[PATCH 0/8] various fixes

[PATCH 0/8] various fixes

[Luis Chamberlain]

I took a GPU out for a spin for PCI passthrough and found that the
vagrant removal nuked some things for it, this fixes that and enhances
the PCI passthrough listing with sensible names.

While at it, this also includes a fix to gest us through the debian
trixie to forky transition, allowing our host and target nodes to
upgrade.

Luis Chamberlain (8):
  libvirt: fix PCI-E passthrough configuration after Vagrant removal
  gen-dynamic-pci: fix PCI passthrough target guest default value
  gen-dynamic-pci: add GPU detection and naming support
  guestfs / devconfig: allow for upgrade from Debian testing to forky
  guestfs: add OVMF package for UEFI firmware support
  gen_nodes: fix duplicate pcipassthrough keys in YAML output
  gen_nodes: fix YAML indentation in pcipassthrough template
  guestfs: fix PCIe passthrough device attachment

 kconfigs/Kconfig.libvirt                      |  43 ++++
 .../dynamic-kconfig/gen-dynamic-pci.py        | 198 +++++++++++++++++-
 .../tasks/install-deps/debian/main.yml        |   9 +
 .../gen_nodes/templates/gen_nodes_list.j2     |  14 +-
 .../gen_nodes/templates/guestfs_nodes.j2      |   2 +-
 playbooks/roles/gen_nodes/templates/hosts.j2  |   4 +-
 .../roles/guestfs/tasks/bringup/main.yml      |   2 +-
 .../tasks/install-deps/debian/main.yml        |   9 +
 scripts/check_pciepassthrough_kconfig.sh      |  15 +-
 9 files changed, 275 insertions(+), 21 deletions(-)

-- 
2.47.2

[PATCH 1/8] libvirt: fix PCI-E passthrough configuration after Vagrant removal

[Luis Chamberlain]

The PCI-E passthrough functionality was broken when Vagrant support was
removed. Two issues prevented the passthrough options from appearing in
menuconfig:

1. The check_pciepassthrough_kconfig.sh script was gutted during the
   Vagrant removal and always returned "n", preventing the passthrough
   Kconfig from being detected.

2. The source line for including the generated passthrough Kconfig was
   removed but never updated to the new location.

This commit fixes both issues and additionally improves the user
interface by:

- Restoring the check script to properly detect if the passthrough
  Kconfig file exists
- Adding a proper menu structure with "Enable PCI-E passthrough support"
  boolean option
- Creating a dedicated "Configure PCI-E passthrough" submenu for device
  selection
- Including the assignment strategy choice (manual vs round-robin)

Now users can properly enable and configure PCI-E passthrough through
a clean menu interface instead of having all device options scattered
in the main libvirt menu.

Fixes: 89b7ef82fd91 ("kconfigs: fix Kconfig references after vagrant removal")
Generated-by: Claude AI
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
---
 kconfigs/Kconfig.libvirt                 | 43 ++++++++++++++++++++++++
 scripts/check_pciepassthrough_kconfig.sh | 15 ++++++++-
 2 files changed, 57 insertions(+), 1 deletion(-)

diff --git a/kconfigs/Kconfig.libvirt b/kconfigs/Kconfig.libvirt
index 0f26699..61502c1 100644
--- a/kconfigs/Kconfig.libvirt
+++ b/kconfigs/Kconfig.libvirt
@@ -462,6 +462,49 @@ config HAVE_LIBVIRT_PCIE_PASSTHROUGH
 	bool
 	default $(shell, scripts/check_pciepassthrough_kconfig.sh passthrough_libvirt.generated)
 
+config KDEVOPS_LIBVIRT_PCIE_PASSTHROUGH
+	bool "Enable PCI-E passthrough support"
+	depends on HAVE_LIBVIRT_PCIE_PASSTHROUGH
+	depends on LIBVIRT
+	depends on LIBVIRT_MACHINE_TYPE_Q35
+	default n
+	help
+	  Enable this to configure PCI-E passthrough of devices from the host
+	  to guest VMs. This requires the q35 machine type and proper IOMMU
+	  configuration on the host.
+
+	  If enabled, you can select which PCI devices to pass through to
+	  which guests in the passthrough configuration menu.
+
+if KDEVOPS_LIBVIRT_PCIE_PASSTHROUGH
+
+menu "Configure PCI-E passthrough"
+
+choice
+	prompt "PCI-E passthrough assignment strategy"
+	default KDEVOPS_LIBVIRT_PCIE_PASSTHROUGH_TYPE_EACH
+
+config KDEVOPS_LIBVIRT_PCIE_PASSTHROUGH_TYPE_EACH
+	bool "Manually assign devices to specific guests"
+	help
+	  This lets you manually specify which PCI device gets assigned to
+	  which specific guest. Each device can be assigned to exactly one
+	  guest.
+
+config KDEVOPS_LIBVIRT_PCIE_PASSTHROUGH_TYPE_ROUNDROBIN
+	bool "Round-robin device assignment"
+	help
+	  Automatically distribute PCI devices among guests in a round-robin
+	  fashion. This is useful when you have multiple similar devices
+	  (like NVMe drives) and want to distribute them evenly.
+
+endchoice
+
+source "Kconfig.passthrough_libvirt.generated"
+
+endmenu
+
+endif # KDEVOPS_LIBVIRT_PCIE_PASSTHROUGH
 
 choice
 	prompt "Machine type to use"
diff --git a/scripts/check_pciepassthrough_kconfig.sh b/scripts/check_pciepassthrough_kconfig.sh
index 5d83ecf..e631085 100755
--- a/scripts/check_pciepassthrough_kconfig.sh
+++ b/scripts/check_pciepassthrough_kconfig.sh
@@ -1,4 +1,17 @@
 #!/bin/bash
 # SPDX-License-Identifier: copyleft-next-0.3.1
 
-echo n
+# Check if the PCI passthrough Kconfig file exists
+KCONFIG_FILE="$1"
+
+if [ -z "$KCONFIG_FILE" ]; then
+    echo n
+    exit 0
+fi
+
+# Check both with and without Kconfig. prefix
+if [ -f "Kconfig.${KCONFIG_FILE}" ] || [ -f "${KCONFIG_FILE}" ]; then
+    echo y
+else
+    echo n
+fi
-- 
2.47.2

[PATCH 2/8] gen-dynamic-pci: fix PCI passthrough target guest default value

[Luis Chamberlain]

Fix two issues preventing PCI passthrough configuration from appearing
in generated nodes files:

1. Template generation was producing "None" in YAML files due to
   incorrect include path and missing None handling in the Jinja2
   template. Fixed by correcting the include path and adding a
   replace filter to remove "None" values.

2. PCI passthrough target_guest field was defaulting to empty string,
   preventing devices from being assigned to guests. Fixed by setting
   the Kconfig default to reference KDEVOPS_HOSTS_PREFIX, which
   automatically assigns devices to the configured host prefix (e.g.,
   "debian13").

With these fixes, PCI passthrough configuration now correctly appears
in the generated guestfs/kdevops_nodes.yaml file when devices are
enabled and target_guest is properly configured.

Generated-by: Claude AI
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
---
 playbooks/python/workflows/dynamic-kconfig/gen-dynamic-pci.py | 4 ++--
 playbooks/roles/gen_nodes/templates/guestfs_nodes.j2          | 2 +-
 playbooks/roles/gen_nodes/templates/hosts.j2                  | 4 ++--
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/playbooks/python/workflows/dynamic-kconfig/gen-dynamic-pci.py b/playbooks/python/workflows/dynamic-kconfig/gen-dynamic-pci.py
index 107bb25..f86ce74 100755
--- a/playbooks/python/workflows/dynamic-kconfig/gen-dynamic-pci.py
+++ b/playbooks/python/workflows/dynamic-kconfig/gen-dynamic-pci.py
@@ -95,9 +95,9 @@ def add_pcie_kconfig_name(config_name, sdevice):
 def add_pcie_kconfig_target(config_name, sdevice):
     sys.stdout.write("config %s_TARGET_GUEST\n" % (config_name))
     sys.stdout.write(
-        '\tstring  "Taret guest to offload %s"\n' % (strip_kconfig_name(sdevice))
+        '\tstring  "Target guest to offload %s"\n' % (strip_kconfig_name(sdevice))
     )
-    sys.stdout.write('\tdefault ""\n')
+    sys.stdout.write('\tdefault KDEVOPS_HOSTS_PREFIX\n')
     sys.stdout.write("\tdepends on %s\n" % config_name)
     sys.stdout.write("\tdepends on KDEVOPS_LIBVIRT_PCIE_PASSTHROUGH_TYPE_EACH\n")
     sys.stdout.write("\thelp\n")
diff --git a/playbooks/roles/gen_nodes/templates/guestfs_nodes.j2 b/playbooks/roles/gen_nodes/templates/guestfs_nodes.j2
index 5e87df3..a548f25 100644
--- a/playbooks/roles/gen_nodes/templates/guestfs_nodes.j2
+++ b/playbooks/roles/gen_nodes/templates/guestfs_nodes.j2
@@ -1,3 +1,3 @@
 ---
 guestfs_nodes:
-{% include './templates/hosts.j2' %}
+{% include 'hosts.j2' -%}
diff --git a/playbooks/roles/gen_nodes/templates/hosts.j2 b/playbooks/roles/gen_nodes/templates/hosts.j2
index ca4d2cb..a7f8b82 100644
--- a/playbooks/roles/gen_nodes/templates/hosts.j2
+++ b/playbooks/roles/gen_nodes/templates/hosts.j2
@@ -1,6 +1,6 @@
-{% from "gen_nodes_list.j2" import gen_nodes_list %}
+{% from "gen_nodes_list.j2" import gen_nodes_list -%}
 {{ gen_nodes_list( nodes, kdevops_baseline_and_dev, 100,
 		   pcie_passthrough_enable,
 		   pcie_passthrough_target_type,
 		   pcie_passthrough_target,
-		   pcie_passthrough_devices ) }}
+		   pcie_passthrough_devices ) | replace("None", "") }}
-- 
2.47.2

[PATCH 3/8] gen-dynamic-pci: add GPU detection and naming support

[Luis Chamberlain]

Add specialized detection and naming support for GPU devices in the
PCI passthrough configuration generator, similar to existing NVMe
support. This provides cleaner, more informative names for GPUs in
the menuconfig interface.

Features added:
- Automatic detection of AMD, NVIDIA, and Intel GPUs based on device
  names and common GPU keywords
- GPU memory detection from sysfs when available (via mem_info_vram_total
  or resource file size estimation)
- Clean formatting of GPU names with vendor prefixes and memory info
- Support for various GPU families:
  * AMD: Radeon, Instinct, FirePro, RDNA, GCN series
  * NVIDIA: GeForce, Quadro, Tesla, RTX/GTX series
  * Intel: UHD Graphics, HD Graphics, Iris, Arc series

Example output:
- Before: "0000:03:00.0 IOMMU group None - Device 0834"
- After: "0000:03:00.0 IOMMU group None - GPU - Radeon Pro VII/Radeon Instinct MI50 32GB"

This makes it much easier for users to identify and select GPUs for
passthrough in the configuration menu.

Generated-by: Claude AI
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
---
 .../dynamic-kconfig/gen-dynamic-pci.py        | 194 +++++++++++++++++-
 1 file changed, 187 insertions(+), 7 deletions(-)

diff --git a/playbooks/python/workflows/dynamic-kconfig/gen-dynamic-pci.py b/playbooks/python/workflows/dynamic-kconfig/gen-dynamic-pci.py
index f86ce74..52824c7 100755
--- a/playbooks/python/workflows/dynamic-kconfig/gen-dynamic-pci.py
+++ b/playbooks/python/workflows/dynamic-kconfig/gen-dynamic-pci.py
@@ -64,11 +64,156 @@ def get_special_device_nvme(pci_id, IOMMUGroup):
     )
 
 
-def get_kconfig_device_name(pci_id, sdevice, IOMMUGroup):
+def get_gpu_memory_info(pci_id):
+    """Try to get GPU memory information from sysfs."""
+    # Try to get memory info from various possible locations
+    mem_paths = [
+        sys_bus_prefix + pci_id + "/mem_info_vram_total",
+        sys_bus_prefix + pci_id + "/drm/card0/mem_info_vram_total",
+    ]
+
+    for mem_path in mem_paths:
+        if os.path.isfile(mem_path):
+            try:
+                with open(mem_path, "r") as f:
+                    mem_bytes = int(f.read().strip())
+                    mem_gb = mem_bytes / (1024 * 1024 * 1024)
+                    return f"{mem_gb:.0f}GB"
+            except:
+                pass
+
+    # Try to get memory from resource file size (less accurate)
+    resource_path = sys_bus_prefix + pci_id + "/resource0"
+    if os.path.isfile(resource_path):
+        try:
+            size = os.path.getsize(resource_path)
+            if size > 0:
+                size_gb = size / (1024 * 1024 * 1024)
+                if size_gb >= 1:
+                    return f"{size_gb:.0f}GB"
+        except:
+            pass
+
+    return None
+
+
+def get_special_device_gpu(pci_id, device_name, IOMMUGroup):
+    """Generate a nice display name for GPU devices."""
+    pci_id_name = strip_kconfig_name(pci_id)
+
+    # Clean up the device name to extract the GPU model
+    gpu_model = device_name
+
+    # Common GPU name patterns to clean up
+    replacements = [
+        ("[AMD/ATI]", "AMD"),
+        ("Advanced Micro Devices, Inc.", "AMD"),
+        ("NVIDIA Corporation", "NVIDIA"),
+        ("Intel Corporation", "Intel"),
+    ]
+
+    for old, new in replacements:
+        gpu_model = gpu_model.replace(old, new).strip()
+
+    # Try to extract specific GPU model from brackets
+    import re
+
+    bracket_match = re.search(r"\[([^\]]+)\]", gpu_model)
+    if bracket_match:
+        model_name = bracket_match.group(1)
+        # Keep the vendor prefix if it's a clean model name
+        if "Radeon" in model_name or "GeForce" in model_name or "Intel" in model_name:
+            gpu_model = model_name
+        else:
+            # Prepend vendor if needed
+            if "AMD" in gpu_model and "Radeon" not in model_name:
+                gpu_model = f"AMD {model_name}"
+            elif (
+                "NVIDIA" in gpu_model
+                and "GeForce" not in model_name
+                and "Quadro" not in model_name
+            ):
+                gpu_model = f"NVIDIA {model_name}"
+            else:
+                gpu_model = model_name
+
+    # Remove any existing memory info from the model name (e.g., "32GB" at the end)
+    gpu_model = re.sub(r"\s+\d+GB\s*$", "", gpu_model)
+
+    # Try to get memory info
+    mem_info = get_gpu_memory_info(pci_id)
+
+    # Build the display name
+    if mem_info:
+        display_name = (
+            f"{pci_id_name} IOMMU group {IOMMUGroup} - GPU - {gpu_model} {mem_info}"
+        )
+    else:
+        display_name = f"{pci_id_name} IOMMU group {IOMMUGroup} - GPU - {gpu_model}"
+
+    return display_name
+
+
+def is_gpu_device(device_name):
+    """Check if a device is a GPU based on its name."""
+    gpu_keywords = [
+        # AMD/ATI
+        "Radeon",
+        "Vega",
+        "Navi",
+        "RDNA",
+        "GCN",
+        "Polaris",
+        "Fiji",
+        "Instinct",
+        "FirePro",
+        "FireGL",
+        "RX",
+        "AMD.*GPU",
+        # NVIDIA
+        "GeForce",
+        "Quadro",
+        "Tesla",
+        "NVIDIA.*GPU",
+        "GTX",
+        "RTX",
+        "Titan",
+        "NVS",
+        "GRID",
+        # Intel
+        "Intel.*Graphics",
+        "UHD Graphics",
+        "HD Graphics",
+        "Iris",
+        "Arc",
+        "Xe Graphics",
+        # Generic
+        "VGA compatible controller",
+        "Display controller",
+        "3D controller",
+        "Graphics",
+    ]
+
+    device_lower = device_name.lower()
+    for keyword in gpu_keywords:
+        if keyword.lower() in device_lower:
+            return True
+    return False
+
+
+def get_kconfig_device_name(
+    pci_id, sdevice, IOMMUGroup, vendor_name=None, device_name=None
+):
     default_name = "%s IOMMU group %s - %s" % (pci_id, IOMMUGroup, sdevice)
     special_name = None
+
+    # Check for NVMe devices
     if os.path.isdir(sys_bus_prefix + pci_id + "/nvme"):
         special_name = get_special_device_nvme(pci_id, IOMMUGroup)
+    # Check for GPU devices
+    elif device_name and is_gpu_device(device_name):
+        special_name = get_special_device_gpu(pci_id, device_name, IOMMUGroup)
+
     if not special_name:
         return strip_kconfig_name(default_name)
     return strip_kconfig_name(special_name)
@@ -107,10 +252,21 @@ def add_pcie_kconfig_target(config_name, sdevice):
 
 
 def add_pcie_kconfig_entry(
-    pci_id, sdevice, domain, bus, slot, function, IOMMUGroup, config_id
+    pci_id,
+    sdevice,
+    domain,
+    bus,
+    slot,
+    function,
+    IOMMUGroup,
+    config_id,
+    vendor_name=None,
+    device_name=None,
 ):
     prefix = passthrough_prefix + "_%04d" % config_id
-    name = get_kconfig_device_name(pci_id, sdevice, IOMMUGroup)
+    name = get_kconfig_device_name(
+        pci_id, sdevice, IOMMUGroup, vendor_name, device_name
+    )
     add_pcie_kconfig_name(prefix, name)
     add_pcie_kconfig_target(prefix, sdevice)
     add_pcie_kconfig_string(prefix, pci_id, "pci_id")
@@ -123,7 +279,9 @@ def add_pcie_kconfig_entry(
     add_pcie_kconfig_string(prefix, function, "function")
 
 
-def add_new_device(slot, sdevice, IOMMUGroup, possible_id):
+def add_new_device(
+    slot, sdevice, IOMMUGroup, possible_id, vendor_name=None, device_name=None
+):
     # Example expeced format 0000:2d:00.0
     m = re.match(
         r"^(?P<DOMAIN>\w+):" "(?P<BUS>\w+):" "(?P<MSLOT>\w+)\." "(?P<FUNCTION>\w+)$",
@@ -154,7 +312,16 @@ def add_new_device(slot, sdevice, IOMMUGroup, possible_id):
         )
 
     add_pcie_kconfig_entry(
-        slot, sdevice, domain, bus, mslot, function, IOMMUGroup, possible_id
+        slot,
+        sdevice,
+        domain,
+        bus,
+        mslot,
+        function,
+        IOMMUGroup,
+        possible_id,
+        vendor_name,
+        device_name,
     )
 
     return possible_id
@@ -184,6 +351,8 @@ def main():
     slot = -1
     sdevice = None
     IOMMUGroup = None
+    vendor_name = None
+    device_name = None
 
     for line in all_lines:
         line = line.strip()
@@ -197,20 +366,31 @@ def main():
         if tag == "Slot":
             if sdevice:
                 num_candidate_devices = add_new_device(
-                    slot, sdevice, IOMMUGroup, num_candidate_devices
+                    slot,
+                    sdevice,
+                    IOMMUGroup,
+                    num_candidate_devices,
+                    vendor_name,
+                    device_name,
                 )
             slot = data
             sdevice = None
             IOMMUGroup = None
+            vendor_name = None
+            device_name = None
         elif tag == "SDevice":
             sdevice = data
         elif tag == "IOMMUGroup":
             IOMMUGroup = data
+        elif tag == "Vendor":
+            vendor_name = data
+        elif tag == "Device":
+            device_name = data
 
     # Handle the last device
     if sdevice and slot:
         num_candidate_devices = add_new_device(
-            slot, sdevice, IOMMUGroup, num_candidate_devices
+            slot, sdevice, IOMMUGroup, num_candidate_devices, vendor_name, device_name
         )
 
     add_pcie_kconfig_string(passthrough_prefix, num_candidate_devices, "NUM_DEVICES")
-- 
2.47.2

[PATCH 4/8] guestfs / devconfig: allow for upgrade from Debian testing to forky

[Luis Chamberlain]

Debian Trixie (13) was just released as stable, and the testing
repository has transitioned from codename 'trixie' to 'forky'. This
causes apt-get update to fail with:

  E:Repository 'http://mirror.keystealth.org/debian testing InRelease'
  changed its 'Codename' value from 'trixie' to 'forky'

Fix by adding a preliminary apt-get update with the
--allow-releaseinfo-change flag to accept the repository metadata
changes before installing packages. This ensures the playbook can
handle Debian testing transitions gracefully.

We do this for both the host through the guestfs playbook and
and the target node through the devconfig playbook.

Generated-by: Claude AI
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
---
 .../roles/devconfig/tasks/install-deps/debian/main.yml   | 9 +++++++++
 .../roles/guestfs/tasks/install-deps/debian/main.yml     | 8 ++++++++
 2 files changed, 17 insertions(+)

diff --git a/playbooks/roles/devconfig/tasks/install-deps/debian/main.yml b/playbooks/roles/devconfig/tasks/install-deps/debian/main.yml
index 6ffe00f..140c3a1 100644
--- a/playbooks/roles/devconfig/tasks/install-deps/debian/main.yml
+++ b/playbooks/roles/devconfig/tasks/install-deps/debian/main.yml
@@ -45,6 +45,15 @@
   ignore_errors: yes
   tags: firstconfig
 
+- name: Update apt cache accepting release info changes
+  become: yes
+  become_method: sudo
+  ansible.builtin.command:
+    cmd: apt-get update --allow-releaseinfo-change
+  changed_when: false
+  ignore_errors: true
+  tags: firstconfig
+
 - name: Upgrade Packages
   become: yes
   become_method: sudo
diff --git a/playbooks/roles/guestfs/tasks/install-deps/debian/main.yml b/playbooks/roles/guestfs/tasks/install-deps/debian/main.yml
index 0529db6..dc69c65 100644
--- a/playbooks/roles/guestfs/tasks/install-deps/debian/main.yml
+++ b/playbooks/roles/guestfs/tasks/install-deps/debian/main.yml
@@ -1,4 +1,12 @@
 ---
+- name: Update apt cache accepting release info changes
+  become: true
+  become_method: ansible.builtin.sudo
+  ansible.builtin.command:
+    cmd: apt-get update --allow-releaseinfo-change
+  changed_when: false
+  ignore_errors: true
+
 - name: Install guestfs dependencies for Debian
   become: true
   become_method: ansible.builtin.sudo
-- 
2.47.2

[PATCH 5/8] guestfs: add OVMF package for UEFI firmware support

[Luis Chamberlain]

The OVMF (Open Virtual Machine Firmware) package provides UEFI firmware
files required by libvirt when creating VMs with UEFI support. Without
this package, libvirt fails with:

  libvirtError: operation failed: Unable to find 'efi' firmware that
  is compatible with the current configuration

Add ovmf to the Debian dependencies to ensure UEFI firmware is available
for VM creation, particularly when guestfs_requires_uefi is enabled.

Generated-by: Claude AI
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
---
 playbooks/roles/guestfs/tasks/install-deps/debian/main.yml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/playbooks/roles/guestfs/tasks/install-deps/debian/main.yml b/playbooks/roles/guestfs/tasks/install-deps/debian/main.yml
index dc69c65..1626e75 100644
--- a/playbooks/roles/guestfs/tasks/install-deps/debian/main.yml
+++ b/playbooks/roles/guestfs/tasks/install-deps/debian/main.yml
@@ -17,4 +17,5 @@
       - isc-dhcp-client
       - python3-lxml
       - python3-libvirt
+      - ovmf
     state: present
-- 
2.47.2

[PATCH 6/8] gen_nodes: fix duplicate pcipassthrough keys in YAML output

[Luis Chamberlain]

The template was generating multiple 'pcipassthrough' keys for each
device, causing YAML parsing warnings about duplicate mapping keys.

Fixed by collecting all devices for a target guest first, then
outputting a single 'pcipassthrough' key with all devices under it.
This uses Jinja2's namespace feature to maintain state across loop
iterations.

Generated-by: Claude AI
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
---
 .../roles/gen_nodes/templates/gen_nodes_list.j2   | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/playbooks/roles/gen_nodes/templates/gen_nodes_list.j2 b/playbooks/roles/gen_nodes/templates/gen_nodes_list.j2
index e8f5e34..6ad2af1 100644
--- a/playbooks/roles/gen_nodes/templates/gen_nodes_list.j2
+++ b/playbooks/roles/gen_nodes/templates/gen_nodes_list.j2
@@ -1,18 +1,19 @@
 {%- macro gen_pci_passthrough_each(n, device_list) -%}
-{%- set found_pcie_devices_on_node = False -%}
+{%- set ns = namespace(has_devices=false, device_index=1) -%}
 {%- for d in device_list %}
 {%- if n == d.target_guest -%}
-{%- if not found_pcie_devices_on_node -%}
-{% set found_pcie_devices_on_node = True %}
+{%- if not ns.has_devices -%}
+{%- set ns.has_devices = true %}
     pcipassthrough:
-{% endif %}
-      passthrough{{ loop.index | string }}:
+{% endif -%}
+      passthrough{{ ns.device_index | string }}:
         domain: {{ d.domain }}
         bus: {{ d.bus }}
         slot: {{ d.slot }}
         function: {{ d.function }}
-{% endif %}
-{% endfor -%}
+{%- set ns.device_index = ns.device_index + 1 -%}
+{%- endif %}
+{%- endfor -%}
 {%- endmacro -%}
 {%- macro gen_pci_passthrough( device_list ) -%}
 {%- for d in device_list %}
-- 
2.47.2

[PATCH 7/8] gen_nodes: fix YAML indentation in pcipassthrough template

[Luis Chamberlain]

The template was generating malformed YAML with improper indentation
and missing newlines between passthrough entries. This caused parsing
errors: "Colons in unquoted values must be followed by a non-space
character."

Fixed by:
- Removing the trailing whitespace stripping after 'pcipassthrough:'
- Ensuring proper indentation for all passthrough entries
- Maintaining consistent whitespace control throughout the macro

The generated YAML now correctly produces a single 'pcipassthrough'
key with properly indented sub-entries for all devices.

Generated-by: Claude AI
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
---
 playbooks/roles/gen_nodes/templates/gen_nodes_list.j2 | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/playbooks/roles/gen_nodes/templates/gen_nodes_list.j2 b/playbooks/roles/gen_nodes/templates/gen_nodes_list.j2
index 6ad2af1..bb4e053 100644
--- a/playbooks/roles/gen_nodes/templates/gen_nodes_list.j2
+++ b/playbooks/roles/gen_nodes/templates/gen_nodes_list.j2
@@ -5,14 +5,13 @@
 {%- if not ns.has_devices -%}
 {%- set ns.has_devices = true %}
     pcipassthrough:
-{% endif -%}
-      passthrough{{ ns.device_index | string }}:
+{% endif %}      passthrough{{ ns.device_index | string }}:
         domain: {{ d.domain }}
         bus: {{ d.bus }}
         slot: {{ d.slot }}
         function: {{ d.function }}
-{%- set ns.device_index = ns.device_index + 1 -%}
-{%- endif %}
+{% set ns.device_index = ns.device_index + 1 -%}
+{%- endif -%}
 {%- endfor -%}
 {%- endmacro -%}
 {%- macro gen_pci_passthrough( device_list ) -%}
-- 
2.47.2

[PATCH 8/8] guestfs: fix PCIe passthrough device attachment

[Luis Chamberlain]

The virsh attach-device command was receiving the entire file
information dictionary instead of just the file path, causing:

  error: Failed to open file '{'path': '/path/to/file.xml', ...}'

Fixed by using {{ item.path }} instead of {{ item }} to extract
just the file path from the find results.

Generated-by: Claude AI
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
---
 playbooks/roles/guestfs/tasks/bringup/main.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/playbooks/roles/guestfs/tasks/bringup/main.yml b/playbooks/roles/guestfs/tasks/bringup/main.yml
index af95c5d..deb0846 100644
--- a/playbooks/roles/guestfs/tasks/bringup/main.yml
+++ b/playbooks/roles/guestfs/tasks/bringup/main.yml
@@ -119,7 +119,7 @@
           - "virsh"
           - "attach-device"
           - "{{ inventory_hostname }}"
-          - "{{ item }}"
+          - "{{ item.path }}"
           - "--config"
       loop: "{{ passthrough_devices.files }}"
       loop_control:
-- 
2.47.2

Re: [PATCH 0/8] various fixes

[Chuck Lever]

On 8/11/25 4:12 PM, Luis Chamberlain wrote:
> I took a GPU out for a spin for PCI passthrough and found that the
> vagrant removal nuked some things for it, this fixes that and enhances
> the PCI passthrough listing with sensible names.

Very pleased to see this!

Reviewed-by: Chuck Lever <chuck.lever@oracle.com>


> While at it, this also includes a fix to gest us through the debian
> trixie to forky transition, allowing our host and target nodes to
> upgrade.
> 
> Luis Chamberlain (8):
>   libvirt: fix PCI-E passthrough configuration after Vagrant removal
>   gen-dynamic-pci: fix PCI passthrough target guest default value
>   gen-dynamic-pci: add GPU detection and naming support
>   guestfs / devconfig: allow for upgrade from Debian testing to forky
>   guestfs: add OVMF package for UEFI firmware support
>   gen_nodes: fix duplicate pcipassthrough keys in YAML output
>   gen_nodes: fix YAML indentation in pcipassthrough template
>   guestfs: fix PCIe passthrough device attachment
> 
>  kconfigs/Kconfig.libvirt                      |  43 ++++
>  .../dynamic-kconfig/gen-dynamic-pci.py        | 198 +++++++++++++++++-
>  .../tasks/install-deps/debian/main.yml        |   9 +
>  .../gen_nodes/templates/gen_nodes_list.j2     |  14 +-
>  .../gen_nodes/templates/guestfs_nodes.j2      |   2 +-
>  playbooks/roles/gen_nodes/templates/hosts.j2  |   4 +-
>  .../roles/guestfs/tasks/bringup/main.yml      |   2 +-
>  .../tasks/install-deps/debian/main.yml        |   9 +
>  scripts/check_pciepassthrough_kconfig.sh      |  15 +-
>  9 files changed, 275 insertions(+), 21 deletions(-)
> 


-- 
Chuck Lever

Re: [PATCH 0/8] various fixes

[Luis Chamberlain]

On Mon, Aug 11, 2025 at 04:38:10PM -0400, Chuck Lever wrote:
> On 8/11/25 4:12 PM, Luis Chamberlain wrote:
> > I took a GPU out for a spin for PCI passthrough and found that the
> > vagrant removal nuked some things for it, this fixes that and enhances
> > the PCI passthrough listing with sensible names.
> 
> Very pleased to see this!
> 
> Reviewed-by: Chuck Lever <chuck.lever@oracle.com>

Great thanks, pushed, I added just one more minor fix patch onto
the queue for make destroy.

  Luis