- uint64_t aligned_pos = cursor.start & ~(uint64_t)3;
- uint64_t bytes = 4 - (cursor.start & 3);
- uint32_t shift = (cursor.start & 3) * 8;
- uint32_t mask = 0xffffffff << shift;
-
- if (cursor.size < bytes) {
- mask &= 0xffffffff >> (bytes - cursor.size) * 8;
- bytes = cursor.size;
- }
-
- if (mask != 0xffffffff) {
- spin_lock_irqsave(&adev->mmio_idx_lock, flags);
- WREG32_NO_KIQ(mmMM_INDEX, ((uint32_t)aligned_pos) | 0x80000000);
- WREG32_NO_KIQ(mmMM_INDEX_HI, aligned_pos >> 31);
- value = RREG32_NO_KIQ(mmMM_DATA);
- if (write) {
- value &= ~mask;
- value |= (*(uint32_t *)buf << shift) & mask;
- WREG32_NO_KIQ(mmMM_DATA, value);
- }
- spin_unlock_irqrestore(&adev->mmio_idx_lock, flags);
- if (!write) {
- value = (value & mask) >> shift;
- memcpy(buf, &value, bytes);
- }
- } else {
- bytes = cursor.size & ~0x3ULL;
- amdgpu_device_vram_access(adev, cursor.start,
- (uint32_t *)buf, bytes,
- write);
+ size_t count, size = cursor.size;
+ loff_t pos = cursor.start;
+
+ count = amdgpu_device_aper_access(adev, pos, buf, size, write);
+ size -= count;
+ if (size) {
+ /* using MM to access rest vram and handle un-aligned address */
+ pos += count;
+ buf += count;
+ amdgpu_ttm_vram_mm_access(adev, pos, buf, size, write);