Support ACPI Generic Initiator Affinity to associate
PCI devices with NUMA proximity domains
Add GenericInitiatorAffinity struct
Add from_pci_bdf() to encode PCI Segment:Bus:Device.Function
Add from_acpi_device() for ACPI device handles (future use)
Generate SRAT Type 5 entries for nodes with device_id
Improve create_slit_table() to check distance symmetry when
forward distance is missing
Track device ID to BDF mappings in DeviceManager
Includes comprehensive unit tests
Signed-off-by: Saravanan D <saravanand@crusoe.ai>
Validate device_id in numa config is mutually
exclusive with cpus and memory_zones
Add NumaConfig::validate() and modify NumaConfig::parse()
Add ValidationError::InvalidNumaConfig for detailed error
messages
Include unit tests covering valid and invalid configs
Signed-off-by: Saravanan D <saravanand@crusoe.ai>
Add an optional device_id string field to NumaConfig for identifying
PCI devices associated with a NUMA node. This is used by the Generic
Initiator support to map devices to their proximity domain.
Update OpenAPI spec (cloud-hypervisor.yaml) to include the
new device_id field in the NumaConfig schema.
The device_id is optional and parsed from the --numa parameter:
--numa "device_id=<device_id>,distances=[...],..."
The optional field is accepted but not used.
Signed-off-by: Saravanan D <saravanand@crusoe.ai>
The documentation says guest_numa_id is required to be unique and
therefore the parser() giving default value for non-existing
guest_numa_id with .unwrap_or(0) is dangerous.
Return a validation error if guest_numa_id is not provided instead
of silently defaulting to 0.
Signed-off-by: Saravanan D <saravanand@crusoe.ai>
When a VFIO device with multiple MMIO regions is hot-unplugged, each
region must be individually matched and removed from the DeviceManager's
mmio_regions list. Compare per-region rather than building an aggregate
across all regions, which would never match any individual entry.
Also remove the now-unused HashSet import.
Signed-off-by: Damian Barabonkov <dbctl@pm.me>
Replace `clone()` with `take()` when retrieving device configurations
from `DeviceManager.config`.
This avoids unnecessarily copying the device configuration lists (e.g.,
`disks`, `net`, `fs`) when they are being processed and subsequently
moved out of the configuration. This optimization improves performance
by reducing memory allocations and cloning overhead.
Signed-off-by: Changyuan Lyu <changyuanl@google.com>
Backing files (e.g. for QCOW2) interact badly with landlock since they
are not obvious from the initial VM configuration. Only enable their use
with an explicit option.
Signed-off-by: Rob Bradford <rbradford@meta.com>
Set HV_PARTITION_CREATION_FLAG_SMT_ENABLED_GUEST when the guest
topology has more than one thread per core. This allows the
hypervisor to schedule guest VPs correctly on SMT-enabled hosts.
Without this flag, the hypervisor schedules guest VPs incorrectly,
causing SMT unusable.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
The kernel allows madvise on shared memory if
/sys/kernel/mm/transparent_hugepage/shmem_enabled is set.
Always try and configure THP via madvise when
the user requests THP be enabled.
If this fails, only a warning log is emitted and THP won't be enabled.
Signed-off-by: Champ-Goblem <cameron@northflank.com>
When restoring from snapshot with shared=false, write access to the
backing file is not required. Opening it read-only allows restore to
succeed on read-only media and overlay lower layers while preserving
MAP_PRIVATE semantics.
Signed-off-by: Rowen-Ye <rowenye1@gmail.com>
TDX builds its own ACPI tables in `create_acpi_tables_tdx` so it will
return None in the standard `create_acpi_tables` function and the
assertion for `rsdp_addr` will fail.
Signed-off-by: Zhibin Li <banlu.lzb@antgroup.com>
Recent changes related to arm64 support in MSHV exposed
inconsistencies in the VM initialization and CVM boot paths.
The VM creation flow currently diverges across multiple scenarios,
including regular MSHV, CVM, and arm64, with each path performing
guest initialization steps in a different order.
Certain platform-specific requirements further constrain the ordering
of operations, such as the timing of address space creation,
IGVM loading, interrupt controller setup, and payload loading. For
CVM case address-space creation must be done after IGVM loading, and
PSP measurement. For Regular and arm64 this memory initialization
must be done early. For MSHV, vm.init() and sev_snp.init() are called in
different order which is run time and build time conditionally checked.
Additionally, while the KVM initialization path differs slightly
from MSHV, it shares common logic that is currently split across
separate conditional and build-time code paths, contributing to
fragmentation of the overall flow.
This change restructures the VM creation and initialization sequence
to better align shared logic, enforce scenario-specific ordering
constraints, and ensure consistent and correct behavior across all
supported configurations. In doing so, it restores proper CVM boot
behavior and improves the maintainability of the initialization code.
Signed-off-by: Muminul Islam <muislam@microsoft.com>
The IORT table's ID mapping uses a 256-ID offset per PCI segment to
ensure unique device IDs across all segments. This partitioning scheme
(output_base = 256 * segment_id) must match the device ID encoding used
in KVM MSI routing configuration [1].
This mapping assumes one bus per PCI segment, and supports up to 256 PCI
segments in the system.
[1] c9374d87ac
Signed-off-by: Bo Chen <bchen@crusoe.ai>
The IORT specification (Revision E.b, Table 12) defines the ITS Group
Node structure with an ITS Identifiers array following the node header.
Although the IORT table is zero-initialized, this commit adds an
explicit write of the ITS identifier value (0) for clarity and spec
compliance.
This ITS identifier must match the `translation_id` field in the MADT
GIC ITS structure to ensure proper interrupt routing on ARM platforms.
Signed-off-by: Bo Chen <bchen@crusoe.ai>
The current IORT table implementation is based on IORT Spec revision E.b
[1], as evidenced by:
* The PCI root complex node revision being set to `3`
* The code being updated in late 2021 [2] when revision E.b was the
latest version
This patch ensures the IORT table is properly generated according to
this specification revision, fixing three issues:
1. The IORT table revision should be `3` rather than `2` (see Table 2 in
the spec [1])
2. The GIC ITS group node revision should be `1` rather than `0`
(see Table 12 in the spec [1])
3. The "Memory access properties" and "ATS Attribute" fields of the PCI
root complex node was set incorrectly - specifically the MAF (Memory
Access Flags) including CPM and DACS bits (see Tables 14, 15, and 17
in the spec [1])
[1] https://developer.arm.com/documentation/den0049/eb/?lang=en
[2] https://github.com/cloud-hypervisor/cloud-hypervisor/pull/3356
Signed-off-by: Bo Chen <bchen@crusoe.ai>
It should always succeed and is apparently implicitly called by libc or
some dependency somewhere.
Signed-off-by: Demi Marie Obenour <demiobenour@gmail.com>
On MSHV, exposing multithreaded CPU topologies requires setting the
PROCESSORS_PER_SOCKET partition property so that CPUID.0xB reports
correct logical processor counts and topology levels to the guest.
This property must be set after all vCPUs are configured, as the
hypervisor uses the complete vCPU layout to derive and report CPU
topology information.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
Virtio PCI devices are created in a set of nested functions. In each
of this functions a vector is created to add created devices to, only
to be appended to the vector of the higher nesting level. Those nested
vectors are unnecessary as we can directly write to the member of
`DeviceManager`.
Signed-off-by: Pascal Scholz <pascal.scholz@cyberus-technology.de>
On-behalf-of: SAP pascal.scholz@sap.com
In [0] we refactored some Arc<Mutex<T>> parameters to &Mutex<T>> to
satisfy clippy's needless_pass_by_value lint. Nevertheless, this is also
not so idiomatic, so as a follow-up, we put the responsibility to lock
objects to the caller side (only where this is not strictly needed by
the callee).
While on it, I also tried to pass vm_config directly into
pre_create_console_devices() which would clean up some code, but then
we have interleaving mutable and immutable borrows of the Vmm, which
are denied by the borrow checker.
Signed-off-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
On-behalf-of: SAP philipp.schuster@sap.com
We enable AMX tile state components via the hypervisor (as introduced
in the previous commit) instead of doing this inline in the body of
`CpuManager::new`.
Signed-off-by: Oliver Anderson <oliver.anderson@cyberus-technology.de>
On-behalf-of: SAP oliver.anderson@sap.com
Add an option in the CLI to enable nested support.
Adding an option --cpu nested=on|off to the CPU
argument to enable nested support for Microsoft
Hypervisor. By default it is enabled for both KVM
and MSHV.
Signed-off-by: Muminul Islam <muislam@microsoft.com>
Create HypervisorVmConfig early and pass the
struct to VM creation API in the vmm crate. Getting
rid of multiple conditional parameter.
Signed-off-by: Muminul Islam <muislam@microsoft.com>
The MSRs are constant at boot time so rather than creating a vector in
the boot_msr_entries() method instead reaturn a reference to static MSR
array data.
Signed-off-by: Rob Bradford <rbradford@rivosinc.com>
Previously, state that we accumulated during the migration process in
the receiver was kept in `mut Option` variables or HashMaps. The
problem is that it is unclear in the code when these variables can be
safely used. It's also difficult to add new state, such as the state
we need to handle the upcoming feature for performing the migration
using multiple connections.
To solve this, I've refactored the code to use the state machine
pattern. Each state carries the data it needs to. Any state that
didn't arrive yet (memory_files, memory_manager) cannot be accessed
until we are in the proper state.
Some benefits that fall out of this:
- We now respond to all requests, even invalid ones, at least with an
error message.
- Any error handling a request will result in an error message being
sent.
- There is only a single place where responses are sent and thus it's
very hard to mess this up in the code.
- The main protocol state machine fits on a screen.
I would argue that especially the error cases are now much more
consistent. There is still a lot to be done. There is still state
transfer via self.vm and similar. In an ideal world, this would also
be carried by the state machine. I also want to see better handling of
payloads, which are still handled all over the place, but this change
is already too big. :)
Co-authored-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
On-behalf-of: SAP julian.stecklina@sap.com
On-behalf-of: SAP philipp.schuster@sap.com
Signed-off-by: Julian Stecklina <julian.stecklina@cyberus-technology.de>
Signed-off-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
... and nuke some Option<> while I was there. Given that HashMap has a
usable default and we end up passing an empty HashMap anyway, just get
rid of the Option.
On-behalf-of: SAP julian.stecklina@sap.com
Signed-off-by: Julian Stecklina <julian.stecklina@cyberus-technology.de>
This is a follow-up of [0].
# Advantages
- This saves dozens of unneeded clone()s across the whole code base
- Makes it much easier to reason about how parameters are used
(often we passed owned Arc/Rc versions without actually needing
ownership)
# Exceptions
For certain code paths, the alternatives would require awkward or overly
complex code, and in some cases the functions are the logical owners of
the values they take. In those cases, I've added
#[allow(clippy::needless_pass_by_value)].
This does not mean that one should not improve this in the future.
[0] 6a86c157af
Signed-off-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
On-behalf-of: SAP philipp.schuster@sap.com
Fix the condition to correctly identify a new PTY connection.
Fixes: 287887c99 (vmm: fix console IO safety)
Signed-off-by: Praveen K Paladugu <prapal@linux.microsoft.com>
Previously, `into_raw_fd()` transferred fd ownership to epoll while
keeping separate clones in `reader` and `writer`, causing leaks when
the stream was closed. Now, `reader` owns the fd and epoll borrows it.
When the FD is closed in the kernel, reader will be reset to None.
Signed-off-by: Praveen K Paladugu <prapal@linux.microsoft.com>
When enabling amx feature, we should call arch_prctl to request
permission to use tile data for guest. The permission should be
requested before the first vcpu is created, so we need to call
arch_prctl in vmm thread. This patch adds the arch_prctl syscall for
vmm_thread_rules.
Fixes: #7516
Signed-off-by: Songqian Li <sionli@tencent.com>
This helps to uncover expensive and needless clones in the code base.
For example, I prevented extensive clones in the snapshot path where
(nested) BTreeMap's have been cloned over and over again. Further,
the lint helps devs to much better reason about the ownership of
parameters.
All of these changes have been done manually with the necessary
caution. A few structs that are cheap to clone are now `copy` so that
this lint won't trigger for them.
I didn't enable the lint so far as it is a massive rabbit hole and
needs much more fixes. Nevertheless, it is very useful.
Signed-off-by: Philipp Schuster <philipp.schuster@cyberus-technology.de>
On-behalf-of: SAP philipp.schuster@sap.com
