*
* @adev: amdgpu_device pointer
* @exclusive: fence we need to sync to
- * @gtt_flags: flags as they are used for GTT
* @pages_addr: DMA addresses to use for mapping
* @vm: requested vm
* @mapping: mapped range and flags to use for the update
*/
static int amdgpu_vm_bo_split_mapping(struct amdgpu_device *adev,
struct dma_fence *exclusive,
- uint64_t gtt_flags,
dma_addr_t *pages_addr,
struct amdgpu_vm *vm,
struct amdgpu_bo_va_mapping *mapping,
}
if (pages_addr) {
- if (flags == gtt_flags)
- src = adev->gart.table_addr +
- (addr >> AMDGPU_GPU_PAGE_SHIFT) * 8;
- else
- max_entries = min(max_entries, 16ull * 1024ull);
+ max_entries = min(max_entries, 16ull * 1024ull);
addr = 0;
} else if (flags & AMDGPU_PTE_VALID) {
addr += adev->vm_manager.vram_base_offset;
struct amdgpu_vm *vm = bo_va->base.vm;
struct amdgpu_bo_va_mapping *mapping;
dma_addr_t *pages_addr = NULL;
- uint64_t gtt_flags, flags;
struct ttm_mem_reg *mem;
struct drm_mm_node *nodes;
struct dma_fence *exclusive;
+ uint64_t flags;
int r;
if (clear || !bo_va->base.bo) {
exclusive = reservation_object_get_excl(bo->tbo.resv);
}
- if (bo) {
+ if (bo)
flags = amdgpu_ttm_tt_pte_flags(adev, bo->tbo.ttm, mem);
- gtt_flags = (amdgpu_ttm_is_bound(bo->tbo.ttm) &&
- adev == amdgpu_ttm_adev(bo->tbo.bdev)) ?
- flags : 0;
- } else {
+ else
flags = 0x0;
- gtt_flags = ~0x0;
- }
spin_lock(&vm->status_lock);
if (!list_empty(&bo_va->base.vm_status))
spin_unlock(&vm->status_lock);
list_for_each_entry(mapping, &bo_va->invalids, list) {
- r = amdgpu_vm_bo_split_mapping(adev, exclusive,
- gtt_flags, pages_addr, vm,
+ r = amdgpu_vm_bo_split_mapping(adev, exclusive, pages_addr, vm,
mapping, flags, nodes,
&bo_va->last_pt_update);
if (r)