drm/amdgpu/ttm: optimize vram access in amdgpu_ttm_access_memory()

1. using vram aper to access vram if possible
2. avoid MM_INDEX/MM_DATA is not working when mmio protect feature is
enabled.
Signed-off-by: Kevin Wang <kevin1.wang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c

@@ -1393,6 +1393,41 @@ static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
 	return ttm_bo_eviction_valuable(bo, place);
 }
 
+static void amdgpu_ttm_vram_mm_access(struct amdgpu_device *adev, loff_t pos,
+				      void *buf, size_t size, bool write)
+{
+	while (size) {
+		uint64_t aligned_pos = ALIGN_DOWN(pos, 4);
+		uint64_t bytes = 4 - (pos & 0x3);
+		uint32_t shift = (pos & 0x3) * 8;
+		uint32_t mask = 0xffffffff << shift;
+		uint32_t value = 0;
+
+		if (size < bytes) {
+			mask &= 0xffffffff >> (bytes - size) * 8;
+			bytes = size;
+		}
+
+		if (mask != 0xffffffff) {
+			amdgpu_device_mm_access(adev, aligned_pos, &value, 4, false);
+			if (write) {
+				value &= ~mask;
+				value |= (*(uint32_t *)buf << shift) & mask;
+				amdgpu_device_mm_access(adev, aligned_pos, &value, 4, true);
+			} else {
+				value = (value & mask) >> shift;
+				memcpy(buf, &value, bytes);
+			}
+		} else {
+			amdgpu_device_mm_access(adev, aligned_pos, buf, 4, write);
+		}
+
+		pos += bytes;
+		buf += bytes;
+		size -= bytes;
+	}
+}
+
 /**
  * amdgpu_ttm_access_memory - Read or Write memory that backs a buffer object.
  *
@@ -1412,8 +1447,6 @@ static int amdgpu_ttm_access_memory(struct ttm_buffer_object *bo,
 	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
 	struct amdgpu_device *adev = amdgpu_ttm_adev(abo->tbo.bdev);
 	struct amdgpu_res_cursor cursor;
-	unsigned long flags;
-	uint32_t value = 0;
 	int ret = 0;
 
 	if (bo->resource->mem_type != TTM_PL_VRAM)
@@ -1421,41 +1454,21 @@ static int amdgpu_ttm_access_memory(struct ttm_buffer_object *bo,
 
 	amdgpu_res_first(bo->resource, offset, len, &cursor);
 	while (cursor.remaining) {
-		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;
+		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);
 		}
 
-		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);
-		}
-
-		ret += bytes;
-		buf = (uint8_t *)buf + bytes;
-		amdgpu_res_next(&cursor, bytes);
+		ret += cursor.size;
+		buf += cursor.size;
+		amdgpu_res_next(&cursor, cursor.size);
 	}
 
 	return ret;