tree: https://git.kernel.org/pub/scm/linux/kernel/git/geert/renesas-drivers.git master head: 051b9eadcc17b76b87726d74d18d1b741837adcc commit: 64207500b99223e7ba854a4ab5d170a542fc0375 [16/58] Merge remote-tracking branch 'drm/drm-next' into renesas-drivers config: loongarch-allmodconfig (https://download.01.org/0day-ci/archive/20250521/202505210600.iPS7fzOk-lkp@xxxxxxxxx/config) compiler: loongarch64-linux-gcc (GCC) 14.2.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20250521/202505210600.iPS7fzOk-lkp@xxxxxxxxx/reproduce) If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <lkp@xxxxxxxxx> | Closes: https://lore.kernel.org/oe-kbuild-all/202505210600.iPS7fzOk-lkp@xxxxxxxxx/ All errors (new ones prefixed by >>): drivers/gpu/drm/xe/xe_svm.c: In function 'xe_svm_find_vma_start': >> drivers/gpu/drm/xe/xe_svm.c:712:16: error: implicit declaration of function 'drm_gpusvm_find_vma_start'; did you mean 'xe_svm_find_vma_start'? [-Wimplicit-function-declaration] 712 | return drm_gpusvm_find_vma_start(&vm->svm.gpusvm, | ^~~~~~~~~~~~~~~~~~~~~~~~~ | xe_svm_find_vma_start drivers/gpu/drm/xe/xe_svm.c: In function 'xe_svm_handle_pagefault': >> drivers/gpu/drm/xe/xe_svm.c:858:31: error: 'struct xe_device' has no member named 'atomic_svm_timeslice_ms' 858 | vm->xe->atomic_svm_timeslice_ms : 0, | ^~ vim +712 drivers/gpu/drm/xe/xe_svm.c 694 695 /** 696 * xe_svm_find_vma_start - Find start of CPU VMA 697 * @vm: xe_vm pointer 698 * @start: start address 699 * @end: end address 700 * @vma: Pointer to struct xe_vma 701 * 702 * 703 * This function searches for a cpu vma, within the specified 704 * range [start, end] in the given VM. It adjusts the range based on the 705 * xe_vma start and end addresses. If no cpu VMA is found, it returns ULONG_MAX. 706 * 707 * Return: The starting address of the VMA within the range, 708 * or ULONG_MAX if no VMA is found 709 */ 710 u64 xe_svm_find_vma_start(struct xe_vm *vm, u64 start, u64 end, struct xe_vma *vma) 711 { > 712 return drm_gpusvm_find_vma_start(&vm->svm.gpusvm, 713 max(start, xe_vma_start(vma)), 714 min(end, xe_vma_end(vma))); 715 } 716 717 #if IS_ENABLED(CONFIG_DRM_XE_DEVMEM_MIRROR) 718 static struct xe_vram_region *tile_to_vr(struct xe_tile *tile) 719 { 720 return &tile->mem.vram; 721 } 722 723 /** 724 * xe_svm_alloc_vram()- Allocate device memory pages for range, 725 * migrating existing data. 726 * @vm: The VM. 727 * @tile: tile to allocate vram from 728 * @range: SVM range 729 * @ctx: DRM GPU SVM context 730 * 731 * Return: 0 on success, error code on failure. 732 */ 733 int xe_svm_alloc_vram(struct xe_vm *vm, struct xe_tile *tile, 734 struct xe_svm_range *range, 735 const struct drm_gpusvm_ctx *ctx) 736 { 737 struct mm_struct *mm = vm->svm.gpusvm.mm; 738 struct xe_vram_region *vr = tile_to_vr(tile); 739 struct drm_buddy_block *block; 740 struct list_head *blocks; 741 struct xe_bo *bo; 742 ktime_t end = 0; 743 int err; 744 745 range_debug(range, "ALLOCATE VRAM"); 746 747 if (!mmget_not_zero(mm)) 748 return -EFAULT; 749 mmap_read_lock(mm); 750 751 retry: 752 bo = xe_bo_create_locked(tile_to_xe(tile), NULL, NULL, 753 xe_svm_range_size(range), 754 ttm_bo_type_device, 755 XE_BO_FLAG_VRAM_IF_DGFX(tile) | 756 XE_BO_FLAG_CPU_ADDR_MIRROR); 757 if (IS_ERR(bo)) { 758 err = PTR_ERR(bo); 759 if (xe_vm_validate_should_retry(NULL, err, &end)) 760 goto retry; 761 goto unlock; 762 } 763 764 drm_gpusvm_devmem_init(&bo->devmem_allocation, 765 vm->xe->drm.dev, mm, 766 &gpusvm_devmem_ops, 767 &tile->mem.vram.dpagemap, 768 xe_svm_range_size(range)); 769 770 blocks = &to_xe_ttm_vram_mgr_resource(bo->ttm.resource)->blocks; 771 list_for_each_entry(block, blocks, link) 772 block->private = vr; 773 774 xe_bo_get(bo); 775 err = drm_gpusvm_migrate_to_devmem(&vm->svm.gpusvm, &range->base, 776 &bo->devmem_allocation, ctx); 777 if (err) 778 xe_svm_devmem_release(&bo->devmem_allocation); 779 780 xe_bo_unlock(bo); 781 xe_bo_put(bo); 782 783 unlock: 784 mmap_read_unlock(mm); 785 mmput(mm); 786 787 return err; 788 } 789 #endif 790 791 static bool supports_4K_migration(struct xe_device *xe) 792 { 793 if (xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K) 794 return false; 795 796 return true; 797 } 798 799 /** 800 * xe_svm_range_needs_migrate_to_vram() - SVM range needs migrate to VRAM or not 801 * @range: SVM range for which migration needs to be decided 802 * @vma: vma which has range 803 * @preferred_region_is_vram: preferred region for range is vram 804 * 805 * Return: True for range needing migration and migration is supported else false 806 */ 807 bool xe_svm_range_needs_migrate_to_vram(struct xe_svm_range *range, struct xe_vma *vma, 808 bool preferred_region_is_vram) 809 { 810 struct xe_vm *vm = range_to_vm(&range->base); 811 u64 range_size = xe_svm_range_size(range); 812 813 if (!range->base.flags.migrate_devmem || !preferred_region_is_vram) 814 return false; 815 816 xe_assert(vm->xe, IS_DGFX(vm->xe)); 817 818 if (preferred_region_is_vram && xe_svm_range_in_vram(range)) { 819 drm_info(&vm->xe->drm, "Range is already in VRAM\n"); 820 return false; 821 } 822 823 if (preferred_region_is_vram && range_size <= SZ_64K && !supports_4K_migration(vm->xe)) { 824 drm_warn(&vm->xe->drm, "Platform doesn't support SZ_4K range migration\n"); 825 return false; 826 } 827 828 return true; 829 } 830 831 /** 832 * xe_svm_handle_pagefault() - SVM handle page fault 833 * @vm: The VM. 834 * @vma: The CPU address mirror VMA. 835 * @gt: The gt upon the fault occurred. 836 * @fault_addr: The GPU fault address. 837 * @atomic: The fault atomic access bit. 838 * 839 * Create GPU bindings for a SVM page fault. Optionally migrate to device 840 * memory. 841 * 842 * Return: 0 on success, negative error code on error. 843 */ 844 int xe_svm_handle_pagefault(struct xe_vm *vm, struct xe_vma *vma, 845 struct xe_gt *gt, u64 fault_addr, 846 bool atomic) 847 { 848 struct drm_gpusvm_ctx ctx = { 849 .read_only = xe_vma_read_only(vma), 850 .devmem_possible = IS_DGFX(vm->xe) && 851 IS_ENABLED(CONFIG_DRM_XE_DEVMEM_MIRROR), 852 .check_pages_threshold = IS_DGFX(vm->xe) && 853 IS_ENABLED(CONFIG_DRM_XE_DEVMEM_MIRROR) ? SZ_64K : 0, 854 .devmem_only = atomic && IS_DGFX(vm->xe) && 855 IS_ENABLED(CONFIG_DRM_XE_DEVMEM_MIRROR), 856 .timeslice_ms = atomic && IS_DGFX(vm->xe) && 857 IS_ENABLED(CONFIG_DRM_XE_DEVMEM_MIRROR) ? > 858 vm->xe->atomic_svm_timeslice_ms : 0, 859 }; 860 struct xe_svm_range *range; 861 struct drm_exec exec; 862 struct dma_fence *fence; 863 struct xe_tile *tile = gt_to_tile(gt); 864 int migrate_try_count = ctx.devmem_only ? 3 : 1; 865 ktime_t end = 0; 866 int err; 867 868 lockdep_assert_held_write(&vm->lock); 869 xe_assert(vm->xe, xe_vma_is_cpu_addr_mirror(vma)); 870 871 xe_gt_stats_incr(gt, XE_GT_STATS_ID_SVM_PAGEFAULT_COUNT, 1); 872 873 retry: 874 /* Always process UNMAPs first so view SVM ranges is current */ 875 err = xe_svm_garbage_collector(vm); 876 if (err) 877 return err; 878 879 range = xe_svm_range_find_or_insert(vm, fault_addr, vma, &ctx); 880 881 if (IS_ERR(range)) 882 return PTR_ERR(range); 883 884 if (ctx.devmem_only && !range->base.flags.migrate_devmem) 885 return -EACCES; 886 887 if (xe_svm_range_is_valid(range, tile, ctx.devmem_only)) 888 return 0; 889 890 range_debug(range, "PAGE FAULT"); 891 892 if (--migrate_try_count >= 0 && 893 xe_svm_range_needs_migrate_to_vram(range, vma, IS_DGFX(vm->xe))) { 894 err = xe_svm_alloc_vram(vm, tile, range, &ctx); 895 ctx.timeslice_ms <<= 1; /* Double timeslice if we have to retry */ 896 if (err) { 897 if (migrate_try_count || !ctx.devmem_only) { 898 drm_dbg(&vm->xe->drm, 899 "VRAM allocation failed, falling back to retrying fault, asid=%u, errno=%pe\n", 900 vm->usm.asid, ERR_PTR(err)); 901 goto retry; 902 } else { 903 drm_err(&vm->xe->drm, 904 "VRAM allocation failed, retry count exceeded, asid=%u, errno=%pe\n", 905 vm->usm.asid, ERR_PTR(err)); 906 return err; 907 } 908 } 909 } 910 911 range_debug(range, "GET PAGES"); 912 err = xe_svm_range_get_pages(vm, range, &ctx); 913 /* Corner where CPU mappings have changed */ 914 if (err == -EOPNOTSUPP || err == -EFAULT || err == -EPERM) { 915 ctx.timeslice_ms <<= 1; /* Double timeslice if we have to retry */ 916 if (migrate_try_count > 0 || !ctx.devmem_only) { 917 if (err == -EOPNOTSUPP) { 918 range_debug(range, "PAGE FAULT - EVICT PAGES"); 919 drm_gpusvm_range_evict(&vm->svm.gpusvm, 920 &range->base); 921 } 922 drm_dbg(&vm->xe->drm, 923 "Get pages failed, falling back to retrying, asid=%u, gpusvm=%p, errno=%pe\n", 924 vm->usm.asid, &vm->svm.gpusvm, ERR_PTR(err)); 925 range_debug(range, "PAGE FAULT - RETRY PAGES"); 926 goto retry; 927 } else { 928 drm_err(&vm->xe->drm, 929 "Get pages failed, retry count exceeded, asid=%u, gpusvm=%p, errno=%pe\n", 930 vm->usm.asid, &vm->svm.gpusvm, ERR_PTR(err)); 931 } 932 } 933 if (err) { 934 range_debug(range, "PAGE FAULT - FAIL PAGE COLLECT"); 935 goto err_out; 936 } 937 938 range_debug(range, "PAGE FAULT - BIND"); 939 940 retry_bind: 941 drm_exec_init(&exec, 0, 0); 942 drm_exec_until_all_locked(&exec) { 943 err = drm_exec_lock_obj(&exec, vm->gpuvm.r_obj); 944 drm_exec_retry_on_contention(&exec); 945 if (err) { 946 drm_exec_fini(&exec); 947 goto err_out; 948 } 949 950 fence = xe_vm_range_rebind(vm, vma, range, BIT(tile->id)); 951 if (IS_ERR(fence)) { 952 drm_exec_fini(&exec); 953 err = PTR_ERR(fence); 954 if (err == -EAGAIN) { 955 ctx.timeslice_ms <<= 1; /* Double timeslice if we have to retry */ 956 range_debug(range, "PAGE FAULT - RETRY BIND"); 957 goto retry; 958 } 959 if (xe_vm_validate_should_retry(&exec, err, &end)) 960 goto retry_bind; 961 goto err_out; 962 } 963 } 964 drm_exec_fini(&exec); 965 966 dma_fence_wait(fence, false); 967 dma_fence_put(fence); 968 969 err_out: 970 971 return err; 972 } 973 -- 0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki
