Add a new id option to the VFIO hotplug command so that it matches the
VFIO coldplug semantic.
This is done by refactoring the existing code for VFIO hotplug, where
VmAddDeviceData structure is replaced by DeviceConfig. This structure is
the one used whenever a VFIO device is coldplugged, which is why it
makes sense to reuse it for the hotplug codepath.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Add the ability to specify the "id" associated with a device, by adding
an extra option to the parameter --device.
This new option is not mandatory, and by default, the VMM will take care
of finding a unique identifier.
If the identifier provided by the user through this new option is not
unique, an error will be thrown and the VM won't be started.
Fixes#881
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The 32 bits MMIO address space is handled separately from the 64 bits
one. For this reason, we need to invoke the appropriate freeing function
to remove a range from this address space.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Now that PciDevice trait has a dedicated function to remove the bars,
the DeviceManager can invoke this function whenever a PCI device is
unplugged from the VM.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Upon removal of a PCI device, make sure we don't hold onto the device ID
as it could be reused for another device later.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
In order to handle the case where devices are very often plugged and
unplugged from a VM, we need to handle the PCI device ID allocation
better.
Any PCI device could be removed, which means we cannot simply rely on
the vector size to give the next available PCI device ID.
That's why this patch stores in memory the information about the 32
slots availability. Based on this information, whenever a new slot is
needed, the code can correctly provide an available ID, or simply return
an error because all slots are taken.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Add a new field to the DeviceConfig, allowing the VMM to allocate a name
to the VFIO devices.
By identifying a VFIO device with a unique name, we can make sure a user
can properly unplug it at any time.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This commit introduces the new command "remove-device" that will let a
user hot-unplug a VFIO PCI device from an already running VM.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This commit implements the eject function so that a VFIO device will be
removed from any bus it might sit on, and from any list it might be
stored in.
The idea is to reach a point where there is no reference of the device
anywhere in the code, so that the Drop implementation will be invoked
and so that the device will be fully removed from the VMM.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
When the guest OS is done removing a PCI device, it will invoke the _EJ0
method from ACPI, associated with the device. This will trigger a port
IO write to a region known by the VMM. Upon this writing, the VMM will
trap the VM exit and retrieve the written value.
Based on the value, the VMM will invoke its eject_device() method to
finalize the removal of the device.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
As we try to keep track of every PCI device related to the VM, we don't
want to have separate lists depending on the concrete type associated
with the PciDevice trait. Also, we want to be able to cast the actual
type into any trait or concrete type.
The most efficient way to solve all these issues is to store every
device as an Arc<dyn Any + Send + Sync>. This gives the ability to
downcast into the appropriate concrete type, and then to cast back into
any trait that we might need.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Add a new list storing the device names across the entire codebase. VFIO
devices are added to the list whenever a new one is created. By default,
each VFIO device is given a name "vfioX" where X is the first available
integer.
Along with this new list of names, another list is created, grouping PCI
device's name with its associated b/d/f. This will be useful to keep
track of the created devices so that we can implement unplug
functionality.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The Vm structure was used to store a strong reference to the IO bus.
This is not needed anymore since the AddressManager is logically the
one holding this strong reference. This has been made possible by the
introduction of Weak references on the Bus structure itself.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Now that the BusDevice devices are stored as Weak references by the
IO and MMIO buses, there's no need to use Weak references from the
DeviceManager anymore.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Now that the BusDevice devices are stored as Weak references by the IO
and MMIO buses, there's no need to use Weak references from the PciBus
anymore.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The point is to make sure the DeviceManager holds a strong reference of
each BusDevice inserted on the IO and MMIO buses. This will allow these
buses to hold Weak references onto the BusDevice devices.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The method add_vfio_device() from the DeviceManager needs to be mutable
if we want later to be able to update some internal fields from the
DeviceManager from this same function.
This commit simply takes care of making the necessary changes to change
this function as mutable.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
By inserting the DeviceManager on the IO bus, we introduced some cyclic
dependency:
DeviceManager ---> AddressManager ---> Bus ---> BusDevice
^ |
| |
+---------------------------------------------+
This cycle needs to be broken by inserting a Weak reference instead of
an Arc (considered as a strong reference).
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Ensures the configuration is updated after a new device has been
hotplugged. In the event of a reboot, this means the new VM will be
started with the new device that had been previously hotplugged.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This commit finalizes the VFIO PCI hotplug support, based on all the
previous commits preparing for it.
One thing to notice, this does not support vIOMMU yet. This means we can
hotplug VFIO PCI devices, but we cannot attach them to an existing or a
new virtio-iommu device.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This factorization is very important as it will allow both the standard
codepath and the VFIO PCI hotplug codepath to rely on the same function
to perform the addition of a new VFIO PCI device.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This commit introduces the new command "add-device" that will let a user
hotplug a VFIO PCI device to an already running VM.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Through the BusDevice implementation from the DeviceManager, and by
inserting the DeviceManager on the IO bus for a specific IO port range,
the VMM now has the ability to handle PCI device hotplug.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
In anticipation of inserting the DeviceManager on the IO/MMIO buses,
the DeviceManager must implement the BusDevice trait.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Create a small method that will perform both hotplug of all the devices
identified by PCIU bitmap, and then perform the hotunplug of all the
devices identified by the PCID bitmap.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The _EJ0 method provides the guest OS a way to notify the VMM that the
device has been properly ejected from the guest OS. Only after this
point, the VMM can fully remove the device.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This new PHPR device in the DSDT table introduces some specific
operation regions and the associated fields.
PCIU stands for "PCI up", which identifies PCI devices that must be
added.
PCID stands for "PCI down", which identifies PCI devices that must be
removed.
B0EJ stands for "Bus 0 eject", which identifies which device on the bus
has been ejected by the guest OS.
Thanks to these fields, the VMM and the guest OS can communicate while
performing hotplug/hotunplug operations.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Adds the DVNT method to the PCI0 device in the DSDT table. This new
method is responsible for checking each slot and notify the guest OS if
one of the slots is supposed to be added or removed.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This commit introduces the ACPI support for describing the 32 device
slots attached to the main PCI host bridge.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
In anticipation of the support for device hotplug, this commit moves the
DeviceManager object into an Arc<Mutex<>> when the DeviceManager is
being created. The reason is, we need the DeviceManager to implement the
BusDevice trait and then provide it to the IO bus, so that IO accesses
related to device hotplug can be handled correctly.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Relying on the latest vm-memory version, including the freshly
introduced structure GuestMemoryAtomic, this patch replaces every
occurrence of Arc<ArcSwap<GuestMemoryMmap> with
GuestMemoryAtomic<GuestMemoryMmap>.
The point is to rely on the common RCU-like implementation from
vm-memory so that we don't have to do it from Cloud-Hypervisor.
Fixes#735
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
If no socket is supplied when enabling "vhost_user=true" on "--disk"
follow the "exe" path in the /proc entry for this process and launch the
network backend (via the vmm_path field.)
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
When a virtio-fs device is created with a dedicated shared region, by
default the region should be mapped as PROT_NONE so that no pages can be
faulted in.
It's only when the guest performs the mount of the virtiofs filesystem
that we can expect the VMM, on behalf of the backend, to perform some
new mappings in the reserved shared window, using PROT_READ and/or
PROT_WRITE.
Fixes#763
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
If no socket is supplied when enabling "vhost_user=true" on "--net"
follow the "exe" path in the /proc entry for this process and launch the
network backend (via the vmm_path field.)
Currently this only supports creating a new tap interface as the network
backend also only supports that.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
It is necessary to do this at the start of the VMM execution rather than
later as it must be done in the main thread in order to satisfy the
checks required by PTRACE_MODE_READ_FSCREDS (see proc(5) and
ptrace(2))
The alternative is to run as CAP_SYS_PTRACE but that has its
disadvantages.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Having the InterruptManager trait depend on an InterruptType forces
implementations into supporting potentially very different kind of
interrupts from the same code base. What we're defining through the
current, interrupt type based create_group() method is a need for having
different interrupt managers for different kind of interrupts.
By associating the InterruptManager trait to an interrupt group
configuration type, we create a cleaner design to support that need as
we're basically saying that one interrupt manager should have the single
responsibility of supporting one kind of interrupt (defined through its
configuration).
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
We create 2 different interrupt managers for separately handling
creation of legacy and MSI interrupt groups.
Doing so allows us to have a cleaner interrupt manager and IOAPIC
initialization path. It also prepares for an InterruptManager trait
design improvement where we remove the interrupt source type dependency
by associating an interrupt configuration type to the trait.
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
A reference to the VmFd is stored on the AddressManager so it is not
necessary to pass in the VmInfo into all methods that need it as it can
be obtained from the AddressManager.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
The DeviceManager has a reference to the MemoryManager so use that to
get the GuestMemoryMmap rather than the version stored in the VmInfo
struct.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Remove the use of vm_info in methods to get the config and instead use
the config stored on the DeviceManager itself.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Remove some in/out parameters and instead rely on them as members of the
&mut self parameter. This prepares the way to more easily store state on
the DeviceManager.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Remove some in/out parameters and instead rely on them as members of the
&mut self parameter. A follow-up commit will change the callee functions
that create the devices themselves.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Modify these functions to take an &mut self and become methods on
DeviceManager. This allows the removal of some in/out parameters and
leads the way to further refactoring and simplification.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
The MemoryManager should only be included on the I/O bus when doing ACPI
builds as that is the only time it will be interrogated.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Currently the MemoryManager is only used on the ACPI code paths after
the DeviceManager has been created. This will change in a future commit
as part of the refactoring so for now always include it but name it with
underscore prefix to indicate it might not always be used.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Now that devices attached to the virtual IOMMU are described through
virtio configuration, there is no need for the DeviceManager to store
the list of IDs for all these devices. Instead, things are handled
locally when PCI devices are being added.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Instead of relying on the ACPI tables to describe the devices attached
to the virtual IOMMU, let's use the virtio topology, as the ACPI support
is getting deprecated.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Add a socket and vhost_user parameter to this option so that the same
configuration option can be used for both virtio-block and
vhost-user-block. For now it is necessary to specify both vhost_user
and socket parameters as auto activation is not yet implemented. The wce
parameter for supporting "Write Cache Enabling" is also added to the
disk configuration.
The original command line parameter is still supported for now and will
be removed in a future release.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Add a socket and vhost_user parameter to this option so that the same
configuration option can be used for both virtio-net and vhost-user-net.
For now it is necessary to specify both vhost_user and socket parameters
as auto activation is not yet implemented. The original command line
parameter is still supported for now.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
This commit improves the existing virtio-blk implementation, allowing
for better I/O performance. The cost for the end user is to accept
allocating more vCPUs to the virtual machine, so that multiple I/O
threads can run in parallel.
One thing to notice, the amount of vCPUs must be egal or superior to the
amount of queues dedicated to the virtio-blk device.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Devices like virtio-pmem and virtio-fs require some dedicated memory
region to be mapped. The memory mapping from the DeviceManager is being
replaced by the usage of MmapRegion from the vm-memory crate.
The unmap will happen automatically when the MmapRegion will be dropped,
which should happen when the DeviceManager gets dropped.
Fixes#240
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Move GED device reporting of required device type to scan into an MMIO
region rather than an I/O port.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Rather than have the MemoryManager device sit on the I/O bus allocate
space for MMIO and add it to the MMIO bus.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
This commit relies on the interrupt manager and the resulting interrupt
source group to abstract the knowledge about KVM and how interrupts are
updated and delivered.
This allows the entire "devices" crate to be freed from kvm_ioctls and
kvm_bindings dependencies.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The interrupt manager is passed to the IOAPIC creation, and the IOAPIC
now creates an InterruptSourceGroup for MSI interrupts based on it.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
By introducing a new InterruptManager dedicated to the IOAPIC, we don't
have to solve the chicken and eggs problem about which of the
InterruptManager or the Ioapic should be created first. It's also
totally fine to have two interrupt manager instances as they both share
the same list of GSI routes and the same allocator.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
vhost_user_blk already has it, so it's only fair to give it to
virtio-blk too. Extend DiskConfig with a 'direct' property, honor
it while opening the file backing the disk image, and pass it to
vm_virtio::RawFile.
Fixes#631
Signed-off-by: Sergio Lopez <slp@redhat.com>
vhost_user_blk already has it, so it's only fair to give it to
virtio-blk too. Extend DiskConfig with a 'readonly' properly, and pass
it to vm_virtio::Block.
Signed-off-by: Sergio Lopez <slp@redhat.com>
The build is run against "--all-features", "pci,acpi", "pci" and "mmio"
separately. The clippy validation must be run against the same set of
features in order to validate the code is correct.
Because of these new checks, this commit includes multiple fixes
related to the errors generated when manually running the checks.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
There's no need for assign_irq() or assign_msix() functions from the
PciDevice trait, as we can see it's never used anywhere in the codebase.
That's why it's better to remove these methods from the trait, and
slightly adapt the existing code.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Since the InterruptManager is never stored into any structure, it should
be passed as a reference instead of being cloned.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This commit replaces the way legacy interrupts were handled with the
brand new implementation of the legacy InterruptSourceGroup for KVM.
Additionally, since it removes the last bit relying on the Interrupt
trait, the trait and its implementation can be removed from the
codebase.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This commit replaces the way legacy interrupts were handled with the
brand new implementation of the legacy InterruptSourceGroup for KVM.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This commit replaces the way legacy interrupts were handled with the
brand new implementation of the legacy InterruptSourceGroup for KVM.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
By having a reference to the IOAPIC, the KvmInterruptManager is going
to be able to initialize properly the legacy interrupt source group.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
In order to let the InterruptManager be shared across both PCI and MMIO
devices, this commit moves the initialization earlier in the code.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Based on all the previous changes, we can at this point replace the
entire interrupt management with the implementation of InterruptManager
and InterruptSourceGroup traits.
By using KvmInterruptManager from the DeviceManager, we can provide both
VirtioPciDevice and VfioPciDevice a way to pick the kind of
InterruptSourceGroup they want to create. Because they choose the type
of interrupt to be MSI/MSI-X, they will be given a MsiInterruptGroup.
Both MsixConfig and MsiConfig are responsible for the update of the GSI
routes, which is why, by passing the MsiInterruptGroup to them, they can
still perform the GSI route management without knowing implementation
details. That's where the InterruptSourceGroup is powerful, as it
provides a generic way to manage interrupt, no matter the type of
interrupt and no matter which hypervisor might be in use.
Once the full replacement has been achieved, both SystemAllocator and
KVM specific dependencies can be removed.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Callbacks are not the most Rust idiomatic way of programming. The right
way is to use a Trait to provide multiple implementation of the same
interface.
Additionally, a Trait will allow for multiple functions to be defined
while using callbacks means that a new callback must be introduced for
each new function we want to add.
For these two reasons, the current commit modifies the existing
VirtioInterrupt callback into a Trait of the same name.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Because MsixConfig will be responsible for updating KVM GSI routes at
some point, it is necessary that it can access the list of routes
contained by gsi_msi_routes.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Because MsixConfig will be responsible for updating the KVM GSI routes
at some point, it must have access to the VmFd to invoke the KVM ioctl
KVM_SET_GSI_ROUTING.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The point here is to let MsixConfig take care of the GSI allocation,
which means the SystemAllocator must be passed from the vmm crate all
the way down to the pci crate.
Once this is done, the GSI allocation and irq_fd creation is performed
by MsixConfig directly.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Because we will need to share the same list of GSI routes across
multiple PCI devices (virtio-pci, VFIO), this commit moves the creation
of such list to a higher level location in the code.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Use RawFile as backend instead of File. This allows us to abstract
the access to the actual image with a specialized layer, so we have a
place where we can deal with the low-level peculiarities.
Signed-off-by: Sergio Lopez <slp@redhat.com>
Doing I/O on an image opened with O_DIRECT requires to adhere to
certain restrictions, requiring the following elements to be aligned:
- Address of the source/destination memory buffer.
- File offset.
- Length of the data to be read/written.
The actual alignment value depends on various elements, and according
to open(2) "(...) there is currently no filesystem-independent
interface for an application to discover these restrictions (...)".
To discover such value, we iterate through a list of alignments
(currently, 512 and 4096) calling pread() with each one and checking
if the operation succeeded.
We also extend RawFile so it can be used as a backend for QcowFile,
so the later can be easily adapted to support O_DIRECT too.
Signed-off-by: Sergio Lopez <slp@redhat.com>
Update the common part in net_util.rs under vm-virtio to add mq
support, meanwhile enable mq for virtio-net device, vhost-user-net
device and vhost-user-net backend. Multiple threads will be created,
one thread will be responsible to handle one queue pair separately.
To gain the better performance, it requires to have the same amount
of vcpus as queue pair numbers defined for the net device, due to
the cpu affinity.
Multiple thread support is not added for vhost-user-net backend
currently, it will be added in future.
Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
Add support to allow VMMs to open the same tap device many times, it will
create multiple file descriptors meanwhile.
Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
Since the common parts are put into net_util.rs under vm-virtio,
refactoring code for virtio-net device, vhost-user-net device
and backend to shrink the code size and improve readability
meanwhile.
Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
Use independent bits for storing whether there is a CPU or memory device
changed when reporting changes via ACPI GED interrupt. This prevents a
later notification squashing an earlier one and ensure that hotplugging
both CPU and memory at the same time succeeds.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
If there is a GED interrupt and the field indicates that the memory
device has changed triggers a scan of the memory devices.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
When the value is read from the I/O port via the ACPI AML functions to
determine what has been triggered the notifiction value is reset
preventing a second read from exposing the value. If we need support
multiple types of GED notification (such as memory hotplug) then we
should avoid reading the value multiple times.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
If neither PCI or MMIO are built in, we should not bother creating any
virtio devices at all.
When building a minimal VMM made of a kernel with an initramfs and a
serial console, the RNG virtio device is still created even though there
is no way it can ever get probed.
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>