Per-device CFMWS base for multi-GPU CXL passthrough

.github/workflows/build-source-package.yml

@@ -0,0 +1,115 @@
+# Builds a Debian source package for QEMU
+
+name: Build Debian Source Package
+
+on:
+  push:
+    branches:
+      - 'nvidia/latest'
+      - 'nvidia_stable-10.1'
+
+  # Allow manual triggering of the workflow from the GitHub UI
+  workflow_dispatch:
+
+jobs:
+  build-source-package:
+    # arm64 is the primary platform of interest for this project
+    runs-on: ubuntu-24.04
+
+    steps:
+      # Step 1: Check out the repository's code
+      - name: Checkout repository
+        uses: actions/checkout@v6
+        with:
+          # Fetch all history for all tags and branches, which can be necessary
+          # for tools like git-buildpackage to determine version numbers.
+          fetch-depth: 0
+
+      # Step 2: Install Debian packaging tools and build dependencies
+      - name: Install Build Dependencies
+        run: |
+          sudo apt-get update
+
+          sudo DEBIAN_FRONTEND=noninteractive apt-get install -y devscripts equivs fakeroot git-buildpackage
+
+          # Use mk-build-deps to parse debian/control, create a dummy package
+          # with all build dependencies, and install it. This ensures all
+          # required build dependencies are present.
+          # The --no-install-recommends flag keeps the environment minimal.
+          sudo mk-build-deps -i --tool="apt-get -y --no-install-recommends"
+          # Remove debs that this generates from the working directory after install
+          rm -f qemu-build-deps*
+
+      # Step 3: Generate Upstream Tarball
+      - name: Generate Upstream Tarball
+        run: |
+          # Extract source name and upstream version from debian/changelog.
+          # The first sed command strips any epoch (e.g., "1:") from the start of the version.
+          # The second sed command strips the Debian revision from the end of the version string.
+          # dpkg-parsechangelog is in the dpkg-dev package, a dependency of devscripts.
+          export SOURCE_NAME=$(dpkg-parsechangelog -S Source)
+          export UPSTREAM_VERSION=$(dpkg-parsechangelog -S Version | sed 's/^[0-9]\+://' | sed 's/-[^-]*$//')
+
+          git config --global user.email "robot@builder.local"
+          git config --global user.name "Builder Robot"
+          # Create the .orig.tar.gz in the parent directory.
+          ./debian/build-source-package-from-git.sh
+
+      # Step 4: Upload Workspace for Debugging
+      - name: Upload Workspace for Debugging
+        uses: actions/upload-artifact@v7
+        with:
+          name: pre-build-workspace
+          path: .
+
+      # Step 5: Build the Debian source package
+      - name: Build Source Package
+        run: |
+          # Use debuild to create the source package.
+          # -S: Build a source package only (no binaries).
+          # -us: Do not sign the source package.
+          # -uc: Do not sign the .changes file.
+          # This command will find and use the .orig.tar.gz we just created.
+          current_dir=$(pwd)
+          cd deb
+          # There should only be one thing in here, which is the tarball we'll extract
+          # so wildcard should be OK in this instance.
+          ls
+          tar -xvf $(ls -d * | head -n 1)
+          ls
+          cd $(ls -d */ | head -n 1)
+          debuild -S -us -uc
+          cd "$current_dir"
+
+      # Step 6: Log MD5 Checksums
+      - name: Log MD5 Checksums as Annotations
+        run: |
+          # Change to the parent directory where artifacts were created.
+          current_dir=$(pwd)
+          cd deb
+          mkdir source_package_artifacts
+
+          # Iterate over all the generated build artifacts.
+          for file in *.dsc *.orig.tar.* *.debian.tar.* *.changes *source.build *source.buildinfo; do
+            # Check if files matching the pattern exist before processing.
+            if [ -f "$file" ]; then
+              # Calculate the MD5 sum and format it for the annotation.
+              # The output of md5sum is "CHECKSUM  FILENAME".
+              checksum=$(md5sum "$file")
+
+              # Echo the checksum as a workflow notice annotation.
+              # This will make it easily visible in the GitHub Actions UI.
+              echo "::notice title=MD5 Checksum::$checksum"
+              mv "$file" source_package_artifacts
+            fi
+          done
+          mv source_package_artifacts "$current_dir"
+
+      # Step 7: Upload the generated source package files as artifacts
+      - name: Upload Debian Source Package Artifacts
+        uses: actions/upload-artifact@v7
+        with:
+          # Name of the artifact bundle to be displayed in the GitHub UI
+          name: debian-source-package
+
+          path: source_package_artifacts

backends/iommufd.c

@@ -390,16 +390,23 @@ bool iommufd_backend_get_dirty_bitmap(IOMMUFDBackend *be,
     return true;
 }
 
+/*
+ * @type can carry a desired HW info type defined in the uapi headers. If caller
+ * doesn't have one, indicating it wants the default type, then @type should be
+ * zeroed (i.e. IOMMU_HW_INFO_TYPE_DEFAULT).
+ */
 bool iommufd_backend_get_device_info(IOMMUFDBackend *be, uint32_t devid,
                                      uint32_t *type, void *data, uint32_t len,
                                      uint64_t *caps, uint8_t *max_pasid_log2,
                                      Error **errp)
 {
     struct iommu_hw_info info = {
+        .flags = (*type) ? IOMMU_HW_INFO_FLAG_INPUT_TYPE : 0,
         .size = sizeof(info),
         .dev_id = devid,
         .data_len = len,
         .data_uptr = (uintptr_t)data,
+        .in_data_type = *type,
     };
 
     if (ioctl(be->fd, IOMMU_GET_HW_INFO, &info)) {
@@ -456,6 +463,7 @@ bool iommufd_backend_invalidate_cache(IOMMUFDBackend *be, uint32_t id,
 
 bool iommufd_backend_alloc_viommu(IOMMUFDBackend *be, uint32_t dev_id,
                                   uint32_t viommu_type, uint32_t hwpt_id,
+                                  void *data_ptr, uint32_t data_len,
                                   uint32_t *out_viommu_id, Error **errp)
 {
     int ret;
@@ -464,11 +472,14 @@ bool iommufd_backend_alloc_viommu(IOMMUFDBackend *be, uint32_t dev_id,
         .type = viommu_type,
         .dev_id = dev_id,
         .hwpt_id = hwpt_id,
+        .data_len = data_len,
+        .data_uptr = (uintptr_t)data_ptr,
     };
 
     ret = ioctl(be->fd, IOMMU_VIOMMU_ALLOC, &alloc_viommu);
 
     trace_iommufd_backend_alloc_viommu(be->fd, dev_id, viommu_type, hwpt_id,
+                                       (uintptr_t)data_ptr, data_len,
                                        alloc_viommu.out_viommu_id, ret);
     if (ret) {
         error_setg_errno(errp, errno, "IOMMU_VIOMMU_ALLOC failed");
@@ -538,24 +549,77 @@ bool iommufd_backend_alloc_veventq(IOMMUFDBackend *be, uint32_t viommu_id,
     return true;
 }
 
-bool host_iommu_device_iommufd_attach_hwpt(HostIOMMUDeviceIOMMUFD *idev,
+bool iommufd_backend_alloc_hw_queue(IOMMUFDBackend *be, uint32_t viommu_id,
+                                    uint32_t queue_type, uint32_t index,
+                                    uint64_t addr, uint64_t length,
+                                    uint32_t *out_hw_queue_id, Error **errp)
+{
+    int ret;
+    struct iommu_hw_queue_alloc alloc_hw_queue = {
+        .size = sizeof(alloc_hw_queue),
+        .flags = 0,
+        .viommu_id = viommu_id,
+        .type = queue_type,
+        .index = index,
+        .nesting_parent_iova = addr,
+        .length = length,
+    };
+
+    ret = ioctl(be->fd, IOMMU_HW_QUEUE_ALLOC, &alloc_hw_queue);
+
+    trace_iommufd_backend_alloc_hw_queue(be->fd, viommu_id, queue_type,
+                                         index, addr, length,
+                                         alloc_hw_queue.out_hw_queue_id, ret);
+    if (ret) {
+        error_setg_errno(errp, errno, "IOMMU_HW_QUEUE_ALLOC failed");
+        return false;
+    }
+
+    g_assert(out_hw_queue_id);
+    *out_hw_queue_id = alloc_hw_queue.out_hw_queue_id;
+    return true;
+}
+
+/*
+ * Helper to mmap HW MMIO regions exposed via iommufd for a vIOMMU instance.
+ * The caller is responsible for unmapping the mapped region.
+ */
+bool iommufd_backend_viommu_mmap(IOMMUFDBackend *be, uint32_t viommu_id,
+                                 uint64_t size, off_t offset, void **out_ptr,
+                                 Error **errp)
+{
+    g_assert(viommu_id);
+    g_assert(out_ptr);
+
+    *out_ptr = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, be->fd,
+                   offset);
+    trace_iommufd_backend_viommu_mmap(be->fd, viommu_id, size, offset);
+    if (*out_ptr == MAP_FAILED) {
+        error_setg_errno(errp, errno, "IOMMUFD vIOMMU mmap failed");
+        return false;
+    }
+
+    return true;
+}
+
+bool host_iommu_device_iommufd_attach_hwpt(HostIOMMUDeviceIOMMUFD *hiodi,
                                            uint32_t hwpt_id, Error **errp)
 {
-    HostIOMMUDeviceIOMMUFDClass *idevc =
-        HOST_IOMMU_DEVICE_IOMMUFD_GET_CLASS(idev);
+    HostIOMMUDeviceIOMMUFDClass *hiodic =
+        HOST_IOMMU_DEVICE_IOMMUFD_GET_CLASS(hiodi);
 
-    g_assert(idevc->attach_hwpt);
-    return idevc->attach_hwpt(idev, hwpt_id, errp);
+    g_assert(hiodic->attach_hwpt);
+    return hiodic->attach_hwpt(hiodi, hwpt_id, errp);
 }
 
-bool host_iommu_device_iommufd_detach_hwpt(HostIOMMUDeviceIOMMUFD *idev,
+bool host_iommu_device_iommufd_detach_hwpt(HostIOMMUDeviceIOMMUFD *hiodi,
                                            Error **errp)
 {
-    HostIOMMUDeviceIOMMUFDClass *idevc =
-        HOST_IOMMU_DEVICE_IOMMUFD_GET_CLASS(idev);
+    HostIOMMUDeviceIOMMUFDClass *hiodic =
+        HOST_IOMMU_DEVICE_IOMMUFD_GET_CLASS(hiodi);
 
-    g_assert(idevc->detach_hwpt);
-    return idevc->detach_hwpt(idev, errp);
+    g_assert(hiodic->detach_hwpt);
+    return hiodic->detach_hwpt(hiodi, errp);
 }
 
 static int hiod_iommufd_get_cap(HostIOMMUDevice *hiod, int cap, Error **errp)
@@ -591,10 +655,10 @@ static bool hiod_iommufd_get_pasid_info(HostIOMMUDevice *hiod,
 
 static void hiod_iommufd_class_init(ObjectClass *oc, const void *data)
 {
-    HostIOMMUDeviceClass *hioc = HOST_IOMMU_DEVICE_CLASS(oc);
+    HostIOMMUDeviceClass *hiodc = HOST_IOMMU_DEVICE_CLASS(oc);
 
-    hioc->get_cap = hiod_iommufd_get_cap;
-    hioc->get_pasid_info = hiod_iommufd_get_pasid_info;
+    hiodc->get_cap = hiod_iommufd_get_cap;
+    hiodc->get_pasid_info = hiod_iommufd_get_pasid_info;
 };
 
 static const TypeInfo types[] = {

backends/trace-events

@@ -21,9 +21,11 @@ iommufd_backend_free_id(int iommufd, uint32_t id, int ret) " iommufd=%d id=%d (%
 iommufd_backend_set_dirty(int iommufd, uint32_t hwpt_id, bool start, int ret) " iommufd=%d hwpt=%u enable=%d (%d)"
 iommufd_backend_get_dirty_bitmap(int iommufd, uint32_t hwpt_id, uint64_t iova, uint64_t size, uint64_t flags, uint64_t page_size, int ret) " iommufd=%d hwpt=%u iova=0x%"PRIx64" size=0x%"PRIx64" flags=0x%"PRIx64" page_size=0x%"PRIx64" (%d)"
 iommufd_backend_invalidate_cache(int iommufd, uint32_t id, uint32_t data_type, uint32_t entry_len, uint32_t entry_num, uint32_t done_num, uint64_t data_ptr, int ret) " iommufd=%d id=%u data_type=%u entry_len=%u entry_num=%u done_num=%u data_ptr=0x%"PRIx64" (%d)"
-iommufd_backend_alloc_viommu(int iommufd, uint32_t dev_id, uint32_t type, uint32_t hwpt_id, uint32_t viommu_id, int ret) " iommufd=%d type=%u dev_id=%u hwpt_id=%u viommu_id=%u (%d)"
+iommufd_backend_alloc_viommu(int iommufd, uint32_t dev_id, uint32_t type, uint32_t hwpt_id, uint64_t data_ptr, uint32_t data_len, uint32_t viommu_id, int ret) " iommufd=%d type=%u dev_id=%u hwpt_id=%u data_ptr=0x%"PRIx64" data_len=0x%x viommu_id=%u (%d)"
 iommufd_backend_alloc_vdev(int iommufd, uint32_t dev_id, uint32_t viommu_id, uint64_t virt_id, uint32_t vdev_id, int ret) " iommufd=%d dev_id=%u viommu_id=%u virt_id=0x%"PRIx64" vdev_id=%u (%d)"
 iommufd_viommu_alloc_eventq(int iommufd, uint32_t viommu_id, uint32_t type, uint32_t veventq_id, uint32_t veventq_fd, int ret) " iommufd=%d viommu_id=%u type=%u veventq_id=%u veventq_fd=%u (%d)"
+iommufd_backend_alloc_hw_queue(int iommufd, uint32_t viommu_id, uint32_t queue_type, uint32_t index, uint64_t addr, uint64_t size, uint32_t queue_id, int ret) " iommufd=%d viommu_id=%u queue_type=%u index=%u addr=0x%"PRIx64" size=0x%"PRIx64" queue_id=%u (%d)"
+iommufd_backend_viommu_mmap(int iommufd, uint32_t viommu_id, uint64_t size, uint64_t offset) " iommufd=%d viommu_id=%u size=0x%"PRIx64" offset=0x%"PRIx64
 
 # igvm-cfg.c
 igvm_reset_enter(int type) "type=%u"

debian/README-components-versions

@@ -0,0 +1,74 @@
+Evolution of qemu-related packages in debian
+
+kvm: was a binary package providing /usr/bin/kvm before squeeze.
+
+ In squeeze it has been renamed to qemu-kvm,
+ so kvm becomes a transitional package.
+
+ Probably can be get rid of now.
+
+ No need to conflict with this package, but we can force-remove it
+ by conflicting with it.
+
+ Some packages may still Require/Depend/Suggest: kvm.
+
+
+qemu-kvm: was a binary package providing /usr/bin/kvm in wheezy (v. 1.1)
+
+ This binary become a (deprecated) shell wrapper and was a part of
+ qemu-system-x86 package between versions 1.3.0+dfsg-5 (when qemu-system
+ were split) up to 1.7.0+dfsg-1, with qemu-kvm package being transitional.
+
+ At version 1.7.0+dfsg-2, qemu-kvm packcage is not transitional anymore,
+ and the wrapper is provided by qemu-kvm, not qemu-system-x86.
+
+ Some packages may Require/Depend/Suggest: qemu-kvm.
+
+qemu: initially it was a mix of all of qemu (unrelated, not conflicting
+with qemu-kvm or kvm)
+
+ at 0.11.0-1 it were split to qemu-user, qemu-user-static and qemu-system.
+
+ Also qemu-keymaps is made a separate package at version 0.12.4+dfsg-4
+ (for use together with qemu-kvm and xen and, on ubuntu, qemu-linaro).
+
+ And qemu-utils become a separate package (look when? since the beginning?)
+
+ Now it is a metapackage and a placeholder for some docs
+
+qemu-system: was a package with all qemu-system-* binaries.
+
+ Were split into individual qemu-system-foo and qemu-system-common at version
+ 1.3.0+dfsg-5, so qemu-system becomes a metapackage without its own files.
+
+ The split introduced a bunch of qemu-system-* conflicting+replacing
+ qemu-system<<1.3.0+dfsg-5.
+
+ At version 1.3.0+dfsg-4exp (before split, in experimental), /usr/bin/kvm was
+ a part of qemu-system (and later qemu-system-x86)/
+
+ At the same version, virtfs-proxy-helper binary has been moved from qemu-utils
+ to qemu-system (and later to qemu-system-common).
+
+qemu-system-common: appeared at 1.3.0+dfsg-5, files common for all qemu-system*
+
+ briefly (>=1.3.0+dfsg-4exp, <<1.3.0+dfsg-5), qemu-utils had
+ virtfs-proxy-helper binary which is in qemu-system-common
+
+qemu-keymaps: keymap definition for qemu, qemu-kvm and xen.
+
+ Can be merged into qemu-system-common now, or after qemu-linaro.
+
+ Were split out of qemu at version 0.12.4+dfsg-4
+
+qemu-utils: common utilities (part of qemu-common in ubuntu?)
+
+ In versions between 1.3.0+dfsg-1~exp1 and 1.3.0+dfsg-4exp, wrongly shipped
+ virtfs-proxy-helper binary
+
+qemu-user, qemu-user-static: user-mode qemu emulation
+
+ Were split out of qemu at verson 0.11.0-1 (pre-squeeze),
+ do not conflict/replace anything anymore.
+
+qemu-common: ubuntu package

debian/TODO

@@ -0,0 +1,16 @@
+* fix other binaries (s390-zipl, ppc_rom)
+
+* permissions/ownership for /dev/vhost_net???
+
+* startup script for qemu-guest-agent: check dependencies & runlevels.
+
+* maybe provide activation for udev & systemd:
+
+   SUBSYSTEM=="virtio-ports", ATTR{name}=="org.qemu.guest_agent.0", \
+     TAG+="systemd" ENV{SYSTEMD_WANTS}="qemu-guest-agent.service"
+
+  and dev-virtio/ports-org.qemu.guest_agent.0.device activation
+  for systemd.
+
+* decide what to do with qemu-bridge-helper (#691138)
+