petri/vm/openvmm/

modify.rs

// Copyright (c) Microsoft Corporation.
// Licensed under the MIT License.

//! Helpers to modify a [`PetriVmConfigOpenVmm`] from its defaults.

// TODO: Delete all modification functions that are not backend-specific
// from this file, add necessary settings to the backend-agnostic
// `PetriVmConfig`, and add corresponding functions to `PetriVmBuilder`.

use super::MANA_INSTANCE;
use super::NIC_MAC_ADDRESS;
use super::PetriVmConfigOpenVmm;
use chipset_resources::battery::BatteryDeviceHandleX64;
use chipset_resources::battery::HostBatteryUpdate;
use disk_backend_resources::LayeredDiskHandle;
use disk_backend_resources::layer::RamDiskLayerHandle;
use gdma_resources::GdmaDeviceHandle;
use gdma_resources::VportDefinition;
use get_resources::ged::IgvmAttestTestConfig;
use guid::Guid;
use net_backend_resources::mac_address::MacAddress;
use nvme_resources::NamespaceDefinition;
use nvme_resources::NvmeControllerHandle;
use openvmm_defs::config::Config;
use openvmm_defs::config::DeviceVtl;
use openvmm_defs::config::LoadMode;
use openvmm_defs::config::PcieDeviceConfig;
use openvmm_defs::config::PcieIommuConfig;
use openvmm_defs::config::PcieMmioRangeConfig;
use openvmm_defs::config::PciePortConfig;
use openvmm_defs::config::PcieRootComplexConfig;
use openvmm_defs::config::PcieSwitchConfig;
use openvmm_defs::config::VpciDeviceConfig;
use openvmm_defs::config::Vtl2BaseAddressType;
use vm_resource::IntoResource;
use vmotherboard::ChipsetDeviceHandle;

impl PetriVmConfigOpenVmm {
    /// Enable the VTL0 alias map.
    // TODO: Remove once #912 is fixed.
    pub fn with_vtl0_alias_map(mut self) -> Self {
        self.config
            .hypervisor
            .with_vtl2
            .as_mut()
            .expect("Not an openhcl config.")
            .vtl0_alias_map = true;
        self
    }

    /// Enable the battery for the VM.
    pub fn with_battery(mut self) -> Self {
        if self.resources.properties.is_openhcl {
            self.ged.as_mut().unwrap().enable_battery = true;
        } else {
            self.config.chipset_devices.push(ChipsetDeviceHandle {
                name: "battery".to_string(),
                resource: BatteryDeviceHandleX64 {
                    battery_status_recv: {
                        let (tx, rx) = mesh::channel();
                        tx.send(HostBatteryUpdate::default_present());
                        rx
                    },
                }
                .into_resource(),
            });
            if let LoadMode::Uefi { enable_battery, .. } = &mut self.config.load_mode {
                *enable_battery = true;
            }
        }
        self
    }

    /// Set test config for the GED's IGVM attest request handler
    pub fn with_igvm_attest_test_config(mut self, config: IgvmAttestTestConfig) -> Self {
        if !self.resources.properties.is_openhcl {
            panic!("IGVM Attest test config is only supported for OpenHCL.")
        };

        let ged = self.ged.as_mut().expect("No GED to configure TPM");

        ged.igvm_attest_test_config = Some(config);

        self
    }

    /// Override the SMBIOS identity delivered to the guest, regardless of how
    /// the VM is loaded.
    ///
    /// For OpenHCL the identity is forwarded to the paravisor over the Guest
    /// Emulation Transport (GET), which synthesizes the guest's DMI tables from
    /// it. For direct OpenVMM boot (Linux direct, UEFI, or PCAT) it is applied
    /// to the loader's SMBIOS config. Each load path honors only the subset of
    /// fields it can express and fails closed on the rest.
    pub fn with_smbios(mut self, f: impl FnOnce(&mut smbios_defs::SmbiosConfig)) -> Self {
        if self.resources.properties.is_openhcl {
            let ged = self.ged.as_mut().expect("OpenHCL config must have a GED.");
            f(&mut ged.smbios);
        } else {
            let smbios = match &mut self.config.load_mode {
                LoadMode::Linux { smbios, .. }
                | LoadMode::Uefi { smbios, .. }
                | LoadMode::Pcat { smbios, .. } => &mut **smbios,
                LoadMode::Igvm { .. } | LoadMode::None => {
                    panic!("SMBIOS configuration is not supported for this load mode.")
                }
            };
            f(smbios);
        }
        self
    }

    /// Enable a synthnic for the VM.
    ///
    /// Uses a mana emulator and the paravisor if a paravisor is present.
    pub fn with_nic(mut self) -> Self {
        let endpoint = net_backend_resources::consomme::ConsommeHandle {
            cidr: None,
            ports: Vec::new(),
            recv: None,
        }
        .into_resource();
        if let Some(vtl2_settings) = self.runtime_config.vtl2_settings.as_mut() {
            self.config.vpci_devices.push(VpciDeviceConfig {
                vtl: DeviceVtl::Vtl2,
                instance_id: MANA_INSTANCE,
                resource: GdmaDeviceHandle {
                    vports: vec![VportDefinition {
                        mac_address: NIC_MAC_ADDRESS,
                        endpoint,
                    }],
                }
                .into_resource(),
                vnode: None,
            });

            vtl2_settings.dynamic.as_mut().unwrap().nic_devices.push(
                vtl2_settings_proto::NicDeviceLegacy {
                    instance_id: MANA_INSTANCE.to_string(),
                    subordinate_instance_id: None,
                    max_sub_channels: None,
                },
            );
        } else {
            const NETVSP_INSTANCE: Guid = guid::guid!("c6c46cc3-9302-4344-b206-aef65e5bd0a2");
            self.config.vmbus_devices.push((
                DeviceVtl::Vtl0,
                netvsp_resources::NetvspHandle {
                    instance_id: NETVSP_INSTANCE,
                    mac_address: NIC_MAC_ADDRESS,
                    endpoint,
                    max_queues: None,
                }
                .into_resource(),
            ));
        }

        self
    }

    /// Add a PCIe NIC to the VM using the MANA emulator.
    pub fn with_pcie_nic(mut self, port_name: &str, mac_address: MacAddress) -> Self {
        let endpoint = net_backend_resources::consomme::ConsommeHandle {
            cidr: None,
            ports: Vec::new(),
            recv: None,
        }
        .into_resource();
        self.config.pcie_devices.push(PcieDeviceConfig {
            port_name: port_name.to_string(),
            resource: GdmaDeviceHandle {
                vports: vec![VportDefinition {
                    mac_address,
                    endpoint,
                }],
            }
            .into_resource(),
        });

        self
    }

    /// Add a PCIe NVMe device to the VM using the NVMe emulator.
    pub fn with_pcie_nvme(mut self, port_name: &str, subsystem_id: Guid) -> Self {
        self.config.pcie_devices.push(PcieDeviceConfig {
            port_name: port_name.to_string(),
            resource: NvmeControllerHandle {
                subsystem_id,
                max_io_queues: 64,
                msix_count: 64,
                namespaces: vec![NamespaceDefinition {
                    nsid: 1,
                    disk: LayeredDiskHandle::single_layer(RamDiskLayerHandle {
                        len: Some(1024 * 1024),
                        sector_size: None,
                    })
                    .into_resource(),
                    read_only: false,
                }],
                requests: None,
            }
            .into_resource(),
        });

        self
    }

    /// Enable a virtio-net NIC for the VM backed by Consomme.
    ///
    /// This exposes a virtio-net device on a PCIe root port, suitable for
    /// guests running virtio drivers (e.g. Linux with UEFI boot).
    pub fn with_virtio_nic(mut self, port_name: &str, mac_address: MacAddress) -> Self {
        let endpoint = net_backend_resources::consomme::ConsommeHandle {
            cidr: None,
            ports: Vec::new(),
            recv: None,
        }
        .into_resource();

        self.config.pcie_devices.push(PcieDeviceConfig {
            port_name: port_name.to_string(),
            resource: virtio_resources::VirtioPciDeviceHandle(
                virtio_resources::net::VirtioNetHandle {
                    max_queues: None,
                    mac_address,
                    endpoint,
                }
                .into_resource(),
            )
            .into_resource(),
        });

        self
    }

    /// Add a virtio-net NIC with consomme and TCP port forwarding for
    /// pipette. Used for Windows no-vmbus guests where virtio-vsock is
    /// unavailable.
    ///
    /// This configures consomme to forward the pipette TCP port from the
    /// host into the guest, so the petri framework can connect to the
    /// pipette agent over TCP.
    pub fn with_tcp_pipette_nic(mut self, port_name: &str, mac_address: MacAddress) -> Self {
        let (port_send, port_recv) = mesh::oneshot();
        let endpoint = net_backend_resources::consomme::ConsommeHandle {
            cidr: None,
            ports: vec![net_backend_resources::consomme::HostPortConfig {
                protocol: net_backend_resources::consomme::HostPortProtocol::Tcp,
                host_address: Some(net_backend_resources::consomme::HostIpAddress::Ipv4(
                    std::net::Ipv4Addr::LOCALHOST,
                )),
                host_port: net_backend_resources::consomme::HostPort::Dynamic(port_send),
                guest_port: pipette_client::PIPETTE_PORT as u16,
            }],
            recv: None,
        }
        .into_resource();
        self.config.pcie_devices.push(PcieDeviceConfig {
            port_name: port_name.to_string(),
            resource: virtio_resources::VirtioPciDeviceHandle(
                virtio_resources::net::VirtioNetHandle {
                    max_queues: None,
                    mac_address,
                    endpoint,
                }
                .into_resource(),
            )
            .into_resource(),
        });
        self.resources.tcp_pipette_port = Some(port_recv);
        self
    }

    /// Request nested virtualization support from the host hypervisor.
    pub fn with_nested_virt(mut self) -> Self {
        self.config.hypervisor.nested_virt = true;
        self
    }

    /// Enable a synthnic for the VM backed by the Windows vmswitch
    /// DirectIO (`-net dio`) backend.
    ///
    /// `switch_id`, when `None`, defaults to the Hyper-V Default Switch.
    /// This requires the host to have Hyper-V installed and the chosen
    /// switch available; tests that call this method should pre-resolve
    /// a switch via [`super::find_switch`] (or an equivalent runtime
    /// probe) and bail out with a clear error when the host does not
    /// meet those requirements. The method itself panics if the switch
    /// cannot be opened or a port cannot be created.
    ///
    /// The created vmswitch port handle is held in the petri (parent)
    /// process for the lifetime of the VM. The kernel switch port object
    /// is reference counted, so keeping the handle alive in this process
    /// keeps the port usable from the child VMM process.
    #[cfg(windows)]
    pub fn with_dio_nic(mut self, switch_id: Option<Guid>) -> Self {
        let switch_port_id = vmswitch::kernel::SwitchPortId {
            switch: switch_id.unwrap_or(vmswitch::hcn::DEFAULT_SWITCH),
            port: Guid::new_random(),
        };
        let _ = vmswitch::hcn::Network::open(&switch_port_id.switch)
            .unwrap_or_else(|e| panic!("could not find switch {}: {e}", switch_port_id.switch));
        let switch_port = vmswitch::kernel::SwitchPort::new(&switch_port_id)
            .expect("failed to create vmswitch DIO port");
        self.resources._switch_ports.push(switch_port);

        let endpoint = net_backend_resources::dio::WindowsDirectIoHandle {
            switch_port_id: net_backend_resources::dio::SwitchPortId {
                switch: switch_port_id.switch,
                port: switch_port_id.port,
            },
        }
        .into_resource();

        if let Some(vtl2_settings) = self.runtime_config.vtl2_settings.as_mut() {
            self.config.vpci_devices.push(VpciDeviceConfig {
                vtl: DeviceVtl::Vtl2,
                instance_id: MANA_INSTANCE,
                resource: GdmaDeviceHandle {
                    vports: vec![VportDefinition {
                        mac_address: NIC_MAC_ADDRESS,
                        endpoint,
                    }],
                }
                .into_resource(),
                vnode: None,
            });

            vtl2_settings.dynamic.as_mut().unwrap().nic_devices.push(
                vtl2_settings_proto::NicDeviceLegacy {
                    instance_id: MANA_INSTANCE.to_string(),
                    subordinate_instance_id: None,
                    max_sub_channels: None,
                },
            );
        } else {
            const NETVSP_DIO_INSTANCE: Guid = guid::guid!("d1ff4c5a-1b3c-4f0d-8e10-1b9d8b1d1cee");
            self.config.vmbus_devices.push((
                DeviceVtl::Vtl0,
                netvsp_resources::NetvspHandle {
                    instance_id: NETVSP_DIO_INSTANCE,
                    mac_address: NIC_MAC_ADDRESS,
                    endpoint,
                    max_queues: None,
                }
                .into_resource(),
            ));
        }

        self
    }

    /// Load with the specified VTL2 relocation mode.
    pub fn with_vtl2_relocation_mode(mut self, mode: Vtl2BaseAddressType) -> Self {
        let LoadMode::Igvm {
            vtl2_base_address, ..
        } = &mut self.config.load_mode
        else {
            panic!("vtl2 relocation mode is only supported for OpenHCL firmware")
        };
        *vtl2_base_address = mode;
        self
    }

    /// Use a file-backed memory region instead of anonymous RAM.
    ///
    /// The file at the given path will be created (or opened) and sized to
    /// match the VM's configured memory. Guest memory is then backed by
    /// this file, which persists across snapshot save/restore.
    ///
    /// This forces shared (non-private) memory, since a file-backed mapping
    /// is incompatible with private anonymous RAM. Panics if the caller
    /// explicitly requested private memory via
    /// [`MemoryConfig::private_memory`](crate::MemoryConfig::private_memory),
    /// rather than silently downgrading it.
    pub fn with_memory_backing_file(mut self, path: impl Into<std::path::PathBuf>) -> Self {
        assert_ne!(
            self.requested_private_memory,
            Some(true),
            "with_memory_backing_file forces shared memory, which conflicts with \
             the explicitly requested private memory"
        );
        self.memory_backing_file = Some(path.into());
        for node in &mut self.config.numa.nodes {
            if let Some(mem) = &mut node.mem {
                mem.private_memory = false;
            }
        }
        self
    }

    /// Use explicit hugetlb-backed guest memory.
    ///
    /// This forces shared (non-private) memory, since hugetlb backing
    /// requires a file-backed mapping rather than private anonymous RAM.
    /// Panics if the caller explicitly requested private memory via
    /// [`MemoryConfig::private_memory`](crate::MemoryConfig::private_memory),
    /// rather than silently downgrading it.
    pub fn with_hugepages(mut self, hugepage_size: Option<u64>) -> Self {
        assert_ne!(
            self.requested_private_memory,
            Some(true),
            "with_hugepages forces shared memory, which conflicts with the \
             explicitly requested private memory"
        );
        for node in &mut self.config.numa.nodes {
            if let Some(mem) = &mut node.mem {
                mem.hugepages = true;
                mem.hugepage_size = hugepage_size;
                mem.private_memory = false;
            }
        }
        self
    }

    /// Add a symmetric PCIe topology to the VM based on some basic scale factors
    ///
    /// All root ports are named according to their index within their parent
    /// using the naming scheme `sXrcYrpZ`. For example, the third root port on
    /// the fourth root complex in segment 0 would be named `s0rc3rp2`.
    ///
    /// This may be called multiple times to build asymmetric topologies (e.g. a
    /// different number of root complexes per segment). Each call appends its
    /// root complexes to segments numbered after any added by previous calls,
    /// so the segment numbers in the `sXrcY` names continue from where the last
    /// call left off.
    pub fn with_pcie_root_topology(
        mut self,
        segment_count: u64,
        root_complex_per_segment: u64,
        root_ports_per_root_complex: u64,
    ) -> Self {
        const LOW_MMIO_SIZE: u64 = 64 * 1024 * 1024; // 64 MB
        const HIGH_MMIO_SIZE: u64 = 1024 * 1024 * 1024; // 1 GB

        // Offset the segments and global indices added by this call so that it
        // can be called multiple times. New segments are numbered after any
        // existing ones, and the global index continues from the existing
        // root complex count.
        let segment_base = self
            .config
            .pcie_root_complexes
            .iter()
            .map(|rc| u64::from(rc.segment) + 1)
            .max()
            .unwrap_or(0);
        let index_base = self.config.pcie_root_complexes.len() as u64;

        // Add the root complexes to the VM
        for segment_offset in 0..segment_count {
            let segment = segment_base + segment_offset;
            let bus_count_per_rc = 256 / root_complex_per_segment;
            for rc_index_in_segment in 0..root_complex_per_segment {
                let index =
                    index_base + segment_offset * root_complex_per_segment + rc_index_in_segment;
                let name = format!("s{}rc{}", segment, rc_index_in_segment);

                let start_bus = rc_index_in_segment * bus_count_per_rc;
                let end_bus = start_bus + bus_count_per_rc - 1;

                let ports = (0..root_ports_per_root_complex)
                    .map(|i| PciePortConfig {
                        name: format!("s{}rc{}rp{}", segment, rc_index_in_segment, i),
                        devfn: None,
                        hotplug: true,
                        acs_capabilities_supported: Some(0),
                        cxl: false,
                        pasid: false,
                    })
                    .collect();

                self.config.pcie_root_complexes.push(PcieRootComplexConfig {
                    index: index.try_into().unwrap(),
                    name,
                    segment: segment.try_into().unwrap(),
                    start_bus: start_bus.try_into().unwrap(),
                    end_bus: end_bus.try_into().unwrap(),
                    low_mmio: PcieMmioRangeConfig::Dynamic {
                        size: LOW_MMIO_SIZE,
                    },
                    high_mmio: PcieMmioRangeConfig::Dynamic {
                        size: HIGH_MMIO_SIZE,
                    },
                    cxl: None,
                    ports,
                    iommu: None,
                    vnode: None,
                    preserve_bars: false,
                });
            }
        }

        self
    }

    /// Add a PCIe switch to the VM.
    pub fn with_pcie_switch(
        mut self,
        port_name: &str,
        switch_name: &str,
        port_count: u8,
        hotplug: bool,
    ) -> Self {
        self.config.pcie_switches.push(PcieSwitchConfig {
            name: switch_name.to_string(),
            parent_port: port_name.to_string(),
            ports: (0..port_count)
                .map(|i| PciePortConfig {
                    name: format!("{switch_name}-downstream-{i}"),
                    devfn: None,
                    hotplug,
                    acs_capabilities_supported: Some(0),
                    cxl: false,
                    pasid: false,
                })
                .collect(),
        });
        self
    }

    /// Enable SMMUv3 IOMMU on the specified root complexes (aarch64 only).
    ///
    /// Each name must match a root complex added via
    /// [`with_pcie_root_topology`](Self::with_pcie_root_topology). The SMMU
    /// provides stage 1 IOVA translation for devices behind those root
    /// complexes.
    pub fn with_smmu(mut self, rc_names: &[&str]) -> Self {
        for name in rc_names {
            self.pending_iommu.push((
                name.to_string(),
                PcieIommuConfig::Smmu {
                    accel: false,
                    oas: openvmm_defs::config::SmmuOas::Auto,
                },
            ));
        }
        self
    }

    /// Enable an accelerated (iommufd-nested) SMMUv3 on the specified root
    /// complexes (aarch64 only).
    ///
    /// Like [`with_smmu`](Self::with_smmu), but the SMMU programs the host
    /// IOMMU for hardware nested stage-1 translation, so VFIO devices behind
    /// these root complexes are permitted (and their guest-programmed stage-1
    /// tables are honored via a host nested HWPT). Requires a host SMMU that
    /// supports iommufd nesting.
    pub fn with_smmu_accel(mut self, rc_names: &[&str]) -> Self {
        for name in rc_names {
            self.pending_iommu.push((
                name.to_string(),
                PcieIommuConfig::Smmu {
                    accel: true,
                    oas: openvmm_defs::config::SmmuOas::Auto,
                },
            ));
        }
        self
    }

    /// Enable AMD IOMMU (AMD-Vi) on the specified root complexes.
    ///
    /// Each name must match a root complex added via
    /// [`with_pcie_root_topology`](Self::with_pcie_root_topology). The IOMMU
    /// appears at device 0 function 0 on each listed root complex; PCIe
    /// devices behind those root complexes have DMA translated through
    /// guest-programmed page tables and MSIs remapped through the interrupt
    /// remapping table.
    pub fn with_amd_iommu(mut self, rc_names: &[&str]) -> Self {
        for name in rc_names {
            self.pending_iommu
                .push((name.to_string(), PcieIommuConfig::AmdVi));
        }
        self
    }

    /// Enable Intel VT-d IOMMU on the specified root complexes.
    ///
    /// Each name must match a root complex added via
    /// [`with_pcie_root_topology`](Self::with_pcie_root_topology). The IOMMU
    /// is a platform device discovered via the ACPI DMAR table; PCIe devices
    /// behind those root complexes have DMA translated through
    /// guest-programmed page tables and MSIs remapped through the interrupt
    /// remapping table.
    pub fn with_intel_vtd(mut self, rc_names: &[&str]) -> Self {
        for name in rc_names {
            self.pending_iommu
                .push((name.to_string(), PcieIommuConfig::IntelVtd));
        }
        self
    }

    /// This is intended for special one-off use cases. As soon as something
    /// is needed in multiple tests we should consider making it a supported
    /// pattern.
    pub fn with_custom_config(mut self, f: impl FnOnce(&mut Config)) -> Self {
        f(&mut self.config);
        self
    }

    /// Specifies whether VTL2 should be allowed to access VTL0 memory before it
    /// sets any VTL protections.
    ///
    /// This is needed just for the TMK VMM, and only until it gains support for
    /// setting VTL protections.
    pub fn with_allow_early_vtl0_access(mut self, allow: bool) -> Self {
        self.config
            .hypervisor
            .with_vtl2
            .as_mut()
            .unwrap()
            .late_map_vtl0_memory =
            (!allow).then_some(openvmm_defs::config::LateMapVtl0MemoryPolicy::InjectException);

        self
    }
}