build: Release v51.0

This release contains security fixes. Details can be found in the
release notes.

Signed-off-by: Bo Chen <bchen@crusoe.ai>

.config/nextest.toml

@@ -0,0 +1,3 @@
+[profile.default]
+# Don't let one individual test run for more than 10 minutes
+slow-timeout = { period = "60s", terminate-after = 10 }

.github/dependabot.yml

@@ -1,21 +1,68 @@
 version: 2
 updates:
 - package-ecosystem: cargo
-  directory: "/"
+  directories:
+    - "/"
+    - "/fuzz"
   schedule:
-    interval: daily
-  open-pull-requests-limit: 1
+    interval: weekly
   allow:
-  - dependency-type: direct
-  - dependency-type: indirect
+    - dependency-name: "acpi_tables"
+    - dependency-name: "kvm-bindings"
+    - dependency-name: "kvm-ioctls"
+    - dependency-name: "linux-loader"
+    - dependency-name: "micro_http"
+    - dependency-name: "mshv-bindings"
+    - dependency-name: "mshv-ioctls"
+    - dependency-name: "seccompiler"
+    - dependency-name: "vfio-bindings"
+    - dependency-name: "vfio-ioctls"
+    - dependency-name: "vfio_user"
+    - dependency-name: "vhost"
+    - dependency-name: "vhost-user-backend"
+    - dependency-name: "virtio-bindings"
+    - dependency-name: "virtio-queue"
+    - dependency-name: "vm-fdt"
+    - dependency-name: "vm-memory"
+    - dependency-name: "vmm-sys-util"
+  groups:
+      rust-vmm:
+        patterns:
+          - "*"
 - package-ecosystem: cargo
-  directory: "/fuzz"
+  directories:
+    - "/"
+    - "/fuzz"
   schedule:
-    interval: daily
-  open-pull-requests-limit: 1
+    interval: weekly
   allow:
-  - dependency-type: direct
-  - dependency-type: indirect
+    - dependency-type: all
+  ignore:
+    - dependency-name: "acpi_tables"
+    - dependency-name: "kvm-bindings"
+    - dependency-name: "kvm-ioctls"
+    - dependency-name: "linux-loader"
+    - dependency-name: "micro_http"
+    - dependency-name: "mshv-bindings"
+    - dependency-name: "mshv-ioctls"
+    - dependency-name: "seccompiler"
+    - dependency-name: "vfio-bindings"
+    - dependency-name: "vfio-ioctls"
+    - dependency-name: "vfio_user"
+    - dependency-name: "vhost"
+    - dependency-name: "vhost-user-backend"
+    - dependency-name: "virtio-bindings"
+    - dependency-name: "virtio-queue"
+    - dependency-name: "vm-fdt"
+    - dependency-name: "vm-memory"
+    - dependency-name: "vmm-sys-util"
+  groups:
+      non-rust-vmm:
+        patterns:
+          - "*"
+  # Makes it possible to have another config for the same directory.
+  # https://github.com/dependabot/dependabot-core/issues/1778#issuecomment-1988140219
+  target-branch: main
 - package-ecosystem: github-actions
   directory: "/"
   schedule:

.github/workflows/audit.yaml

@@ -10,7 +10,7 @@ jobs:
     name: Audit
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
       - uses: actions-rust-lang/audit@v1
         with:
           token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/build.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor Build
 on: [pull_request, merge_group]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -15,13 +15,13 @@ jobs:
           - stable
           - beta
           - nightly
-          - "1.83.0"
+          - "1.89.0"
         target:
           - x86_64-unknown-linux-gnu
           - x86_64-unknown-linux-musl
     steps:
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
 
@@ -35,34 +35,40 @@ jobs:
           target: ${{ matrix.target }}
 
       - name: Build (default features)
-        run: cargo rustc --locked --bin cloud-hypervisor -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: cargo build --locked --bin cloud-hypervisor
 
       - name: Build (kvm)
-        run: cargo rustc --locked --bin cloud-hypervisor --no-default-features --features "kvm"  -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: cargo build --locked --bin cloud-hypervisor --no-default-features --features "kvm"
 
       - name: Build (default features + tdx)
-        run: cargo rustc --locked --bin cloud-hypervisor --features "tdx" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: cargo build --locked --bin cloud-hypervisor --features "tdx"
 
       - name: Build (default features + dbus_api)
-        run: cargo rustc --locked --bin cloud-hypervisor --features "dbus_api" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: cargo build --locked --bin cloud-hypervisor --features "dbus_api"
 
       - name: Build (default features + guest_debug)
-        run: cargo rustc --locked --bin cloud-hypervisor --features "guest_debug" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: cargo build --locked --bin cloud-hypervisor --features "guest_debug"
 
       - name: Build (default features + pvmemcontrol)
-        run: cargo rustc --locked --bin cloud-hypervisor --features "pvmemcontrol" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: cargo build --locked --bin cloud-hypervisor --features "pvmemcontrol"
+
+      - name: Build (default features + fw_cfg)
+        run: cargo build --locked --bin cloud-hypervisor --features "fw_cfg"
+
+      - name: Build (default features + ivshmem)
+        run: cargo build --locked --bin cloud-hypervisor --features "ivshmem"
 
       - name: Build (mshv)
-        run: cargo rustc --locked --bin cloud-hypervisor --no-default-features --features "mshv"  -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: cargo build --locked --bin cloud-hypervisor --no-default-features --features "mshv"
 
       - name: Build (sev_snp)
-        run: cargo rustc --locked --bin cloud-hypervisor --no-default-features --features "sev_snp"  -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: cargo build --locked --bin cloud-hypervisor --no-default-features --features "sev_snp"
 
       - name: Build (igvm)
-        run: cargo rustc --locked --bin cloud-hypervisor --no-default-features --features "igvm"  -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: cargo build --locked --bin cloud-hypervisor --no-default-features --features "igvm"
 
       - name: Build (mshv + kvm)
-        run: cargo rustc --locked --bin cloud-hypervisor --no-default-features --features "mshv,kvm"  -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: cargo build --locked --bin cloud-hypervisor --no-default-features --features "mshv,kvm"
 
       - name: Release Build (default features)
         run: cargo build --locked --all --release --target=${{ matrix.target }}

.github/workflows/dco.yaml

@@ -3,12 +3,12 @@ on: [pull_request, merge_group]
 
 jobs:
   check:
-    name: DCO Check ("Signed-Off-By")
+    name: DCO Check ("Signed-off-by")
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v6
     - name: Set up Python 3.x
-      uses: actions/setup-python@v5
+      uses: actions/setup-python@v6
       with:
         python-version: '3.x'
     - name: Check DCO

.github/workflows/docker-image.yaml

@@ -6,7 +6,7 @@ on:
   pull_request:
     paths: resources/Dockerfile
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 env:
@@ -18,7 +18,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
 
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v3
@@ -41,7 +41,7 @@ jobs:
           images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
           # generate Docker tags based on the following events/attributes
           tags: |
-            type=raw,value=20250412-0
+            type=raw,value=20251114-0
             type=sha
 
       - name: Build and push

.github/workflows/formatting.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor Code Formatting
 on: [pull_request, merge_group]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -19,7 +19,7 @@ jobs:
       RUSTFLAGS: -D warnings
     steps:
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
       - name: Install Rust toolchain (${{ matrix.rust }})
         uses: dtolnay/rust-toolchain@stable
         with:

.github/workflows/fuzz-build.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor Cargo Fuzz Build
 on: [pull_request, merge_group]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -18,7 +18,7 @@ jobs:
       RUSTFLAGS: -D warnings
     steps:
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
       - name: Install Rust toolchain (${{ matrix.rust }})
         uses: dtolnay/rust-toolchain@stable
         with:

.github/workflows/gitlint.yaml

@@ -8,12 +8,12 @@ jobs:
     runs-on: ubuntu-latest
     steps:
     - name: Checkout repository
-      uses: actions/checkout@v4
+      uses: actions/checkout@v6
       with:
         ref: ${{ github.event.pull_request.head.sha }}
         fetch-depth: 0
     - name: Set up Python 3.10
-      uses: actions/setup-python@v5
+      uses: actions/setup-python@v6
       with:
         python-version: "3.10"
     - name: Install dependencies

.github/workflows/hadolint.yaml

@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout code
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
 
       - name: Lint Dockerfile
         uses: hadolint/hadolint-action@master

.github/workflows/integration-arm64.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor Tests (ARM64)
 on: [pull_request, merge_group]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -13,7 +13,7 @@ jobs:
       - name: Fix workspace permissions
         run: sudo chown -R runner:runner ${GITHUB_WORKSPACE}
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
       - name: Run unit tests (musl)

.github/workflows/integration-metrics.yaml

@@ -12,7 +12,7 @@ jobs:
       METRICS_PUBLISH_KEY: ${{ secrets.METRICS_PUBLISH_KEY }}
     steps:
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
       - name: Run metrics tests

.github/workflows/integration-rate-limiter.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor Tests (Rate-Limiter)
 on: [merge_group, pull_request]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -13,7 +13,7 @@ jobs:
     steps:
       - name: Code checkout
         if: ${{ github.event_name != 'pull_request' }}
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
       - name: Run rate-limiter integration tests

.github/workflows/integration-vfio.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor Tests (VFIO)
 on: [merge_group, pull_request]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -16,7 +16,7 @@ jobs:
         run: sudo chown -R runner:runner ${GITHUB_WORKSPACE}
       - name: Code checkout
         if: ${{ github.event_name != 'pull_request' }}
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
       - name: Run VFIO integration tests

.github/workflows/integration-windows.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor Tests (Windows Guest)
 on: [merge_group, pull_request]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -11,7 +11,7 @@ jobs:
     steps:
       - name: Code checkout
         if: ${{ github.event_name != 'pull_request' }}
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
       - name: Install Docker

.github/workflows/integration-x86-64.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor Tests (x86-64)
 on: [pull_request, merge_group]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -17,7 +17,7 @@ jobs:
     steps:
       - name: Code checkout
         if: ${{ github.event_name != 'pull_request' || (matrix.runner == 'garm-jammy' && matrix.libc == 'gnu') }}
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
       - name: Install Docker

.github/workflows/lychee.yaml

@@ -1,16 +1,45 @@
 name: Link Check (lychee)
-on:
-  pull_request
-
+on: pull_request
 jobs:
   link_check:
     name: Link Check
     runs-on: ubuntu-latest
     steps:
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
+        with:
+          # Fetch the entire history so git diff can compare against the base branch
+          fetch-depth: 0
+      - name: Get changed files in PR
+        id: changed-files
+        uses: tj-actions/changed-files@v47 # Using a dedicated action for robustness
+        with:
+          # Compare the HEAD of the PR with the merge-base (where the PR branches off)
+          base_sha: ${{ github.event.pull_request.base.sha }}
 
-      - name: Link Availability Check
+      # NEW STEP: Print all changed-files outputs for verification
+      - name: Verify Changed Files
+        run: |
+          echo "--- tj-actions/changed-files Outputs ---"
+          echo "any_changed: ${{ steps.changed-files.outputs.any_changed }}"
+          echo "all_changed_files: ${{ steps.changed-files.outputs.all_changed_files }}"
+          echo "added_files: ${{ steps.changed-files.outputs.added_files }}"
+          echo "modified_files: ${{ steps.changed-files.outputs.modified_files }}"
+          echo "deleted_files: ${{ steps.changed-files.outputs.deleted_files }}"
+          echo "renamed_files: ${{ steps.changed-files.outputs.renamed_files }}"
+          echo "----------------------------------------"
+          # This will also show if the all_changed_files string is empty or not
+          if [ -n "${{ steps.changed-files.outputs.all_changed_files }}" ]; then
+            echo "Detected changes: all_changed_files output is NOT empty."
+          else
+            echo "No changes detected: all_changed_files output IS empty."
+          fi
+      - name: Link Availability Check (Diff Only)
+        # MODIFIED: Only run lychee if the 'all_changed_files' output is not an empty string
+        if: ${{ steps.changed-files.outputs.all_changed_files != '' }}
         uses: lycheeverse/lychee-action@master
         with:
-          args: --verbose --config .lychee.toml .
+          # Pass the space-separated list of changed files to lychee
+          args: --verbose --config .lychee.toml ${{ steps.changed-files.outputs.all_changed_files }}
+          failIfEmpty: false
+          fail: true

.github/workflows/mshv-infra.yaml

@@ -0,0 +1,248 @@
+name: MSHV Infra Setup
+on:
+  workflow_call:
+    inputs:
+      ARCH:
+        description: 'Architecture for the VM'
+        required: true
+        type: string
+      KEY:
+        description: 'SSH Key Name'
+        required: true
+        type: string
+      OS_DISK_SIZE:
+        description: 'OS Disk Size in GB'
+        required: true
+        type: string
+      RG:
+        description: 'Resource Group Name'
+        required: true
+        type: string
+      VM_SKU:
+        description: 'VM SKU'
+        required: true
+        type: string
+    secrets:
+      MI_CLIENT_ID:
+        required: true
+      RUNNER_RG:
+        required: true
+      STORAGE_ACCOUNT_PATHS:
+        required: true
+      ARCH_SOURCE_PATH:
+        required: true
+      USERNAME:
+        required: true
+    outputs:
+      RG_NAME:
+        description: 'Resource group of the VM'
+        value: ${{ jobs.infra-setup.outputs.RG_NAME }}
+      VM_NAME:
+        description: 'Name of the VM'
+        value: ${{ jobs.infra-setup.outputs.VM_NAME }}
+      PRIVATE_IP:
+        description: 'Private IP of the VM'
+        value: ${{ jobs.infra-setup.outputs.PRIVATE_IP }}
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
+  cancel-in-progress: true
+jobs:
+  infra-setup:
+    name: ${{ inputs.ARCH }} VM Provision
+    runs-on: mshv
+    outputs:
+      RG_NAME: ${{ steps.rg-setup.outputs.RG_NAME }}
+      VM_NAME: ${{ steps.vm-setup.outputs.VM_NAME }}
+      PRIVATE_IP: ${{ steps.get-vm-ip.outputs.PRIVATE_IP }}
+    steps:
+      - name: Install & login to AZ CLI
+        env:
+          MI_CLIENT_ID: ${{ secrets.MI_CLIENT_ID }}
+        run: |
+          set -e
+          echo "Installing Azure CLI if not already installed"
+          if ! command -v az &>/dev/null; then
+            curl -sL https://aka.ms/InstallAzureCLIDeb | sudo bash
+          else
+            echo "Azure CLI already installed"
+          fi
+          az --version
+          echo "Logging into Azure CLI using Managed Identity"
+          az login --identity --client-id ${MI_CLIENT_ID}
+
+      - name: Get Location
+        id: get-location
+        env:
+          SKU: ${{ inputs.VM_SKU }}
+          STORAGE_ACCOUNT_PATHS: ${{ secrets.STORAGE_ACCOUNT_PATHS }}
+        run: |
+          set -e
+          # Extract vCPU count from SKU (e.g., "Standard_D2s_v3" => 2)
+          vcpu=$(echo "$SKU" | sed -n 's/^Standard_[A-Za-z]\+\([0-9]\+\).*/\1/p')
+          if [[ -z "$vcpu" ]]; then
+            echo "Cannot extract vCPU count from SKU: $SKU"
+            exit 1
+          fi
+
+          SUPPORTED_LOCATIONS=$(echo "$STORAGE_ACCOUNT_PATHS" | jq -r 'to_entries[] | .key')
+
+          for location in $SUPPORTED_LOCATIONS; do
+            family=$(az vm list-skus --size "$SKU" --location "$location" --resource-type "virtualMachines" --query '[0].family' -o tsv)
+            if [[ -z "$family" ]]; then
+              echo "Cannot determine VM family for SKU: $SKU in $location"
+              continue
+            fi
+
+            usage=$(az vm list-usage --location "$location" --query "[?name.value=='$family'] | [0]" -o json)
+            current=$(echo "$usage" | jq -r '.currentValue')
+            limit=$(echo "$usage" | jq -r '.limit')
+
+            if [[ $((limit - current)) -ge $vcpu ]]; then
+              echo "Sufficient quota found in $location"
+              echo "location=$location" >> "$GITHUB_OUTPUT"
+              exit 0
+            fi
+          done
+
+          echo "No location found with sufficient vCPU quota for SKU: $SKU"
+          exit 1
+
+      - name: Create Resource Group
+        id: rg-setup
+        env:
+          LOCATION: ${{ steps.get-location.outputs.location }}
+          RG: ${{ inputs.RG }}
+          STORAGE_ACCOUNT_PATHS: ${{ secrets.STORAGE_ACCOUNT_PATHS }}
+        run: |
+          set -e
+          echo "Creating Resource Group: $RG"
+          # Create the resource group
+          echo "Creating resource group in location: ${LOCATION}"
+          az group create --name ${RG} --location ${LOCATION}
+          echo "RG_NAME=${RG}" >> $GITHUB_OUTPUT
+          echo "Resource group created successfully."
+
+      - name: Generate SSH Key
+        id: generate-ssh-key
+        env:
+          KEY: ${{ inputs.KEY }}
+        run: |
+          set -e
+          echo "Generating SSH key: $KEY"
+          mkdir -p ~/.ssh
+          ssh-keygen -t rsa -b 4096 -f ~/.ssh/${KEY} -N ""
+
+      - name: Create VM
+        id: vm-setup
+        env:
+          KEY: ${{ inputs.KEY }}
+          LOCATION: ${{ steps.get-location.outputs.location }}
+          OS_DISK_SIZE: ${{ inputs.OS_DISK_SIZE }}
+          RG: ${{ inputs.RG }}
+          RUNNER_RG: ${{ secrets.RUNNER_RG }}
+          USERNAME: ${{ secrets.USERNAME }}
+          VM_SKU: ${{ inputs.VM_SKU }}
+          VM_IMAGE_NAME: ${{ inputs.ARCH }}_${{ steps.get-location.outputs.location }}_image
+          VM_NAME: ${{ inputs.ARCH }}_${{ steps.get-location.outputs.location }}_${{ github.run_id }}
+        run: |
+          set -e
+          echo "Creating $VM_SKU VM: $VM_NAME"
+
+          # Extract subnet ID from the runner VM
+          echo "Retrieving subnet ID..."
+          SUBNET_ID=$(az network vnet list --resource-group ${RUNNER_RG} --query "[?contains(location, '${LOCATION}')].{SUBNETS:subnets}" | jq -r ".[0].SUBNETS[0].id")
+          if [[ -z "${SUBNET_ID}" ]]; then
+            echo "ERROR: Failed to retrieve Subnet ID."
+            exit 1
+          fi
+
+          # Extract image ID from the runner VM
+          echo "Retrieving image ID..."
+          IMAGE_ID=$(az image show --resource-group ${RUNNER_RG} --name ${VM_IMAGE_NAME} --query "id" -o tsv)
+          if [[ -z "${IMAGE_ID}" ]]; then
+            echo "ERROR: Failed to retrieve Image ID."
+            exit 1
+          fi
+
+          # Create VM
+          az vm create \
+            --resource-group ${RG} \
+            --name ${VM_NAME} \
+            --subnet ${SUBNET_ID} \
+            --size ${VM_SKU} \
+            --location ${LOCATION} \
+            --image ${IMAGE_ID} \
+            --os-disk-size-gb ${OS_DISK_SIZE} \
+            --public-ip-sku Standard \
+            --storage-sku Premium_LRS \
+            --public-ip-address "" \
+            --admin-username ${USERNAME} \
+            --ssh-key-value  ~/.ssh/${KEY}.pub \
+            --security-type Standard \
+            --output json
+
+          az vm boot-diagnostics enable --name ${VM_NAME} --resource-group ${RG}
+
+          echo "VM_NAME=${VM_NAME}" >> $GITHUB_OUTPUT
+          echo "VM creation process completed successfully."
+
+      - name: Get VM Private IP
+        id: get-vm-ip
+        env:
+          RG: ${{ inputs.RG }}
+          VM_NAME: ${{ inputs.ARCH }}_${{ steps.get-location.outputs.location }}_${{ github.run_id }}
+        run: |
+          set -e
+          echo "Retrieving VM Private IP address..."
+          # Retrieve VM Private IP address
+          PRIVATE_IP=$(az vm show -g ${RG} -n ${VM_NAME} -d --query privateIps -o tsv)
+          if [[ -z "$PRIVATE_IP" ]]; then
+            echo "ERROR: Failed to retrieve private IP address."
+            exit 1
+          fi
+          echo "PRIVATE_IP=$PRIVATE_IP" >> $GITHUB_OUTPUT
+
+      - name: Wait for SSH availability
+        env:
+          KEY: ${{ inputs.KEY }}
+          PRIVATE_IP: ${{ steps.get-vm-ip.outputs.PRIVATE_IP }}
+          USERNAME: ${{ secrets.USERNAME }}
+        run: |
+          echo "Waiting for SSH to be accessible..."
+          timeout 120 bash -c 'until ssh -o StrictHostKeyChecking=no -i ~/.ssh/${KEY} ${USERNAME}@${PRIVATE_IP} "exit" 2>/dev/null; do sleep 5; done'
+          echo "VM is accessible!"
+
+      - name: Remove Old Host Key
+        env:
+          PRIVATE_IP: ${{ steps.get-vm-ip.outputs.PRIVATE_IP }}
+        run: |
+          set -e
+          echo "Removing the old host key"
+          ssh-keygen -R $PRIVATE_IP
+
+      - name: SSH into VM and Install Dependencies
+        env:
+          KEY: ${{ inputs.KEY }}
+          PRIVATE_IP: ${{ steps.get-vm-ip.outputs.PRIVATE_IP }}
+          USERNAME: ${{ secrets.USERNAME }}
+        run: |
+          set -e
+          ssh -i ~/.ssh/${KEY} -o StrictHostKeyChecking=no ${USERNAME}@${PRIVATE_IP} << EOF
+            set -e
+            echo "Logged in successfully."
+            echo "Installing dependencies..."
+            sudo tdnf install -y git moby-engine moby-cli clang llvm pkg-config make gcc glibc-devel
+            echo "Installing Rust..."
+            curl -sSf https://sh.rustup.rs | sh -s -- --default-toolchain stable --profile default -y
+            export PATH="\$HOME/.cargo/bin:\$PATH"
+            cargo --version
+            sudo mkdir -p /etc/docker/
+            echo '{"default-ulimits":{"nofile":{"Hard":65535,"Name":"nofile","Soft":65535}}}' | sudo tee /etc/docker/daemon.json
+            sudo systemctl stop docker
+            sudo systemctl enable docker.service
+            sudo systemctl enable containerd.service
+            sudo systemctl start docker
+            sudo groupadd -f docker
+            sudo usermod -a -G docker ${USERNAME}
+            sudo systemctl restart docker
+          EOF

.github/workflows/mshv-integration.yaml

@@ -0,0 +1,130 @@
+name: Cloud Hypervisor Tests (MSHV) (x86_64)
+on: [pull_request_target, merge_group]
+
+jobs:
+  infra-setup:
+    name: MSHV Infra Setup (x86_64)
+    uses: ./.github/workflows/mshv-infra.yaml
+    with:
+      ARCH: x86_64
+      KEY: azure_key_${{ github.run_id }}
+      OS_DISK_SIZE: 512
+      RG: MSHV-INTEGRATION-${{ github.run_id }}
+      VM_SKU: Standard_D16s_v5
+    secrets:
+      MI_CLIENT_ID: ${{ secrets.MSHV_MI_CLIENT_ID }}
+      RUNNER_RG: ${{ secrets.MSHV_RUNNER_RG }}
+      STORAGE_ACCOUNT_PATHS: ${{ secrets.MSHV_STORAGE_ACCOUNT_PATHS }}
+      ARCH_SOURCE_PATH: ${{ secrets.MSHV_X86_SOURCE_PATH }}
+      USERNAME: ${{ secrets.MSHV_USERNAME }}
+
+  run-tests:
+    name: Integration Tests (x86_64)
+    needs: infra-setup
+    if: ${{ always() && needs.infra-setup.result == 'success' }}
+    runs-on: mshv
+    steps:
+      - name: Run integration tests
+        timeout-minutes: 60
+        env:
+          KEY: azure_key_${{ github.run_id }}
+          PR_NUMBER: ${{ github.event.pull_request.number }}
+          REPO_URL: https://github.com/cloud-hypervisor/cloud-hypervisor.git
+          REPO_DIR: cloud-hypervisor
+          PRIVATE_IP: ${{ needs.infra-setup.outputs.PRIVATE_IP }}
+          RG: MSHV-${{ github.run_id }}
+          USERNAME: ${{ secrets.MSHV_USERNAME }}
+        run: |
+          set -e
+          echo "Connecting to the VM via SSH..."
+          ssh -i ~/.ssh/${KEY} -o StrictHostKeyChecking=no ${USERNAME}@${PRIVATE_IP} << EOF
+            set -e
+            echo "Logged in successfully."
+            export PATH="\$HOME/.cargo/bin:\$PATH"
+
+            if [[ "${{ github.event_name }}" == "pull_request_target" ]]; then
+              git clone --depth 1 "$REPO_URL" "$REPO_DIR"
+              cd "$REPO_DIR"
+              git fetch origin pull/${{ github.event.pull_request.number }}/merge
+              git checkout FETCH_HEAD
+            else
+              git clone --depth 1 --single-branch --branch "${{ github.ref_name }}" "$REPO_URL" "$REPO_DIR"
+              cd "$REPO_DIR"
+            fi
+
+            echo "Loading VDPA kernel modules..."
+            sudo modprobe vdpa
+            sudo modprobe vhost_vdpa
+            sudo modprobe vdpa_sim
+            sudo modprobe vdpa_sim_blk
+            sudo modprobe vdpa_sim_net
+
+            echo "Creating VDPA devices..."
+            sudo vdpa dev add name vdpa-blk0 mgmtdev vdpasim_blk
+            sudo vdpa dev add name vdpa-blk1 mgmtdev vdpasim_blk
+            sudo vdpa dev add name vdpa-blk2 mgmtdev vdpasim_net
+
+            echo "Setting permissions..."
+            for i in 0 1 2; do
+              dev="/dev/vhost-vdpa-$i"
+              if [ -e "$dev" ]; then
+                sudo chown $USER:$USER "$dev"
+                sudo chmod 660 "$dev"
+              else
+                echo "Warning: Device $dev not found"
+              fi
+            done
+
+            sudo ./scripts/dev_cli.sh tests --hypervisor mshv --integration
+          EOF
+
+      - name: Dump dmesg
+        if: always()
+        continue-on-error: true
+        env:
+          KEY: azure_key_${{ github.run_id }}
+          PRIVATE_IP: ${{ needs.infra-setup.outputs.PRIVATE_IP }}
+          USERNAME: ${{ secrets.MSHV_USERNAME }}
+        run: |
+          ssh -i ~/.ssh/${KEY} -o StrictHostKeyChecking=no ${USERNAME}@${PRIVATE_IP} << EOF
+            sudo dmesg
+          EOF
+
+      - name: Dump serial console logs
+        if: always()
+        continue-on-error: true
+        env:
+          RG_NAME: ${{ needs.infra-setup.outputs.RG_NAME }}
+          VM_NAME: ${{ needs.infra-setup.outputs.VM_NAME }}
+        run: |
+          set -e
+          az vm boot-diagnostics get-boot-log --name "${VM_NAME}" --resource-group "${RG_NAME}" | jq -r
+
+  cleanup:
+    name: Cleanup
+    needs: run-tests
+    if: always()
+    runs-on: mshv
+    steps:
+      - name: Delete RG
+        env:
+          RG: MSHV-INTEGRATION-${{ github.run_id }}
+        run: |
+          if az group exists --name ${RG}; then
+            az group delete --name ${RG} --yes --no-wait
+          else
+            echo "Resource Group ${RG} does not exist. Skipping deletion."
+          fi
+          echo "Cleanup process completed."
+
+      - name: Delete SSH Key
+        env:
+          KEY: azure_key_${{ github.run_id }}
+        run: |
+          if [ -f ~/.ssh/${KEY} ]; then
+            rm -f ~/.ssh/${KEY} ~/.ssh/${KEY}.pub
+            echo "SSH key deleted successfully."
+          else
+            echo "SSH key does not exist. Skipping deletion."
+          fi
+          echo "Cleanup process completed."

.github/workflows/openapi.yaml

@@ -6,7 +6,7 @@ jobs:
     runs-on: ubuntu-latest
     container: openapitools/openapi-generator-cli
     steps:
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v6
     - name: Validate OpenAPI
       env:
         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/package-consistency.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor Consistency
 on: [pull_request, merge_group]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
 

.github/workflows/preview-riscv64-build.yaml

@@ -0,0 +1,30 @@
+name: Cloud Hypervisor RISC-V 64-bit kvm build Preview
+on: [pull_request, merge_group]
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
+  cancel-in-progress: true
+
+jobs:
+  build:
+    name: Cargo
+    runs-on: riscv64-qemu-host
+    strategy:
+      fail-fast: false
+
+    steps:
+      - name: Code checkout
+        uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+
+      - name: Install Rust toolchain
+        run: /opt/scripts/exec-in-qemu.sh rustup default 1.89.0
+
+      - name: Build test (kvm)
+        run: /opt/scripts/exec-in-qemu.sh cargo build --locked --no-default-features --features "kvm" -p cloud-hypervisor
+
+      - name: Clippy test (kvm)
+        run: /opt/scripts/exec-in-qemu.sh cargo clippy --locked --no-default-features --features "kvm" -p cloud-hypervisor
+
+      - name: Check no files were modified
+        run: test -z "$(git status --porcelain)"

.github/workflows/preview-riscv64-modules.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor RISC-V 64-bit Preview
 on: [pull_request, merge_group]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -19,18 +19,18 @@ jobs:
 
     steps:
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
 
       - name: Install Rust toolchain
-        run: /opt/scripts/exec-in-qemu.sh rustup default 1.83.0
+        run: /opt/scripts/exec-in-qemu.sh rustup default 1.89.0
 
       - name: Build ${{ matrix.module }} Module (kvm)
-        run: /opt/scripts/exec-in-qemu.sh cargo rustc --locked -p ${{ matrix.module }} --no-default-features --features "kvm" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: /opt/scripts/exec-in-qemu.sh cargo build --locked -p ${{ matrix.module }} --no-default-features --features "kvm"
 
       - name: Clippy ${{ matrix.module }} Module (kvm)
-        run: /opt/scripts/exec-in-qemu.sh cargo clippy --locked -p ${{ matrix.module }} --no-default-features --features "kvm" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+        run: /opt/scripts/exec-in-qemu.sh cargo clippy --locked -p ${{ matrix.module }} --no-default-features --features "kvm" -- -D warnings
 
       - name: Test ${{ matrix.module }} Module (kvm)
         run: /opt/scripts/exec-in-qemu.sh cargo test --locked -p ${{ matrix.module }} --no-default-features --features "kvm"

.github/workflows/quality.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor Quality Checks
 on: [pull_request, merge_group]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -29,7 +29,7 @@ jobs:
 
     steps:
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
 
@@ -56,7 +56,7 @@ jobs:
           cross-version: 3e0957637b49b1bbced23ad909170650c5b70635
           toolchain: ${{ matrix.rust }}
           target: ${{ matrix.target }}
-          args: --locked --all --all-targets --no-default-features --tests --examples --features "kvm" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+          args: --locked --all --all-targets --no-default-features --tests --examples --features "kvm" -- -D warnings
 
       - name: Clippy (mshv)
         uses: houseabsolute/actions-rust-cross@v1
@@ -65,7 +65,7 @@ jobs:
           cross-version: 3e0957637b49b1bbced23ad909170650c5b70635
           toolchain: ${{ matrix.rust }}
           target: ${{ matrix.target }}
-          args: --locked --all --all-targets --no-default-features --tests --examples --features "mshv" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+          args: --locked --all --all-targets --no-default-features --tests --examples --features "mshv" -- -D warnings
 
       - name: Clippy (mshv + kvm)
         uses: houseabsolute/actions-rust-cross@v1
@@ -74,7 +74,7 @@ jobs:
           cross-version: 3e0957637b49b1bbced23ad909170650c5b70635
           toolchain: ${{ matrix.rust }}
           target: ${{ matrix.target }}
-          args: --locked --all --all-targets --no-default-features --tests --examples --features "mshv,kvm" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+          args: --locked --all --all-targets --no-default-features --tests --examples --features "mshv,kvm" -- -D warnings
 
       - name: Clippy (default features)
         uses: houseabsolute/actions-rust-cross@v1
@@ -83,7 +83,7 @@ jobs:
           cross-version: 3e0957637b49b1bbced23ad909170650c5b70635
           toolchain: ${{ matrix.rust }}
           target: ${{ matrix.target }}
-          args: --locked --all --all-targets --tests --examples -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+          args: --locked --all --all-targets --tests --examples -- -D warnings
 
       - name: Clippy (default features + guest_debug)
         uses: houseabsolute/actions-rust-cross@v1
@@ -92,7 +92,7 @@ jobs:
           cross-version: 3e0957637b49b1bbced23ad909170650c5b70635
           toolchain: ${{ matrix.rust }}
           target: ${{ matrix.target }}
-          args: --locked --all --all-targets --tests --examples --features "guest_debug" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+          args: --locked --all --all-targets --tests --examples --features "guest_debug" -- -D warnings
 
       - name: Clippy (default features + pvmemcontrol)
         uses: houseabsolute/actions-rust-cross@v1
@@ -101,7 +101,7 @@ jobs:
           cross-version: 3e0957637b49b1bbced23ad909170650c5b70635
           toolchain: ${{ matrix.rust }}
           target: ${{ matrix.target }}
-          args: --locked --all --all-targets --tests --examples --features "pvmemcontrol" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+          args: --locked --all --all-targets --tests --examples --features "pvmemcontrol" -- -D warnings
 
       - name: Clippy (default features + tracing)
         uses: houseabsolute/actions-rust-cross@v1
@@ -110,7 +110,22 @@ jobs:
           cross-version: 3e0957637b49b1bbced23ad909170650c5b70635
           toolchain: ${{ matrix.rust }}
           target: ${{ matrix.target }}
-          args: --locked --all --all-targets --tests --examples --features "tracing" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+          args: --locked --all --all-targets --tests --examples --features "tracing" -- -D warnings
+      - name: Clippy (default features + fw_cfg)
+        uses: actions-rs/cargo@v1
+        with:
+          use-cross: ${{ matrix.target != 'x86_64-unknown-linux-gnu' }}
+          command: clippy
+          args: --target=${{ matrix.target }} --locked --all --all-targets --tests --examples --features "fw_cfg" -- -D warnings
+
+      - name: Clippy (default features + ivshmem)
+        uses: houseabsolute/actions-rust-cross@v1
+        with:
+          command: clippy
+          cross-version: 3e0957637b49b1bbced23ad909170650c5b70635
+          toolchain: ${{ matrix.rust }}
+          target: ${{ matrix.target }}
+          args: --locked --all --all-targets --tests --examples --features "ivshmem" -- -D warnings
 
       - name: Clippy (sev_snp)
         if: ${{ matrix.target == 'x86_64-unknown-linux-gnu' }}
@@ -120,7 +135,7 @@ jobs:
           cross-version: 3e0957637b49b1bbced23ad909170650c5b70635
           toolchain: ${{ matrix.rust }}
           target: ${{ matrix.target }}
-          args: --locked --all --all-targets --no-default-features --tests --examples --features "sev_snp" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+          args: --locked --all --all-targets --no-default-features --tests --examples --features "sev_snp" -- -D warnings
 
       - name: Clippy (igvm)
         if: ${{ matrix.target == 'x86_64-unknown-linux-gnu' }}
@@ -130,7 +145,7 @@ jobs:
           cross-version: 3e0957637b49b1bbced23ad909170650c5b70635
           toolchain: ${{ matrix.rust }}
           target: ${{ matrix.target }}
-          args: --locked --all --all-targets --no-default-features --tests --examples --features "igvm" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+          args: --locked --all --all-targets --no-default-features --tests --examples --features "igvm" -- -D warnings
 
       - name: Clippy (kvm + tdx)
         if: ${{ matrix.target == 'x86_64-unknown-linux-gnu' }}
@@ -140,7 +155,7 @@ jobs:
           cross-version: 3e0957637b49b1bbced23ad909170650c5b70635
           toolchain: ${{ matrix.rust }}
           target: ${{ matrix.target }}
-          args: --locked --all --all-targets --no-default-features --tests --examples --features "tdx,kvm" -- -D warnings -D clippy::undocumented_unsafe_blocks -W clippy::assertions_on_result_states
+          args: --locked --all --all-targets --no-default-features --tests --examples --features "tdx,kvm" -- -D warnings
 
       - name: Check build did not modify any files
         run: test -z "$(git status --porcelain)"
@@ -150,6 +165,6 @@ jobs:
     name: Typos / Spellcheck
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
       # Executes "typos ."
-      - uses: crate-ci/typos@v1.34.0
+      - uses: crate-ci/typos@v1.43.5

.github/workflows/release.yaml

@@ -1,7 +1,7 @@
 name: Cloud Hypervisor Release
 on: [create, merge_group]
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 env:
   GITHUB_TOKEN: ${{ github.token }}
@@ -29,7 +29,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
       - name: Install musl-gcc
         if: contains(matrix.platform.target, 'musl')
         run: sudo apt install -y musl-tools
@@ -45,7 +45,7 @@ jobs:
           target: ${{ matrix.platform.target }}
           args: ${{ matrix.platform.args  }}
           strip: true
-          toolchain: "1.83.0"
+          toolchain: "1.89.0"
       - name: Copy Release Binaries
         if: github.event_name == 'create' && github.event.ref_type == 'tag'
         shell: bash
@@ -54,7 +54,7 @@ jobs:
           cp target/${{ matrix.platform.target }}/release/ch-remote ./${{ matrix.platform.name_ch_remote }}
       - name: Upload Release Artifacts
         if: github.event_name == 'create' && github.event.ref_type == 'tag'
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v6
         with:
           name: Artifacts for ${{ matrix.platform.target }}
           path: |
@@ -80,7 +80,7 @@ jobs:
           github.event_name == 'create' && github.event.ref_type == 'tag' &&
           matrix.platform.target == 'x86_64-unknown-linux-gnu'
         id: upload-release-cloud-hypervisor-vendored-sources
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v6
         with:
           path: cloud-hypervisor-${{ github.event.ref }}.tar.xz
           name: cloud-hypervisor-${{ github.event.ref }}.tar.xz

.github/workflows/reuse.yaml

@@ -7,6 +7,6 @@ jobs:
     name: REUSE Compliance Check
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v6
     - name: REUSE Compliance Check
-      uses: fsfe/reuse-action@v5
+      uses: fsfe/reuse-action@v6

.github/workflows/shlint.yaml

@@ -12,7 +12,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
     - name: Checkout repository
-      uses: actions/checkout@v4
+      uses: actions/checkout@v6
     - name: Run the shell script checkers
       uses: luizm/action-sh-checker@master
       env:

.github/workflows/taplo.yaml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Code checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
       - name: Install Rust toolchain
         uses: dtolnay/rust-toolchain@stable
       - name: Install build dependencies

.gitlint

@@ -1,7 +1,7 @@
 [general]
 extra-path=scripts/gitlint/rules
 regex-style-search=true
-ignore=body-max-line-length
+ignore=body-max-line-length,body-hard-tab
 
 [ignore-by-author-name]
 regex=dependabot

.lychee.toml

@@ -1,9 +1,9 @@
 verbose = "info"
 
+exclude_path = [".lychee.toml"]
+
 exclude = [
     # Availability of links below should be manually verified.
-    # Page for intel SGX support, returns 403 while querying.
-    '^https://www.intel.com/content/www/us/en/developer/tools/software-guard-extensions/linux-overview.html',
     # Page for intel TDX support, returns 403 while querying.
     '^https://www.intel.com/content/www/us/en/developer/tools/trust-domain-extensions/overview.html',
     # Page for TPM, returns 403 while querying.
@@ -14,8 +14,23 @@ exclude = [
 
     # OSDev has added bot protection and accesses my result in 403 Forbidden.
     '^https://wiki.osdev.org',
+    # Exclude all pages with $ in the URL since $XXX is a variable
+    "\\$.*",
+    # Exclude local files
+    "file://.*",
+
+    # ARM documentation returns 403 Forbidden for automated CI checks.
+    '^http://infocenter\.arm\.com',
+    '^https://developer\.arm\.com',
+
+    # Ignore internal/unsupported protocols seen in logs
+    '^tcp://192\.168\.1\.10',
 ]
 
+# Exclude loopback addresses
+exclude_loopback = true
+
+
 max_retries = 3
 
 retry_wait_time = 5

.rustfmt.toml

@@ -1,4 +1,4 @@
-edition = "2021"
+edition = "2024"
 group_imports="StdExternalCrate"
 imports_granularity="Module"
 

.typos.toml

@@ -11,6 +11,7 @@ CLASSE = "CLASSE"
 Dake = "Dake"
 EXTINT = "EXTINT"
 INOUT = "INOUT"
+MSIS = "MSIS" # MSIs (Message Signaled Interrupt)
 SME = "SME" # Secure Memory Encryption
 THR = "THR" # Transmitter Holding Register
 TRANSLATER = "TRANSLATER"
@@ -20,5 +21,8 @@ liness = "liness"
 outout = "outout"
 
 [default.extend-identifiers]
+consts = "consts"
 fo = "fo"
 fpr = "fpr"
+# Public Linux API
+msg_controllen = "msg_controllen"

CONTRIBUTING.md

@@ -14,11 +14,33 @@ License](https://opensource.org/licenses/Apache-2.0).
 ## Coding Style
 
 We follow the [Rust Style](https://github.com/rust-lang/rust/tree/HEAD/src/doc/style-guide/src)
-convention and enforce it through the Continuous Integration (CI) process calling into `rustfmt`
-for each submitted Pull Request (PR).
+convention and enforce it through the Continuous Integration (CI) process calling into `rustfmt`,
+`clippy`, and other well-known code quality tool of the ecosystem for each submitted Pull Request (PR).
 
 ## Basic Checks
 
+```sh
+# We currently rely on nightly-only formatting features
+cargo +nightly fmt --all 
+cargo check --all-targets --tests
+cargo clippy --all-targets --tests
+# Please note that this will not execute integration tests.
+cargo test --all-targets --tests
+
+# To lint your last three commits
+gitlint --commits "HEAD~3..HEAD"
+```
+
+### \[Optional\] Run Integration Tests
+
+_Caution: These tests are taking a long time to complete (40+ mins) and need special setup._
+
+```sh
+ bash ./scripts/dev_cli.sh tests --integration -- --test-filter '<optionally filter test by name pattern>' 
+```
+
+### Setup Commit Hook
+
 Please consider creating the following hook as `.git/hooks/pre-commit` in order
 to ensure basic correctness of your code. You can extend this further if you
 have specific features that you regularly develop against.
@@ -26,9 +48,9 @@ have specific features that you regularly develop against.
 ```sh
 #!/bin/sh
 
-cargo fmt -- --check || exit 1
-cargo check --locked --all --all-targets --tests || exit 1
-cargo clippy --locked --all --all-targets --tests -- -D warnings || exit 1
+cargo +nightly fmt --all -- --check || exit 1
+cargo check --locked --all-targets --tests || exit 1
+cargo clippy --locked --all-targets --tests -- -D warnings || exit 1
 ```
 
 You will need to `chmod +x .git/hooks/pre-commit` to have it run on every
@@ -114,3 +136,13 @@ Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
 
 Then, after the corresponding PR is merged, GitHub will automatically close that issue when parsing the
 [commit message](https://help.github.com/articles/closing-issues-via-commit-messages/).
+
+## AI Generated Code
+
+Our policy is to decline any contributions known to contain contents
+generated or derived from using Large Language Models (LLMs). This
+includes ChatGPT, Gemini, Claude, Copilot and similar tools.
+
+The goal is to avoid ambiguity in license compliance and optimize the
+use of limited project resources, especially for code review and
+maintenance. This policy can be revisited as LLMs evolve and mature.

Cargo.toml

@@ -1,21 +1,6 @@
-[package]
-authors = ["The Cloud Hypervisor Authors"]
-build = "build.rs"
-default-run = "cloud-hypervisor"
-description = "Open source Virtual Machine Monitor (VMM) that runs on top of KVM & MSHV"
-edition = "2021"
-homepage = "https://github.com/cloud-hypervisor/cloud-hypervisor"
-license = "Apache-2.0 AND BSD-3-Clause"
-name = "cloud-hypervisor"
-version = "47.0.0"
-# Minimum buildable version:
-# Keep in sync with version in .github/workflows/build.yaml
-# Policy on MSRV (see #4318):
-# Can only be bumped if satisfying any of the following:
-# a.) A dependency requires it,
-# b.) If we want to use a new feature and that MSRV is at least 6 months old,
-# c.) There is a security issue that is addressed by the toolchain update.
-rust-version = "1.83.0"
+# Cloud Hypervisor Workspace
+#
+# The main crate producing the binaries is in `./cloud-hypervisor`.
 
 [profile.release]
 codegen-units = 1
@@ -28,57 +13,12 @@ debug = true
 inherits = "release"
 strip = false
 
-[dependencies]
-anyhow = "1.0.94"
-api_client = { path = "api_client" }
-clap = { version = "4.5.13", features = ["string"] }
-dhat = { version = "0.3.3", optional = true }
-env_logger = { workspace = true }
-epoll = "4.3.3"
-event_monitor = { path = "event_monitor" }
-hypervisor = { path = "hypervisor" }
-libc = "0.2.167"
-log = { version = "0.4.22", features = ["std"] }
-option_parser = { path = "option_parser" }
-seccompiler = { workspace = true }
-serde_json = { workspace = true }
-signal-hook = "0.3.18"
-thiserror = { workspace = true }
-tpm = { path = "tpm" }
-tracer = { path = "tracer" }
-vm-memory = { workspace = true }
-vmm = { path = "vmm" }
-vmm-sys-util = { workspace = true }
-zbus = { version = "5.7.1", optional = true }
-
-[dev-dependencies]
-dirs = "6.0.0"
-net_util = { path = "net_util" }
-serde_json = { workspace = true }
-test_infra = { path = "test_infra" }
-wait-timeout = "0.2.0"
-
-# Please adjust `vmm::feature_list()` accordingly when changing the
-# feature list below
-[features]
-dbus_api = ["vmm/dbus_api", "zbus"]
-default = ["io_uring", "kvm"]
-dhat-heap = ["dhat", "vmm/dhat-heap"]       # For heap profiling
-guest_debug = ["vmm/guest_debug"]
-igvm = ["mshv", "vmm/igvm"]
-io_uring = ["vmm/io_uring"]
-kvm = ["vmm/kvm"]
-mshv = ["vmm/mshv"]
-pvmemcontrol = ["vmm/pvmemcontrol"]
-sev_snp = ["igvm", "mshv", "vmm/sev_snp"]
-tdx = ["vmm/tdx"]
-tracing = ["tracer/tracing", "vmm/tracing"]
-
 [workspace]
 members = [
   "api_client",
   "arch",
   "block",
+  "cloud-hypervisor",
   "devices",
   "event_monitor",
   "hypervisor",
@@ -100,37 +40,82 @@ members = [
   "vm-virtio",
   "vmm",
 ]
+package.edition = "2024"
+resolver = "3"
 
 [workspace.dependencies]
 # rust-vmm crates
-acpi_tables = { git = "https://github.com/rust-vmm/acpi_tables", branch = "main" }
-kvm-bindings = "0.12.0"
-kvm-ioctls = "0.22.0"
-linux-loader = "0.13.0"
-mshv-bindings = "0.5.2"
-mshv-ioctls = "0.5.2"
+acpi_tables = "0.2.0"
+kvm-bindings = "0.12.1"
+kvm-ioctls = "0.22.1"
+linux-loader = "0.13.1"
+mshv-bindings = "0.6.7"
+mshv-ioctls = "0.6.7"
 seccompiler = "0.5.0"
-vfio-bindings = { version = "0.5.0", default-features = false }
-vfio-ioctls = { version = "0.5.0", default-features = false }
-vfio_user = { version = "0.1.0", default-features = false }
+vfio-bindings = { version = "0.6.0", default-features = false }
+vfio-ioctls = { version = "0.5.1", default-features = false }
+vfio_user = { version = "0.1.1", default-features = false }
 vhost = { version = "0.14.0", default-features = false }
 vhost-user-backend = { version = "0.20.0", default-features = false }
 virtio-bindings = "0.2.6"
 virtio-queue = "0.16.0"
-vm-fdt = { git = "https://github.com/rust-vmm/vm-fdt", branch = "main" }
+vm-fdt = "0.3.0"
 vm-memory = "0.16.1"
 vmm-sys-util = "0.14.0"
 
 # igvm crates
-# TODO: bump to 0.3.5 release
-igvm = { git = "https://github.com/microsoft/igvm", branch = "main" }
-igvm_defs = { git = "https://github.com/microsoft/igvm", branch = "main" }
+igvm = "0.4.0"
+igvm_defs = "0.4.0"
 
 # serde crates
-serde_json = "1.0.120"
+serde = "1.0.228"
+serde_json = "1.0.149"
+serde_with = { version = "3.16.1", default-features = false }
 
 # other crates
+anyhow = "1.0.101"
+bitflags = "2.11.0"
+byteorder = "1.5.0"
+cfg-if = "1.0.4"
+clap = "4.5.59"
+dhat = "0.3.3"
+dirs = "6.0.0"
 env_logger = "0.11.8"
-thiserror = "2.0.12"
-uuid = { version = "1.17.0" }
-zerocopy = { version = "0.8.26", default-features = false }
+epoll = "4.4.0"
+flume = "0.12.0"
+itertools = "0.14.0"
+libc = "0.2.182"
+log = "0.4.29"
+signal-hook = "0.4.3"
+thiserror = "2.0.18"
+uuid = { version = "1.21.0" }
+wait-timeout = "0.2.1"
+zerocopy = { version = "0.8.39", default-features = false }
+
+[workspace.lints.clippy]
+# Any clippy lint (group) in alphabetical order:
+# https://rust-lang.github.io/rust-clippy/master/index.html
+
+# Groups
+all = "deny"         # shorthand for the other groups but here for compleness
+complexity = "deny"
+correctness = "deny"
+perf = "deny"
+style = "deny"
+suspicious = "deny"
+
+# Individual Lints
+assertions_on_result_states = "deny"
+if_not_else = "deny"
+manual_string_new = "deny"
+map_unwrap_or = "deny"
+needless_pass_by_value = "deny"
+redundant_else = "deny"
+semicolon_if_nothing_returned = "deny"
+undocumented_unsafe_blocks = "deny"
+uninlined_format_args = "deny"
+unnecessary_semicolon = "deny"
+
+[workspace.lints.rust]
+# `level = warn` is irrelevant here but mandatory for rustc/cargo
+unexpected_cfgs = { level = "warn", check-cfg = ['cfg(devcli_testenv)'] }

README.md

@@ -153,7 +153,7 @@ interface will be enabled as per `network-config` details.
 $ sudo setcap cap_net_admin+ep ./cloud-hypervisor
 $ ./create-cloud-init.sh
 $ ./cloud-hypervisor \
-	--kernel ./hypervisor-fw \
+	--firmware ./hypervisor-fw \
 	--disk path=focal-server-cloudimg-amd64.raw path=/tmp/ubuntu-cloudinit.img \
 	--cpus boot=4 \
 	--memory size=1024M \
@@ -175,6 +175,18 @@ $ ./cloud-hypervisor \
 	--console off
 ```
 
+## Booting: `--firmware` vs `--kernel`
+
+The following scenarios are supported by Cloud Hypervisor to bootstrap a VM, i.e.,
+to load a payload/bootitem(s):
+
+- Provide firmware
+- Provide kernel \[+ cmdline\]\ [+ initrd\]
+
+Please note that our Cloud Hypervisor firmware (`hypervisor-fw`) has a Xen PVH
+boot entry, therefore it can also be booted via the `--kernel` parameter, as 
+seen in some examples.
+
 ### Custom Kernel and Disk Image
 
 #### Building your Kernel

api_client/Cargo.toml

@@ -1,9 +1,13 @@
 [package]
 authors = ["The Cloud Hypervisor Authors"]
-edition = "2021"
+edition.workspace = true
+license = "Apache-2.0"
 name = "api_client"
 version = "0.1.0"
 
 [dependencies]
 thiserror = { workspace = true }
 vmm-sys-util = { workspace = true }
+
+[lints]
+workspace = true

api_client/src/lib.rs

@@ -118,12 +118,11 @@ fn parse_http_response(socket: &mut dyn Read) -> Result<Option<String>, Error> {
             }
         }
 
-        if let Some(body_offset) = body_offset {
-            if let Some(content_length) = content_length {
-                if res.len() >= content_length + body_offset {
-                    break;
-                }
-            }
+        if let Some(body_offset) = body_offset
+            && let Some(content_length) = content_length
+            && res.len() >= content_length + body_offset
+        {
+            break;
         }
     }
     let body_string = content_length.and(body_offset.map(|o| String::from(&res[o..])));
@@ -143,7 +142,7 @@ pub fn simple_api_full_command_with_fds_and_response<T: Read + Write + ScmSocket
     method: &str,
     full_command: &str,
     request_body: Option<&str>,
-    request_fds: Vec<RawFd>,
+    request_fds: &[RawFd],
 ) -> Result<Option<String>, Error> {
     socket
         .send_with_fds(
@@ -151,7 +150,7 @@ pub fn simple_api_full_command_with_fds_and_response<T: Read + Write + ScmSocket
                 "{method} /api/v1/{full_command} HTTP/1.1\r\nHost: localhost\r\nAccept: */*\r\n"
             )
             .as_bytes()],
-            &request_fds,
+            request_fds,
         )
         .map_err(Error::SocketSendFds)?;
 
@@ -179,7 +178,7 @@ pub fn simple_api_full_command_with_fds<T: Read + Write + ScmSocket>(
     method: &str,
     full_command: &str,
     request_body: Option<&str>,
-    request_fds: Vec<RawFd>,
+    request_fds: &[RawFd],
 ) -> Result<(), Error> {
     let response = simple_api_full_command_with_fds_and_response(
         socket,
@@ -189,8 +188,8 @@ pub fn simple_api_full_command_with_fds<T: Read + Write + ScmSocket>(
         request_fds,
     )?;
 
-    if response.is_some() {
-        println!("{}", response.unwrap());
+    if let Some(response) = response {
+        println!("{response}");
     }
 
     Ok(())
@@ -202,7 +201,7 @@ pub fn simple_api_full_command<T: Read + Write + ScmSocket>(
     full_command: &str,
     request_body: Option<&str>,
 ) -> Result<(), Error> {
-    simple_api_full_command_with_fds(socket, method, full_command, request_body, Vec::new())
+    simple_api_full_command_with_fds(socket, method, full_command, request_body, &[])
 }
 
 pub fn simple_api_full_command_and_response<T: Read + Write + ScmSocket>(
@@ -211,13 +210,7 @@ pub fn simple_api_full_command_and_response<T: Read + Write + ScmSocket>(
     full_command: &str,
     request_body: Option<&str>,
 ) -> Result<Option<String>, Error> {
-    simple_api_full_command_with_fds_and_response(
-        socket,
-        method,
-        full_command,
-        request_body,
-        Vec::new(),
-    )
+    simple_api_full_command_with_fds_and_response(socket, method, full_command, request_body, &[])
 }
 
 pub fn simple_api_command_with_fds<T: Read + Write + ScmSocket>(
@@ -225,7 +218,7 @@ pub fn simple_api_command_with_fds<T: Read + Write + ScmSocket>(
     method: &str,
     c: &str,
     request_body: Option<&str>,
-    request_fds: Vec<RawFd>,
+    request_fds: &[RawFd],
 ) -> Result<(), Error> {
     // Create the full VM command. For VMM commands, use
     // simple_api_full_command().
@@ -240,5 +233,5 @@ pub fn simple_api_command<T: Read + Write + ScmSocket>(
     c: &str,
     request_body: Option<&str>,
 ) -> Result<(), Error> {
-    simple_api_command_with_fds(socket, method, c, request_body, Vec::new())
+    simple_api_command_with_fds(socket, method, c, request_body, &[])
 }

arch/Cargo.toml

@@ -1,23 +1,24 @@
 [package]
 authors = ["The Chromium OS Authors"]
-edition = "2021"
+edition.workspace = true
 name = "arch"
 version = "0.1.0"
 
 [features]
 default = []
+fw_cfg = []
 kvm = ["hypervisor/kvm"]
 sev_snp = []
 tdx = []
 
 [dependencies]
-anyhow = "1.0.94"
-byteorder = "1.5.0"
+anyhow = { workspace = true }
+byteorder = { workspace = true }
 hypervisor = { path = "../hypervisor" }
-libc = "0.2.167"
+libc = { workspace = true }
 linux-loader = { workspace = true, features = ["bzimage", "elf", "pe"] }
-log = "0.4.22"
-serde = { version = "1.0.208", features = ["derive", "rc"] }
+log = { workspace = true }
+serde = { workspace = true, features = ["derive", "rc"] }
 thiserror = { workspace = true }
 uuid = { workspace = true }
 vm-memory = { workspace = true, features = ["backend-bitmap", "backend-mmap"] }
@@ -26,3 +27,6 @@ vmm-sys-util = { workspace = true, features = ["with-serde"] }
 [target.'cfg(any(target_arch = "aarch64", target_arch = "riscv64"))'.dependencies]
 fdt_parser = { version = "0.1.5", package = "fdt" }
 vm-fdt = { workspace = true }
+
+[lints]
+workspace = true

arch/src/aarch64/fdt.rs

@@ -19,14 +19,16 @@ use hypervisor::arch::aarch64::regs::{
     AARCH64_ARCH_TIMER_HYP_IRQ, AARCH64_ARCH_TIMER_PHYS_NONSECURE_IRQ,
     AARCH64_ARCH_TIMER_PHYS_SECURE_IRQ, AARCH64_ARCH_TIMER_VIRT_IRQ, AARCH64_PMU_IRQ,
 };
+use log::{debug, info, warn};
 use thiserror::Error;
 use vm_fdt::{FdtWriter, FdtWriterResult};
 use vm_memory::{Address, Bytes, GuestMemory, GuestMemoryError, GuestMemoryRegion};
 
 use super::super::{DeviceType, GuestMemoryMmap, InitramfsConfig};
 use super::layout::{
-    GIC_V2M_COMPATIBLE, IRQ_BASE, MEM_32BIT_DEVICES_SIZE, MEM_32BIT_DEVICES_START, MEM_PCI_IO_SIZE,
-    MEM_PCI_IO_START, PCI_HIGH_BASE, PCI_MMIO_CONFIG_SIZE_PER_SEGMENT, SPI_BASE, SPI_NUM,
+    GIC_V2M_COMPATIBLE, GICV2M_SPI_BASE, GICV2M_SPI_NUM, IRQ_BASE, MEM_32BIT_DEVICES_SIZE,
+    MEM_32BIT_DEVICES_START, MEM_PCI_IO_SIZE, MEM_PCI_IO_START, PCI_HIGH_BASE,
+    PCI_MMIO_CONFIG_SIZE_PER_SEGMENT,
 };
 use crate::{NumaNodes, PciSpaceInfo};
 
@@ -86,6 +88,7 @@ pub enum Error {
 }
 type Result<T> = result::Result<T, Error>;
 
+#[derive(Copy, Clone)]
 pub enum CacheLevel {
     /// L1 data cache
     L1D = 0,
@@ -109,12 +112,7 @@ pub fn get_cache_size(cache_level: CacheLevel) -> u32 {
     }
 
     let file_path = Path::new(&file_directory);
-    if !file_path.exists() {
-        warn!("File: {} does not exist.", file_directory);
-        0
-    } else {
-        info!("File: {} exist.", file_directory);
-
+    if file_path.exists() {
         let src = fs::read_to_string(file_directory).expect("File not exists or file corrupted.");
         // The content of the file is as simple as a size, like: "32K"
         let src = src.trim();
@@ -128,6 +126,8 @@ pub fn get_cache_size(cache_level: CacheLevel) -> u32 {
                 "G" => 1024u32.pow(3),
                 _ => 1,
             }
+    } else {
+        0
     }
 }
 
@@ -143,14 +143,11 @@ pub fn get_cache_coherency_line_size(cache_level: CacheLevel) -> u32 {
     }
 
     let file_path = Path::new(&file_directory);
-    if !file_path.exists() {
-        warn!("File: {} does not exist.", file_directory);
-        0
-    } else {
-        info!("File: {} exist.", file_directory);
-
+    if file_path.exists() {
         let src = fs::read_to_string(file_directory).expect("File not exists or file corrupted.");
         src.trim().parse::<u32>().unwrap()
+    } else {
+        0
     }
 }
 
@@ -166,14 +163,11 @@ pub fn get_cache_number_of_sets(cache_level: CacheLevel) -> u32 {
     }
 
     let file_path = Path::new(&file_directory);
-    if !file_path.exists() {
-        warn!("File: {} does not exist.", file_directory);
-        0
-    } else {
-        info!("File: {} exist.", file_directory);
-
+    if file_path.exists() {
         let src = fs::read_to_string(file_directory).expect("File not exists or file corrupted.");
         src.trim().parse::<u32>().unwrap()
+    } else {
+        0
     }
 }
 
@@ -194,12 +188,7 @@ pub fn get_cache_shared(cache_level: CacheLevel) -> bool {
     }
 
     let file_path = Path::new(&file_directory);
-    if !file_path.exists() {
-        warn!("File: {} does not exist.", file_directory);
-        result = false;
-    } else {
-        info!("File: {} exist.", file_directory);
-
+    if file_path.exists() {
         let src = fs::read_to_string(file_directory).expect("File not exists or file corrupted.");
         let src = src.trim();
         if src.is_empty() {
@@ -207,6 +196,8 @@ pub fn get_cache_shared(cache_level: CacheLevel) -> bool {
         } else {
             result = src.contains('-') || src.contains(',');
         }
+    } else {
+        result = false;
     }
 
     result
@@ -217,8 +208,8 @@ pub fn get_cache_shared(cache_level: CacheLevel) -> bool {
 pub fn create_fdt<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::BuildHasher>(
     guest_mem: &GuestMemoryMmap,
     cmdline: &str,
-    vcpu_mpidr: Vec<u64>,
-    vcpu_topology: Option<(u8, u8, u8)>,
+    vcpu_mpidr: &[u64],
+    vcpu_topology: Option<(u16, u16, u16, u16)>,
     device_info: &HashMap<(DeviceType, String), T, S>,
     gic_device: &Arc<Mutex<dyn Vgic>>,
     initrd: &Option<InitramfsConfig>,
@@ -231,8 +222,8 @@ pub fn create_fdt<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::BuildHash
     let mut fdt = FdtWriter::new().unwrap();
 
     // For an explanation why these nodes were introduced in the blob take a look at
-    // https://github.com/torvalds/linux/blob/master/Documentation/devicetree/booting-without-of.txt#L845
-    // Look for "Required nodes and properties".
+    // the "Device Node Requirements" chapter of the Devicetree Specification.
+    // https://www.devicetree.org/specifications/
 
     // Header or the root node as per above mentioned documentation.
     let root_node = fdt.begin_node("")?;
@@ -244,7 +235,7 @@ pub fn create_fdt<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::BuildHash
     // This is not mandatory but we use it to point the root node to the node
     // containing description of the interrupt controller for this VM.
     fdt.property_u32("interrupt-parent", GIC_PHANDLE)?;
-    create_cpu_nodes(&mut fdt, &vcpu_mpidr, vcpu_topology, numa_nodes)?;
+    create_cpu_nodes(&mut fdt, vcpu_mpidr, vcpu_topology, numa_nodes)?;
     create_memory_node(&mut fdt, guest_mem, numa_nodes)?;
     create_chosen_node(&mut fdt, cmdline, initrd)?;
     create_gic_node(&mut fdt, gic_device)?;
@@ -268,10 +259,10 @@ pub fn create_fdt<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::BuildHash
     Ok(fdt_final)
 }
 
-pub fn write_fdt_to_memory(fdt_final: Vec<u8>, guest_mem: &GuestMemoryMmap) -> Result<()> {
+pub fn write_fdt_to_memory(fdt_final: &[u8], guest_mem: &GuestMemoryMmap) -> Result<()> {
     // Write FDT to memory.
     guest_mem
-        .write_slice(fdt_final.as_slice(), super::layout::FDT_START)
+        .write_slice(fdt_final, super::layout::FDT_START)
         .map_err(Error::WriteFdtToMemory)?;
     Ok(())
 }
@@ -280,7 +271,7 @@ pub fn write_fdt_to_memory(fdt_final: Vec<u8>, guest_mem: &GuestMemoryMmap) -> R
 fn create_cpu_nodes(
     fdt: &mut FdtWriter,
     vcpu_mpidr: &[u64],
-    vcpu_topology: Option<(u8, u8, u8)>,
+    vcpu_topology: Option<(u16, u16, u16, u16)>,
     numa_nodes: &NumaNodes,
 ) -> FdtWriterResult<()> {
     // See https://github.com/torvalds/linux/blob/master/Documentation/devicetree/bindings/arm/cpus.yaml.
@@ -289,8 +280,11 @@ fn create_cpu_nodes(
     fdt.property_u32("#size-cells", 0x0)?;
 
     let num_cpus = vcpu_mpidr.len();
-    let (threads_per_core, cores_per_package, packages) = vcpu_topology.unwrap_or((1, 1, 1));
-    let max_cpus: u32 = (threads_per_core * cores_per_package * packages).into();
+    let (threads_per_core, cores_per_die, dies_per_package, packages) =
+        vcpu_topology.unwrap_or((1, 1, 1, 1));
+    let cores_per_package = cores_per_die * dies_per_package;
+    let max_cpus: u32 =
+        threads_per_core as u32 * cores_per_die as u32 * dies_per_package as u32 * packages as u32;
 
     // Add cache info.
     // L1 Data Cache Info.
@@ -319,10 +313,7 @@ fn create_cpu_nodes(
 
     let cache_path = Path::new("/sys/devices/system/cpu/cpu0/cache");
     let cache_exist: bool = cache_path.exists();
-    if !cache_exist {
-        warn!("cache sysfs system does not exist.");
-    } else {
-        info!("cache sysfs system exists.");
+    if cache_exist {
         // L1 Data Cache Info.
         l1_d_cache_size = get_cache_size(CacheLevel::L1D);
         l1_d_cache_line_size = get_cache_coherency_line_size(CacheLevel::L1D);
@@ -350,6 +341,19 @@ fn create_cpu_nodes(
         if l3_cache_size != 0 {
             l3_cache_shared = get_cache_shared(CacheLevel::L3);
         }
+    } else {
+        warn!("cache sysfs system does not exist.");
+    }
+
+    // Arm boot protocol requires a minimal Device Tree
+    // https://docs.kernel.org/arch/arm64/booting.html
+    // As Generic initiators are supported only in ACPI
+    // When a guest kernel does not boot under "acpi=force" mode it can
+    // hang due to conflicting numa information present in FDT which
+    // does not support Generic Initiators
+    let has_generic_initiator = numa_nodes.values().any(|node| node.device_id.is_some());
+    if has_generic_initiator {
+        info!("Skipping NUMA CPU node encoding in FDT with Generic Initiator devices");
     }
 
     for (cpu_id, mpidr) in vcpu_mpidr.iter().enumerate().take(num_cpus) {
@@ -366,11 +370,13 @@ fn create_cpu_nodes(
         fdt.property_u32("reg", (mpidr & 0x7FFFFF) as u32)?;
         fdt.property_u32("phandle", cpu_id as u32 + FIRST_VCPU_PHANDLE)?;
 
-        // Add `numa-node-id` property if there is any numa config.
-        if numa_nodes.len() > 1 {
+        // Skipping NUMA encoding in FDT when Generic Initiator devices
+        // are present allowed such guest kernels to boot properly and
+        // rely solely on ACPI tables to setup NUMA
+        if numa_nodes.len() > 1 && !has_generic_initiator {
             for numa_node_idx in 0..numa_nodes.len() {
                 let numa_node = numa_nodes.get(&(numa_node_idx as u32));
-                if numa_node.unwrap().cpus.contains(&(cpu_id as u8)) {
+                if numa_node.unwrap().cpus.contains(&(cpu_id as u32)) {
                     fdt.property_u32("numa-node-id", numa_node_idx as u32)?;
                 }
             }
@@ -423,9 +429,6 @@ fn create_cpu_nodes(
 
                 fdt.end_node(l2_cache_node)?;
             }
-            if l2_cache_size != 0 && l2_cache_shared {
-                warn!("L2 cache shared with other cpus");
-            }
         }
 
         fdt.end_node(cpu_node)?;
@@ -462,7 +465,8 @@ fn create_cpu_nodes(
     }
 
     if let Some(topology) = vcpu_topology {
-        let (threads_per_core, cores_per_package, packages) = topology;
+        let (threads_per_core, cores_per_die, dies_per_package, packages) = topology;
+        let cores_per_package = cores_per_die * dies_per_package;
         let cpu_map_node = fdt.begin_node("cpu-map")?;
 
         // Create device tree nodes with regard of above mapping.
@@ -510,7 +514,14 @@ fn create_memory_node(
 ) -> FdtWriterResult<()> {
     // See https://github.com/torvalds/linux/blob/58ae0b51506802713aa0e9956d1853ba4c722c98/Documentation/devicetree/bindings/numa.txt
     // for NUMA setting in memory node.
-    if numa_nodes.len() > 1 {
+    let has_generic_initiator = numa_nodes.values().any(|node| node.device_id.is_some());
+    if has_generic_initiator {
+        info!("Skipping NUMA memory node encoding in FDT with Generic Initiator devices");
+    }
+    // Skipping NUMA encoding in FDT when Generic Initiator devices
+    // are present allowed guest kernels to boot and
+    // rely solely on ACPI tables to setup NUMA
+    if numa_nodes.len() > 1 && !has_generic_initiator {
         for numa_node_idx in 0..numa_nodes.len() {
             let numa_node = numa_nodes.get(&(numa_node_idx as u32));
             let mut mem_reg_prop: Vec<u64> = Vec::new();
@@ -527,12 +538,15 @@ fn create_memory_node(
                     node_memory_addr = memory_region_start_addr;
                 }
             }
-            let memory_node_name = format!("memory@{node_memory_addr:x}");
-            let memory_node = fdt.begin_node(&memory_node_name)?;
-            fdt.property_string("device_type", "memory")?;
-            fdt.property_array_u64("reg", &mem_reg_prop)?;
-            fdt.property_u32("numa-node-id", numa_node_idx as u32)?;
-            fdt.end_node(memory_node)?;
+            // Only create a memory node if this NUMA node has memory regions
+            if !mem_reg_prop.is_empty() {
+                let memory_node_name = format!("memory@{node_memory_addr:x}");
+                let memory_node = fdt.begin_node(&memory_node_name)?;
+                fdt.property_string("device_type", "memory")?;
+                fdt.property_array_u64("reg", &mem_reg_prop)?;
+                fdt.property_u32("numa-node-id", numa_node_idx as u32)?;
+                fdt.end_node(memory_node)?;
+            }
         }
     } else {
         // Note: memory regions from "GuestMemory" are sorted and non-zero sized.
@@ -678,8 +692,8 @@ fn create_gic_node(fdt: &mut FdtWriter, gic_device: &Arc<Mutex<dyn Vgic>>) -> Fd
         fdt.property_array_u64("reg", &msi_reg_prop)?;
 
         if msi_compatibility == GIC_V2M_COMPATIBLE {
-            fdt.property_u32("arm,msi-base-spi", SPI_BASE)?;
-            fdt.property_u32("arm,msi-num-spis", SPI_NUM)?;
+            fdt.property_u32("arm,msi-base-spi", GICV2M_SPI_BASE)?;
+            fdt.property_u32("arm,msi-num-spis", GICV2M_SPI_NUM)?;
         }
 
         fdt.end_node(msic_node)?;
@@ -850,6 +864,21 @@ fn create_gpio_node<T: DeviceInfoForFdt + Clone + Debug>(
     Ok(())
 }
 
+// https://www.kernel.org/doc/Documentation/devicetree/bindings/arm/fw-cfg.txt
+#[cfg(feature = "fw_cfg")]
+fn create_fw_cfg_node<T: DeviceInfoForFdt + Clone + Debug>(
+    fdt: &mut FdtWriter,
+    dev_info: &T,
+) -> FdtWriterResult<()> {
+    // FwCfg node
+    let fw_cfg_node = fdt.begin_node(&format!("fw-cfg@{:x}", dev_info.addr()))?;
+    fdt.property("compatible", b"qemu,fw-cfg-mmio\0")?;
+    fdt.property_array_u64("reg", &[dev_info.addr(), dev_info.length()])?;
+    fdt.end_node(fw_cfg_node)?;
+
+    Ok(())
+}
+
 fn create_devices_node<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::BuildHasher>(
     fdt: &mut FdtWriter,
     dev_info: &HashMap<(DeviceType, String), T, S>,
@@ -865,6 +894,8 @@ fn create_devices_node<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::Buil
             DeviceType::Virtio(_) => {
                 ordered_virtio_device.push(info);
             }
+            #[cfg(feature = "fw_cfg")]
+            DeviceType::FwCfg => create_fw_cfg_node(fdt, info)?,
         }
     }
 
@@ -994,39 +1025,39 @@ fn create_pci_nodes(
         fdt.property_array_u32("msi-map", &msi_map)?;
         fdt.property_u32("msi-parent", MSI_PHANDLE)?;
 
-        if pci_device_info_elem.pci_segment_id == 0 {
-            if let Some(virtio_iommu_bdf) = virtio_iommu_bdf {
-                // See kernel document Documentation/devicetree/bindings/pci/pci-iommu.txt
-                // for 'iommu-map' attribute setting.
-                let iommu_map = [
-                    0_u32,
-                    VIRTIO_IOMMU_PHANDLE,
-                    0_u32,
-                    virtio_iommu_bdf,
-                    virtio_iommu_bdf + 1,
-                    VIRTIO_IOMMU_PHANDLE,
-                    virtio_iommu_bdf + 1,
-                    0xffff - virtio_iommu_bdf,
-                ];
-                fdt.property_array_u32("iommu-map", &iommu_map)?;
+        if pci_device_info_elem.pci_segment_id == 0
+            && let Some(virtio_iommu_bdf) = virtio_iommu_bdf
+        {
+            // See kernel document Documentation/devicetree/bindings/pci/pci-iommu.txt
+            // for 'iommu-map' attribute setting.
+            let iommu_map = [
+                0_u32,
+                VIRTIO_IOMMU_PHANDLE,
+                0_u32,
+                virtio_iommu_bdf,
+                virtio_iommu_bdf + 1,
+                VIRTIO_IOMMU_PHANDLE,
+                virtio_iommu_bdf + 1,
+                0xffff - virtio_iommu_bdf,
+            ];
+            fdt.property_array_u32("iommu-map", &iommu_map)?;
 
-                // See kernel document Documentation/devicetree/bindings/virtio/iommu.txt
-                // for virtio-iommu node settings.
-                let virtio_iommu_node_name = format!("virtio_iommu@{virtio_iommu_bdf:x}");
-                let virtio_iommu_node = fdt.begin_node(&virtio_iommu_node_name)?;
-                fdt.property_u32("#iommu-cells", 1)?;
-                fdt.property_string("compatible", "virtio,pci-iommu")?;
+            // See kernel document Documentation/devicetree/bindings/virtio/iommu.txt
+            // for virtio-iommu node settings.
+            let virtio_iommu_node_name = format!("virtio_iommu@{virtio_iommu_bdf:x}");
+            let virtio_iommu_node = fdt.begin_node(&virtio_iommu_node_name)?;
+            fdt.property_u32("#iommu-cells", 1)?;
+            fdt.property_string("compatible", "virtio,pci-iommu")?;
 
-                // 'reg' is a five-cell address encoded as
-                // (phys.hi phys.mid phys.lo size.hi size.lo). phys.hi should contain the
-                // device's BDF as 0b00000000 bbbbbbbb dddddfff 00000000. The other cells
-                // should be zero.
-                let reg = [virtio_iommu_bdf << 8, 0_u32, 0_u32, 0_u32, 0_u32];
-                fdt.property_array_u32("reg", &reg)?;
-                fdt.property_u32("phandle", VIRTIO_IOMMU_PHANDLE)?;
+            // 'reg' is a five-cell address encoded as
+            // (phys.hi phys.mid phys.lo size.hi size.lo). phys.hi should contain the
+            // device's BDF as 0b00000000 bbbbbbbb dddddfff 00000000. The other cells
+            // should be zero.
+            let reg = [virtio_iommu_bdf << 8, 0_u32, 0_u32, 0_u32, 0_u32];
+            fdt.property_array_u32("reg", &reg)?;
+            fdt.property_u32("phandle", VIRTIO_IOMMU_PHANDLE)?;
 
-                fdt.end_node(virtio_iommu_node)?;
-            }
+            fdt.end_node(virtio_iommu_node)?;
         }
 
         fdt.end_node(pci_node)?;
@@ -1036,6 +1067,22 @@ fn create_pci_nodes(
 }
 
 fn create_distance_map_node(fdt: &mut FdtWriter, numa_nodes: &NumaNodes) -> FdtWriterResult<()> {
+    // When Generic Initiator nodes are present, skip ALL FDT NUMA information.
+    // Let ACPI (which supports Generic Initiator via SRAT Type 5) handle the entire NUMA topology.
+    // FDT cannot represent Generic Initiator nodes, and mixing FDT + ACPI NUMA info causes conflicts.
+    let has_generic_initiator = numa_nodes.values().any(|node| node.device_id.is_some());
+    if has_generic_initiator {
+        info!("Skipping NUMA distance map encoding in FDT with Generic Initiator devices");
+        return Ok(());
+    }
+    // At this point, we know there are no Generic Initiator nodes
+    let mut numa_ids: Vec<u32> = numa_nodes.keys().cloned().collect();
+
+    // If we only have one node, no distance map is needed
+    if numa_ids.len() <= 1 {
+        return Ok(());
+    }
+
     let distance_map_node = fdt.begin_node("distance-map")?;
     fdt.property_string("compatible", "numa-distance-map-v1")?;
     // Construct the distance matrix.
@@ -1048,26 +1095,33 @@ fn create_distance_map_node(fdt: &mut FdtWriter, numa_nodes: &NumaNodes) -> FdtW
     // a value greater than 10.
     // 4. distance-matrix should have entries in lexicographical ascending
     // order of nodes.
+    numa_ids.sort_unstable(); // lexicographical order
     let mut distance_matrix = Vec::new();
-    for numa_node_idx in 0..numa_nodes.len() {
-        let numa_node = numa_nodes.get(&(numa_node_idx as u32));
-        for dest_numa_node in 0..numa_node.unwrap().distances.len() + 1 {
-            if numa_node_idx == dest_numa_node {
-                distance_matrix.push(numa_node_idx as u32);
-                distance_matrix.push(dest_numa_node as u32);
+    // Iterate over actual numa IDs instead of 0..len()
+    for numa_id in numa_ids.iter() {
+        let numa_node = &numa_nodes[numa_id];
+        for dest_numa_id in numa_ids.iter() {
+            if *numa_id == *dest_numa_id {
+                distance_matrix.push(*numa_id);
+                distance_matrix.push(*dest_numa_id);
                 distance_matrix.push(10_u32);
                 continue;
             }
 
-            distance_matrix.push(numa_node_idx as u32);
-            distance_matrix.push(dest_numa_node as u32);
-            distance_matrix.push(
-                *numa_node
-                    .unwrap()
-                    .distances
-                    .get(&(dest_numa_node as u32))
-                    .unwrap() as u32,
-            );
+            distance_matrix.push(*numa_id);
+            distance_matrix.push(*dest_numa_id);
+            // Use user-specified distance, checking both directions for symmetry
+            let distance = if let Some(&dist) = numa_node.distances.get(dest_numa_id) {
+                // Forward direction: current node -> dest node
+                dist
+            } else if let Some(dest_node) = numa_nodes.get(dest_numa_id) {
+                // Reverse direction for symmetry: dest node -> current node
+                dest_node.distances.get(numa_id).copied().unwrap_or(20)
+            } else {
+                // Default distance when neither direction is specified
+                20
+            };
+            distance_matrix.push(distance as u32);
         }
     }
     fdt.property_array_u32("distance-matrix", distance_matrix.as_ref())?;
@@ -1144,7 +1198,7 @@ fn print_node(node: fdt_parser::node::FdtNode<'_, '_>, n_spaces: usize) {
                 array,
                 indent = (n_spaces + 2)
             );
-        };
+        }
     }
 
     // Print children nodes if there is any
@@ -1152,3 +1206,118 @@ fn print_node(node: fdt_parser::node::FdtNode<'_, '_>, n_spaces: usize) {
         print_node(child, n_spaces + 2);
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use std::collections::BTreeMap;
+
+    use super::*;
+    use crate::NumaNode;
+
+    // Helper function to create a simple NumaNode for testing
+    fn create_test_numa_node(cpus: Vec<u32>, device_id: Option<String>) -> NumaNode {
+        NumaNode {
+            memory_regions: Vec::new(),
+            hotplug_regions: Vec::new(),
+            cpus,
+            pci_segments: Vec::new(),
+            distances: BTreeMap::new(),
+            memory_zones: Vec::new(),
+            device_id,
+        }
+    }
+
+    #[test]
+    fn test_fdt_generic_initiator_detection_and_skip() {
+        // No Generic Initiator - should not skip FDT NUMA
+        let mut numa_nodes = BTreeMap::new();
+        numa_nodes.insert(0, create_test_numa_node(vec![0, 1], None));
+        numa_nodes.insert(1, create_test_numa_node(vec![2, 3], None));
+
+        let has_gi = numa_nodes.values().any(|node| node.device_id.is_some());
+        assert!(
+            !has_gi,
+            "Should not detect Generic Initiator when none present"
+        );
+
+        // One Generic Initiator - should skip FDT NUMA
+        let mut numa_nodes = BTreeMap::new();
+        numa_nodes.insert(0, create_test_numa_node(vec![0, 1], None));
+        numa_nodes.insert(1, create_test_numa_node(vec![], Some("vfio0".to_string())));
+
+        let has_gi = numa_nodes.values().any(|node| node.device_id.is_some());
+        assert!(has_gi, "Should detect Generic Initiator when present");
+
+        let mut fdt = FdtWriter::new().unwrap();
+        let result = create_distance_map_node(&mut fdt, &numa_nodes);
+        assert!(result.is_ok(), "Should skip distance map when GI present");
+
+        // Multiple Generic Initiators - should skip FDT NUMA
+        let mut numa_nodes = BTreeMap::new();
+        numa_nodes.insert(0, create_test_numa_node(vec![0, 1], None));
+        numa_nodes.insert(1, create_test_numa_node(vec![], Some("vfio0".to_string())));
+        numa_nodes.insert(2, create_test_numa_node(vec![], Some("vfio1".to_string())));
+
+        let has_gi = numa_nodes.values().any(|node| node.device_id.is_some());
+        assert!(has_gi, "Should detect multiple Generic Initiators");
+    }
+
+    #[test]
+    fn test_fdt_distance_map() {
+        // Single NUMA node - should skip distance map
+        let mut numa_nodes = BTreeMap::new();
+        numa_nodes.insert(0, create_test_numa_node(vec![0, 1], None));
+
+        let mut fdt = FdtWriter::new().unwrap();
+        let result = create_distance_map_node(&mut fdt, &numa_nodes);
+        assert!(result.is_ok(), "Should skip distance map for single node");
+
+        // Empty NUMA nodes - should handle gracefully
+        let numa_nodes = BTreeMap::new();
+        let mut fdt = FdtWriter::new().unwrap();
+        let result = create_distance_map_node(&mut fdt, &numa_nodes);
+        assert!(result.is_ok(), "Should handle empty NUMA nodes");
+
+        // Non-contiguous NUMA IDs (0, 2, 5) with distance symmetry
+        let mut numa_nodes = BTreeMap::new();
+
+        let mut node0 = create_test_numa_node(vec![0], None);
+        node0.distances.insert(2, 20);
+        // node0 has no explicit distance to node5
+
+        let mut node2 = create_test_numa_node(vec![1], None);
+        node2.distances.insert(0, 20);
+        node2.distances.insert(5, 25);
+
+        let mut node5 = create_test_numa_node(vec![2], None);
+        node5.distances.insert(0, 30);
+        node5.distances.insert(2, 25);
+        // node5->node0 (should be used for node0->node5)
+
+        numa_nodes.insert(0, node0);
+        numa_nodes.insert(2, node2);
+        numa_nodes.insert(5, node5);
+
+        // Verify IDs are sorted lexicographically
+        let mut numa_ids: Vec<u32> = numa_nodes.keys().cloned().collect();
+        numa_ids.sort_unstable();
+        assert_eq!(numa_ids, vec![0, 2, 5]);
+
+        let mut fdt = FdtWriter::new().unwrap();
+        let result = create_distance_map_node(&mut fdt, &numa_nodes);
+        assert!(
+            result.is_ok(),
+            "Should handle non-contiguous IDs and symmetry"
+        );
+
+        // Default distance (20) when no distance specified in either direction
+        let mut numa_nodes = BTreeMap::new();
+        numa_nodes.insert(0, create_test_numa_node(vec![0], None));
+        numa_nodes.insert(1, create_test_numa_node(vec![1], None));
+        // Neither node has distance to the other
+
+        let mut fdt = FdtWriter::new().unwrap();
+        let result = create_distance_map_node(&mut fdt, &numa_nodes);
+        assert!(result.is_ok(), "Should default to 20 for missing distances");
+    }
+}

arch/src/aarch64/layout.rs

@@ -139,11 +139,11 @@ pub const IRQ_BASE: u32 = 32;
 /// Number of supported interrupts
 pub const IRQ_NUM: u32 = 256;
 
-/// Base SPI interrupt number
-pub const SPI_BASE: u32 = 32;
+/// Base SPI interrupt number for the GICv2M MSI frame
+pub const GICV2M_SPI_BASE: u32 = 128;
 
-/// Total number of SPIs
-pub const SPI_NUM: u32 = 64;
+/// Total number of SPIs for the GICv2M MSI frame
+pub const GICV2M_SPI_NUM: u32 = 64;
 
 /// GICv2M compatible string
 pub const GIC_V2M_COMPATIBLE: &str = "arm,gic-v2m-frame";

arch/src/aarch64/mod.rs

@@ -15,7 +15,7 @@ use std::sync::{Arc, Mutex};
 
 use hypervisor::arch::aarch64::gic::Vgic;
 use hypervisor::arch::aarch64::regs::MPIDR_EL1;
-use log::{log_enabled, Level};
+use log::{Level, log_enabled};
 use thiserror::Error;
 use vm_memory::{Address, GuestAddress, GuestMemory, GuestMemoryAtomic};
 
@@ -66,8 +66,8 @@ pub struct EntryPoint {
 
 /// Configure the specified VCPU, and return its MPIDR.
 pub fn configure_vcpu(
-    vcpu: &Arc<dyn hypervisor::Vcpu>,
-    id: u8,
+    vcpu: &dyn hypervisor::Vcpu,
+    id: u32,
     boot_setup: Option<(EntryPoint, &GuestMemoryAtomic<GuestMemoryMmap>)>,
 ) -> super::Result<u64> {
     if let Some((kernel_entry_point, _guest_memory)) = boot_setup {
@@ -125,8 +125,8 @@ pub fn arch_memory_regions() -> Vec<(GuestAddress, usize, RegionType)> {
 pub fn configure_system<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::BuildHasher>(
     guest_mem: &GuestMemoryMmap,
     cmdline: &str,
-    vcpu_mpidr: Vec<u64>,
-    vcpu_topology: Option<(u8, u8, u8)>,
+    vcpu_mpidr: &[u64],
+    vcpu_topology: Option<(u16, u16, u16, u16)>,
     device_info: &HashMap<(DeviceType, String), T, S>,
     initrd: &Option<super::InitramfsConfig>,
     pci_space_info: &[PciSpaceInfo],
@@ -154,7 +154,7 @@ pub fn configure_system<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::Bui
         fdt::print_fdt(&fdt_final);
     }
 
-    fdt::write_fdt_to_memory(fdt_final, guest_mem).map_err(Error::WriteFdtToMemory)?;
+    fdt::write_fdt_to_memory(&fdt_final, guest_mem).map_err(Error::WriteFdtToMemory)?;
 
     Ok(())
 }
@@ -180,7 +180,7 @@ pub fn initramfs_load_addr(
     }
 }
 
-pub fn get_host_cpu_phys_bits(hypervisor: &Arc<dyn hypervisor::Hypervisor>) -> u8 {
+pub fn get_host_cpu_phys_bits(hypervisor: &dyn hypervisor::Hypervisor) -> u8 {
     let host_cpu_phys_bits = hypervisor.get_host_ipa_limit().try_into().unwrap();
     if host_cpu_phys_bits == 0 {
         // Host kernel does not support `get_host_ipa_limit`,
@@ -192,7 +192,7 @@ pub fn get_host_cpu_phys_bits(hypervisor: &Arc<dyn hypervisor::Hypervisor>) -> u
 }
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use super::*;
 
     #[test]

arch/src/lib.rs

@@ -8,9 +8,6 @@
 //! Implements platform specific functionality.
 //! Supported platforms: x86_64, aarch64, riscv64.
 
-#[macro_use]
-extern crate log;
-
 use std::collections::BTreeMap;
 use std::sync::Arc;
 use std::{fmt, result};
@@ -18,9 +15,6 @@ use std::{fmt, result};
 use serde::{Deserialize, Serialize};
 use thiserror::Error;
 
-#[cfg(target_arch = "x86_64")]
-use crate::x86_64::SgxEpcSection;
-
 type GuestMemoryMmap = vm_memory::GuestMemoryMmap<vm_memory::bitmap::AtomicBitmap>;
 type GuestRegionMmap = vm_memory::GuestRegionMmap<vm_memory::bitmap::AtomicBitmap>;
 
@@ -84,9 +78,9 @@ pub mod aarch64;
 
 #[cfg(target_arch = "aarch64")]
 pub use aarch64::{
-    arch_memory_regions, configure_system, configure_vcpu, fdt::DeviceInfoForFdt,
-    get_host_cpu_phys_bits, initramfs_load_addr, layout, layout::CMDLINE_MAX_SIZE,
-    layout::IRQ_BASE, uefi, EntryPoint, _NSIG,
+    _NSIG, EntryPoint, arch_memory_regions, configure_system, configure_vcpu,
+    fdt::DeviceInfoForFdt, get_host_cpu_phys_bits, initramfs_load_addr, layout,
+    layout::CMDLINE_MAX_SIZE, layout::IRQ_BASE, uefi,
 };
 
 /// Module for riscv64 related functionality.
@@ -95,9 +89,9 @@ pub mod riscv64;
 
 #[cfg(target_arch = "riscv64")]
 pub use riscv64::{
-    arch_memory_regions, configure_system, configure_vcpu, fdt::DeviceInfoForFdt,
-    get_host_cpu_phys_bits, initramfs_load_addr, layout, layout::CMDLINE_MAX_SIZE,
-    layout::IRQ_BASE, EntryPoint, _NSIG,
+    _NSIG, EntryPoint, arch_memory_regions, configure_system, configure_vcpu,
+    fdt::DeviceInfoForFdt, get_host_cpu_phys_bits, initramfs_load_addr, layout,
+    layout::CMDLINE_MAX_SIZE, layout::IRQ_BASE, uefi,
 };
 
 #[cfg(target_arch = "x86_64")]
@@ -105,10 +99,9 @@ pub mod x86_64;
 
 #[cfg(target_arch = "x86_64")]
 pub use x86_64::{
-    arch_memory_regions, configure_system, configure_vcpu, generate_common_cpuid,
-    generate_ram_ranges, get_host_cpu_phys_bits, initramfs_load_addr, layout,
-    layout::CMDLINE_MAX_SIZE, layout::CMDLINE_START, regs, CpuidConfig, CpuidFeatureEntry,
-    EntryPoint, _NSIG,
+    _NSIG, CpuidConfig, CpuidFeatureEntry, EntryPoint, arch_memory_regions, configure_system,
+    configure_vcpu, generate_common_cpuid, generate_ram_ranges, get_host_cpu_phys_bits,
+    initramfs_load_addr, layout, layout::CMDLINE_MAX_SIZE, layout::CMDLINE_START, regs,
 };
 
 /// Safe wrapper for `sysconf(_SC_PAGESIZE)`.
@@ -123,12 +116,11 @@ fn pagesize() -> usize {
 pub struct NumaNode {
     pub memory_regions: Vec<Arc<GuestRegionMmap>>,
     pub hotplug_regions: Vec<Arc<GuestRegionMmap>>,
-    pub cpus: Vec<u8>,
+    pub cpus: Vec<u32>,
     pub pci_segments: Vec<u16>,
     pub distances: BTreeMap<u32, u8>,
     pub memory_zones: Vec<String>,
-    #[cfg(target_arch = "x86_64")]
-    pub sgx_epc_sections: Vec<SgxEpcSection>,
+    pub device_id: Option<String>,
 }
 
 pub type NumaNodes = BTreeMap<u32, NumaNode>;
@@ -155,6 +147,9 @@ pub enum DeviceType {
     /// Device Type: GPIO.
     #[cfg(target_arch = "aarch64")]
     Gpio,
+    /// Device Type: fw_cfg.
+    #[cfg(feature = "fw_cfg")]
+    FwCfg,
 }
 
 /// Default (smallest) memory page size for the supported architectures.

arch/src/riscv64/fdt.rs

@@ -15,6 +15,7 @@ use std::{cmp, result, str};
 
 use byteorder::{BigEndian, ByteOrder};
 use hypervisor::arch::riscv64::aia::Vaia;
+use log::debug;
 use thiserror::Error;
 use vm_fdt::{FdtWriter, FdtWriterResult};
 use vm_memory::{Address, Bytes, GuestMemory, GuestMemoryError, GuestMemoryRegion};
@@ -65,6 +66,7 @@ pub fn create_fdt<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::BuildHash
     guest_mem: &GuestMemoryMmap,
     cmdline: &str,
     num_vcpu: u32,
+    isa_string: &str,
     device_info: &HashMap<(DeviceType, String), T, S>,
     aia_device: &Arc<Mutex<dyn Vaia>>,
     initrd: &Option<InitramfsConfig>,
@@ -84,7 +86,7 @@ pub fn create_fdt<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::BuildHash
     // Properties
     fdt.property_u32("#address-cells", ADDRESS_CELLS)?;
     fdt.property_u32("#size-cells", SIZE_CELLS)?;
-    create_cpu_nodes(&mut fdt, num_vcpu)?;
+    create_cpu_nodes(&mut fdt, num_vcpu, isa_string)?;
     create_memory_node(&mut fdt, guest_mem)?;
     create_chosen_node(&mut fdt, cmdline, initrd)?;
     create_aia_node(&mut fdt, aia_device)?;
@@ -99,16 +101,16 @@ pub fn create_fdt<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::BuildHash
     Ok(fdt_final)
 }
 
-pub fn write_fdt_to_memory(fdt_final: Vec<u8>, guest_mem: &GuestMemoryMmap) -> Result<()> {
+pub fn write_fdt_to_memory(fdt_final: &[u8], guest_mem: &GuestMemoryMmap) -> Result<()> {
     // Write FDT to memory.
     guest_mem
-        .write_slice(fdt_final.as_slice(), super::layout::FDT_START)
+        .write_slice(fdt_final, super::layout::FDT_START)
         .map_err(Error::WriteFdtToMemory)?;
     Ok(())
 }
 
 // Following are the auxiliary function for creating the different nodes that we append to our FDT.
-fn create_cpu_nodes(fdt: &mut FdtWriter, num_cpus: u32) -> FdtWriterResult<()> {
+fn create_cpu_nodes(fdt: &mut FdtWriter, num_cpus: u32, isa_string: &str) -> FdtWriterResult<()> {
     // See https://elixir.bootlin.com/linux/v6.10/source/Documentation/devicetree/bindings/riscv/cpus.yaml
     let cpus = fdt.begin_node("cpus")?;
     // As per documentation, on RISC-V 64-bit systems value should be set to 1.
@@ -119,11 +121,11 @@ fn create_cpu_nodes(fdt: &mut FdtWriter, num_cpus: u32) -> FdtWriterResult<()> {
     fdt.property_u32("timebase-frequency", timebase_frequency)?;
 
     for cpu_index in 0..num_cpus {
-        let cpu = fdt.begin_node(&format!("cpu@{:x}", cpu_index))?;
+        let cpu = fdt.begin_node(&format!("cpu@{cpu_index:x}"))?;
         fdt.property_string("device_type", "cpu")?;
         fdt.property_string("compatible", "riscv")?;
         fdt.property_string("mmu-type", "sv48")?;
-        fdt.property_string("riscv,isa", "rv64imafdc_smaia_ssaia")?;
+        fdt.property_string("riscv,isa", isa_string)?;
         fdt.property_string("status", "okay")?;
         fdt.property_u32("reg", cpu_index)?;
         fdt.property_u32("phandle", CPU_BASE_PHANDLE + cpu_index)?;
@@ -184,7 +186,7 @@ fn create_memory_node(fdt: &mut FdtWriter, guest_mem: &GuestMemoryMmap) -> FdtWr
     }
 
     let ram_start = super::layout::RAM_START.raw_value();
-    let memory_node_name = format!("memory@{:x}", ram_start);
+    let memory_node_name = format!("memory@{ram_start:x}");
     let memory_node = fdt.begin_node(&memory_node_name)?;
     fdt.property_string("device_type", "memory")?;
     fdt.property_array_u64("reg", &mem_reg_property)?;
@@ -448,10 +450,7 @@ fn print_node(node: fdt_parser::node::FdtNode<'_, '_>, n_spaces: usize) {
         //   - At first, try to convert it to CStr and print,
         //   - If failed, print it as u32 array.
         let value_result = match CStr::from_bytes_with_nul(value) {
-            Ok(value_cstr) => match value_cstr.to_str() {
-                Ok(value_str) => Some(value_str),
-                Err(_e) => None,
-            },
+            Ok(value_cstr) => value_cstr.to_str().ok(),
             Err(_e) => None,
         };
 
@@ -474,7 +473,7 @@ fn print_node(node: fdt_parser::node::FdtNode<'_, '_>, n_spaces: usize) {
                 array,
                 indent = (n_spaces + 2)
             );
-        };
+        }
     }
 
     // Print children nodes if there is any

arch/src/riscv64/layout.rs

@@ -44,16 +44,23 @@
 //           |                                                               |
 //           |                            APLICs                             |
 //           |                                                               |
+//    4 MB   +---------------------------------------------------------------+
+//           |                          UEFI flash                           |
 //    0 GB   +---------------------------------------------------------------+
 //
 //
 
 use vm_memory::GuestAddress;
 
+/// 0x0 ~ 0x40_0000 (4 MiB) is reserved to UEFI
+/// UEFI binary size is required less than 3 MiB, reserving 4 MiB is enough.
+pub const UEFI_START: GuestAddress = GuestAddress(0);
+pub const UEFI_SIZE: u64 = 0x040_0000;
+
 /// AIA related devices
 /// See https://elixir.bootlin.com/linux/v6.10/source/arch/riscv/include/uapi/asm/kvm.h
-/// 0x0 ~ 0x0400_0000 (64 MiB) resides APLICs
-pub const APLIC_START: GuestAddress = GuestAddress(0);
+/// 0x40_0000 ~ 0x0400_0000 (64 MiB) resides APLICs
+pub const APLIC_START: GuestAddress = GuestAddress(0x40_0000);
 pub const APLIC_SIZE: u64 = 0x4000;
 
 /// 0x0400_0000 ~ 0x0800_0000 (64 MiB) resides IMSICs
@@ -91,7 +98,12 @@ pub const CMDLINE_MAX_SIZE: usize = 1024;
 pub const FDT_START: GuestAddress = RAM_START;
 pub const FDT_MAX_SIZE: u64 = 0x1_0000;
 
-/// Kernel start after FDT
+/// Put ACPI table above dtb
+pub const ACPI_START: GuestAddress = GuestAddress(RAM_START.0 + FDT_MAX_SIZE);
+pub const ACPI_MAX_SIZE: u64 = 0x20_0000;
+pub const RSDP_POINTER: GuestAddress = ACPI_START;
+
+/// Kernel start after FDT and ACPI
 pub const KERNEL_START: GuestAddress = GuestAddress(RAM_START.0 + FDT_MAX_SIZE);
 
 /// Pci high memory base

arch/src/riscv64/mod.rs

@@ -7,19 +7,28 @@
 pub mod fdt;
 /// Layout for this riscv64 system.
 pub mod layout;
+/// Module for loading UEFI binary.
+pub mod uefi;
 
 use std::collections::HashMap;
 use std::fmt::Debug;
+use std::fs::File;
+use std::io::{BufRead, BufReader};
 use std::sync::{Arc, Mutex};
 
 use hypervisor::arch::riscv64::aia::Vaia;
-use log::{log_enabled, Level};
+use log::{Level, log_enabled};
 use thiserror::Error;
 use vm_memory::{Address, GuestAddress, GuestMemory, GuestMemoryAtomic};
 
 pub use self::fdt::DeviceInfoForFdt;
 use crate::{DeviceType, GuestMemoryMmap, PciSpaceInfo, RegionType};
 
+pub const CLOUDHV_IRQCHIP_NUM_MSIS: u16 = 255;
+pub const CLOUDHV_IRQCHIP_NUM_SOURCES: u8 = 96;
+pub const CLOUDHV_IRQCHIP_NUM_PRIO_BITS: u8 = 3;
+pub const CLOUDHV_IRQCHIP_MAX_GUESTS_BITS: u8 = 3;
+pub const CLOUDHV_IRQCHIP_MAX_GUESTS: u8 = (1 << CLOUDHV_IRQCHIP_MAX_GUESTS_BITS) - 1;
 pub const _NSIG: i32 = 65;
 
 /// Errors thrown while configuring riscv64 system.
@@ -44,6 +53,22 @@ pub enum Error {
     /// Error configuring the general purpose registers
     #[error("Error configuring the general purpose registers")]
     RegsConfiguration(#[source] hypervisor::HypervisorCpuError),
+
+    /// Error opening /proc/cpuinfo
+    #[error("Error opening /proc/cpuinfo")]
+    OpenCpuInfo(#[source] std::io::Error),
+
+    /// Error reading /proc/cpuinfo
+    #[error("Error reading /proc/cpuinfo")]
+    ReadCpuInfo(#[source] std::io::Error),
+
+    /// Invalid ISA string
+    #[error("Invalid ISA string: {0}")]
+    InvalidIsaString(String),
+
+    /// Error parsing /proc/cpuinfo
+    #[error("Error parsing /proc/cpuinfo")]
+    CpuInfoParsing,
 }
 
 #[derive(Debug, Copy, Clone)]
@@ -56,8 +81,8 @@ pub struct EntryPoint {
 
 /// Configure the specified VCPU, and return its MPIDR.
 pub fn configure_vcpu(
-    vcpu: &Arc<dyn hypervisor::Vcpu>,
-    id: u8,
+    vcpu: &dyn hypervisor::Vcpu,
+    id: u32,
     boot_setup: Option<(EntryPoint, &GuestMemoryAtomic<GuestMemoryMmap>)>,
 ) -> super::Result<()> {
     if let Some((kernel_entry_point, _guest_memory)) = boot_setup {
@@ -97,6 +122,43 @@ pub fn arch_memory_regions() -> Vec<(GuestAddress, usize, RegionType)> {
     ]
 }
 
+// Read the first "isa" string from /proc/cpuinfo and filter out the H extension,
+// while correctly preserving multi-letter extensions.
+fn isa_string_from_host() -> Result<String, Error> {
+    let file = File::open("/proc/cpuinfo").map_err(Error::OpenCpuInfo)?;
+    let reader = BufReader::new(file);
+
+    for line in reader.lines() {
+        let line = line.map_err(Error::ReadCpuInfo)?;
+        let trimmed_line = line.trim();
+
+        if trimmed_line.starts_with("isa") {
+            let parts: Vec<&str> = trimmed_line.split(':').collect();
+            if parts.len() == 2 {
+                let isa_string = parts[1].trim();
+
+                // Split the string by underscores to separate single letter vs long-form
+                // extensions
+                let mut components: Vec<String> =
+                    isa_string.split('_').map(|s| s.to_string()).collect();
+
+                if components.is_empty() {
+                    return Err(Error::InvalidIsaString(isa_string.to_string()));
+                }
+
+                // Remove H extension if present in single letter extensions
+                let first_component = components[0].chars().filter(|&c| c != 'h').collect();
+
+                components[0] = first_component;
+
+                return Ok(components.join("_"));
+            }
+        }
+    }
+
+    Err(Error::CpuInfoParsing)
+}
+
 /// Configures the system and should be called once per vm before starting vcpu threads.
 #[allow(clippy::too_many_arguments)]
 pub fn configure_system<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::BuildHasher>(
@@ -108,10 +170,12 @@ pub fn configure_system<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::Bui
     pci_space_info: &[PciSpaceInfo],
     aia_device: &Arc<Mutex<dyn Vaia>>,
 ) -> super::Result<()> {
+    let isa_string = isa_string_from_host()?;
     let fdt_final = fdt::create_fdt(
         guest_mem,
         cmdline,
         num_vcpu,
+        &isa_string,
         device_info,
         aia_device,
         initrd,
@@ -123,7 +187,7 @@ pub fn configure_system<T: DeviceInfoForFdt + Clone + Debug, S: ::std::hash::Bui
         fdt::print_fdt(&fdt_final);
     }
 
-    fdt::write_fdt_to_memory(fdt_final, guest_mem).map_err(Error::WriteFdtToMemory)?;
+    fdt::write_fdt_to_memory(&fdt_final, guest_mem).map_err(Error::WriteFdtToMemory)?;
 
     Ok(())
 }
@@ -149,12 +213,12 @@ pub fn initramfs_load_addr(
     }
 }
 
-pub fn get_host_cpu_phys_bits(_hypervisor: &Arc<dyn hypervisor::Hypervisor>) -> u8 {
+pub fn get_host_cpu_phys_bits(_hypervisor: &dyn hypervisor::Hypervisor) -> u8 {
     40
 }
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use super::*;
 
     #[test]

arch/src/riscv64/uefi.rs

@@ -0,0 +1,50 @@
+// Copyright 2020 Arm Limited (or its affiliates). All rights reserved.
+//
+// SPDX-License-Identifier: Apache-2.0
+
+use std::io::{Read, Seek, SeekFrom};
+use std::os::fd::AsFd;
+use std::result;
+
+use thiserror::Error;
+use vm_memory::{GuestAddress, GuestMemory};
+
+/// Errors thrown while loading UEFI binary
+#[derive(Debug, Error)]
+pub enum Error {
+    /// Unable to seek to UEFI image start.
+    #[error("Unable to seek to UEFI image start")]
+    SeekUefiStart,
+    /// Unable to seek to UEFI image end.
+    #[error("Unable to seek to UEFI image end")]
+    SeekUefiEnd,
+    /// UEFI image too big.
+    #[error("UEFI image too big")]
+    UefiTooBig,
+    /// Unable to read UEFI image
+    #[error("Unable to read UEFI image")]
+    ReadUefiImage,
+}
+type Result<T> = result::Result<T, Error>;
+
+pub fn load_uefi<F, M: GuestMemory>(
+    guest_mem: &M,
+    guest_addr: GuestAddress,
+    uefi_image: &mut F,
+) -> Result<()>
+where
+    F: Read + Seek + AsFd,
+{
+    let uefi_size = uefi_image
+        .seek(SeekFrom::End(0))
+        .map_err(|_| Error::SeekUefiEnd)? as usize;
+
+    // edk2 image on virtual platform is smaller than 3M
+    if uefi_size > 0x300000 {
+        return Err(Error::UefiTooBig);
+    }
+    uefi_image.rewind().map_err(|_| Error::SeekUefiStart)?;
+    guest_mem
+        .read_exact_volatile_from(guest_addr, &mut uefi_image.as_fd(), uefi_size)
+        .map_err(|_| Error::ReadUefiImage)
+}

arch/src/x86_64/interrupts.rs

@@ -6,7 +6,6 @@
 // found in the LICENSE-BSD-3-Clause file.
 
 use std::result;
-use std::sync::Arc;
 
 pub type Result<T> = result::Result<T, hypervisor::HypervisorCpuError>;
 
@@ -24,7 +23,7 @@ pub fn set_apic_delivery_mode(reg: u32, mode: u32) -> u32 {
 ///
 /// # Arguments
 /// * `vcpu` - The VCPU object to configure.
-pub fn set_lint(vcpu: &Arc<dyn hypervisor::Vcpu>) -> Result<()> {
+pub fn set_lint(vcpu: &dyn hypervisor::Vcpu) -> Result<()> {
     let mut klapic = vcpu.get_lapic()?;
 
     let lvt_lint0 = klapic.get_klapic_reg(APIC_LVT0);

arch/src/x86_64/mod.rs

@@ -6,43 +6,55 @@
 // Portions Copyright 2017 The Chromium OS Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE-BSD-3-Clause file.
-use std::sync::Arc;
+
 pub mod interrupts;
 pub mod layout;
+pub mod regs;
+
+#[cfg(feature = "tdx")]
+pub mod tdx;
+
 mod mpspec;
 mod mptable;
-pub mod regs;
-use std::collections::BTreeMap;
+mod smbios;
+
+use std::arch::x86_64;
 use std::mem;
 
-use hypervisor::arch::x86::{CpuIdEntry, CPUID_FLAG_VALID_INDEX};
+use hypervisor::arch::x86::{CPUID_FLAG_VALID_INDEX, CpuIdEntry};
 use hypervisor::{CpuVendor, HypervisorCpuError, HypervisorError};
 use linux_loader::loader::bootparam::{boot_params, setup_header};
 use linux_loader::loader::elf::start_info::{
     hvm_memmap_table_entry, hvm_modlist_entry, hvm_start_info,
 };
+use log::{debug, error, info};
 use thiserror::Error;
 use vm_memory::{
     Address, Bytes, GuestAddress, GuestAddressSpace, GuestMemory, GuestMemoryAtomic,
-    GuestMemoryRegion, GuestUsize,
+    GuestMemoryRegion,
 };
 
 use crate::{GuestMemoryMmap, InitramfsConfig, RegionType};
-mod smbios;
-use std::arch::x86_64;
-#[cfg(feature = "tdx")]
-pub mod tdx;
+
+// While modern architectures support more than 255 CPUs via x2APIC,
+// legacy devices such as mptable support at most 254 CPUs.
+pub const MAX_SUPPORTED_CPUS_LEGACY: u32 = 254;
 
 // CPUID feature bits
 #[cfg(feature = "kvm")]
 const TSC_DEADLINE_TIMER_ECX_BIT: u8 = 24; // tsc deadline timer ecx bit.
 const HYPERVISOR_ECX_BIT: u8 = 31; // Hypervisor ecx bit.
+const VMX_ECX_BIT: u8 = 5; // VMX for Intel
+const SVM_ECX_BIT: u8 = 2; // SVM for AMD
 const MTRR_EDX_BIT: u8 = 12; // Hypervisor ecx bit.
 const INVARIANT_TSC_EDX_BIT: u8 = 8; // Invariant TSC bit on 0x8000_0007 EDX
 const AMX_BF16: u8 = 22; // AMX tile computation on bfloat16 numbers
 const AMX_TILE: u8 = 24; // AMX tile load/store instructions
 const AMX_INT8: u8 = 25; // AMX tile computation on 8-bit integers
 
+const AMX_FP16: u8 = 21; // AMX tile computation on fp16 numbers
+const AMX_COMPLEX: u8 = 8; // AMX tile computation on complex numbers
+
 // KVM feature bits
 #[cfg(feature = "tdx")]
 const KVM_FEATURE_CLOCKSOURCE_BIT: u8 = 0;
@@ -57,6 +69,8 @@ const KVM_FEATURE_ASYNC_PF_VMEXIT_BIT: u8 = 10;
 #[cfg(feature = "tdx")]
 const KVM_FEATURE_STEAL_TIME_BIT: u8 = 5;
 
+const KVM_FEATURE_MSI_EXT_DEST_ID: u8 = 15;
+
 pub const _NSIG: i32 = 65;
 
 #[derive(Debug, Copy, Clone)]
@@ -73,55 +87,7 @@ pub struct EntryPoint {
 const E820_RAM: u32 = 1;
 const E820_RESERVED: u32 = 2;
 
-#[derive(Clone)]
-pub struct SgxEpcSection {
-    start: GuestAddress,
-    size: GuestUsize,
-}
-
-impl SgxEpcSection {
-    pub fn new(start: GuestAddress, size: GuestUsize) -> Self {
-        SgxEpcSection { start, size }
-    }
-    pub fn start(&self) -> GuestAddress {
-        self.start
-    }
-    pub fn size(&self) -> GuestUsize {
-        self.size
-    }
-}
-
-#[derive(Clone)]
-pub struct SgxEpcRegion {
-    start: GuestAddress,
-    size: GuestUsize,
-    epc_sections: BTreeMap<String, SgxEpcSection>,
-}
-
-impl SgxEpcRegion {
-    pub fn new(start: GuestAddress, size: GuestUsize) -> Self {
-        SgxEpcRegion {
-            start,
-            size,
-            epc_sections: BTreeMap::new(),
-        }
-    }
-    pub fn start(&self) -> GuestAddress {
-        self.start
-    }
-    pub fn size(&self) -> GuestUsize {
-        self.size
-    }
-    pub fn epc_sections(&self) -> &BTreeMap<String, SgxEpcSection> {
-        &self.epc_sections
-    }
-    pub fn insert(&mut self, id: String, epc_section: SgxEpcSection) {
-        self.epc_sections.insert(id, epc_section);
-    }
-}
-
 pub struct CpuidConfig {
-    pub sgx_epc_sections: Option<Vec<SgxEpcSection>>,
     pub phys_bits: u8,
     pub kvm_hyperv: bool,
     #[cfg(feature = "tdx")]
@@ -163,18 +129,6 @@ pub enum Error {
     #[error("Error setting up SMBIOS table")]
     SmbiosSetup(#[source] smbios::Error),
 
-    /// Could not find any SGX EPC section
-    #[error("Could not find any SGX EPC section")]
-    NoSgxEpcSection,
-
-    /// Missing SGX CPU feature
-    #[error("Missing SGX CPU feature")]
-    MissingSgxFeature,
-
-    /// Missing SGX_LC CPU feature
-    #[error("Missing SGX_LC CPU feature")]
-    MissingSgxLaunchControlFeature,
-
     /// Error getting supported CPUID through the hypervisor (kvm/mshv) API
     #[error("Error getting supported CPUID through the hypervisor API")]
     CpuidGetSupported(#[source] HypervisorError),
@@ -209,11 +163,11 @@ pub enum Error {
     E820Configuration,
 }
 
-pub fn get_x2apic_id(cpu_id: u32, topology: Option<(u8, u8, u8)>) -> u32 {
+pub fn get_x2apic_id(cpu_id: u32, topology: Option<(u16, u16, u16, u16)>) -> u32 {
     if let Some(t) = topology {
-        let thread_mask_width = u8::BITS - (t.0 - 1).leading_zeros();
-        let core_mask_width = u8::BITS - (t.1 - 1).leading_zeros();
-        let die_mask_width = u8::BITS - (t.2 - 1).leading_zeros();
+        let thread_mask_width = u16::BITS - (t.0 - 1).leading_zeros();
+        let core_mask_width = u16::BITS - (t.1 - 1).leading_zeros();
+        let die_mask_width = u16::BITS - (t.2 - 1).leading_zeros();
 
         let thread_id = cpu_id % (t.0 as u32);
         let core_id = cpu_id / (t.0 as u32) % (t.1 as u32);
@@ -229,6 +183,13 @@ pub fn get_x2apic_id(cpu_id: u32, topology: Option<(u8, u8, u8)>) -> u32 {
     cpu_id
 }
 
+pub fn get_max_x2apic_id(topology: (u16, u16, u16, u16)) -> u32 {
+    get_x2apic_id(
+        (topology.0 as u32 * topology.1 as u32 * topology.2 as u32 * topology.3 as u32) - 1,
+        Some(topology),
+    )
+}
+
 #[derive(Copy, Clone, Debug)]
 pub enum CpuidReg {
     EAX,
@@ -327,7 +288,7 @@ impl CpuidPatch {
         }
     }
 
-    pub fn patch_cpuid(cpuid: &mut [CpuIdEntry], patches: Vec<CpuidPatch>) {
+    pub fn patch_cpuid(cpuid: &mut [CpuIdEntry], patches: &[CpuidPatch]) {
         for entry in cpuid {
             for patch in patches.iter() {
                 if entry.function == patch.function && entry.index == patch.index {
@@ -454,7 +415,7 @@ impl CpuidFeatureEntry {
                 feature_reg: CpuidReg::EDX,
                 compatible_check: CpuidCompatibleCheck::BitwiseSubset,
             },
-            // KVM CPUID bits: https://www.kernel.org/doc/html/latest/virt/kvm/cpuid.html
+            // KVM CPUID bits: https://www.kernel.org/doc/html/latest/virt/kvm/x86/cpuid.html
             // Leaf 0x4000_0000, EAX/EBX/ECX/EDX, KVM CPUID SIGNATURE
             CpuidFeatureEntry {
                 function: 0x4000_0000,
@@ -573,9 +534,13 @@ impl CpuidFeatureEntry {
                 error!(
                     "Detected incompatible CPUID entry: leaf={:#02x} (subleaf={:#02x}), register='{:?}', \
                     compatible_check='{:?}', source VM feature='{:#04x}', destination VM feature'{:#04x}'.",
-                    entry.function, entry.index, entry.feature_reg,
-                    entry.compatible_check, src_vm_feature, dest_vm_feature
-                    );
+                    entry.function,
+                    entry.index,
+                    entry.feature_reg,
+                    entry.compatible_check,
+                    src_vm_feature,
+                    dest_vm_feature
+                );
 
                 compatible = false;
             }
@@ -591,9 +556,10 @@ impl CpuidFeatureEntry {
 }
 
 pub fn generate_common_cpuid(
-    hypervisor: &Arc<dyn hypervisor::Hypervisor>,
+    hypervisor: &dyn hypervisor::Hypervisor,
     config: &CpuidConfig,
 ) -> super::Result<Vec<CpuIdEntry>> {
+    #[allow(unused_unsafe)]
     // SAFETY: cpuid called with valid leaves
     if unsafe { x86_64::__cpuid(1) }.ecx & (1 << HYPERVISOR_ECX_BIT) == 1 << HYPERVISOR_ECX_BIT {
         // SAFETY: cpuid called with valid leaves
@@ -660,18 +626,14 @@ pub fn generate_common_cpuid(
         .get_supported_cpuid()
         .map_err(Error::CpuidGetSupported)?;
 
-    CpuidPatch::patch_cpuid(&mut cpuid, cpuid_patches);
-
-    if let Some(sgx_epc_sections) = &config.sgx_epc_sections {
-        update_cpuid_sgx(&mut cpuid, sgx_epc_sections)?;
-    }
+    CpuidPatch::patch_cpuid(&mut cpuid, &cpuid_patches);
 
     #[cfg(feature = "tdx")]
     let tdx_capabilities = if config.tdx {
         let caps = hypervisor
             .tdx_capabilities()
             .map_err(Error::TdxCapabilities)?;
-        info!("TDX capabilities {:#?}", caps);
+        info!("TDX capabilities {caps:#?}");
         Some(caps)
     } else {
         None
@@ -682,8 +644,14 @@ pub fn generate_common_cpuid(
         match entry.function {
             // Clear AMX related bits if the AMX feature is not enabled
             0x7 => {
-                if !config.amx && entry.index == 0 {
-                    entry.edx &= !((1 << AMX_BF16) | (1 << AMX_TILE) | (1 << AMX_INT8))
+                if !config.amx {
+                    if entry.index == 0 {
+                        entry.edx &= !((1 << AMX_BF16) | (1 << AMX_TILE) | (1 << AMX_INT8));
+                    }
+                    if entry.index == 1 {
+                        entry.eax &= !(1 << AMX_FP16);
+                        entry.edx &= !(1 << AMX_COMPLEX);
+                    }
                 }
             }
             0xd =>
@@ -705,9 +673,29 @@ pub fn generate_common_cpuid(
                     }
                 }
             }
+            0x1d => {
+                // Tile Information (purely AMX related).
+                if !config.amx {
+                    entry.eax = 0;
+                    entry.ebx = 0;
+                    entry.ecx = 0;
+                    entry.edx = 0;
+                }
+            }
+            0x1e => {
+                // TMUL information (purely AMX related)
+                if !config.amx {
+                    entry.eax = 0;
+                    entry.ebx = 0;
+                    entry.ecx = 0;
+                    entry.edx = 0;
+                }
+            }
+
             // Copy host L1 cache details if not populated by KVM
             0x8000_0005 => {
                 if entry.eax == 0 && entry.ebx == 0 && entry.ecx == 0 && entry.edx == 0 {
+                    #[allow(unused_unsafe)]
                     // SAFETY: cpuid called with valid leaves
                     if unsafe { std::arch::x86_64::__cpuid(0x8000_0000).eax } >= 0x8000_0005 {
                         // SAFETY: cpuid called with valid leaves
@@ -722,8 +710,10 @@ pub fn generate_common_cpuid(
             // Copy host L2 cache details if not populated by KVM
             0x8000_0006 => {
                 if entry.eax == 0 && entry.ebx == 0 && entry.ecx == 0 && entry.edx == 0 {
+                    #[allow(unused_unsafe)]
                     // SAFETY: cpuid called with valid leaves
                     if unsafe { std::arch::x86_64::__cpuid(0x8000_0000).eax } >= 0x8000_0006 {
+                        #[allow(unused_unsafe)]
                         // SAFETY: cpuid called with valid leaves
                         let leaf = unsafe { std::arch::x86_64::__cpuid(0x8000_0006) };
                         entry.eax = leaf.eax;
@@ -738,6 +728,10 @@ pub fn generate_common_cpuid(
                 entry.eax = (entry.eax & 0xffff_ff00) | (config.phys_bits as u32 & 0xff);
             }
             0x4000_0001 => {
+                // Enable KVM_FEATURE_MSI_EXT_DEST_ID. This allows the guest to target
+                // device interrupts to cpus with APIC IDs > 254 without interrupt remapping.
+                entry.eax |= 1 << KVM_FEATURE_MSI_EXT_DEST_ID;
+
                 // These features are not supported by TDX
                 #[cfg(feature = "tdx")]
                 if config.tdx {
@@ -746,7 +740,7 @@ pub fn generate_common_cpuid(
                         | (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT)
                         | (1 << KVM_FEATURE_ASYNC_PF_BIT)
                         | (1 << KVM_FEATURE_ASYNC_PF_VMEXIT_BIT)
-                        | (1 << KVM_FEATURE_STEAL_TIME_BIT))
+                        | (1 << KVM_FEATURE_STEAL_TIME_BIT));
                 }
             }
             _ => {}
@@ -757,6 +751,7 @@ pub fn generate_common_cpuid(
     for i in 0x8000_0002..=0x8000_0004 {
         cpuid.retain(|c| c.function != i);
         // SAFETY: call cpuid with valid leaves
+        #[allow(unused_unsafe)]
         let leaf = unsafe { std::arch::x86_64::__cpuid(i) };
         cpuid.push(CpuIdEntry {
             function: i,
@@ -818,16 +813,18 @@ pub fn generate_common_cpuid(
     Ok(cpuid)
 }
 
+#[allow(clippy::too_many_arguments)]
 pub fn configure_vcpu(
-    vcpu: &Arc<dyn hypervisor::Vcpu>,
-    id: u8,
+    vcpu: &dyn hypervisor::Vcpu,
+    id: u32,
     boot_setup: Option<(EntryPoint, &GuestMemoryAtomic<GuestMemoryMmap>)>,
     cpuid: Vec<CpuIdEntry>,
     kvm_hyperv: bool,
     cpu_vendor: CpuVendor,
-    topology: Option<(u8, u8, u8)>,
+    topology: (u16, u16, u16, u16),
+    nested: bool,
 ) -> super::Result<()> {
-    let x2apic_id = get_x2apic_id(id as u32, topology);
+    let x2apic_id = get_x2apic_id(id, Some(topology));
 
     // Per vCPU CPUID changes; common are handled via generate_common_cpuid()
     let mut cpuid = cpuid;
@@ -844,45 +841,48 @@ pub fn configure_vcpu(
             entry.ebx &= 0xffffff;
             entry.ebx |= x2apic_id << 24;
             apic_id_patched = true;
+            if !nested {
+                // Disable nested virtualization for Intel
+                entry.ecx &= !(1 << VMX_ECX_BIT);
+            }
+            break;
+        }
+        if entry.function == 0x8000_0001 {
+            if !nested {
+                // Disable the nested virtualization for AMD
+                entry.ecx &= !(1 << SVM_ECX_BIT);
+            }
             break;
         }
     }
     assert!(apic_id_patched);
 
-    if let Some(t) = topology {
-        update_cpuid_topology(&mut cpuid, t.0, t.1, t.2, cpu_vendor, id);
-    }
+    update_cpuid_topology(
+        &mut cpuid, topology.0, topology.1, topology.2, topology.3, cpu_vendor, id,
+    );
 
     // The TSC frequency CPUID leaf should not be included when running with HyperV emulation
-    if !kvm_hyperv {
-        if let Some(tsc_khz) = vcpu.tsc_khz().map_err(Error::GetTscFrequency)? {
-            // Need to check that the TSC doesn't vary with dynamic frequency
-            // SAFETY: cpuid called with valid leaves
-            if unsafe { std::arch::x86_64::__cpuid(0x8000_0007) }.edx
-                & (1u32 << INVARIANT_TSC_EDX_BIT)
-                > 0
-            {
-                CpuidPatch::set_cpuid_reg(
-                    &mut cpuid,
-                    0x4000_0000,
-                    None,
-                    CpuidReg::EAX,
-                    0x4000_0010,
-                );
-                cpuid.retain(|c| c.function != 0x4000_0010);
-                cpuid.push(CpuIdEntry {
-                    function: 0x4000_0010,
-                    eax: tsc_khz,
-                    ebx: 1000000, /* LAPIC resolution of 1ns (freq: 1GHz) is hardcoded in KVM's
-                                   * APIC_BUS_CYCLE_NS */
-                    ..Default::default()
-                });
-            };
+    if !kvm_hyperv && let Some(tsc_khz) = vcpu.tsc_khz().map_err(Error::GetTscFrequency)? {
+        // Need to check that the TSC doesn't vary with dynamic frequency
+        #[allow(unused_unsafe)]
+        // SAFETY: cpuid called with valid leaves
+        if unsafe { std::arch::x86_64::__cpuid(0x8000_0007) }.edx & (1u32 << INVARIANT_TSC_EDX_BIT)
+            > 0
+        {
+            CpuidPatch::set_cpuid_reg(&mut cpuid, 0x4000_0000, None, CpuidReg::EAX, 0x4000_0010);
+            cpuid.retain(|c| c.function != 0x4000_0010);
+            cpuid.push(CpuIdEntry {
+                function: 0x4000_0010,
+                eax: tsc_khz,
+                ebx: 1000000, /* LAPIC resolution of 1ns (freq: 1GHz) is hardcoded in KVM's
+                               * APIC_BUS_CYCLE_NS */
+                ..Default::default()
+            });
         }
     }
 
     for c in &cpuid {
-        debug!("{}", c);
+        debug!("{c}");
     }
 
     vcpu.set_cpuid2(&cpuid)
@@ -896,7 +896,15 @@ pub fn configure_vcpu(
     if let Some((kernel_entry_point, guest_memory)) = boot_setup {
         regs::setup_regs(vcpu, kernel_entry_point).map_err(Error::RegsConfiguration)?;
         regs::setup_fpu(vcpu).map_err(Error::FpuConfiguration)?;
-        regs::setup_sregs(&guest_memory.memory(), vcpu).map_err(Error::SregsConfiguration)?;
+
+        // CPUs are required (by Intel sdm spec) to boot in x2apic mode if any
+        // of the apic IDs is larger than 255. Experimentally, the Linux kernel
+        // does not recognize the last vCPU if x2apic is not enabled when
+        // there are 256 vCPUs in a flat hierarchy (i.e. max x2apic ID is 255),
+        // so we need to enable x2apic in this case as well.
+        let enable_x2_apic_mode = get_max_x2apic_id(topology) > MAX_SUPPORTED_CPUS_LEGACY;
+        regs::setup_sregs(&guest_memory.memory(), vcpu, enable_x2_apic_mode)
+            .map_err(Error::SregsConfiguration)?;
     }
     interrupts::set_lint(vcpu).map_err(|e| Error::LocalIntConfiguration(e.into()))?;
     Ok(())
@@ -946,14 +954,13 @@ pub fn configure_system(
     cmdline_addr: GuestAddress,
     cmdline_size: usize,
     initramfs: &Option<InitramfsConfig>,
-    _num_cpus: u8,
+    _num_cpus: u32,
     setup_header: Option<setup_header>,
     rsdp_addr: Option<GuestAddress>,
-    sgx_epc_region: Option<SgxEpcRegion>,
     serial_number: Option<&str>,
     uuid: Option<&str>,
     oem_strings: Option<&[&str]>,
-    topology: Option<(u8, u8, u8)>,
+    topology: Option<(u16, u16, u16, u16)>,
 ) -> super::Result<()> {
     // Write EBDA address to location where ACPICA expects to find it
     guest_mem
@@ -969,10 +976,10 @@ pub fn configure_system(
     mptable::setup_mptable(offset, guest_mem, _num_cpus, topology).map_err(Error::MpTableSetup)?;
 
     // Check that the RAM is not smaller than the RSDP start address
-    if let Some(rsdp_addr) = rsdp_addr {
-        if rsdp_addr.0 > guest_mem.last_addr().0 {
-            return Err(super::Error::RsdpPastRamEnd);
-        }
+    if let Some(rsdp_addr) = rsdp_addr
+        && rsdp_addr.0 > guest_mem.last_addr().0
+    {
+        return Err(super::Error::RsdpPastRamEnd);
     }
 
     match setup_header {
@@ -983,15 +990,8 @@ pub fn configure_system(
             initramfs,
             hdr,
             rsdp_addr,
-            sgx_epc_region,
-        ),
-        None => configure_pvh(
-            guest_mem,
-            cmdline_addr,
-            initramfs,
-            rsdp_addr,
-            sgx_epc_region,
         ),
+        None => configure_pvh(guest_mem, cmdline_addr, initramfs, rsdp_addr),
     }
 }
 
@@ -1048,17 +1048,15 @@ pub fn generate_ram_ranges(guest_mem: &GuestMemoryMmap) -> super::Result<Vec<Ram
             && (first_region_end <= &mem_32bit_reserved_start))
         {
             error!(
-                "Unexpected first memory region layout: (start: 0x{:08x}, end: 0x{:08x}).
-                high_ram_start: 0x{:08x}, mem_32bit_reserved_start: 0x{:08x}",
-                first_region_start, first_region_end, high_ram_start, mem_32bit_reserved_start
+                "Unexpected first memory region layout: (start: 0x{first_region_start:08x}, end: 0x{first_region_end:08x}).
+                high_ram_start: 0x{high_ram_start:08x}, mem_32bit_reserved_start: 0x{mem_32bit_reserved_start:08x}"
             );
 
             return Err(super::Error::MemmapTableSetup);
         }
 
         info!(
-            "first usable physical memory range, start: 0x{:08x}, end: 0x{:08x}",
-            high_ram_start, first_region_end
+            "first usable physical memory range, start: 0x{high_ram_start:08x}, end: 0x{first_region_end:08x}"
         );
 
         (high_ram_start, *first_region_end)
@@ -1083,7 +1081,6 @@ fn configure_pvh(
     cmdline_addr: GuestAddress,
     initramfs: &Option<InitramfsConfig>,
     rsdp_addr: Option<GuestAddress>,
-    sgx_epc_region: Option<SgxEpcRegion>,
 ) -> super::Result<()> {
     const XEN_HVM_START_MAGIC_VALUE: u32 = 0x336ec578;
 
@@ -1149,15 +1146,6 @@ fn configure_pvh(
         E820_RESERVED,
     );
 
-    if let Some(sgx_epc_region) = sgx_epc_region {
-        add_memmap_entry(
-            &mut memmap,
-            sgx_epc_region.start().raw_value(),
-            sgx_epc_region.size(),
-            E820_RESERVED,
-        );
-    }
-
     start_info.memmap_entries = memmap.len() as u32;
 
     // Copy the vector with the memmap table to the MEMMAP_START address
@@ -1204,7 +1192,6 @@ fn configure_32bit_entry(
     initramfs: &Option<InitramfsConfig>,
     setup_hdr: setup_header,
     rsdp_addr: Option<GuestAddress>,
-    sgx_epc_region: Option<SgxEpcRegion>,
 ) -> super::Result<()> {
     const KERNEL_LOADER_OTHER: u8 = 0xff;
 
@@ -1260,15 +1247,6 @@ fn configure_32bit_entry(
         E820_RESERVED,
     )?;
 
-    if let Some(sgx_epc_region) = sgx_epc_region {
-        add_e820_entry(
-            &mut params,
-            sgx_epc_region.start().raw_value(),
-            sgx_epc_region.size(),
-            E820_RESERVED,
-        )?;
-    }
-
     if let Some(rsdp_addr) = rsdp_addr {
         params.acpi_rsdp_addr = rsdp_addr.0;
     }
@@ -1333,8 +1311,9 @@ pub fn initramfs_load_addr(
     Ok(aligned_addr)
 }
 
-pub fn get_host_cpu_phys_bits(hypervisor: &Arc<dyn hypervisor::Hypervisor>) -> u8 {
+pub fn get_host_cpu_phys_bits(hypervisor: &dyn hypervisor::Hypervisor) -> u8 {
     // SAFETY: call cpuid with valid leaves
+    #[allow(unused_unsafe)]
     unsafe {
         let leaf = x86_64::__cpuid(0x8000_0000);
 
@@ -1361,30 +1340,35 @@ pub fn get_host_cpu_phys_bits(hypervisor: &Arc<dyn hypervisor::Hypervisor>) -> u
 
 fn update_cpuid_topology(
     cpuid: &mut Vec<CpuIdEntry>,
-    threads_per_core: u8,
-    cores_per_die: u8,
-    dies_per_package: u8,
+    threads_per_core: u16,
+    cores_per_die: u16,
+    dies_per_package: u16,
+    packages: u16,
     cpu_vendor: CpuVendor,
-    id: u8,
+    id: u32,
 ) {
     let x2apic_id = get_x2apic_id(
-        id as u32,
-        Some((threads_per_core, cores_per_die, dies_per_package)),
+        id,
+        Some((threads_per_core, cores_per_die, dies_per_package, packages)),
     );
 
-    let thread_width = 8 - (threads_per_core - 1).leading_zeros();
-    let core_width = (8 - (cores_per_die - 1).leading_zeros()) + thread_width;
-    let die_width = (8 - (dies_per_package - 1).leading_zeros()) + core_width;
+    // Note: the topology defined here is per "package" (~NUMA node).
+    let thread_width = u16::BITS - (threads_per_core - 1).leading_zeros();
+    let core_width = u16::BITS - (cores_per_die - 1).leading_zeros() + thread_width;
+    let die_width = u16::BITS - (dies_per_package - 1).leading_zeros() + core_width;
 
+    // The very old way: a flat number of logical CPUs per package: CPUID.1H:EBX[23:16] bits.
+    let core_count = dies_per_package as u32 * cores_per_die as u32 * threads_per_core as u32;
     let mut cpu_ebx = CpuidPatch::get_cpuid_reg(cpuid, 0x1, None, CpuidReg::EBX).unwrap_or(0);
-    cpu_ebx |= ((dies_per_package as u32) * (cores_per_die as u32) * (threads_per_core as u32))
-        & (0xff << 16);
+    cpu_ebx &= !(0xff << 16);
+    cpu_ebx |= (core_count & 0xff) << 16;
     CpuidPatch::set_cpuid_reg(cpuid, 0x1, None, CpuidReg::EBX, cpu_ebx);
 
     let mut cpu_edx = CpuidPatch::get_cpuid_reg(cpuid, 0x1, None, CpuidReg::EDX).unwrap_or(0);
     cpu_edx |= 1 << 28;
     CpuidPatch::set_cpuid_reg(cpuid, 0x1, None, CpuidReg::EDX, cpu_edx);
 
+    // The legacy way: threads+cores per package.
     // CPU Topology leaf 0xb
     CpuidPatch::set_cpuid_reg(cpuid, 0xb, Some(0), CpuidReg::EAX, thread_width);
     CpuidPatch::set_cpuid_reg(
@@ -1407,6 +1391,7 @@ fn update_cpuid_topology(
     CpuidPatch::set_cpuid_reg(cpuid, 0xb, Some(1), CpuidReg::ECX, 2 << 8);
     CpuidPatch::set_cpuid_reg(cpuid, 0xb, Some(1), CpuidReg::EDX, x2apic_id);
 
+    // The modern way: many-level hierarchy (but we here only support four levels).
     // CPU Topology leaf 0x1f
     CpuidPatch::set_cpuid_reg(cpuid, 0x1f, Some(0), CpuidReg::EAX, thread_width);
     CpuidPatch::set_cpuid_reg(
@@ -1417,6 +1402,7 @@ fn update_cpuid_topology(
         u32::from(threads_per_core),
     );
     CpuidPatch::set_cpuid_reg(cpuid, 0x1f, Some(0), CpuidReg::ECX, 1 << 8);
+    CpuidPatch::set_cpuid_reg(cpuid, 0x1f, Some(0), CpuidReg::EDX, x2apic_id);
 
     CpuidPatch::set_cpuid_reg(cpuid, 0x1f, Some(1), CpuidReg::EAX, core_width);
     CpuidPatch::set_cpuid_reg(
@@ -1427,6 +1413,7 @@ fn update_cpuid_topology(
         u32::from(cores_per_die * threads_per_core),
     );
     CpuidPatch::set_cpuid_reg(cpuid, 0x1f, Some(1), CpuidReg::ECX, 2 << 8);
+    CpuidPatch::set_cpuid_reg(cpuid, 0x1f, Some(1), CpuidReg::EDX, x2apic_id);
 
     CpuidPatch::set_cpuid_reg(cpuid, 0x1f, Some(2), CpuidReg::EAX, die_width);
     CpuidPatch::set_cpuid_reg(
@@ -1437,6 +1424,7 @@ fn update_cpuid_topology(
         u32::from(dies_per_package * cores_per_die * threads_per_core),
     );
     CpuidPatch::set_cpuid_reg(cpuid, 0x1f, Some(2), CpuidReg::ECX, 5 << 8);
+    CpuidPatch::set_cpuid_reg(cpuid, 0x1f, Some(2), CpuidReg::EDX, x2apic_id);
 
     if matches!(cpu_vendor, CpuVendor::AMD) {
         CpuidPatch::set_cpuid_reg(
@@ -1483,7 +1471,7 @@ fn update_cpuid_topology(
                     edx_bit: Some(28),
                 },
             ];
-            CpuidPatch::patch_cpuid(cpuid, cpuid_patches);
+            CpuidPatch::patch_cpuid(cpuid, &cpuid_patches);
             CpuidPatch::set_cpuid_reg(
                 cpuid,
                 0x8000_0008,
@@ -1497,59 +1485,8 @@ fn update_cpuid_topology(
         }
     }
 }
-
-// The goal is to update the CPUID sub-leaves to reflect the number of EPC
-// sections exposed to the guest.
-fn update_cpuid_sgx(
-    cpuid: &mut Vec<CpuIdEntry>,
-    epc_sections: &[SgxEpcSection],
-) -> Result<(), Error> {
-    // Something's wrong if there's no EPC section.
-    if epc_sections.is_empty() {
-        return Err(Error::NoSgxEpcSection);
-    }
-    // We can't go further if the hypervisor does not support SGX feature.
-    if !CpuidPatch::is_feature_enabled(cpuid, 0x7, 0, CpuidReg::EBX, 2) {
-        return Err(Error::MissingSgxFeature);
-    }
-    // We can't go further if the hypervisor does not support SGX_LC feature.
-    if !CpuidPatch::is_feature_enabled(cpuid, 0x7, 0, CpuidReg::ECX, 30) {
-        return Err(Error::MissingSgxLaunchControlFeature);
-    }
-
-    // Get host CPUID for leaf 0x12, subleaf 0x2. This is to retrieve EPC
-    // properties such as confidentiality and integrity.
-    // SAFETY: call cpuid with valid leaves
-    let leaf = unsafe { std::arch::x86_64::__cpuid_count(0x12, 0x2) };
-
-    for (i, epc_section) in epc_sections.iter().enumerate() {
-        let subleaf_idx = i + 2;
-        let start = epc_section.start().raw_value();
-        let size = epc_section.size();
-        let eax = (start & 0xffff_f000) as u32 | 0x1;
-        let ebx = (start >> 32) as u32;
-        let ecx = (size & 0xffff_f000) as u32 | (leaf.ecx & 0xf);
-        let edx = (size >> 32) as u32;
-        // CPU Topology leaf 0x12
-        CpuidPatch::set_cpuid_reg(cpuid, 0x12, Some(subleaf_idx as u32), CpuidReg::EAX, eax);
-        CpuidPatch::set_cpuid_reg(cpuid, 0x12, Some(subleaf_idx as u32), CpuidReg::EBX, ebx);
-        CpuidPatch::set_cpuid_reg(cpuid, 0x12, Some(subleaf_idx as u32), CpuidReg::ECX, ecx);
-        CpuidPatch::set_cpuid_reg(cpuid, 0x12, Some(subleaf_idx as u32), CpuidReg::EDX, edx);
-    }
-
-    // Add one NULL entry to terminate the dynamic list
-    let subleaf_idx = epc_sections.len() + 2;
-    // CPU Topology leaf 0x12
-    CpuidPatch::set_cpuid_reg(cpuid, 0x12, Some(subleaf_idx as u32), CpuidReg::EAX, 0);
-    CpuidPatch::set_cpuid_reg(cpuid, 0x12, Some(subleaf_idx as u32), CpuidReg::EBX, 0);
-    CpuidPatch::set_cpuid_reg(cpuid, 0x12, Some(subleaf_idx as u32), CpuidReg::ECX, 0);
-    CpuidPatch::set_cpuid_reg(cpuid, 0x12, Some(subleaf_idx as u32), CpuidReg::EDX, 0);
-
-    Ok(())
-}
-
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use linux_loader::loader::bootparam::boot_e820_entry;
 
     use super::*;
@@ -1578,7 +1515,6 @@ mod tests {
             None,
             None,
             None,
-            None,
         );
         config_err.unwrap_err();
 
@@ -1603,7 +1539,6 @@ mod tests {
             None,
             None,
             None,
-            None,
         )
         .unwrap();
 
@@ -1633,7 +1568,6 @@ mod tests {
             None,
             None,
             None,
-            None,
         )
         .unwrap();
 
@@ -1649,7 +1583,6 @@ mod tests {
             None,
             None,
             None,
-            None,
         )
         .unwrap();
     }
@@ -1721,22 +1654,27 @@ mod tests {
 
     #[test]
     fn test_get_x2apic_id() {
-        let x2apic_id = get_x2apic_id(0, Some((2, 3, 1)));
+        let x2apic_id = get_x2apic_id(0, Some((2, 3, 1, 1)));
         assert_eq!(x2apic_id, 0);
 
-        let x2apic_id = get_x2apic_id(1, Some((2, 3, 1)));
+        let x2apic_id = get_x2apic_id(1, Some((2, 3, 1, 1)));
         assert_eq!(x2apic_id, 1);
 
-        let x2apic_id = get_x2apic_id(2, Some((2, 3, 1)));
+        let x2apic_id = get_x2apic_id(2, Some((2, 3, 1, 1)));
         assert_eq!(x2apic_id, 2);
 
-        let x2apic_id = get_x2apic_id(6, Some((2, 3, 1)));
+        let x2apic_id = get_x2apic_id(6, Some((2, 3, 1, 1)));
         assert_eq!(x2apic_id, 8);
 
-        let x2apic_id = get_x2apic_id(7, Some((2, 3, 1)));
+        let x2apic_id = get_x2apic_id(7, Some((2, 3, 1, 1)));
         assert_eq!(x2apic_id, 9);
 
-        let x2apic_id = get_x2apic_id(8, Some((2, 3, 1)));
+        let x2apic_id = get_x2apic_id(8, Some((2, 3, 1, 1)));
         assert_eq!(x2apic_id, 10);
+
+        let x2apic_id = get_x2apic_id(257, Some((1, 312, 1, 1)));
+        assert_eq!(x2apic_id, 257);
+
+        assert_eq!(255, get_max_x2apic_id((1, 256, 1, 1)));
     }
 }

arch/src/x86_64/mptable.rs

@@ -8,12 +8,14 @@
 use std::{mem, result, slice};
 
 use libc::c_uchar;
+use log::{info, warn};
 use thiserror::Error;
 use vm_memory::{Address, ByteValued, Bytes, GuestAddress, GuestMemory, GuestMemoryError};
 
+use super::MAX_SUPPORTED_CPUS_LEGACY;
+use crate::GuestMemoryMmap;
 use crate::layout::{APIC_START, HIGH_RAM_START, IOAPIC_START};
 use crate::x86_64::{get_x2apic_id, mpspec};
-use crate::GuestMemoryMmap;
 
 // This is a workaround to the Rust enforcement specifying that any implementation of a foreign
 // trait (in this case `ByteValued`) where:
@@ -61,9 +63,6 @@ pub enum Error {
     /// Failure while zeroing out the memory for the MP table.
     #[error("Failure while zeroing out the memory for the MP table")]
     Clear(#[source] GuestMemoryError),
-    /// Number of CPUs exceeds the maximum supported CPUs
-    #[error("Number of CPUs exceeds the maximum supported CPUs")]
-    TooManyCpus,
     /// Failure to write the MP floating pointer.
     #[error("Failure to write the MP floating pointer")]
     WriteMpfIntel(#[source] GuestMemoryError),
@@ -89,11 +88,6 @@ pub enum Error {
 
 pub type Result<T> = result::Result<T, Error>;
 
-// With APIC/xAPIC, there are only 255 APIC IDs available. And IOAPIC occupies
-// one APIC ID, so only 254 CPUs at maximum may be supported. Actually it's
-// a large number for FC usecases.
-pub const MAX_SUPPORTED_CPUS: u32 = 254;
-
 // Most of these variables are sourced from the Intel MP Spec 1.4.
 const SMP_MAGIC_IDENT: &[c_uchar; 4] = b"_MP_";
 const MPC_SIGNATURE: &[c_uchar; 4] = b"PCMP";
@@ -121,7 +115,7 @@ fn mpf_intel_compute_checksum(v: &mpspec::mpf_intel) -> u8 {
     (!checksum).wrapping_add(1)
 }
 
-fn compute_mp_size(num_cpus: u8) -> usize {
+fn compute_mp_size(num_cpus: u32) -> usize {
     mem::size_of::<MpfIntelWrapper>()
         + mem::size_of::<MpcTableWrapper>()
         + mem::size_of::<MpcCpuWrapper>() * (num_cpus as usize)
@@ -135,14 +129,15 @@ fn compute_mp_size(num_cpus: u8) -> usize {
 pub fn setup_mptable(
     offset: GuestAddress,
     mem: &GuestMemoryMmap,
-    num_cpus: u8,
-    topology: Option<(u8, u8, u8)>,
+    num_cpus: u32,
+    topology: Option<(u16, u16, u16, u16)>,
 ) -> Result<()> {
     if num_cpus > 0 {
         let cpu_id_max = num_cpus - 1;
-        let x2apic_id_max = get_x2apic_id(cpu_id_max.into(), topology);
-        if x2apic_id_max >= MAX_SUPPORTED_CPUS {
-            return Err(Error::TooManyCpus);
+        let x2apic_id_max = get_x2apic_id(cpu_id_max, topology);
+        if x2apic_id_max >= MAX_SUPPORTED_CPUS_LEGACY {
+            info!("Skipping mptable creation due to too many CPUs");
+            return Ok(());
         }
     }
 
@@ -157,7 +152,7 @@ pub fn setup_mptable(
     }
 
     let mut checksum: u8 = 0;
-    let ioapicid: u8 = MAX_SUPPORTED_CPUS as u8 + 1;
+    let ioapicid: u8 = MAX_SUPPORTED_CPUS_LEGACY as u8 + 1;
 
     // The checked_add here ensures the all of the following base_mp.unchecked_add's will be without
     // overflow.
@@ -195,7 +190,7 @@ pub fn setup_mptable(
         for cpu_id in 0..num_cpus {
             let mut mpc_cpu = MpcCpuWrapper(mpspec::mpc_cpu::default());
             mpc_cpu.0.type_ = mpspec::MP_PROCESSOR as u8;
-            mpc_cpu.0.apicid = get_x2apic_id(cpu_id as u32, topology) as u8;
+            mpc_cpu.0.apicid = get_x2apic_id(cpu_id, topology) as u8;
             mpc_cpu.0.apicver = APIC_VERSION;
             mpc_cpu.0.cpuflag = mpspec::CPU_ENABLED as u8
                 | if cpu_id == 0 {
@@ -303,7 +298,7 @@ pub fn setup_mptable(
 }
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use vm_memory::bitmap::BitmapSlice;
     use vm_memory::{GuestUsize, VolatileMemoryError, VolatileSlice, WriteVolatile};
 
@@ -394,11 +389,11 @@ mod tests {
     fn cpu_entry_count() {
         let mem = GuestMemoryMmap::from_ranges(&[(
             MPTABLE_START,
-            compute_mp_size(MAX_SUPPORTED_CPUS as u8),
+            compute_mp_size(MAX_SUPPORTED_CPUS_LEGACY),
         )])
         .unwrap();
 
-        for i in 0..MAX_SUPPORTED_CPUS as u8 {
+        for i in 0..MAX_SUPPORTED_CPUS_LEGACY {
             setup_mptable(MPTABLE_START, &mem, i, None).unwrap();
 
             let mpf_intel: MpfIntelWrapper = mem.read_obj(MPTABLE_START).unwrap();
@@ -428,11 +423,9 @@ mod tests {
 
     #[test]
     fn cpu_entry_count_max() {
-        let cpus = MAX_SUPPORTED_CPUS + 1;
-        let mem =
-            GuestMemoryMmap::from_ranges(&[(MPTABLE_START, compute_mp_size(cpus as u8))]).unwrap();
+        let cpus = MAX_SUPPORTED_CPUS_LEGACY + 1;
+        let mem = GuestMemoryMmap::from_ranges(&[(MPTABLE_START, compute_mp_size(cpus))]).unwrap();
 
-        let result = setup_mptable(MPTABLE_START, &mem, cpus as u8, None);
-        result.unwrap_err();
+        setup_mptable(MPTABLE_START, &mem, cpus, None).unwrap();
     }
 }

arch/src/x86_64/regs.rs

@@ -6,7 +6,6 @@
 // Portions Copyright 2017 The Chromium OS Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE-BSD-3-Clause file.
-use std::sync::Arc;
 use std::{mem, result};
 
 use hypervisor::arch::x86::gdt::{gdt_entry, segment_from_gdt};
@@ -67,7 +66,7 @@ pub type Result<T> = result::Result<T, Error>;
 /// # Arguments
 ///
 /// * `vcpu` - Structure for the VCPU that holds the VCPU's fd.
-pub fn setup_fpu(vcpu: &Arc<dyn hypervisor::Vcpu>) -> Result<()> {
+pub fn setup_fpu(vcpu: &dyn hypervisor::Vcpu) -> Result<()> {
     let fpu: FpuState = FpuState {
         fcw: 0x37f,
         mxcsr: 0x1f80,
@@ -82,8 +81,8 @@ pub fn setup_fpu(vcpu: &Arc<dyn hypervisor::Vcpu>) -> Result<()> {
 /// # Arguments
 ///
 /// * `vcpu` - Structure for the VCPU that holds the VCPU's fd.
-pub fn setup_msrs(vcpu: &Arc<dyn hypervisor::Vcpu>) -> Result<()> {
-    vcpu.set_msrs(&vcpu.boot_msr_entries())
+pub fn setup_msrs(vcpu: &dyn hypervisor::Vcpu) -> Result<()> {
+    vcpu.set_msrs(vcpu.boot_msr_entries())
         .map_err(Error::SetModelSpecificRegisters)?;
 
     Ok(())
@@ -95,7 +94,7 @@ pub fn setup_msrs(vcpu: &Arc<dyn hypervisor::Vcpu>) -> Result<()> {
 ///
 /// * `vcpu` - Structure for the VCPU that holds the VCPU's fd.
 /// * `entry_point` - Description of the boot entry to set up.
-pub fn setup_regs(vcpu: &Arc<dyn hypervisor::Vcpu>, entry_point: EntryPoint) -> Result<()> {
+pub fn setup_regs(vcpu: &dyn hypervisor::Vcpu, entry_point: EntryPoint) -> Result<()> {
     let mut regs = vcpu.create_standard_regs();
     match entry_point.setup_header {
         None => {
@@ -109,7 +108,7 @@ pub fn setup_regs(vcpu: &Arc<dyn hypervisor::Vcpu>, entry_point: EntryPoint) ->
             regs.set_rsp(BOOT_STACK_POINTER.raw_value());
             regs.set_rsi(ZERO_PAGE_START.raw_value());
         }
-    };
+    }
     vcpu.set_regs(&regs).map_err(Error::SetBaseRegisters)
 }
 
@@ -119,9 +118,13 @@ pub fn setup_regs(vcpu: &Arc<dyn hypervisor::Vcpu>, entry_point: EntryPoint) ->
 ///
 /// * `mem` - The memory that will be passed to the guest.
 /// * `vcpu` - Structure for the VCPU that holds the VCPU's fd.
-pub fn setup_sregs(mem: &GuestMemoryMmap, vcpu: &Arc<dyn hypervisor::Vcpu>) -> Result<()> {
+pub fn setup_sregs(
+    mem: &GuestMemoryMmap,
+    vcpu: &dyn hypervisor::Vcpu,
+    enable_x2_apic_mode: bool,
+) -> Result<()> {
     let mut sregs: SpecialRegisters = vcpu.get_sregs().map_err(Error::GetStatusRegisters)?;
-    configure_segments_and_sregs(mem, &mut sregs)?;
+    configure_segments_and_sregs(mem, &mut sregs, enable_x2_apic_mode)?;
     vcpu.set_sregs(&sregs).map_err(Error::SetStatusRegisters)
 }
 
@@ -148,6 +151,7 @@ fn write_idt_value(val: u64, guest_mem: &GuestMemoryMmap) -> Result<()> {
 pub fn configure_segments_and_sregs(
     mem: &GuestMemoryMmap,
     sregs: &mut SpecialRegisters,
+    enable_x2_apic_mode: bool,
 ) -> Result<()> {
     let gdt_table: [u64; BOOT_GDT_MAX] = {
         // Configure GDT entries as specified by PVH boot protocol
@@ -183,11 +187,16 @@ pub fn configure_segments_and_sregs(
     sregs.cr0 = CR0_PE;
     sregs.cr4 = 0;
 
+    if enable_x2_apic_mode {
+        const X2APIC_ENABLE_BIT: u64 = 1 << 10;
+        sregs.apic_base |= X2APIC_ENABLE_BIT;
+    }
+
     Ok(())
 }
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use vm_memory::GuestAddress;
 
     use super::*;
@@ -204,7 +213,7 @@ mod tests {
     fn segments_and_sregs() {
         let mut sregs: SpecialRegisters = Default::default();
         let gm = create_guest_mem();
-        configure_segments_and_sregs(&gm, &mut sregs).unwrap();
+        configure_segments_and_sregs(&gm, &mut sregs, false).unwrap();
         assert_eq!(0x0, read_u64(&gm, BOOT_GDT_START));
         assert_eq!(
             0xcf9b000000ffff,

arch/src/x86_64/smbios.rs

@@ -12,8 +12,8 @@ use thiserror::Error;
 use uuid::Uuid;
 use vm_memory::{Address, ByteValued, Bytes, GuestAddress};
 
-use crate::layout::SMBIOS_START;
 use crate::GuestMemoryMmap;
+use crate::layout::SMBIOS_START;
 
 #[derive(Debug, Error)]
 pub enum Error {
@@ -33,8 +33,8 @@ pub enum Error {
     #[error("Failure to write additional data to memory")]
     WriteData,
     /// Failure to parse uuid, uuid format may be error
-    #[error("Failure to parse uuid")]
-    ParseUuid(#[source] uuid::Error),
+    #[error("Failure to parse uuid: {1}")]
+    ParseUuid(#[source] uuid::Error, String),
 }
 
 pub type Result<T> = result::Result<T, Error>;
@@ -198,7 +198,7 @@ pub fn setup_smbios(
         let uuid_number = uuid
             .map(Uuid::parse_str)
             .transpose()
-            .map_err(Error::ParseUuid)?
+            .map_err(|e| Error::ParseUuid(e, uuid.unwrap().to_string()))?
             .unwrap_or(Uuid::nil());
         let smbios_sysinfo = SmbiosSysInfo {
             r#type: SYSTEM_INFORMATION,
@@ -272,7 +272,7 @@ pub fn setup_smbios(
 }
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use super::*;
 
     #[test]

arch/src/x86_64/tdx/mod.rs

@@ -5,6 +5,7 @@ use std::fs::File;
 use std::io::{Read, Seek, SeekFrom};
 use std::str::FromStr;
 
+use log::{debug, info};
 use thiserror::Error;
 use uuid::Uuid;
 use vm_memory::{ByteValued, Bytes, GuestAddress, GuestMemoryError};
@@ -108,7 +109,7 @@ fn tdvf_descriptor_offset(file: &mut File) -> Result<(SeekFrom, bool), TdvfError
                 u16::from_le_bytes(table[offset - 18..offset - 16].try_into().unwrap()) as usize;
             debug!(
                 "Entry GUID = {}, size = {}",
-                entry_uuid.hyphenated().to_string(),
+                entry_uuid.hyphenated(),
                 entry_size
             );
 
@@ -304,7 +305,7 @@ fn align_hob(v: u64) -> u64 {
 
 impl TdHob {
     fn update_offset<T>(&mut self) {
-        self.current_offset = align_hob(self.current_offset + std::mem::size_of::<T>() as u64)
+        self.current_offset = align_hob(self.current_offset + std::mem::size_of::<T>() as u64);
     }
 
     pub fn start(offset: u64) -> TdHob {
@@ -519,7 +520,7 @@ impl TdHob {
 }
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use super::*;
 
     #[test]
@@ -528,7 +529,7 @@ mod tests {
         let mut f = std::fs::File::open("tdvf.fd").unwrap();
         let (sections, _) = parse_tdvf_sections(&mut f).unwrap();
         for section in sections {
-            eprintln!("{section:x?}")
+            eprintln!("{section:x?}");
         }
     }
 }

block/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 authors = ["The Chromium OS Authors", "The Cloud Hypervisor Authors"]
-edition = "2021"
+edition.workspace = true
 name = "block"
 version = "0.1.0"
 
@@ -9,14 +9,16 @@ default = []
 io_uring = ["dep:io-uring"]
 
 [dependencies]
-byteorder = "1.5.0"
+bitflags = { workspace = true }
+byteorder = { workspace = true }
 crc-any = "2.5.0"
-io-uring = { version = "0.6.4", optional = true }
-libc = "0.2.167"
-log = "0.4.22"
+flate2 = "1.1"
+io-uring = { version = "0.7.11", optional = true }
+libc = { workspace = true }
+log = { workspace = true }
 remain = "0.2.15"
-serde = { version = "1.0.208", features = ["derive"] }
-smallvec = "1.13.2"
+serde = { workspace = true, features = ["derive"] }
+smallvec = "1.15.1"
 thiserror = { workspace = true }
 uuid = { workspace = true, features = ["v4"] }
 virtio-bindings = { workspace = true }
@@ -28,3 +30,7 @@ vm-memory = { workspace = true, features = [
 ] }
 vm-virtio = { path = "../vm-virtio" }
 vmm-sys-util = { workspace = true }
+zstd = "0.13"
+
+[lints]
+workspace = true

block/src/async_io.rs

@@ -8,7 +8,7 @@ use std::os::fd::{AsRawFd, OwnedFd, RawFd};
 use thiserror::Error;
 use vmm_sys_util::eventfd::EventFd;
 
-use crate::DiskTopology;
+use crate::{BatchRequest, DiskTopology};
 
 #[derive(Error, Debug)]
 pub enum DiskFileError {
@@ -18,6 +18,12 @@ pub enum DiskFileError {
     /// Failed creating a new AsyncIo.
     #[error("Failed creating a new AsyncIo")]
     NewAsyncIo(#[source] std::io::Error),
+    /// Unsupported operation.
+    #[error("Unsupported operation")]
+    Unsupported,
+    /// Resize failed
+    #[error("Resize failed")]
+    ResizeError(#[source] std::io::Error),
 }
 
 pub type DiskFileResult<T> = std::result::Result<T, DiskFileError>;
@@ -56,11 +62,34 @@ impl AsRawFd for BorrowedDiskFd<'_> {
 /// This allows abstracting over raw image formats as well as structured
 /// image formats.
 pub trait DiskFile: Send {
-    fn size(&mut self) -> DiskFileResult<u64>;
+    /// Returns the logical disk size a guest will see.
+    ///
+    /// For raw formats, this is equal to [`Self::physical_size`]. For file formats
+    /// that wrap disk images in a container (e.g. QCOW2), this refers to the
+    /// effective size that the guest will see.
+    fn logical_size(&mut self) -> DiskFileResult<u64>;
+    /// Returns the physical size of the underlying file.
+    fn physical_size(&mut self) -> DiskFileResult<u64>;
     fn new_async_io(&self, ring_depth: u32) -> DiskFileResult<Box<dyn AsyncIo>>;
     fn topology(&mut self) -> DiskTopology {
         DiskTopology::default()
     }
+    fn resize(&mut self, _size: u64) -> DiskFileResult<()> {
+        Err(DiskFileError::Unsupported)
+    }
+
+    /// Indicates support for sparse operations (punch hole, write zeroes, discard).
+    /// Override to return true when supported.
+    fn supports_sparse_operations(&self) -> bool {
+        false
+    }
+
+    /// Indicates support for zero flag optimization in WRITE_ZEROES. Override
+    /// to return true when supported.
+    fn supports_zero_flag(&self) -> bool {
+        false
+    }
+
     /// Returns the file descriptor of the underlying disk image file.
     ///
     /// The file descriptor is supposed to be used for `fcntl()` calls but no
@@ -79,6 +108,15 @@ pub enum AsyncIoError {
     /// Failed synchronizing file.
     #[error("Failed synchronizing file")]
     Fsync(#[source] std::io::Error),
+    /// Failed punching hole.
+    #[error("Failed punching hole")]
+    PunchHole(#[source] std::io::Error),
+    /// Failed writing zeroes.
+    #[error("Failed writing zeroes")]
+    WriteZeroes(#[source] std::io::Error),
+    /// Failed submitting batch requests.
+    #[error("Failed submitting batch requests")]
+    SubmitBatchRequests(#[source] std::io::Error),
 }
 
 pub type AsyncIoResult<T> = std::result::Result<T, AsyncIoError>;
@@ -98,5 +136,13 @@ pub trait AsyncIo: Send {
         user_data: u64,
     ) -> AsyncIoResult<()>;
     fn fsync(&mut self, user_data: Option<u64>) -> AsyncIoResult<()>;
+    fn punch_hole(&mut self, offset: u64, length: u64, user_data: u64) -> AsyncIoResult<()>;
+    fn write_zeroes(&mut self, offset: u64, length: u64, user_data: u64) -> AsyncIoResult<()>;
     fn next_completed_request(&mut self) -> Option<(u64, i32)>;
+    fn batch_requests_enabled(&self) -> bool {
+        false
+    }
+    fn submit_batch_requests(&mut self, _batch_request: &[BatchRequest]) -> AsyncIoResult<()> {
+        Ok(())
+    }
 }

block/src/fcntl.rs

@@ -101,13 +101,52 @@ impl LockState {
     }
 }
 
+/// The granularity of the advisory lock.
+///
+/// The granularity has significant implications in typical cloud deployments
+/// with network storage. The Linux kernel will sync advisory locks to network
+/// file systems, but these backends may have different policies and handle
+/// locks differently. For example, Netapp speaks a NFS API but will treat
+/// advisory OFD locks for the whole file as mandatory locks, whereas byte-range
+/// locks for the whole file will remain advisory [0].
+///
+/// As it is a valid use case to prevent multiple CHV instances from accessing
+/// the same disk but disk management software (e.g., Cinder in OpenStack)
+/// should be able to snapshot disks while VMs are running, we need special
+/// control over the lock granularity. Therefore, it is a valid use case to lock
+/// the whole byte range of a disk image without technically locking the whole
+/// file - to get the best of both worlds.
+///
+/// [0] https://kb.netapp.com/on-prem/ontap/da/NAS/NAS-KBs/How_is_Mandatory_Locking_supported_for_NFSv4_on_ONTAP_9
+#[derive(Clone, Copy, Debug)]
+pub enum LockGranularity {
+    WholeFile,
+    ByteRange(u64 /* from, inclusive */, u64 /* len */),
+}
+
+impl LockGranularity {
+    const fn l_start(self) -> u64 {
+        match self {
+            LockGranularity::WholeFile => 0,
+            LockGranularity::ByteRange(start, _) => start,
+        }
+    }
+
+    const fn l_len(self) -> u64 {
+        match self {
+            LockGranularity::WholeFile => 0, /* EOF */
+            LockGranularity::ByteRange(_, len) => len,
+        }
+    }
+}
+
 /// Returns a [`struct@libc::flock`] structure for the whole file.
-const fn get_flock(lock_type: LockType) -> libc::flock {
+const fn get_flock(lock_type: LockType, granularity: LockGranularity) -> libc::flock {
     libc::flock {
         l_type: lock_type.to_libc_val() as libc::c_short,
         l_whence: libc::SEEK_SET as libc::c_short,
-        l_start: 0,
-        l_len: 0, /* EOF */
+        l_start: granularity.l_start() as libc::c_long,
+        l_len: granularity.l_len() as libc::c_long,
         l_pid: 0, /* filled by callee */
     }
 }
@@ -122,8 +161,13 @@ const fn get_flock(lock_type: LockType) -> libc::flock {
 /// - `file`: The file to acquire a lock for [`LockType`]. The file's state will
 ///   be logically mutated, but not technically.
 /// - `lock_type`: The [`LockType`]
-pub fn try_acquire_lock<Fd: AsRawFd>(file: Fd, lock_type: LockType) -> Result<(), LockError> {
-    let flock = get_flock(lock_type);
+/// - `granularity`: The [`LockGranularity`].
+pub fn try_acquire_lock<Fd: AsRawFd>(
+    file: &Fd,
+    lock_type: LockType,
+    granularity: LockGranularity,
+) -> Result<(), LockError> {
+    let flock = get_flock(lock_type, granularity);
 
     let res = fcntl(file.as_raw_fd(), FcntlArg::F_OFD_SETLK(&flock));
     match res {
@@ -146,8 +190,9 @@ pub fn try_acquire_lock<Fd: AsRawFd>(file: Fd, lock_type: LockType) -> Result<()
 ///
 /// # Parameters
 /// - `file`: The file to clear all locks for [`LockType`].
-pub fn clear_lock<Fd: AsRawFd>(file: Fd) -> Result<(), LockError> {
-    try_acquire_lock(file, LockType::Unlock)
+/// - `granularity`: The [`LockGranularity`].
+pub fn clear_lock<Fd: AsRawFd>(file: &Fd, granularity: LockGranularity) -> Result<(), LockError> {
+    try_acquire_lock(file, LockType::Unlock, granularity)
 }
 
 /// Returns the current lock state using [`fcntl`] with respect to the given
@@ -155,8 +200,12 @@ pub fn clear_lock<Fd: AsRawFd>(file: Fd) -> Result<(), LockError> {
 ///
 /// # Parameters
 /// - `file`: The file for which to get the lock state.
-pub fn get_lock_state<Fd: AsRawFd>(file: Fd) -> Result<LockState, LockError> {
-    let mut flock = get_flock(LockType::Write);
+/// - `granularity`: The [`LockGranularity`].
+pub fn get_lock_state<Fd: AsRawFd>(
+    file: &Fd,
+    granularity: LockGranularity,
+) -> Result<LockState, LockError> {
+    let mut flock = get_flock(LockType::Write, granularity);
     let res = fcntl(file.as_raw_fd(), FcntlArg::F_OFD_GETLK(&mut flock));
     match res {
         0 => {

block/src/fixed_vhd.rs

@@ -6,8 +6,8 @@ use std::fs::File;
 use std::io::{Read, Seek, SeekFrom, Write};
 use std::os::unix::io::{AsRawFd, RawFd};
 
-use crate::vhd::VhdFooter;
 use crate::BlockBackend;
+use crate::vhd::VhdFooter;
 
 #[derive(Debug)]
 pub struct FixedVhd {
@@ -75,9 +75,17 @@ impl Seek for FixedVhd {
 }
 
 impl BlockBackend for FixedVhd {
-    fn size(&self) -> std::result::Result<u64, crate::Error> {
+    fn logical_size(&self) -> Result<u64, crate::Error> {
         Ok(self.size)
     }
+
+    /// Returns the physical size of the underlying file.
+    fn physical_size(&self) -> Result<u64, crate::Error> {
+        self.file
+            .metadata()
+            .map(|m| m.len())
+            .map_err(crate::Error::GetFileMetadata)
+    }
 }
 
 impl Clone for FixedVhd {

block/src/fixed_vhd_async.rs

@@ -12,7 +12,7 @@ use crate::async_io::{
 };
 use crate::fixed_vhd::FixedVhd;
 use crate::raw_async::RawFileAsync;
-use crate::BlockBackend;
+use crate::{BatchRequest, BlockBackend};
 
 pub struct FixedVhdDiskAsync(FixedVhd);
 
@@ -23,14 +23,22 @@ impl FixedVhdDiskAsync {
 }
 
 impl DiskFile for FixedVhdDiskAsync {
-    fn size(&mut self) -> DiskFileResult<u64> {
-        Ok(self.0.size().unwrap())
+    fn logical_size(&mut self) -> DiskFileResult<u64> {
+        Ok(self.0.logical_size().unwrap())
+    }
+
+    fn physical_size(&mut self) -> DiskFileResult<u64> {
+        Ok(self.0.physical_size().unwrap())
     }
 
     fn new_async_io(&self, ring_depth: u32) -> DiskFileResult<Box<dyn AsyncIo>> {
         Ok(Box::new(
-            FixedVhdAsync::new(self.0.as_raw_fd(), ring_depth, self.0.size().unwrap())
-                .map_err(DiskFileError::NewAsyncIo)?,
+            FixedVhdAsync::new(
+                self.0.as_raw_fd(),
+                ring_depth,
+                self.0.logical_size().unwrap(),
+            )
+            .map_err(DiskFileError::NewAsyncIo)?,
         ) as Box<dyn AsyncIo>)
     }
 
@@ -106,4 +114,24 @@ impl AsyncIo for FixedVhdAsync {
     fn next_completed_request(&mut self) -> Option<(u64, i32)> {
         self.raw_file_async.next_completed_request()
     }
+
+    fn punch_hole(&mut self, _offset: u64, _length: u64, _user_data: u64) -> AsyncIoResult<()> {
+        Err(AsyncIoError::PunchHole(std::io::Error::other(
+            "punch_hole not supported for fixed VHD",
+        )))
+    }
+
+    fn write_zeroes(&mut self, _offset: u64, _length: u64, _user_data: u64) -> AsyncIoResult<()> {
+        Err(AsyncIoError::WriteZeroes(std::io::Error::other(
+            "write_zeroes not supported for fixed VHD",
+        )))
+    }
+
+    fn batch_requests_enabled(&self) -> bool {
+        true
+    }
+
+    fn submit_batch_requests(&mut self, batch_request: &[BatchRequest]) -> AsyncIoResult<()> {
+        self.raw_file_async.submit_batch_requests(batch_request)
+    }
 }

block/src/fixed_vhd_sync.rs

@@ -7,12 +7,12 @@ use std::os::unix::io::{AsRawFd, RawFd};
 
 use vmm_sys_util::eventfd::EventFd;
 
+use crate::BlockBackend;
 use crate::async_io::{
     AsyncIo, AsyncIoError, AsyncIoResult, BorrowedDiskFd, DiskFile, DiskFileError, DiskFileResult,
 };
 use crate::fixed_vhd::FixedVhd;
 use crate::raw_sync::RawFileSync;
-use crate::BlockBackend;
 
 pub struct FixedVhdDiskSync(FixedVhd);
 
@@ -23,13 +23,23 @@ impl FixedVhdDiskSync {
 }
 
 impl DiskFile for FixedVhdDiskSync {
-    fn size(&mut self) -> DiskFileResult<u64> {
-        Ok(self.0.size().unwrap())
+    fn logical_size(&mut self) -> DiskFileResult<u64> {
+        Ok(self.0.logical_size().unwrap())
+    }
+
+    fn physical_size(&mut self) -> DiskFileResult<u64> {
+        self.0.physical_size().map_err(|e| {
+            let io_inner = match e {
+                crate::Error::GetFileMetadata(e) => e,
+                _ => unreachable!(),
+            };
+            DiskFileError::Size(io_inner)
+        })
     }
 
     fn new_async_io(&self, _ring_depth: u32) -> DiskFileResult<Box<dyn AsyncIo>> {
         Ok(Box::new(
-            FixedVhdSync::new(self.0.as_raw_fd(), self.0.size().unwrap())
+            FixedVhdSync::new(self.0.as_raw_fd(), self.0.logical_size().unwrap())
                 .map_err(DiskFileError::NewAsyncIo)?,
         ) as Box<dyn AsyncIo>)
     }
@@ -103,4 +113,16 @@ impl AsyncIo for FixedVhdSync {
     fn next_completed_request(&mut self) -> Option<(u64, i32)> {
         self.raw_file_sync.next_completed_request()
     }
+
+    fn punch_hole(&mut self, _offset: u64, _length: u64, _user_data: u64) -> AsyncIoResult<()> {
+        Err(AsyncIoError::PunchHole(std::io::Error::other(
+            "punch_hole not supported for fixed VHD",
+        )))
+    }
+
+    fn write_zeroes(&mut self, _offset: u64, _length: u64, _user_data: u64) -> AsyncIoResult<()> {
+        Err(AsyncIoError::WriteZeroes(std::io::Error::other(
+            "write_zeroes not supported for fixed VHD",
+        )))
+    }
 }

block/src/lib.rs

@@ -8,9 +8,6 @@
 //
 // SPDX-License-Identifier: Apache-2.0 AND BSD-3-Clause
 
-#[macro_use]
-extern crate log;
-
 pub mod async_io;
 pub mod fcntl;
 pub mod fixed_vhd;
@@ -31,21 +28,22 @@ pub mod vhd;
 pub mod vhdx;
 pub mod vhdx_sync;
 
-use std::alloc::{alloc_zeroed, dealloc, Layout};
+use std::alloc::{Layout, alloc_zeroed, dealloc};
 use std::collections::VecDeque;
-use std::fmt::Debug;
+use std::fmt::{self, Debug};
 use std::fs::File;
 use std::io::{self, IoSlice, IoSliceMut, Read, Seek, SeekFrom, Write};
 use std::os::linux::fs::MetadataExt;
 use std::os::unix::io::AsRawFd;
 use std::path::Path;
-use std::sync::{Arc, MutexGuard};
+use std::str::FromStr;
 use std::time::Instant;
 use std::{cmp, result};
 
 #[cfg(feature = "io_uring")]
-use io_uring::{opcode, IoUring, Probe};
-use libc::{ioctl, S_IFBLK, S_IFMT};
+use io_uring::{IoUring, Probe, opcode};
+use libc::{S_IFBLK, S_IFMT, ioctl};
+use log::{debug, error, info, warn};
 use serde::{Deserialize, Serialize};
 use smallvec::SmallVec;
 use thiserror::Error;
@@ -53,7 +51,7 @@ use virtio_bindings::virtio_blk::*;
 use virtio_queue::DescriptorChain;
 use vm_memory::bitmap::Bitmap;
 use vm_memory::{
-    ByteValued, Bytes, GuestAddress, GuestMemory, GuestMemoryError, GuestMemoryLoadGuard,
+    Address, ByteValued, Bytes, GuestAddress, GuestMemory, GuestMemoryError, GuestMemoryLoadGuard,
 };
 use vm_virtio::{AccessPlatform, Translatable};
 use vmm_sys_util::eventfd::EventFd;
@@ -83,8 +81,8 @@ pub enum Error {
     DetectImageType(#[source] std::io::Error),
     #[error("Failure in fixed vhd")]
     FixedVhdError(#[source] std::io::Error),
-    #[error("Getting a block's metadata fails for any reason")]
-    GetFileMetadata,
+    #[error("Getting a block's metadata failed")]
+    GetFileMetadata(#[source] std::io::Error),
     #[error("The requested operation would cause a seek beyond disk end")]
     InvalidOffset,
     #[error("Failure in qcow")]
@@ -99,7 +97,7 @@ pub enum Error {
 
 fn build_device_id(disk_path: &Path) -> result::Result<String, Error> {
     let blk_metadata = match disk_path.metadata() {
-        Err(_) => return Err(Error::GetFileMetadata),
+        Err(e) => return Err(Error::GetFileMetadata(e)),
         Ok(m) => m,
     };
     // This is how kvmtool does it.
@@ -123,7 +121,7 @@ pub fn build_serial(disk_path: &Path) -> Vec<u8> {
             // This will also zero out any leftover bytes.
             let disk_id = m.as_bytes();
             let bytes_to_copy = cmp::min(disk_id.len(), VIRTIO_BLK_ID_BYTES as usize);
-            default_serial[..bytes_to_copy].clone_from_slice(&disk_id[..bytes_to_copy])
+            default_serial[..bytes_to_copy].clone_from_slice(&disk_id[..bytes_to_copy]);
         }
     }
     default_serial
@@ -139,6 +137,8 @@ pub enum ExecuteError {
     Read(#[source] GuestMemoryError),
     #[error("Failed to read_exact")]
     ReadExact(#[source] io::Error),
+    #[error("Can't execute an operation other than `read` or `get_id` on a read-only device")]
+    ReadOnly,
     #[error("Failed to seek")]
     Seek(#[source] io::Error),
     #[error("Failed to write")]
@@ -157,6 +157,10 @@ pub enum ExecuteError {
     AsyncWrite(#[source] AsyncIoError),
     #[error("failed to async flush")]
     AsyncFlush(#[source] AsyncIoError),
+    #[error("Failed to async punch hole")]
+    AsyncPunchHole(#[source] AsyncIoError),
+    #[error("Failed to async write zeroes")]
+    AsyncWriteZeroes(#[source] AsyncIoError),
     #[error("Failed allocating a temporary buffer")]
     TemporaryBufferAllocation(#[source] io::Error),
 }
@@ -168,6 +172,7 @@ impl ExecuteError {
             ExecuteError::Flush(_) => VIRTIO_BLK_S_IOERR,
             ExecuteError::Read(_) => VIRTIO_BLK_S_IOERR,
             ExecuteError::ReadExact(_) => VIRTIO_BLK_S_IOERR,
+            ExecuteError::ReadOnly => VIRTIO_BLK_S_IOERR,
             ExecuteError::Seek(_) => VIRTIO_BLK_S_IOERR,
             ExecuteError::Write(_) => VIRTIO_BLK_S_IOERR,
             ExecuteError::WriteAll(_) => VIRTIO_BLK_S_IOERR,
@@ -177,6 +182,8 @@ impl ExecuteError {
             ExecuteError::AsyncRead(_) => VIRTIO_BLK_S_IOERR,
             ExecuteError::AsyncWrite(_) => VIRTIO_BLK_S_IOERR,
             ExecuteError::AsyncFlush(_) => VIRTIO_BLK_S_IOERR,
+            ExecuteError::AsyncPunchHole(_) => VIRTIO_BLK_S_IOERR,
+            ExecuteError::AsyncWriteZeroes(_) => VIRTIO_BLK_S_IOERR,
             ExecuteError::TemporaryBufferAllocation(_) => VIRTIO_BLK_S_IOERR,
         };
         status as u8
@@ -189,6 +196,8 @@ pub enum RequestType {
     Out,
     Flush,
     GetDeviceId,
+    Discard,
+    WriteZeroes,
     Unsupported(u32),
 }
 
@@ -202,6 +211,8 @@ pub fn request_type<B: Bitmap + 'static>(
         VIRTIO_BLK_T_OUT => Ok(RequestType::Out),
         VIRTIO_BLK_T_FLUSH => Ok(RequestType::Flush),
         VIRTIO_BLK_T_GET_ID => Ok(RequestType::GetDeviceId),
+        VIRTIO_BLK_T_DISCARD => Ok(RequestType::Discard),
+        VIRTIO_BLK_T_WRITE_ZEROES => Ok(RequestType::WriteZeroes),
         t => Ok(RequestType::Unsupported(t)),
     }
 }
@@ -229,6 +240,20 @@ pub struct AlignedOperation {
     layout: Layout,
 }
 
+pub struct BatchRequest {
+    pub offset: libc::off_t,
+    pub iovecs: SmallVec<[libc::iovec; DEFAULT_DESCRIPTOR_VEC_SIZE]>,
+    pub user_data: u64,
+    pub request_type: RequestType,
+}
+
+pub struct ExecuteAsync {
+    // `true` if the execution will complete asynchronously
+    pub async_complete: bool,
+    // request need to be batched for submission if any
+    pub batch_request: Option<BatchRequest>,
+}
+
 #[derive(Debug)]
 pub struct Request {
     pub request_type: RequestType,
@@ -243,7 +268,7 @@ pub struct Request {
 impl Request {
     pub fn parse<B: Bitmap + 'static>(
         desc_chain: &mut DescriptorChain<GuestMemoryLoadGuard<vm_memory::GuestMemoryMmap<B>>>,
-        access_platform: Option<&Arc<dyn AccessPlatform>>,
+        access_platform: Option<&dyn AccessPlatform>,
     ) -> result::Result<Request, Error> {
         let hdr_desc = desc_chain
             .next()
@@ -276,22 +301,21 @@ impl Request {
             .next()
             .ok_or(Error::DescriptorChainTooShort)
             .inspect_err(|_| {
-                error!("Only head descriptor present: request = {:?}", req);
+                error!("Only head descriptor present: request = {req:?}");
             })?;
 
-        if !desc.has_next() {
-            status_desc = desc;
-            // Only flush requests are allowed to skip the data descriptor.
-            if req.request_type != RequestType::Flush {
-                error!("Need a data descriptor: request = {:?}", req);
-                return Err(Error::DescriptorChainTooShort);
-            }
-        } else {
+        if desc.has_next() {
             req.data_descriptors.reserve_exact(1);
             while desc.has_next() {
                 if desc.is_write_only() && req.request_type == RequestType::Out {
                     return Err(Error::UnexpectedWriteOnlyDescriptor);
                 }
+                if desc.is_write_only() && req.request_type == RequestType::Discard {
+                    return Err(Error::UnexpectedWriteOnlyDescriptor);
+                }
+                if desc.is_write_only() && req.request_type == RequestType::WriteZeroes {
+                    return Err(Error::UnexpectedWriteOnlyDescriptor);
+                }
                 if !desc.is_write_only() && req.request_type == RequestType::In {
                     return Err(Error::UnexpectedReadOnlyDescriptor);
                 }
@@ -308,10 +332,17 @@ impl Request {
                     .next()
                     .ok_or(Error::DescriptorChainTooShort)
                     .inspect_err(|_| {
-                        error!("DescriptorChain corrupted: request = {:?}", req);
+                        error!("DescriptorChain corrupted: request = {req:?}");
                     })?;
             }
             status_desc = desc;
+        } else {
+            status_desc = desc;
+            // Only flush requests are allowed to skip the data descriptor.
+            if req.request_type != RequestType::Flush {
+                error!("Need a data descriptor: request = {req:?}");
+                return Err(Error::DescriptorChainTooShort);
+            }
         }
 
         // The status MUST always be writable.
@@ -381,8 +412,14 @@ impl Request {
                     mem.write_slice(serial, *data_addr)
                         .map_err(ExecuteError::Write)?;
                 }
+                RequestType::Discard => {
+                    return Err(ExecuteError::Unsupported(VIRTIO_BLK_T_DISCARD));
+                }
+                RequestType::WriteZeroes => {
+                    return Err(ExecuteError::Unsupported(VIRTIO_BLK_T_WRITE_ZEROES));
+                }
                 RequestType::Unsupported(t) => return Err(ExecuteError::Unsupported(t)),
-            };
+            }
         }
         Ok(len)
     }
@@ -393,8 +430,9 @@ impl Request {
         disk_nsectors: u64,
         disk_image: &mut dyn AsyncIo,
         serial: &[u8],
+        disable_sector0_writes: bool,
         user_data: u64,
-    ) -> result::Result<bool, ExecuteError> {
+    ) -> result::Result<ExecuteAsync, ExecuteError> {
         let sector = self.sector;
         let request_type = self.request_type;
         let offset = (sector << SECTOR_SHIFT) as libc::off_t;
@@ -424,14 +462,17 @@ impl Request {
 
             let origin_ptr = mem
                 .get_slice(data_addr, data_len)
-                .map_err(ExecuteError::GetHostAddress)?
-                .ptr_guard();
+                .map_err(ExecuteError::GetHostAddress)?;
+            assert!(origin_ptr.len() >= data_len);
+            let origin_ptr = origin_ptr.ptr_guard();
 
             // Verify the buffer alignment.
             // In case it's not properly aligned, an intermediate buffer is
             // created with the correct alignment, and a copy from/to the
             // origin buffer is performed, depending on the type of operation.
-            let iov_base = if (origin_ptr.as_ptr() as u64) % SECTOR_SIZE != 0 {
+            let iov_base = if (origin_ptr.as_ptr() as u64).is_multiple_of(SECTOR_SIZE) {
+                origin_ptr.as_ptr() as *mut libc::c_void
+            } else {
                 let layout = Layout::from_size_align(data_len, SECTOR_SIZE as usize).unwrap();
                 // SAFETY: layout has non-zero size
                 let aligned_ptr = unsafe { alloc_zeroed(layout) };
@@ -459,8 +500,6 @@ impl Request {
                 });
 
                 aligned_ptr as *mut libc::c_void
-            } else {
-                origin_ptr.as_ptr() as *mut libc::c_void
             };
 
             let iovec = libc::iovec {
@@ -470,6 +509,10 @@ impl Request {
             iovecs.push(iovec);
         }
 
+        let mut ret = ExecuteAsync {
+            async_complete: true,
+            batch_request: None,
+        };
         // Queue operations expected to be submitted.
         match request_type {
             RequestType::In => {
@@ -479,14 +522,32 @@ impl Request {
                         .bitmap()
                         .mark_dirty(0, *data_len as usize);
                 }
-                disk_image
-                    .read_vectored(offset, &iovecs, user_data)
-                    .map_err(ExecuteError::AsyncRead)?;
+                if disk_image.batch_requests_enabled() {
+                    ret.batch_request = Some(BatchRequest {
+                        offset,
+                        iovecs,
+                        user_data,
+                        request_type,
+                    });
+                } else {
+                    disk_image
+                        .read_vectored(offset, &iovecs, user_data)
+                        .map_err(ExecuteError::AsyncRead)?;
+                }
             }
             RequestType::Out => {
-                disk_image
-                    .write_vectored(offset, &iovecs, user_data)
-                    .map_err(ExecuteError::AsyncWrite)?;
+                if disk_image.batch_requests_enabled() {
+                    ret.batch_request = Some(BatchRequest {
+                        offset,
+                        iovecs,
+                        user_data,
+                        request_type,
+                    });
+                } else {
+                    disk_image
+                        .write_vectored(offset, &iovecs, user_data)
+                        .map_err(ExecuteError::AsyncWrite)?;
+                }
             }
             RequestType::Flush => {
                 disk_image
@@ -504,12 +565,78 @@ impl Request {
                 }
                 mem.write_slice(serial, data_addr)
                     .map_err(ExecuteError::Write)?;
-                return Ok(false);
+                ret.async_complete = false;
+                return Ok(ret);
+            }
+            RequestType::Discard => {
+                let (data_addr, data_len) = if self.data_descriptors.len() == 1 {
+                    (self.data_descriptors[0].0, self.data_descriptors[0].1)
+                } else {
+                    return Err(ExecuteError::BadRequest(Error::TooManyDescriptors));
+                };
+
+                if data_len < 16 {
+                    return Err(ExecuteError::BadRequest(Error::DescriptorLengthTooSmall));
+                }
+
+                let mut discard_sector = [0u8; 8];
+                let mut discard_num_sectors = [0u8; 4];
+                mem.read_slice(&mut discard_sector, data_addr)
+                    .map_err(ExecuteError::Read)?;
+                mem.read_slice(&mut discard_num_sectors, data_addr.checked_add(8).unwrap())
+                    .map_err(ExecuteError::Read)?;
+
+                let discard_sector = u64::from_le_bytes(discard_sector);
+
+                if discard_sector == 0 && disable_sector0_writes {
+                    return Err(ExecuteError::BadRequest(Error::InvalidOffset));
+                }
+
+                let discard_num_sectors = u32::from_le_bytes(discard_num_sectors);
+
+                let discard_offset = discard_sector * SECTOR_SIZE;
+                let discard_length = (discard_num_sectors as u64) * SECTOR_SIZE;
+
+                disk_image
+                    .punch_hole(discard_offset, discard_length, user_data)
+                    .map_err(ExecuteError::AsyncPunchHole)?;
+            }
+            RequestType::WriteZeroes => {
+                let (data_addr, data_len) = if self.data_descriptors.len() == 1 {
+                    (self.data_descriptors[0].0, self.data_descriptors[0].1)
+                } else {
+                    return Err(ExecuteError::BadRequest(Error::TooManyDescriptors));
+                };
+
+                if data_len < 16 {
+                    return Err(ExecuteError::BadRequest(Error::DescriptorLengthTooSmall));
+                }
+
+                let mut wz_sector = [0u8; 8];
+
+                let mut wz_num_sectors = [0u8; 4];
+                mem.read_slice(&mut wz_sector, data_addr)
+                    .map_err(ExecuteError::Read)?;
+                mem.read_slice(&mut wz_num_sectors, data_addr.checked_add(8).unwrap())
+                    .map_err(ExecuteError::Read)?;
+
+                let wz_sector = u64::from_le_bytes(wz_sector);
+                let wz_num_sectors = u32::from_le_bytes(wz_num_sectors);
+
+                let wz_offset = wz_sector * SECTOR_SIZE;
+                if wz_offset == 0 && disable_sector0_writes {
+                    return Err(ExecuteError::BadRequest(Error::InvalidOffset));
+                }
+                let wz_length = (wz_num_sectors as u64) * SECTOR_SIZE;
+
+                disk_image
+                    .write_zeroes(wz_offset, wz_length, user_data)
+                    .map_err(ExecuteError::AsyncWriteZeroes)?;
             }
             RequestType::Unsupported(t) => return Err(ExecuteError::Unsupported(t)),
         }
 
-        Ok(true)
+        Ok(ret)
     }
 
     pub fn complete_async(&mut self) -> result::Result<(), Error> {
@@ -524,7 +651,7 @@ impl Request {
                         aligned_operation.aligned_ptr as *const u8,
                         aligned_operation.origin_ptr as *mut u8,
                         aligned_operation.size,
-                    )
+                    );
                 };
             }
 
@@ -535,7 +662,7 @@ impl Request {
                 dealloc(
                     aligned_operation.aligned_ptr as *mut u8,
                     aligned_operation.layout,
-                )
+                );
             };
         }
 
@@ -543,7 +670,7 @@ impl Request {
     }
 
     pub fn set_writeback(&mut self, writeback: bool) {
-        self.writeback = writeback
+        self.writeback = writeback;
     }
 }
 
@@ -606,7 +733,7 @@ pub fn block_io_uring_is_supported() -> bool {
         let io_uring = match IoUring::new(1) {
             Ok(io_uring) => io_uring,
             Err(e) => {
-                info!("{} failed to create io_uring instance: {}", error_msg, e);
+                info!("{error_msg} failed to create io_uring instance: {e}");
                 return false;
             }
         };
@@ -619,26 +746,26 @@ pub fn block_io_uring_is_supported() -> bool {
         match submitter.register_probe(&mut probe) {
             Ok(_) => {}
             Err(e) => {
-                info!("{} failed to register a probe: {}", error_msg, e);
+                info!("{error_msg} failed to register a probe: {e}");
                 return false;
             }
         }
 
         // Check IORING_OP_FSYNC is supported
         if !probe.is_supported(opcode::Fsync::CODE) {
-            info!("{} IORING_OP_FSYNC operation not supported", error_msg);
+            info!("{error_msg} IORING_OP_FSYNC operation not supported");
             return false;
         }
 
         // Check IORING_OP_READV is supported
         if !probe.is_supported(opcode::Readv::CODE) {
-            info!("{} IORING_OP_READV operation not supported", error_msg);
+            info!("{error_msg} IORING_OP_READV operation not supported");
             return false;
         }
 
         // Check IORING_OP_WRITEV is supported
         if !probe.is_supported(opcode::Writev::CODE) {
-            info!("{} IORING_OP_WRITEV operation not supported", error_msg);
+            info!("{error_msg} IORING_OP_WRITEV operation not supported");
             return false;
         }
 
@@ -646,10 +773,146 @@ pub fn block_io_uring_is_supported() -> bool {
     }
 }
 
-pub trait AsyncAdaptor<F>
-where
-    F: Read + Write + Seek,
-{
+/// Probe whether the file/device supports punch hole and zero range
+pub fn probe_sparse_support(file: &File) -> bool {
+    let fd = file.as_raw_fd();
+
+    let is_block_device = {
+        let mut stat = std::mem::MaybeUninit::<libc::stat>::uninit();
+        // SAFETY: FFI call with valid fd and buffer
+        let ret = unsafe { libc::fstat(fd, stat.as_mut_ptr()) };
+        if ret != 0 {
+            warn!(
+                "Failed to stat file descriptor for sparse probe: {}",
+                io::Error::last_os_error()
+            );
+            return false;
+        }
+        // SAFETY: stat result is valid at this point
+        unsafe { (*stat.as_ptr()).st_mode & S_IFMT == S_IFBLK }
+    };
+
+    if is_block_device {
+        probe_block_device_sparse_support(fd)
+    } else {
+        probe_file_sparse_support(fd)
+    }
+}
+
+/// Probe sparse support for a regular file using fallocate().
+fn probe_file_sparse_support(fd: libc::c_int) -> bool {
+    const FALLOC_FL_KEEP_SIZE: libc::c_int = 0x01;
+    const FALLOC_FL_PUNCH_HOLE: libc::c_int = 0x02;
+    const FALLOC_FL_ZERO_RANGE: libc::c_int = 0x10;
+
+    // SAFETY: FFI call with valid fd
+    let file_size = unsafe { libc::lseek(fd, 0, libc::SEEK_END) };
+    if file_size < 0 {
+        let err = io::Error::last_os_error();
+        warn!("Failed to get file size for sparse probe: {err}");
+        return false;
+    }
+
+    // SAFETY: FFI call with valid fd, probing past EOF is safe with KEEP_SIZE
+    let punch_hole =
+        unsafe { libc::fallocate(fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, file_size, 1) }
+            == 0;
+
+    if !punch_hole {
+        let err = io::Error::last_os_error();
+        if err.raw_os_error() == Some(libc::EOPNOTSUPP) {
+            debug!("File does not support FALLOC_FL_PUNCH_HOLE: {err}");
+        } else {
+            debug!("PUNCH_HOLE probe returned unexpected error: {err}");
+        }
+    }
+
+    // SAFETY: FFI call with valid fd, probing past EOF is safe with KEEP_SIZE
+    let zero_range =
+        unsafe { libc::fallocate(fd, FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE, file_size, 1) }
+            == 0;
+
+    if !zero_range {
+        let err = io::Error::last_os_error();
+        if err.raw_os_error() == Some(libc::EOPNOTSUPP) {
+            debug!("File does not support FALLOC_FL_ZERO_RANGE: {err}");
+        }
+    }
+
+    let supported = punch_hole || zero_range;
+    info!(
+        "Probed file sparse support: punch_hole={punch_hole}, zero_range={zero_range} => {supported}"
+    );
+    supported
+}
+
+/// Probe sparse support for a block device using ioctls.
+fn probe_block_device_sparse_support(fd: libc::c_int) -> bool {
+    ioctl_io_nr!(BLKDISCARD, 0x12, 119);
+    ioctl_io_nr!(BLKZEROOUT, 0x12, 127);
+
+    let range: [u64; 2] = [0, 0];
+
+    // SAFETY: FFI call with valid fd and valid range buffer
+    let punch_hole = unsafe { ioctl(fd, BLKDISCARD() as _, &range) } == 0;
+
+    if !punch_hole {
+        let err = io::Error::last_os_error();
+        debug!("Block device BLKDISCARD probe returned: {err}");
+    }
+
+    // SAFETY: FFI call with valid fd and valid range buffer
+    let zero_range = unsafe { ioctl(fd, BLKZEROOUT() as _, &range) } == 0;
+
+    if !zero_range {
+        let err = io::Error::last_os_error();
+        debug!("Block device BLKZEROOUT probe returned: {err}");
+    }
+
+    let supported = punch_hole || zero_range;
+    info!(
+        "Probed block device sparse support: punch_hole={punch_hole}, zero_range={zero_range} => {supported}"
+    );
+    supported
+}
+
+/// Preallocate disk space for a disk image file.
+///
+/// Uses `fallocate()` to allocate all disk space upfront, ensuring storage
+/// availability and reducing fragmentation. Allocating all blocks upfront is
+/// more likely to place them contiguously than allocating on demand during
+/// random writes.
+pub fn preallocate_disk<P: AsRef<Path>>(file: &File, path: P) {
+    let size = match file.metadata() {
+        Ok(m) => m.len(),
+        Err(e) => {
+            warn!("Failed to get metadata for {:?}: {}", path.as_ref(), e);
+            return;
+        }
+    };
+
+    if size == 0 {
+        return;
+    }
+
+    // SAFETY: FFI call with valid file descriptor and size
+    let ret = unsafe { libc::fallocate(file.as_raw_fd(), 0, 0, size as libc::off_t) };
+
+    if ret != 0 {
+        warn!(
+            "Failed to preallocate disk space for {:?}: {}",
+            path.as_ref(),
+            io::Error::last_os_error()
+        );
+    } else {
+        debug!(
+            "Preallocated {size} bytes for disk image {:?}",
+            path.as_ref()
+        );
+    }
+}
+
+pub trait AsyncAdaptor {
     fn read_vectored_sync(
         &mut self,
         offset: libc::off_t,
@@ -657,7 +920,10 @@ where
         user_data: u64,
         eventfd: &EventFd,
         completion_list: &mut VecDeque<(u64, i32)>,
-    ) -> AsyncIoResult<()> {
+    ) -> AsyncIoResult<()>
+    where
+        Self: Read + Seek,
+    {
         // Convert libc::iovec into IoSliceMut
         let mut slices: SmallVec<[IoSliceMut; DEFAULT_DESCRIPTOR_VEC_SIZE]> =
             SmallVec::with_capacity(iovecs.len());
@@ -669,15 +935,13 @@ where
         }
 
         let result = {
-            let mut file = self.file();
-
             // Move the cursor to the right offset
-            file.seek(SeekFrom::Start(offset as u64))
+            self.seek(SeekFrom::Start(offset as u64))
                 .map_err(AsyncIoError::ReadVectored)?;
 
             let mut r = 0;
             for b in slices.iter_mut() {
-                r += file.read(b).map_err(AsyncIoError::ReadVectored)?;
+                r += self.read(b).map_err(AsyncIoError::ReadVectored)?;
             }
             r
         };
@@ -695,7 +959,10 @@ where
         user_data: u64,
         eventfd: &EventFd,
         completion_list: &mut VecDeque<(u64, i32)>,
-    ) -> AsyncIoResult<()> {
+    ) -> AsyncIoResult<()>
+    where
+        Self: Write + Seek,
+    {
         // Convert libc::iovec into IoSlice
         let mut slices: SmallVec<[IoSlice; DEFAULT_DESCRIPTOR_VEC_SIZE]> =
             SmallVec::with_capacity(iovecs.len());
@@ -707,15 +974,13 @@ where
         }
 
         let result = {
-            let mut file = self.file();
-
             // Move the cursor to the right offset
-            file.seek(SeekFrom::Start(offset as u64))
+            self.seek(SeekFrom::Start(offset as u64))
                 .map_err(AsyncIoError::WriteVectored)?;
 
             let mut r = 0;
             for b in slices.iter() {
-                r += file.write(b).map_err(AsyncIoError::WriteVectored)?;
+                r += self.write(b).map_err(AsyncIoError::WriteVectored)?;
             }
             r
         };
@@ -731,12 +996,13 @@ where
         user_data: Option<u64>,
         eventfd: &EventFd,
         completion_list: &mut VecDeque<(u64, i32)>,
-    ) -> AsyncIoResult<()> {
+    ) -> AsyncIoResult<()>
+    where
+        Self: Write,
+    {
         let result: i32 = {
-            let mut file = self.file();
-
             // Flush
-            file.flush().map_err(AsyncIoError::Fsync)?;
+            self.flush().map_err(AsyncIoError::Fsync)?;
 
             0
         };
@@ -748,15 +1014,46 @@ where
 
         Ok(())
     }
-
-    fn file(&mut self) -> MutexGuard<'_, F>;
 }
 
+#[derive(Serialize, Deserialize, Clone, Copy, Debug, PartialEq, Eq, Default)]
 pub enum ImageType {
     FixedVhd,
     Qcow2,
     Raw,
     Vhdx,
+    #[default]
+    Unknown,
+}
+
+impl fmt::Display for ImageType {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            ImageType::FixedVhd => write!(f, "vhd"),
+            ImageType::Qcow2 => write!(f, "qcow2"),
+            ImageType::Raw => write!(f, "raw"),
+            ImageType::Vhdx => write!(f, "vhdx"),
+            ImageType::Unknown => write!(f, "unknown"),
+        }
+    }
+}
+
+pub enum ImageTypeParseError {
+    InvalidValue(String),
+}
+
+impl FromStr for ImageType {
+    type Err = ImageTypeParseError;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        match s.to_lowercase().as_str() {
+            "vhd" => Ok(ImageType::FixedVhd),
+            "qcow2" => Ok(ImageType::Qcow2),
+            "raw" => Ok(ImageType::Raw),
+            "vhdx" => Ok(ImageType::Vhdx),
+            _ => Err(ImageTypeParseError::InvalidValue(s.to_string())),
+        }
+    }
 }
 
 const QCOW_MAGIC: u32 = 0x5146_49fb;
@@ -798,7 +1095,14 @@ pub fn detect_image_type(f: &mut File) -> std::io::Result<ImageType> {
 }
 
 pub trait BlockBackend: Read + Write + Seek + Send + Debug {
-    fn size(&self) -> Result<u64, Error>;
+    /// Returns the logical disk size a guest will see.
+    ///
+    /// For raw formats, this is equal to [`Self::physical_size`]. For file formats
+    /// that wrap disk images in a container (e.g. QCOW2), this refers to the
+    /// effective size that the guest will see.
+    fn logical_size(&self) -> Result<u64, Error>;
+    /// Returns the physical size of the underlying file.
+    fn physical_size(&self) -> Result<u64, Error>;
 }
 
 #[derive(Debug)]
@@ -825,6 +1129,7 @@ ioctl_io_nr!(BLKPBSZGET, 0x12, 123);
 ioctl_io_nr!(BLKIOMIN, 0x12, 120);
 ioctl_io_nr!(BLKIOOPT, 0x12, 121);
 
+#[derive(Copy, Clone)]
 enum BlockSize {
     LogicalBlock,
     PhysicalBlock,
@@ -864,7 +1169,7 @@ impl DiskTopology {
         };
         if ret != 0 {
             return Err(std::io::Error::last_os_error());
-        };
+        }
 
         Ok(block_size)
     }

block/src/qcow/decoder.rs

@@ -0,0 +1,87 @@
+// Copyright 2025 The Cloud Hypervisor Authors. All rights reserved.
+//
+// SPDX-License-Identifier: Apache-2.0
+
+use thiserror::Error;
+
+#[derive(Debug, Error)]
+pub enum Error {
+    #[error("Zlib decompress error")]
+    ZlibDecompress(#[source] flate2::DecompressError),
+    #[error("Zlib unexpected status: {0:?}")]
+    ZlibUnexpectedStatus(flate2::Status),
+    #[error("Zstd decompress error")]
+    ZstdDecompress(#[source] std::io::Error),
+    #[error("Zstd: failed to fill buffer")]
+    ZstdFillBuffer(#[source] std::io::Error),
+}
+
+pub type Result<T> = std::result::Result<T, Error>;
+
+/// Generic trait for decoding zlib/zstd formats
+pub trait Decoder {
+    fn decode(&self, input: &[u8], output: &mut [u8]) -> Result<usize>;
+}
+
+#[derive(Default)]
+pub struct ZlibDecoder {}
+
+impl Decoder for ZlibDecoder {
+    fn decode(&self, input: &[u8], output: &mut [u8]) -> Result<usize> {
+        use flate2::{Decompress, FlushDecompress, Status};
+
+        let mut decompressor = Decompress::new(false);
+        let status = decompressor
+            .decompress(input, output, FlushDecompress::Finish)
+            .map_err(Error::ZlibDecompress)?;
+        if status == Status::StreamEnd {
+            Ok(decompressor.total_out() as usize)
+        } else {
+            Err(Error::ZlibUnexpectedStatus(status))
+        }
+    }
+}
+
+#[derive(Default)]
+pub struct ZstdDecoder {}
+
+impl Decoder for ZstdDecoder {
+    fn decode(&self, input: &[u8], output: &mut [u8]) -> Result<usize> {
+        use std::io::Read;
+
+        let mut decoder = zstd::stream::read::Decoder::new(input).map_err(Error::ZstdDecompress)?;
+        let decoded_size = decoder.read(output).map_err(Error::ZstdFillBuffer)?;
+        Ok(decoded_size)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_zlib_decode() {
+        let d = ZlibDecoder::default();
+        let valid_input = vec![99, 96, 100, 98, 6, 0];
+        let mut output1 = vec![0; 4];
+        d.decode(&valid_input, &mut output1).unwrap();
+        assert_eq!(&output1, b"\x00\x01\x02\x03");
+
+        let invalid_input = vec![1, 2, 3, 4];
+        let mut output2 = vec![0; 1024];
+        d.decode(&invalid_input, &mut output2).unwrap_err();
+    }
+
+    #[test]
+    fn test_zstd_decode() {
+        let d = ZstdDecoder::default();
+        let valid_input = vec![40, 181, 47, 253, 32, 2, 17, 0, 0, 1, 254];
+        let mut output1 = vec![0; 2];
+        d.decode(&valid_input, &mut output1).unwrap();
+        assert_eq!(&output1, b"\x01\xfe");
+
+        let invalid_input = vec![1, 2, 3, 4];
+        let mut output2 = vec![0; 1024];
+        d.decode(&invalid_input, &mut output2).unwrap_err();
+    }
+}

block/src/qcow/qcow_raw_file.rs

@@ -4,7 +4,8 @@
 //
 // SPDX-License-Identifier: Apache-2.0 AND BSD-3-Clause
 
-use std::io::{self, BufWriter, Seek, SeekFrom, Write};
+use std::fmt::Debug;
+use std::io::{self, BufWriter, Read, Seek, SeekFrom, Write};
 use std::mem::size_of;
 use std::os::fd::{AsRawFd, RawFd};
 
@@ -13,25 +14,184 @@ use vmm_sys_util::write_zeroes::WriteZeroes;
 
 use super::RawFile;
 
+// Type aliases for the refcount read/write function pointers
+type RefcountReader = fn(&mut RawFile, usize) -> io::Result<Vec<u64>>;
+type RefcountWriter = fn(&mut RawFile, &[u64]) -> io::Result<()>;
+
+/// Big-endian file access trait.
+pub(super) trait BeUint: Sized + Copy {
+    fn from_be_slice(bytes: &[u8]) -> u64;
+    fn read_be<R: Read>(r: &mut R) -> io::Result<Self>;
+    fn write_be<W: Write>(w: &mut W, val: Self) -> io::Result<()>;
+}
+
+impl BeUint for u8 {
+    #[inline(always)]
+    fn from_be_slice(bytes: &[u8]) -> u64 {
+        bytes[0] as u64
+    }
+    #[inline(always)]
+    fn read_be<R: Read>(r: &mut R) -> io::Result<Self> {
+        r.read_u8()
+    }
+    #[inline(always)]
+    fn write_be<W: Write>(w: &mut W, val: Self) -> io::Result<()> {
+        w.write_u8(val)
+    }
+}
+
+impl BeUint for u16 {
+    #[inline(always)]
+    fn from_be_slice(bytes: &[u8]) -> u64 {
+        u16::from_be_bytes([bytes[0], bytes[1]]) as u64
+    }
+    #[inline(always)]
+    fn read_be<R: Read>(r: &mut R) -> io::Result<Self> {
+        r.read_u16::<BigEndian>()
+    }
+    #[inline(always)]
+    fn write_be<W: Write>(w: &mut W, val: Self) -> io::Result<()> {
+        w.write_u16::<BigEndian>(val)
+    }
+}
+
+impl BeUint for u32 {
+    #[inline(always)]
+    fn from_be_slice(bytes: &[u8]) -> u64 {
+        u32::from_be_bytes([bytes[0], bytes[1], bytes[2], bytes[3]]) as u64
+    }
+    #[inline(always)]
+    fn read_be<R: Read>(r: &mut R) -> io::Result<Self> {
+        r.read_u32::<BigEndian>()
+    }
+    #[inline(always)]
+    fn write_be<W: Write>(w: &mut W, val: Self) -> io::Result<()> {
+        w.write_u32::<BigEndian>(val)
+    }
+}
+
+impl BeUint for u64 {
+    #[inline(always)]
+    fn from_be_slice(bytes: &[u8]) -> u64 {
+        u64::from_be_bytes([
+            bytes[0], bytes[1], bytes[2], bytes[3], bytes[4], bytes[5], bytes[6], bytes[7],
+        ])
+    }
+    #[inline(always)]
+    fn read_be<R: Read>(r: &mut R) -> io::Result<Self> {
+        r.read_u64::<BigEndian>()
+    }
+    #[inline(always)]
+    fn write_be<W: Write>(w: &mut W, val: Self) -> io::Result<()> {
+        w.write_u64::<BigEndian>(val)
+    }
+}
+
+/// Read byte-aligned refcounts.
+fn read_refcount<T: BeUint>(file: &mut RawFile, count: usize) -> io::Result<Vec<u64>> {
+    let bytes_per_entry = size_of::<T>();
+    let mut data = vec![0u8; count * bytes_per_entry];
+    file.read_exact(&mut data)?;
+    Ok(data
+        .chunks_exact(bytes_per_entry)
+        .map(T::from_be_slice)
+        .collect())
+}
+
+/// Write byte-aligned refcounts.
+fn write_refcount<T: BeUint + TryFrom<u64>>(file: &mut RawFile, table: &[u64]) -> io::Result<()>
+where
+    <T as TryFrom<u64>>::Error: Debug,
+{
+    let bytes_per_entry = size_of::<T>();
+    let mut buffer = BufWriter::with_capacity(table.len() * bytes_per_entry, file);
+    for &val in table {
+        let converted = T::try_from(val).expect("refcount values are validated on increment");
+        T::write_be(&mut buffer, converted)?;
+    }
+    buffer.flush()
+}
+
+/// Read sub-byte refcounts. Bit 0 is the least significant bit.
+fn read_refcount_subbyte<const BITS: usize>(
+    file: &mut RawFile,
+    count: usize,
+) -> io::Result<Vec<u64>> {
+    const { assert!(BITS == 1 || BITS == 2 || BITS == 4) };
+    let entries_per_byte = 8 / BITS;
+    let mask = (1u64 << BITS) - 1;
+    let bytes_needed = count.div_ceil(entries_per_byte);
+    let mut bytes = vec![0u8; bytes_needed];
+    file.read_exact(&mut bytes)?;
+
+    let mut table = vec![0u64; count];
+    for (i, val) in table.iter_mut().enumerate() {
+        let byte_idx = i / entries_per_byte;
+        let bit_offset = (i % entries_per_byte) * BITS;
+        *val = (bytes[byte_idx] as u64 >> bit_offset) & mask;
+    }
+    Ok(table)
+}
+
+/// Write sub-byte refcounts. Bit 0 is the least significant bit.
+fn write_refcount_subbyte<const BITS: usize>(file: &mut RawFile, table: &[u64]) -> io::Result<()> {
+    const { assert!(BITS == 1 || BITS == 2 || BITS == 4) };
+    let entries_per_byte = 8 / BITS;
+    let mask = (1u64 << BITS) - 1;
+    let mut buffer = BufWriter::with_capacity(table.len().div_ceil(entries_per_byte), file);
+
+    for chunk in table.chunks(entries_per_byte) {
+        let mut byte = 0u8;
+        for (i, &val) in chunk.iter().enumerate() {
+            let bit_offset = i * BITS;
+            byte |= ((val & mask) << bit_offset) as u8;
+        }
+        buffer.write_u8(byte)?;
+    }
+    buffer.flush()
+}
+
 /// A qcow file. Allows reading/writing clusters and appending clusters.
 #[derive(Debug)]
 pub struct QcowRawFile {
     file: RawFile,
     cluster_size: u64,
     cluster_mask: u64,
+    refcount_block_entries: u64,
+    read_refcount_fn: RefcountReader,
+    write_refcount_fn: RefcountWriter,
 }
 
 impl QcowRawFile {
     /// Creates a `QcowRawFile` from the given `File`, `None` is returned if `cluster_size` is not
-    /// a power of two.
-    pub fn from(file: RawFile, cluster_size: u64) -> Option<Self> {
+    /// a power of two or refcount_bits is invalid.
+    pub fn from(file: RawFile, cluster_size: u64, refcount_bits: u64) -> Option<Self> {
         if !cluster_size.is_power_of_two() {
             return None;
         }
+
+        let (read_refcount_fn, write_refcount_fn): (RefcountReader, RefcountWriter) =
+            match refcount_bits {
+                1 => (read_refcount_subbyte::<1>, write_refcount_subbyte::<1>),
+                2 => (read_refcount_subbyte::<2>, write_refcount_subbyte::<2>),
+                4 => (read_refcount_subbyte::<4>, write_refcount_subbyte::<4>),
+                8 => (read_refcount::<u8>, write_refcount::<u8>),
+                16 => (read_refcount::<u16>, write_refcount::<u16>),
+                32 => (read_refcount::<u32>, write_refcount::<u32>),
+                64 => (read_refcount::<u64>, write_refcount::<u64>),
+                _ => return None,
+            };
+
+        // For sub-byte refcounts (1,2,4 bits), entries pack multiple per byte
+        let refcount_block_entries = cluster_size * 8 / refcount_bits;
+
         Some(QcowRawFile {
             file,
             cluster_size,
             cluster_mask: cluster_size - 1,
+            refcount_block_entries,
+            read_refcount_fn,
+            write_refcount_fn,
         })
     }
 
@@ -61,47 +221,65 @@ impl QcowRawFile {
         self.read_pointer_table(offset, count, mask)
     }
 
-    /// Writes `table` of u64 pointers to `offset` in the file.
-    /// `non_zero_flags` will be ORed with all non-zero values in `table`.
-    /// writing.
-    pub fn write_pointer_table(
+    /// Internal helper for creating a buffered writer for pointer tables.
+    #[inline]
+    fn setup_pointer_table_writer<T>(
         &mut self,
         offset: u64,
-        table: &[u64],
-        non_zero_flags: u64,
-    ) -> io::Result<()> {
+        entries: &impl Iterator<Item = T>,
+    ) -> io::Result<BufWriter<RawFile>> {
         self.file.seek(SeekFrom::Start(offset))?;
-        let mut buffer = BufWriter::with_capacity(std::mem::size_of_val(table), &mut self.file);
-        for addr in table {
-            let val = if *addr == 0 {
-                0
-            } else {
-                *addr | non_zero_flags
-            };
-            buffer.write_u64::<BigEndian>(val)?;
+        let my_file = self.file.try_clone()?;
+        let capacity = entries.size_hint().0 * size_of::<u64>();
+        Ok(BufWriter::with_capacity(capacity, my_file))
+    }
+
+    /// Writes a pointer table to `offset` in the file.
+    /// Entries are computed on-the-fly by the callback.
+    pub fn write_pointer_table<'a, T: Copy + 'a>(
+        &mut self,
+        offset: u64,
+        entries: impl Iterator<Item = &'a T>,
+        mut f: impl FnMut(&mut QcowRawFile, T) -> io::Result<u64>,
+    ) -> io::Result<()> {
+        let mut buffer = self.setup_pointer_table_writer(offset, &entries)?;
+
+        for addr in entries {
+            let entry = f(self, *addr)?;
+            u64::write_be(&mut buffer, entry)?;
         }
+        buffer.flush()?;
+        Ok(())
+    }
+
+    /// Writes a pointer table directly without transforming values.
+    pub fn write_pointer_table_direct<'a>(
+        &mut self,
+        offset: u64,
+        entries: impl Iterator<Item = &'a u64>,
+    ) -> io::Result<()> {
+        let mut buffer = self.setup_pointer_table_writer(offset, &entries)?;
+
+        for &entry in entries {
+            u64::write_be(&mut buffer, entry)?;
+        }
+        buffer.flush()?;
         Ok(())
     }
 
     /// Read a refcount block from the file and returns a Vec containing the block.
     /// Always returns a cluster's worth of data.
-    pub fn read_refcount_block(&mut self, offset: u64) -> io::Result<Vec<u16>> {
-        let count = self.cluster_size / size_of::<u16>() as u64;
-        let mut table = vec![0; count as usize];
+    #[inline]
+    pub fn read_refcount_block(&mut self, offset: u64) -> io::Result<Vec<u64>> {
         self.file.seek(SeekFrom::Start(offset))?;
-        self.file.read_u16_into::<BigEndian>(&mut table)?;
-        Ok(table)
+        (self.read_refcount_fn)(&mut self.file, self.refcount_block_entries as usize)
     }
 
     /// Writes a refcount block to the file.
-    pub fn write_refcount_block(&mut self, offset: u64, table: &[u16]) -> io::Result<()> {
+    #[inline]
+    pub fn write_refcount_block(&mut self, offset: u64, table: &[u64]) -> io::Result<()> {
         self.file.seek(SeekFrom::Start(offset))?;
-        let mut buffer = BufWriter::with_capacity(std::mem::size_of_val(table), &mut self.file);
-
-        for count in table {
-            buffer.write_u16::<BigEndian>(*count)?;
-        }
-        Ok(())
+        (self.write_refcount_fn)(&mut self.file, table)
     }
 
     /// Allocates a new cluster at the end of the current file, return the address.
@@ -120,6 +298,11 @@ impl QcowRawFile {
         Ok(Some(new_cluster_address))
     }
 
+    /// Returns a reference to the underlying file.
+    pub fn file(&self) -> &RawFile {
+        &self.file
+    }
+
     /// Returns a mutable reference to the underlying file.
     pub fn file_mut(&mut self) -> &mut RawFile {
         &mut self.file
@@ -135,6 +318,11 @@ impl QcowRawFile {
         address & self.cluster_mask
     }
 
+    /// Returns the base address of the cluster containing `address`.
+    pub fn cluster_address(&self, address: u64) -> u64 {
+        address & !self.cluster_mask
+    }
+
     /// Zeros out a cluster in the file.
     pub fn zero_cluster(&mut self, address: u64) -> io::Result<()> {
         let cluster_size = self.cluster_size as usize;
@@ -144,11 +332,15 @@ impl QcowRawFile {
     }
 
     /// Writes
-    pub fn write_cluster(&mut self, address: u64, data: Vec<u8>) -> io::Result<()> {
+    pub fn write_cluster(&mut self, address: u64, data: &[u8]) -> io::Result<()> {
         let cluster_size = self.cluster_size as usize;
         self.file.seek(SeekFrom::Start(address))?;
         self.file.write_all(&data[0..cluster_size])
     }
+
+    pub fn physical_size(&self) -> Result<u64, std::io::Error> {
+        self.file.metadata().map(|m| m.len())
+    }
 }
 
 impl Clone for QcowRawFile {
@@ -157,6 +349,9 @@ impl Clone for QcowRawFile {
             file: self.file.try_clone().expect("QcowRawFile cloning failed"),
             cluster_size: self.cluster_size,
             cluster_mask: self.cluster_mask,
+            refcount_block_entries: self.refcount_block_entries,
+            read_refcount_fn: self.read_refcount_fn,
+            write_refcount_fn: self.write_refcount_fn,
         }
     }
 }

block/src/qcow/raw_file.rs

@@ -8,13 +8,14 @@
 //
 // SPDX-License-Identifier: Apache-2.0 AND BSD-3-Clause
 
-use std::alloc::{alloc_zeroed, dealloc, Layout};
+use std::alloc::{Layout, alloc_zeroed, dealloc};
 use std::fs::{File, Metadata};
 use std::io::{self, Read, Seek, SeekFrom, Write};
 use std::os::unix::io::{AsRawFd, RawFd};
 use std::slice;
 
 use libc::c_void;
+use vmm_sys_util::file_traits::FileSync;
 use vmm_sys_util::seek_hole::SeekHole;
 use vmm_sys_util::write_zeroes::{PunchHole, WriteZeroesAt};
 
@@ -89,9 +90,9 @@ impl RawFile {
 
         let align64: u64 = self.alignment.try_into().unwrap();
 
-        (self.position % align64 == 0)
-            && ((buf.as_ptr() as usize) % self.alignment == 0)
-            && (buf.len() % self.alignment == 0)
+        self.position.is_multiple_of(align64)
+            && (buf.as_ptr() as usize).is_multiple_of(self.alignment)
+            && buf.len().is_multiple_of(self.alignment)
     }
 
     pub fn set_len(&self, size: u64) -> std::io::Result<()> {
@@ -122,6 +123,17 @@ impl RawFile {
     pub fn is_direct(&self) -> bool {
         self.direct_io
     }
+
+    /// Returns true if the file was opened with write access.
+    pub fn is_writable(&self) -> bool {
+        // SAFETY: fcntl with F_GETFL is safe and doesn't modify the file descriptor
+        let flags = unsafe { libc::fcntl(self.file.as_raw_fd(), libc::F_GETFL) };
+        if flags < 0 {
+            return false;
+        }
+        let access_mode = flags & libc::O_ACCMODE;
+        access_mode == libc::O_WRONLY || access_mode == libc::O_RDWR
+    }
 }
 
 impl Read for RawFile {
@@ -259,7 +271,7 @@ impl Write for RawFile {
                 // SAFETY: tmp_ptr was allocated by alloc_zeroed with layout
                 unsafe { dealloc(tmp_ptr, layout) };
                 return Err(io::Error::last_os_error());
-            };
+            }
 
             tmp_buf[file_offset..(file_offset + buf_len)].copy_from_slice(buf);
 
@@ -327,6 +339,12 @@ impl PunchHole for RawFile {
     }
 }
 
+impl FileSync for RawFile {
+    fn fsync(&mut self) -> std::io::Result<()> {
+        self.file.fsync()
+    }
+}
+
 impl SeekHole for RawFile {
     fn seek_hole(&mut self, offset: u64) -> std::io::Result<Option<u64>> {
         match self.file.seek_hole(offset) {
@@ -354,7 +372,11 @@ impl SeekHole for RawFile {
 }
 
 impl BlockBackend for RawFile {
-    fn size(&self) -> std::result::Result<u64, crate::Error> {
+    fn logical_size(&self) -> std::result::Result<u64, crate::Error> {
+        Ok(self.metadata().map_err(crate::Error::RawFileError)?.len())
+    }
+
+    fn physical_size(&self) -> std::result::Result<u64, crate::Error> {
         Ok(self.metadata().map_err(crate::Error::RawFileError)?.len())
     }
 }

block/src/qcow/refcount.rs

@@ -20,6 +20,9 @@ pub enum Error {
     /// `InvalidIndex` - Address requested isn't within the range of the disk.
     #[error("Address requested is not within the range of the disk")]
     InvalidIndex,
+    /// `RefblockUnaligned` - Refcount block offset is not cluster aligned.
+    #[error("Refcount block offset {0:#x} is not cluster aligned")]
+    RefblockUnaligned(u64),
     /// `NeedCluster` - Handle this error by reading the cluster and calling the function again.
     #[error("Cluster with addr={0} needs to be read")]
     NeedCluster(u64),
@@ -29,6 +32,13 @@ pub enum Error {
     /// `ReadingRefCounts` - Error reading the file into the refcount cache.
     #[error("Failed to read the file into the refcount cache")]
     ReadingRefCounts(#[source] io::Error),
+    /// `RefcountOverflow` - Refcount value exceeds maximum for the refcount width.
+    #[error("Refcount value {value} exceeds {refcount_bits}-bit max ({max})")]
+    RefcountOverflow {
+        value: u64,
+        max: u64,
+        refcount_bits: u64,
+    },
 }
 
 pub type Result<T> = std::result::Result<T, Error>;
@@ -38,16 +48,19 @@ pub type Result<T> = std::result::Result<T, Error>;
 pub struct RefCount {
     ref_table: VecCache<u64>,
     refcount_table_offset: u64,
-    refblock_cache: CacheMap<VecCache<u16>>,
+    refblock_cache: CacheMap<VecCache<u64>>,
     refcount_block_entries: u64, // number of refcounts in a cluster.
     cluster_size: u64,
     max_valid_cluster_offset: u64,
+    max_refcount: u64,  // maximum refcount value for this image's refcount_order
+    refcount_bits: u64, // number of bits per refcount entry
 }
 
 impl RefCount {
     /// Creates a `RefCount` from `file`, reading the refcount table from `refcount_table_offset`.
     /// `refcount_table_entries` specifies the number of refcount blocks used by this image.
     /// `refcount_block_entries` indicates the number of refcounts in each refcount block.
+    /// `refcount_bits` is the number of bits per refcount (1, 2, 4, 8, 16, 32, or 64).
     /// Each refcount table entry points to a refcount block.
     pub fn new(
         raw_file: &mut QcowRawFile,
@@ -55,6 +68,7 @@ impl RefCount {
         refcount_table_entries: u64,
         refcount_block_entries: u64,
         cluster_size: u64,
+        refcount_bits: u64,
     ) -> io::Result<RefCount> {
         let ref_table = VecCache::from_vec(raw_file.read_pointer_table(
             refcount_table_offset,
@@ -63,6 +77,11 @@ impl RefCount {
         )?);
         let max_valid_cluster_index = (ref_table.len() as u64) * refcount_block_entries - 1;
         let max_valid_cluster_offset = max_valid_cluster_index * cluster_size;
+        let max_refcount = if refcount_bits >= 64 {
+            u64::MAX
+        } else {
+            (1u64 << refcount_bits) - 1
+        };
         Ok(RefCount {
             ref_table,
             refcount_table_offset,
@@ -70,6 +89,8 @@ impl RefCount {
             refcount_block_entries,
             cluster_size,
             max_valid_cluster_offset,
+            max_refcount,
+            refcount_bits,
         })
     }
 
@@ -92,9 +113,17 @@ impl RefCount {
         &mut self,
         raw_file: &mut QcowRawFile,
         cluster_address: u64,
-        refcount: u16,
-        mut new_cluster: Option<(u64, VecCache<u16>)>,
+        refcount: u64,
+        mut new_cluster: Option<(u64, VecCache<u64>)>,
     ) -> Result<Option<u64>> {
+        if refcount > self.max_refcount {
+            return Err(Error::RefcountOverflow {
+                value: refcount,
+                max: self.max_refcount,
+                refcount_bits: self.refcount_bits,
+            });
+        }
+
         let (table_index, block_index) = self.get_refcount_index(cluster_address);
 
         let block_addr_disk = *self.ref_table.get(table_index).ok_or(Error::InvalidIndex)?;
@@ -119,7 +148,9 @@ impl RefCount {
         }
 
         // Unwrap is safe here as the entry was filled directly above.
-        let dropped_cluster = if !self.refblock_cache.get(table_index).unwrap().dirty() {
+        let dropped_cluster = if self.refblock_cache.get(table_index).unwrap().dirty() {
+            None
+        } else {
             // Free the previously used block and use a new one. Writing modified counts to new
             // blocks keeps the on-disk state consistent even if it's out of date.
             if let Some((addr, _)) = new_cluster.take() {
@@ -128,8 +159,6 @@ impl RefCount {
             } else {
                 return Err(Error::NeedNewCluster);
             }
-        } else {
-            None
         };
 
         self.refblock_cache.get_mut(table_index).unwrap()[block_index] = refcount;
@@ -156,11 +185,8 @@ impl RefCount {
     /// Returns true if the table changed since the previous `flush_table()` call.
     pub fn flush_table(&mut self, raw_file: &mut QcowRawFile) -> io::Result<bool> {
         if self.ref_table.dirty() {
-            raw_file.write_pointer_table(
-                self.refcount_table_offset,
-                self.ref_table.get_values(),
-                0,
-            )?;
+            raw_file
+                .write_pointer_table_direct(self.refcount_table_offset, self.ref_table.iter())?;
             self.ref_table.mark_clean();
             Ok(true)
         } else {
@@ -173,12 +199,15 @@ impl RefCount {
         &mut self,
         raw_file: &mut QcowRawFile,
         address: u64,
-    ) -> Result<u16> {
+    ) -> Result<u64> {
         let (table_index, block_index) = self.get_refcount_index(address);
         let block_addr_disk = *self.ref_table.get(table_index).ok_or(Error::InvalidIndex)?;
         if block_addr_disk == 0 {
             return Ok(0);
         }
+        if block_addr_disk & (self.cluster_size - 1) != 0 {
+            return Err(Error::RefblockUnaligned(block_addr_disk));
+        }
         if !self.refblock_cache.contains_key(table_index) {
             let table = VecCache::from_vec(
                 raw_file
@@ -205,7 +234,7 @@ impl RefCount {
         &mut self,
         raw_file: &mut QcowRawFile,
         table_index: usize,
-    ) -> Result<Option<&[u16]>> {
+    ) -> Result<Option<&[u64]>> {
         let block_addr_disk = *self.ref_table.get(table_index).ok_or(Error::InvalidIndex)?;
         if block_addr_disk == 0 {
             return Ok(None);

block/src/qcow/vec_cache.rs

@@ -4,10 +4,10 @@
 //
 // SPDX-License-Identifier: Apache-2.0 AND BSD-3-Clause
 
-use std::collections::hash_map::IterMut;
 use std::collections::HashMap;
+use std::collections::hash_map::IterMut;
 use std::io;
-use std::ops::{Index, IndexMut};
+use std::ops::{Deref, Index, IndexMut};
 use std::slice::SliceIndex;
 
 /// Trait that allows for checking if an implementor is dirty. Useful for types that are cached so
@@ -62,6 +62,21 @@ impl<T: 'static + Copy + Default> VecCache<T> {
     pub fn len(&self) -> usize {
         self.vec.len()
     }
+
+    /// Extends the cache capacity to `new_len` elements.
+    ///
+    /// No-op if `new_len <= self.len()`. Allocates a new buffer, copies
+    /// existing data, and fills new elements with default values.
+    /// Marks the cache as dirty.
+    pub fn extend(&mut self, new_len: usize) {
+        if new_len <= self.vec.len() {
+            return;
+        }
+        let mut new_vec = vec![Default::default(); new_len];
+        new_vec[..self.vec.len()].copy_from_slice(&self.vec);
+        self.vec = new_vec.into_boxed_slice();
+        self.dirty = true;
+    }
 }
 
 impl<T: 'static + Copy + Default> Cacheable for VecCache<T> {
@@ -85,6 +100,14 @@ impl<T: 'static + Copy + Default> IndexMut<usize> for VecCache<T> {
     }
 }
 
+impl<T: 'static + Copy + Default> Deref for VecCache<T> {
+    type Target = [T];
+
+    fn deref(&self) -> &[T] {
+        &self.vec
+    }
+}
+
 #[derive(Clone, Debug)]
 pub struct CacheMap<T: Cacheable> {
     capacity: usize,
@@ -123,10 +146,10 @@ impl<T: Cacheable> CacheMap<T> {
         if self.map.len() == self.capacity {
             // TODO(dgreid) - smarter eviction strategy.
             let to_evict = *self.map.iter().next().unwrap().0;
-            if let Some(evicted) = self.map.remove(&to_evict) {
-                if evicted.dirty() {
-                    write_callback(to_evict, evicted)?;
-                }
+            if let Some(evicted) = self.map.remove(&to_evict)
+                && evicted.dirty()
+            {
+                write_callback(to_evict, evicted)?;
             }
         }
         self.map.insert(index, block);
@@ -135,7 +158,7 @@ impl<T: Cacheable> CacheMap<T> {
 }
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use super::*;
 
     struct NumCache(());

block/src/qcow_sync.rs

@@ -4,44 +4,90 @@
 
 use std::collections::VecDeque;
 use std::fs::File;
-use std::io::{Seek, SeekFrom};
+use std::io::{self, Seek, SeekFrom};
 use std::os::fd::AsRawFd;
-use std::sync::{Arc, Mutex, MutexGuard};
+use std::sync::{Arc, Mutex};
 
 use vmm_sys_util::eventfd::EventFd;
+use vmm_sys_util::write_zeroes::PunchHole;
 
 use crate::async_io::{
-    AsyncIo, AsyncIoResult, BorrowedDiskFd, DiskFile, DiskFileError, DiskFileResult,
+    AsyncIo, AsyncIoError, AsyncIoResult, BorrowedDiskFd, DiskFile, DiskFileError, DiskFileResult,
 };
-use crate::qcow::{QcowFile, RawFile, Result as QcowResult};
-use crate::AsyncAdaptor;
+use crate::qcow::{Error as QcowError, MAX_NESTING_DEPTH, QcowFile, RawFile, Result as QcowResult};
+use crate::{AsyncAdaptor, BlockBackend};
 
 pub struct QcowDiskSync {
+    // FIXME: The Mutex serializes all QCOW2 I/O operations across queues, which
+    // is necessary for correctness but eliminates any parallelism benefit from
+    // multiqueue. QcowFile has internal mutable state (L2 cache, refcounts, file
+    // position) that is not safe to share across threads via Clone.
+    //
+    // A proper fix would require restructuring QcowFile to separate metadata
+    // operations (which need synchronization) from data I/O (which could be
+    // parallelized with per queue file descriptors). See #7560 for details.
     qcow_file: Arc<Mutex<QcowFile>>,
 }
 
 impl QcowDiskSync {
-    pub fn new(file: File, direct_io: bool) -> QcowResult<Self> {
+    pub fn new(file: File, direct_io: bool, backing_files: bool, sparse: bool) -> QcowResult<Self> {
+        let max_nesting_depth = if backing_files { MAX_NESTING_DEPTH } else { 0 };
+        let qcow_file = QcowFile::from_with_nesting_depth(
+            RawFile::new(file, direct_io),
+            max_nesting_depth,
+            sparse,
+        )
+        .map_err(|e| match e {
+            QcowError::MaxNestingDepthExceeded if !backing_files => QcowError::BackingFilesDisabled,
+            other => other,
+        })?;
         Ok(QcowDiskSync {
-            qcow_file: Arc::new(Mutex::new(QcowFile::from(RawFile::new(file, direct_io))?)),
+            qcow_file: Arc::new(Mutex::new(qcow_file)),
         })
     }
 }
 
 impl DiskFile for QcowDiskSync {
-    fn size(&mut self) -> DiskFileResult<u64> {
-        let mut file = self.qcow_file.lock().unwrap();
+    fn logical_size(&mut self) -> DiskFileResult<u64> {
+        self.qcow_file
+            .lock()
+            .unwrap()
+            .seek(SeekFrom::End(0))
+            .map_err(DiskFileError::Size)
+    }
 
-        file.seek(SeekFrom::End(0)).map_err(DiskFileError::Size)
+    fn physical_size(&mut self) -> DiskFileResult<u64> {
+        self.qcow_file.lock().unwrap().physical_size().map_err(|e| {
+            let io_inner = match e {
+                crate::Error::GetFileMetadata(e) => e,
+                _ => unreachable!(),
+            };
+            DiskFileError::Size(io_inner)
+        })
     }
 
     fn new_async_io(&self, _ring_depth: u32) -> DiskFileResult<Box<dyn AsyncIo>> {
-        Ok(Box::new(QcowSync::new(self.qcow_file.clone())) as Box<dyn AsyncIo>)
+        Ok(Box::new(QcowSync::new(Arc::clone(&self.qcow_file))) as Box<dyn AsyncIo>)
+    }
+
+    fn resize(&mut self, size: u64) -> DiskFileResult<()> {
+        self.qcow_file
+            .lock()
+            .unwrap()
+            .resize(size)
+            .map_err(|e| DiskFileError::ResizeError(io::Error::other(e)))
+    }
+
+    fn supports_sparse_operations(&self) -> bool {
+        true
+    }
+
+    fn supports_zero_flag(&self) -> bool {
+        true
     }
 
     fn fd(&mut self) -> BorrowedDiskFd<'_> {
-        let lock = self.qcow_file.lock().unwrap();
-        BorrowedDiskFd::new(lock.as_raw_fd())
+        BorrowedDiskFd::new(self.qcow_file.lock().unwrap().as_raw_fd())
     }
 }
 
@@ -62,11 +108,7 @@ impl QcowSync {
     }
 }
 
-impl AsyncAdaptor<QcowFile> for Arc<Mutex<QcowFile>> {
-    fn file(&mut self) -> MutexGuard<'_, QcowFile> {
-        self.lock().unwrap()
-    }
-}
+impl AsyncAdaptor for QcowFile {}
 
 impl AsyncIo for QcowSync {
     fn notifier(&self) -> &EventFd {
@@ -79,7 +121,7 @@ impl AsyncIo for QcowSync {
         iovecs: &[libc::iovec],
         user_data: u64,
     ) -> AsyncIoResult<()> {
-        self.qcow_file.read_vectored_sync(
+        self.qcow_file.lock().unwrap().read_vectored_sync(
             offset,
             iovecs,
             user_data,
@@ -94,7 +136,7 @@ impl AsyncIo for QcowSync {
         iovecs: &[libc::iovec],
         user_data: u64,
     ) -> AsyncIoResult<()> {
-        self.qcow_file.write_vectored_sync(
+        self.qcow_file.lock().unwrap().write_vectored_sync(
             offset,
             iovecs,
             user_data,
@@ -104,11 +146,359 @@ impl AsyncIo for QcowSync {
     }
 
     fn fsync(&mut self, user_data: Option<u64>) -> AsyncIoResult<()> {
-        self.qcow_file
-            .fsync_sync(user_data, &self.eventfd, &mut self.completion_list)
+        self.qcow_file.lock().unwrap().fsync_sync(
+            user_data,
+            &self.eventfd,
+            &mut self.completion_list,
+        )
     }
 
     fn next_completed_request(&mut self) -> Option<(u64, i32)> {
         self.completion_list.pop_front()
     }
+
+    fn punch_hole(&mut self, offset: u64, length: u64, user_data: u64) -> AsyncIoResult<()> {
+        // For QCOW2, punch_hole calls deallocate_cluster
+        let result = self
+            .qcow_file
+            .lock()
+            .unwrap()
+            .punch_hole(offset, length)
+            .map(|_| 0i32)
+            .map_err(AsyncIoError::PunchHole);
+
+        match result {
+            Ok(res) => {
+                self.completion_list.push_back((user_data, res));
+                self.eventfd.write(1).unwrap();
+                Ok(())
+            }
+            Err(e) => {
+                // CRITICAL: Always signal completion even on error to avoid hangs
+                let errno = if let AsyncIoError::PunchHole(io_err) = &e {
+                    let err = io_err.raw_os_error().unwrap_or(libc::EIO);
+                    -err
+                } else {
+                    -libc::EIO
+                };
+                self.completion_list.push_back((user_data, errno));
+                self.eventfd.write(1).unwrap();
+                Ok(())
+            }
+        }
+    }
+
+    fn write_zeroes(&mut self, offset: u64, length: u64, user_data: u64) -> AsyncIoResult<()> {
+        // For QCOW2, write_zeroes is implemented by deallocating clusters via punch_hole.
+        // This is more efficient than writing actual zeros and reduces disk usage.
+        // Unallocated clusters inherently read as zero in the QCOW2 format.
+        let result = self
+            .qcow_file
+            .lock()
+            .unwrap()
+            .punch_hole(offset, length)
+            .map(|_| 0i32)
+            .map_err(AsyncIoError::WriteZeroes);
+
+        match result {
+            Ok(res) => {
+                self.completion_list.push_back((user_data, res));
+                self.eventfd.write(1).unwrap();
+                Ok(())
+            }
+            Err(e) => {
+                // Always signal completion even on error to avoid hangs
+                let errno = if let AsyncIoError::WriteZeroes(io_err) = &e {
+                    let err = io_err.raw_os_error().unwrap_or(libc::EIO);
+                    -err
+                } else {
+                    -libc::EIO
+                };
+                self.completion_list.push_back((user_data, errno));
+                self.eventfd.write(1).unwrap();
+                Ok(())
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod unit_tests {
+    use std::io::{Read, Seek, SeekFrom, Write};
+
+    use vmm_sys_util::tempfile::TempFile;
+
+    use super::*;
+    use crate::qcow::{QcowFile, QcowHeader, RawFile};
+
+    #[test]
+    fn test_qcow_async_punch_hole_completion() {
+        // Create a QCOW2 image with valid header
+        let temp_file = TempFile::new().unwrap();
+        let raw_file = RawFile::new(temp_file.into_file(), false);
+        let file_size = 1024 * 1024 * 100; // 100MB
+        let mut qcow_file = QcowFile::new(raw_file, 3, file_size, true).unwrap();
+
+        // Write some data
+        let data = vec![0xDD; 128 * 1024]; // 128KB
+        let offset = 0;
+        qcow_file.seek(SeekFrom::Start(offset)).unwrap();
+        qcow_file.write_all(&data).unwrap();
+        qcow_file.flush().unwrap();
+
+        // Create async wrapper
+        let qcow_file = Arc::new(Mutex::new(qcow_file));
+        let mut async_qcow = QcowSync::new(qcow_file.clone());
+
+        // Punch hole
+        async_qcow
+            .punch_hole(offset, data.len() as u64, 100)
+            .unwrap();
+
+        // Verify completion event was generated
+        let (user_data, result) = async_qcow.next_completed_request().unwrap();
+        assert_eq!(user_data, 100);
+        assert_eq!(result, 0, "punch_hole should succeed");
+
+        // Verify data reads as zeros
+        let mut read_buf = vec![0; data.len()];
+        qcow_file
+            .lock()
+            .unwrap()
+            .seek(SeekFrom::Start(offset))
+            .unwrap();
+        qcow_file.lock().unwrap().read_exact(&mut read_buf).unwrap();
+        assert!(
+            read_buf.iter().all(|&b| b == 0),
+            "Punched hole should read as zeros"
+        );
+    }
+
+    #[test]
+    fn test_qcow_async_write_zeroes_completion() {
+        // Create a QCOW2 image with valid header
+        let temp_file = TempFile::new().unwrap();
+        let raw_file = RawFile::new(temp_file.into_file(), false);
+        let file_size = 1024 * 1024 * 100; // 100MB
+        let mut qcow_file = QcowFile::new(raw_file, 3, file_size, true).unwrap();
+
+        // Write some data
+        let data = vec![0xEE; 256 * 1024]; // 256KB
+        let offset = 64 * 1024; // Start at 64KB offset
+        qcow_file.seek(SeekFrom::Start(offset)).unwrap();
+        qcow_file.write_all(&data).unwrap();
+        qcow_file.flush().unwrap();
+
+        // Create async wrapper
+        let qcow_file = Arc::new(Mutex::new(qcow_file));
+        let mut async_qcow = QcowSync::new(qcow_file.clone());
+
+        // Write zeros
+        async_qcow
+            .write_zeroes(offset, data.len() as u64, 200)
+            .unwrap();
+
+        // Verify completion event was generated
+        let (user_data, result) = async_qcow.next_completed_request().unwrap();
+        assert_eq!(user_data, 200);
+        assert_eq!(result, 0, "write_zeroes should succeed");
+
+        // Verify data reads as zeros
+        let mut read_buf = vec![0; data.len()];
+        qcow_file
+            .lock()
+            .unwrap()
+            .seek(SeekFrom::Start(offset))
+            .unwrap();
+        qcow_file.lock().unwrap().read_exact(&mut read_buf).unwrap();
+        assert!(
+            read_buf.iter().all(|&b| b == 0),
+            "Zeroed region should read as zeros"
+        );
+    }
+
+    #[test]
+    fn test_qcow_async_multiple_operations() {
+        // Create a QCOW2 image with valid header
+        let temp_file = TempFile::new().unwrap();
+        let raw_file = RawFile::new(temp_file.into_file(), false);
+        let file_size = 1024 * 1024 * 100; // 100MB
+        let mut qcow_file = QcowFile::new(raw_file, 3, file_size, true).unwrap();
+
+        // Write data at multiple offsets
+        let data = vec![0xFF; 64 * 1024]; // 64KB chunks
+        for i in 0..4 {
+            let offset = i * 128 * 1024; // 128KB spacing
+            qcow_file.seek(SeekFrom::Start(offset)).unwrap();
+            qcow_file.write_all(&data).unwrap();
+        }
+        qcow_file.flush().unwrap();
+
+        // Create async wrapper
+        let qcow_file = Arc::new(Mutex::new(qcow_file));
+        let mut async_qcow = QcowSync::new(qcow_file.clone());
+
+        // Queue multiple punch_hole operations
+        async_qcow.punch_hole(0, 64 * 1024, 1).unwrap();
+        async_qcow.punch_hole(128 * 1024, 64 * 1024, 2).unwrap();
+        async_qcow.punch_hole(256 * 1024, 64 * 1024, 3).unwrap();
+
+        // Verify all completions
+        let (user_data, result) = async_qcow.next_completed_request().unwrap();
+        assert_eq!(user_data, 1);
+        assert_eq!(result, 0);
+
+        let (user_data, result) = async_qcow.next_completed_request().unwrap();
+        assert_eq!(user_data, 2);
+        assert_eq!(result, 0);
+
+        let (user_data, result) = async_qcow.next_completed_request().unwrap();
+        assert_eq!(user_data, 3);
+        assert_eq!(result, 0);
+
+        // Verify no more completions
+        assert!(async_qcow.next_completed_request().is_none());
+    }
+
+    #[test]
+    fn test_qcow_punch_hole_with_shared_instance() {
+        // This test verifies that with Arc<Mutex<>>, multiple async I/O operations
+        // share the same QcowFile instance and see each other's changes.
+
+        // Create a QCOW2 image
+        let temp_file = TempFile::new().unwrap();
+        let raw_file = RawFile::new(temp_file.into_file(), false);
+        let file_size = 1024 * 1024 * 100; // 100MB
+        let mut qcow_file = QcowFile::new(raw_file, 3, file_size, true).unwrap();
+
+        // Write some data at offset 0
+        let data = vec![0xAB; 128 * 1024]; // 128KB of 0xAB pattern
+        let offset = 0;
+        qcow_file.seek(SeekFrom::Start(offset)).unwrap();
+        qcow_file.write_all(&data).unwrap();
+        qcow_file.flush().unwrap();
+
+        let qcow_shared = Arc::new(Mutex::new(qcow_file));
+
+        // First async I/O: punch hole
+        let mut async_qcow1 = QcowSync::new(qcow_shared.clone());
+        async_qcow1
+            .punch_hole(offset, data.len() as u64, 100)
+            .unwrap();
+
+        // Verify punch_hole completed
+        let (user_data, result) = async_qcow1.next_completed_request().unwrap();
+        assert_eq!(user_data, 100);
+        assert_eq!(result, 0, "punch_hole should succeed");
+
+        // Second async I/O: read from same shared instance
+        // This should see the deallocated cluster because they share the same QcowFile
+        let mut read_buf = vec![0xFF; data.len()];
+        qcow_shared
+            .lock()
+            .unwrap()
+            .seek(SeekFrom::Start(offset))
+            .unwrap();
+        qcow_shared
+            .lock()
+            .unwrap()
+            .read_exact(&mut read_buf)
+            .unwrap();
+
+        // The read should return zeros because the cluster was deallocated
+        assert!(
+            read_buf.iter().all(|&b| b == 0),
+            "After punch_hole, shared QcowFile instance should read zeros from deallocated cluster"
+        );
+    }
+
+    #[test]
+    fn test_qcow_disk_sync_punch_hole_with_new_async_io() {
+        // This test simulates the EXACT real usage pattern: QcowDiskSync.new_async_io()
+        // creates a new QcowSync with a cloned QcowFile for each I/O operation.
+
+        use std::io::Write;
+
+        use crate::async_io::DiskFile;
+
+        // Create a QCOW2 image
+        let temp_file = TempFile::new().unwrap();
+        let file_size = 1024 * 1024 * 100; // 100MB
+
+        {
+            let raw_file = RawFile::new(temp_file.as_file().try_clone().unwrap(), false);
+            let mut qcow_file = QcowFile::new(raw_file, 3, file_size, true).unwrap();
+
+            // Write data at offset 1MB - use single cluster (64KB) to simplify test
+            let data = vec![0xCD; 64 * 1024]; // 64KB (one cluster)
+            let offset = 1024 * 1024u64;
+            qcow_file.seek(SeekFrom::Start(offset)).unwrap();
+            qcow_file.write_all(&data).unwrap();
+            qcow_file.flush().unwrap();
+        }
+
+        // Open with QcowDiskSync (like real code does)
+        let disk =
+            QcowDiskSync::new(temp_file.as_file().try_clone().unwrap(), false, true, true).unwrap();
+
+        // First async I/O: punch hole (simulates DISCARD command)
+        let mut async_io1 = disk.new_async_io(1).unwrap();
+        let offset = 1024 * 1024u64;
+        let length = 64 * 1024u64; // Single cluster
+        async_io1.punch_hole(offset, length, 1).unwrap();
+        let (user_data, result) = async_io1.next_completed_request().unwrap();
+        assert_eq!(user_data, 1);
+        assert_eq!(result, 0, "punch_hole should succeed");
+        drop(async_io1);
+
+        // Second async I/O: read from the same location (simulates READ command)
+        let mut async_io2 = disk.new_async_io(1).unwrap();
+        let mut read_buf = vec![0xFF; length as usize];
+        let iovec = libc::iovec {
+            iov_base: read_buf.as_mut_ptr() as *mut libc::c_void,
+            iov_len: read_buf.len(),
+        };
+
+        // These assertions are critical to prevent compiler optimization bugs
+        // that can reorder operations. Without them, the test can fail even
+        // though the QCOW2 implementation is correct.
+        assert_eq!(iovec.iov_base as *const u8, read_buf.as_ptr());
+        assert_eq!(iovec.iov_len, read_buf.len());
+
+        async_io2
+            .read_vectored(offset as libc::off_t, &[iovec], 2)
+            .unwrap();
+
+        let (user_data, result) = async_io2.next_completed_request().unwrap();
+        assert_eq!(user_data, 2);
+        assert_eq!(
+            result as usize, length as usize,
+            "read should complete successfully"
+        );
+
+        // Verify the data is all zeros
+        assert!(
+            read_buf.iter().all(|&b| b == 0),
+            "After punch_hole via new_async_io, read should return zeros"
+        );
+    }
+
+    #[test]
+    fn backing_files_disabled_error() {
+        let header =
+            QcowHeader::create_for_size_and_path(3, 0x10_0000, Some("/path/to/backing/file"))
+                .expect("Failed to create header.");
+        let temp_file = TempFile::new().unwrap();
+        let mut raw_file = RawFile::new(temp_file.as_file().try_clone().unwrap(), false);
+        header
+            .write_to(&mut raw_file)
+            .expect("Failed to write header.");
+
+        let file = temp_file.into_file();
+        match QcowDiskSync::new(file, false, false, true) {
+            Err(QcowError::BackingFilesDisabled) => {}
+            Err(other) => panic!("Expected BackingFilesDisabled, got: {other:?}"),
+            Ok(_) => panic!("Expected BackingFilesDisabled error, but succeeded"),
+        }
+    }
 }

block/src/raw_async.rs

@@ -6,13 +6,14 @@ use std::fs::File;
 use std::io::{Error, Seek, SeekFrom};
 use std::os::unix::io::{AsRawFd, RawFd};
 
-use io_uring::{opcode, types, IoUring};
+use io_uring::{IoUring, opcode, types};
+use log::warn;
 use vmm_sys_util::eventfd::EventFd;
 
 use crate::async_io::{
     AsyncIo, AsyncIoError, AsyncIoResult, BorrowedDiskFd, DiskFile, DiskFileError, DiskFileResult,
 };
-use crate::DiskTopology;
+use crate::{BatchRequest, DiskTopology, RequestType, probe_sparse_support};
 
 pub struct RawFileDisk {
     file: File,
@@ -25,12 +26,19 @@ impl RawFileDisk {
 }
 
 impl DiskFile for RawFileDisk {
-    fn size(&mut self) -> DiskFileResult<u64> {
+    fn logical_size(&mut self) -> DiskFileResult<u64> {
         self.file
             .seek(SeekFrom::End(0))
             .map_err(DiskFileError::Size)
     }
 
+    fn physical_size(&mut self) -> DiskFileResult<u64> {
+        self.file
+            .metadata()
+            .map(|m| m.len())
+            .map_err(DiskFileError::Size)
+    }
+
     fn new_async_io(&self, ring_depth: u32) -> DiskFileResult<Box<dyn AsyncIo>> {
         Ok(Box::new(
             RawFileAsync::new(self.file.as_raw_fd(), ring_depth)
@@ -47,6 +55,14 @@ impl DiskFile for RawFileDisk {
         }
     }
 
+    fn resize(&mut self, size: u64) -> DiskFileResult<()> {
+        self.file.set_len(size).map_err(DiskFileError::ResizeError)
+    }
+
+    fn supports_sparse_operations(&self) -> bool {
+        probe_sparse_support(&self.file)
+    }
+
     fn fd(&mut self) -> BorrowedDiskFd<'_> {
         BorrowedDiskFd::new(self.file.as_raw_fd())
     }
@@ -97,7 +113,7 @@ impl AsyncIo for RawFileAsync {
                     .build()
                     .user_data(user_data),
             )
-            .map_err(|_| AsyncIoError::ReadVectored(Error::other("Submission queue is full")))?
+            .map_err(|_| AsyncIoError::ReadVectored(Error::other("Submission queue is full")))?;
         };
 
         // Update the submission queue and submit new operations to the
@@ -125,7 +141,7 @@ impl AsyncIo for RawFileAsync {
                     .build()
                     .user_data(user_data),
             )
-            .map_err(|_| AsyncIoError::WriteVectored(Error::other("Submission queue is full")))?
+            .map_err(|_| AsyncIoError::WriteVectored(Error::other("Submission queue is full")))?;
         };
 
         // Update the submission queue and submit new operations to the
@@ -147,7 +163,7 @@ impl AsyncIo for RawFileAsync {
                         .build()
                         .user_data(user_data),
                 )
-                .map_err(|_| AsyncIoError::Fsync(Error::other("Submission queue is full")))?
+                .map_err(|_| AsyncIoError::Fsync(Error::other("Submission queue is full")))?;
             };
 
             // Update the submission queue and submit new operations to the
@@ -168,4 +184,131 @@ impl AsyncIo for RawFileAsync {
             .next()
             .map(|entry| (entry.user_data(), entry.result()))
     }
+
+    fn batch_requests_enabled(&self) -> bool {
+        true
+    }
+
+    fn submit_batch_requests(&mut self, batch_request: &[BatchRequest]) -> AsyncIoResult<()> {
+        if !self.batch_requests_enabled() {
+            return Ok(());
+        }
+
+        let (submitter, mut sq, _) = self.io_uring.split();
+        let mut submitted = false;
+
+        for req in batch_request {
+            match req.request_type {
+                RequestType::In => {
+                    // SAFETY: we know the file descriptor is valid and we
+                    // relied on vm-memory to provide the buffer address.
+                    unsafe {
+                        sq.push(
+                            &opcode::Readv::new(
+                                types::Fd(self.fd),
+                                req.iovecs.as_ptr(),
+                                req.iovecs.len() as u32,
+                            )
+                            .offset(req.offset as u64)
+                            .build()
+                            .user_data(req.user_data),
+                        )
+                        .map_err(|_| {
+                            AsyncIoError::ReadVectored(Error::other("Submission queue is full"))
+                        })?;
+                    };
+                    submitted = true;
+                }
+                RequestType::Out => {
+                    // SAFETY: we know the file descriptor is valid and we
+                    // relied on vm-memory to provide the buffer address.
+                    unsafe {
+                        sq.push(
+                            &opcode::Writev::new(
+                                types::Fd(self.fd),
+                                req.iovecs.as_ptr(),
+                                req.iovecs.len() as u32,
+                            )
+                            .offset(req.offset as u64)
+                            .build()
+                            .user_data(req.user_data),
+                        )
+                        .map_err(|_| {
+                            AsyncIoError::WriteVectored(Error::other("Submission queue is full"))
+                        })?;
+                    };
+                    submitted = true;
+                }
+                _ => {
+                    unreachable!("Unexpected batch request type: {:?}", req.request_type)
+                }
+            }
+        }
+
+        // Only submit if we actually queued something
+        if submitted {
+            // Update the submission queue and submit new operations to the
+            // io_uring instance.
+            sq.sync();
+            submitter
+                .submit()
+                .map_err(AsyncIoError::SubmitBatchRequests)?;
+        }
+
+        Ok(())
+    }
+
+    fn punch_hole(&mut self, offset: u64, length: u64, user_data: u64) -> AsyncIoResult<()> {
+        let (submitter, mut sq, _) = self.io_uring.split();
+
+        const FALLOC_FL_PUNCH_HOLE: i32 = 0x02;
+        const FALLOC_FL_KEEP_SIZE: i32 = 0x01;
+        let mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
+
+        // SAFETY: The file descriptor is known to be valid.
+        unsafe {
+            sq.push(
+                &opcode::Fallocate::new(types::Fd(self.fd), length)
+                    .offset(offset)
+                    .mode(mode)
+                    .build()
+                    .user_data(user_data),
+            )
+            .map_err(|e| {
+                AsyncIoError::PunchHole(Error::other(format!("Submission queue is full: {e:?}")))
+            })?;
+        };
+
+        sq.sync();
+        submitter.submit().map_err(AsyncIoError::PunchHole)?;
+
+        Ok(())
+    }
+
+    fn write_zeroes(&mut self, offset: u64, length: u64, user_data: u64) -> AsyncIoResult<()> {
+        let (submitter, mut sq, _) = self.io_uring.split();
+
+        const FALLOC_FL_ZERO_RANGE: i32 = 0x10;
+        const FALLOC_FL_KEEP_SIZE: i32 = 0x01;
+        let mode = FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE;
+
+        // SAFETY: The file descriptor is known to be valid.
+        unsafe {
+            sq.push(
+                &opcode::Fallocate::new(types::Fd(self.fd), length)
+                    .offset(offset)
+                    .mode(mode)
+                    .build()
+                    .user_data(user_data),
+            )
+            .map_err(|e| {
+                AsyncIoError::WriteZeroes(Error::other(format!("Submission queue is full: {e:?}")))
+            })?;
+        };
+
+        sq.sync();
+        submitter.submit().map_err(AsyncIoError::WriteZeroes)?;
+
+        Ok(())
+    }
 }

block/src/raw_async_aio.rs

@@ -9,13 +9,14 @@ use std::fs::File;
 use std::io::{Seek, SeekFrom};
 use std::os::unix::io::{AsRawFd, RawFd};
 
+use log::warn;
 use vmm_sys_util::aio;
 use vmm_sys_util::eventfd::EventFd;
 
 use crate::async_io::{
     AsyncIo, AsyncIoError, AsyncIoResult, BorrowedDiskFd, DiskFile, DiskFileError, DiskFileResult,
 };
-use crate::DiskTopology;
+use crate::{DiskTopology, probe_sparse_support};
 
 pub struct RawFileDiskAio {
     file: File,
@@ -28,12 +29,19 @@ impl RawFileDiskAio {
 }
 
 impl DiskFile for RawFileDiskAio {
-    fn size(&mut self) -> DiskFileResult<u64> {
+    fn logical_size(&mut self) -> DiskFileResult<u64> {
         self.file
             .seek(SeekFrom::End(0))
             .map_err(DiskFileError::Size)
     }
 
+    fn physical_size(&mut self) -> DiskFileResult<u64> {
+        self.file
+            .metadata()
+            .map(|m| m.len())
+            .map_err(DiskFileError::Size)
+    }
+
     fn new_async_io(&self, ring_depth: u32) -> DiskFileResult<Box<dyn AsyncIo>> {
         Ok(Box::new(
             RawFileAsyncAio::new(self.file.as_raw_fd(), ring_depth)
@@ -50,6 +58,10 @@ impl DiskFile for RawFileDiskAio {
         }
     }
 
+    fn supports_sparse_operations(&self) -> bool {
+        probe_sparse_support(&self.file)
+    }
+
     fn fd(&mut self) -> BorrowedDiskFd<'_> {
         BorrowedDiskFd::new(self.file.as_raw_fd())
     }
@@ -153,4 +165,16 @@ impl AsyncIo for RawFileAsyncAio {
             Some((events[0].data, events[0].res as i32))
         }
     }
+
+    fn punch_hole(&mut self, _offset: u64, _length: u64, _user_data: u64) -> AsyncIoResult<()> {
+        Err(AsyncIoError::PunchHole(std::io::Error::other(
+            "punch_hole not supported with AIO backend",
+        )))
+    }
+
+    fn write_zeroes(&mut self, _offset: u64, _length: u64, _user_data: u64) -> AsyncIoResult<()> {
+        Err(AsyncIoError::WriteZeroes(std::io::Error::other(
+            "write_zeroes not supported with AIO backend",
+        )))
+    }
 }

block/src/raw_sync.rs

@@ -7,12 +7,13 @@ use std::fs::File;
 use std::io::{Seek, SeekFrom};
 use std::os::unix::io::{AsRawFd, RawFd};
 
+use log::warn;
 use vmm_sys_util::eventfd::EventFd;
 
 use crate::async_io::{
     AsyncIo, AsyncIoError, AsyncIoResult, BorrowedDiskFd, DiskFile, DiskFileError, DiskFileResult,
 };
-use crate::DiskTopology;
+use crate::{DiskTopology, probe_sparse_support};
 
 pub struct RawFileDiskSync {
     file: File,
@@ -25,12 +26,19 @@ impl RawFileDiskSync {
 }
 
 impl DiskFile for RawFileDiskSync {
-    fn size(&mut self) -> DiskFileResult<u64> {
+    fn logical_size(&mut self) -> DiskFileResult<u64> {
         self.file
             .seek(SeekFrom::End(0))
             .map_err(DiskFileError::Size)
     }
 
+    fn physical_size(&mut self) -> DiskFileResult<u64> {
+        self.file
+            .metadata()
+            .map(|m| m.len())
+            .map_err(DiskFileError::Size)
+    }
+
     fn new_async_io(&self, _ring_depth: u32) -> DiskFileResult<Box<dyn AsyncIo>> {
         Ok(Box::new(RawFileSync::new(self.file.as_raw_fd())) as Box<dyn AsyncIo>)
     }
@@ -44,6 +52,10 @@ impl DiskFile for RawFileDiskSync {
         }
     }
 
+    fn supports_sparse_operations(&self) -> bool {
+        probe_sparse_support(&self.file)
+    }
+
     fn fd(&mut self) -> BorrowedDiskFd<'_> {
         BorrowedDiskFd::new(self.file.as_raw_fd())
     }
@@ -138,4 +150,226 @@ impl AsyncIo for RawFileSync {
     fn next_completed_request(&mut self) -> Option<(u64, i32)> {
         self.completion_list.pop_front()
     }
+
+    fn punch_hole(&mut self, offset: u64, length: u64, user_data: u64) -> AsyncIoResult<()> {
+        const FALLOC_FL_PUNCH_HOLE: i32 = 0x02;
+        const FALLOC_FL_KEEP_SIZE: i32 = 0x01;
+        let mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
+
+        // SAFETY: FFI call with valid arguments
+        let result = unsafe {
+            libc::fallocate(
+                self.fd as libc::c_int,
+                mode,
+                offset as libc::off_t,
+                length as libc::off_t,
+            )
+        };
+        if result < 0 {
+            return Err(AsyncIoError::PunchHole(std::io::Error::last_os_error()));
+        }
+
+        self.completion_list.push_back((user_data, result));
+        self.eventfd.write(1).unwrap();
+
+        Ok(())
+    }
+
+    fn write_zeroes(&mut self, offset: u64, length: u64, user_data: u64) -> AsyncIoResult<()> {
+        const FALLOC_FL_ZERO_RANGE: i32 = 0x10;
+        const FALLOC_FL_KEEP_SIZE: i32 = 0x01;
+        let mode = FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE;
+
+        // SAFETY: FFI call with valid arguments
+        let result = unsafe {
+            libc::fallocate(
+                self.fd as libc::c_int,
+                mode,
+                offset as libc::off_t,
+                length as libc::off_t,
+            )
+        };
+        if result < 0 {
+            return Err(AsyncIoError::WriteZeroes(std::io::Error::last_os_error()));
+        }
+
+        self.completion_list.push_back((user_data, result));
+        self.eventfd.write(1).unwrap();
+
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod unit_tests {
+    use std::io::{Read, Seek, SeekFrom, Write};
+
+    use vmm_sys_util::tempfile::TempFile;
+
+    use super::*;
+
+    #[test]
+    fn test_punch_hole() {
+        let temp_file = TempFile::new().unwrap();
+        let mut file = temp_file.into_file();
+
+        // Write 4MB of data
+        let data = vec![0xAA; 4 * 1024 * 1024];
+        file.write_all(&data).unwrap();
+        file.sync_all().unwrap();
+
+        // Create async IO instance
+        let mut async_io = RawFileSync::new(file.as_raw_fd());
+
+        // Punch hole in the middle (1MB at offset 1MB)
+        let offset = 1024 * 1024;
+        let length = 1024 * 1024;
+        async_io.punch_hole(offset, length, 1).unwrap();
+
+        // Check completion
+        let (user_data, result) = async_io.next_completed_request().unwrap();
+        assert_eq!(user_data, 1);
+        assert_eq!(result, 0);
+
+        // Verify the hole reads as zeros
+        file.seek(SeekFrom::Start(offset)).unwrap();
+        let mut read_buf = vec![0; length as usize];
+        file.read_exact(&mut read_buf).unwrap();
+        assert!(
+            read_buf.iter().all(|&b| b == 0),
+            "Punched hole should read as zeros"
+        );
+
+        // Verify data before hole is intact
+        file.seek(SeekFrom::Start(0)).unwrap();
+        let mut read_buf = vec![0; 1024];
+        file.read_exact(&mut read_buf).unwrap();
+        assert!(
+            read_buf.iter().all(|&b| b == 0xAA),
+            "Data before hole should be intact"
+        );
+
+        // Verify data after hole is intact
+        file.seek(SeekFrom::Start(offset + length)).unwrap();
+        let mut read_buf = vec![0; 1024];
+        file.read_exact(&mut read_buf).unwrap();
+        assert!(
+            read_buf.iter().all(|&b| b == 0xAA),
+            "Data after hole should be intact"
+        );
+    }
+
+    #[test]
+    fn test_write_zeroes() {
+        let temp_file = TempFile::new().unwrap();
+        let mut file = temp_file.into_file();
+
+        // Write 4MB of data
+        let data = vec![0xBB; 4 * 1024 * 1024];
+        file.write_all(&data).unwrap();
+        file.sync_all().unwrap();
+
+        // Create async IO instance
+        let mut async_io = RawFileSync::new(file.as_raw_fd());
+
+        // Write zeros in the middle (512KB at offset 2MB)
+        let offset = 2 * 1024 * 1024;
+        let length = 512 * 1024;
+        let write_zeroes_result = async_io.write_zeroes(offset, length, 2);
+
+        // FALLOC_FL_ZERO_RANGE might not be supported on all filesystems (e.g., tmpfs)
+        // If it fails with ENOTSUP, skip the test
+        if let Err(AsyncIoError::WriteZeroes(ref e)) = write_zeroes_result
+            && (e.raw_os_error() == Some(libc::EOPNOTSUPP)
+                || e.raw_os_error() == Some(libc::ENOTSUP))
+        {
+            eprintln!(
+                "Skipping test_write_zeroes: filesystem doesn't support FALLOC_FL_ZERO_RANGE"
+            );
+            return;
+        }
+        write_zeroes_result.unwrap();
+
+        // Check completion
+        let (user_data, result) = async_io.next_completed_request().unwrap();
+        assert_eq!(user_data, 2);
+        assert_eq!(result, 0);
+
+        // Verify the zeroed region reads as zeros
+        file.seek(SeekFrom::Start(offset)).unwrap();
+        let mut read_buf = vec![0; length as usize];
+        file.read_exact(&mut read_buf).unwrap();
+        assert!(
+            read_buf.iter().all(|&b| b == 0),
+            "Zeroed region should read as zeros"
+        );
+
+        // Verify data before zeroed region is intact
+        file.seek(SeekFrom::Start(offset - 1024)).unwrap();
+        let mut read_buf = vec![0; 1024];
+        file.read_exact(&mut read_buf).unwrap();
+        assert!(
+            read_buf.iter().all(|&b| b == 0xBB),
+            "Data before zeroed region should be intact"
+        );
+
+        // Verify data after zeroed region is intact
+        file.seek(SeekFrom::Start(offset + length)).unwrap();
+        let mut read_buf = vec![0; 1024];
+        file.read_exact(&mut read_buf).unwrap();
+        assert!(
+            read_buf.iter().all(|&b| b == 0xBB),
+            "Data after zeroed region should be intact"
+        );
+    }
+
+    #[test]
+    fn test_punch_hole_multiple_operations() {
+        let temp_file = TempFile::new().unwrap();
+        let mut file = temp_file.into_file();
+
+        // Write 8MB of data
+        let data = vec![0xCC; 8 * 1024 * 1024];
+        file.write_all(&data).unwrap();
+        file.sync_all().unwrap();
+
+        // Create async IO instance
+        let mut async_io = RawFileSync::new(file.as_raw_fd());
+
+        // Punch multiple holes
+        async_io.punch_hole(1024 * 1024, 512 * 1024, 10).unwrap();
+        async_io
+            .punch_hole(3 * 1024 * 1024, 512 * 1024, 11)
+            .unwrap();
+        async_io
+            .punch_hole(5 * 1024 * 1024, 512 * 1024, 12)
+            .unwrap();
+
+        // Check all completions
+        let (user_data, result) = async_io.next_completed_request().unwrap();
+        assert_eq!(user_data, 10);
+        assert_eq!(result, 0);
+
+        let (user_data, result) = async_io.next_completed_request().unwrap();
+        assert_eq!(user_data, 11);
+        assert_eq!(result, 0);
+
+        let (user_data, result) = async_io.next_completed_request().unwrap();
+        assert_eq!(user_data, 12);
+        assert_eq!(result, 0);
+
+        // Verify all holes read as zeros
+        file.seek(SeekFrom::Start(1024 * 1024)).unwrap();
+        let mut read_buf = vec![0; 512 * 1024];
+        file.read_exact(&mut read_buf).unwrap();
+        assert!(read_buf.iter().all(|&b| b == 0));
+
+        file.seek(SeekFrom::Start(3 * 1024 * 1024)).unwrap();
+        file.read_exact(&mut read_buf).unwrap();
+        assert!(read_buf.iter().all(|&b| b == 0));
+
+        file.seek(SeekFrom::Start(5 * 1024 * 1024)).unwrap();
+        file.read_exact(&mut read_buf).unwrap();
+        assert!(read_buf.iter().all(|&b| b == 0));
+    }
 }

block/src/vhd.rs

@@ -5,7 +5,7 @@
 use std::fs::File;
 use std::io::{Seek, SeekFrom};
 
-use crate::{read_aligned_block_size, DiskTopology};
+use crate::{DiskTopology, read_aligned_block_size};
 
 #[derive(Clone, Copy)]
 pub struct VhdFooter {
@@ -117,13 +117,13 @@ pub fn is_fixed_vhd(f: &mut File) -> std::io::Result<bool> {
 }
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use std::fs::File;
     use std::io::{Seek, SeekFrom, Write};
 
     use vmm_sys_util::tempfile::TempFile;
 
-    use super::{is_fixed_vhd, VhdFooter};
+    use super::{VhdFooter, is_fixed_vhd};
 
     fn valid_fixed_vhd_footer() -> Vec<u8> {
         vec![

block/src/vhdx/mod.rs

@@ -12,11 +12,11 @@ use remain::sorted;
 use thiserror::Error;
 use uuid::Uuid;
 
+use crate::BlockBackend;
 use crate::vhdx::vhdx_bat::{BatEntry, VhdxBatError};
 use crate::vhdx::vhdx_header::{RegionInfo, RegionTableEntry, VhdxHeader, VhdxHeaderError};
 use crate::vhdx::vhdx_io::VhdxIoError;
 use crate::vhdx::vhdx_metadata::{DiskSpec, VhdxMetadataError};
-use crate::BlockBackend;
 
 mod vhdx_bat;
 mod vhdx_header;
@@ -187,11 +187,11 @@ impl Seek for Vhdx {
             }
         };
 
-        if let Some(o) = new_offset {
-            if o <= self.virtual_disk_size() {
-                self.current_offset = o;
-                return Ok(o);
-            }
+        if let Some(o) = new_offset
+            && o <= self.virtual_disk_size()
+        {
+            self.current_offset = o;
+            return Ok(o);
         }
 
         Err(std::io::Error::new(
@@ -202,9 +202,16 @@ impl Seek for Vhdx {
 }
 
 impl BlockBackend for Vhdx {
-    fn size(&self) -> std::result::Result<u64, crate::Error> {
+    fn logical_size(&self) -> std::result::Result<u64, crate::Error> {
         Ok(self.virtual_disk_size())
     }
+
+    fn physical_size(&self) -> std::result::Result<u64, crate::Error> {
+        self.file
+            .metadata()
+            .map(|m| m.len())
+            .map_err(crate::Error::GetFileMetadata)
+    }
 }
 
 impl Clone for Vhdx {

block/src/vhdx/vhdx_header.rs

@@ -2,8 +2,6 @@
 //
 // SPDX-License-Identifier: Apache-2.0
 
-extern crate log;
-
 use std::collections::btree_map::BTreeMap;
 use std::fs::File;
 use std::io::{self, Read, Seek, SeekFrom, Write};

block/src/vhdx/vhdx_io.rs

@@ -35,9 +35,7 @@ pub enum VhdxIoError {
 pub type Result<T> = std::result::Result<T, VhdxIoError>;
 
 macro_rules! align {
-    ($n:expr, $align:expr) => {{
-        $n.div_ceil($align) * $align
-    }};
+    ($n:expr, $align:expr) => {{ $n.div_ceil($align) * $align }};
 }
 
 #[derive(Default)]
@@ -130,7 +128,7 @@ pub fn read(
             _ => {
                 return Err(VhdxIoError::InvalidBatEntryState);
             }
-        };
+        }
         sector_count -= sector.free_sectors;
         sector_index += sector.free_sectors;
         read_count += sector.free_bytes as usize;
@@ -212,7 +210,7 @@ pub fn write(
             _ => {
                 return Err(VhdxIoError::InvalidBatEntryState);
             }
-        };
+        }
         sector_count -= sector.free_sectors;
         sector_index += sector.free_sectors;
         write_count += sector.free_bytes as usize;

block/src/vhdx_sync.rs

@@ -5,31 +5,40 @@
 use std::collections::VecDeque;
 use std::fs::File;
 use std::os::fd::AsRawFd;
-use std::sync::{Arc, Mutex, MutexGuard};
 
 use vmm_sys_util::eventfd::EventFd;
 
 use crate::async_io::{
-    AsyncIo, AsyncIoResult, BorrowedDiskFd, DiskFile, DiskFileError, DiskFileResult,
+    AsyncIo, AsyncIoError, AsyncIoResult, BorrowedDiskFd, DiskFile, DiskFileError, DiskFileResult,
 };
 use crate::vhdx::{Result as VhdxResult, Vhdx};
-use crate::AsyncAdaptor;
+use crate::{AsyncAdaptor, BlockBackend, Error};
 
 pub struct VhdxDiskSync {
-    vhdx_file: Arc<Mutex<Vhdx>>,
+    vhdx_file: Vhdx,
 }
 
 impl VhdxDiskSync {
     pub fn new(f: File) -> VhdxResult<Self> {
         Ok(VhdxDiskSync {
-            vhdx_file: Arc::new(Mutex::new(Vhdx::new(f)?)),
+            vhdx_file: Vhdx::new(f)?,
         })
     }
 }
 
 impl DiskFile for VhdxDiskSync {
-    fn size(&mut self) -> DiskFileResult<u64> {
-        Ok(self.vhdx_file.lock().unwrap().virtual_disk_size())
+    fn logical_size(&mut self) -> DiskFileResult<u64> {
+        Ok(self.vhdx_file.virtual_disk_size())
+    }
+
+    fn physical_size(&mut self) -> DiskFileResult<u64> {
+        self.vhdx_file.physical_size().map_err(|e| {
+            let io_inner = match e {
+                Error::GetFileMetadata(e) => e,
+                _ => unreachable!(),
+            };
+            DiskFileError::Size(io_inner)
+        })
     }
 
     fn new_async_io(&self, _ring_depth: u32) -> DiskFileResult<Box<dyn AsyncIo>> {
@@ -40,19 +49,18 @@ impl DiskFile for VhdxDiskSync {
     }
 
     fn fd(&mut self) -> BorrowedDiskFd<'_> {
-        let lock = self.vhdx_file.lock().unwrap();
-        BorrowedDiskFd::new(lock.as_raw_fd())
+        BorrowedDiskFd::new(self.vhdx_file.as_raw_fd())
     }
 }
 
 pub struct VhdxSync {
-    vhdx_file: Arc<Mutex<Vhdx>>,
+    vhdx_file: Vhdx,
     eventfd: EventFd,
     completion_list: VecDeque<(u64, i32)>,
 }
 
 impl VhdxSync {
-    pub fn new(vhdx_file: Arc<Mutex<Vhdx>>) -> std::io::Result<Self> {
+    pub fn new(vhdx_file: Vhdx) -> std::io::Result<Self> {
         Ok(VhdxSync {
             vhdx_file,
             eventfd: EventFd::new(libc::EFD_NONBLOCK)?,
@@ -61,11 +69,7 @@ impl VhdxSync {
     }
 }
 
-impl AsyncAdaptor<Vhdx> for Arc<Mutex<Vhdx>> {
-    fn file(&mut self) -> MutexGuard<'_, Vhdx> {
-        self.lock().unwrap()
-    }
-}
+impl AsyncAdaptor for Vhdx {}
 
 impl AsyncIo for VhdxSync {
     fn notifier(&self) -> &EventFd {
@@ -110,4 +114,16 @@ impl AsyncIo for VhdxSync {
     fn next_completed_request(&mut self) -> Option<(u64, i32)> {
         self.completion_list.pop_front()
     }
+
+    fn punch_hole(&mut self, _offset: u64, _length: u64, _user_data: u64) -> AsyncIoResult<()> {
+        Err(AsyncIoError::PunchHole(std::io::Error::other(
+            "punch_hole not supported for VHDX",
+        )))
+    }
+
+    fn write_zeroes(&mut self, _offset: u64, _length: u64, _user_data: u64) -> AsyncIoResult<()> {
+        Err(AsyncIoError::WriteZeroes(std::io::Error::other(
+            "write_zeroes not supported for VHDX",
+        )))
+    }
 }

cloud-hypervisor/Cargo.toml

@@ -0,0 +1,70 @@
+[package]
+authors = ["The Cloud Hypervisor Authors"]
+build = "build.rs"
+default-run = "cloud-hypervisor"
+description = "Open source Virtual Machine Monitor (VMM) that runs on top of KVM & MSHV"
+edition = "2024"
+homepage = "https://github.com/cloud-hypervisor/cloud-hypervisor"
+license = "Apache-2.0 AND BSD-3-Clause"
+name = "cloud-hypervisor"
+version = "51.0.0"
+# Minimum buildable version:
+# Keep in sync with version in .github/workflows/build.yaml
+# Policy on MSRV (see #4318):
+# Can only be bumped if satisfying any of the following:
+# a.) A dependency requires it,
+# b.) If we want to use a new feature and that MSRV is at least 6 months old,
+# c.) There is a security issue that is addressed by the toolchain update.
+rust-version = "1.89.0"
+
+[dependencies]
+anyhow = { workspace = true }
+api_client = { path = "../api_client" }
+clap = { workspace = true, features = ["string"] }
+dhat = { workspace = true, optional = true }
+env_logger = { workspace = true }
+epoll = { workspace = true }
+event_monitor = { path = "../event_monitor" }
+hypervisor = { path = "../hypervisor" }
+libc = { workspace = true }
+log = { workspace = true, features = ["std"] }
+option_parser = { path = "../option_parser" }
+seccompiler = { workspace = true }
+serde_json = { workspace = true }
+signal-hook = { workspace = true }
+thiserror = { workspace = true }
+tpm = { path = "../tpm" }
+tracer = { path = "../tracer" }
+vm-memory = { workspace = true }
+vmm = { path = "../vmm" }
+vmm-sys-util = { workspace = true }
+zbus = { version = "5.13.2", optional = true }
+
+[dev-dependencies]
+block = { path = "../block" }
+dirs = { workspace = true }
+net_util = { path = "../net_util" }
+serde_json = { workspace = true }
+test_infra = { path = "../test_infra" }
+wait-timeout = { workspace = true }
+
+# Please adjust `vmm::feature_list()` accordingly when changing the
+# feature list below
+[features]
+dbus_api = ["vmm/dbus_api", "zbus"]
+default = ["io_uring", "kvm"]
+dhat-heap = ["dhat", "vmm/dhat-heap"]       # For heap profiling
+fw_cfg = ["vmm/fw_cfg"]
+guest_debug = ["vmm/guest_debug"]
+igvm = ["mshv", "vmm/igvm"]
+io_uring = ["vmm/io_uring"]
+ivshmem = ["vmm/ivshmem"]
+kvm = ["vmm/kvm"]
+mshv = ["vmm/mshv"]
+pvmemcontrol = ["vmm/pvmemcontrol"]
+sev_snp = ["igvm", "mshv", "vmm/sev_snp"]
+tdx = ["vmm/tdx"]
+tracing = ["tracer/tracing", "vmm/tracing"]
+
+[lints]
+workspace = true

cloud-hypervisor/build.rs

@@ -9,14 +9,13 @@ use std::process::Command;
 fn main() {
     let mut version = "v".to_owned() + env!("CARGO_PKG_VERSION");
 
-    if let Ok(git_out) = Command::new("git").args(["describe", "--dirty"]).output() {
-        if git_out.status.success() {
-            if let Ok(git_out_str) = String::from_utf8(git_out.stdout) {
-                version = git_out_str;
-                // Pop the trailing newline.
-                version.pop();
-            }
-        }
+    if let Ok(git_out) = Command::new("git").args(["describe", "--dirty"]).output()
+        && git_out.status.success()
+        && let Ok(git_out_str) = String::from_utf8(git_out.stdout)
+    {
+        version = git_out_str;
+        // Pop the trailing newline.
+        version.pop();
     }
 
     // Append CH_EXTRA_VERSION to version if it is set.

cloud-hypervisor/src/bin/ch-remote.rs

@@ -13,8 +13,8 @@ use std::os::unix::net::UnixStream;
 use std::process;
 
 use api_client::{
-    simple_api_command, simple_api_command_with_fds, simple_api_full_command,
-    Error as ApiClientError,
+    Error as ApiClientError, simple_api_command, simple_api_command_with_fds,
+    simple_api_full_command,
 };
 use clap::{Arg, ArgAction, ArgMatches, Command};
 use log::error;
@@ -65,6 +65,8 @@ enum Error {
     ReadingStdin(#[source] std::io::Error),
     #[error("Error reading from file")]
     ReadingFile(#[source] std::io::Error),
+    #[error("Invalid disk size")]
+    InvalidDiskSize(#[source] ByteSizedParseError),
 }
 
 enum TargetApi<'a> {
@@ -320,6 +322,22 @@ fn rest_api_do_command(matches: &ArgMatches, socket: &mut UnixStream) -> ApiResu
             )?;
             simple_api_command(socket, "PUT", "resize", Some(&resize)).map_err(Error::HttpApiClient)
         }
+        Some("resize-disk") => {
+            let resize_disk = resize_disk_config(
+                matches
+                    .subcommand_matches("resize-disk")
+                    .unwrap()
+                    .get_one::<String>("disk")
+                    .unwrap(),
+                matches
+                    .subcommand_matches("resize-disk")
+                    .unwrap()
+                    .get_one::<String>("size")
+                    .unwrap(),
+            )?;
+            simple_api_command(socket, "PUT", "resize-disk", Some(&resize_disk))
+                .map_err(Error::HttpApiClient)
+        }
         Some("resize-zone") => {
             let resize_zone = resize_zone_config(
                 matches
@@ -399,7 +417,7 @@ fn rest_api_do_command(matches: &ArgMatches, socket: &mut UnixStream) -> ApiResu
                     .get_one::<String>("net_config")
                     .unwrap(),
             )?;
-            simple_api_command_with_fds(socket, "PUT", "add-net", Some(&net_config), fds)
+            simple_api_command_with_fds(socket, "PUT", "add-net", Some(&net_config), &fds)
                 .map_err(Error::HttpApiClient)
         }
         Some("add-user-device") => {
@@ -454,7 +472,7 @@ fn rest_api_do_command(matches: &ArgMatches, socket: &mut UnixStream) -> ApiResu
                     .get_one::<String>("restore_config")
                     .unwrap(),
             )?;
-            simple_api_command_with_fds(socket, "PUT", "restore", Some(&restore_config), fds)
+            simple_api_command_with_fds(socket, "PUT", "restore", Some(&restore_config), &fds)
                 .map_err(Error::HttpApiClient)
         }
         Some("coredump") => {
@@ -725,7 +743,7 @@ fn resize_config(
     memory: Option<&str>,
     balloon: Option<&str>,
 ) -> Result<String, Error> {
-    let desired_vcpus: Option<u8> = if let Some(cpus) = cpus {
+    let desired_vcpus: Option<u32> = if let Some(cpus) = cpus {
         Some(cpus.parse().map_err(Error::InvalidCpuCount)?)
     } else {
         None
@@ -762,6 +780,15 @@ fn resize_config(
     Ok(serde_json::to_string(&resize).unwrap())
 }
 
+fn resize_disk_config(id: &str, size: &str) -> Result<String, Error> {
+    let resize_disk = vmm::api::VmResizeDiskData {
+        id: id.to_owned(),
+        desired_size: size.parse::<ByteSized>().map_err(Error::InvalidDiskSize)?.0,
+    };
+
+    Ok(serde_json::to_string(&resize_disk).unwrap())
+}
+
 fn resize_zone_config(id: &str, size: &str) -> Result<String, Error> {
     let resize_zone = vmm::api::VmResizeZoneData {
         id: id.to_owned(),
@@ -1022,6 +1049,20 @@ fn get_cli_commands_sorted() -> Box<[Command]> {
                     .help("New memory size in bytes (supports K/M/G suffix)")
                     .num_args(1),
             ),
+        Command::new("resize-disk")
+            .about("Resize an attached disk")
+            .arg(
+                Arg::new("disk")
+                    .long("disk")
+                    .help("Disk identifier")
+                    .num_args(1),
+            )
+            .arg(
+                Arg::new("size")
+                    .long("size")
+                    .help("New disk size")
+                    .num_args(1),
+            ),
         Command::new("resize-zone")
             .about("Resize a memory zone")
             .arg(
@@ -1128,13 +1169,16 @@ fn main() {
             process::exit(1);
         }
         _ => {
-            error!("Please either provide the api-socket option or dbus-service-name and dbus-object-path options");
+            error!(
+                "Please either provide the api-socket option or dbus-service-name and dbus-object-path options"
+            );
             process::exit(1);
         }
     };
 
     if let Err(top_error) = target_api.do_command(&matches) {
         // Helper to join strings with a newline.
+        #[allow(clippy::needless_pass_by_value)]
         fn join_strs(mut acc: String, next: String) -> String {
             if !acc.is_empty() {
                 acc.push('\n');
@@ -1159,7 +1203,7 @@ fn main() {
             if let Some(api_client::Error::ServerResponse(status_code, body)) =
                 error.downcast_ref::<api_client::Error>()
             {
-                let body = body.as_ref().map(|body| body.as_str()).unwrap_or("");
+                let body = body.as_ref().map_or("", |body| body.as_str());
 
                 // Retrieve the list of error messages back.
                 let lines: Vec<&str> = match serde_json::from_str(body) {
@@ -1199,11 +1243,11 @@ fn main() {
             server_api_error_display_modifier,
         );
         process::exit(1)
-    };
+    }
 }
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use std::cmp::Ordering;
 
     use super::*;

cloud-hypervisor/src/lib.rs

@@ -19,12 +19,15 @@ pub fn cli_print_error_chain<'a>(
         &'a (dyn Error + 'static),
     ) -> Option<String>,
 ) {
-    let msg = format!("Error: {component} exited with the following");
+    // Debug info.
+    error!("Fatal error: {top_error:?}");
+
+    eprint!("Error: {component} exited with the following ");
     if top_error.source().is_none() {
-        error!("{msg} error:");
-        error!("  {top_error}");
+        eprintln!("error:");
+        eprintln!("  {top_error}");
     } else {
-        error!("{msg} chain of errors:");
+        eprintln!("chain of errors:");
         std::iter::successors(Some(top_error), |sub_error| {
             // Dereference necessary to mitigate rustc compiler bug.
             // See <https://github.com/rust-lang/rust/issues/141673>
@@ -34,13 +37,10 @@ pub fn cli_print_error_chain<'a>(
         .for_each(|(level, error)| {
             // Special case: handling of HTTP Server responses in ch-remote
             if let Some(message) = display_modifier(level, 2, error) {
-                error!("{message}");
+                eprintln!("{message}");
             } else {
-                error!("  {level}: {error}");
+                eprintln!("  {level}: {error}");
             }
         });
     }
-
-    error!("");
-    error!("Debug Info: {top_error:?}");
 }

cloud-hypervisor/src/main.rs

@@ -8,27 +8,29 @@ mod test_util;
 
 use std::fs::File;
 use std::os::unix::io::{AsRawFd, FromRawFd, RawFd};
-use std::sync::mpsc::channel;
 use std::sync::Mutex;
+use std::sync::mpsc::channel;
 use std::{env, io};
 
 use clap::{Arg, ArgAction, ArgGroup, ArgMatches, Command};
 use event_monitor::event;
 use libc::EFD_NONBLOCK;
-use log::{error, warn, LevelFilter};
+use log::{LevelFilter, error, info, warn};
 use option_parser::OptionParser;
 use seccompiler::SeccompAction;
 use signal_hook::consts::SIGSYS;
 use thiserror::Error;
-#[cfg(feature = "dbus_api")]
-use vmm::api::dbus::{dbus_api_graceful_shutdown, DBusApiOptions};
-use vmm::api::http::http_api_graceful_shutdown;
 use vmm::api::ApiAction;
+#[cfg(feature = "dbus_api")]
+use vmm::api::dbus::{DBusApiOptions, dbus_api_graceful_shutdown};
+use vmm::api::http::http_api_graceful_shutdown;
 use vmm::config::{RestoreConfig, VmParams};
 use vmm::landlock::{Landlock, LandlockError};
 use vmm::vm_config;
-#[cfg(target_arch = "x86_64")]
-use vmm::vm_config::SgxEpcConfig;
+#[cfg(feature = "fw_cfg")]
+use vmm::vm_config::FwCfgConfig;
+#[cfg(feature = "ivshmem")]
+use vmm::vm_config::IvshmemConfig;
 use vmm::vm_config::{
     BalloonConfig, DeviceConfig, DiskConfig, FsConfig, LandlockConfig, NetConfig, NumaConfig,
     PciSegmentConfig, PmemConfig, RateLimiterGroupConfig, TpmConfig, UserDeviceConfig, VdpaConfig,
@@ -131,29 +133,25 @@ impl log::Log for Logger {
 
         let now = std::time::Instant::now();
         let duration = now.duration_since(self.start);
+        let duration_s = duration.as_secs_f32();
 
-        if record.file().is_some() && record.line().is_some() {
-            write!(
-                *(*(self.output.lock().unwrap())),
-                "cloud-hypervisor: {:.6?}: <{}> {}:{}:{} -- {}\r\n",
-                duration,
-                std::thread::current().name().unwrap_or("anonymous"),
-                record.level(),
-                record.file().unwrap(),
-                record.line().unwrap(),
-                record.args()
-            )
+        let location = if let (Some(file), Some(line)) = (record.file(), record.line()) {
+            format!("{file}:{line}")
         } else {
-            write!(
-                *(*(self.output.lock().unwrap())),
-                "cloud-hypervisor: {:.6?}: <{}> {}:{} -- {}\r\n",
-                duration,
-                std::thread::current().name().unwrap_or("anonymous"),
-                record.level(),
-                record.target(),
-                record.args()
-            )
-        }
+            record.target().to_string()
+        };
+
+        let mut out = self.output.lock().unwrap();
+        write!(
+            &mut *out,
+            // 10: 6 decimal places + sep => whole seconds in range `0..=999` properly aligned
+            "cloud-hypervisor: {:>10.6?}s: <{}> {}:{} -- {}\r\n",
+            duration_s,
+            std::thread::current().name().unwrap_or("anonymous"),
+            record.level(),
+            location,
+            record.args(),
+        )
         .ok();
     }
     fn flush(&self) {}
@@ -215,22 +213,11 @@ fn get_cli_options_sorted(
                     topology=<threads_per_core>:<cores_per_die>:<dies_per_package>:<packages>,\
                     kvm_hyperv=on|off,max_phys_bits=<maximum_number_of_physical_bits>,\
                     affinity=<list_of_vcpus_with_their_associated_cpuset>,\
-                    features=<list_of_features_to_enable>",
+                    features=<list_of_features_to_enable>,\
+                    nested=on|off",
             )
             .default_value(default_vcpus)
             .group("vm-config"),
-        #[cfg(target_arch = "x86_64")]
-        Arg::new("debug-console")
-            .long("debug-console")
-            .help("Debug console: off|pty|tty|file=</path/to/a/file>,iobase=<port in hex>")
-            .default_value("off,iobase=0xe9")
-            .group("vm-config"),
-        #[cfg(feature = "dbus_api")]
-        Arg::new("dbus-service-name")
-            .long("dbus-service-name")
-            .help("Well known name of the device")
-            .num_args(1)
-            .group("vmm-config"),
         #[cfg(feature = "dbus_api")]
         Arg::new("dbus-object-path")
             .long("dbus-object-path")
@@ -238,12 +225,24 @@ fn get_cli_options_sorted(
             .num_args(1)
             .group("vmm-config"),
         #[cfg(feature = "dbus_api")]
+        Arg::new("dbus-service-name")
+            .long("dbus-service-name")
+            .help("Well known name of the device")
+            .num_args(1)
+            .group("vmm-config"),
+        #[cfg(feature = "dbus_api")]
         Arg::new("dbus-system-bus")
             .long("dbus-system-bus")
             .action(ArgAction::SetTrue)
             .help("Use the system bus instead of a session bus")
             .num_args(0)
             .group("vmm-config"),
+        #[cfg(target_arch = "x86_64")]
+        Arg::new("debug-console")
+            .long("debug-console")
+            .help("Debug console: off|pty|tty|file=</path/to/a/file>,iobase=<port in hex>")
+            .default_value("off,iobase=0xe9")
+            .group("vm-config"),
         Arg::new("device")
             .long("device")
             .help(DeviceConfig::SYNTAX)
@@ -269,6 +268,12 @@ fn get_cli_options_sorted(
             .help(FsConfig::SYNTAX)
             .num_args(1..)
             .group("vm-config"),
+        #[cfg(feature = "fw_cfg")]
+        Arg::new("fw-cfg-config")
+            .long("fw-cfg-config")
+            .help(FwCfgConfig::SYNTAX)
+            .num_args(1)
+            .group("vm-payload"),
         #[cfg(feature = "guest_debug")]
         Arg::new("gdb")
             .long("gdb")
@@ -292,6 +297,12 @@ fn get_cli_options_sorted(
             .help("Path to initramfs image")
             .num_args(1)
             .group("vm-config"),
+        #[cfg(feature = "ivshmem")]
+        Arg::new("ivshmem")
+            .long("ivshmem")
+            .help(IvshmemConfig::SYNTAX)
+            .num_args(1)
+            .group("vm-config"),
         Arg::new("kernel")
             .long("kernel")
             .help(
@@ -413,12 +424,6 @@ fn get_cli_options_sorted(
             .help("Control serial port: off|null|pty|tty|file=</path/to/a/file>|socket=</path/to/a/file>")
             .default_value("null")
             .group("vm-config"),
-        #[cfg(target_arch = "x86_64")]
-        Arg::new("sgx-epc")
-            .long("sgx-epc")
-            .help(SgxEpcConfig::SYNTAX)
-            .num_args(1..)
-            .group("vm-config"),
         Arg::new("tpm")
             .long("tpm")
             .num_args(1)
@@ -482,7 +487,7 @@ fn create_app(default_vcpus: String, default_memory: String, default_rng: String
         .args(args)
 }
 
-fn start_vmm(cmd_arguments: ArgMatches) -> Result<Option<String>, Error> {
+fn start_vmm(cmd_arguments: &ArgMatches) -> Result<Option<String>, Error> {
     let log_level = match cmd_arguments.get_count("v") {
         0 => LevelFilter::Warn,
         1 => LevelFilter::Info,
@@ -553,10 +558,10 @@ fn start_vmm(cmd_arguments: ArgMatches) -> Result<Option<String>, Error> {
         // handler safe functions (writing to stderr) and manipulating signals.
         unsafe {
             signal_hook::low_level::register(signal_hook::consts::SIGSYS, || {
-                eprint!(
+                eprintln!(
                     "\n==== Possible seccomp violation ====\n\
                 Try running with `strace -ff` to identify the cause and open an issue: \
-                https://github.com/cloud-hypervisor/cloud-hypervisor/issues/new\n"
+                https://github.com/cloud-hypervisor/cloud-hypervisor/issues/new"
                 );
                 signal_hook::low_level::emulate_default_handler(SIGSYS).unwrap();
             })
@@ -718,7 +723,7 @@ fn start_vmm(cmd_arguments: ArgMatches) -> Result<Option<String>, Error> {
             cmd_arguments.contains_id("kernel") || cmd_arguments.contains_id("firmware");
 
         if payload_present {
-            let vm_params = VmParams::from_arg_matches(&cmd_arguments);
+            let vm_params = VmParams::from_arg_matches(cmd_arguments);
             let vm_config = VmConfig::parse(vm_params).map_err(Error::ParsingConfig)?;
 
             // Create and boot the VM based off the VM config we just built.
@@ -746,10 +751,10 @@ fn start_vmm(cmd_arguments: ArgMatches) -> Result<Option<String>, Error> {
         Ok(())
     })();
 
-    if r.is_err() {
-        if let Err(e) = exit_evt.write(1) {
-            warn!("writing to exit EventFd: {e}");
-        }
+    if r.is_err()
+        && let Err(e) = exit_evt.write(1)
+    {
+        warn!("writing to exit EventFd: {e}");
     }
 
     if landlock_enable {
@@ -766,7 +771,7 @@ fn start_vmm(cmd_arguments: ArgMatches) -> Result<Option<String>, Error> {
         .map_err(Error::VmmThread)?;
 
     if let Some(api_handle) = vmm_thread_handle.http_api_handle {
-        http_api_graceful_shutdown(api_handle).map_err(Error::HttpApiShutdown)?
+        http_api_graceful_shutdown(api_handle).map_err(Error::HttpApiShutdown)?;
     }
 
     #[cfg(feature = "dbus_api")]
@@ -853,6 +858,8 @@ fn main() {
     compile_error!("Feature 'tdx' and 'sev_snp' are mutually exclusive.");
     #[cfg(all(feature = "sev_snp", not(target_arch = "x86_64")))]
     compile_error!("Feature 'sev_snp' needs target 'x86_64'");
+    #[cfg(all(feature = "fw_cfg", target_arch = "riscv64"))]
+    compile_error!("Feature 'fw_cfg' needs targets 'x86_64' or 'aarch64'");
 
     #[cfg(feature = "dhat-heap")]
     let _profiler = dhat::Profiler::new_heap();
@@ -878,9 +885,10 @@ fn main() {
         warn!("Error expanding FD table: {e}");
     }
 
-    let exit_code = match start_vmm(cmd_arguments) {
+    let exit_code = match start_vmm(&cmd_arguments) {
         Ok(path) => {
             path.map(|s| std::fs::remove_file(s).ok());
+            info!("Cloud Hypervisor exited successfully");
             0
         }
         Err(top_error) => {
@@ -954,6 +962,7 @@ mod unit_tests {
                 max_phys_bits: 46,
                 affinity: None,
                 features: CpuFeatures::default(),
+                nested: true,
             },
             memory: MemoryConfig {
                 size: 536_870_912,
@@ -977,6 +986,8 @@ mod unit_tests {
                 igvm: None,
                 #[cfg(feature = "sev_snp")]
                 host_data: None,
+                #[cfg(feature = "fw_cfg")]
+                fw_cfg_config: None,
             }),
             rate_limit_groups: None,
             disks: None,
@@ -1010,8 +1021,6 @@ mod unit_tests {
             #[cfg(feature = "pvmemcontrol")]
             pvmemcontrol: None,
             iommu: false,
-            #[cfg(target_arch = "x86_64")]
-            sgx_epc: None,
             numa: None,
             watchdog: false,
             #[cfg(feature = "guest_debug")]
@@ -1022,6 +1031,8 @@ mod unit_tests {
             preserved_fds: None,
             landlock_enable: false,
             landlock_rules: None,
+            #[cfg(feature = "ivshmem")]
+            ivshmem: None,
         };
 
         assert_eq!(expected_vm_config, result_vm_config);
@@ -1081,8 +1092,7 @@ mod unit_tests {
 
     #[test]
     fn test_valid_vm_config_memory() {
-        vec![
-            (
+        [(
                 vec!["cloud-hypervisor", "--kernel", "/path/to/kernel", "--memory", "size=1073741824"],
                 r#"{
                     "payload": {"kernel": "/path/to/kernel"},
@@ -1137,8 +1147,7 @@ mod unit_tests {
                     "memory": {"size": 1073741824, "hotplug_method": "VirtioMem", "hotplug_size": 1073741824}
                 }"#,
                 true,
-            ),
-        ]
+            )]
         .iter()
         .for_each(|(cli, openapi, equal)| {
             compare_vm_config_cli_vs_json(cli, openapi, *equal);
@@ -1190,14 +1199,14 @@ mod unit_tests {
                     "--kernel",
                     "/path/to/kernel",
                     "--disk",
-                    "path=/path/to/disk/1",
+                    "path=/path/to/disk/1,image_type=raw",
                     "path=/path/to/disk/2",
                 ],
                 r#"{
                     "payload": {"kernel": "/path/to/kernel"},
                     "disks": [
-                        {"path": "/path/to/disk/1"},
-                        {"path": "/path/to/disk/2"}
+                        {"path": "/path/to/disk/1", "image_type": "Raw"},
+                        {"path": "/path/to/disk/2", "image_type": "Unknown"}
                     ]
                 }"#,
                 true,
@@ -1208,8 +1217,8 @@ mod unit_tests {
                     "--kernel",
                     "/path/to/kernel",
                     "--disk",
-                    "path=/path/to/disk/1",
-                    "path=/path/to/disk/2",
+                    "path=/path/to/disk/1,image_type=raw",
+                    "path=/path/to/disk/2,image_type=qcow2",
                 ],
                 r#"{
                     "payload": {"kernel": "/path/to/kernel"},
@@ -1271,8 +1280,8 @@ mod unit_tests {
                 r#"{
                     "payload": {"kernel": "/path/to/kernel"},
                     "disks": [
-                        {"path": "/path/to/disk/1", "rate_limit_group": "group0"},
-                        {"path": "/path/to/disk/2", "rate_limit_group": "group0"}
+                        {"path": "/path/to/disk/1", "rate_limit_group": "group0", "image_type": "Unknown"},
+                        {"path": "/path/to/disk/2", "rate_limit_group": "group0", "image_type": "Unknown"}
                     ],
                     "rate_limit_groups": [
                         {"id": "group0", "rate_limiter_config": {"bandwidth": {"size": 1000, "one_time_burst": 0, "refill_time": 100}}}
@@ -1289,7 +1298,7 @@ mod unit_tests {
 
     #[test]
     fn test_valid_vm_config_net() {
-        vec![
+        [
             // This test is expected to fail because the default MAC address is
             // randomly generated. There's no way we can have twice the same
             // default value.
@@ -1325,20 +1334,6 @@ mod unit_tests {
                 }"#,
                 true,
             ),
-            (
-                vec![
-                    "cloud-hypervisor", "--kernel", "/path/to/kernel",
-                    "--net",
-                    "mac=12:34:56:78:90:ab,host_mac=34:56:78:90:ab:cd,tap=tap0,ip=1.2.3.4",
-                ],
-                r#"{
-                    "payload": {"kernel": "/path/to/kernel"},
-                    "net": [
-                        {"mac": "12:34:56:78:90:ab", "host_mac": "34:56:78:90:ab:cd", "tap": "tap0", "ip": "1.2.3.4"}
-                    ]
-                }"#,
-                true,
-            ),
             (
                 vec![
                     "cloud-hypervisor", "--kernel", "/path/to/kernel",
@@ -1767,7 +1762,7 @@ mod unit_tests {
     #[test]
     #[cfg(target_arch = "x86_64")]
     fn test_valid_vm_config_devices() {
-        vec![
+        [
             (
                 vec![
                     "cloud-hypervisor",
@@ -2017,6 +2012,6 @@ mod unit_tests {
         let (default_vcpus, default_memory, default_rng) = prepare_default_values();
         let args = get_cli_options_sorted(default_vcpus, default_memory, default_rng);
 
-        assert_args_sorted(|| args.iter())
+        assert_args_sorted(|| args.iter());
     }
 }

devices/Cargo.toml

@@ -1,22 +1,28 @@
 [package]
 authors = ["The Chromium OS Authors"]
-edition = "2021"
+edition.workspace = true
 name = "devices"
 version = "0.1.0"
 
 [dependencies]
 acpi_tables = { workspace = true }
-anyhow = "1.0.94"
+anyhow = { workspace = true }
 arch = { path = "../arch" }
-bitflags = "2.9.0"
-byteorder = "1.5.0"
+bitfield-struct = { version = "0.12.0", optional = true }
+bitflags = { workspace = true }
+byteorder = { workspace = true }
 event_monitor = { path = "../event_monitor" }
 hypervisor = { path = "../hypervisor" }
-libc = "0.2.167"
-log = "0.4.22"
-num_enum = "0.7.2"
+libc = { workspace = true }
+linux-loader = { workspace = true, features = [
+  "bzimage",
+  "elf",
+  "pe",
+], optional = true }
+log = { workspace = true }
+num_enum = "0.7.5"
 pci = { path = "../pci" }
-serde = { version = "1.0.208", features = ["derive"] }
+serde = { workspace = true, features = ["derive"] }
 thiserror = { workspace = true }
 tpm = { path = "../tpm" }
 vm-allocator = { path = "../vm-allocator" }
@@ -28,11 +34,20 @@ vm-memory = { workspace = true, features = [
 ] }
 vm-migration = { path = "../vm-migration" }
 vmm-sys-util = { workspace = true }
+zerocopy = { version = "0.8.39", features = [
+  "alloc",
+  "derive",
+], optional = true }
 
 [target.'cfg(any(target_arch = "aarch64", target_arch = "riscv64"))'.dependencies]
 arch = { path = "../arch" }
 
 [features]
 default = []
+fw_cfg = ["arch/fw_cfg", "bitfield-struct", "linux-loader", "zerocopy"]
+ivshmem = []
 kvm = ["arch/kvm"]
 pvmemcontrol = []
+
+[lints]
+workspace = true

devices/src/acpi.rs

@@ -8,9 +8,10 @@ use std::sync::{Arc, Barrier};
 use std::thread;
 use std::time::Instant;
 
-use acpi_tables::{aml, Aml, AmlSink};
-use vm_device::interrupt::InterruptSourceGroup;
+use acpi_tables::{Aml, AmlSink, aml};
+use log::{error, info, warn};
 use vm_device::BusDevice;
+use vm_device::interrupt::InterruptSourceGroup;
 use vm_memory::GuestAddress;
 use vmm_sys_util::eventfd::EventFd;
 
@@ -44,14 +45,14 @@ impl AcpiShutdownDevice {
 impl BusDevice for AcpiShutdownDevice {
     // Spec has all fields as zero
     fn read(&mut self, _base: u64, _offset: u64, data: &mut [u8]) {
-        data.fill(0)
+        data.fill(0);
     }
 
     fn write(&mut self, _base: u64, _offset: u64, data: &[u8]) -> Option<Arc<Barrier>> {
         if data[0] == 1 {
             info!("ACPI Reboot signalled");
             if let Err(e) = self.reset_evt.write(1) {
-                error!("Error triggering ACPI reset event: {}", e);
+                error!("Error triggering ACPI reset event: {e}");
             }
             // Spin until we are sure the reset_evt has been handled and that when
             // we return from the KVM_RUN we will exit rather than re-enter the guest.
@@ -68,7 +69,7 @@ impl BusDevice for AcpiShutdownDevice {
         if data[0] == (S5_SLEEP_VALUE << SLEEP_VALUE_BIT) | (1 << SLEEP_STATUS_EN_BIT) {
             info!("ACPI Shutdown signalled");
             if let Err(e) = self.exit_evt.write(1) {
-                error!("Error triggering ACPI shutdown event: {}", e);
+                error!("Error triggering ACPI shutdown event: {e}");
             }
             // Spin until we are sure the reset_evt has been handled and that when
             // we return from the KVM_RUN we will exit rather than re-enter the guest.
@@ -213,7 +214,7 @@ impl Aml for AcpiGedDevice {
                 ),
             ],
         )
-        .to_aml_bytes(sink)
+        .to_aml_bytes(sink);
     }
 }
 

devices/src/aia.rs

@@ -3,8 +3,6 @@
 //
 // SPDX-License-Identifier: Apache-2.0 AND BSD-3-Clause
 
-use super::interrupt_controller::{Error, InterruptController};
-extern crate arch;
 use std::result;
 use std::sync::{Arc, Mutex};
 
@@ -19,6 +17,8 @@ use vm_memory::address::Address;
 use vm_migration::{Migratable, Pausable, Snapshottable, Transportable};
 use vmm_sys_util::eventfd::EventFd;
 
+use super::interrupt_controller::{Error, InterruptController};
+
 type Result<T> = result::Result<T, Error>;
 
 // Reserve 32 IRQs for legacy devices.
@@ -39,8 +39,9 @@ pub struct Aia {
 }
 
 impl Aia {
+    #[allow(clippy::needless_pass_by_value)]
     pub fn new(
-        vcpu_count: u8,
+        vcpu_count: u32,
         interrupt_manager: Arc<dyn InterruptManager<GroupConfig = MsiIrqGroupConfig>>,
         vm: Arc<dyn hypervisor::Vm>,
     ) -> Result<Aia> {
@@ -51,9 +52,8 @@ impl Aia {
             })
             .map_err(Error::CreateInterruptSourceGroup)?;
 
-        let vaia = vm
-            .create_vaia(Aia::create_default_config(vcpu_count as u64))
-            .map_err(Error::CreateAia)?;
+        let config = Aia::create_default_config(vcpu_count as u64);
+        let vaia = vm.create_vaia(&config).map_err(Error::CreateAia)?;
 
         let aia = Aia {
             interrupt_source_group,

devices/src/debug_console.rs

@@ -9,6 +9,7 @@ use std::io;
 use std::io::Write;
 use std::sync::{Arc, Barrier};
 
+use log::error;
 use vm_device::BusDevice;
 use vm_migration::{Migratable, MigratableError, Pausable, Snapshot, Snapshottable, Transportable};
 

devices/src/gic.rs

@@ -2,15 +2,13 @@
 //
 // SPDX-License-Identifier: Apache-2.0 AND BSD-3-Clause
 
-use super::interrupt_controller::{Error, InterruptController};
-extern crate arch;
 use std::result;
 use std::sync::{Arc, Mutex};
 
 use anyhow::anyhow;
 use arch::layout;
-use hypervisor::arch::aarch64::gic::{GicState, Vgic, VgicConfig};
 use hypervisor::CpuState;
+use hypervisor::arch::aarch64::gic::{GicState, Vgic, VgicConfig};
 use vm_device::interrupt::{
     InterruptIndex, InterruptManager, InterruptSourceConfig, InterruptSourceGroup,
     LegacyIrqSourceConfig, MsiIrqGroupConfig,
@@ -19,6 +17,8 @@ use vm_memory::address::Address;
 use vm_migration::{Migratable, MigratableError, Pausable, Snapshot, Snapshottable, Transportable};
 use vmm_sys_util::eventfd::EventFd;
 
+use super::interrupt_controller::{Error, InterruptController};
+
 type Result<T> = result::Result<T, Error>;
 
 // Reserve 32 IRQs for legacy devices.
@@ -38,8 +38,9 @@ pub struct Gic {
 }
 
 impl Gic {
+    #[allow(clippy::needless_pass_by_value)]
     pub fn new(
-        vcpu_count: u8,
+        vcpu_count: u32,
         interrupt_manager: Arc<dyn InterruptManager<GroupConfig = MsiIrqGroupConfig>>,
         vm: Arc<dyn hypervisor::Vm>,
     ) -> Result<Gic> {
@@ -50,9 +51,8 @@ impl Gic {
             })
             .map_err(Error::CreateInterruptSourceGroup)?;
 
-        let vgic = vm
-            .create_vgic(Gic::create_default_config(vcpu_count as u64))
-            .map_err(Error::CreateGic)?;
+        let config = Gic::create_default_config(vcpu_count as u64);
+        let vgic = vm.create_vgic(&config).map_err(Error::CreateGic)?;
 
         let gic = Gic {
             interrupt_source_group,
@@ -167,10 +167,7 @@ impl Pausable for Gic {
         // Flush tables to guest RAM
         let vgic = self.vgic.as_ref().unwrap().clone();
         vgic.lock().unwrap().save_data_tables().map_err(|e| {
-            MigratableError::Pause(anyhow!(
-                "Could not save GICv3ITS GIC pending tables {:?}",
-                e
-            ))
+            MigratableError::Pause(anyhow!("Could not save GICv3ITS GIC pending tables {e:?}",))
         })?;
         Ok(())
     }

devices/src/ioapic.rs

@@ -13,12 +13,13 @@ use std::result;
 use std::sync::{Arc, Barrier};
 
 use byteorder::{ByteOrder, LittleEndian};
+use log::{debug, error, trace, warn};
 use serde::{Deserialize, Serialize};
+use vm_device::BusDevice;
 use vm_device::interrupt::{
     InterruptIndex, InterruptManager, InterruptSourceConfig, InterruptSourceGroup,
     MsiIrqGroupConfig, MsiIrqSourceConfig,
 };
-use vm_device::BusDevice;
 use vm_memory::GuestAddress;
 use vm_migration::{Migratable, MigratableError, Pausable, Snapshot, Snapshottable, Transportable};
 use vmm_sys_util::eventfd::EventFd;
@@ -151,13 +152,13 @@ impl BusDevice for Ioapic {
             return;
         }
 
-        debug!("IOAPIC_R @ offset 0x{:x}", offset);
+        debug!("IOAPIC_R @ offset 0x{offset:x}");
 
         let value: u32 = match offset as u8 {
             IOREGSEL_OFF => self.reg_sel,
             IOWIN_OFF => self.ioapic_read(),
             _ => {
-                error!("IOAPIC: failed reading at offset {}", offset);
+                error!("IOAPIC: failed reading at offset {offset}");
                 return;
             }
         };
@@ -171,7 +172,7 @@ impl BusDevice for Ioapic {
             return None;
         }
 
-        debug!("IOAPIC_W @ offset 0x{:x}", offset);
+        debug!("IOAPIC_W @ offset 0x{offset:x}");
 
         let value = LittleEndian::read_u32(data);
 
@@ -179,7 +180,7 @@ impl BusDevice for Ioapic {
             IOREGSEL_OFF => self.reg_sel = value,
             IOWIN_OFF => self.ioapic_write(value),
             _ => {
-                error!("IOAPIC: failed writing at offset {}", offset);
+                error!("IOAPIC: failed writing at offset {offset}");
             }
         }
         None
@@ -190,8 +191,8 @@ impl Ioapic {
     pub fn new(
         id: String,
         apic_address: GuestAddress,
-        interrupt_manager: Arc<dyn InterruptManager<GroupConfig = MsiIrqGroupConfig>>,
-        state: Option<IoapicState>,
+        interrupt_manager: &dyn InterruptManager<GroupConfig = MsiIrqGroupConfig>,
+        state: Option<&IoapicState>,
     ) -> Result<Ioapic> {
         let interrupt_source_group = interrupt_manager
             .create_group(MsiIrqGroupConfig {
@@ -266,7 +267,7 @@ impl Ioapic {
             IOWIN_OFF..=REG_MAX_OFFSET => {
                 let (index, is_high_bits) = decode_irq_from_selector(self.reg_sel as u8);
                 if index > NUM_IOAPIC_PINS {
-                    warn!("IOAPIC index out of range: {}", index);
+                    warn!("IOAPIC index out of range: {index}");
                     return;
                 }
                 if is_high_bits {
@@ -282,7 +283,7 @@ impl Ioapic {
                 // The entry must be updated through the interrupt source
                 // group.
                 if let Err(e) = self.update_entry(index, true) {
-                    error!("Failed updating IOAPIC entry: {:?}", e);
+                    error!("Failed updating IOAPIC entry: {e:?}");
                 }
                 // Store the information this IRQ is now being used.
                 self.used_entries[index] = true;
@@ -303,7 +304,7 @@ impl Ioapic {
             IOWIN_OFF..=REG_MAX_OFFSET => {
                 let (index, is_high_bits) = decode_irq_from_selector(self.reg_sel as u8);
                 if index > NUM_IOAPIC_PINS {
-                    warn!("IOAPIC index out of range: {}", index);
+                    warn!("IOAPIC index out of range: {index}");
                     return 0;
                 }
                 if is_high_bits {

devices/src/ivshmem.rs

@@ -0,0 +1,415 @@
+// Copyright © 2024 Tencent Corporation. All rights reserved.
+//
+// SPDX-License-Identifier: Apache-2.0
+//
+
+use std::any::Any;
+use std::path::PathBuf;
+use std::result;
+use std::sync::atomic::{AtomicU32, Ordering};
+use std::sync::{Arc, Barrier, Mutex};
+
+use anyhow::anyhow;
+use byteorder::{ByteOrder, LittleEndian};
+use log::{debug, error, warn};
+use pci::{
+    BarReprogrammingParams, PCI_CONFIGURATION_ID, PciBarConfiguration, PciBarPrefetchable,
+    PciBarRegionType, PciClassCode, PciConfiguration, PciDevice, PciDeviceError, PciHeaderType,
+    PciSubclass,
+};
+use serde::{Deserialize, Serialize};
+use thiserror::Error;
+use vm_allocator::{AddressAllocator, SystemAllocator};
+use vm_device::{BusDevice, Resource, UserspaceMapping};
+use vm_memory::bitmap::AtomicBitmap;
+use vm_memory::{Address, GuestAddress};
+use vm_migration::{Migratable, MigratableError, Pausable, Snapshot, Snapshottable, Transportable};
+
+const IVSHMEM_BAR0_IDX: usize = 0;
+const IVSHMEM_BAR1_IDX: usize = 1;
+const IVSHMEM_BAR2_IDX: usize = 2;
+
+const IVSHMEM_VENDOR_ID: u16 = 0x1af4;
+const IVSHMEM_DEVICE_ID: u16 = 0x1110;
+
+const IVSHMEM_REG_BAR_SIZE: u64 = 0x100;
+
+type MmapRegion = vm_memory::MmapRegion<AtomicBitmap>;
+
+#[derive(Debug, Error)]
+pub enum IvshmemError {
+    #[error("Failed to retrieve PciConfigurationState: {0}")]
+    RetrievePciConfigurationState(#[source] anyhow::Error),
+    #[error("Failed to retrieve IvshmemDeviceState: {0}")]
+    RetrieveIvshmemDeviceStateState(#[source] anyhow::Error),
+    #[error("Failed to remove user memory region")]
+    RemoveUserMemoryRegion,
+    #[error("Failed to create user memory region.")]
+    CreateUserMemoryRegion,
+    #[error("Failed to create userspace mapping.")]
+    CreateUserspaceMapping,
+    #[error("Failed to remove old userspace mapping.")]
+    RemoveUserspaceMapping,
+}
+
+#[derive(Copy, Clone)]
+pub enum IvshmemSubclass {
+    Other = 0x00,
+}
+
+impl PciSubclass for IvshmemSubclass {
+    fn get_register_value(&self) -> u8 {
+        *self as u8
+    }
+}
+
+pub trait IvshmemOps: Send + Sync {
+    fn map_ram_region(
+        &mut self,
+        start_addr: u64,
+        size: usize,
+        backing_file: Option<PathBuf>,
+    ) -> Result<(Arc<MmapRegion>, UserspaceMapping), IvshmemError>;
+
+    fn unmap_ram_region(&mut self, mapping: UserspaceMapping) -> Result<(), IvshmemError>;
+}
+
+/// Inner-Vm Shared Memory Device (Ivshmem device)
+///
+/// This device can share memory between host and guest(ivshmem-plain)
+/// and share memory between guests(ivshmem-doorbell).
+/// But only ivshmem-plain support now, ivshmem-doorbell doesn't support yet.
+pub struct IvshmemDevice {
+    id: String,
+
+    // ivshmem device registers
+    // (only used for ivshmem-doorbell, ivshmem-doorbell don't support yet)
+    _interrupt_mask: u32,
+    _interrupt_status: Arc<AtomicU32>,
+    _iv_position: u32,
+    _doorbell: u32,
+
+    // PCI configuration registers.
+    configuration: PciConfiguration,
+    bar_regions: Vec<PciBarConfiguration>,
+
+    region_size: u64,
+    ivshmem_ops: Arc<Mutex<dyn IvshmemOps>>,
+    backend_file: Option<PathBuf>,
+    region: Option<Arc<MmapRegion>>,
+    userspace_mapping: Option<UserspaceMapping>,
+}
+
+#[derive(Serialize, Deserialize, Default, Clone)]
+pub struct IvshmemDeviceState {
+    interrupt_mask: u32,
+    interrupt_status: u32,
+    iv_position: u32,
+    doorbell: u32,
+}
+
+impl IvshmemDevice {
+    pub fn new(
+        id: String,
+        region_size: u64,
+        backend_file: Option<PathBuf>,
+        ivshmem_ops: Arc<Mutex<dyn IvshmemOps>>,
+        snapshot: Option<&Snapshot>,
+    ) -> Result<Self, IvshmemError> {
+        let pci_configuration_state = vm_migration::state_from_id(snapshot, PCI_CONFIGURATION_ID)
+            .map_err(|e| {
+            IvshmemError::RetrievePciConfigurationState(anyhow!(
+                "Failed to get PciConfigurationState from Snapshot: {e}",
+            ))
+        })?;
+
+        let state: Option<IvshmemDeviceState> = snapshot
+            .as_ref()
+            .map(|s| s.to_state())
+            .transpose()
+            .map_err(|e| {
+                IvshmemError::RetrieveIvshmemDeviceStateState(anyhow!(
+                    "Failed to get IvshmemDeviceState from Snapshot: {e}",
+                ))
+            })?;
+
+        let configuration = PciConfiguration::new(
+            IVSHMEM_VENDOR_ID,
+            IVSHMEM_DEVICE_ID,
+            0x1,
+            PciClassCode::MemoryController,
+            &IvshmemSubclass::Other,
+            None,
+            PciHeaderType::Device,
+            0,
+            0,
+            None,
+            pci_configuration_state,
+        );
+
+        let device = if let Some(s) = state {
+            IvshmemDevice {
+                id,
+                configuration,
+                bar_regions: vec![],
+                _interrupt_mask: s.interrupt_mask,
+                _interrupt_status: Arc::new(AtomicU32::new(s.interrupt_status)),
+                _iv_position: s.iv_position,
+                _doorbell: s.doorbell,
+                region_size,
+                ivshmem_ops,
+                region: None,
+                userspace_mapping: None,
+                backend_file,
+            }
+        } else {
+            IvshmemDevice {
+                id,
+                configuration,
+                bar_regions: vec![],
+                _interrupt_mask: 0,
+                _interrupt_status: Arc::new(AtomicU32::new(0)),
+                _iv_position: 0,
+                _doorbell: 0,
+                region_size,
+                ivshmem_ops,
+                region: None,
+                userspace_mapping: None,
+                backend_file,
+            }
+        };
+        Ok(device)
+    }
+
+    pub fn set_region(&mut self, region: Arc<MmapRegion>, userspace_mapping: UserspaceMapping) {
+        self.region = Some(region);
+        self.userspace_mapping = Some(userspace_mapping);
+    }
+
+    pub fn config_bar_addr(&self) -> u64 {
+        self.configuration.get_bar_addr(IVSHMEM_BAR0_IDX)
+    }
+
+    pub fn data_bar_addr(&self) -> u64 {
+        self.configuration.get_bar_addr(IVSHMEM_BAR2_IDX)
+    }
+
+    fn state(&self) -> IvshmemDeviceState {
+        IvshmemDeviceState {
+            interrupt_mask: self._interrupt_mask,
+            interrupt_status: self._interrupt_status.load(Ordering::SeqCst),
+            iv_position: self._iv_position,
+            doorbell: self._doorbell,
+        }
+    }
+}
+
+impl BusDevice for IvshmemDevice {
+    fn read(&mut self, base: u64, offset: u64, data: &mut [u8]) {
+        self.read_bar(base, offset, data);
+    }
+
+    fn write(&mut self, base: u64, offset: u64, data: &[u8]) -> Option<Arc<Barrier>> {
+        self.write_bar(base, offset, data)
+    }
+}
+
+impl PciDevice for IvshmemDevice {
+    fn allocate_bars(
+        &mut self,
+        _allocator: &Arc<Mutex<SystemAllocator>>,
+        mmio32_allocator: &mut AddressAllocator,
+        mmio64_allocator: &mut AddressAllocator,
+        resources: Option<Vec<Resource>>,
+    ) -> std::result::Result<Vec<PciBarConfiguration>, PciDeviceError> {
+        let mut bars = Vec::new();
+        let mut bar0_addr = None;
+        let mut bar2_addr = None;
+
+        let restoring = resources.is_some();
+        if let Some(resources) = resources {
+            for resource in resources {
+                match resource {
+                    Resource::PciBar { index, base, .. } => match index {
+                        IVSHMEM_BAR0_IDX => {
+                            bar0_addr = Some(GuestAddress(base));
+                        }
+                        IVSHMEM_BAR1_IDX => {}
+                        IVSHMEM_BAR2_IDX => {
+                            bar2_addr = Some(GuestAddress(base));
+                        }
+                        _ => {
+                            error!("Unexpected pci bar index {index}");
+                        }
+                    },
+                    _ => {
+                        error!("Unexpected resource {resource:?}");
+                    }
+                }
+            }
+            if bar0_addr.is_none() || bar2_addr.is_none() {
+                return Err(PciDeviceError::MissingResource);
+            }
+        }
+
+        // BAR0 holds device registers (256 Byte MMIO)
+        let bar0_addr = mmio32_allocator
+            .allocate(bar0_addr, IVSHMEM_REG_BAR_SIZE, None)
+            .ok_or(PciDeviceError::IoAllocationFailed(IVSHMEM_REG_BAR_SIZE))?;
+        debug!("ivshmem bar0 address 0x{:x}", bar0_addr.0);
+
+        let bar0 = PciBarConfiguration::default()
+            .set_index(IVSHMEM_BAR0_IDX)
+            .set_address(bar0_addr.raw_value())
+            .set_size(IVSHMEM_REG_BAR_SIZE)
+            .set_region_type(PciBarRegionType::Memory32BitRegion)
+            .set_prefetchable(PciBarPrefetchable::NotPrefetchable);
+
+        // BAR1 holds MSI-X table and PBA (only ivshmem-doorbell).
+
+        // BAR2 maps the shared memory object
+        let bar2_size = self.region_size;
+        let bar2_addr = mmio64_allocator
+            .allocate(bar2_addr, bar2_size, None)
+            .ok_or(PciDeviceError::IoAllocationFailed(bar2_size))?;
+        debug!("ivshmem bar2 address 0x{:x}", bar2_addr.0);
+
+        let bar2 = PciBarConfiguration::default()
+            .set_index(IVSHMEM_BAR2_IDX)
+            .set_address(bar2_addr.raw_value())
+            .set_size(bar2_size)
+            .set_region_type(PciBarRegionType::Memory64BitRegion)
+            .set_prefetchable(PciBarPrefetchable::Prefetchable);
+
+        if !restoring {
+            self.configuration
+                .add_pci_bar(&bar0)
+                .map_err(|e| PciDeviceError::IoRegistrationFailed(bar0_addr.raw_value(), e))?;
+            self.configuration
+                .add_pci_bar(&bar2)
+                .map_err(|e| PciDeviceError::IoRegistrationFailed(bar2_addr.raw_value(), e))?;
+        }
+
+        bars.push(bar0);
+        bars.push(bar2);
+        self.bar_regions = bars.clone();
+
+        Ok(bars)
+    }
+
+    fn free_bars(
+        &mut self,
+        _allocator: &mut SystemAllocator,
+        _mmio32_allocator: &mut AddressAllocator,
+        _mmio64_allocator: &mut AddressAllocator,
+    ) -> std::result::Result<(), PciDeviceError> {
+        unimplemented!("Device hotplug  and remove are not supported for ivshmem");
+    }
+
+    fn write_config_register(
+        &mut self,
+        reg_idx: usize,
+        offset: u64,
+        data: &[u8],
+    ) -> (Vec<BarReprogrammingParams>, Option<Arc<Barrier>>) {
+        (
+            self.configuration
+                .write_config_register(reg_idx, offset, data),
+            None,
+        )
+    }
+
+    fn read_config_register(&mut self, reg_idx: usize) -> u32 {
+        self.configuration.read_reg(reg_idx)
+    }
+
+    fn read_bar(&mut self, base: u64, offset: u64, data: &mut [u8]) {
+        debug!("read base {base:x} offset {offset}");
+
+        let mut bar_idx = 0;
+        for (idx, bar) in self.bar_regions.iter().enumerate() {
+            if bar.addr() == base {
+                bar_idx = idx;
+            }
+        }
+        match bar_idx {
+            // bar 0
+            0 => {
+                // ivshmem don't use interrupt, we return zero now.
+                LittleEndian::write_u32(data, 0);
+            }
+            // bar 2
+            1 => warn!("Unexpected read ivshmem memory idx: {offset}"),
+            _ => {
+                warn!("Invalid bar_idx: {bar_idx}");
+            }
+        }
+    }
+
+    fn write_bar(&mut self, base: u64, offset: u64, _data: &[u8]) -> Option<Arc<Barrier>> {
+        debug!("write base {base:x} offset {offset}");
+        warn!("Unexpected write ivshmem memory idx: {offset}");
+        None
+    }
+
+    fn move_bar(&mut self, old_base: u64, new_base: u64) -> result::Result<(), std::io::Error> {
+        if new_base == self.data_bar_addr() {
+            if let Some(old_mapping) = self.userspace_mapping.take() {
+                self.ivshmem_ops
+                    .lock()
+                    .unwrap()
+                    .unmap_ram_region(old_mapping)
+                    .map_err(std::io::Error::other)?;
+            }
+            let (region, new_mapping) = self
+                .ivshmem_ops
+                .lock()
+                .unwrap()
+                .map_ram_region(
+                    new_base,
+                    self.region_size as usize,
+                    self.backend_file.clone(),
+                )
+                .map_err(std::io::Error::other)?;
+            self.set_region(region, new_mapping);
+        }
+        for bar in self.bar_regions.iter_mut() {
+            if bar.addr() == old_base {
+                *bar = bar.set_address(new_base);
+            }
+        }
+
+        Ok(())
+    }
+
+    fn as_any_mut(&mut self) -> &mut dyn Any {
+        self
+    }
+
+    fn id(&self) -> Option<String> {
+        Some(self.id.clone())
+    }
+}
+
+impl Pausable for IvshmemDevice {}
+
+impl Snapshottable for IvshmemDevice {
+    fn id(&self) -> String {
+        self.id.clone()
+    }
+
+    // The snapshot/restore (also live migration) support only work for ivshmem-plain mode.
+    // Additional work is needed for supporting ivshmem-doorbell.
+    fn snapshot(&mut self) -> std::result::Result<Snapshot, MigratableError> {
+        let mut snapshot = Snapshot::new_from_state(&self.state())?;
+
+        // Snapshot PciConfiguration
+        snapshot.add_snapshot(self.configuration.id(), self.configuration.snapshot()?);
+
+        Ok(snapshot)
+    }
+}
+
+impl Transportable for IvshmemDevice {}
+
+impl Migratable for IvshmemDevice {}

devices/src/legacy/cmos.rs

@@ -12,7 +12,8 @@ use std::{mem, thread};
 // https://github.com/rust-lang/libc/issues/1848
 #[cfg_attr(target_env = "musl", allow(deprecated))]
 use libc::time_t;
-use libc::{clock_gettime, gmtime_r, timespec, tm, CLOCK_REALTIME};
+use libc::{CLOCK_REALTIME, clock_gettime, gmtime_r, timespec, tm};
+use log::{info, warn};
 use vm_device::BusDevice;
 use vmm_sys_util::eventfd::EventFd;
 
@@ -87,11 +88,11 @@ impl BusDevice for Cmos {
                         }
                     }
                 } else {
-                    self.data[(self.index & INDEX_MASK) as usize] = data[0]
+                    self.data[(self.index & INDEX_MASK) as usize] = data[0];
                 }
             }
-            o => warn!("bad write offset on CMOS device: {}", o),
-        };
+            o => warn!("bad write offset on CMOS device: {o}"),
+        }
         None
     }
 
@@ -164,7 +165,7 @@ impl BusDevice for Cmos {
                 }
             }
             o => {
-                warn!("bad read offset on CMOS device: {}", o);
+                warn!("bad read offset on CMOS device: {o}");
                 0
             }
         }

devices/src/legacy/debug_port.rs

@@ -6,6 +6,7 @@
 use std::fmt;
 use std::time::Instant;
 
+use log::{error, warn};
 use vm_device::BusDevice;
 
 /// Debug I/O port, see:
@@ -62,7 +63,7 @@ impl DebugPort {
 
 impl BusDevice for DebugPort {
     fn read(&mut self, _base: u64, _offset: u64, _data: &mut [u8]) {
-        error!("Invalid read to debug port")
+        error!("Invalid read to debug port");
     }
 
     fn write(

devices/src/legacy/fw_cfg.rs

@@ -0,0 +1,955 @@
+// Copyright 2025 Google LLC.
+//
+// SPDX-License-Identifier: Apache-2.0
+//
+
+/// Cloud Hypervisor implementation of Qemu's fw_cfg spec
+/// https://www.qemu.org/docs/master/specs/fw_cfg.html
+/// Linux kernel fw_cfg driver header
+/// https://github.com/torvalds/linux/blob/master/include/uapi/linux/qemu_fw_cfg.h
+/// Uploading files to the guest via fw_cfg is supported for all kernels 4.6+ w/ CONFIG_FW_CFG_SYSFS enabled
+/// https://cateee.net/lkddb/web-lkddb/FW_CFG_SYSFS.html
+/// No kernel requirement if above functionality is not required,
+/// only firmware must implement mechanism to interact with this fw_cfg device
+use std::{
+    fs::File,
+    io::{ErrorKind, Read, Result, Seek, SeekFrom},
+    mem::offset_of,
+    os::unix::fs::FileExt,
+    sync::{Arc, Barrier},
+};
+
+use acpi_tables::rsdp::Rsdp;
+use arch::RegionType;
+#[cfg(target_arch = "aarch64")]
+use arch::aarch64::layout::{
+    MEM_32BIT_DEVICES_START, MEM_32BIT_RESERVED_START, RAM_64BIT_START, RAM_START as HIGH_RAM_START,
+};
+#[cfg(target_arch = "x86_64")]
+use arch::layout::{
+    EBDA_START, HIGH_RAM_START, MEM_32BIT_DEVICES_SIZE, MEM_32BIT_DEVICES_START,
+    MEM_32BIT_RESERVED_START, PCI_MMCONFIG_SIZE, PCI_MMCONFIG_START, RAM_64BIT_START,
+};
+use bitfield_struct::bitfield;
+#[cfg(target_arch = "x86_64")]
+use linux_loader::bootparam::boot_params;
+#[cfg(target_arch = "aarch64")]
+use linux_loader::loader::pe::arm64_image_header as boot_params;
+use log::{debug, error};
+use vm_device::BusDevice;
+use vm_memory::bitmap::AtomicBitmap;
+use vm_memory::{
+    ByteValued, Bytes, GuestAddress, GuestAddressSpace, GuestMemoryAtomic, GuestMemoryMmap,
+};
+use vmm_sys_util::sock_ctrl_msg::IntoIovec;
+use zerocopy::{FromBytes, FromZeros, Immutable, IntoBytes};
+
+#[cfg(target_arch = "x86_64")]
+// https://github.com/project-oak/oak/tree/main/stage0_bin#memory-layout
+const STAGE0_START_ADDRESS: GuestAddress = GuestAddress(0xfffe_0000);
+#[cfg(target_arch = "x86_64")]
+const STAGE0_SIZE: usize = 0x2_0000;
+const E820_RAM: u32 = 1;
+const E820_RESERVED: u32 = 2;
+
+#[cfg(target_arch = "x86_64")]
+const PORT_FW_CFG_SELECTOR: u64 = 0x510;
+#[cfg(target_arch = "x86_64")]
+const PORT_FW_CFG_DATA: u64 = 0x511;
+#[cfg(target_arch = "x86_64")]
+const PORT_FW_CFG_DMA_HI: u64 = 0x514;
+#[cfg(target_arch = "x86_64")]
+const PORT_FW_CFG_DMA_LO: u64 = 0x518;
+#[cfg(target_arch = "x86_64")]
+pub const PORT_FW_CFG_BASE: u64 = 0x510;
+#[cfg(target_arch = "x86_64")]
+pub const PORT_FW_CFG_WIDTH: u64 = 0xc;
+#[cfg(target_arch = "aarch64")]
+const PORT_FW_CFG_SELECTOR: u64 = 0x9030008;
+#[cfg(target_arch = "aarch64")]
+const PORT_FW_CFG_DATA: u64 = 0x9030000;
+#[cfg(target_arch = "aarch64")]
+const PORT_FW_CFG_DMA_HI: u64 = 0x9030010;
+#[cfg(target_arch = "aarch64")]
+const PORT_FW_CFG_DMA_LO: u64 = 0x9030014;
+#[cfg(target_arch = "aarch64")]
+pub const PORT_FW_CFG_BASE: u64 = 0x9030000;
+#[cfg(target_arch = "aarch64")]
+pub const PORT_FW_CFG_WIDTH: u64 = 0x10;
+
+const FW_CFG_SIGNATURE: u16 = 0x00;
+const FW_CFG_ID: u16 = 0x01;
+const FW_CFG_KERNEL_SIZE: u16 = 0x08;
+const FW_CFG_INITRD_SIZE: u16 = 0x0b;
+const FW_CFG_KERNEL_DATA: u16 = 0x11;
+const FW_CFG_INITRD_DATA: u16 = 0x12;
+const FW_CFG_CMDLINE_SIZE: u16 = 0x14;
+const FW_CFG_CMDLINE_DATA: u16 = 0x15;
+const FW_CFG_SETUP_SIZE: u16 = 0x17;
+const FW_CFG_SETUP_DATA: u16 = 0x18;
+const FW_CFG_FILE_DIR: u16 = 0x19;
+const FW_CFG_KNOWN_ITEMS: usize = 0x20;
+
+pub const FW_CFG_FILE_FIRST: u16 = 0x20;
+pub const FW_CFG_DMA_SIGNATURE: [u8; 8] = *b"QEMU CFG";
+// https://github.com/torvalds/linux/blob/master/include/uapi/linux/qemu_fw_cfg.h
+pub const FW_CFG_ACPI_ID: &str = "QEMU0002";
+// Reserved (must be enabled)
+const FW_CFG_F_RESERVED: u8 = 1 << 0;
+// DMA Toggle Bit (enabled by default)
+const FW_CFG_F_DMA: u8 = 1 << 1;
+pub const FW_CFG_FEATURE: [u8; 4] = [FW_CFG_F_RESERVED | FW_CFG_F_DMA, 0, 0, 0];
+
+const COMMAND_ALLOCATE: u32 = 0x1;
+const COMMAND_ADD_POINTER: u32 = 0x2;
+const COMMAND_ADD_CHECKSUM: u32 = 0x3;
+
+const ALLOC_ZONE_HIGH: u8 = 0x1;
+const ALLOC_ZONE_FSEG: u8 = 0x2;
+
+const FW_CFG_FILENAME_TABLE_LOADER: &str = "etc/table-loader";
+const FW_CFG_FILENAME_RSDP: &str = "acpi/rsdp";
+const FW_CFG_FILENAME_ACPI_TABLES: &str = "acpi/tables";
+
+#[derive(Debug)]
+pub enum FwCfgContent {
+    Bytes(Vec<u8>),
+    Slice(&'static [u8]),
+    File(u64, File),
+    U32(u32),
+}
+
+struct FwCfgContentAccess<'a> {
+    content: &'a FwCfgContent,
+    offset: u32,
+}
+
+impl Read for FwCfgContentAccess<'_> {
+    fn read(&mut self, buf: &mut [u8]) -> Result<usize> {
+        match self.content {
+            FwCfgContent::File(offset, f) => {
+                Seek::seek(&mut (&*f), SeekFrom::Start(offset + self.offset as u64))?;
+                Read::read(&mut (&*f), buf)
+            }
+            FwCfgContent::Bytes(b) => match b.get(self.offset as usize..) {
+                Some(mut s) => s.read(buf),
+                None => Err(ErrorKind::UnexpectedEof)?,
+            },
+            FwCfgContent::Slice(b) => match b.get(self.offset as usize..) {
+                Some(mut s) => s.read(buf),
+                None => Err(ErrorKind::UnexpectedEof)?,
+            },
+            FwCfgContent::U32(n) => match n.to_le_bytes().get(self.offset as usize..) {
+                Some(mut s) => s.read(buf),
+                None => Err(ErrorKind::UnexpectedEof)?,
+            },
+        }
+    }
+}
+
+impl Default for FwCfgContent {
+    fn default() -> Self {
+        FwCfgContent::Slice(&[])
+    }
+}
+
+impl FwCfgContent {
+    fn size(&self) -> Result<u32> {
+        let ret = match self {
+            FwCfgContent::Bytes(v) => v.len(),
+            FwCfgContent::File(offset, f) => (f.metadata()?.len() - offset) as usize,
+            FwCfgContent::Slice(s) => s.len(),
+            FwCfgContent::U32(n) => size_of_val(n),
+        };
+        u32::try_from(ret).map_err(|_| std::io::ErrorKind::InvalidInput.into())
+    }
+    fn access(&self, offset: u32) -> FwCfgContentAccess<'_> {
+        FwCfgContentAccess {
+            content: self,
+            offset,
+        }
+    }
+}
+
+#[derive(Debug, Default)]
+pub struct FwCfgItem {
+    pub name: String,
+    pub content: FwCfgContent,
+}
+
+/// https://www.qemu.org/docs/master/specs/fw_cfg.html
+#[derive(Debug)]
+pub struct FwCfg {
+    selector: u16,
+    data_offset: u32,
+    dma_address: u64,
+    items: Vec<FwCfgItem>,                           // 0x20 and above
+    known_items: [FwCfgContent; FW_CFG_KNOWN_ITEMS], // 0x0 to 0x19
+    memory: GuestMemoryAtomic<GuestMemoryMmap<AtomicBitmap>>,
+}
+
+#[repr(C)]
+#[derive(Debug, IntoBytes, FromBytes)]
+struct FwCfgDmaAccess {
+    control_be: u32,
+    length_be: u32,
+    address_be: u64,
+}
+
+// https://github.com/torvalds/linux/blob/master/include/uapi/linux/qemu_fw_cfg.h#L67
+#[bitfield(u32)]
+struct AccessControl {
+    // FW_CFG_DMA_CTL_ERROR = 0x01
+    error: bool,
+    // FW_CFG_DMA_CTL_READ = 0x02
+    read: bool,
+    #[bits(1)]
+    _unused2: u8,
+    // FW_CFG_DMA_CTL_SKIP = 0x04
+    skip: bool,
+    #[bits(3)]
+    _unused3: u8,
+    // FW_CFG_DMA_CTL_ERROR = 0x08
+    select: bool,
+    #[bits(7)]
+    _unused4: u8,
+    // FW_CFG_DMA_CTL_WRITE = 0x10
+    write: bool,
+    #[bits(16)]
+    _unused: u32,
+}
+
+#[repr(C)]
+#[derive(Debug, IntoBytes, FromBytes)]
+struct FwCfgFilesHeader {
+    count_be: u32,
+}
+
+pub const FILE_NAME_SIZE: usize = 56;
+
+pub fn create_file_name(name: &str) -> [u8; FILE_NAME_SIZE] {
+    let mut c_name = [0u8; FILE_NAME_SIZE];
+    let c_len = std::cmp::min(FILE_NAME_SIZE - 1, name.len());
+    c_name[0..c_len].copy_from_slice(&name.as_bytes()[0..c_len]);
+    c_name
+}
+
+#[allow(dead_code)]
+#[repr(C, packed)]
+#[derive(Debug, IntoBytes, FromBytes, Clone, Copy)]
+struct BootE820Entry {
+    addr: u64,
+    size: u64,
+    type_: u32,
+}
+
+#[repr(C)]
+#[derive(Debug, IntoBytes, FromBytes)]
+struct FwCfgFile {
+    size_be: u32,
+    select_be: u16,
+    _reserved: u16,
+    name: [u8; FILE_NAME_SIZE],
+}
+
+#[repr(C, align(4))]
+#[derive(Debug, IntoBytes, Immutable)]
+struct Allocate {
+    command: u32,
+    file: [u8; FILE_NAME_SIZE],
+    align: u32,
+    zone: u8,
+    _pad: [u8; 63],
+}
+
+#[repr(C, align(4))]
+#[derive(Debug, IntoBytes, Immutable)]
+struct AddPointer {
+    command: u32,
+    dst: [u8; FILE_NAME_SIZE],
+    src: [u8; FILE_NAME_SIZE],
+    offset: u32,
+    size: u8,
+    _pad: [u8; 7],
+}
+
+#[repr(C, align(4))]
+#[derive(Debug, IntoBytes, Immutable)]
+struct AddChecksum {
+    command: u32,
+    file: [u8; FILE_NAME_SIZE],
+    offset: u32,
+    start: u32,
+    len: u32,
+    _pad: [u8; 56],
+}
+
+fn create_intra_pointer(name: &str, offset: usize, size: u8) -> AddPointer {
+    AddPointer {
+        command: COMMAND_ADD_POINTER,
+        dst: create_file_name(name),
+        src: create_file_name(name),
+        offset: offset as u32,
+        size,
+        _pad: [0; 7],
+    }
+}
+
+fn create_acpi_table_checksum(offset: usize, len: usize) -> AddChecksum {
+    AddChecksum {
+        command: COMMAND_ADD_CHECKSUM,
+        file: create_file_name(FW_CFG_FILENAME_ACPI_TABLES),
+        offset: (offset + offset_of!(AcpiTableHeader, checksum)) as u32,
+        start: offset as u32,
+        len: len as u32,
+        _pad: [0; 56],
+    }
+}
+
+#[repr(C, align(4))]
+#[derive(Debug, Clone, Default, FromBytes, IntoBytes)]
+struct AcpiTableHeader {
+    signature: [u8; 4],
+    length: u32,
+    revision: u8,
+    checksum: u8,
+    oem_id: [u8; 6],
+    oem_table_id: [u8; 8],
+    oem_revision: u32,
+    asl_compiler_id: [u8; 4],
+    asl_compiler_revision: u32,
+}
+
+struct AcpiTable {
+    rsdp: Rsdp,
+    tables: Vec<u8>,
+    table_pointers: Vec<usize>,
+    table_checksums: Vec<(usize, usize)>,
+}
+
+impl AcpiTable {
+    fn pointers(&self) -> &[usize] {
+        &self.table_pointers
+    }
+
+    fn checksums(&self) -> &[(usize, usize)] {
+        &self.table_checksums
+    }
+
+    fn take(self) -> (Rsdp, Vec<u8>) {
+        (self.rsdp, self.tables)
+    }
+}
+
+// Creates fw_cfg items used by firmware to load and verify Acpi tables
+// https://github.com/qemu/qemu/blob/master/hw/acpi/bios-linker-loader.c
+fn create_acpi_loader(acpi_table: AcpiTable) -> [FwCfgItem; 3] {
+    let mut table_loader_bytes: Vec<u8> = Vec::new();
+    let allocate_rsdp = Allocate {
+        command: COMMAND_ALLOCATE,
+        file: create_file_name(FW_CFG_FILENAME_RSDP),
+        align: 4,
+        zone: ALLOC_ZONE_FSEG,
+        _pad: [0; 63],
+    };
+    table_loader_bytes.extend(allocate_rsdp.as_bytes());
+
+    let allocate_tables = Allocate {
+        command: COMMAND_ALLOCATE,
+        file: create_file_name(FW_CFG_FILENAME_ACPI_TABLES),
+        align: 4,
+        zone: ALLOC_ZONE_HIGH,
+        _pad: [0; 63],
+    };
+    table_loader_bytes.extend(allocate_tables.as_bytes());
+
+    for pointer_offset in acpi_table.pointers().iter() {
+        let pointer = create_intra_pointer(FW_CFG_FILENAME_ACPI_TABLES, *pointer_offset, 8);
+        table_loader_bytes.extend(pointer.as_bytes());
+    }
+    for (offset, len) in acpi_table.checksums().iter() {
+        let checksum = create_acpi_table_checksum(*offset, *len);
+        table_loader_bytes.extend(checksum.as_bytes());
+    }
+    let pointer_rsdp_to_xsdt = AddPointer {
+        command: COMMAND_ADD_POINTER,
+        dst: create_file_name(FW_CFG_FILENAME_RSDP),
+        src: create_file_name(FW_CFG_FILENAME_ACPI_TABLES),
+        offset: offset_of!(Rsdp, xsdt_addr) as u32,
+        size: 8,
+        _pad: [0; 7],
+    };
+    table_loader_bytes.extend(pointer_rsdp_to_xsdt.as_bytes());
+    let checksum_rsdp = AddChecksum {
+        command: COMMAND_ADD_CHECKSUM,
+        file: create_file_name(FW_CFG_FILENAME_RSDP),
+        offset: offset_of!(Rsdp, checksum) as u32,
+        start: 0,
+        len: offset_of!(Rsdp, length) as u32,
+        _pad: [0; 56],
+    };
+    let checksum_rsdp_ext = AddChecksum {
+        command: COMMAND_ADD_CHECKSUM,
+        file: create_file_name(FW_CFG_FILENAME_RSDP),
+        offset: offset_of!(Rsdp, extended_checksum) as u32,
+        start: 0,
+        len: size_of::<Rsdp>() as u32,
+        _pad: [0; 56],
+    };
+    table_loader_bytes.extend(checksum_rsdp.as_bytes());
+    table_loader_bytes.extend(checksum_rsdp_ext.as_bytes());
+
+    let table_loader = FwCfgItem {
+        name: FW_CFG_FILENAME_TABLE_LOADER.to_owned(),
+        content: FwCfgContent::Bytes(table_loader_bytes),
+    };
+    let (rsdp, tables) = acpi_table.take();
+    let acpi_rsdp = FwCfgItem {
+        name: FW_CFG_FILENAME_RSDP.to_owned(),
+        content: FwCfgContent::Bytes(rsdp.as_bytes().to_owned()),
+    };
+    let apci_tables = FwCfgItem {
+        name: FW_CFG_FILENAME_ACPI_TABLES.to_owned(),
+        content: FwCfgContent::Bytes(tables),
+    };
+    [table_loader, acpi_rsdp, apci_tables]
+}
+
+impl FwCfg {
+    pub fn new(memory: GuestMemoryAtomic<GuestMemoryMmap<AtomicBitmap>>) -> FwCfg {
+        const DEFAULT_ITEM: FwCfgContent = FwCfgContent::Slice(&[]);
+        let mut known_items = [DEFAULT_ITEM; FW_CFG_KNOWN_ITEMS];
+        known_items[FW_CFG_SIGNATURE as usize] = FwCfgContent::Slice(&FW_CFG_DMA_SIGNATURE);
+        known_items[FW_CFG_ID as usize] = FwCfgContent::Slice(&FW_CFG_FEATURE);
+        let file_buf = Vec::from(FwCfgFilesHeader { count_be: 0 }.as_mut_bytes());
+        known_items[FW_CFG_FILE_DIR as usize] = FwCfgContent::Bytes(file_buf);
+
+        FwCfg {
+            selector: 0,
+            data_offset: 0,
+            dma_address: 0,
+            items: vec![],
+            known_items,
+            memory,
+        }
+    }
+
+    pub fn populate_fw_cfg(
+        &mut self,
+        mem_size: Option<usize>,
+        kernel: Option<File>,
+        initramfs: Option<File>,
+        cmdline: Option<std::ffi::CString>,
+        fw_cfg_item_list: Option<Vec<FwCfgItem>>,
+    ) -> Result<()> {
+        if let Some(mem_size) = mem_size {
+            self.add_e820(mem_size)?;
+        }
+        if let Some(kernel) = kernel {
+            self.add_kernel_data(&kernel)?;
+        }
+        if let Some(cmdline) = cmdline {
+            self.add_kernel_cmdline(cmdline);
+        }
+        if let Some(initramfs) = initramfs {
+            self.add_initramfs_data(&initramfs)?;
+        }
+        if let Some(fw_cfg_item_list) = fw_cfg_item_list {
+            for item in fw_cfg_item_list {
+                self.add_item(item)?;
+            }
+        }
+        Ok(())
+    }
+
+    pub fn add_e820(&mut self, mem_size: usize) -> Result<()> {
+        #[cfg(target_arch = "x86_64")]
+        let mut mem_regions = vec![
+            (GuestAddress(0), EBDA_START.0 as usize, RegionType::Ram),
+            (
+                MEM_32BIT_DEVICES_START,
+                MEM_32BIT_DEVICES_SIZE as usize,
+                RegionType::Reserved,
+            ),
+            (
+                PCI_MMCONFIG_START,
+                PCI_MMCONFIG_SIZE as usize,
+                RegionType::Reserved,
+            ),
+            (STAGE0_START_ADDRESS, STAGE0_SIZE, RegionType::Reserved),
+        ];
+        #[cfg(target_arch = "aarch64")]
+        let mut mem_regions = arch::aarch64::arch_memory_regions();
+        if mem_size < MEM_32BIT_DEVICES_START.0 as usize {
+            mem_regions.push((
+                HIGH_RAM_START,
+                mem_size - HIGH_RAM_START.0 as usize,
+                RegionType::Ram,
+            ));
+        } else {
+            mem_regions.push((
+                HIGH_RAM_START,
+                MEM_32BIT_RESERVED_START.0 as usize - HIGH_RAM_START.0 as usize,
+                RegionType::Ram,
+            ));
+            mem_regions.push((
+                RAM_64BIT_START,
+                mem_size - (MEM_32BIT_DEVICES_START.0 as usize),
+                RegionType::Ram,
+            ));
+        }
+        let mut bytes = vec![];
+        for (addr, size, region) in mem_regions.iter() {
+            let type_ = match region {
+                RegionType::Ram => E820_RAM,
+                RegionType::Reserved => E820_RESERVED,
+                RegionType::SubRegion => continue,
+            };
+            let mut entry = BootE820Entry {
+                addr: addr.0,
+                size: *size as u64,
+                type_,
+            };
+            bytes.extend_from_slice(entry.as_mut_bytes());
+        }
+        let item = FwCfgItem {
+            name: "etc/e820".to_owned(),
+            content: FwCfgContent::Bytes(bytes),
+        };
+        self.add_item(item)
+    }
+
+    fn file_dir_mut(&mut self) -> &mut Vec<u8> {
+        let FwCfgContent::Bytes(file_buf) = &mut self.known_items[FW_CFG_FILE_DIR as usize] else {
+            unreachable!("fw_cfg: selector {FW_CFG_FILE_DIR:#x} should be FwCfgContent::Byte!")
+        };
+        file_buf
+    }
+
+    fn update_count(&mut self) {
+        let mut header = FwCfgFilesHeader {
+            count_be: (self.items.len() as u32).to_be(),
+        };
+        self.file_dir_mut()[0..4].copy_from_slice(header.as_mut_bytes());
+    }
+
+    pub fn add_item(&mut self, item: FwCfgItem) -> Result<()> {
+        let index = self.items.len();
+        let c_name = create_file_name(&item.name);
+        let size = item.content.size()?;
+        let mut cfg_file = FwCfgFile {
+            size_be: size.to_be(),
+            select_be: (FW_CFG_FILE_FIRST + index as u16).to_be(),
+            _reserved: 0,
+            name: c_name,
+        };
+        self.file_dir_mut()
+            .extend_from_slice(cfg_file.as_mut_bytes());
+        self.items.push(item);
+        self.update_count();
+        Ok(())
+    }
+
+    fn dma_read_content(
+        &self,
+        content: &FwCfgContent,
+        offset: u32,
+        len: u32,
+        address: u64,
+    ) -> Result<u32> {
+        let content_size = content.size()?.saturating_sub(offset);
+        let op_size = std::cmp::min(content_size, len);
+        let mut access = content.access(offset);
+        let mut buf = vec![0u8; op_size as usize];
+        access.read_exact(buf.as_mut_bytes())?;
+        let r = self
+            .memory
+            .memory()
+            .write(buf.as_bytes(), GuestAddress(address));
+        match r {
+            Err(e) => {
+                error!("fw_cfg: dma read error: {e:x?}");
+                Err(ErrorKind::InvalidInput.into())
+            }
+            Ok(size) => Ok(size as u32),
+        }
+    }
+
+    fn dma_read(&mut self, selector: u16, len: u32, address: u64) -> Result<()> {
+        let op_size = if let Some(content) = self.known_items.get(selector as usize) {
+            self.dma_read_content(content, self.data_offset, len, address)
+        } else if let Some(item) = self.items.get((selector - FW_CFG_FILE_FIRST) as usize) {
+            self.dma_read_content(&item.content, self.data_offset, len, address)
+        } else {
+            error!("fw_cfg: selector {selector:#x} does not exist.");
+            Err(ErrorKind::NotFound.into())
+        }?;
+        self.data_offset += op_size;
+        Ok(())
+    }
+
+    fn do_dma(&mut self) {
+        let dma_address = self.dma_address;
+        let mut access = FwCfgDmaAccess::new_zeroed();
+        let dma_access = match self
+            .memory
+            .memory()
+            .read(access.as_mut_bytes(), GuestAddress(dma_address))
+        {
+            Ok(_) => access,
+            Err(e) => {
+                error!("fw_cfg: invalid address of dma access {dma_address:#x}: {e:?}");
+                return;
+            }
+        };
+        let control = AccessControl(u32::from_be(dma_access.control_be));
+        if control.select() {
+            self.selector = control.select() as u16;
+        }
+        let len = u32::from_be(dma_access.length_be);
+        let addr = u64::from_be(dma_access.address_be);
+        let ret = if control.read() {
+            self.dma_read(self.selector, len, addr)
+        } else if control.write() {
+            Err(ErrorKind::InvalidInput.into())
+        } else if control.skip() {
+            self.data_offset += len;
+            Ok(())
+        } else {
+            Err(ErrorKind::InvalidData.into())
+        };
+        let mut access_resp = AccessControl(0);
+        if let Err(e) = ret {
+            error!("fw_cfg: dma operation {dma_access:x?}: {e:x?}");
+            access_resp.set_error(true);
+        }
+        if let Err(e) = self.memory.memory().write(
+            &access_resp.0.to_be_bytes(),
+            GuestAddress(dma_address + core::mem::offset_of!(FwCfgDmaAccess, control_be) as u64),
+        ) {
+            error!("fw_cfg: finishing dma: {e:?}");
+        }
+    }
+
+    pub fn add_kernel_data(&mut self, file: &File) -> Result<()> {
+        let mut buffer = vec![0u8; size_of::<boot_params>()];
+        file.read_exact_at(&mut buffer, 0)?;
+        let bp = boot_params::from_mut_slice(&mut buffer).unwrap();
+        #[cfg(target_arch = "x86_64")]
+        {
+            // must set to 4 for backwards compatibility
+            // https://docs.kernel.org/arch/x86/boot.html#the-real-mode-kernel-header
+            if bp.hdr.setup_sects == 0 {
+                bp.hdr.setup_sects = 4;
+            }
+            // wildcard boot loader type
+            bp.hdr.type_of_loader = 0xff;
+        }
+        #[cfg(target_arch = "aarch64")]
+        let kernel_start = bp.text_offset;
+        #[cfg(target_arch = "x86_64")]
+        let kernel_start = (bp.hdr.setup_sects as usize + 1) * 512;
+        self.known_items[FW_CFG_SETUP_SIZE as usize] = FwCfgContent::U32(buffer.len() as u32);
+        self.known_items[FW_CFG_SETUP_DATA as usize] = FwCfgContent::Bytes(buffer);
+        self.known_items[FW_CFG_KERNEL_SIZE as usize] =
+            FwCfgContent::U32(file.metadata()?.len() as u32 - kernel_start as u32);
+        self.known_items[FW_CFG_KERNEL_DATA as usize] =
+            FwCfgContent::File(kernel_start as u64, file.try_clone()?);
+        Ok(())
+    }
+
+    pub fn add_kernel_cmdline(&mut self, s: std::ffi::CString) {
+        let bytes = s.into_bytes_with_nul();
+        self.known_items[FW_CFG_CMDLINE_SIZE as usize] = FwCfgContent::U32(bytes.len() as u32);
+        self.known_items[FW_CFG_CMDLINE_DATA as usize] = FwCfgContent::Bytes(bytes);
+    }
+
+    pub fn add_acpi(
+        &mut self,
+        rsdp: Rsdp,
+        tables: Vec<u8>,
+        table_checksums: Vec<(usize, usize)>,
+        table_pointers: Vec<usize>,
+    ) -> Result<()> {
+        let acpi_table = AcpiTable {
+            rsdp,
+            tables,
+            table_checksums,
+            table_pointers,
+        };
+        let [table_loader, acpi_rsdp, apci_tables] = create_acpi_loader(acpi_table);
+        self.add_item(table_loader)?;
+        self.add_item(acpi_rsdp)?;
+        self.add_item(apci_tables)
+    }
+
+    pub fn add_initramfs_data(&mut self, file: &File) -> Result<()> {
+        let initramfs_size = file.metadata()?.len();
+        self.known_items[FW_CFG_INITRD_SIZE as usize] = FwCfgContent::U32(initramfs_size as _);
+        self.known_items[FW_CFG_INITRD_DATA as usize] = FwCfgContent::File(0, file.try_clone()?);
+        Ok(())
+    }
+
+    fn read_content(content: &FwCfgContent, offset: u32, data: &mut [u8], size: u32) -> Option<u8> {
+        let start = offset as usize;
+        let end = start + size as usize;
+        match content {
+            FwCfgContent::Bytes(b) => {
+                if b.len() >= size as usize {
+                    data.copy_from_slice(&b[start..end]);
+                }
+            }
+            FwCfgContent::Slice(s) => {
+                if s.len() >= size as usize {
+                    data.copy_from_slice(&s[start..end]);
+                }
+            }
+            FwCfgContent::File(o, f) => {
+                f.read_exact_at(data, o + offset as u64).ok()?;
+            }
+            FwCfgContent::U32(n) => {
+                let bytes = n.to_le_bytes();
+                data.copy_from_slice(&bytes[start..end]);
+            }
+        }
+        Some(size as u8)
+    }
+
+    fn read_data(&mut self, data: &mut [u8], size: u32) -> u8 {
+        let ret = if let Some(content) = self.known_items.get(self.selector as usize) {
+            Self::read_content(content, self.data_offset, data, size)
+        } else if let Some(item) = self.items.get((self.selector - FW_CFG_FILE_FIRST) as usize) {
+            Self::read_content(&item.content, self.data_offset, data, size)
+        } else {
+            error!("fw_cfg: selector {:#x} does not exist.", self.selector);
+            None
+        };
+        if let Some(val) = ret {
+            self.data_offset += size;
+            val
+        } else {
+            0
+        }
+    }
+}
+
+impl BusDevice for FwCfg {
+    fn read(&mut self, _base: u64, offset: u64, data: &mut [u8]) {
+        let port = offset + PORT_FW_CFG_BASE;
+        let size = data.len();
+        match (port, size) {
+            (PORT_FW_CFG_SELECTOR, _) => {
+                error!("fw_cfg: selector register is write-only.");
+            }
+            (PORT_FW_CFG_DATA, _) => _ = self.read_data(data, size as u32),
+            (PORT_FW_CFG_DMA_HI, 4) => {
+                let addr = self.dma_address;
+                let addr_hi = (addr >> 32) as u32;
+                data.copy_from_slice(&addr_hi.to_be_bytes());
+            }
+            (PORT_FW_CFG_DMA_LO, 4) => {
+                let addr = self.dma_address;
+                let addr_lo = (addr & 0xffff_ffff) as u32;
+                data.copy_from_slice(&addr_lo.to_be_bytes());
+            }
+            _ => {
+                debug!(
+                    "fw_cfg: read from unknown port {port:#x}: {size:#x} bytes and offset {offset:#x}."
+                );
+            }
+        }
+    }
+
+    fn write(&mut self, _base: u64, offset: u64, data: &[u8]) -> Option<Arc<Barrier>> {
+        let port = offset + PORT_FW_CFG_BASE;
+        let size = data.size();
+        match (port, size) {
+            (PORT_FW_CFG_SELECTOR, 2) => {
+                let mut buf = [0u8; 2];
+                buf[..size].copy_from_slice(&data[..size]);
+                #[cfg(target_arch = "x86_64")]
+                let val = u16::from_le_bytes(buf);
+                #[cfg(target_arch = "aarch64")]
+                let val = u16::from_be_bytes(buf);
+                self.selector = val;
+                self.data_offset = 0;
+            }
+            (PORT_FW_CFG_DATA, 1) => error!("fw_cfg: data register is read-only."),
+            (PORT_FW_CFG_DMA_HI, 4) => {
+                let mut buf = [0u8; 4];
+                buf[..size].copy_from_slice(&data[..size]);
+                let val = u32::from_be_bytes(buf);
+                self.dma_address &= 0xffff_ffff;
+                self.dma_address |= (val as u64) << 32;
+            }
+            (PORT_FW_CFG_DMA_LO, 4) => {
+                let mut buf = [0u8; 4];
+                buf[..size].copy_from_slice(&data[..size]);
+                let val = u32::from_be_bytes(buf);
+                self.dma_address &= !0xffff_ffff;
+                self.dma_address |= val as u64;
+                self.do_dma();
+            }
+            _ => debug!(
+                "fw_cfg: write to unknown port {port:#x}: {size:#x} bytes and offset {offset:#x} ."
+            ),
+        }
+        None
+    }
+}
+
+#[cfg(test)]
+mod unit_tests {
+    use std::ffi::CString;
+    use std::io::Write;
+
+    use vmm_sys_util::tempfile::TempFile;
+
+    use super::*;
+
+    #[cfg(target_arch = "x86_64")]
+    const SELECTOR_OFFSET: u64 = 0;
+    #[cfg(target_arch = "aarch64")]
+    const SELECTOR_OFFSET: u64 = 8;
+    #[cfg(target_arch = "x86_64")]
+    const DATA_OFFSET: u64 = 1;
+    #[cfg(target_arch = "aarch64")]
+    const DATA_OFFSET: u64 = 0;
+    #[cfg(target_arch = "x86_64")]
+    const DMA_OFFSET: u64 = 4;
+    #[cfg(target_arch = "aarch64")]
+    const DMA_OFFSET: u64 = 16;
+
+    #[test]
+    fn test_signature() {
+        let gm = GuestMemoryAtomic::new(
+            GuestMemoryMmap::from_ranges(&[(GuestAddress(0), RAM_64BIT_START.0 as usize)]).unwrap(),
+        );
+
+        let mut fw_cfg = FwCfg::new(gm);
+
+        let mut data = vec![0u8];
+
+        let mut sig_iter = FW_CFG_DMA_SIGNATURE.into_iter();
+        fw_cfg.write(0, SELECTOR_OFFSET, &[FW_CFG_SIGNATURE as u8, 0]);
+        loop {
+            if let Some(char) = sig_iter.next() {
+                fw_cfg.read(0, DATA_OFFSET, &mut data);
+                assert_eq!(data[0], char);
+            } else {
+                return;
+            }
+        }
+    }
+    #[test]
+    fn test_kernel_cmdline() {
+        let gm = GuestMemoryAtomic::new(
+            GuestMemoryMmap::from_ranges(&[(GuestAddress(0), RAM_64BIT_START.0 as usize)]).unwrap(),
+        );
+
+        let mut fw_cfg = FwCfg::new(gm);
+
+        let cmdline = *b"cmdline\0";
+
+        fw_cfg.add_kernel_cmdline(CString::from_vec_with_nul(cmdline.to_vec()).unwrap());
+
+        let mut data = vec![0u8];
+
+        let mut cmdline_iter = cmdline.into_iter();
+        fw_cfg.write(0, SELECTOR_OFFSET, &[FW_CFG_CMDLINE_DATA as u8, 0]);
+        loop {
+            if let Some(char) = cmdline_iter.next() {
+                fw_cfg.read(0, DATA_OFFSET, &mut data);
+                assert_eq!(data[0], char);
+            } else {
+                return;
+            }
+        }
+    }
+
+    #[test]
+    fn test_initram_fs() {
+        let gm = GuestMemoryAtomic::new(
+            GuestMemoryMmap::from_ranges(&[(GuestAddress(0), RAM_64BIT_START.0 as usize)]).unwrap(),
+        );
+
+        let mut fw_cfg = FwCfg::new(gm);
+
+        let temp = TempFile::new().unwrap();
+        let mut temp_file = temp.as_file();
+
+        let initram_content = b"this is the initramfs";
+        let written = temp_file.write(initram_content);
+        assert_eq!(written.unwrap(), 21);
+        let _ = fw_cfg.add_initramfs_data(temp_file);
+
+        let mut data = vec![0u8];
+
+        let mut initram_iter = (*initram_content).into_iter();
+        fw_cfg.write(0, SELECTOR_OFFSET, &[FW_CFG_INITRD_DATA as u8, 0]);
+        loop {
+            if let Some(char) = initram_iter.next() {
+                fw_cfg.read(0, DATA_OFFSET, &mut data);
+                assert_eq!(data[0], char);
+            } else {
+                return;
+            }
+        }
+    }
+
+    #[test]
+    fn test_dma() {
+        let code = [
+            0xba, 0xf8, 0x03, 0x00, 0xd8, 0x04, b'0', 0xee, 0xb0, b'\n', 0xee, 0xf4,
+        ];
+
+        let content = FwCfgContent::Bytes(code.to_vec());
+
+        let mem_size = 0x1000;
+        let load_addr = GuestAddress(0x1000);
+        let mem: GuestMemoryMmap<AtomicBitmap> =
+            GuestMemoryMmap::from_ranges(&[(load_addr, mem_size)]).unwrap();
+
+        // Note: In firmware we would just allocate FwCfgDmaAccess struct
+        // and use address of struct (&) as dma address
+        let mut access_control = AccessControl(0);
+        // bit 1 = read access
+        access_control.set_read(true);
+        // length of data to access
+        let length_be = (code.len() as u32).to_be();
+        // guest address for data
+        let code_address = 0x1900_u64;
+        let address_be = code_address.to_be();
+        let mut access = FwCfgDmaAccess {
+            control_be: access_control.0.to_be(), // bit(1) = read bit
+            length_be,
+            address_be,
+        };
+        // access address is where to put the code
+        let access_address = GuestAddress(load_addr.0);
+        let address_bytes = access_address.0.to_be_bytes();
+        let dma_lo: [u8; 4] = address_bytes[0..4].try_into().unwrap();
+        let dma_hi: [u8; 4] = address_bytes[4..8].try_into().unwrap();
+
+        // writing the FwCfgDmaAccess to mem (this would just be self.dma_access.as_ref() in guest)
+        let _ = mem.write(access.as_mut_bytes(), access_address);
+        let mem_m = GuestMemoryAtomic::new(mem.clone());
+        let mut fw_cfg = FwCfg::new(mem_m);
+        let cfg_item = FwCfgItem {
+            name: "code".to_string(),
+            content,
+        };
+        let _ = fw_cfg.add_item(cfg_item);
+
+        let mut data = [0u8; 12];
+
+        let _ = mem.read(&mut data, GuestAddress(code_address));
+        assert_ne!(data, code);
+
+        fw_cfg.write(0, SELECTOR_OFFSET, &[FW_CFG_FILE_FIRST as u8, 0]);
+        fw_cfg.write(0, DMA_OFFSET, &dma_lo);
+        fw_cfg.write(0, DMA_OFFSET + 4, &dma_hi);
+        let _ = mem.read(&mut data, GuestAddress(code_address));
+        assert_eq!(data, code);
+    }
+}

devices/src/legacy/fwdebug.rs

@@ -9,6 +9,7 @@
 
 use std::sync::{Arc, Barrier};
 
+use log::error;
 use vm_device::BusDevice;
 
 /// Provides firmware debug output via I/O port controls
@@ -26,9 +27,9 @@ impl BusDevice for FwDebugDevice {
     /// Upon read return the magic value to indicate that there is a debug port
     fn read(&mut self, _base: u64, _offset: u64, data: &mut [u8]) {
         if data.len() == 1 {
-            data[0] = 0xe9
+            data[0] = 0xe9;
         } else {
-            error!("Invalid read size on debug port: {}", data.len())
+            error!("Invalid read size on debug port: {}", data.len());
         }
     }
 
@@ -36,7 +37,7 @@ impl BusDevice for FwDebugDevice {
         if data.len() == 1 {
             print!("{}", data[0] as char);
         } else {
-            error!("Invalid write size on debug port: {}", data.len())
+            error!("Invalid write size on debug port: {}", data.len());
         }
 
         None

devices/src/legacy/gpio_pl061.rs

@@ -10,10 +10,11 @@
 use std::sync::{Arc, Barrier};
 use std::{io, result};
 
+use log::warn;
 use serde::{Deserialize, Serialize};
 use thiserror::Error;
-use vm_device::interrupt::InterruptSourceGroup;
 use vm_device::BusDevice;
+use vm_device::interrupt::InterruptSourceGroup;
 use vm_migration::{Migratable, MigratableError, Pausable, Snapshot, Snapshottable, Transportable};
 
 use crate::{read_le_u32, write_le_u32};
@@ -28,10 +29,10 @@ const GPIORIE: u64 = 0x414; // Raw Interrupt Status Register
 const GPIOMIS: u64 = 0x418; // Masked Interrupt Status Register
 const GPIOIC: u64 = 0x41c; // Interrupt Clear Register
 const GPIOAFSEL: u64 = 0x420; // Mode Control Select Register
-                              // From 0x424 to 0xFDC => reserved space.
-                              // From 0xFE0 to 0xFFC => Peripheral and PrimeCell Identification Registers which are Read Only registers.
-                              // These registers can conceptually be treated as a 32-bit register, and PartNumber[11:0] is used to identify the peripheral.
-                              // We are putting the expected values (look at 'Reset value' column from above mentioned document) in an array.
+// From 0x424 to 0xFDC => reserved space.
+// From 0xFE0 to 0xFFC => Peripheral and PrimeCell Identification Registers which are Read Only registers.
+// These registers can conceptually be treated as a 32-bit register, and PartNumber[11:0] is used to identify the peripheral.
+// We are putting the expected values (look at 'Reset value' column from above mentioned document) in an array.
 const GPIO_ID: [u8; 8] = [0x61, 0x10, 0x14, 0x00, 0x0d, 0xf0, 0x05, 0xb1];
 // ID Margins
 const GPIO_ID_LOW: u64 = 0xfe0;
@@ -261,7 +262,7 @@ impl BusDevice for Gpio {
             let index = ((offset - GPIO_ID_LOW) >> 2) as usize;
             value = u32::from(GPIO_ID[index]);
         } else if offset < OFS_DATA {
-            value = self.data & ((offset >> 2) as u32)
+            value = self.data & ((offset >> 2) as u32);
         } else {
             value = match offset {
                 GPIODIR => self.dir,
@@ -294,7 +295,7 @@ impl BusDevice for Gpio {
         if data.len() <= 4 {
             let value = read_le_u32(data);
             if let Err(e) = self.handle_write(offset, value) {
-                warn!("Failed to write to GPIO PL061 device: {}", e);
+                warn!("Failed to write to GPIO PL061 device: {e}");
             }
         } else {
             warn!(
@@ -323,7 +324,7 @@ impl Transportable for Gpio {}
 impl Migratable for Gpio {}
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use vm_device::interrupt::{InterruptIndex, InterruptSourceConfig};
     use vmm_sys_util::eventfd::EventFd;
 

devices/src/legacy/i8042.rs

@@ -8,6 +8,7 @@ use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::{Arc, Barrier};
 use std::thread;
 
+use log::{error, info};
 use vm_device::BusDevice;
 use vmm_sys_util::eventfd::EventFd;
 
@@ -45,7 +46,7 @@ impl BusDevice for I8042Device {
         if data.len() == 1 && data[0] == 0xfe && offset == 3 {
             info!("i8042 reset signalled");
             if let Err(e) = self.reset_evt.write(1) {
-                error!("Error triggering i8042 reset event: {}", e);
+                error!("Error triggering i8042 reset event: {e}");
             }
             // Spin until we are sure the reset_evt has been handled and that when
             // we return from the KVM_RUN we will exit rather than re-enter the guest.

devices/src/legacy/mod.rs

@@ -8,6 +8,8 @@
 mod cmos;
 #[cfg(target_arch = "x86_64")]
 mod debug_port;
+#[cfg(feature = "fw_cfg")]
+pub mod fw_cfg;
 #[cfg(target_arch = "x86_64")]
 mod fwdebug;
 #[cfg(target_arch = "aarch64")]
@@ -22,6 +24,8 @@ mod uart_pl011;
 pub use self::cmos::Cmos;
 #[cfg(target_arch = "x86_64")]
 pub use self::debug_port::DebugPort;
+#[cfg(feature = "fw_cfg")]
+pub use self::fw_cfg::FwCfg;
 #[cfg(target_arch = "x86_64")]
 pub use self::fwdebug::FwDebugDevice;
 #[cfg(target_arch = "aarch64")]

devices/src/legacy/rtc_pl031.rs

@@ -15,6 +15,7 @@ use std::result;
 use std::sync::{Arc, Barrier};
 use std::time::Instant;
 
+use log::warn;
 use thiserror::Error;
 use vm_device::BusDevice;
 
@@ -31,11 +32,11 @@ const RTCIMSC: u64 = 0x10; // Interrupt Mask Set or Clear Register.
 const RTCRIS: u64 = 0x14; // Raw Interrupt Status.
 const RTCMIS: u64 = 0x18; // Masked Interrupt Status.
 const RTCICR: u64 = 0x1c; // Interrupt Clear Register.
-                          // From 0x020 to 0xFDC => reserved space.
-                          // From 0xFE0 to 0x1000 => Peripheral and PrimeCell Identification Registers which are Read Only registers.
-                          // AMBA standard devices have CIDs (Cell IDs) and PIDs (Peripheral IDs). The linux kernel will look for these in order to assert the identity
-                          // of these devices (i.e look at the `amba_device_try_add` function).
-                          // We are putting the expected values (look at 'Reset value' column from above mentioned document) in an array.
+// From 0x020 to 0xFDC => reserved space.
+// From 0xFE0 to 0x1000 => Peripheral and PrimeCell Identification Registers which are Read Only registers.
+// AMBA standard devices have CIDs (Cell IDs) and PIDs (Peripheral IDs). The linux kernel will look for these in order to assert the identity
+// of these devices (i.e look at the `amba_device_try_add` function).
+// We are putting the expected values (look at 'Reset value' column from above mentioned document) in an array.
 const PL031_ID: [u8; 8] = [0x31, 0x10, 0x14, 0x00, 0x0d, 0xf0, 0x05, 0xb1];
 // We are only interested in the margins.
 const AMBA_ID_LOW: u64 = 0xFE0;
@@ -201,12 +202,11 @@ impl BusDevice for Rtc {
         if data.len() <= 4 {
             let v = read_le_u32(data);
             if let Err(e) = self.handle_write(offset, v) {
-                warn!("Failed to write to RTC PL031 device: {}", e);
+                warn!("Failed to write to RTC PL031 device: {e}");
             }
         } else {
             warn!(
-                "Invalid RTC PL031 write: offset {}, data length {}",
-                offset,
+                "Invalid RTC PL031 write: offset {offset}, data length {}",
                 data.len()
             );
         }
@@ -216,7 +216,7 @@ impl BusDevice for Rtc {
 }
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use super::*;
     use crate::{
         read_be_u16, read_be_u32, read_le_i32, read_le_u16, read_le_u64, write_be_u16,

devices/src/legacy/serial.rs

@@ -10,8 +10,8 @@ use std::sync::{Arc, Barrier};
 use std::{io, result};
 
 use serde::{Deserialize, Serialize};
-use vm_device::interrupt::InterruptSourceGroup;
 use vm_device::BusDevice;
+use vm_device::interrupt::InterruptSourceGroup;
 use vm_migration::{Migratable, MigratableError, Pausable, Snapshot, Snapshottable, Transportable};
 use vmm_sys_util::errno::Result;
 
@@ -215,7 +215,7 @@ impl Serial {
     fn thr_empty(&mut self) -> Result<()> {
         if self.is_thr_intr_enabled() {
             self.add_intr_bit(IIR_THR_BIT);
-            self.trigger_interrupt()?
+            self.trigger_interrupt()?;
         }
         Ok(())
     }
@@ -223,7 +223,7 @@ impl Serial {
     fn recv_data(&mut self) -> Result<()> {
         if self.is_recv_intr_enabled() {
             self.add_intr_bit(IIR_RECV_BIT);
-            self.trigger_interrupt()?
+            self.trigger_interrupt()?;
         }
         self.line_status |= LSR_DATA_BIT;
         Ok(())
@@ -240,10 +240,10 @@ impl Serial {
     fn handle_write(&mut self, offset: u8, v: u8) -> Result<()> {
         match offset {
             DLAB_LOW if self.is_dlab_set() => {
-                self.baud_divisor = (self.baud_divisor & 0xff00) | u16::from(v)
+                self.baud_divisor = (self.baud_divisor & 0xff00) | u16::from(v);
             }
             DLAB_HIGH if self.is_dlab_set() => {
-                self.baud_divisor = (self.baud_divisor & 0x00ff) | ((u16::from(v)) << 8)
+                self.baud_divisor = (self.baud_divisor & 0x00ff) | ((u16::from(v)) << 8);
             }
             DATA => {
                 if self.is_loop() {
@@ -340,7 +340,7 @@ impl Transportable for Serial {}
 impl Migratable for Serial {}
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use std::sync::Mutex;
 
     use vm_device::interrupt::{InterruptIndex, InterruptSourceConfig};

devices/src/legacy/uart_pl011.rs

@@ -11,10 +11,11 @@ use std::sync::{Arc, Barrier};
 use std::time::Instant;
 use std::{io, result};
 
+use log::{debug, warn};
 use serde::{Deserialize, Serialize};
 use thiserror::Error;
-use vm_device::interrupt::InterruptSourceGroup;
 use vm_device::BusDevice;
+use vm_device::interrupt::InterruptSourceGroup;
 use vm_migration::{Migratable, MigratableError, Pausable, Snapshot, Snapshottable, Transportable};
 
 use crate::{read_le_u32, write_le_u32};
@@ -322,7 +323,7 @@ impl Pl011 {
                 self.handle_debug();
             }
             off => {
-                debug!("PL011: Bad write offset, offset: {}", off);
+                debug!("PL011: Bad write offset, offset: {off}");
                 return Err(Error::BadWriteOffset(off));
             }
         }
@@ -424,12 +425,11 @@ impl BusDevice for Pl011 {
         if data.len() <= 4 {
             let v = read_le_u32(data);
             if let Err(e) = self.handle_write(offset, v) {
-                warn!("Failed to write to PL011 device: {}", e);
+                warn!("Failed to write to PL011 device: {e}");
             }
         } else {
             warn!(
-                "Invalid PL011 write: offset {}, data length {}",
-                offset,
+                "Invalid PL011 write: offset {offset}, data length {}",
                 data.len()
             );
         }
@@ -453,7 +453,7 @@ impl Transportable for Pl011 {}
 impl Migratable for Pl011 {}
 
 #[cfg(test)]
-mod tests {
+mod unit_tests {
     use std::sync::Mutex;
 
     use vm_device::interrupt::{InterruptIndex, InterruptSourceConfig};

devices/src/lib.rs

@@ -7,13 +7,6 @@
 
 //! Emulates virtual and hardware devices.
 
-#[macro_use]
-extern crate bitflags;
-#[macro_use]
-extern crate event_monitor;
-#[macro_use]
-extern crate log;
-
 pub mod acpi;
 #[cfg(target_arch = "riscv64")]
 pub mod aia;
@@ -24,6 +17,8 @@ pub mod gic;
 pub mod interrupt_controller;
 #[cfg(target_arch = "x86_64")]
 pub mod ioapic;
+#[cfg(feature = "ivshmem")]
+pub mod ivshmem;
 pub mod legacy;
 #[cfg(feature = "pvmemcontrol")]
 pub mod pvmemcontrol;
@@ -32,8 +27,12 @@ pub mod pvpanic;
 #[cfg(not(target_arch = "riscv64"))]
 pub mod tpm;
 
+use bitflags::bitflags;
+
 pub use self::acpi::{AcpiGedDevice, AcpiPmTimerDevice, AcpiShutdownDevice};
-pub use self::pvpanic::{PvPanicDevice, PVPANIC_DEVICE_MMIO_SIZE};
+#[cfg(feature = "ivshmem")]
+pub use self::ivshmem::IvshmemDevice;
+pub use self::pvpanic::{PVPANIC_DEVICE_MMIO_SIZE, PvPanicDevice};
 
 bitflags! {
     pub struct AcpiNotificationFlags: u8 {