diff --git a/wgpu-core/src/device/queue.rs b/wgpu-core/src/device/queue.rs
index 10df7aaf1..22261c13e 100644
--- a/wgpu-core/src/device/queue.rs
+++ b/wgpu-core/src/device/queue.rs
@@ -860,7 +860,6 @@ impl Queue {
                 wgt::BufferSize::new(stage_bytes_per_row as u64 * block_rows_in_copy as u64)
                     .unwrap();
             let mut staging_buffer = StagingBuffer::new(&self.device, stage_size)?;
-            let copy_bytes_per_row = stage_bytes_per_row.min(bytes_per_row) as usize;
             for layer in 0..size.depth_or_array_layers {
                 let rows_offset = layer * rows_per_image;
                 for row in rows_offset..rows_offset + height_in_blocks {
@@ -871,7 +870,7 @@ impl Queue {
                             data,
                             src_offset as isize,
                             dst_offset as isize,
-                            copy_bytes_per_row,
+                            bytes_in_last_row as usize,
                         )
                     }
                 }
diff --git a/wgpu-core/src/resource.rs b/wgpu-core/src/resource.rs
index d8e80702f..afdf73e3f 100644
--- a/wgpu-core/src/resource.rs
+++ b/wgpu-core/src/resource.rs
@@ -1076,6 +1076,13 @@ impl StagingBuffer {
         size: usize,
     ) {
         unsafe {
+            debug_assert!(
+                (src_offset + size as isize) as usize <= data.len(),
+                "src_offset + size must be in-bounds: src_offset = {}, size = {}, data.len() = {}",
+                src_offset,
+                size,
+                data.len()
+            );
             core::ptr::copy_nonoverlapping(
                 data.as_ptr().offset(src_offset),
                 self.ptr.as_ptr().offset(dst_offset),