diff --git a/cts_runner/test.lst b/cts_runner/test.lst
index 9c8a4f3180..1abb27fd4d 100644
--- a/cts_runner/test.lst
+++ b/cts_runner/test.lst
@@ -4,6 +4,7 @@
 // ```
 unittests:*
 webgpu:api,operation,buffers,createBindGroup:buffer_binding_resource:*
+webgpu:api,operation,buffers,map:mapAsync,read:*
 webgpu:api,operation,command_buffer,basic:*
 webgpu:api,operation,command_buffer,copyBufferToBuffer:*
 fails-if(vulkan) webgpu:api,operation,command_buffer,copyTextureToTexture:copy_depth_stencil:format="depth24plus"
diff --git a/tests/tests/wgpu-gpu/cloneable_types.rs b/tests/tests/wgpu-gpu/cloneable_types.rs
index fd578f6cc9..ffae024ff3 100644
--- a/tests/tests/wgpu-gpu/cloneable_types.rs
+++ b/tests/tests/wgpu-gpu/cloneable_types.rs
@@ -28,9 +28,6 @@ fn cloneable_buffers(ctx: TestingContext) {
 
     buffer.unmap();
 
-    // This is actually a bug, we should not need to call submit to make the buffer contents visible.
-    ctx.queue.submit([]);
-
     let cloned_buffer = buffer.clone();
     let cloned_buffer_contents = buffer_contents.clone();
 
diff --git a/wgpu-core/src/conv.rs b/wgpu-core/src/conv.rs
index 0003c991d2..3c1a1bf4f5 100644
--- a/wgpu-core/src/conv.rs
+++ b/wgpu-core/src/conv.rs
@@ -26,7 +26,7 @@ pub fn is_valid_external_image_copy_dst_texture_format(format: wgt::TextureForma
     }
 }
 
-pub fn map_buffer_usage(usage: wgt::BufferUsages) -> wgt::BufferUses {
+pub fn map_buffer_usage(usage: wgt::BufferUsages, mapped_at_creation: bool) -> wgt::BufferUses {
     let mut u = wgt::BufferUses::empty();
     u.set(
         wgt::BufferUses::MAP_READ,
@@ -34,7 +34,8 @@ pub fn map_buffer_usage(usage: wgt::BufferUsages) -> wgt::BufferUses {
     );
     u.set(
         wgt::BufferUses::MAP_WRITE,
-        usage.contains(wgt::BufferUsages::MAP_WRITE),
+        usage.contains(wgt::BufferUsages::MAP_WRITE)
+            || (usage.contains(wgt::BufferUsages::MAP_READ) && mapped_at_creation),
     );
     u.set(
         wgt::BufferUses::COPY_SRC,
diff --git a/wgpu-core/src/device/resource.rs b/wgpu-core/src/device/resource.rs
index 94834746c5..bb171b6b93 100644
--- a/wgpu-core/src/device/resource.rs
+++ b/wgpu-core/src/device/resource.rs
@@ -900,7 +900,7 @@ impl Device {
             }
         }
 
-        let mut usage = conv::map_buffer_usage(desc.usage);
+        let mut usage = conv::map_buffer_usage(desc.usage, desc.mapped_at_creation);
 
         if desc.usage.contains(wgt::BufferUsages::INDIRECT) {
             self.require_downlevel_flags(wgt::DownlevelFlags::INDIRECT_EXECUTION)?;
@@ -990,7 +990,10 @@ impl Device {
 
         let buffer_use = if !desc.mapped_at_creation {
             wgt::BufferUses::empty()
-        } else if desc.usage.contains(wgt::BufferUsages::MAP_WRITE) {
+        } else if desc
+            .usage
+            .intersects(wgt::BufferUsages::MAP_WRITE | wgt::BufferUsages::MAP_READ)
+        {
             // buffer is mappable, so we are just doing that at start
             let map_size = buffer.size;
             let mapping = if map_size == 0 {
diff --git a/wgpu-hal/src/gles/device.rs b/wgpu-hal/src/gles/device.rs
index e85809e604..a508de976f 100644
--- a/wgpu-hal/src/gles/device.rs
+++ b/wgpu-hal/src/gles/device.rs
@@ -526,7 +526,10 @@ impl crate::Device for super::Device {
                 .private_caps
                 .contains(PrivateCapabilities::BUFFER_ALLOCATION);
 
-        if emulate_map && desc.usage.intersects(wgt::BufferUses::MAP_WRITE) {
+        if emulate_map
+            && (desc.usage.intersects(wgt::BufferUses::MAP_WRITE)
+                && !desc.usage.intersects(wgt::BufferUses::MAP_READ))
+        {
             return Ok(super::Buffer {
                 raw: None,
                 target,
@@ -692,12 +695,21 @@ impl crate::Device for super::Device {
     }
     unsafe fn unmap_buffer(&self, buffer: &super::Buffer) {
         if let Some(raw) = buffer.raw {
-            if buffer.data.is_none() {
-                let gl = &self.shared.context.lock();
-                unsafe { gl.bind_buffer(buffer.target, Some(raw)) };
-                unsafe { gl.unmap_buffer(buffer.target) };
-                unsafe { gl.bind_buffer(buffer.target, None) };
-                *lock(&buffer.offset_of_current_mapping) = 0;
+            match &buffer.data {
+                None => {
+                    let gl = &self.shared.context.lock();
+                    unsafe { gl.bind_buffer(buffer.target, Some(raw)) };
+                    unsafe { gl.unmap_buffer(buffer.target) };
+                    unsafe { gl.bind_buffer(buffer.target, None) };
+                    *lock(&buffer.offset_of_current_mapping) = 0;
+                }
+                Some(data) => {
+                    let gl = &self.shared.context.lock();
+                    let data = lock(&*data);
+                    unsafe { gl.bind_buffer(buffer.target, Some(raw)) };
+                    unsafe { gl.buffer_sub_data_u8_slice(buffer.target, 0, &data) };
+                    unsafe { gl.bind_buffer(buffer.target, None) };
+                }
             }
         }
     }