Clarify PCI in placement scenario and storagemgmt use

Signed-off-by: Bohdan Dobrelia <[email protected]>

Author: bogdando

examples/dt/nova/nova04delta/README.md

@@ -1,6 +1,6 @@
 # Nova GPU Passthrough (VFIO)
 
-This directory contains the necessary configurations to deploy OpenStack with Nova configured for full GPU device passthrough (VFIO). This setup allows entire physical GPUs on compute nodes to be passed directly to virtual machines, providing near-native performance.
+This directory contains the necessary configurations to deploy OpenStack with Nova configured for full GPU device passthrough (VFIO). This setup allows entire physical GPUs on compute nodes to be passed directly to virtual machines, providing near-native performance. Nova control plane is configured for requesting PCI devices from Placement.
 
 ## Overview
 
@@ -30,6 +30,11 @@ The following parameters are crucial for host-level configuration:
 
 ## Nova Configuration
 
+Nova control plane is configured for requesting PCI devices from Placement
+through `resources:VGPU=X` flavor extra specs.
+That is a contrary to the legacy mode where PCI devices used to be requested through
+`pci_passthrough:alias` flavor extra specs.
+
 ### Control Plane (`examples/dt/nova/nova04delta/service-values.yaml`)
 
 *   `[pci]alias`: Creates an alias for a specific GPU type. This allows users to request a GPU by a friendly name (e.g., `nvidia_a2`) when creating a VM.
@@ -69,6 +74,4 @@ In addition to PCI device configuration, the `nova.compute.conf` section include
 
 ## Guest VM
 
-To use the passthrough GPU, the guest operating system inside the VM must have the appropriate native NVIDIA driver installed. You will need a standard NVIDIA driver. Do not use vGPU-enabled guest drivers. The GPU will appear as a physical PCI device within the guest.
-
-// NOTE(bogdando) OSPRH-18880: keep in mind that for PCI in Nova mode, a flavor must use pci_passthrough:alias property. While for PCI in Placement mode, it must be resources:VGPU=X annotation. See https://docs.openstack.org/nova/latest/admin/pci-passthrough.html.
+To use the passthrough GPU, the guest operating system inside the VM must have the appropriate native NVIDIA driver installed. You will need a standard NVIDIA driver. Do not use vGPU-enabled guest drivers. The GPU will appear as a physical PCI device within the guest.

examples/va/nvidia-vfio-passthrough/README.md

@@ -14,6 +14,20 @@ with nested virtualization and requires real baremetal hosts.
 
 ## Environment
 
+### Nova
+
+Nova control plane is configured for requesting PCI devices from Placement
+by setting `resources:VGPU=X` in flavor extra specs.
+That is a contrary to the legacy mode where PCI devices used to be requested through
+`pci_passthrough:alias` flavor extra specs.
+
+### Guest VM
+
+To use the passthrough GPU, the guest operating system inside the VM must have
+the appropriate native NVIDIA driver installed. You will need a standard NVIDIA
+driver. Do not use vGPU-enabled guest drivers. The GPU will appear as a physical
+PCI device within the guest.
+
 ### Nodes
 
 | Role                        | Machine Type | Count |
@@ -40,6 +54,10 @@ with nested virtualization and requires real baremetal hosts.
 | storagemgmt | VLAN tagged | 172.20.0.0/24     |
 | tenant      | VLAN tagged | 172.19.0.0/24     |
 
+NOTE: A network attach definition is not provided for storagemgmt because only
+RHEL EDPM nodes are going to use it, while control plane pods on OpenShift do
+not need it.
+
 ## Stages
 All stages must be executed in the order listed below. Everything is required unless otherwise indicated.
 

examples/va/nvidia-vfio-passthrough/control-plane/nncp/values.yaml

@@ -123,6 +123,16 @@ data:
           "range_end": "172.18.0.70"
         }
       }
+  storagemgmt:  # used on RHEL, not used on OpenShift
+    dnsDomain: storagemgmt.example.com
+    subnets:
+      - allocationRanges:
+          - end: 172.20.0.250
+            start: 172.20.0.100
+        cidr: 172.20.0.0/24
+        name: subnet1
+        vlan: 23
+    mtu: 1500
   tenant:
     dnsDomain: tenant.example.com
     subnets: