[Test Purpose] BMI: add bare-metal-instance platform

examples/runbook/bmi.yml

@@ -0,0 +1,73 @@
+# BMI (Bare-Metal Instance) example runbook.
+#
+# Deploys two GB200 BMIs behind a jumphost using the Bicep template at
+# ``lisa/sut_orchestrator/bmi/bmi.bicep`` (compiled to
+# ``autogen_bmi_template.json``) and runs the smoke test suite against them.
+#
+# Required variables (override via -v on the command line or env vars):
+#   - subscription_id            : Azure subscription for the RG.
+#   - bmi_image_sig              : SIG image version resource ID.
+#   - bmi_private_key_file       : SSH key whose ``.pub`` is installed on BMIs.
+#
+# Optional auth overrides:
+#   - jumphost_password          : If set, jumphost uses password auth.
+#                                  Leave empty to auto-generate an SSH
+#                                  keypair (recommended).
+#
+# Optional jumphost overrides (have sensible defaults):
+#   - jumphost_vm_size           : Jumphost VM size (default Standard_DS2_v2).
+#   - jumphost_image             : Jumphost image URN
+#                                  (publisher:offer:sku:version).
+#   - jumphost_username          : Jumphost admin user (default lisatest).
+
+name: bmi_smoke
+test_project: BMI
+test_pass: smoke
+
+import_builtin_tests: true
+
+variable:
+  - name: subscription_id
+    value: ""
+  - name: location
+    value: "southeastus5"
+  - name: bmi_vm_size
+    value: "ND144isr_ETH_GB200_metal_v6"
+  - name: bmi_host_sku
+    value: "GPCv6GB200S186_ETH_metal-Type1"
+  - name: bmi_image_sig
+    value: ""
+  - name: bmi_admin_username
+    value: "azhpcuser"
+  - name: bmi_private_key_file
+    value: "./id_rsa"
+  - name: jumphost_vm_size
+    value: "Standard_DS2_v2"
+  - name: jumphost_image
+    value: "Canonical:0001-com-ubuntu-server-jammy:22_04-lts-gen2:latest"
+  - name: jumphost_username
+    value: "lisatest"
+  - name: jumphost_password
+    value: ""
+    is_secret: true
+
+platform:
+  - type: bmi
+    bmi:
+      subscription_id: $(subscription_id)
+      location: $(location)
+      bmi_vm_size: $(bmi_vm_size)
+      bmi_host_sku: $(bmi_host_sku)
+      bmi_image_sig: $(bmi_image_sig)
+      bmi_admin_username: $(bmi_admin_username)
+      bmi_private_key_file: $(bmi_private_key_file)
+      jumphost_vm_size: $(jumphost_vm_size)
+      jumphost_image: $(jumphost_image)
+      jumphost_username: $(jumphost_username)
+      jumphost_password: $(jumphost_password)
+      delete_on_cleanup: true
+
+testcase:
+  - criteria:
+      area: provisioning
+      priority: 0

lisa/mixin_modules.py

@@ -20,6 +20,14 @@
 import lisa.runners.lisa_runner  # noqa: F401
 import lisa.sut_orchestrator.ready  # noqa: F401
 
+# BMI (Azure bare-metal instance) platform. Optional: depends on azure-mgmt-*
+# and paramiko. Imported lazily so a missing azure SDK does not break LISA
+# for users that only run other platforms.
+try:
+    import lisa.sut_orchestrator.bmi.platform_  # noqa: F401
+except ModuleNotFoundError as e:
+    print(f"bmi package is not installed. [{e}]")
+
 try:
     import lisa.notifiers.html  # noqa: F401
 except ModuleNotFoundError as e:

lisa/nic.py

@@ -529,6 +529,13 @@ def _load_nics(self) -> None:
             sriov_match = self.__nic_pci_device_regex.search(line)
             if sriov_match:
                 nic_name, lower, pci_slot = sriov_match.groups()
+                # Skip virtual master devices (VRF, bridge, bond) that own
+                # physical NICs via /sys/class/net/<master>/lower_* symlinks.
+                # These are not real interfaces and have no device/driver
+                # backing, so they must not be added to self.nics (which is
+                # iterated by _get_nics_driver, load_nics_info, etc.).
+                if nic_name not in self._nic_names:
+                    continue
                 used_module = lspci.get_used_module(pci_slot)
                 self.append(
                     NicInfo(
@@ -665,13 +672,85 @@ def _get_default_nic(self) -> None:
         self.default_nic, self.default_nic_route = self._node.tools[
             Ip
         ].get_default_route_info()
+        # On baremetal/VRF-enabled hosts, the default route may go via a
+        # virtual master device (VRF, bridge, bond) which has no PCI/driver
+        # backing under /sys/class/net/<name>/device. Resolve such masters to
+        # an enslaved physical NIC so downstream tools (driver lookup,
+        # datapath detection, perf tests) operate on a real interface.
+        if self.default_nic not in self._nic_names:
+            resolved = self._resolve_master_to_slave_nic(self.default_nic)
+            if resolved:
+                self._node.log.debug(
+                    f"Default route NIC '{self.default_nic}' is a virtual "
+                    f"master device; using enslaved physical NIC "
+                    f"'{resolved}' instead."
+                )
+                self.default_nic = resolved
+            else:
+                # No slave under the master (e.g. VRF with the physical NIC
+                # not enslaved). Fall back to the first physical NIC that
+                # owns a global-scope IPv4 address.
+                resolved = self._find_first_nic_with_global_ipv4()
+                if resolved:
+                    self._node.log.debug(
+                        f"Default route NIC '{self.default_nic}' is not a "
+                        f"physical interface and has no slaves; using "
+                        f"physical NIC '{resolved}' (first with global "
+                        f"IPv4) instead."
+                    )
+                    self.default_nic = resolved
         assert_that(self.default_nic in self._nic_names).described_as(
             (
                 f"ERROR: NIC name found as default {self.default_nic} "
                 f"was not in original list of nics {repr(self._nic_names)}."
             )
         ).is_true()
 
+    def _resolve_master_to_slave_nic(self, master: str) -> str:
+        # Find a physical NIC enslaved to `master` (VRF, bridge, bond).
+        # `ip -o link show` lines containing "master <name>" identify slaves.
+        # Returns the first slave that is also in self._nic_names.
+        result = self._node.execute(
+            "ip -o link show",
+            shell=True,
+            sudo=True,
+        )
+        if result.exit_code != 0:
+            return ""
+        for line in result.stdout.splitlines():
+            if f"master {master} " not in line and not line.rstrip().endswith(
+                f"master {master}"
+            ):
+                continue
+            # Line format: "<idx>: <name>[@<parent>]: <flags> ... master <m> ..."
+            parts = line.split(":", 2)
+            if len(parts) < 2:
+                continue
+            name = parts[1].strip().split("@")[0].strip()
+            if name in self._nic_names:
+                return name
+        return ""
+
+    def _find_first_nic_with_global_ipv4(self) -> str:
+        # Pick the first physical NIC carrying a global-scope IPv4 address.
+        # `ip -o -4 addr show scope global` lists one address per line:
+        #   "<idx>: <name>    inet 10.0.1.4/24 ..."
+        result = self._node.execute(
+            "ip -o -4 addr show scope global",
+            shell=True,
+            sudo=True,
+        )
+        if result.exit_code != 0:
+            return ""
+        for line in result.stdout.splitlines():
+            parts = line.split()
+            if len(parts) < 2:
+                continue
+            name = parts[1].split("@")[0]
+            if name in self._nic_names:
+                return name
+        return ""
+
     def is_module_reloadable(self, module_name: str) -> bool:
         return self._node.tools[KernelConfig].is_built_as_module(
             self._device_module_map[module_name].config

lisa/sut_orchestrator/__init__.py

@@ -1,6 +1,11 @@
 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT license.
-from lisa.util.constants import PLATFORM_BAREMETAL, PLATFORM_HYPERV, PLATFORM_READY
+from lisa.util.constants import (
+    PLATFORM_BAREMETAL,
+    PLATFORM_BMI,
+    PLATFORM_HYPERV,
+    PLATFORM_READY,
+)
 
 AZURE = "azure"
 QEMU = "qemu"
@@ -10,3 +15,4 @@
 READY = PLATFORM_READY
 BAREMETAL = PLATFORM_BAREMETAL
 HYPERV = PLATFORM_HYPERV
+BMI = PLATFORM_BMI

lisa/sut_orchestrator/azure/credential.py

@@ -1,7 +1,7 @@
 import os
 from dataclasses import dataclass
 from enum import Enum
-from typing import Any, Type, cast
+from typing import Any, Optional, Type, cast
 
 from azure.identity import (
     AzureCliCredential,
@@ -101,7 +101,9 @@ def __init__(
         if runbook.type:
             self._credential_type = runbook.type
         else:
-            self._credential_type = AzureCredentialType.DefaultAzureCredential  # CodeQL [SM05139] Okay use of DefaultAzureCredential as it is only used in development # noqa E501
+            self._credential_type = (
+                AzureCredentialType.DefaultAzureCredential
+            )  # CodeQL [SM05139] Okay use of DefaultAzureCredential as it is only used in development # noqa E501
 
         self._log.debug(f"Credential type: {self._credential_type}")
         self._cloud = cloud
@@ -169,7 +171,7 @@ def get_credential(self) -> Any:
         return AzureCredential with related schema
         """
         additional_tenants = ["*"] if self._allow_all_tenants else None
-        return DefaultAzureCredential( # CodeQL [SM05139] Okay use of DefaultAzureCredential as it is only used in development # noqa E501
+        return DefaultAzureCredential(  # CodeQL [SM05139] Okay use of DefaultAzureCredential as it is only used in development # noqa E501
             cloud=self._cloud,
             additionally_allowed_tenants=additional_tenants,
         )
@@ -427,3 +429,40 @@ def get_credential(self) -> Any:
             tenant_id=self._tenant_id,
             additionally_allowed_tenants=additionally_allowed_tenants,
         )
+
+
+def build_compute_credential(
+    *,
+    credential_schema: Optional[AzureCredentialSchema],
+    service_principal_tenant_id: str,
+    service_principal_client_id: str,
+    service_principal_key: str,
+    azure_arm_access_token: str,
+    cloud: Cloud,
+    logger: Logger,
+) -> Any:
+    """
+    Build a TokenCredential from the same auth surface AzurePlatformSchema
+    exposes (credential subobject + service-principal fields + ARM token).
+    Mirrors ``AzurePlatform._initialize_credential`` resolution order so
+    other Azure-based platforms (e.g. BMI) get identical auth behavior.
+    """
+    if credential_schema:
+        factory = subclasses.Factory[AzureCredential](AzureCredential)
+        azure_credential: AzureCredential = factory.create_by_runbook(
+            credential_schema, cloud=cloud, logger=logger
+        )
+        return azure_credential.get_credential()
+
+    if service_principal_tenant_id:
+        os.environ["AZURE_TENANT_ID"] = service_principal_tenant_id
+    if service_principal_client_id:
+        os.environ["AZURE_CLIENT_ID"] = service_principal_client_id
+    if service_principal_key:
+        os.environ["AZURE_CLIENT_SECRET"] = service_principal_key
+
+    return get_static_access_token(
+        azure_arm_access_token
+    ) or DefaultAzureCredential(  # CodeQL [SM05139] dev fallback # noqa E501
+        authority=cloud.endpoints.active_directory,
+    )

lisa/sut_orchestrator/bmi/__init__.py

@@ -0,0 +1,5 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT license.
+
+# Register BMI pluggy hookspecs on package import so extensions can hook in.
+from . import hooks  # noqa: F401