reloc, |
|
- | 339 | uint64_t target_offset) |
|
- | 340 | { |
|
- | 341 | struct drm_device *dev = obj->base.dev; |
|
- | 342 | uint32_t page_offset = offset_in_page(reloc->offset); |
|
- | 343 | uint64_t delta = (int)reloc->delta + target_offset; |
|
- | 344 | char *vaddr; |
|
- | 345 | int ret; |
|
- | 346 | ||
- | 347 | ret = i915_gem_object_set_to_gtt_domain(obj, true); |
|
- | 348 | if (ret) |
|
- | 349 | return ret; |
|
- | 350 | ||
- | 351 | vaddr = kmap_atomic(i915_gem_object_get_page(obj, |
|
- | 352 | reloc->offset >> PAGE_SHIFT)); |
|
- | 353 | clflush_write32(vaddr + page_offset, lower_32_bits(delta)); |
|
- | 354 | ||
- | 355 | if (INTEL_INFO(dev)->gen >= 8) { |
|
- | 356 | page_offset = offset_in_page(page_offset + sizeof(uint32_t)); |
|
- | 357 | ||
- | 358 | if (page_offset == 0) { |
|
- | 359 | kunmap_atomic(vaddr); |
|
- | 360 | vaddr = kmap_atomic(i915_gem_object_get_page(obj, |
|
- | 361 | (reloc->offset + sizeof(uint32_t)) >> PAGE_SHIFT)); |
|
- | 362 | } |
|
- | 363 | ||
- | 364 | clflush_write32(vaddr + page_offset, upper_32_bits(delta)); |
|
- | 365 | } |
|
Line 322... | Line 366... | ||
322 | 366 | ||
323 | 367 | kunmap_atomic(vaddr); |
|
Line 324... | Line 368... | ||
324 | 368 | ||
Line 348... | Line 392... | ||
348 | 392 | ||
349 | /* Sandybridge PPGTT errata: We need a global gtt mapping for MI and |
393 | /* Sandybridge PPGTT errata: We need a global gtt mapping for MI and |
350 | * pipe_control writes because the gpu doesn't properly redirect them |
394 | * pipe_control writes because the gpu doesn't properly redirect them |
351 | * through the ppgtt for non_secure batchbuffers. */ |
395 | * through the ppgtt for non_secure batchbuffers. */ |
352 | if (unlikely(IS_GEN6(dev) && |
396 | if (unlikely(IS_GEN6(dev) && |
353 | reloc->write_domain == I915_GEM_DOMAIN_INSTRUCTION && |
- | |
354 | !(target_vma->bound & GLOBAL_BIND))) |
397 | reloc->write_domain == I915_GEM_DOMAIN_INSTRUCTION)) { |
355 | target_vma->bind_vma(target_vma, target_i915_obj->cache_level, |
398 | ret = i915_vma_bind(target_vma, target_i915_obj->cache_level, |
- | 399 | PIN_GLOBAL); |
|
- | 400 | if (WARN_ONCE(ret, "Unexpected failure to bind target VMA!")) |
|
- | 401 | return ret; |
|
Line 356... | Line 402... | ||
356 | GLOBAL_BIND); |
402 | } |
357 | 403 | ||
358 | /* Validate that the target is in a valid r/w GPU domain */ |
404 | /* Validate that the target is in a valid r/w GPU domain */ |
359 | if (unlikely(reloc->write_domain & (reloc->write_domain - 1))) { |
405 | if (unlikely(reloc->write_domain & (reloc->write_domain - 1))) { |
Line 407... | Line 453... | ||
407 | 453 | ||
Line 408... | Line 454... | ||
408 | /* We can't wait for rendering with pagefaults disabled */ |
454 | /* We can't wait for rendering with pagefaults disabled */ |
409 | 455 | ||
410 | if (use_cpu_reloc(obj)) |
456 | if (use_cpu_reloc(obj)) |
411 | ret = relocate_entry_cpu(obj, reloc, target_offset); |
457 | ret = relocate_entry_cpu(obj, reloc, target_offset); |
- | 458 | else if (obj->map_and_fenceable) |
|
- | 459 | ret = relocate_entry_gtt(obj, reloc, target_offset); |
|
- | 460 | else if (1) |
|
- | 461 | ret = relocate_entry_clflush(obj, reloc, target_offset); |
|
- | 462 | else { |
|
- | 463 | WARN_ONCE(1, "Impossible case in relocation handling\n"); |
|
Line 412... | Line 464... | ||
412 | else |
464 | ret = -ENODEV; |
413 | ret = relocate_entry_gtt(obj, reloc, target_offset); |
465 | } |
Line 414... | Line 466... | ||
414 | 466 | ||
Line 505... | Line 557... | ||
505 | // pagefault_enable(); |
557 | // pagefault_enable(); |
Line 506... | Line 558... | ||
506 | 558 | ||
507 | return ret; |
559 | return ret; |
Line -... | Line 560... | ||
- | 560 | } |
|
- | 561 | ||
- | 562 | static bool only_mappable_for_reloc(unsigned int flags) |
|
- | 563 | { |
|
- | 564 | return (flags & (EXEC_OBJECT_NEEDS_FENCE | __EXEC_OBJECT_NEEDS_MAP)) == |
|
- | 565 | __EXEC_OBJECT_NEEDS_MAP; |
|
508 | } |
566 | } |
509 | 567 | ||
510 | static int |
568 | static int |
511 | i915_gem_execbuffer_reserve_vma(struct i915_vma *vma, |
569 | i915_gem_execbuffer_reserve_vma(struct i915_vma *vma, |
512 | struct intel_engine_cs *ring, |
570 | struct intel_engine_cs *ring, |
513 | bool *need_reloc) |
571 | bool *need_reloc) |
514 | { |
572 | { |
515 | struct drm_i915_gem_object *obj = vma->obj; |
573 | struct drm_i915_gem_object *obj = vma->obj; |
516 | struct drm_i915_gem_exec_object2 *entry = vma->exec_entry; |
574 | struct drm_i915_gem_exec_object2 *entry = vma->exec_entry; |
Line 517... | Line 575... | ||
517 | uint64_t flags; |
575 | uint64_t flags; |
518 | int ret; |
- | |
519 | - | ||
520 | flags = 0; |
576 | int ret; |
521 | if (entry->flags & __EXEC_OBJECT_NEEDS_MAP) |
577 | |
- | 578 | flags = PIN_USER; |
|
- | 579 | if (entry->flags & EXEC_OBJECT_NEEDS_GTT) |
|
- | 580 | flags |= PIN_GLOBAL; |
|
- | 581 | ||
- | 582 | if (!drm_mm_node_allocated(&vma->node)) { |
|
- | 583 | /* Wa32bitGeneralStateOffset & Wa32bitInstructionBaseOffset, |
|
- | 584 | * limit address to the first 4GBs for unflagged objects. |
|
- | 585 | */ |
|
- | 586 | if ((entry->flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS) == 0) |
|
522 | flags |= PIN_GLOBAL | PIN_MAPPABLE; |
587 | flags |= PIN_ZONE_4G; |
523 | if (entry->flags & EXEC_OBJECT_NEEDS_GTT) |
588 | if (entry->flags & __EXEC_OBJECT_NEEDS_MAP) |
- | 589 | flags |= PIN_GLOBAL | PIN_MAPPABLE; |
|
- | 590 | if (entry->flags & __EXEC_OBJECT_NEEDS_BIAS) |
|
- | 591 | flags |= BATCH_OFFSET_BIAS | PIN_OFFSET_BIAS; |
|
Line 524... | Line 592... | ||
524 | flags |= PIN_GLOBAL; |
592 | if ((flags & PIN_MAPPABLE) == 0) |
- | 593 | flags |= PIN_HIGH; |
|
- | 594 | } |
|
- | 595 | ||
- | 596 | ret = i915_gem_object_pin(obj, vma->vm, entry->alignment, flags); |
|
- | 597 | if ((ret == -ENOSPC || ret == -E2BIG) && |
|
525 | if (entry->flags & __EXEC_OBJECT_NEEDS_BIAS) |
598 | only_mappable_for_reloc(entry->flags)) |
526 | flags |= BATCH_OFFSET_BIAS | PIN_OFFSET_BIAS; |
599 | ret = i915_gem_object_pin(obj, vma->vm, |
Line 527... | Line 600... | ||
527 | 600 | entry->alignment, |
|
Line 585... | Line 658... | ||
585 | 658 | ||
586 | if (entry->alignment && |
659 | if (entry->alignment && |
587 | vma->node.start & (entry->alignment - 1)) |
660 | vma->node.start & (entry->alignment - 1)) |
Line 588... | Line -... | ||
588 | return true; |
- | |
589 | - | ||
590 | if (entry->flags & __EXEC_OBJECT_NEEDS_MAP && !obj->map_and_fenceable) |
- | |
591 | return true; |
661 | return true; |
592 | 662 | ||
593 | if (entry->flags & __EXEC_OBJECT_NEEDS_BIAS && |
663 | if (entry->flags & __EXEC_OBJECT_NEEDS_BIAS && |
Line -... | Line 664... | ||
- | 664 | vma->node.start < BATCH_OFFSET_BIAS) |
|
- | 665 | return true; |
|
- | 666 | ||
- | 667 | /* avoid costly ping-pong once a batch bo ended up non-mappable */ |
|
- | 668 | if (entry->flags & __EXEC_OBJECT_NEEDS_MAP && !obj->map_and_fenceable) |
|
- | 669 | return !only_mappable_for_reloc(entry->flags); |
|
- | 670 | ||
- | 671 | if ((entry->flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS) == 0 && |
|
594 | vma->node.start < BATCH_OFFSET_BIAS) |
672 | (vma->node.start + vma->node.size - 1) >> 32) |
595 | return true; |
673 | return true; |
Line 596... | Line 674... | ||
596 | 674 | ||
597 | return false; |
675 | return false; |
598 | } |
676 | } |
- | 677 | ||
599 | 678 | static int |
|
600 | static int |
679 | i915_gem_execbuffer_reserve(struct intel_engine_cs *ring, |
601 | i915_gem_execbuffer_reserve(struct intel_engine_cs *ring, |
680 | struct list_head *vmas, |
602 | struct list_head *vmas, |
681 | struct intel_context *ctx, |
603 | bool *need_relocs) |
682 | bool *need_relocs) |
Line 620... | Line 699... | ||
620 | 699 | ||
621 | vma = list_first_entry(vmas, struct i915_vma, exec_list); |
700 | vma = list_first_entry(vmas, struct i915_vma, exec_list); |
622 | obj = vma->obj; |
701 | obj = vma->obj; |
Line -... | Line 702... | ||
- | 702 | entry = vma->exec_entry; |
|
- | 703 | ||
- | 704 | if (ctx->flags & CONTEXT_NO_ZEROMAP) |
|
623 | entry = vma->exec_entry; |
705 | entry->flags |= __EXEC_OBJECT_NEEDS_BIAS; |
624 | 706 | ||
625 | if (!has_fenced_gpu_access) |
707 | if (!has_fenced_gpu_access) |
626 | entry->flags &= ~EXEC_OBJECT_NEEDS_FENCE; |
708 | entry->flags &= ~EXEC_OBJECT_NEEDS_FENCE; |
627 | need_fence = |
709 | need_fence = |
Line 697... | Line 779... | ||
697 | i915_gem_execbuffer_relocate_slow(struct drm_device *dev, |
779 | i915_gem_execbuffer_relocate_slow(struct drm_device *dev, |
698 | struct drm_i915_gem_execbuffer2 *args, |
780 | struct drm_i915_gem_execbuffer2 *args, |
699 | struct drm_file *file, |
781 | struct drm_file *file, |
700 | struct intel_engine_cs *ring, |
782 | struct intel_engine_cs *ring, |
701 | struct eb_vmas *eb, |
783 | struct eb_vmas *eb, |
702 | struct drm_i915_gem_exec_object2 *exec) |
784 | struct drm_i915_gem_exec_object2 *exec, |
- | 785 | struct intel_context *ctx) |
|
703 | { |
786 | { |
704 | struct drm_i915_gem_relocation_entry *reloc; |
787 | struct drm_i915_gem_relocation_entry *reloc; |
705 | struct i915_address_space *vm; |
788 | struct i915_address_space *vm; |
706 | struct i915_vma *vma; |
789 | struct i915_vma *vma; |
707 | bool need_relocs; |
790 | bool need_relocs; |
Line 723... | Line 806... | ||
723 | 806 | ||
724 | total = 0; |
807 | total = 0; |
725 | for (i = 0; i < count; i++) |
808 | for (i = 0; i < count; i++) |
Line 726... | Line 809... | ||
726 | total += exec[i].relocation_count; |
809 | total += exec[i].relocation_count; |
727 | 810 | ||
728 | reloc_offset = malloc(count * sizeof(*reloc_offset)); |
811 | reloc_offset = __builtin_malloc(count * sizeof(*reloc_offset)); |
729 | reloc = malloc(total * sizeof(*reloc)); |
812 | reloc = __builtin_malloc(total * sizeof(*reloc)); |
730 | if (reloc == NULL || reloc_offset == NULL) { |
813 | if (reloc == NULL || reloc_offset == NULL) { |
731 | kfree(reloc); |
814 | kfree(reloc); |
732 | kfree(reloc_offset); |
815 | kfree(reloc_offset); |
Line 783... | Line 866... | ||
783 | ret = eb_lookup_vmas(eb, exec, args, vm, file); |
866 | ret = eb_lookup_vmas(eb, exec, args, vm, file); |
784 | if (ret) |
867 | if (ret) |
785 | goto err; |
868 | goto err; |
Line 786... | Line 869... | ||
786 | 869 | ||
787 | need_relocs = (args->flags & I915_EXEC_NO_RELOC) == 0; |
870 | need_relocs = (args->flags & I915_EXEC_NO_RELOC) == 0; |
788 | ret = i915_gem_execbuffer_reserve(ring, &eb->vmas, &need_relocs); |
871 | ret = i915_gem_execbuffer_reserve(ring, &eb->vmas, ctx, &need_relocs); |
789 | if (ret) |
872 | if (ret) |
Line 790... | Line 873... | ||
790 | goto err; |
873 | goto err; |
791 | 874 | ||
Line 808... | Line 891... | ||
808 | kfree(reloc_offset); |
891 | kfree(reloc_offset); |
809 | return ret; |
892 | return ret; |
810 | } |
893 | } |
Line 811... | Line 894... | ||
811 | 894 | ||
812 | static int |
895 | static int |
813 | i915_gem_execbuffer_move_to_gpu(struct intel_engine_cs *ring, |
896 | i915_gem_execbuffer_move_to_gpu(struct drm_i915_gem_request *req, |
814 | struct list_head *vmas) |
897 | struct list_head *vmas) |
- | 898 | { |
|
815 | { |
899 | const unsigned other_rings = ~intel_ring_flag(req->ring); |
816 | struct i915_vma *vma; |
900 | struct i915_vma *vma; |
817 | uint32_t flush_domains = 0; |
901 | uint32_t flush_domains = 0; |
818 | bool flush_chipset = false; |
902 | bool flush_chipset = false; |
Line 819... | Line 903... | ||
819 | int ret; |
903 | int ret; |
820 | 904 | ||
- | 905 | list_for_each_entry(vma, vmas, exec_list) { |
|
- | 906 | struct drm_i915_gem_object *obj = vma->obj; |
|
821 | list_for_each_entry(vma, vmas, exec_list) { |
907 | |
822 | struct drm_i915_gem_object *obj = vma->obj; |
908 | if (obj->active & other_rings) { |
823 | ret = i915_gem_object_sync(obj, ring); |
909 | ret = i915_gem_object_sync(obj, req->ring, &req); |
- | 910 | if (ret) |
|
Line 824... | Line 911... | ||
824 | if (ret) |
911 | return ret; |
825 | return ret; |
912 | } |
Line 826... | Line 913... | ||
826 | 913 | ||
827 | if (obj->base.write_domain & I915_GEM_DOMAIN_CPU) |
914 | if (obj->base.write_domain & I915_GEM_DOMAIN_CPU) |
Line 828... | Line 915... | ||
828 | flush_chipset |= i915_gem_clflush_object(obj, false); |
915 | flush_chipset |= i915_gem_clflush_object(obj, false); |
829 | 916 | ||
Line 830... | Line 917... | ||
830 | flush_domains |= obj->base.write_domain; |
917 | flush_domains |= obj->base.write_domain; |
831 | } |
918 | } |
Line 832... | Line 919... | ||
832 | 919 | ||
833 | if (flush_chipset) |
920 | if (flush_chipset) |
834 | i915_gem_chipset_flush(ring->dev); |
921 | i915_gem_chipset_flush(req->ring->dev); |
835 | 922 | ||
836 | if (flush_domains & I915_GEM_DOMAIN_GTT) |
923 | if (flush_domains & I915_GEM_DOMAIN_GTT) |
Line 837... | Line 924... | ||
837 | wmb(); |
924 | wmb(); |
838 | 925 | ||
839 | /* Unconditionally invalidate gpu caches and ensure that we do flush |
926 | /* Unconditionally invalidate gpu caches and ensure that we do flush |
840 | * any residual writes from the previous batch. |
927 | * any residual writes from the previous batch. |
841 | */ |
928 | */ |
Line -... | Line 929... | ||
- | 929 | return intel_ring_invalidate_all_caches(req); |
|
- | 930 | } |
|
- | 931 | ||
- | 932 | static bool |
|
- | 933 | i915_gem_check_execbuffer(struct drm_i915_gem_execbuffer2 *exec) |
|
- | 934 | { |
|
- | 935 | if (exec->flags & __I915_EXEC_UNKNOWN_FLAGS) |
|
- | 936 | return false; |
|
- | 937 | ||
- | 938 | /* Kernel clipping was a DRI1 misfeature */ |
|
- | 939 | if (exec->num_cliprects || exec->cliprects_ptr) |
|
842 | return intel_ring_invalidate_all_caches(ring); |
940 | return false; |
- | 941 | ||
- | 942 | if (exec->DR4 == 0xffffffff) { |
|
- | 943 | DRM_DEBUG("UXA submitting garbage DR4, fixing up\n"); |
|
843 | } |
944 | exec->DR4 = 0; |
Line 844... | Line 945... | ||
844 | 945 | } |
|
845 | static bool |
946 | if (exec->DR1 || exec->DR4) |
846 | i915_gem_check_execbuffer(struct drm_i915_gem_execbuffer2 *exec) |
947 | return false; |
Line 870... | Line 971... | ||
870 | int length; /* limited by fault_in_pages_readable() */ |
971 | int length; /* limited by fault_in_pages_readable() */ |
Line 871... | Line 972... | ||
871 | 972 | ||
872 | if (exec[i].flags & invalid_flags) |
973 | if (exec[i].flags & invalid_flags) |
Line -... | Line 974... | ||
- | 974 | return -EINVAL; |
|
- | 975 | ||
- | 976 | if (exec[i].alignment && !is_power_of_2(exec[i].alignment)) |
|
873 | return -EINVAL; |
977 | return -EINVAL; |
874 | 978 | ||
875 | /* First check for malicious input causing overflow in |
979 | /* First check for malicious input causing overflow in |
876 | * the worst case where we need to allocate the entire |
980 | * the worst case where we need to allocate the entire |
877 | * relocation tree as a single array. |
981 | * relocation tree as a single array. |
Line 911... | Line 1015... | ||
911 | DRM_DEBUG("Context %u tried to submit while banned\n", ctx_id); |
1015 | DRM_DEBUG("Context %u tried to submit while banned\n", ctx_id); |
912 | return ERR_PTR(-EIO); |
1016 | return ERR_PTR(-EIO); |
913 | } |
1017 | } |
Line 914... | Line 1018... | ||
914 | 1018 | ||
915 | if (i915.enable_execlists && !ctx->engine[ring->id].state) { |
1019 | if (i915.enable_execlists && !ctx->engine[ring->id].state) { |
916 | int ret = intel_lr_context_deferred_create(ctx, ring); |
1020 | int ret = intel_lr_context_deferred_alloc(ctx, ring); |
917 | if (ret) { |
1021 | if (ret) { |
918 | DRM_DEBUG("Could not create LRC %u: %d\n", ctx_id, ret); |
1022 | DRM_DEBUG("Could not create LRC %u: %d\n", ctx_id, ret); |
919 | return ERR_PTR(ret); |
1023 | return ERR_PTR(ret); |
920 | } |
1024 | } |
Line 923... | Line 1027... | ||
923 | return ctx; |
1027 | return ctx; |
924 | } |
1028 | } |
Line 925... | Line 1029... | ||
925 | 1029 | ||
926 | void |
1030 | void |
927 | i915_gem_execbuffer_move_to_active(struct list_head *vmas, |
1031 | i915_gem_execbuffer_move_to_active(struct list_head *vmas, |
928 | struct intel_engine_cs *ring) |
1032 | struct drm_i915_gem_request *req) |
929 | { |
1033 | { |
930 | u32 seqno = intel_ring_get_seqno(ring); |
1034 | struct intel_engine_cs *ring = i915_gem_request_get_ring(req); |
Line 931... | Line 1035... | ||
931 | struct i915_vma *vma; |
1035 | struct i915_vma *vma; |
932 | 1036 | ||
933 | list_for_each_entry(vma, vmas, exec_list) { |
1037 | list_for_each_entry(vma, vmas, exec_list) { |
934 | struct drm_i915_gem_exec_object2 *entry = vma->exec_entry; |
1038 | struct drm_i915_gem_exec_object2 *entry = vma->exec_entry; |
935 | struct drm_i915_gem_object *obj = vma->obj; |
1039 | struct drm_i915_gem_object *obj = vma->obj; |
Line -... | Line 1040... | ||
- | 1040 | u32 old_read = obj->base.read_domains; |
|
936 | u32 old_read = obj->base.read_domains; |
1041 | u32 old_write = obj->base.write_domain; |
937 | u32 old_write = obj->base.write_domain; |
1042 | |
938 | 1043 | obj->dirty = 1; /* be paranoid */ |
|
939 | obj->base.write_domain = obj->base.pending_write_domain; |
1044 | obj->base.write_domain = obj->base.pending_write_domain; |
Line 940... | Line 1045... | ||
940 | if (obj->base.write_domain == 0) |
1045 | if (obj->base.write_domain == 0) |
941 | obj->base.pending_read_domains |= obj->base.read_domains; |
1046 | obj->base.pending_read_domains |= obj->base.read_domains; |
942 | obj->base.read_domains = obj->base.pending_read_domains; |
- | |
943 | 1047 | obj->base.read_domains = obj->base.pending_read_domains; |
|
Line 944... | Line 1048... | ||
944 | i915_vma_move_to_active(vma, ring); |
1048 | |
Line 945... | Line 1049... | ||
945 | if (obj->base.write_domain) { |
1049 | i915_vma_move_to_active(vma, req); |
946 | obj->dirty = 1; |
1050 | if (obj->base.write_domain) { |
947 | obj->last_write_seqno = seqno; |
1051 | i915_gem_request_assign(&obj->last_write_req, req); |
948 | 1052 | ||
949 | intel_fb_obj_invalidate(obj, ring); |
1053 | intel_fb_obj_invalidate(obj, ORIGIN_CS); |
950 | 1054 | ||
951 | /* update for the implicit flush after a batch */ |
1055 | /* update for the implicit flush after a batch */ |
952 | obj->base.write_domain &= ~I915_GEM_GPU_DOMAINS; |
1056 | obj->base.write_domain &= ~I915_GEM_GPU_DOMAINS; |
953 | } |
1057 | } |
954 | if (entry->flags & EXEC_OBJECT_NEEDS_FENCE) { |
1058 | if (entry->flags & EXEC_OBJECT_NEEDS_FENCE) { |
Line 963... | Line 1067... | ||
963 | trace_i915_gem_object_change_domain(obj, old_read, old_write); |
1067 | trace_i915_gem_object_change_domain(obj, old_read, old_write); |
964 | } |
1068 | } |
965 | } |
1069 | } |
Line 966... | Line 1070... | ||
966 | 1070 | ||
967 | void |
1071 | void |
968 | i915_gem_execbuffer_retire_commands(struct drm_device *dev, |
- | |
969 | struct drm_file *file, |
- | |
970 | struct intel_engine_cs *ring, |
- | |
971 | struct drm_i915_gem_object *obj) |
1072 | i915_gem_execbuffer_retire_commands(struct i915_execbuffer_params *params) |
972 | { |
1073 | { |
973 | /* Unconditionally force add_request to emit a full flush. */ |
1074 | /* Unconditionally force add_request to emit a full flush. */ |
Line 974... | Line 1075... | ||
974 | ring->gpu_caches_dirty = true; |
1075 | params->ring->gpu_caches_dirty = true; |
975 | 1076 | ||
976 | /* Add a breadcrumb for the completion of the batch buffer */ |
1077 | /* Add a breadcrumb for the completion of the batch buffer */ |
Line 977... | Line 1078... | ||
977 | (void)__i915_add_request(ring, file, obj, NULL); |
1078 | __i915_add_request(params->request, params->batch_obj, true); |
978 | } |
1079 | } |
979 | 1080 | ||
980 | static int |
1081 | static int |
- | 1082 | i915_reset_gen7_sol_offsets(struct drm_device *dev, |
|
981 | i915_reset_gen7_sol_offsets(struct drm_device *dev, |
1083 | struct drm_i915_gem_request *req) |
982 | struct intel_engine_cs *ring) |
1084 | { |
Line 983... | Line 1085... | ||
983 | { |
1085 | struct intel_engine_cs *ring = req->ring; |
984 | struct drm_i915_private *dev_priv = dev->dev_private; |
1086 | struct drm_i915_private *dev_priv = dev->dev_private; |
985 | int ret, i; |
1087 | int ret, i; |
986 | 1088 | ||
Line 987... | Line 1089... | ||
987 | if (!IS_GEN7(dev) || ring != &dev_priv->ring[RCS]) { |
1089 | if (!IS_GEN7(dev) || ring != &dev_priv->ring[RCS]) { |
988 | DRM_DEBUG("sol reset is gen7/rcs only\n"); |
1090 | DRM_DEBUG("sol reset is gen7/rcs only\n"); |
989 | return -EINVAL; |
1091 | return -EINVAL; |
Line 990... | Line 1092... | ||
990 | } |
1092 | } |
991 | 1093 | ||
Line 1002... | Line 1104... | ||
1002 | intel_ring_advance(ring); |
1104 | intel_ring_advance(ring); |
Line 1003... | Line 1105... | ||
1003 | 1105 | ||
1004 | return 0; |
1106 | return 0; |
Line 1005... | Line 1107... | ||
1005 | } |
1107 | } |
1006 | 1108 | ||
- | 1109 | static struct drm_i915_gem_object* |
|
- | 1110 | i915_gem_execbuffer_parse(struct intel_engine_cs *ring, |
|
1007 | static int |
1111 | struct drm_i915_gem_exec_object2 *shadow_exec_entry, |
- | 1112 | struct eb_vmas *eb, |
|
- | 1113 | struct drm_i915_gem_object *batch_obj, |
|
1008 | i915_emit_box(struct intel_engine_cs *ring, |
1114 | u32 batch_start_offset, |
1009 | struct drm_clip_rect *box, |
1115 | u32 batch_len, |
- | 1116 | bool is_master) |
|
- | 1117 | { |
|
1010 | int DR1, int DR4) |
1118 | struct drm_i915_gem_object *shadow_batch_obj; |
Line 1011... | Line 1119... | ||
1011 | { |
1119 | struct i915_vma *vma; |
1012 | int ret; |
1120 | int ret; |
1013 | 1121 | ||
1014 | if (box->y2 <= box->y1 || box->x2 <= box->x1 || |
- | |
1015 | box->y2 <= 0 || box->x2 <= 0) { |
1122 | shadow_batch_obj = i915_gem_batch_pool_get(&ring->batch_pool, |
1016 | DRM_ERROR("Bad box %d,%d..%d,%d\n", |
- | |
Line 1017... | Line 1123... | ||
1017 | box->x1, box->y1, box->x2, box->y2); |
1123 | PAGE_ALIGN(batch_len)); |
- | 1124 | if (IS_ERR(shadow_batch_obj)) |
|
1018 | return -EINVAL; |
1125 | return shadow_batch_obj; |
- | 1126 | ||
- | 1127 | ret = i915_parse_cmds(ring, |
|
- | 1128 | batch_obj, |
|
1019 | } |
1129 | shadow_batch_obj, |
1020 | 1130 | batch_start_offset, |
|
Line 1021... | Line -... | ||
1021 | if (INTEL_INFO(ring->dev)->gen >= 4) { |
- | |
1022 | ret = intel_ring_begin(ring, 4); |
- | |
1023 | if (ret) |
- | |
1024 | return ret; |
- | |
1025 | - | ||
1026 | intel_ring_emit(ring, GFX_OP_DRAWRECT_INFO_I965); |
1131 | batch_len, |
1027 | intel_ring_emit(ring, (box->x1 & 0xffff) | box->y1 << 16); |
1132 | is_master); |
1028 | intel_ring_emit(ring, ((box->x2 - 1) & 0xffff) | (box->y2 - 1) << 16); |
1133 | if (ret) |
Line 1029... | Line 1134... | ||
1029 | intel_ring_emit(ring, DR4); |
1134 | goto err; |
1030 | } else { |
- | |
1031 | ret = intel_ring_begin(ring, 6); |
- | |
1032 | if (ret) |
- | |
1033 | return ret; |
- | |
1034 | - | ||
1035 | intel_ring_emit(ring, GFX_OP_DRAWRECT_INFO); |
- | |
1036 | intel_ring_emit(ring, DR1); |
- | |
Line -... | Line 1135... | ||
- | 1135 | ||
- | 1136 | ret = i915_gem_obj_ggtt_pin(shadow_batch_obj, 0, 0); |
|
- | 1137 | if (ret) |
|
1037 | intel_ring_emit(ring, (box->x1 & 0xffff) | box->y1 << 16); |
1138 | goto err; |
- | 1139 | ||
- | 1140 | i915_gem_object_unpin_pages(shadow_batch_obj); |
|
- | 1141 | ||
- | 1142 | memset(shadow_exec_entry, 0, sizeof(*shadow_exec_entry)); |
|
- | 1143 | ||
1038 | intel_ring_emit(ring, ((box->x2 - 1) & 0xffff) | (box->y2 - 1) << 16); |
1144 | vma = i915_gem_obj_to_ggtt(shadow_batch_obj); |
- | 1145 | vma->exec_entry = shadow_exec_entry; |
|
Line -... | Line 1146... | ||
- | 1146 | vma->exec_entry->flags = __EXEC_OBJECT_HAS_PIN; |
|
- | 1147 | drm_gem_object_reference(&shadow_batch_obj->base); |
|
- | 1148 | list_add_tail(&vma->exec_list, &eb->vmas); |
|
- | 1149 | ||
- | 1150 | shadow_batch_obj->base.pending_read_domains = I915_GEM_DOMAIN_COMMAND; |
|
- | 1151 | ||
- | 1152 | return shadow_batch_obj; |
|
Line 1039... | Line 1153... | ||
1039 | intel_ring_emit(ring, DR4); |
1153 | |
1040 | intel_ring_emit(ring, 0); |
1154 | err: |
1041 | } |
- | |
1042 | intel_ring_advance(ring); |
- | |
1043 | 1155 | i915_gem_object_unpin_pages(shadow_batch_obj); |
|
1044 | return 0; |
1156 | if (ret == -EACCES) /* unhandled chained batch */ |
1045 | } |
- | |
1046 | - | ||
1047 | 1157 | return batch_obj; |
|
1048 | int |
1158 | else |
- | 1159 | return ERR_PTR(ret); |
|
1049 | i915_gem_ringbuffer_submission(struct drm_device *dev, struct drm_file *file, |
1160 | } |
1050 | struct intel_engine_cs *ring, |
1161 | |
1051 | struct intel_context *ctx, |
1162 | int |
1052 | struct drm_i915_gem_execbuffer2 *args, |
1163 | i915_gem_ringbuffer_submission(struct i915_execbuffer_params *params, |
1053 | struct list_head *vmas, |
1164 | struct drm_i915_gem_execbuffer2 *args, |
1054 | struct drm_i915_gem_object *batch_obj, |
- | |
1055 | u64 exec_start, u32 flags) |
- | |
1056 | { |
- | |
1057 | struct drm_clip_rect *cliprects = NULL; |
- | |
1058 | struct drm_i915_private *dev_priv = dev->dev_private; |
- | |
1059 | u64 exec_len; |
- | |
1060 | int instp_mode; |
- | |
1061 | u32 instp_mask; |
- | |
1062 | int i, ret = 0; |
- | |
1063 | - | ||
1064 | if (args->num_cliprects != 0) { |
- | |
1065 | if (ring != &dev_priv->ring[RCS]) { |
- | |
1066 | DRM_DEBUG("clip rectangles are only valid with the render ring\n"); |
- | |
1067 | return -EINVAL; |
- | |
1068 | } |
- | |
1069 | - | ||
1070 | if (INTEL_INFO(dev)->gen >= 5) { |
- | |
1071 | DRM_DEBUG("clip rectangles are only valid on pre-gen5\n"); |
- | |
1072 | return -EINVAL; |
- | |
1073 | } |
- | |
1074 | - | ||
1075 | if (args->num_cliprects > UINT_MAX / sizeof(*cliprects)) { |
- | |
1076 | DRM_DEBUG("execbuf with %u cliprects\n", |
- | |
1077 | args->num_cliprects); |
- | |
1078 | return -EINVAL; |
- | |
1079 | } |
- | |
1080 | - | ||
1081 | cliprects = kcalloc(args->num_cliprects, |
- | |
1082 | sizeof(*cliprects), |
- | |
1083 | GFP_KERNEL); |
- | |
1084 | if (cliprects == NULL) { |
- | |
1085 | ret = -ENOMEM; |
- | |
1086 | goto error; |
- | |
1087 | } |
- | |
1088 | - | ||
1089 | if (copy_from_user(cliprects, |
- | |
1090 | to_user_ptr(args->cliprects_ptr), |
- | |
1091 | sizeof(*cliprects)*args->num_cliprects)) { |
- | |
1092 | ret = -EFAULT; |
- | |
1093 | goto error; |
- | |
1094 | } |
- | |
1095 | } else { |
- | |
1096 | if (args->DR4 == 0xffffffff) { |
- | |
Line 1097... | Line 1165... | ||
1097 | DRM_DEBUG("UXA submitting garbage DR4, fixing up\n"); |
1165 | struct list_head *vmas) |
1098 | args->DR4 = 0; |
1166 | { |
1099 | } |
1167 | struct drm_device *dev = params->dev; |
Line 1100... | Line 1168... | ||
1100 | 1168 | struct intel_engine_cs *ring = params->ring; |
|
1101 | if (args->DR1 || args->DR4 || args->cliprects_ptr) { |
1169 | struct drm_i915_private *dev_priv = dev->dev_private; |
1102 | DRM_DEBUG("0 cliprects but dirt in cliprects fields\n"); |
1170 | u64 exec_start, exec_len; |
- | 1171 | int instp_mode; |
|
- | 1172 | u32 instp_mask; |
|
- | 1173 | int ret; |
|
Line 1103... | Line 1174... | ||
1103 | return -EINVAL; |
1174 | |
1104 | } |
1175 | ret = i915_gem_execbuffer_move_to_gpu(params->request, vmas); |
1105 | } |
1176 | if (ret) |
1106 | 1177 | return ret; |
|
1107 | ret = i915_gem_execbuffer_move_to_gpu(ring, vmas); |
1178 | |
1108 | if (ret) |
1179 | ret = i915_switch_context(params->request); |
1109 | goto error; |
1180 | if (ret) |
1110 | 1181 | return ret; |
|
1111 | ret = i915_switch_context(ring, ctx); |
1182 | |
1112 | if (ret) |
- | |
1113 | goto error; |
1183 | WARN(params->ctx->ppgtt && params->ctx->ppgtt->pd_dirty_rings & (1< |
Line 1114... | Line 1184... | ||
1114 | 1184 | "%s didn't clear reload\n", ring->name); |
|
1115 | instp_mode = args->flags & I915_EXEC_CONSTANTS_MASK; |
1185 | |
1116 | instp_mask = I915_EXEC_CONSTANTS_MASK; |
1186 | instp_mode = args->flags & I915_EXEC_CONSTANTS_MASK; |
1117 | switch (instp_mode) { |
1187 | instp_mask = I915_EXEC_CONSTANTS_MASK; |
1118 | case I915_EXEC_CONSTANTS_REL_GENERAL: |
- | |
1119 | case I915_EXEC_CONSTANTS_ABSOLUTE: |
1188 | switch (instp_mode) { |
Line 1120... | Line 1189... | ||
1120 | case I915_EXEC_CONSTANTS_REL_SURFACE: |
1189 | case I915_EXEC_CONSTANTS_REL_GENERAL: |
1121 | if (instp_mode != 0 && ring != &dev_priv->ring[RCS]) { |
1190 | case I915_EXEC_CONSTANTS_ABSOLUTE: |
1122 | DRM_DEBUG("non-0 rel constants mode on non-RCS\n"); |
1191 | case I915_EXEC_CONSTANTS_REL_SURFACE: |
1123 | ret = -EINVAL; |
1192 | if (instp_mode != 0 && ring != &dev_priv->ring[RCS]) { |
1124 | goto error; |
- | |
1125 | } |
1193 | DRM_DEBUG("non-0 rel constants mode on non-RCS\n"); |
Line 1126... | Line 1194... | ||
1126 | 1194 | return -EINVAL; |
|
1127 | if (instp_mode != dev_priv->relative_constants_mode) { |
1195 | } |
1128 | if (INTEL_INFO(dev)->gen < 4) { |
1196 | |
1129 | DRM_DEBUG("no rel constants on pre-gen4\n"); |
1197 | if (instp_mode != dev_priv->relative_constants_mode) { |
1130 | ret = -EINVAL; |
1198 | if (INTEL_INFO(dev)->gen < 4) { |
1131 | goto error; |
1199 | DRM_DEBUG("no rel constants on pre-gen4\n"); |
1132 | } |
1200 | return -EINVAL; |
1133 | 1201 | } |
|
1134 | if (INTEL_INFO(dev)->gen > 5 && |
- | |
1135 | instp_mode == I915_EXEC_CONSTANTS_REL_SURFACE) { |
1202 | |
Line 1136... | Line 1203... | ||
1136 | DRM_DEBUG("rel surface constants mode invalid on gen5+\n"); |
1203 | if (INTEL_INFO(dev)->gen > 5 && |
1137 | ret = -EINVAL; |
1204 | instp_mode == I915_EXEC_CONSTANTS_REL_SURFACE) { |
1138 | goto error; |
1205 | DRM_DEBUG("rel surface constants mode invalid on gen5+\n"); |
1139 | } |
1206 | return -EINVAL; |
1140 | 1207 | } |
|
Line 1141... | Line 1208... | ||
1141 | /* The HW changed the meaning on this bit on gen6 */ |
1208 | |
1142 | if (INTEL_INFO(dev)->gen >= 6) |
1209 | /* The HW changed the meaning on this bit on gen6 */ |
1143 | instp_mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE; |
1210 | if (INTEL_INFO(dev)->gen >= 6) |
1144 | } |
1211 | instp_mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE; |
Line 1163... | Line 1230... | ||
1163 | 1230 | ||
1164 | dev_priv->relative_constants_mode = instp_mode; |
1231 | dev_priv->relative_constants_mode = instp_mode; |
Line 1165... | Line 1232... | ||
1165 | } |
1232 | } |
1166 | 1233 | ||
1167 | if (args->flags & I915_EXEC_GEN7_SOL_RESET) { |
1234 | if (args->flags & I915_EXEC_GEN7_SOL_RESET) { |
1168 | ret = i915_reset_gen7_sol_offsets(dev, ring); |
1235 | ret = i915_reset_gen7_sol_offsets(dev, params->request); |
1169 | if (ret) |
1236 | if (ret) |
Line 1170... | Line 1237... | ||
1170 | goto error; |
1237 | return ret; |
1171 | } |
- | |
1172 | 1238 | } |
|
1173 | exec_len = args->batch_len; |
- | |
1174 | if (cliprects) { |
1239 | |
1175 | for (i = 0; i < args->num_cliprects; i++) { |
- | |
1176 | ret = i915_emit_box(ring, &cliprects[i], |
- | |
Line 1177... | Line -... | ||
1177 | args->DR1, args->DR4); |
- | |
1178 | if (ret) |
- | |
1179 | goto error; |
- | |
1180 | - | ||
1181 | ret = ring->dispatch_execbuffer(ring, |
- | |
1182 | exec_start, exec_len, |
- | |
1183 | flags); |
- | |
1184 | if (ret) |
1240 | exec_len = args->batch_len; |
1185 | goto error; |
1241 | exec_start = params->batch_obj_vm_offset + |
1186 | } |
1242 | params->args_batch_start_offset; |
1187 | } else { |
1243 | |
1188 | ret = ring->dispatch_execbuffer(ring, |
1244 | ret = ring->dispatch_execbuffer(params->request, |
1189 | exec_start, exec_len, |
- | |
Line 1190... | Line 1245... | ||
1190 | flags); |
1245 | exec_start, exec_len, |
Line 1191... | Line 1246... | ||
1191 | if (ret) |
1246 | params->dispatch_flags); |
1192 | return ret; |
1247 | if (ret) |
Line 1193... | Line -... | ||
1193 | } |
- | |
1194 | - | ||
1195 | trace_i915_gem_ring_dispatch(ring, intel_ring_get_seqno(ring), flags); |
1248 | return ret; |
1196 | 1249 | ||
Line 1197... | Line 1250... | ||
1197 | i915_gem_execbuffer_move_to_active(vmas, ring); |
1250 | trace_i915_gem_ring_dispatch(params->request, params->dispatch_flags); |
1198 | i915_gem_execbuffer_retire_commands(dev, file, ring, batch_obj); |
1251 | |
1199 | 1252 | i915_gem_execbuffer_move_to_active(vmas, params->request); |
|
Line 1259... | Line 1312... | ||
1259 | struct drm_i915_gem_exec_object2 *exec) |
1312 | struct drm_i915_gem_exec_object2 *exec) |
1260 | { |
1313 | { |
1261 | struct drm_i915_private *dev_priv = dev->dev_private; |
1314 | struct drm_i915_private *dev_priv = dev->dev_private; |
1262 | struct eb_vmas *eb; |
1315 | struct eb_vmas *eb; |
1263 | struct drm_i915_gem_object *batch_obj; |
1316 | struct drm_i915_gem_object *batch_obj; |
- | 1317 | struct drm_i915_gem_exec_object2 shadow_exec_entry; |
|
1264 | struct intel_engine_cs *ring; |
1318 | struct intel_engine_cs *ring; |
1265 | struct intel_context *ctx; |
1319 | struct intel_context *ctx; |
1266 | struct i915_address_space *vm; |
1320 | struct i915_address_space *vm; |
- | 1321 | struct i915_execbuffer_params params_master; /* XXX: will be removed later */ |
|
- | 1322 | struct i915_execbuffer_params *params = ¶ms_master; |
|
1267 | const u32 ctx_id = i915_execbuffer2_get_context_id(*args); |
1323 | const u32 ctx_id = i915_execbuffer2_get_context_id(*args); |
1268 | u64 exec_start = args->batch_start_offset; |
- | |
1269 | u32 flags; |
1324 | u32 dispatch_flags; |
1270 | int ret; |
1325 | int ret; |
1271 | bool need_relocs; |
1326 | bool need_relocs; |
Line 1272... | Line 1327... | ||
1272 | 1327 | ||
1273 | if (!i915_gem_check_execbuffer(args)) |
1328 | if (!i915_gem_check_execbuffer(args)) |
Line 1274... | Line 1329... | ||
1274 | return -EINVAL; |
1329 | return -EINVAL; |
1275 | 1330 | ||
1276 | ret = validate_exec_list(dev, exec, args->buffer_count); |
1331 | ret = validate_exec_list(dev, exec, args->buffer_count); |
Line 1277... | Line 1332... | ||
1277 | if (ret) |
1332 | if (ret) |
1278 | return ret; |
1333 | return ret; |
Line 1279... | Line 1334... | ||
1279 | 1334 | ||
1280 | flags = 0; |
1335 | dispatch_flags = 0; |
1281 | if (args->flags & I915_EXEC_SECURE) { |
1336 | if (args->flags & I915_EXEC_SECURE) { |
1282 | 1337 | ||
Line 1283... | Line 1338... | ||
1283 | flags |= I915_DISPATCH_SECURE; |
1338 | dispatch_flags |= I915_DISPATCH_SECURE; |
1284 | } |
1339 | } |
1285 | if (args->flags & I915_EXEC_IS_PINNED) |
1340 | if (args->flags & I915_EXEC_IS_PINNED) |
1286 | flags |= I915_DISPATCH_PINNED; |
1341 | dispatch_flags |= I915_DISPATCH_PINNED; |
1287 | 1342 | ||
Line -... | Line 1343... | ||
- | 1343 | if ((args->flags & I915_EXEC_RING_MASK) > LAST_USER_RING) { |
|
- | 1344 | DRM_DEBUG("execbuf with unknown ring: %d\n", |
|
- | 1345 | (int)(args->flags & I915_EXEC_RING_MASK)); |
|
- | 1346 | return -EINVAL; |
|
- | 1347 | } |
|
- | 1348 | ||
- | 1349 | if (((args->flags & I915_EXEC_RING_MASK) != I915_EXEC_BSD) && |
|
1288 | if ((args->flags & I915_EXEC_RING_MASK) > LAST_USER_RING) { |
1350 | ((args->flags & I915_EXEC_BSD_MASK) != 0)) { |
1289 | DRM_DEBUG("execbuf with unknown ring: %d\n", |
1351 | DRM_DEBUG("execbuf with non bsd ring but with invalid " |
1290 | (int)(args->flags & I915_EXEC_RING_MASK)); |
1352 | "bsd dispatch flags: %d\n", (int)(args->flags)); |
1291 | return -EINVAL; |
1353 | return -EINVAL; |
1292 | } |
1354 | } |
- | 1355 | ||
- | 1356 | if ((args->flags & I915_EXEC_RING_MASK) == I915_EXEC_DEFAULT) |
|
- | 1357 | ring = &dev_priv->ring[RCS]; |
|
1293 | 1358 | else if ((args->flags & I915_EXEC_RING_MASK) == I915_EXEC_BSD) { |
|
1294 | if ((args->flags & I915_EXEC_RING_MASK) == I915_EXEC_DEFAULT) |
1359 | if (HAS_BSD2(dev)) { |
- | 1360 | int ring_id; |
|
- | 1361 | ||
- | 1362 | switch (args->flags & I915_EXEC_BSD_MASK) { |
|
- | 1363 | case I915_EXEC_BSD_DEFAULT: |
|
- | 1364 | ring_id = gen8_dispatch_bsd_ring(dev, file); |
|
- | 1365 | ring = &dev_priv->ring[ring_id]; |
|
- | 1366 | break; |
|
- | 1367 | case I915_EXEC_BSD_RING1: |
|
- | 1368 | ring = &dev_priv->ring[VCS]; |
|
- | 1369 | break; |
|
- | 1370 | case I915_EXEC_BSD_RING2: |
|
- | 1371 | ring = &dev_priv->ring[VCS2]; |
|
1295 | ring = &dev_priv->ring[RCS]; |
1372 | break; |
1296 | else if ((args->flags & I915_EXEC_RING_MASK) == I915_EXEC_BSD) { |
1373 | default: |
1297 | if (HAS_BSD2(dev)) { |
1374 | DRM_DEBUG("execbuf with unknown bsd ring: %d\n", |
1298 | int ring_id; |
1375 | (int)(args->flags & I915_EXEC_BSD_MASK)); |
Line 1312... | Line 1389... | ||
1312 | if (args->buffer_count < 1) { |
1389 | if (args->buffer_count < 1) { |
1313 | DRM_DEBUG("execbuf with %d buffers\n", args->buffer_count); |
1390 | DRM_DEBUG("execbuf with %d buffers\n", args->buffer_count); |
1314 | return -EINVAL; |
1391 | return -EINVAL; |
1315 | } |
1392 | } |
Line -... | Line 1393... | ||
- | 1393 | ||
- | 1394 | if (args->flags & I915_EXEC_RESOURCE_STREAMER) { |
|
- | 1395 | if (!HAS_RESOURCE_STREAMER(dev)) { |
|
- | 1396 | DRM_DEBUG("RS is only allowed for Haswell, Gen8 and above\n"); |
|
- | 1397 | return -EINVAL; |
|
- | 1398 | } |
|
- | 1399 | if (ring->id != RCS) { |
|
- | 1400 | DRM_DEBUG("RS is not available on %s\n", |
|
- | 1401 | ring->name); |
|
- | 1402 | return -EINVAL; |
|
- | 1403 | } |
|
- | 1404 | ||
- | 1405 | dispatch_flags |= I915_DISPATCH_RS; |
|
- | 1406 | } |
|
1316 | 1407 | ||
Line 1317... | Line 1408... | ||
1317 | intel_runtime_pm_get(dev_priv); |
1408 | intel_runtime_pm_get(dev_priv); |
1318 | 1409 | ||
1319 | ret = i915_mutex_lock_interruptible(dev); |
1410 | ret = i915_mutex_lock_interruptible(dev); |
Line 1332... | Line 1423... | ||
1332 | if (ctx->ppgtt) |
1423 | if (ctx->ppgtt) |
1333 | vm = &ctx->ppgtt->base; |
1424 | vm = &ctx->ppgtt->base; |
1334 | else |
1425 | else |
1335 | vm = &dev_priv->gtt.base; |
1426 | vm = &dev_priv->gtt.base; |
Line -... | Line 1427... | ||
- | 1427 | ||
- | 1428 | memset(¶ms_master, 0x00, sizeof(params_master)); |
|
1336 | 1429 | ||
1337 | eb = eb_create(args); |
1430 | eb = eb_create(args); |
1338 | if (eb == NULL) { |
1431 | if (eb == NULL) { |
1339 | i915_gem_context_unreference(ctx); |
1432 | i915_gem_context_unreference(ctx); |
1340 | mutex_unlock(&dev->struct_mutex); |
1433 | mutex_unlock(&dev->struct_mutex); |
Line 1350... | Line 1443... | ||
1350 | /* take note of the batch buffer before we might reorder the lists */ |
1443 | /* take note of the batch buffer before we might reorder the lists */ |
1351 | batch_obj = eb_get_batch(eb); |
1444 | batch_obj = eb_get_batch(eb); |
Line 1352... | Line 1445... | ||
1352 | 1445 | ||
1353 | /* Move the objects en-masse into the GTT, evicting if necessary. */ |
1446 | /* Move the objects en-masse into the GTT, evicting if necessary. */ |
1354 | need_relocs = (args->flags & I915_EXEC_NO_RELOC) == 0; |
1447 | need_relocs = (args->flags & I915_EXEC_NO_RELOC) == 0; |
1355 | ret = i915_gem_execbuffer_reserve(ring, &eb->vmas, &need_relocs); |
1448 | ret = i915_gem_execbuffer_reserve(ring, &eb->vmas, ctx, &need_relocs); |
1356 | if (ret) |
1449 | if (ret) |
Line 1357... | Line 1450... | ||
1357 | goto err; |
1450 | goto err; |
1358 | 1451 | ||
1359 | /* The objects are in their final locations, apply the relocations. */ |
1452 | /* The objects are in their final locations, apply the relocations. */ |
1360 | if (need_relocs) |
1453 | if (need_relocs) |
1361 | ret = i915_gem_execbuffer_relocate(eb); |
1454 | ret = i915_gem_execbuffer_relocate(eb); |
1362 | if (ret) { |
1455 | if (ret) { |
1363 | if (ret == -EFAULT) { |
1456 | if (ret == -EFAULT) { |
1364 | ret = i915_gem_execbuffer_relocate_slow(dev, args, file, ring, |
1457 | ret = i915_gem_execbuffer_relocate_slow(dev, args, file, ring, |
1365 | eb, exec); |
1458 | eb, exec, ctx); |
1366 | BUG_ON(!mutex_is_locked(&dev->struct_mutex)); |
1459 | BUG_ON(!mutex_is_locked(&dev->struct_mutex)); |
1367 | } |
1460 | } |
1368 | if (ret) |
1461 | if (ret) |
Line 1373... | Line 1466... | ||
1373 | if (batch_obj->base.pending_write_domain) { |
1466 | if (batch_obj->base.pending_write_domain) { |
1374 | DRM_DEBUG("Attempting to use self-modifying batch buffer\n"); |
1467 | DRM_DEBUG("Attempting to use self-modifying batch buffer\n"); |
1375 | ret = -EINVAL; |
1468 | ret = -EINVAL; |
1376 | goto err; |
1469 | goto err; |
1377 | } |
1470 | } |
- | 1471 | ||
1378 | batch_obj->base.pending_read_domains |= I915_GEM_DOMAIN_COMMAND; |
1472 | params->args_batch_start_offset = args->batch_start_offset; |
- | 1473 | ||
1379 | #if 0 |
1474 | #if 0 |
1380 | if (i915_needs_cmd_parser(ring)) { |
1475 | if (i915_needs_cmd_parser(ring) && args->batch_len) { |
- | 1476 | struct drm_i915_gem_object *parsed_batch_obj; |
|
- | 1477 | ||
1381 | ret = i915_parse_cmds(ring, |
1478 | parsed_batch_obj = i915_gem_execbuffer_parse(ring, |
- | 1479 | &shadow_exec_entry, |
|
- | 1480 | eb, |
|
1382 | batch_obj, |
1481 | batch_obj, |
1383 | args->batch_start_offset, |
1482 | args->batch_start_offset, |
- | 1483 | args->batch_len, |
|
1384 | file->is_master); |
1484 | file->is_master); |
1385 | if (ret) { |
1485 | if (IS_ERR(parsed_batch_obj)) { |
1386 | if (ret != -EACCES) |
1486 | ret = PTR_ERR(parsed_batch_obj); |
1387 | goto err; |
1487 | goto err; |
1388 | } else { |
1488 | } |
- | 1489 | ||
1389 | /* |
1490 | /* |
1390 | * XXX: Actually do this when enabling batch copy... |
1491 | * parsed_batch_obj == batch_obj means batch not fully parsed: |
- | 1492 | * Accept, but don't promote to secure. |
|
- | 1493 | */ |
|
- | 1494 | ||
- | 1495 | if (parsed_batch_obj != batch_obj) { |
|
- | 1496 | /* |
|
- | 1497 | * Batch parsed and accepted: |
|
1391 | * |
1498 | * |
1392 | * Set the DISPATCH_SECURE bit to remove the NON_SECURE bit |
1499 | * Set the DISPATCH_SECURE bit to remove the NON_SECURE |
1393 | * from MI_BATCH_BUFFER_START commands issued in the |
1500 | * bit from MI_BATCH_BUFFER_START commands issued in |
1394 | * dispatch_execbuffer implementations. We specifically don't |
1501 | * the dispatch_execbuffer implementations. We |
- | 1502 | * specifically don't want that set on batches the |
|
1395 | * want that set when the command parser is enabled. |
1503 | * command parser has accepted. |
1396 | */ |
1504 | */ |
- | 1505 | dispatch_flags |= I915_DISPATCH_SECURE; |
|
- | 1506 | params->args_batch_start_offset = 0; |
|
- | 1507 | batch_obj = parsed_batch_obj; |
|
1397 | } |
1508 | } |
1398 | } |
1509 | } |
1399 | - | ||
1400 | #endif |
1510 | #endif |
- | 1511 | ||
- | 1512 | batch_obj->base.pending_read_domains |= I915_GEM_DOMAIN_COMMAND; |
|
- | 1513 | ||
1401 | /* snb/ivb/vlv conflate the "batch in ppgtt" bit with the "non-secure |
1514 | /* snb/ivb/vlv conflate the "batch in ppgtt" bit with the "non-secure |
1402 | * batch" bit. Hence we need to pin secure batches into the global gtt. |
1515 | * batch" bit. Hence we need to pin secure batches into the global gtt. |
1403 | * hsw should have this fixed, but bdw mucks it up again. */ |
1516 | * hsw should have this fixed, but bdw mucks it up again. */ |
1404 | if (flags & I915_DISPATCH_SECURE) { |
1517 | if (dispatch_flags & I915_DISPATCH_SECURE) { |
1405 | /* |
1518 | /* |
1406 | * So on first glance it looks freaky that we pin the batch here |
1519 | * So on first glance it looks freaky that we pin the batch here |
1407 | * outside of the reservation loop. But: |
1520 | * outside of the reservation loop. But: |
1408 | * - The batch is already pinned into the relevant ppgtt, so we |
1521 | * - The batch is already pinned into the relevant ppgtt, so we |
1409 | * already have the backing storage fully allocated. |
1522 | * already have the backing storage fully allocated. |
1410 | * - No other BO uses the global gtt (well contexts, but meh), |
1523 | * - No other BO uses the global gtt (well contexts, but meh), |
1411 | * so we don't really have issues with mutliple objects not |
1524 | * so we don't really have issues with multiple objects not |
1412 | * fitting due to fragmentation. |
1525 | * fitting due to fragmentation. |
1413 | * So this is actually safe. |
1526 | * So this is actually safe. |
1414 | */ |
1527 | */ |
1415 | ret = i915_gem_obj_ggtt_pin(batch_obj, 0, 0); |
1528 | ret = i915_gem_obj_ggtt_pin(batch_obj, 0, 0); |
1416 | if (ret) |
1529 | if (ret) |
1417 | goto err; |
1530 | goto err; |
Line 1418... | Line 1531... | ||
1418 | 1531 | ||
1419 | exec_start += i915_gem_obj_ggtt_offset(batch_obj); |
1532 | params->batch_obj_vm_offset = i915_gem_obj_ggtt_offset(batch_obj); |
1420 | } else |
1533 | } else |
- | 1534 | params->batch_obj_vm_offset = i915_gem_obj_offset(batch_obj, vm); |
|
- | 1535 | ||
- | 1536 | /* Allocate a request for this batch buffer nice and early. */ |
|
- | 1537 | ret = i915_gem_request_alloc(ring, ctx, ¶ms->request); |
|
- | 1538 | if (ret) |
|
- | 1539 | goto err_batch_unpin; |
|
- | 1540 | ||
- | 1541 | ret = i915_gem_request_add_to_client(params->request, file); |
|
- | 1542 | if (ret) |
|
- | 1543 | goto err_batch_unpin; |
|
- | 1544 | ||
- | 1545 | /* |
|
- | 1546 | * Save assorted stuff away to pass through to *_submission(). |
|
- | 1547 | * NB: This data should be 'persistent' and not local as it will |
|
- | 1548 | * kept around beyond the duration of the IOCTL once the GPU |
|
- | 1549 | * scheduler arrives. |
|
- | 1550 | */ |
|
- | 1551 | params->dev = dev; |
|
- | 1552 | params->file = file; |
|
- | 1553 | params->ring = ring; |
|
- | 1554 | params->dispatch_flags = dispatch_flags; |
|
- | 1555 | params->batch_obj = batch_obj; |
|
Line 1421... | Line 1556... | ||
1421 | exec_start += i915_gem_obj_offset(batch_obj, vm); |
1556 | params->ctx = ctx; |
1422 | - | ||
Line -... | Line 1557... | ||
- | 1557 | ||
1423 | ret = dev_priv->gt.do_execbuf(dev, file, ring, ctx, args, |
1558 | ret = dev_priv->gt.execbuf_submit(params, args, &eb->vmas); |
1424 | &eb->vmas, batch_obj, exec_start, flags); |
1559 | |
1425 | 1560 | err_batch_unpin: |
|
1426 | /* |
1561 | /* |
1427 | * FIXME: We crucially rely upon the active tracking for the (ppgtt) |
1562 | * FIXME: We crucially rely upon the active tracking for the (ppgtt) |
1428 | * batch vma for correctness. For less ugly and less fragility this |
1563 | * batch vma for correctness. For less ugly and less fragility this |
1429 | * needs to be adjusted to also track the ggtt batch vma properly as |
1564 | * needs to be adjusted to also track the ggtt batch vma properly as |
1430 | * active. |
1565 | * active. |
- | 1566 | */ |
|
1431 | */ |
1567 | if (dispatch_flags & I915_DISPATCH_SECURE) |
1432 | if (flags & I915_DISPATCH_SECURE) |
1568 | i915_gem_object_ggtt_unpin(batch_obj); |
1433 | i915_gem_object_ggtt_unpin(batch_obj); |
1569 | |
1434 | err: |
1570 | err: |
Line -... | Line 1571... | ||
- | 1571 | /* the request owns the ref now */ |
|
- | 1572 | i915_gem_context_unreference(ctx); |
|
- | 1573 | eb_destroy(eb); |
|
- | 1574 | ||
- | 1575 | /* |
|
- | 1576 | * If the request was created but not successfully submitted then it |
|
- | 1577 | * must be freed again. If it was submitted then it is being tracked |
|
- | 1578 | * on the active request list and no clean up is required here. |
|
1435 | /* the request owns the ref now */ |
1579 | */ |
Line 1436... | Line 1580... | ||
1436 | i915_gem_context_unreference(ctx); |
1580 | if (ret && params->request) |
1437 | eb_destroy(eb); |
1581 | i915_gem_request_cancel(params->request); |
1438 | 1582 |