diff --git a/vhost-device-gpu/README.md b/vhost-device-gpu/README.md
index a049b30..72e925a 100644
--- a/vhost-device-gpu/README.md
+++ b/vhost-device-gpu/README.md
@@ -18,8 +18,8 @@ A virtio-gpu device using the vhost-user protocol.
 
   -g, --gpu-mode <GPU_MODE>
           The mode specifies which backend implementation to use
-          
-          [possible values: virglrenderer, gfxstream]
+
+          [possible values: virglrenderer, gfxstream, null]
 
   -c, --capset <CAPSET>
           Comma separated list of enabled capsets
@@ -63,8 +63,9 @@ A virtio-gpu device using the vhost-user protocol.
           Print version
 ```
 
-_NOTE_: Option `-g, --gpu-mode` can only accept the `gfxstream` value if the
-crate has been built with the `backend-gfxstream` feature, which is the default.
+_NOTE_: Option `-g, --gpu-mode` can only accept the `virglrenderer` or `gfxstream`
+values if the crate has been built with the `backend-virgl` or `backend-gfxstream`
+features respectively (both are enabled by default). The `null` mode is always available.
 
 ## Limitations
 
@@ -87,7 +88,7 @@ Because blob resources are not yet supported, some capsets are limited:
 - gfxstream-vulkan and gfxstream-gles support are exposed, but can practically only be used for display output, there is no hardware acceleration yet.
 ## Features
 
-This crate supports two GPU backends: gfxstream (default) and virglrenderer.
+This crate supports three GPU backends: virglrenderer, gfxstream (both enabled by default), and null.
 
 The **virglrenderer** backend uses the [virglrenderer-rs](https://crates.io/crates/virglrenderer-rs)
 crate, which provides Rust bindings to the native virglrenderer library. It translates
@@ -98,24 +99,36 @@ The **gfxstream** backend leverages the [rutabaga_gfx](https://crates.io/crates/
 crate. With gfxstream rendering mode, GLES and Vulkan calls are forwarded to the host
 with minimal modification.
 
+The **null** backend is a no-op implementation that accepts all GPU commands but performs
+no actual rendering. This backend is primarily intended for testing and CI purposes.
+
 Install the development packages for your distro, then build with:
 
 ```session
 $ cargo build
 ```
 
-gfxstream support is compiled by default, it can be disabled by not building with the `backend-gfxstream` feature flag, for example:
+Both virglrenderer and gfxstream support are compiled by default. The null backend is
+always available. To build with specific backends:
 
 ```session
+# Build with only virglrenderer (and null)
+$ cargo build --no-default-features --features backend-virgl
+
+# Build with only gfxstream (and null)
+$ cargo build --no-default-features --features backend-gfxstream
+
+# Build with null backend only (for testing)
 $ cargo build --no-default-features
 ```
 
 ## Examples
 
-First start the daemon on the host machine using either of the 2 gpu modes:
+First start the daemon on the host machine using one of the available gpu modes:
 
 1) `virglrenderer` (if the crate has been compiled with the feature `backend-virgl`)
 2) `gfxstream` (if the crate has been compiled with the feature `backend-gfxstream`)
+3) `null` (always available, for testing)
 
 ```shell
 host# vhost-device-gpu --socket-path /tmp/gpu.socket --gpu-mode virglrenderer
diff --git a/vhost-device-gpu/src/backend/mod.rs b/vhost-device-gpu/src/backend/mod.rs
index c3fe127..1927fd1 100644
--- a/vhost-device-gpu/src/backend/mod.rs
+++ b/vhost-device-gpu/src/backend/mod.rs
@@ -2,8 +2,10 @@
 //
 // SPDX-License-Identifier: Apache-2.0 or BSD-3-Clause
 
+#[cfg(any(feature = "backend-virgl", feature = "backend-gfxstream"))]
 mod common;
 #[cfg(feature = "backend-gfxstream")]
 pub mod gfxstream;
+pub mod null;
 #[cfg(feature = "backend-virgl")]
 pub mod virgl;
diff --git a/vhost-device-gpu/src/backend/null.rs b/vhost-device-gpu/src/backend/null.rs
new file mode 100644
index 0000000..6be1b3e
--- /dev/null
+++ b/vhost-device-gpu/src/backend/null.rs
@@ -0,0 +1,463 @@
+// Null backend
+// Copyright 2025 Red Hat Inc
+//
+// SPDX-License-Identifier: Apache-2.0 or BSD-3-Clause
+
+use log::trace;
+use rutabaga_gfx::RutabagaFence;
+use vhost::vhost_user::{
+    gpu_message::{VhostUserGpuCursorPos, VhostUserGpuEdidRequest},
+    GpuBackend,
+};
+use vm_memory::{GuestAddress, GuestMemoryMmap, VolatileSlice};
+use vmm_sys_util::eventfd::EventFd;
+
+use crate::{
+    gpu_types::{ResourceCreate3d, Transfer3DDesc, VirtioGpuRing},
+    protocol::{virtio_gpu_rect, GpuResponse, VirtioGpuResult},
+    renderer::Renderer,
+    GpuConfig,
+};
+
+pub struct NullAdapter {
+    _gpu_backend: GpuBackend,
+}
+
+impl NullAdapter {
+    pub fn new(
+        _queue_ctl: &vhost_user_backend::VringRwLock,
+        _config: &GpuConfig,
+        gpu_backend: GpuBackend,
+    ) -> Self {
+        trace!("NullAdapter created");
+        Self {
+            _gpu_backend: gpu_backend,
+        }
+    }
+}
+
+impl Renderer for NullAdapter {
+    fn resource_create_3d(
+        &mut self,
+        _resource_id: u32,
+        _args: ResourceCreate3d,
+    ) -> VirtioGpuResult {
+        trace!("NullAdapter::resource_create_3d - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn unref_resource(&mut self, _resource_id: u32) -> VirtioGpuResult {
+        trace!("NullAdapter::unref_resource - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn transfer_write(
+        &mut self,
+        _ctx_id: u32,
+        _resource_id: u32,
+        _transfer: Transfer3DDesc,
+    ) -> VirtioGpuResult {
+        trace!("NullAdapter::transfer_write - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn transfer_write_2d(
+        &mut self,
+        _ctx_id: u32,
+        _resource_id: u32,
+        _transfer: Transfer3DDesc,
+    ) -> VirtioGpuResult {
+        trace!("NullAdapter::transfer_write_2d - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn transfer_read(
+        &mut self,
+        _ctx_id: u32,
+        _resource_id: u32,
+        _transfer: Transfer3DDesc,
+        _buf: Option<VolatileSlice>,
+    ) -> VirtioGpuResult {
+        trace!("NullAdapter::transfer_read - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn attach_backing(
+        &mut self,
+        _resource_id: u32,
+        _mem: &GuestMemoryMmap,
+        _vecs: Vec<(GuestAddress, usize)>,
+    ) -> VirtioGpuResult {
+        trace!("NullAdapter::attach_backing - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn detach_backing(&mut self, _resource_id: u32) -> VirtioGpuResult {
+        trace!("NullAdapter::detach_backing - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn update_cursor(
+        &mut self,
+        _resource_id: u32,
+        _cursor_pos: VhostUserGpuCursorPos,
+        _hot_x: u32,
+        _hot_y: u32,
+    ) -> VirtioGpuResult {
+        trace!("NullAdapter::update_cursor - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn move_cursor(
+        &mut self,
+        _resource_id: u32,
+        _cursor: VhostUserGpuCursorPos,
+    ) -> VirtioGpuResult {
+        trace!("NullAdapter::move_cursor - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn resource_assign_uuid(&self, _resource_id: u32) -> VirtioGpuResult {
+        trace!("NullAdapter::resource_assign_uuid - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn get_capset_info(&self, _capset_index: u32) -> VirtioGpuResult {
+        trace!("NullAdapter::get_capset_info - no capsets");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn get_capset(&self, _capset_id: u32, _capset_version: u32) -> VirtioGpuResult {
+        trace!("NullAdapter::get_capset - no capsets");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn create_context(
+        &mut self,
+        _ctx_id: u32,
+        _context_init: u32,
+        _context_name: Option<&str>,
+    ) -> VirtioGpuResult {
+        trace!("NullAdapter::create_context - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn destroy_context(&mut self, _ctx_id: u32) -> VirtioGpuResult {
+        trace!("NullAdapter::destroy_context - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn context_attach_resource(&mut self, _ctx_id: u32, _resource_id: u32) -> VirtioGpuResult {
+        trace!("NullAdapter::context_attach_resource - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn context_detach_resource(&mut self, _ctx_id: u32, _resource_id: u32) -> VirtioGpuResult {
+        trace!("NullAdapter::context_detach_resource - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn submit_command(
+        &mut self,
+        _ctx_id: u32,
+        _commands: &mut [u8],
+        _fence_ids: &[u64],
+    ) -> VirtioGpuResult {
+        trace!("NullAdapter::submit_command - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn create_fence(&mut self, _rutabaga_fence: RutabagaFence) -> VirtioGpuResult {
+        trace!("NullAdapter::create_fence - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn process_fence(
+        &mut self,
+        _ring: VirtioGpuRing,
+        _fence_id: u64,
+        _desc_index: u16,
+        _len: u32,
+    ) -> bool {
+        trace!("NullAdapter::process_fence - no-op");
+        true
+    }
+
+    fn get_event_poll_fd(&self) -> Option<EventFd> {
+        trace!("NullAdapter::get_event_poll_fd - no-op");
+        None
+    }
+
+    fn event_poll(&self) {
+        trace!("NullAdapter::event_poll - no-op");
+    }
+
+    fn force_ctx_0(&self) {
+        trace!("NullAdapter::force_ctx_0 - no-op");
+    }
+
+    fn display_info(&self) -> VirtioGpuResult {
+        trace!("NullAdapter::display_info - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn get_edid(&self, _edid_req: VhostUserGpuEdidRequest) -> VirtioGpuResult {
+        trace!("NullAdapter::get_edid - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn set_scanout(
+        &mut self,
+        _scanout_id: u32,
+        _resource_id: u32,
+        _rect: virtio_gpu_rect,
+    ) -> VirtioGpuResult {
+        trace!("NullAdapter::set_scanout - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn flush_resource(&mut self, _resource_id: u32, _rect: virtio_gpu_rect) -> VirtioGpuResult {
+        trace!("NullAdapter::flush_resource - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn resource_create_blob(
+        &mut self,
+        _ctx_id: u32,
+        _resource_id: u32,
+        _blob_id: u64,
+        _size: u64,
+        _blob_mem: u32,
+        _blob_flags: u32,
+    ) -> VirtioGpuResult {
+        trace!("NullAdapter::resource_create_blob - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn resource_map_blob(&mut self, _resource_id: u32, _offset: u64) -> VirtioGpuResult {
+        trace!("NullAdapter::resource_map_blob - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+
+    fn resource_unmap_blob(&mut self, _resource_id: u32) -> VirtioGpuResult {
+        trace!("NullAdapter::resource_unmap_blob - no-op");
+        Ok(GpuResponse::OkNoData)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::os::unix::net::UnixStream;
+
+    use vhost_user_backend::{VringRwLock, VringT};
+    use vm_memory::{GuestAddress, GuestMemoryAtomic, GuestMemoryMmap};
+
+    use super::*;
+    use crate::{GpuFlags, GpuMode};
+
+    fn create_null_adapter() -> NullAdapter {
+        let (_, backend) = UnixStream::pair().unwrap();
+        let gpu_backend = GpuBackend::from_stream(backend);
+        let mem = GuestMemoryAtomic::new(
+            GuestMemoryMmap::<()>::from_ranges(&[(GuestAddress(0), 0x1000)]).unwrap(),
+        );
+        let vring = VringRwLock::new(mem, 0x100).unwrap();
+        let config = GpuConfig::new(GpuMode::Null, None, GpuFlags::default()).unwrap();
+
+        NullAdapter::new(&vring, &config, gpu_backend)
+    }
+
+    #[test]
+    fn test_null_adapter_creation() {
+        // Verify that NullAdapter can be successfully created
+        let _adapter = create_null_adapter();
+    }
+
+    #[test]
+    fn test_null_adapter_resource_operations() {
+        let mut adapter = create_null_adapter();
+
+        // Verify resource creation returns success without doing anything
+        let resource_create = ResourceCreate3d {
+            target: 2,
+            format: 1,
+            bind: 1,
+            width: 640,
+            height: 480,
+            depth: 1,
+            array_size: 1,
+            last_level: 0,
+            nr_samples: 0,
+            flags: 0,
+        };
+        let result = adapter.resource_create_3d(1, resource_create);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify unreferencing a resource succeeds
+        let result = adapter.unref_resource(1);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify attaching and detaching backing memory succeeds
+        let mem = GuestMemoryMmap::<()>::from_ranges(&[(GuestAddress(0), 0x1000)]).unwrap();
+        let result = adapter.attach_backing(1, &mem, vec![]);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        let result = adapter.detach_backing(1);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+    }
+
+    #[test]
+    fn test_null_adapter_transfer_operations() {
+        let mut adapter = create_null_adapter();
+        let transfer = Transfer3DDesc::new_2d(0, 0, 640, 480, 0);
+
+        // Verify 3D transfer write succeeds
+        let result = adapter.transfer_write(0, 1, transfer);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify 2D transfer write succeeds
+        let result = adapter.transfer_write_2d(0, 1, transfer);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify transfer read succeeds
+        let result = adapter.transfer_read(0, 1, transfer, None);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+    }
+
+    #[test]
+    fn test_null_adapter_context_operations() {
+        let mut adapter = create_null_adapter();
+
+        // Verify context creation succeeds
+        let result = adapter.create_context(1, 0, Some("test"));
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify attaching a resource to a context succeeds
+        let result = adapter.context_attach_resource(1, 1);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify detaching a resource from a context succeeds
+        let result = adapter.context_detach_resource(1, 1);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify context destruction succeeds
+        let result = adapter.destroy_context(1);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+    }
+
+    #[test]
+    fn test_null_adapter_display_operations() {
+        let mut adapter = create_null_adapter();
+
+        // Verify getting display info succeeds
+        let result = adapter.display_info();
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify getting EDID info succeeds
+        let result = adapter.get_edid(VhostUserGpuEdidRequest { scanout_id: 0 });
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify setting scanout succeeds
+        let result = adapter.set_scanout(0, 1, virtio_gpu_rect::default());
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify flushing a resource succeeds
+        let result = adapter.flush_resource(1, virtio_gpu_rect::default());
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+    }
+
+    #[test]
+    fn test_null_adapter_cursor_operations() {
+        let mut adapter = create_null_adapter();
+        let cursor_pos = VhostUserGpuCursorPos {
+            scanout_id: 0,
+            x: 0,
+            y: 0,
+        };
+
+        // Verify updating cursor succeeds
+        let result = adapter.update_cursor(1, cursor_pos, 0, 0);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify moving cursor succeeds
+        let result = adapter.move_cursor(1, cursor_pos);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+    }
+
+    #[test]
+    fn test_null_adapter_capset_operations() {
+        let adapter = create_null_adapter();
+
+        // Verify getting capset info returns success (null backend has no capsets)
+        let result = adapter.get_capset_info(0);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify getting capset returns success (null backend has no capsets)
+        let result = adapter.get_capset(0, 0);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+    }
+
+    #[test]
+    fn test_null_adapter_command_operations() {
+        let mut adapter = create_null_adapter();
+        let mut commands = vec![0u8; 64];
+
+        // Verify submitting commands succeeds
+        let result = adapter.submit_command(1, &mut commands, &[]);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+    }
+
+    #[test]
+    fn test_null_adapter_fence_operations() {
+        let mut adapter = create_null_adapter();
+        let fence = RutabagaFence {
+            flags: 0,
+            fence_id: 1,
+            ctx_id: 0,
+            ring_idx: 0,
+        };
+
+        // Verify creating a fence succeeds
+        let result = adapter.create_fence(fence);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify processing fence returns true (fence is immediately ready in null
+        // backend)
+        let ready = adapter.process_fence(VirtioGpuRing::Global, 1, 0, 0);
+        assert!(ready);
+    }
+
+    #[test]
+    fn test_null_adapter_blob_operations() {
+        let mut adapter = create_null_adapter();
+
+        // Verify blob resource creation succeeds
+        let result = adapter.resource_create_blob(0, 1, 1, 4096, 0, 0);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify mapping blob resource succeeds
+        let result = adapter.resource_map_blob(1, 0);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify unmapping blob resource succeeds
+        let result = adapter.resource_unmap_blob(1);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+    }
+
+    #[test]
+    fn test_null_adapter_misc_operations() {
+        let adapter = create_null_adapter();
+
+        // Verify assigning UUID to resource succeeds
+        let result = adapter.resource_assign_uuid(1);
+        assert!(matches!(result, Ok(GpuResponse::OkNoData)));
+
+        // Verify no event poll fd is provided (null backend has no events)
+        let event_fd = adapter.get_event_poll_fd();
+        assert!(event_fd.is_none());
+
+        // Verify event polling and force_ctx_0 don't panic (they're no-ops)
+        adapter.event_poll();
+        adapter.force_ctx_0();
+    }
+}
diff --git a/vhost-device-gpu/src/device.rs b/vhost-device-gpu/src/device.rs
index 2064aac..1847310 100644
--- a/vhost-device-gpu/src/device.rs
+++ b/vhost-device-gpu/src/device.rs
@@ -77,6 +77,7 @@ use crate::backend::gfxstream::GfxstreamAdapter;
 #[cfg(feature = "backend-virgl")]
 use crate::backend::virgl::VirglRendererAdapter;
 use crate::{
+    backend::null::NullAdapter,
     gpu_types::{ResourceCreate3d, Transfer3DDesc, VirtioGpuRing},
     protocol::{
         virtio_gpu_ctrl_hdr, virtio_gpu_ctx_create, virtio_gpu_get_edid,
@@ -630,6 +631,17 @@ impl VhostUserGpuBackendInner {
                 device_event,
                 vrings
             ),
+
+            GpuMode::Null => handle_adapter!(
+                NullAdapter,
+                TLS_NULL,
+                |control_vring, gpu_backend| {
+                    NullAdapter::new(control_vring, &self.gpu_config, gpu_backend)
+                },
+                self,
+                device_event,
+                vrings
+            ),
         }
     }
 
diff --git a/vhost-device-gpu/src/gpu_types.rs b/vhost-device-gpu/src/gpu_types.rs
index 0caf067..6c9df9e 100644
--- a/vhost-device-gpu/src/gpu_types.rs
+++ b/vhost-device-gpu/src/gpu_types.rs
@@ -4,6 +4,7 @@
 
 /// Generates an implementation of `From<Transfer3DDesc>` for any compatible
 /// target struct.
+#[cfg(any(feature = "backend-virgl", feature = "backend-gfxstream"))]
 macro_rules! impl_transfer3d_from_desc {
     ($target:path) => {
         impl From<Transfer3DDesc> for $target {
@@ -25,6 +26,7 @@ macro_rules! impl_transfer3d_from_desc {
     };
 }
 
+#[cfg(any(feature = "backend-virgl", feature = "backend-gfxstream"))]
 macro_rules! impl_from_resource_create3d {
     ($target:ty) => {
         impl From<ResourceCreate3d> for $target {
@@ -48,7 +50,9 @@ macro_rules! impl_from_resource_create3d {
 
 use std::{collections::BTreeMap, os::raw::c_void};
 
+#[cfg(feature = "backend-gfxstream")]
 use rutabaga_gfx::Transfer3D;
+#[cfg(feature = "backend-virgl")]
 use virglrenderer::Transfer3D as VirglTransfer3D;
 
 use crate::protocol::virtio_gpu_rect;
@@ -87,8 +91,10 @@ impl Transfer3DDesc {
 }
 // Invoke the macro for both targets
 // rutabaga_gfx::Transfer3D
+#[cfg(feature = "backend-gfxstream")]
 impl_transfer3d_from_desc!(Transfer3D);
 // virglrenderer::Transfer3D
+#[cfg(feature = "backend-virgl")]
 impl_transfer3d_from_desc!(VirglTransfer3D);
 
 // These are neutral types that can be used by all backends
@@ -136,7 +142,9 @@ pub struct ResourceCreate3d {
 }
 
 // Invoke the macro for both targets
+#[cfg(feature = "backend-gfxstream")]
 impl_from_resource_create3d!(rutabaga_gfx::ResourceCreate3D);
+#[cfg(feature = "backend-virgl")]
 impl_from_resource_create3d!(virglrenderer::ResourceCreate3D);
 
 #[derive(Debug, Clone, Copy)]
diff --git a/vhost-device-gpu/src/lib.rs b/vhost-device-gpu/src/lib.rs
index b710f68..a50ca42 100644
--- a/vhost-device-gpu/src/lib.rs
+++ b/vhost-device-gpu/src/lib.rs
@@ -44,6 +44,7 @@ pub enum GpuMode {
     VirglRenderer,
     #[cfg(feature = "backend-gfxstream")]
     Gfxstream,
+    Null,
 }
 
 impl Display for GpuMode {
@@ -53,6 +54,7 @@ impl Display for GpuMode {
             Self::VirglRenderer => write!(f, "virglrenderer"),
             #[cfg(feature = "backend-gfxstream")]
             Self::Gfxstream => write!(f, "gfxstream"),
+            Self::Null => write!(f, "null"),
         }
     }
 }
@@ -79,7 +81,12 @@ bitflags! {
 
 impl Display for GpuCapset {
     fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
+        if self.is_empty() {
+            return write!(f, "none");
+        }
+
         let mut first = true;
+        #[allow(unused_assignments)]
         for capset in self.iter() {
             if !first {
                 write!(f, ", ")?;
@@ -88,15 +95,15 @@ impl Display for GpuCapset {
 
             match capset {
                 #[cfg(feature = "backend-virgl")]
-                Self::VIRGL => write!(f, "virgl"),
+                Self::VIRGL => write!(f, "virgl")?,
                 #[cfg(feature = "backend-virgl")]
-                Self::VIRGL2 => write!(f, "virgl2"),
+                Self::VIRGL2 => write!(f, "virgl2")?,
                 #[cfg(feature = "backend-gfxstream")]
-                Self::GFXSTREAM_VULKAN => write!(f, "gfxstream-vulkan"),
+                Self::GFXSTREAM_VULKAN => write!(f, "gfxstream-vulkan")?,
                 #[cfg(feature = "backend-gfxstream")]
-                Self::GFXSTREAM_GLES => write!(f, "gfxstream-gles"),
+                Self::GFXSTREAM_GLES => write!(f, "gfxstream-gles")?,
                 _ => panic!("Unknown capset {:#x}", self.bits()),
-            }?;
+            }
         }
 
         Ok(())
@@ -165,6 +172,7 @@ impl GpuConfig {
             GpuMode::VirglRenderer => Self::DEFAULT_VIRGLRENDER_CAPSET_MASK,
             #[cfg(feature = "backend-gfxstream")]
             GpuMode::Gfxstream => Self::DEFAULT_GFXSTREAM_CAPSET_MASK,
+            GpuMode::Null => GpuCapset::empty(),
         }
     }
 
@@ -174,6 +182,7 @@ impl GpuConfig {
             GpuMode::VirglRenderer => GpuCapset::ALL_VIRGLRENDERER_CAPSETS,
             #[cfg(feature = "backend-gfxstream")]
             GpuMode::Gfxstream => GpuCapset::ALL_GFXSTREAM_CAPSETS,
+            GpuMode::Null => GpuCapset::empty(),
         };
         for capset in capset.iter() {
             if !supported_capset_mask.contains(capset) {
diff --git a/vhost-device-gpu/src/main.rs b/vhost-device-gpu/src/main.rs
index 55ef8dd..90019c0 100644
--- a/vhost-device-gpu/src/main.rs
+++ b/vhost-device-gpu/src/main.rs
@@ -12,6 +12,9 @@ use vhost_device_gpu::{start_backend, GpuCapset, GpuConfig, GpuConfigError, GpuF
 
 #[derive(ValueEnum, Debug, Copy, Clone, Eq, PartialEq)]
 #[repr(u64)]
+// __Null is a placeholder to prevent a zero-variant enum when building with
+// --no-default-features, not an implementation of the non-exhaustive pattern
+#[allow(clippy::manual_non_exhaustive)]
 pub enum CapsetName {
     /// [virglrenderer] OpenGL implementation, superseded by Virgl2
     #[cfg(feature = "backend-virgl")]
@@ -32,10 +35,20 @@ pub enum CapsetName {
     /// hardware acceleration yet
     #[cfg(feature = "backend-gfxstream")]
     GfxstreamGles = GpuCapset::GFXSTREAM_GLES.bits(),
+
+    /// Placeholder variant to prevent zero-variant enum when no backend
+    /// features are enabled. The null backend doesn't use capsets, so this
+    /// maps to GpuCapset::empty().
+    #[doc(hidden)]
+    __Null = 0,
 }
 
 impl From<CapsetName> for GpuCapset {
     fn from(capset_name: CapsetName) -> GpuCapset {
+        if matches!(capset_name, CapsetName::__Null) {
+            return GpuCapset::empty();
+        }
+
         GpuCapset::from_bits(capset_name as u64)
             .expect("Internal error: CapsetName enum is incorrectly defined")
     }