feat(render): add output-specific DMA-BUF feedback (#53)

Co-authored-by: Samuel Onoja <[email protected]>

compositor/src/backends/direct.rs

@@ -8,6 +8,7 @@ use std::time::{Duration, Instant};
 pub(crate) mod capture;
 mod crtc_assignment;
 mod diagnostics;
+pub(crate) mod feedback;
 pub(crate) mod gamma;
 mod lid;
 mod mirror;
@@ -22,14 +23,15 @@ use diagnostics::{Diagnostics, PrimaryOutcome};
 use output_power::{OutputPower, Power};
 use redraw::{AnimationFallback, SchedulingMetrics};
 use scheduling::*;
-use smithay::backend::allocator::Fourcc;
 use smithay::backend::allocator::gbm::{GbmAllocator, GbmBufferFlags, GbmDevice};
+use smithay::backend::allocator::{Buffer, Fourcc};
 use smithay::backend::drm::compositor::{DrmCompositor, FrameFlags, PrimaryPlaneElement};
 use smithay::backend::drm::exporter::gbm::GbmFramebufferExporter;
 use smithay::backend::drm::{DrmDevice, DrmDeviceFd, DrmEvent, DrmEventTime, DrmNode, NodeType};
 use smithay::backend::egl::{EGLContext, EGLDisplay};
 use smithay::backend::input::InputEvent;
 use smithay::backend::libinput::{LibinputInputBackend, LibinputSessionInterface};
+use smithay::backend::renderer::element::{RenderElement, UnderlyingStorage};
 use smithay::backend::renderer::gles::{GlesRenderer, GlesTexture};
 use smithay::backend::renderer::utils::CommitCounter;
 use smithay::backend::renderer::{ImportDma, ImportMemWl, Renderer};
@@ -277,6 +279,7 @@ struct DirectOutput {
     capture_texture: Option<GlesTexture>,
     render_metrics: RenderMetrics,
     diagnostics: Diagnostics,
+    dmabuf_feedback: feedback::OutputFeedback,
     frame_pending: bool,
     completion: recovery::Completion,
     completion_timer: Option<RegistrationToken>,
@@ -413,6 +416,7 @@ pub fn init(event_loop: &mut EventLoop<Ferese>, state: &mut Ferese) -> Result<()
                             output.gamma.suspend();
                             cancel_output_timers(&handle, output);
                             output.surface.reset_buffers();
+                            output.dmabuf_feedback.invalidate();
                             output.primary_commit = None;
                             output.capture_texture = None;
                             output.frame_pending = false;
@@ -971,6 +975,7 @@ fn open_device(state: &mut Ferese, path: &Path) -> Result<DrmNode, Box<dyn Error
         .dmabuf_state
         .create_global_with_default_feedback::<Ferese>(&display_handle, &feedback);
     state.dmabuf_imports.register(dmabuf_global);
+    state.dmabuf_feedback.register(dmabuf_global, feedback);
     state
         .direct_backend
         .as_mut()
@@ -1334,6 +1339,22 @@ fn render_output(
                 )
                 .inspect_err(|_| output.diagnostics.render_failed())?;
 
+            let primary_format = match &result.primary_element {
+                PrimaryPlaneElement::Swapchain(primary) => Some(primary.buffer().format()),
+                PrimaryPlaneElement::Element(element) => match element.underlying_storage(&mut device.renderer) {
+                    Some(UnderlyingStorage::Wayland(buffer)) => smithay::wayland::dmabuf::get_dmabuf(buffer)
+                        .ok()
+                        .map(|buffer| buffer.format()),
+                    _ => None,
+                },
+            };
+            let feedback_candidates = feedback::candidates(
+                &mut device.renderer,
+                &elements,
+                &output.output,
+                flags.contains(FrameFlags::ALLOW_PRIMARY_PLANE_SCANOUT),
+            );
+
             if result.needs_sync()
                 && let PrimaryPlaneElement::Swapchain(primary) = &result.primary_element
             {
@@ -1347,6 +1368,9 @@ fn render_output(
                         .into_iter()
                         .any(|cursor| state.has_pending_screencopy(&output.output, cursor))
                 });
+                output
+                    .dmabuf_feedback
+                    .observed_frame(primary_format, &result.states, feedback_candidates);
                 if output.mirror_source.is_some() {
                     capture::capture_mirror(state, &mut device.renderer, &scene_output, &frame, &mut output);
                 } else {
@@ -1420,6 +1444,9 @@ fn render_output(
                 flags.contains(FrameFlags::ALLOW_PRIMARY_PLANE_SCANOUT),
             );
             capture::after_queue(queued, || {
+                output
+                    .dmabuf_feedback
+                    .observed_frame(primary_format, &visibility, feedback_candidates);
                 if output.mirror_source.is_none() {
                     state.display_presentation.queued(&output.output, &visibility);
                 }
@@ -1663,6 +1690,7 @@ fn remove_device(state: &mut Ferese, node: DrmNode) {
 
     state.loop_handle.remove(device.notifier);
     state.dmabuf_imports.remove(device.dmabuf_global);
+    state.dmabuf_feedback.remove(device.dmabuf_global);
     state
         .dmabuf_state
         .disable_global::<Ferese>(&state.display_handle, &device.dmabuf_global);
@@ -2252,6 +2280,12 @@ fn create_direct_output(
     surface.set_output_mode_source((&output).into());
     let render_metrics = RenderMetrics::from_environment(output.name());
     let diagnostics = Diagnostics::from_environment(output.name());
+    let dmabuf_feedback = feedback::OutputFeedback::new(
+        render_node.dev_id(),
+        node.dev_id(),
+        renderer.dmabuf_formats(),
+        surface.surface().plane_info().formats.clone(),
+    );
 
     tracing::info!(?crtc, connector = %connector, "initialized DRM output");
     Ok(DirectOutput {
@@ -2274,6 +2308,7 @@ fn create_direct_output(
         capture_texture: None,
         render_metrics,
         diagnostics,
+        dmabuf_feedback,
         frame_pending: false,
         completion: Default::default(),
         completion_timer: None,

compositor/src/backends/direct/feedback.rs

@@ -0,0 +1,478 @@
+//! Conservative output feedback for the existing primary-plane policy.
+use std::collections::{HashMap, HashSet};
+
+use smithay::backend::allocator::Format;
+use smithay::backend::allocator::format::FormatSet;
+use smithay::backend::renderer::element::{
+    Element, Id, RenderElement, RenderElementPresentationState, RenderElementStates, RenderingReason,
+};
+use smithay::backend::renderer::gles::GlesRenderer;
+use smithay::backend::renderer::utils::with_renderer_surface_state;
+use smithay::output::Output;
+use smithay::reexports::wayland_protocols::wp::linux_dmabuf::zv1::server::zwp_linux_dmabuf_feedback_v1::TrancheFlags;
+use smithay::utils::{Physical, Rectangle, Scale};
+use smithay::wayland::dmabuf::{DmabufFeedback, DmabufFeedbackBuilder, DmabufGlobal};
+
+use crate::Ferese;
+use crate::render::AnimatedWindowRenderElement;
+
+fn compatible(render: &FormatSet, primary: &FormatSet, observed: Format) -> bool {
+    // Do not equate implicit and explicit modifiers, alpha and opaque FourCCs,
+    // or formats from cursor/overlay planes. Exact intersection preserves import.
+    render.contains(&observed) && primary.contains(&observed)
+}
+
+pub(super) struct OutputFeedback {
+    render_device: libc::dev_t,
+    scanout_device: libc::dev_t,
+    render_formats: FormatSet,
+    primary_formats: FormatSet,
+    observed: Option<Format>,
+    preferred: Option<DmabufFeedback>,
+    visible: HashSet<Id>,
+    candidates: HashSet<Id>,
+}
+
+impl OutputFeedback {
+    pub fn new(
+        render_device: libc::dev_t,
+        scanout_device: libc::dev_t,
+        render_formats: FormatSet,
+        primary_formats: FormatSet,
+    ) -> Self {
+        Self {
+            render_device,
+            scanout_device,
+            render_formats,
+            primary_formats,
+            observed: None,
+            preferred: None,
+            visible: HashSet::new(),
+            candidates: HashSet::new(),
+        }
+    }
+
+    pub fn invalidate(&mut self) {
+        self.observed = None;
+        self.preferred = None;
+        self.visible.clear();
+        self.candidates.clear();
+    }
+
+    fn observe_format(&mut self, format: Option<Format>) {
+        let Some(format) = format else {
+            return;
+        };
+        if self.observed == Some(format) {
+            return;
+        }
+        self.observed = Some(format);
+        self.preferred = None;
+        if compatible(&self.render_formats, &self.primary_formats, format) {
+            // Restrict the preference to the observed primary format/modifier:
+            // without ALLOW_PRIMARY_PLANE_SCANOUT_ANY Smithay compares the full
+            // format with the swapchain. Renderer-only formats remain in fallback.
+            match DmabufFeedbackBuilder::new(self.render_device, self.render_formats.iter().copied())
+                .add_preference_tranche(self.scanout_device, Some(TrancheFlags::Scanout), [format])
+                .build()
+            {
+                Ok(feedback) => self.preferred = Some(feedback),
+                Err(error) => tracing::warn!(%error, "output DMA-BUF feedback unavailable; retaining render fallback"),
+            }
+        }
+    }
+
+    pub fn observed_frame(&mut self, format: Option<Format>, states: &RenderElementStates, candidates: HashSet<Id>) {
+        self.observe_format(format);
+        self.visible = states
+            .states
+            .iter()
+            .filter(|(_, state)| crate::display_presentation::is_visible(**state))
+            .map(|(id, _)| id.clone())
+            .collect();
+        self.candidates = candidates
+            .into_iter()
+            .filter(|id| {
+                states.states.get(id).is_some_and(|state| {
+                    crate::display_presentation::is_visible(*state)
+                        && state.presentation_state
+                            != RenderElementPresentationState::Rendering {
+                                reason: Some(RenderingReason::ScanoutFailed),
+                            }
+                })
+            })
+            .collect();
+    }
+}
+
+pub(super) fn candidates(
+    renderer: &mut GlesRenderer,
+    elements: &[AnimatedWindowRenderElement],
+    output: &Output,
+    allowed: bool,
+) -> HashSet<Id> {
+    if !allowed {
+        return HashSet::new();
+    }
+    let Some(mode) = output.current_mode() else {
+        return HashSet::new();
+    };
+    let bounds = Rectangle::from_size(output.current_transform().transform_size(mode.size));
+    let scale = output.current_scale().fractional_scale().into();
+    elements
+        .iter()
+        .filter(|&element| {
+            candidate_geometry(element, bounds, scale, || {
+                element.underlying_storage(renderer).is_some()
+            })
+        })
+        .map(|element| element.id().clone())
+        .collect()
+}
+
+fn candidate_geometry(
+    element: &impl Element,
+    bounds: Rectangle<i32, Physical>,
+    scale: Scale<f64>,
+    has_storage: impl FnOnce() -> bool,
+) -> bool {
+    let geometry = element.geometry(scale);
+    if element.alpha() != 1.0 || !geometry.contains_rect(bounds) {
+        return false;
+    }
+    // A single opaque rectangle covering the output is deliberately more
+    // conservative than reconstructing an arbitrary union of client regions.
+    element
+        .opaque_regions(scale)
+        .iter()
+        .any(|region| Rectangle::new(region.loc + geometry.loc, region.size).contains_rect(bounds))
+        && has_storage()
+}
+
+struct OutputView<'a> {
+    global: DmabufGlobal,
+    feedback: &'a OutputFeedback,
+    eligible: Option<Id>,
+}
+
+fn select<'a>(
+    global: DmabufGlobal,
+    surface: &Id,
+    outputs: impl IntoIterator<Item = OutputView<'a>>,
+) -> Option<&'a DmabufFeedback> {
+    let mut visible = outputs
+        .into_iter()
+        .filter(|output| output.feedback.visible.contains(surface));
+    let output = visible.next()?;
+    // A surface visible on multiple outputs gets render fallback rather than
+    // whichever output happened to render last. Cross-GPU import is not assumed.
+    if visible.next().is_some()
+        || output.global != global
+        || output.eligible.as_ref() != Some(surface)
+        || !output.feedback.candidates.contains(surface)
+    {
+        return None;
+    }
+    output.feedback.preferred.as_ref()
+}
+
+pub(crate) fn refresh(state: &mut Ferese) {
+    let mut preferred = HashMap::new();
+    if let Some(backend) = state.direct_backend.as_ref() {
+        for (surface, global) in state.dmabuf_feedback.requested() {
+            let id = Id::from(surface);
+            let outputs = backend
+                .devices
+                .values()
+                .flat_map(|device| device.outputs.values().map(move |output| (device, output)))
+                .filter(|(_, output)| state.output_id(&output.output).is_some());
+            let views = outputs.map(|(device, output)| {
+                let eligible = (backend.can_render()
+                    && device.drm.is_active()
+                    && output.power.can_render()
+                    && output.mirror_source.is_none()
+                    && !state.session_lock.active()
+                    && !state.output_has_animations(&output.output)
+                    && state.output_has_fullscreen(&output.output))
+                .then(|| {
+                    state
+                        .output_id(&output.output)
+                        .and_then(|id| state.output_workspaces.active_workspace(id))
+                        .and_then(|workspace| state.workspaces.workspace(workspace))
+                        .and_then(|workspace| workspace.fullscreen)
+                        .filter(|id| {
+                            state
+                                .current_window_presentation(*id)
+                                .is_some_and(|presentation| presentation.alpha() == 1.0 && presentation.dim == 0.0)
+                        })
+                        .and_then(|id| state.windows.window(id))
+                        .and_then(|window| window.toplevel())
+                        .filter(|toplevel| {
+                            with_renderer_surface_state(toplevel.wl_surface(), |state| state.buffer().is_some())
+                                .unwrap_or(false)
+                        })
+                        .map(|toplevel| Id::from(toplevel.wl_surface()))
+                })
+                .flatten();
+                OutputView {
+                    global: device.dmabuf_global,
+                    feedback: &output.dmabuf_feedback,
+                    eligible,
+                }
+            });
+            if let Some(feedback) = select(global, &id, views) {
+                preferred.insert(surface.clone(), feedback.clone());
+            }
+        }
+    }
+    state.dmabuf_feedback.refresh(&preferred);
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use smithay::backend::allocator::{Fourcc, Modifier};
+    use smithay::backend::renderer::element::{RenderElementPresentationState, RenderElementState};
+    use smithay::reexports::wayland_server::Display;
+    use smithay::wayland::dmabuf::DmabufState;
+
+    fn format(code: Fourcc, modifier: Modifier) -> Format {
+        Format { code, modifier }
+    }
+    fn formats(values: &[Format]) -> FormatSet {
+        values.iter().copied().collect()
+    }
+    fn output(render: &[Format], primary: &[Format]) -> OutputFeedback {
+        OutputFeedback::new(10, 20, formats(render), formats(primary))
+    }
+    fn create_globals() -> (Display<Ferese>, DmabufGlobal, DmabufGlobal) {
+        let display = Display::new().unwrap();
+        let mut state = DmabufState::new();
+        let feedback = DmabufFeedbackBuilder::new(10, [format(Fourcc::Argb8888, Modifier::Linear)])
+            .build()
+            .unwrap();
+        let global = state.create_global_with_default_feedback::<Ferese>(&display.handle(), &feedback);
+        let other = state.create_global_with_default_feedback::<Ferese>(&display.handle(), &feedback);
+        (display, global, other)
+    }
+    fn visible(output: &mut OutputFeedback, id: &Id, f: Format) {
+        let mut states = RenderElementStates::default();
+        states.states.insert(
+            id.clone(),
+            RenderElementState {
+                visible_area: 100,
+                presentation_state: RenderElementPresentationState::Rendering { reason: None },
+            },
+        );
+        output.observed_frame(Some(f), &states, HashSet::from([id.clone()]));
+    }
+
+    #[test]
+    fn preferences_require_exact_renderer_and_primary_intersection() {
+        let linear = format(Fourcc::Argb8888, Modifier::Linear);
+        let implicit = format(Fourcc::Argb8888, Modifier::Invalid);
+        let opaque = format(Fourcc::Xrgb8888, Modifier::Linear);
+        assert!(compatible(&formats(&[linear, implicit]), &formats(&[linear]), linear));
+        assert!(!compatible(&formats(&[linear]), &formats(&[implicit]), linear));
+        assert!(!compatible(&formats(&[linear]), &formats(&[opaque]), linear));
+        assert!(!compatible(&formats(&[linear]), &formats(&[implicit]), implicit));
+        assert!(!compatible(&formats(&[opaque]), &formats(&[linear]), linear));
+    }
+
+    #[test]
+    fn observed_format_limits_preference_and_retains_renderer_fallback() {
+        let linear = format(Fourcc::Argb8888, Modifier::Linear);
+        let other = format(Fourcc::Abgr8888, Modifier::Linear);
+        let mut o = output(&[linear, other], &[linear, other]);
+        assert!(o.preferred.is_none());
+        o.observe_format(Some(linear));
+        let expected = DmabufFeedbackBuilder::new(10, [linear, other])
+            .add_preference_tranche(20, Some(TrancheFlags::Scanout), [linear])
+            .build()
+            .unwrap();
+        assert_eq!(o.preferred.as_ref(), Some(&expected));
+        o.observe_format(Some(other));
+        assert_ne!(o.preferred.as_ref(), Some(&expected));
+        o.observe_format(Some(format(Fourcc::Rgb565, Modifier::Linear)));
+        assert!(o.preferred.is_none());
+    }
+
+    #[test]
+    fn migration_multi_output_cross_gpu_and_policy_changes_fall_back() {
+        let (_display, global, other_global) = create_globals();
+        let id = Id::new();
+        let f = format(Fourcc::Argb8888, Modifier::Linear);
+        let mut a = output(&[f], &[f]);
+        let mut b = output(&[f], &[f]);
+        visible(&mut a, &id, f);
+        visible(&mut b, &id, f);
+        fn view(global: DmabufGlobal, feedback: &OutputFeedback, eligible: Option<Id>) -> OutputView<'_> {
+            OutputView {
+                global,
+                feedback,
+                eligible,
+            }
+        }
+        assert!(select(global, &id, [view(global, &a, Some(id.clone()))]).is_some());
+        assert!(select(global, &id, [view(global, &a, None)]).is_none());
+        assert!(
+            select(
+                global,
+                &id,
+                [view(global, &a, Some(id.clone())), view(global, &b, Some(id.clone()))]
+            )
+            .is_none()
+        );
+        assert!(select(other_global, &id, [view(global, &a, Some(id.clone()))]).is_none());
+        a.invalidate();
+        assert!(select(global, &id, [view(global, &a, Some(id.clone()))]).is_none());
+        assert!(
+            select(
+                global,
+                &id,
+                [view(global, &a, Some(id.clone())), view(global, &b, Some(id.clone()))]
+            )
+            .is_some()
+        );
+        b.candidates.clear();
+        assert!(select(global, &id, [view(global, &b, Some(id.clone()))]).is_none());
+    }
+
+    #[test]
+    fn reconfiguration_invalidates_observations_until_a_new_frame() {
+        let (_display, global, _) = create_globals();
+        let id = Id::new();
+        let f = format(Fourcc::Argb8888, Modifier::Linear);
+        let mut o = output(&[f], &[f]);
+        visible(&mut o, &id, f);
+        o.invalidate();
+        assert!(o.observed.is_none() && o.preferred.is_none() && o.visible.is_empty() && o.candidates.is_empty());
+        o.observe_format(None);
+        assert!(o.preferred.is_none());
+        visible(&mut o, &id, f);
+        assert!(
+            select(
+                global,
+                &id,
+                [OutputView {
+                    global,
+                    feedback: &o,
+                    eligible: Some(id.clone())
+                }]
+            )
+            .is_some()
+        );
+    }
+
+    #[test]
+    fn candidate_geometry_requires_covering_opaque_content_and_storage() {
+        use smithay::backend::renderer::element::Kind;
+        use smithay::backend::renderer::element::solid::SolidColorRenderElement;
+        use smithay::backend::renderer::utils::CommitCounter;
+        use smithay::utils::Transform;
+        for transform in [Transform::Normal, Transform::_90] {
+            let bounds = Rectangle::from_size(transform.transform_size((100, 60).into()));
+            for scale in [1.0, 1.25, 2.0] {
+                let solid = |geometry, alpha| {
+                    SolidColorRenderElement::new(
+                        Id::new(),
+                        geometry,
+                        CommitCounter::default(),
+                        [1., 1., 1., alpha],
+                        Kind::Unspecified,
+                    )
+                };
+                let opaque = solid(bounds, 1.0);
+                assert!(candidate_geometry(&opaque, bounds, scale.into(), || true));
+                assert!(
+                    !candidate_geometry(&opaque, bounds, scale.into(), || false),
+                    "clipped storage must remain hidden"
+                );
+                let faded = solid(bounds, 0.8);
+                assert!(!candidate_geometry(&faded, bounds, scale.into(), || panic!(
+                    "faded elements must not probe storage"
+                )));
+                let mut partial = bounds;
+                partial.size.w -= 1;
+                assert!(!candidate_geometry(
+                    &solid(partial, 1.0),
+                    bounds,
+                    scale.into(),
+                    || panic!("partial content must not probe storage")
+                ));
+            }
+        }
+    }
+
+    #[test]
+    fn failed_scanout_and_occlusion_withdraw_preferences_but_format_hints_can_recover() {
+        let (_display, global, _) = create_globals();
+        let id = Id::new();
+        let f = format(Fourcc::Argb8888, Modifier::Linear);
+        let mut o = output(&[f], &[f]);
+        for (state, expected) in [
+            (
+                RenderElementState {
+                    visible_area: 100,
+                    presentation_state: RenderElementPresentationState::Rendering {
+                        reason: Some(RenderingReason::FormatUnsupported),
+                    },
+                },
+                true,
+            ),
+            (
+                RenderElementState {
+                    visible_area: 100,
+                    presentation_state: RenderElementPresentationState::Rendering {
+                        reason: Some(RenderingReason::ScanoutFailed),
+                    },
+                },
+                false,
+            ),
+            (
+                RenderElementState {
+                    visible_area: 0,
+                    presentation_state: RenderElementPresentationState::Skipped,
+                },
+                false,
+            ),
+            (
+                RenderElementState {
+                    visible_area: 100,
+                    presentation_state: RenderElementPresentationState::ZeroCopy,
+                },
+                true,
+            ),
+        ] {
+            let states = RenderElementStates {
+                states: HashMap::from([(id.clone(), state)]),
+            };
+            o.observed_frame(Some(f), &states, HashSet::from([id.clone()]));
+            assert_eq!(
+                select(
+                    global,
+                    &id,
+                    [OutputView {
+                        global,
+                        feedback: &o,
+                        eligible: Some(id.clone())
+                    }]
+                )
+                .is_some(),
+                expected
+            );
+        }
+    }
+
+    #[test]
+    fn queue_failure_never_publishes_prepared_feedback_observations() {
+        let f = format(Fourcc::Argb8888, Modifier::Linear);
+        let id = Id::new();
+        let mut o = output(&[f], &[f]);
+        let failed = super::super::capture::after_queue(Err::<(), _>("queue failed"), || visible(&mut o, &id, f));
+        assert!(failed.is_err());
+        assert!(o.visible.is_empty() && o.preferred.is_none());
+        super::super::capture::after_queue(Ok::<(), ()>(()), || visible(&mut o, &id, f)).unwrap();
+        assert!(o.visible.contains(&id) && o.preferred.is_some());
+    }
+}

compositor/src/backends/direct/topology.rs

@@ -274,6 +274,7 @@ pub(super) fn reconcile_outputs(state: &mut Ferese, reactivate: bool) {
                 for (crtc, output) in &mut device.outputs {
                     cancel_output_timers(&state.loop_handle, output);
                     output.surface.reset_buffers();
+                    output.dmabuf_feedback.invalidate();
                     output.primary_commit = None;
                     output.capture_texture = None;
                     output.frame_pending = false;

compositor/src/backends/direct/transaction.rs

@@ -262,6 +262,7 @@ impl crate::output_transaction::HardwareTransaction for DeviceTransaction<'_> {
             .surface
             .set_connectors(&[selection.connector.handle()])
             .map_err(|error| error.to_string())?;
+        output.dmabuf_feedback.invalidate();
         prepare(
             &mut output.surface,
             &mut self.device.renderer,
@@ -280,6 +281,7 @@ impl crate::output_transaction::HardwareTransaction for DeviceTransaction<'_> {
             let replacement = self.staged.remove(&crtc).unwrap();
             let previous = self.device.outputs.get_mut(&crtc).unwrap();
             previous.surface = replacement.surface;
+            previous.dmabuf_feedback = replacement.dmabuf_feedback;
             self.replaced.insert(crtc);
         }
         self.staged
@@ -345,11 +347,13 @@ impl crate::output_transaction::HardwareTransaction for DeviceTransaction<'_> {
             .map_err(|error| error.to_string())?;
             let previous = self.device.outputs.get_mut(&crtc).unwrap();
             previous.surface = replacement.surface;
+            previous.dmabuf_feedback = replacement.dmabuf_feedback;
             previous.surface.commit_frame().map_err(|error| error.to_string())?;
         }
         let Some(output) = self.device.outputs.get_mut(&crtc) else {
             return Ok(());
         };
+        output.dmabuf_feedback.invalidate();
         output.surface.reset_state().map_err(|error| error.to_string())?;
         let restored = output
             .surface

compositor/src/dmabuf_feedback.rs

@@ -0,0 +1,78 @@
+//! Allocation hints retain the import global's renderer as a composition fallback.
+use std::collections::HashMap;
+
+use smithay::reexports::wayland_server::Resource;
+use smithay::reexports::wayland_server::protocol::wl_surface::WlSurface;
+use smithay::wayland::compositor::with_states;
+use smithay::wayland::dmabuf::{DmabufFeedback, DmabufGlobal, SurfaceDmabufFeedbackState};
+
+struct Surface {
+    global: DmabufGlobal,
+    fallback: DmabufFeedback,
+}
+
+#[derive(Default)]
+pub(crate) struct FeedbackManager {
+    globals: HashMap<DmabufGlobal, DmabufFeedback>,
+    surfaces: HashMap<WlSurface, Surface>,
+}
+
+impl FeedbackManager {
+    pub fn register(&mut self, global: DmabufGlobal, render: DmabufFeedback) {
+        self.globals.insert(global, render);
+    }
+
+    pub fn new_surface(&mut self, surface: &WlSurface, global: DmabufGlobal) -> Option<DmabufFeedback> {
+        let fallback = self.globals.get(&global)?.clone();
+        self.surfaces.insert(
+            surface.clone(),
+            Surface {
+                global,
+                fallback: fallback.clone(),
+            },
+        );
+        Some(fallback)
+    }
+
+    pub fn requested(&self) -> impl Iterator<Item = (&WlSurface, DmabufGlobal)> {
+        self.surfaces.iter().map(|(surface, record)| (surface, record.global))
+    }
+
+    pub fn refresh(&mut self, preferred: &HashMap<WlSurface, DmabufFeedback>) {
+        self.surfaces.retain(|surface, _| surface.is_alive());
+        for (surface, record) in &self.surfaces {
+            let feedback = if self.globals.contains_key(&record.global) {
+                preferred.get(surface).unwrap_or(&record.fallback)
+            } else {
+                &record.fallback
+            };
+            // Smithay owns subscriptions, protocol messages and equality suppression.
+            with_states(surface, |states| {
+                if let Some(state) = SurfaceDmabufFeedbackState::from_states(states) {
+                    state.set_feedback(feedback);
+                }
+            });
+        }
+    }
+
+    pub fn remove(&mut self, global: DmabufGlobal) {
+        self.globals.remove(&global);
+        for (surface, record) in self.surfaces.iter().filter(|(_, record)| record.global == global) {
+            // V4 requires a main-device tranche. Retain last-known render
+            // capabilities, withdraw scanout hints and retire the global. Bound
+            // resources cannot import through the retired global or another GPU.
+            with_states(surface, |states| {
+                if let Some(state) = SurfaceDmabufFeedbackState::from_states(states) {
+                    state.set_feedback(&record.fallback);
+                }
+            });
+        }
+    }
+
+    pub fn destroyed(&mut self, surface: &WlSurface) {
+        self.surfaces.remove(surface);
+    }
+}
+
+#[cfg(test)]
+mod tests;

compositor/src/dmabuf_feedback/tests.rs

@@ -0,0 +1,205 @@
+//! Real v4 feedback subscriptions over a private Wayland socket; no DRM/EGL.
+use std::io::Read;
+use std::os::unix::net::UnixStream;
+
+use smithay::backend::allocator::{Format, Fourcc, Modifier};
+use smithay::output::{Mode, Output, PhysicalProperties, Subpixel};
+use smithay::reexports::calloop::EventLoop;
+use smithay::reexports::wayland_protocols::wp::linux_dmabuf::zv1::server::zwp_linux_dmabuf_feedback_v1::TrancheFlags;
+use smithay::reexports::wayland_server::Display;
+use smithay::wayland::dmabuf::DmabufFeedbackBuilder;
+
+use super::*;
+use crate::Ferese;
+use crate::state::{dispatch_test_client as dispatch, test_client_request as request, test_window};
+
+fn messages(wire: &mut UnixStream) -> Vec<(u32, u32, Vec<u32>)> {
+    wire.set_nonblocking(true).unwrap();
+    let mut bytes = Vec::new();
+    let _ = wire.read_to_end(&mut bytes);
+    wire.set_nonblocking(false).unwrap();
+    let mut offset = 0;
+    let mut result = Vec::new();
+    while offset < bytes.len() {
+        let object = u32::from_ne_bytes(bytes[offset..offset + 4].try_into().unwrap());
+        let header = u32::from_ne_bytes(bytes[offset + 4..offset + 8].try_into().unwrap());
+        let size = (header >> 16) as usize;
+        result.push((
+            object,
+            header & 0xffff,
+            bytes[offset + 8..offset + size]
+                .chunks_exact(4)
+                .map(|word| u32::from_ne_bytes(word.try_into().unwrap()))
+                .collect(),
+        ));
+        offset += size;
+    }
+    result
+}
+
+fn flags(events: &[(u32, u32, Vec<u32>)], object: u32) -> Vec<u32> {
+    events
+        .iter()
+        .filter(|(id, opcode, _)| *id == object && *opcode == 6)
+        .map(|(_, _, args)| args[0])
+        .collect()
+}
+
+fn done(events: &[(u32, u32, Vec<u32>)], object: u32) -> usize {
+    events
+        .iter()
+        .filter(|(id, opcode, _)| *id == object && *opcode == 0)
+        .count()
+}
+
+#[test]
+fn subscriptions_update_deduplicate_and_retire_device_preferences() {
+    if !crate::startup_tests::private_runtime(
+        "dmabuf_feedback::tests::subscriptions_update_deduplicate_and_retire_device_preferences",
+    ) {
+        return;
+    }
+    let mut events = EventLoop::try_new().unwrap();
+    let mut config = crate::config::Config::default().runtime_config().unwrap();
+    config.wallpaper.path = None;
+    let mut state = Ferese::new(&mut events, Display::new().unwrap(), config).unwrap();
+    let output = Output::new(
+        "feedback-test".into(),
+        PhysicalProperties {
+            size: (0, 0).into(),
+            subpixel: Subpixel::Unknown,
+            make: "test".into(),
+            model: "test".into(),
+        },
+    );
+    output.change_current_state(
+        Some(Mode {
+            size: (100, 100).into(),
+            refresh: 60_000,
+        }),
+        None,
+        None,
+        None,
+    );
+    state.space.map_output(&output, (0, 0));
+    state.register_output(&output, "feedback-test".into());
+    let f = Format {
+        code: Fourcc::Argb8888,
+        modifier: Modifier::Linear,
+    };
+    let fallback = DmabufFeedbackBuilder::new(10, [f]).build().unwrap();
+    let preferred = DmabufFeedbackBuilder::new(10, [f])
+        .add_preference_tranche(20, Some(TrancheFlags::Scanout), [f])
+        .build()
+        .unwrap();
+    let global = state
+        .dmabuf_state
+        .create_global_with_default_feedback::<Ferese>(&state.display_handle, &fallback);
+    state.dmabuf_feedback.register(global, fallback.clone());
+    let (window, mut wire) = test_window(&mut state, &mut events, 0xffabcdef);
+    let surface = window.toplevel().unwrap().wl_surface().clone();
+    request(&mut wire, 1, 1, &[7], None); // wl_display.get_registry
+    dispatch(&mut events, &mut state);
+    let registry = messages(&mut wire);
+    let dmabuf_name = registry
+        .iter()
+        .find_map(|(object, opcode, args)| {
+            if *object != 7 || *opcode != 0 {
+                return None;
+            }
+            let bytes = args[2..].iter().flat_map(|word| word.to_ne_bytes()).collect::<Vec<_>>();
+            bytes.starts_with(b"zwp_linux_dmabuf_v1\0").then_some(args[0])
+        })
+        .expect("DMA-BUF v4 global");
+    let name = b"zwp_linux_dmabuf_v1\0";
+    let mut padded = name.to_vec();
+    padded.resize(name.len().next_multiple_of(4), 0);
+    let bind = [dmabuf_name, name.len() as u32]
+        .into_iter()
+        .chain(
+            padded
+                .chunks_exact(4)
+                .map(|word| u32::from_ne_bytes(word.try_into().unwrap())),
+        )
+        .chain([4, 8])
+        .collect::<Vec<_>>();
+    request(&mut wire, 7, 0, &bind, None);
+    request(&mut wire, 8, 3, &[9, 2], None); // get_surface_feedback
+    dispatch(&mut events, &mut state);
+    let initial = messages(&mut wire);
+    assert_eq!(flags(&initial, 9), [0]);
+    assert_eq!(done(&initial, 9), 1);
+    assert_eq!(state.dmabuf_feedback.requested().count(), 1);
+    let main = initial.iter().find(|(id, opcode, _)| *id == 9 && *opcode == 2).unwrap();
+    assert_eq!(main.2[0] as usize, std::mem::size_of::<libc::dev_t>());
+    assert_eq!(main.2[1], 10, "main device remains the renderer");
+
+    let selected = HashMap::from([(surface.clone(), preferred.clone())]);
+    state.dmabuf_feedback.refresh(&selected);
+    state.display_handle.flush_clients().unwrap();
+    let changed = messages(&mut wire);
+    assert_eq!(flags(&changed, 9), [TrancheFlags::Scanout.bits(), 0]);
+    assert_eq!(done(&changed, 9), 1);
+    let target = changed.iter().find(|(id, opcode, _)| *id == 9 && *opcode == 4).unwrap();
+    assert_eq!(target.2[1], 20, "scanout tranche targets the primary DRM device");
+    state.dmabuf_feedback.refresh(&selected);
+    state.display_handle.flush_clients().unwrap();
+    assert!(messages(&mut wire).is_empty(), "Smithay suppresses unchanged feedback");
+
+    request(&mut wire, 8, 3, &[10, 2], None);
+    dispatch(&mut events, &mut state);
+    let second = messages(&mut wire);
+    assert_eq!(flags(&second, 10), [TrancheFlags::Scanout.bits(), 0]);
+    state.dmabuf_feedback.refresh(&HashMap::new()); // migration/ineligibility fallback
+    state.display_handle.flush_clients().unwrap();
+    let fallback_events = messages(&mut wire);
+    assert_eq!(flags(&fallback_events, 9), [0]);
+    assert_eq!(flags(&fallback_events, 10), [0]);
+    assert_eq!(done(&fallback_events, 9), 1);
+    assert_eq!(done(&fallback_events, 10), 1);
+
+    request(&mut wire, 9, 0, &[], None);
+    request(&mut wire, 10, 0, &[], None);
+    dispatch(&mut events, &mut state);
+    messages(&mut wire);
+    request(&mut wire, 8, 3, &[11, 2], None);
+    request(&mut wire, 8, 2, &[12], None); // get_default_feedback
+    dispatch(&mut events, &mut state);
+    let renewed = messages(&mut wire);
+    assert_eq!(flags(&renewed, 11), [0], "renewal starts with safe render feedback");
+    assert_eq!(flags(&renewed, 12), [0]);
+    state.dmabuf_feedback.refresh(&selected);
+    state.display_handle.flush_clients().unwrap();
+    let promoted = messages(&mut wire);
+    assert_eq!(flags(&promoted, 11), [TrancheFlags::Scanout.bits(), 0]);
+    state.dmabuf_feedback.remove(global);
+    state
+        .dmabuf_state
+        .disable_global::<Ferese>(&state.display_handle, &global);
+    state.display_handle.flush_clients().unwrap();
+    let retired = messages(&mut wire);
+    assert_eq!(done(&retired, 11), 1);
+    assert_eq!(
+        flags(&retired, 11),
+        [0],
+        "retirement keeps a valid main-device render tranche"
+    );
+    assert_eq!(
+        done(&retired, 12),
+        0,
+        "default renderer capabilities remain unchanged on existing bindings"
+    );
+    assert!(
+        retired
+            .iter()
+            .any(|(id, opcode, args)| *id == 7 && *opcode == 1 && args[0] == dmabuf_name)
+    );
+    state.dmabuf_feedback.refresh(&selected); // A stale choice cannot resurrect a retired preference.
+    state.display_handle.flush_clients().unwrap();
+    assert!(messages(&mut wire).is_empty());
+    assert!(state.dmabuf_feedback.new_surface(&surface, global).is_none());
+    drop(wire);
+    dispatch(&mut events, &mut state);
+    state.dmabuf_feedback.refresh(&HashMap::new());
+    assert_eq!(state.dmabuf_feedback.requested().count(), 0);
+}

compositor/src/handlers/compositor.rs

@@ -84,6 +84,7 @@ impl CompositorHandler for Ferese {
     }
 
     fn destroyed(&mut self, surface: &WlSurface) {
+        self.dmabuf_feedback.destroyed(surface);
         if self.session_lock.active() {
             self.session_lock
                 .surfaces
@@ -112,6 +113,14 @@ impl DmabufHandler for Ferese {
         &mut self.dmabuf_state
     }
 
+    fn new_surface_feedback(
+        &mut self,
+        surface: &WlSurface,
+        global: &DmabufGlobal,
+    ) -> Option<smithay::wayland::dmabuf::DmabufFeedback> {
+        self.dmabuf_feedback.new_surface(surface, *global)
+    }
+
     fn dmabuf_imported(&mut self, global: &DmabufGlobal, dmabuf: Dmabuf, notifier: ImportNotifier) {
         self.dmabuf_imports
             .enqueue(*global, dmabuf, notifier, std::time::Instant::now());

compositor/src/main.rs

@@ -5,6 +5,7 @@ mod config;
 mod cursor;
 mod daemon;
 mod display_presentation;
+mod dmabuf_feedback;
 mod dmabuf_imports;
 mod effects;
 mod floating;
@@ -243,6 +244,7 @@ fn after_dispatch_with_redraw(state: &mut Ferese, mut redraw: impl FnMut(&mut Fe
         warn!(%error, "could not schedule wallpaper upload retry");
     }
 
+    crate::backends::direct::feedback::refresh(state);
     state.publish_ipc_events();
 
     // Clients need registry/sync/configure replies even when no frame is drawn.

compositor/src/state.rs

@@ -514,6 +514,7 @@ pub struct Ferese {
     pub direct_backend: Option<DirectBackendState>,
     _ipc_socket: Option<IpcSocketGuard>,
     pub(crate) dmabuf_imports: crate::dmabuf_imports::ImportManager,
+    pub(crate) dmabuf_feedback: crate::dmabuf_feedback::FeedbackManager,
     pub(crate) activation_inputs: crate::handlers::activation::InputHistory,
     pub(crate) pending_screencopies: Vec<PendingScreencopy>,
     // Screenshot requests outlive the readback that filled them: the
@@ -818,6 +819,7 @@ impl Ferese {
             direct_backend: None,
             _ipc_socket: None,
             dmabuf_imports: crate::dmabuf_imports::ImportManager::default(),
+            dmabuf_feedback: Default::default(),
             activation_inputs: Default::default(),
             pending_screencopies: Vec::new(),
             screenshot: Coordinator::new(),

docs/README.md

@@ -46,6 +46,7 @@ and [guides at that tag](https://github.com/ferese-wm/ferese/tree/v0.1.0-beta/do
 - [Animation model](animation-model.md): geometry, spring motion, gestures and prediction.
 - [Desktop reconciliation](desktop-reconciliation.md): topology publication and presentation ownership.
 - [Rendering diagnostics](rendering-performance.md): outcome counters, capture cost and repeatable hardware baselines.
+- [DMA-BUF feedback](dmabuf-feedback.md): conservative allocation hints, fallback and hardware validation.
 - [Output management tests](output-management-testing.md): profiles, hotplug and DRM failures.
 - [Nested resize checks](nested-resize-testing.md).
 - [Client soak testing](native-soak-testing.md).

docs/development.md

@@ -30,6 +30,7 @@ host. Hardware checks remain separate.
 - [Live Activities checks](#live-activities-checks): core, CLI, shell hooks, installer setup and native rendering.
 - [Desktop reconciliation](desktop-reconciliation.md): topology publication, hardware failure and presentation ownership.
 - [Rendering diagnostics](rendering-performance.md): per-output outcomes and repeatable hardware baselines.
+- [DMA-BUF feedback](dmabuf-feedback.md): conservative allocation hints, fallback and hardware validation.
 - [Output management tests](output-management-testing.md): profiles, hotplug, mirroring, lid handling and failed DRM commits.
 - [Nested resize checks](nested-resize-testing.md): wallpaper and bar responsiveness.
 - [Client soak testing](native-soak-testing.md): repeated client lifecycle and resource growth.

docs/dmabuf-feedback.md

@@ -0,0 +1,94 @@
+# DMA-BUF allocation feedback
+
+DRM sessions publish renderer formats through a Linux DMA-BUF v4 global for each
+render device. Its default feedback identifies that renderer's render node (or
+its primary node when no render node is available). Per-surface feedback starts
+with this same render fallback; requesting feedback does not require a mapped
+window or a scanout-capable buffer.
+
+After a successful display queue, or a successful render returning no display
+changes, Ferese can add an output-specific `scanout` preference. It uses the
+primary plane selected by that output's DRM surface, not a union of all planes.
+Only an exact FourCC/modifier pair shared by that plane and the renderer is
+preferred. The pair must have been observed on the primary result. This narrow
+choice follows the current policy's full-format comparison with the swapchain;
+it does not advertise broader primary formats, cursor formats or overlay formats.
+Implicit modifiers are not equated with explicit linear modifiers, and alpha
+formats are not equated with their opaque counterparts.
+
+The preference's target device is the output's primary DRM node; its main device
+and renderer fallback remain the originating global's render device. Every
+preferred pair is also renderer-importable. Failed preference construction retains
+render feedback rather than failing an already queued display frame.
+
+## Surface selection and invalidation
+
+Only the current mapped fullscreen toplevel can receive the preference. Its last
+successful frame must expose underlying buffer storage, opaque content covering
+the output and full opacity. Rounded shader clipping continues to hide storage.
+Current opacity or dimming also withdraws the preference immediately.
+The surface must be visible on exactly one registered output, and that output
+must belong to the originating import global. Animation, overview, locking,
+paused/inactive outputs and mirror targets receive render fallback. A failed
+scanout reported by Smithay withdraws the preference; a format-unsupported result
+can still receive a compatible allocation hint.
+
+Feedback selection refreshes after event dispatch, before flushing protocol
+messages, even when no frame is drawn. Migration, fullscreen exit, unmapping and
+changes in eligibility restore render-only feedback. Multiple visible outputs do
+not race to replace one another's feedback. Subsurfaces, popups, layer surfaces
+and cursors currently retain render-only feedback; eligibility expansion is a
+separate hardware-gated phase.
+
+Mode changes, surface replacement, rollback and session pause/reactivation
+invalidate output observations. A fresh successful frame is needed before a
+preference returns. Retiring a GPU withdraws scanout preferences and retires its
+global. Version 4
+requires a main-device tranche, so existing bindings retain last-known renderer
+capabilities rather than receiving empty feedback or being retargeted to another
+GPU. These retained capabilities do not promise device availability: imports
+through that retired global are rejected. A reconnected device publishes a new
+global. Destroyed surface requests are pruned.
+Smithay owns the subscriptions, format-table transport, tranche encoding and
+suppression of unchanged updates. Its feedback state is shared by all subscriptions
+for one surface; the initial live request determines the originating global.
+Destroy all feedback objects before requesting the surface through a different
+global to change that association.
+
+Cross-GPU migration keeps the originating renderer's fallback instead of
+advertising unverified formats or retargeting imports to another GPU. This does
+not establish cross-GPU composition compatibility: imports are still checked
+through the originating global's renderer, and the existing destination-renderer
+behavior needs physical validation. Version 3 clients have no per-surface feedback
+updates; their existing import checks remain authoritative.
+
+Allocation feedback is an optimization hint, not proof that a particular scene
+or client buffer will scan out. The compositor still owns plane assignment,
+clipping, blur, capture privacy, lock protection, gamma and composition fallback.
+No plane eligibility flags change here.
+
+## Checks and hardware gate
+
+```sh
+cargo test --locked -p ferese backends::direct::feedback -- --nocapture
+cargo test --locked -p ferese dmabuf_feedback -- --nocapture
+```
+
+Deterministic tests cover exact intersections, device identities, observed format
+changes, candidate geometry, queue failure, migration, multiple outputs, policy
+withdrawal, scanout failure and reconfiguration. A private Wayland v4 client checks
+initial render feedback, subscriptions, preference/fallback changes, unchanged
+update suppression, feedback renewal and retired-device preference withdrawal. It does not
+create a real DMA-BUF, use DRM or establish hardware acceptance.
+
+Before expanding eligibility, record the build, GPU/driver/kernel, output modes,
+scale/transforms, application and its actual allocated FourCC/modifier. Decode
+v4 feedback device IDs and format tables and verify preferences against the live
+renderer and selected primary plane. Test fullscreen entry/exit, animations,
+opacity/clipping, popups, sharing and lock/unlock; compare captured pixels and
+allocation outcomes. Exercise mixed outputs, mirrors, migration, hotplug,
+reconfiguration/rollback, pause/resume, gamma and failed scanout. Include v3 clients
+and applications that ignore or follow feedback, and explicitly validate
+cross-GPU imports. Use the [output-management hardware checks](output-management-testing.md).
+Physical scanout rate, GPU work, power and presentation latency remain unmeasured
+until those hardware baselines are collected.

docs/rendering-performance.md

@@ -86,5 +86,5 @@ actual framebuffer recapture counts/cost, asynchronous capture encoding and
 window-capture readback separately. Current effect metrics are placeholders;
 `capture_us` covers monitor capture batches only. Exact cursor-only KMS updates
 and detailed scanout rejection causes require additional upstream observability.
-Output/per-surface DMA-BUF feedback and eligibility expansion remain later phases,
-with broader primary and overlay policies validated separately on hardware.
+See [DMA-BUF feedback](dmabuf-feedback.md) for the current allocation hints.
+Broader primary-plane and overlay policies still need hardware validation.