Ensure safety of indirect dispatch (#5714)

by injecting a compute shader that validates the content of the indirect buffer

CHANGELOG.md

@@ -133,6 +133,7 @@ By @bradwerth [#6216](https://github.com/gfx-rs/wgpu/pull/6216).
 - Call `flush_mapped_ranges` when unmapping write-mapped buffers. By @teoxoy in [#6089](https://github.com/gfx-rs/wgpu/pull/6089).
 - When mapping buffers for reading, mark buffers as initialized only when they have `MAP_WRITE` usage. By @teoxoy in [#6178](https://github.com/gfx-rs/wgpu/pull/6178).
 - Add a separate pipeline constants error. By @teoxoy in [#6094](https://github.com/gfx-rs/wgpu/pull/6094).
+- Ensure safety of indirect dispatch by injecting a compute shader that validates the content of the indirect buffer. By @teoxoy in [#5714](https://github.com/gfx-rs/wgpu/pull/5714)
 
 #### GLES / OpenGL
 

tests/tests/dispatch_workgroups_indirect.rs

@@ -0,0 +1,241 @@
+use wgpu_test::{gpu_test, FailureCase, GpuTestConfiguration, TestParameters, TestingContext};
+
+/// Make sure that the num_workgroups builtin works properly (it requires a workaround on D3D12).
+#[gpu_test]
+static NUM_WORKGROUPS_BUILTIN: GpuTestConfiguration = GpuTestConfiguration::new()
+    .parameters(
+        TestParameters::default()
+            .features(wgpu::Features::PUSH_CONSTANTS)
+            .downlevel_flags(
+                wgpu::DownlevelFlags::COMPUTE_SHADERS | wgpu::DownlevelFlags::INDIRECT_EXECUTION,
+            )
+            .limits(wgpu::Limits {
+                max_push_constant_size: 4,
+                ..wgpu::Limits::downlevel_defaults()
+            })
+            .expect_fail(FailureCase::backend(wgt::Backends::DX12)),
+    )
+    .run_async(|ctx| async move {
+        let num_workgroups = [1, 2, 3];
+        let res = run_test(&ctx, &num_workgroups, false).await;
+        assert_eq!(res, num_workgroups);
+    });
+
+/// Make sure that we discard (don't run) the dispatch if its size exceeds the device limit.
+#[gpu_test]
+static DISCARD_DISPATCH: GpuTestConfiguration = GpuTestConfiguration::new()
+    .parameters(
+        TestParameters::default()
+            .features(wgpu::Features::PUSH_CONSTANTS)
+            .downlevel_flags(
+                wgpu::DownlevelFlags::COMPUTE_SHADERS | wgpu::DownlevelFlags::INDIRECT_EXECUTION,
+            )
+            .limits(wgpu::Limits {
+                max_compute_workgroups_per_dimension: 10,
+                max_push_constant_size: 4,
+                ..wgpu::Limits::downlevel_defaults()
+            })
+            .expect_fail(FailureCase::backend(wgt::Backends::DX12)),
+    )
+    .run_async(|ctx| async move {
+        let max = ctx.device.limits().max_compute_workgroups_per_dimension;
+
+        let res = run_test(&ctx, &[max, max, max], false).await;
+        assert_eq!(res, [max; 3]);
+
+        let res = run_test(&ctx, &[max + 1, 1, 1], false).await;
+        assert_eq!(res, [0; 3]);
+
+        let res = run_test(&ctx, &[1, max + 1, 1], false).await;
+        assert_eq!(res, [0; 3]);
+
+        let res = run_test(&ctx, &[1, 1, max + 1], false).await;
+        assert_eq!(res, [0; 3]);
+    });
+
+/// Make sure that resetting the bind groups set by the validation code works properly.
+#[gpu_test]
+static RESET_BIND_GROUPS: GpuTestConfiguration = GpuTestConfiguration::new()
+    .parameters(
+        TestParameters::default()
+            .features(wgpu::Features::PUSH_CONSTANTS)
+            .downlevel_flags(
+                wgpu::DownlevelFlags::COMPUTE_SHADERS | wgpu::DownlevelFlags::INDIRECT_EXECUTION,
+            )
+            .limits(wgpu::Limits {
+                max_push_constant_size: 4,
+                ..wgpu::Limits::downlevel_defaults()
+            }),
+    )
+    .run_async(|ctx| async move {
+        ctx.device.push_error_scope(wgpu::ErrorFilter::Validation);
+
+        let _ = run_test(&ctx, &[0, 0, 0], true).await;
+
+        let error = pollster::block_on(ctx.device.pop_error_scope());
+        assert!(error.map_or(false, |error| {
+            format!("{error}").contains("The current set ComputePipeline with '' label expects a BindGroup to be set at index 0")
+        }));
+    });
+
+async fn run_test(
+    ctx: &TestingContext,
+    num_workgroups: &[u32; 3],
+    forget_to_set_bind_group: bool,
+) -> [u32; 3] {
+    const SHADER_SRC: &str = "
+        struct TestOffsetPc {
+            inner: u32,
+        }
+
+        // `test_offset.inner` should always be 0; we test that resetting the push constant set by the validation code works properly.
+        var<push_constant> test_offset: TestOffsetPc;
+
+        @group(0) @binding(0)
+        var<storage, read_write> out: array<u32, 3>;
+
+        @compute @workgroup_size(1)
+        fn main(@builtin(num_workgroups) num_workgroups: vec3u, @builtin(workgroup_id) workgroup_id: vec3u) {
+            if (all(workgroup_id == vec3u())) {
+                out[0] = num_workgroups.x + test_offset.inner;
+                out[1] = num_workgroups.y + test_offset.inner;
+                out[2] = num_workgroups.z + test_offset.inner;
+            }
+        }
+    ";
+
+    let module = ctx
+        .device
+        .create_shader_module(wgpu::ShaderModuleDescriptor {
+            label: None,
+            source: wgpu::ShaderSource::Wgsl(SHADER_SRC.into()),
+        });
+
+    let bgl = ctx
+        .device
+        .create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
+            label: None,
+            entries: &[wgpu::BindGroupLayoutEntry {
+                binding: 0,
+                visibility: wgt::ShaderStages::COMPUTE,
+                ty: wgpu::BindingType::Buffer {
+                    ty: wgpu::BufferBindingType::Storage { read_only: false },
+                    has_dynamic_offset: false,
+                    min_binding_size: None,
+                },
+                count: None,
+            }],
+        });
+
+    let layout = ctx
+        .device
+        .create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
+            label: None,
+            bind_group_layouts: &[&bgl],
+            push_constant_ranges: &[wgt::PushConstantRange {
+                stages: wgt::ShaderStages::COMPUTE,
+                range: 0..4,
+            }],
+        });
+
+    let pipeline = ctx
+        .device
+        .create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
+            label: None,
+            layout: Some(&layout),
+            module: &module,
+            entry_point: Some("main"),
+            compilation_options: Default::default(),
+            cache: None,
+        });
+
+    let out_buffer = ctx.device.create_buffer(&wgpu::BufferDescriptor {
+        label: None,
+        size: 12,
+        usage: wgpu::BufferUsages::STORAGE | wgpu::BufferUsages::COPY_SRC,
+        mapped_at_creation: false,
+    });
+
+    let readback_buffer = ctx.device.create_buffer(&wgpu::BufferDescriptor {
+        label: None,
+        size: 12,
+        usage: wgpu::BufferUsages::COPY_DST | wgpu::BufferUsages::MAP_READ,
+        mapped_at_creation: false,
+    });
+
+    let bind_group = ctx.device.create_bind_group(&wgpu::BindGroupDescriptor {
+        label: None,
+        layout: &pipeline.get_bind_group_layout(0),
+        entries: &[wgpu::BindGroupEntry {
+            binding: 0,
+            resource: out_buffer.as_entire_binding(),
+        }],
+    });
+
+    let mut res = None;
+
+    for (indirect_offset, indirect_buffer_size) in [
+        // internal src buffer binding size will be buffer.size
+        (0, 12),
+        (4, 4 + 12),
+        (4, 8 + 12),
+        (256 * 2 - 4 - 12, 256 * 2 - 4),
+        // internal src buffer binding size will be 256 * 2 + x
+        (0, 256 * 2 * 2 + 4),
+        (256, 256 * 2 * 2 + 8),
+        (256 + 4, 256 * 2 * 2 + 12),
+        (256 * 2 + 16, 256 * 2 * 2 + 16),
+        (256 * 2 * 2, 256 * 2 * 2 + 32),
+        (256 + 12, 256 * 2 * 2 + 64),
+    ] {
+        let indirect_buffer = ctx.device.create_buffer(&wgpu::BufferDescriptor {
+            label: None,
+            size: indirect_buffer_size,
+            usage: wgpu::BufferUsages::COPY_DST | wgpu::BufferUsages::INDIRECT,
+            mapped_at_creation: false,
+        });
+
+        ctx.queue.write_buffer(
+            &indirect_buffer,
+            indirect_offset,
+            bytemuck::bytes_of(num_workgroups),
+        );
+
+        let mut encoder = ctx.device.create_command_encoder(&Default::default());
+        {
+            let mut compute_pass = encoder.begin_compute_pass(&Default::default());
+            compute_pass.set_pipeline(&pipeline);
+            compute_pass.set_push_constants(0, &[0, 0, 0, 0]);
+            if !forget_to_set_bind_group {
+                compute_pass.set_bind_group(0, Some(&bind_group), &[]);
+            }
+            compute_pass.dispatch_workgroups_indirect(&indirect_buffer, indirect_offset);
+        }
+
+        encoder.copy_buffer_to_buffer(&out_buffer, 0, &readback_buffer, 0, 12);
+
+        ctx.queue.submit(Some(encoder.finish()));
+
+        readback_buffer
+            .slice(..)
+            .map_async(wgpu::MapMode::Read, |_| {});
+
+        ctx.async_poll(wgpu::Maintain::wait())
+            .await
+            .panic_on_timeout();
+
+        let view = readback_buffer.slice(..).get_mapped_range();
+
+        let current_res = *bytemuck::from_bytes(&view);
+        drop(view);
+        readback_buffer.unmap();
+
+        if let Some(past_res) = res {
+            assert_eq!(past_res, current_res);
+        } else {
+            res = Some(current_res);
+        }
+    }
+
+    res.unwrap()
+}

tests/tests/root.rs

@@ -19,6 +19,7 @@ mod clear_texture;
 mod compute_pass_ownership;
 mod create_surface_error;
 mod device;
+mod dispatch_workgroups_indirect;
 mod encoder;
 mod external_texture;
 mod float32_filterable;

wgpu-core/Cargo.toml

@@ -51,6 +51,10 @@ renderdoc = ["hal/renderdoc"]
 ## to the validation carried out at public APIs in all builds.
 strict_asserts = ["wgt/strict_asserts"]
 
+## Validates indirect draw/dispatch calls. This will also enable naga's
+## WGSL frontend since we use a WGSL compute shader to do the validation.
+indirect-validation = ["naga/wgsl-in"]
+
 ## Enables serialization via `serde` on common wgpu types.
 serde = ["dep:serde", "wgt/serde", "arrayvec/serde"]
 

wgpu-core/src/command/bind.rs

@@ -200,13 +200,17 @@ mod compat {
                 entries: (0..hal::MAX_BIND_GROUPS).map(|_| Entry::empty()).collect(),
             }
         }
-        fn make_range(&self, start_index: usize) -> Range<usize> {
+
+        pub fn num_valid_entries(&self) -> usize {
             // find first incompatible entry
-            let end = self
-                .entries
+            self.entries
                 .iter()
                 .position(|e| e.is_incompatible())
-                .unwrap_or(self.entries.len());
+                .unwrap_or(self.entries.len())
+        }
+
+        fn make_range(&self, start_index: usize) -> Range<usize> {
+            let end = self.num_valid_entries();
             start_index..end.max(start_index)
         }
 
@@ -406,6 +410,14 @@ impl Binder {
             .map(move |index| payloads[index].group.as_ref().unwrap())
     }
 
+    #[cfg(feature = "indirect-validation")]
+    pub(super) fn list_valid<'a>(&'a self) -> impl Iterator<Item = (usize, &'a EntryPayload)> + '_ {
+        self.payloads
+            .iter()
+            .take(self.manager.num_valid_entries())
+            .enumerate()
+    }
+
     pub(super) fn check_compatibility<T: Labeled>(
         &self,
         pipeline: &T,

wgpu-core/src/command/compute.rs

@@ -18,7 +18,7 @@ use crate::{
     pipeline::ComputePipeline,
     resource::{
         self, Buffer, DestroyedResourceError, InvalidResourceError, Labeled,
-        MissingBufferUsageError, ParentDevice, Trackable,
+        MissingBufferUsageError, ParentDevice,
     },
     snatch::SnatchGuard,
     track::{ResourceUsageCompatibilityError, Tracker, TrackerIndex, UsageScope},
@@ -216,6 +216,8 @@ struct State<'scope, 'snatch_guard, 'cmd_buf, 'raw_encoder> {
     string_offset: usize,
     active_query: Option<(Arc<resource::QuerySet>, u32)>,
 
+    push_constants: Vec<u32>,
+
     intermediate_trackers: Tracker,
 
     /// Immediate texture inits required because of prior discards. Need to
@@ -443,6 +445,8 @@ impl Global {
             string_offset: 0,
             active_query: None,
 
+            push_constants: Vec::new(),
+
             intermediate_trackers: Tracker::new(),
 
             pending_discard_init_fixups: SurfacesInDiscardState::new(),
@@ -746,6 +750,21 @@ fn set_pipeline(
             }
         }
 
+        // TODO: integrate this in the code below once we simplify push constants
+        state.push_constants.clear();
+        // Note that can only be one range for each stage. See the `MoreThanOnePushConstantRangePerStage` error.
+        if let Some(push_constant_range) =
+            pipeline.layout.push_constant_ranges.iter().find_map(|pcr| {
+                pcr.stages
+                    .contains(wgt::ShaderStages::COMPUTE)
+                    .then_some(pcr.range.clone())
+            })
+        {
+            // Note that non-0 range start doesn't work anyway https://github.com/gfx-rs/wgpu/issues/4502
+            let len = push_constant_range.len() / wgt::PUSH_CONSTANT_ALIGNMENT as usize;
+            state.push_constants.extend(core::iter::repeat(0).take(len));
+        }
+
         // Clear push constant ranges
         let non_overlapping =
             super::bind::compute_nonoverlapping_ranges(&pipeline.layout.push_constant_ranges);
@@ -791,6 +810,10 @@ fn set_push_constant(
         end_offset_bytes,
     )?;
 
+    let offset_in_elements = (offset / wgt::PUSH_CONSTANT_ALIGNMENT) as usize;
+    let size_in_elements = (size_bytes / wgt::PUSH_CONSTANT_ALIGNMENT) as usize;
+    state.push_constants[offset_in_elements..][..size_in_elements].copy_from_slice(data_slice);
+
     unsafe {
         state.raw_encoder.set_push_constants(
             pipeline_layout.raw(),
@@ -841,10 +864,6 @@ fn dispatch_indirect(
         .device
         .require_downlevel_flags(wgt::DownlevelFlags::INDIRECT_EXECUTION)?;
 
-    state
-        .scope
-        .buffers
-        .merge_single(&buffer, hal::BufferUses::INDIRECT)?;
     buffer.check_usage(wgt::BufferUsages::INDIRECT)?;
 
     if offset % 4 != 0 {
@@ -861,7 +880,6 @@ fn dispatch_indirect(
     }
 
     let stride = 3 * 4; // 3 integers, x/y/z group size
-
     state
         .buffer_memory_init_actions
         .extend(buffer.initialization_status.read().create_action(
@@ -870,12 +888,132 @@ fn dispatch_indirect(
             MemoryInitKind::NeedsInitializedMemory,
         ));
 
-    state.flush_states(Some(buffer.tracker_index()))?;
+    #[cfg(feature = "indirect-validation")]
+    {
+        let params = state.device.indirect_validation.as_ref().unwrap().params(
+            &state.device.limits,
+            offset,
+            buffer.size,
+        );
 
-    let buf_raw = buffer.try_raw(&state.snatch_guard)?;
-    unsafe {
-        state.raw_encoder.dispatch_indirect(buf_raw, offset);
+        unsafe {
+            state.raw_encoder.set_compute_pipeline(params.pipeline);
+        }
+
+        unsafe {
+            state.raw_encoder.set_push_constants(
+                params.pipeline_layout,
+                wgt::ShaderStages::COMPUTE,
+                0,
+                &[params.offset_remainder as u32 / 4],
+            );
+        }
+
+        unsafe {
+            state.raw_encoder.set_bind_group(
+                params.pipeline_layout,
+                0,
+                Some(params.dst_bind_group),
+                &[],
+            );
+        }
+        unsafe {
+            state.raw_encoder.set_bind_group(
+                params.pipeline_layout,
+                1,
+                Some(
+                    buffer
+                        .raw_indirect_validation_bind_group
+                        .get(&state.snatch_guard)
+                        .unwrap()
+                        .as_ref(),
+                ),
+                &[params.aligned_offset as u32],
+            );
+        }
+
+        let src_transition = state
+            .intermediate_trackers
+            .buffers
+            .set_single(&buffer, hal::BufferUses::STORAGE_READ);
+        let src_barrier =
+            src_transition.map(|transition| transition.into_hal(&buffer, &state.snatch_guard));
+        unsafe {
+            state.raw_encoder.transition_buffers(src_barrier.as_slice());
+        }
+
+        unsafe {
+            state.raw_encoder.transition_buffers(&[hal::BufferBarrier {
+                buffer: params.dst_buffer,
+                usage: hal::BufferUses::INDIRECT..hal::BufferUses::STORAGE_READ_WRITE,
+            }]);
+        }
+
+        unsafe {
+            state.raw_encoder.dispatch([1, 1, 1]);
+        }
+
+        // reset state
+        {
+            let pipeline = state.pipeline.as_ref().unwrap();
+
+            unsafe {
+                state.raw_encoder.set_compute_pipeline(pipeline.raw());
+            }
+
+            if !state.push_constants.is_empty() {
+                unsafe {
+                    state.raw_encoder.set_push_constants(
+                        pipeline.layout.raw(),
+                        wgt::ShaderStages::COMPUTE,
+                        0,
+                        &state.push_constants,
+                    );
+                }
+            }
+
+            for (i, e) in state.binder.list_valid() {
+                let group = e.group.as_ref().unwrap();
+                let raw_bg = group.try_raw(&state.snatch_guard)?;
+                unsafe {
+                    state.raw_encoder.set_bind_group(
+                        pipeline.layout.raw(),
+                        i as u32,
+                        Some(raw_bg),
+                        &e.dynamic_offsets,
+                    );
+                }
+            }
+        }
+
+        unsafe {
+            state.raw_encoder.transition_buffers(&[hal::BufferBarrier {
+                buffer: params.dst_buffer,
+                usage: hal::BufferUses::STORAGE_READ_WRITE..hal::BufferUses::INDIRECT,
+            }]);
+        }
+
+        state.flush_states(None)?;
+        unsafe {
+            state.raw_encoder.dispatch_indirect(params.dst_buffer, 0);
+        }
+    };
+    #[cfg(not(feature = "indirect-validation"))]
+    {
+        state
+            .scope
+            .buffers
+            .merge_single(&buffer, hal::BufferUses::INDIRECT)?;
+
+        use crate::resource::Trackable;
+        state.flush_states(Some(buffer.tracker_index()))?;
+
+        let buf_raw = buffer.try_raw(&state.snatch_guard)?;
+        unsafe {
+            state.raw_encoder.dispatch_indirect(buf_raw, offset);
+        }
     }
+
     Ok(())
 }
 

wgpu-core/src/device/global.rs

@@ -406,12 +406,12 @@ impl Global {
             trace.add(trace::Action::CreateBuffer(fid.id(), desc.clone()));
         }
 
-        let buffer = device.create_buffer_from_hal(Box::new(hal_buffer), desc);
+        let (buffer, err) = device.create_buffer_from_hal(Box::new(hal_buffer), desc);
 
         let id = fid.assign(buffer);
         api_log!("Device::create_buffer -> {id:?}");
 
-        (id, None)
+        (id, err)
     }
 
     pub fn texture_destroy(&self, texture_id: id::TextureId) -> Result<(), resource::DestroyError> {

wgpu-core/src/device/mod.rs

@@ -36,7 +36,7 @@ pub(crate) const ZERO_BUFFER_SIZE: BufferAddress = 512 << 10;
 // See https://github.com/gfx-rs/wgpu/issues/4589. 60s to reduce the chances of this.
 const CLEANUP_WAIT_MS: u32 = 60000;
 
-const ENTRYPOINT_FAILURE_ERROR: &str = "The given EntryPoint is Invalid";
+pub(crate) const ENTRYPOINT_FAILURE_ERROR: &str = "The given EntryPoint is Invalid";
 
 pub type DeviceDescriptor<'a> = wgt::DeviceDescriptor<Label<'a>>;
 

wgpu-core/src/device/resource.rs

@@ -31,7 +31,7 @@ use crate::{
         UsageScopePool,
     },
     validation::{self, validate_color_attachment_bytes_per_sample},
-    FastHashMap, LabelHelpers as _, PreHashedKey, PreHashedMap,
+    FastHashMap, LabelHelpers, PreHashedKey, PreHashedMap,
 };
 
 use arrayvec::ArrayVec;
@@ -144,6 +144,9 @@ pub struct Device {
     #[cfg(feature = "trace")]
     pub(crate) trace: Mutex<Option<trace::Trace>>,
     pub(crate) usage_scopes: UsageScopePool,
+
+    #[cfg(feature = "indirect-validation")]
+    pub(crate) indirect_validation: Option<crate::indirect_validation::IndirectValidation>,
 }
 
 pub(crate) enum DeferredDestroy {
@@ -175,6 +178,11 @@ impl Drop for Device {
         let fence = unsafe { ManuallyDrop::take(&mut self.fence.write()) };
         pending_writes.dispose(raw.as_ref());
         self.command_allocator.dispose(raw.as_ref());
+        #[cfg(feature = "indirect-validation")]
+        self.indirect_validation
+            .take()
+            .unwrap()
+            .dispose(raw.as_ref());
         unsafe {
             raw.destroy_buffer(zero_buffer);
             raw.destroy_fence(fence);
@@ -261,6 +269,25 @@ impl Device {
         let alignments = adapter.raw.capabilities.alignments.clone();
         let downlevel = adapter.raw.capabilities.downlevel.clone();
 
+        #[cfg(feature = "indirect-validation")]
+        let indirect_validation = if downlevel
+            .flags
+            .contains(wgt::DownlevelFlags::INDIRECT_EXECUTION)
+        {
+            match crate::indirect_validation::IndirectValidation::new(
+                raw_device.as_ref(),
+                &desc.required_limits,
+            ) {
+                Ok(indirect_validation) => Some(indirect_validation),
+                Err(e) => {
+                    log::error!("indirect-validation error: {e:?}");
+                    return Err(DeviceError::Lost);
+                }
+            }
+        } else {
+            None
+        };
+
         Ok(Self {
             raw: ManuallyDrop::new(raw_device),
             adapter: adapter.clone(),
@@ -306,6 +333,8 @@ impl Device {
             ),
             deferred_destroy: Mutex::new(rank::DEVICE_DEFERRED_DESTROY, Vec::new()),
             usage_scopes: Mutex::new(rank::DEVICE_USAGE_SCOPES, Default::default()),
+            #[cfg(feature = "indirect-validation")]
+            indirect_validation,
         })
     }
 
@@ -547,6 +576,13 @@ impl Device {
 
         let mut usage = conv::map_buffer_usage(desc.usage);
 
+        if desc.usage.contains(wgt::BufferUsages::INDIRECT) {
+            self.require_downlevel_flags(wgt::DownlevelFlags::INDIRECT_EXECUTION)?;
+            // We are going to be reading from it, internally;
+            // when validating the content of the buffer
+            usage |= hal::BufferUses::STORAGE_READ | hal::BufferUses::STORAGE_READ_WRITE;
+        }
+
         if desc.mapped_at_creation {
             if desc.size % wgt::COPY_BUFFER_ALIGNMENT != 0 {
                 return Err(resource::CreateBufferError::UnalignedSize);
@@ -586,6 +622,10 @@ impl Device {
         let buffer =
             unsafe { self.raw().create_buffer(&hal_desc) }.map_err(|e| self.handle_hal_error(e))?;
 
+        #[cfg(feature = "indirect-validation")]
+        let raw_indirect_validation_bind_group =
+            self.create_indirect_validation_bind_group(buffer.as_ref(), desc.size, desc.usage)?;
+
         let buffer = Buffer {
             raw: Snatchable::new(buffer),
             device: self.clone(),
@@ -599,6 +639,8 @@ impl Device {
             label: desc.label.to_string(),
             tracking_data: TrackingData::new(self.tracker_indices.buffers.clone()),
             bind_groups: Mutex::new(rank::BUFFER_BIND_GROUPS, Vec::new()),
+            #[cfg(feature = "indirect-validation")]
+            raw_indirect_validation_bind_group,
         };
 
         let buffer = Arc::new(buffer);
@@ -686,7 +728,17 @@ impl Device {
         self: &Arc<Self>,
         hal_buffer: Box<dyn hal::DynBuffer>,
         desc: &resource::BufferDescriptor,
-    ) -> Fallible<Buffer> {
+    ) -> (Fallible<Buffer>, Option<resource::CreateBufferError>) {
+        #[cfg(feature = "indirect-validation")]
+        let raw_indirect_validation_bind_group = match self.create_indirect_validation_bind_group(
+            hal_buffer.as_ref(),
+            desc.size,
+            desc.usage,
+        ) {
+            Ok(ok) => ok,
+            Err(e) => return (Fallible::Invalid(Arc::new(desc.label.to_string())), Some(e)),
+        };
+
         unsafe { self.raw().add_raw_buffer(&*hal_buffer) };
 
         let buffer = Buffer {
@@ -702,6 +754,8 @@ impl Device {
             label: desc.label.to_string(),
             tracking_data: TrackingData::new(self.tracker_indices.buffers.clone()),
             bind_groups: Mutex::new(rank::BUFFER_BIND_GROUPS, Vec::new()),
+            #[cfg(feature = "indirect-validation")]
+            raw_indirect_validation_bind_group,
         };
 
         let buffer = Arc::new(buffer);
@@ -711,7 +765,25 @@ impl Device {
             .buffers
             .insert_single(&buffer, hal::BufferUses::empty());
 
-        Fallible::Valid(buffer)
+        (Fallible::Valid(buffer), None)
+    }
+
+    #[cfg(feature = "indirect-validation")]
+    fn create_indirect_validation_bind_group(
+        &self,
+        raw_buffer: &dyn hal::DynBuffer,
+        buffer_size: u64,
+        usage: wgt::BufferUsages,
+    ) -> Result<Snatchable<Box<dyn hal::DynBindGroup>>, resource::CreateBufferError> {
+        if usage.contains(wgt::BufferUsages::INDIRECT) {
+            let indirect_validation = self.indirect_validation.as_ref().unwrap();
+            let bind_group = indirect_validation
+                .create_src_bind_group(self.raw(), &self.limits, buffer_size, raw_buffer)
+                .map_err(resource::CreateBufferError::IndirectValidationBindGroup)?;
+            Ok(Snatchable::new(bind_group))
+        } else {
+            Ok(Snatchable::empty())
+        }
     }
 
     pub(crate) fn create_texture(

wgpu-core/src/indirect_validation.rs

@@ -0,0 +1,378 @@
+use thiserror::Error;
+
+use crate::{
+    device::DeviceError,
+    pipeline::{CreateComputePipelineError, CreateShaderModuleError},
+};
+
+#[derive(Clone, Debug, Error)]
+#[non_exhaustive]
+pub enum CreateDispatchIndirectValidationPipelineError {
+    #[error(transparent)]
+    DeviceError(#[from] DeviceError),
+    #[error(transparent)]
+    ShaderModule(#[from] CreateShaderModuleError),
+    #[error(transparent)]
+    ComputePipeline(#[from] CreateComputePipelineError),
+}
+
+/// This machinery requires the following limits:
+///
+/// - max_bind_groups: 2,
+/// - max_dynamic_storage_buffers_per_pipeline_layout: 1,
+/// - max_storage_buffers_per_shader_stage: 2,
+/// - max_storage_buffer_binding_size: 3 * min_storage_buffer_offset_alignment,
+/// - max_push_constant_size: 4,
+/// - max_compute_invocations_per_workgroup 1
+///
+/// These are all indirectly satisfied by `DownlevelFlags::INDIRECT_EXECUTION`, which is also
+/// required for this module's functionality to work.
+#[derive(Debug)]
+pub struct IndirectValidation {
+    module: Box<dyn hal::DynShaderModule>,
+    dst_bind_group_layout: Box<dyn hal::DynBindGroupLayout>,
+    src_bind_group_layout: Box<dyn hal::DynBindGroupLayout>,
+    pipeline_layout: Box<dyn hal::DynPipelineLayout>,
+    pipeline: Box<dyn hal::DynComputePipeline>,
+    dst_buffer: Box<dyn hal::DynBuffer>,
+    dst_bind_group: Box<dyn hal::DynBindGroup>,
+}
+
+pub struct Params<'a> {
+    pub pipeline_layout: &'a dyn hal::DynPipelineLayout,
+    pub pipeline: &'a dyn hal::DynComputePipeline,
+    pub dst_buffer: &'a dyn hal::DynBuffer,
+    pub dst_bind_group: &'a dyn hal::DynBindGroup,
+    pub aligned_offset: u64,
+    pub offset_remainder: u64,
+}
+
+impl IndirectValidation {
+    pub fn new(
+        device: &dyn hal::DynDevice,
+        limits: &wgt::Limits,
+    ) -> Result<Self, CreateDispatchIndirectValidationPipelineError> {
+        let max_compute_workgroups_per_dimension = limits.max_compute_workgroups_per_dimension;
+
+        let src = format!(
+            "
+            @group(0) @binding(0)
+            var<storage, read_write> dst: array<u32, 3>;
+            @group(1) @binding(0)
+            var<storage, read> src: array<u32>;
+            struct OffsetPc {{
+                inner: u32,
+            }}
+            var<push_constant> offset: OffsetPc;
+
+            @compute @workgroup_size(1)
+            fn main() {{
+                let src = vec3(src[offset.inner], src[offset.inner + 1], src[offset.inner + 2]);
+                let max_compute_workgroups_per_dimension = {max_compute_workgroups_per_dimension}u;
+                if (
+                    src.x > max_compute_workgroups_per_dimension ||
+                    src.y > max_compute_workgroups_per_dimension ||
+                    src.z > max_compute_workgroups_per_dimension
+                ) {{
+                    dst = array(0u, 0u, 0u);
+                }} else {{
+                    dst = array(src.x, src.y, src.z);
+                }}
+            }}
+        "
+        );
+
+        let module = naga::front::wgsl::parse_str(&src).map_err(|inner| {
+            CreateShaderModuleError::Parsing(naga::error::ShaderError {
+                source: src.clone(),
+                label: None,
+                inner: Box::new(inner),
+            })
+        })?;
+        let info = crate::device::create_validator(
+            wgt::Features::PUSH_CONSTANTS,
+            wgt::DownlevelFlags::empty(),
+            naga::valid::ValidationFlags::all(),
+        )
+        .validate(&module)
+        .map_err(|inner| {
+            CreateShaderModuleError::Validation(naga::error::ShaderError {
+                source: src,
+                label: None,
+                inner: Box::new(inner),
+            })
+        })?;
+        let hal_shader = hal::ShaderInput::Naga(hal::NagaShader {
+            module: std::borrow::Cow::Owned(module),
+            info,
+            debug_source: None,
+        });
+        let hal_desc = hal::ShaderModuleDescriptor {
+            label: None,
+            runtime_checks: false,
+        };
+        let module =
+            unsafe { device.create_shader_module(&hal_desc, hal_shader) }.map_err(|error| {
+                match error {
+                    hal::ShaderError::Device(error) => {
+                        CreateShaderModuleError::Device(DeviceError::from_hal(error))
+                    }
+                    hal::ShaderError::Compilation(ref msg) => {
+                        log::error!("Shader error: {}", msg);
+                        CreateShaderModuleError::Generation
+                    }
+                }
+            })?;
+
+        let dst_bind_group_layout_desc = hal::BindGroupLayoutDescriptor {
+            label: None,
+            flags: hal::BindGroupLayoutFlags::empty(),
+            entries: &[wgt::BindGroupLayoutEntry {
+                binding: 0,
+                visibility: wgt::ShaderStages::COMPUTE,
+                ty: wgt::BindingType::Buffer {
+                    ty: wgt::BufferBindingType::Storage { read_only: false },
+                    has_dynamic_offset: false,
+                    min_binding_size: Some(std::num::NonZeroU64::new(4 * 3).unwrap()),
+                },
+                count: None,
+            }],
+        };
+        let dst_bind_group_layout = unsafe {
+            device
+                .create_bind_group_layout(&dst_bind_group_layout_desc)
+                .map_err(DeviceError::from_hal)?
+        };
+
+        let src_bind_group_layout_desc = hal::BindGroupLayoutDescriptor {
+            label: None,
+            flags: hal::BindGroupLayoutFlags::empty(),
+            entries: &[wgt::BindGroupLayoutEntry {
+                binding: 0,
+                visibility: wgt::ShaderStages::COMPUTE,
+                ty: wgt::BindingType::Buffer {
+                    ty: wgt::BufferBindingType::Storage { read_only: true },
+                    has_dynamic_offset: true,
+                    min_binding_size: Some(std::num::NonZeroU64::new(4 * 3).unwrap()),
+                },
+                count: None,
+            }],
+        };
+        let src_bind_group_layout = unsafe {
+            device
+                .create_bind_group_layout(&src_bind_group_layout_desc)
+                .map_err(DeviceError::from_hal)?
+        };
+
+        let pipeline_layout_desc = hal::PipelineLayoutDescriptor {
+            label: None,
+            flags: hal::PipelineLayoutFlags::FIRST_VERTEX_INSTANCE,
+            bind_group_layouts: &[
+                dst_bind_group_layout.as_ref(),
+                src_bind_group_layout.as_ref(),
+            ],
+            push_constant_ranges: &[wgt::PushConstantRange {
+                stages: wgt::ShaderStages::COMPUTE,
+                range: 0..4,
+            }],
+        };
+        let pipeline_layout = unsafe {
+            device
+                .create_pipeline_layout(&pipeline_layout_desc)
+                .map_err(DeviceError::from_hal)?
+        };
+
+        let pipeline_desc = hal::ComputePipelineDescriptor {
+            label: None,
+            layout: pipeline_layout.as_ref(),
+            stage: hal::ProgrammableStage {
+                module: module.as_ref(),
+                entry_point: "main",
+                constants: &Default::default(),
+                zero_initialize_workgroup_memory: false,
+            },
+            cache: None,
+        };
+        let pipeline =
+            unsafe { device.create_compute_pipeline(&pipeline_desc) }.map_err(|err| match err {
+                hal::PipelineError::Device(error) => {
+                    CreateComputePipelineError::Device(DeviceError::from_hal(error))
+                }
+                hal::PipelineError::Linkage(_stages, msg) => {
+                    CreateComputePipelineError::Internal(msg)
+                }
+                hal::PipelineError::EntryPoint(_stage) => CreateComputePipelineError::Internal(
+                    crate::device::ENTRYPOINT_FAILURE_ERROR.to_string(),
+                ),
+                hal::PipelineError::PipelineConstants(_, error) => {
+                    CreateComputePipelineError::PipelineConstants(error)
+                }
+            })?;
+
+        let dst_buffer_desc = hal::BufferDescriptor {
+            label: None,
+            size: 4 * 3,
+            usage: hal::BufferUses::INDIRECT | hal::BufferUses::STORAGE_READ_WRITE,
+            memory_flags: hal::MemoryFlags::empty(),
+        };
+        let dst_buffer =
+            unsafe { device.create_buffer(&dst_buffer_desc) }.map_err(DeviceError::from_hal)?;
+
+        let dst_bind_group_desc = hal::BindGroupDescriptor {
+            label: None,
+            layout: dst_bind_group_layout.as_ref(),
+            entries: &[hal::BindGroupEntry {
+                binding: 0,
+                resource_index: 0,
+                count: 1,
+            }],
+            buffers: &[hal::BufferBinding {
+                buffer: dst_buffer.as_ref(),
+                offset: 0,
+                size: Some(std::num::NonZeroU64::new(4 * 3).unwrap()),
+            }],
+            samplers: &[],
+            textures: &[],
+            acceleration_structures: &[],
+        };
+        let dst_bind_group = unsafe {
+            device
+                .create_bind_group(&dst_bind_group_desc)
+                .map_err(DeviceError::from_hal)
+        }?;
+
+        Ok(Self {
+            module,
+            dst_bind_group_layout,
+            src_bind_group_layout,
+            pipeline_layout,
+            pipeline,
+            dst_buffer,
+            dst_bind_group,
+        })
+    }
+
+    pub fn create_src_bind_group(
+        &self,
+        device: &dyn hal::DynDevice,
+        limits: &wgt::Limits,
+        buffer_size: u64,
+        buffer: &dyn hal::DynBuffer,
+    ) -> Result<Box<dyn hal::DynBindGroup>, DeviceError> {
+        let binding_size = calculate_src_buffer_binding_size(buffer_size, limits);
+        let hal_desc = hal::BindGroupDescriptor {
+            label: None,
+            layout: self.src_bind_group_layout.as_ref(),
+            entries: &[hal::BindGroupEntry {
+                binding: 0,
+                resource_index: 0,
+                count: 1,
+            }],
+            buffers: &[hal::BufferBinding {
+                buffer,
+                offset: 0,
+                size: Some(std::num::NonZeroU64::new(binding_size).unwrap()),
+            }],
+            samplers: &[],
+            textures: &[],
+            acceleration_structures: &[],
+        };
+        unsafe {
+            device
+                .create_bind_group(&hal_desc)
+                .map_err(DeviceError::from_hal)
+        }
+    }
+
+    pub fn params<'a>(&'a self, limits: &wgt::Limits, offset: u64, buffer_size: u64) -> Params<'a> {
+        // The offset we receive is only required to be aligned to 4 bytes.
+        //
+        // Binding offsets and dynamic offsets are required to be aligned to
+        // min_storage_buffer_offset_alignment (256 bytes by default).
+        //
+        // So, we work around this limitation by calculating an aligned offset
+        // and pass the remainder through a push constant.
+        //
+        // We could bind the whole buffer and only have to pass the offset
+        // through a push constant but we might run into the
+        // max_storage_buffer_binding_size limit.
+        //
+        // See the inner docs of `calculate_src_buffer_binding_size` to
+        // see how we get the appropriate `binding_size`.
+        let alignment = limits.min_storage_buffer_offset_alignment as u64;
+        let binding_size = calculate_src_buffer_binding_size(buffer_size, limits);
+        let aligned_offset = offset - offset % alignment;
+        // This works because `binding_size` is either `buffer_size` or `alignment * 2 + buffer_size % alignment`.
+        let max_aligned_offset = buffer_size - binding_size;
+        let aligned_offset = aligned_offset.min(max_aligned_offset);
+        let offset_remainder = offset - aligned_offset;
+
+        Params {
+            pipeline_layout: self.pipeline_layout.as_ref(),
+            pipeline: self.pipeline.as_ref(),
+            dst_buffer: self.dst_buffer.as_ref(),
+            dst_bind_group: self.dst_bind_group.as_ref(),
+            aligned_offset,
+            offset_remainder,
+        }
+    }
+
+    pub fn dispose(self, device: &dyn hal::DynDevice) {
+        let IndirectValidation {
+            module,
+            dst_bind_group_layout,
+            src_bind_group_layout,
+            pipeline_layout,
+            pipeline,
+            dst_buffer,
+            dst_bind_group,
+        } = self;
+
+        unsafe {
+            device.destroy_bind_group(dst_bind_group);
+            device.destroy_buffer(dst_buffer);
+            device.destroy_compute_pipeline(pipeline);
+            device.destroy_pipeline_layout(pipeline_layout);
+            device.destroy_bind_group_layout(src_bind_group_layout);
+            device.destroy_bind_group_layout(dst_bind_group_layout);
+            device.destroy_shader_module(module);
+        }
+    }
+}
+
+fn calculate_src_buffer_binding_size(buffer_size: u64, limits: &wgt::Limits) -> u64 {
+    let alignment = limits.min_storage_buffer_offset_alignment as u64;
+
+    // We need to choose a binding size that can address all possible sets of 12 contiguous bytes in the buffer taking
+    // into account that the dynamic offset needs to be a multiple of `min_storage_buffer_offset_alignment`.
+
+    // Given the know variables: `offset`, `buffer_size`, `alignment` and the rule `offset + 12 <= buffer_size`.
+
+    // Let `chunks = floor(buffer_size / alignment)`.
+    // Let `chunk` be the interval `[0, chunks]`.
+    // Let `offset = alignment * chunk + r` where `r` is the interval [0, alignment - 4].
+    // Let `binding` be the interval `[offset, offset + 12]`.
+    // Let `aligned_offset = alignment * chunk`.
+    // Let `aligned_binding` be the interval `[aligned_offset, aligned_offset + r + 12]`.
+    // Let `aligned_binding_size = r + 12 = [12, alignment + 8]`.
+    // Let `min_aligned_binding_size = alignment + 8`.
+
+    // `min_aligned_binding_size` is the minimum binding size required to address all 12 contiguous bytes in the buffer
+    // but the last aligned_offset + min_aligned_binding_size might overflow the buffer. In order to avoid this we must
+    // pick a larger `binding_size` that satisfies: `last_aligned_offset + binding_size = buffer_size` and
+    // `binding_size >= min_aligned_binding_size`.
+
+    // Let `buffer_size = alignment * chunks + sr` where `sr` is the interval [0, alignment - 4].
+    // Let `last_aligned_offset = alignment * (chunks - u)` where `u` is the interval [0, chunks].
+    // => `binding_size = buffer_size - last_aligned_offset`
+    // => `binding_size = alignment * chunks + sr - alignment * (chunks - u)`
+    // => `binding_size = alignment * chunks + sr - alignment * chunks + alignment * u`
+    // => `binding_size = sr + alignment * u`
+    // => `min_aligned_binding_size <= sr + alignment * u`
+    // => `alignment + 8 <= sr + alignment * u`
+    // => `u` must be at least 2
+    // => `binding_size = sr + alignment * 2`
+
+    let binding_size = 2 * alignment + (buffer_size % alignment);
+    binding_size.min(buffer_size)
+}

wgpu-core/src/lib.rs

@@ -67,6 +67,8 @@ mod hash_utils;
 pub mod hub;
 pub mod id;
 pub mod identity;
+#[cfg(feature = "indirect-validation")]
+mod indirect_validation;
 mod init_tracker;
 pub mod instance;
 mod lock;

wgpu-core/src/pipeline.rs

@@ -92,7 +92,7 @@ impl ShaderModule {
 #[derive(Clone, Debug, Error)]
 #[non_exhaustive]
 pub enum CreateShaderModuleError {
-    #[cfg(feature = "wgsl")]
+    #[cfg(any(feature = "wgsl", feature = "indirect-validation"))]
     #[error(transparent)]
     Parsing(#[from] ShaderError<naga::front::wgsl::ParseError>),
     #[cfg(feature = "glsl")]

wgpu-core/src/resource.rs

@@ -475,10 +475,18 @@ pub struct Buffer {
     pub(crate) tracking_data: TrackingData,
     pub(crate) map_state: Mutex<BufferMapState>,
     pub(crate) bind_groups: Mutex<Vec<Weak<BindGroup>>>,
+    #[cfg(feature = "indirect-validation")]
+    pub(crate) raw_indirect_validation_bind_group: Snatchable<Box<dyn hal::DynBindGroup>>,
 }
 
 impl Drop for Buffer {
     fn drop(&mut self) {
+        #[cfg(feature = "indirect-validation")]
+        if let Some(raw) = self.raw_indirect_validation_bind_group.take() {
+            unsafe {
+                self.device.raw().destroy_bind_group(raw);
+            }
+        }
         if let Some(raw) = self.raw.take() {
             resource_log!("Destroy raw {}", self.error_ident());
             unsafe {
@@ -737,13 +745,22 @@ impl Buffer {
         let device = &self.device;
 
         let temp = {
-            let raw = match self.raw.snatch(&mut device.snatchable_lock.write()) {
+            let mut snatch_guard = device.snatchable_lock.write();
+
+            let raw = match self.raw.snatch(&mut snatch_guard) {
                 Some(raw) => raw,
                 None => {
                     return Err(DestroyError::AlreadyDestroyed);
                 }
             };
 
+            #[cfg(feature = "indirect-validation")]
+            let raw_indirect_validation_bind_group = self
+                .raw_indirect_validation_bind_group
+                .snatch(&mut snatch_guard);
+
+            drop(snatch_guard);
+
             let bind_groups = {
                 let mut guard = self.bind_groups.lock();
                 mem::take(&mut *guard)
@@ -754,6 +771,8 @@ impl Buffer {
                 device: Arc::clone(&self.device),
                 label: self.label().to_owned(),
                 bind_groups,
+                #[cfg(feature = "indirect-validation")]
+                raw_indirect_validation_bind_group,
             })
         };
 
@@ -789,6 +808,8 @@ pub enum CreateBufferError {
     MaxBufferSize { requested: u64, maximum: u64 },
     #[error(transparent)]
     MissingDownlevelFlags(#[from] MissingDownlevelFlags),
+    #[error("Failed to create bind group for indirect buffer validation: {0}")]
+    IndirectValidationBindGroup(DeviceError),
 }
 
 crate::impl_resource_type!(Buffer);
@@ -804,6 +825,8 @@ pub struct DestroyedBuffer {
     device: Arc<Device>,
     label: String,
     bind_groups: Vec<Weak<BindGroup>>,
+    #[cfg(feature = "indirect-validation")]
+    raw_indirect_validation_bind_group: Option<Box<dyn hal::DynBindGroup>>,
 }
 
 impl DestroyedBuffer {
@@ -820,6 +843,13 @@ impl Drop for DestroyedBuffer {
         }
         drop(deferred);
 
+        #[cfg(feature = "indirect-validation")]
+        if let Some(raw) = self.raw_indirect_validation_bind_group.take() {
+            unsafe {
+                self.device.raw().destroy_bind_group(raw);
+            }
+        }
+
         resource_log!("Destroy raw Buffer (destroyed) {:?}", self.label());
         // SAFETY: We are in the Drop impl and we don't use self.raw anymore after this point.
         let raw = unsafe { ManuallyDrop::take(&mut self.raw) };

wgpu-core/src/snatch.rs

@@ -32,6 +32,12 @@ impl<T> Snatchable<T> {
         }
     }
 
+    pub fn empty() -> Self {
+        Snatchable {
+            value: UnsafeCell::new(None),
+        }
+    }
+
     /// Get read access to the value. Requires a the snatchable lock's read guard.
     pub fn get<'a>(&'a self, _guard: &'a SnatchGuard) -> Option<&'a T> {
         unsafe { (*self.value.get()).as_ref() }

wgpu/Cargo.toml

@@ -130,6 +130,12 @@ features = ["raw-window-handle"]
 workspace = true
 features = ["raw-window-handle"]
 
+# If we are not targeting WebGL, enable indirect-validation.
+# WebGL doesn't support indirect execution so this is not needed.
+[target.'cfg(not(target_arch = "wasm32"))'.dependencies.wgc]
+workspace = true
+features = ["indirect-validation"]
+
 # Enable `wgc` by default on macOS and iOS to allow the `metal` crate feature to
 # enable the Metal backend while being no-op on other targets.
 [target.'cfg(any(target_os = "macos", target_os = "ios"))'.dependencies.wgc]