Commit Graph

173 Commits

Author SHA1 Message Date
Rob Bradford
7df88793a0 vmm: device_manager: Get device range from MemoryManager
This removes the duplication of these values.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-23 10:25:40 +00:00
Rob Bradford
61cfe3e72d vmm: Obtain sequential KVM memory slot numbers from MemoryManager
This removes the need to handle a mutable integer and also centralises
the allocation of these slot numbers.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-23 10:25:40 +00:00
Rob Bradford
260cebb8cf vmm: Introduce MemoryManager
The memory manager is responsible for setting up the guest memory and in
the long term will also handle addition of guest memory.

In this commit move code for creating the backing memory and populating
the allocator into the new implementation trying to make as minimal
changes to other code as possible.

Follow on commits will further reduce some of the duplicated code.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-23 10:25:40 +00:00
Samuel Ortiz
37557c8b35 vmm: vm: Implement the Pausable trait
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-12-12 08:50:36 +01:00
Samuel Ortiz
9756fc2dd0 vmm: cpu_manager: Implement the Pausable trait
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-12-12 08:50:36 +01:00
Samuel Ortiz
35dd1523c9 vmm: device_manager: Implement the Pausable trait
Since the Snapshotable placeholder and Migratable traits are provided as
well, the DeviceManager object and all its objects are now Migratable.

All Migratable devices are tracked as Arc<Mutex<dyn Migratable>>
references.

Keeping track of all migratable devices allows for implementing the
Migratable trait for the DeviceManager structure, making the whole
device model potentially migratable.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-12-12 08:50:36 +01:00
Rob Bradford
f994665610 vmm: Reduce the minimum IRQ constant
Now that the GED device does not use a hardcoded IRQ number the starting
IRQ number can be restored (needed for the hardcoded serial port IRQ.)

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-09 16:58:00 +00:00
Rob Bradford
9b1ba14f2d vmm: Delegate device related ACPI DSDT table work to DeviceManager
Move the code for handling the creation of the DSDT entries for devices
into the DeviceManager.

This will make it easier to handle device hotplug and also in the future
remove some hardcoded ACPI constants.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-06 17:44:00 +00:00
Rob Bradford
60e6609011 vmm: Delegate CPU related ACPI tables to CpuManager
Move the code for generating the MADT (APIC) table and the DSDT
generation for CPU related functionality into the CpuManager.

There is no functional change just code rearrangement.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-06 17:44:00 +00:00
Rob Bradford
59d01712ad vmm: Remove kernel based IOAPIC handling from the device manager
Previously the device setup code assumed that if no IOAPIC was passed in
then the device should be added to the kernel irqchip. As an earlier
change meant that there was always a userspace IOAPIC this kernel based
code can be removed.

The accessor still returns an Option type to leave scope for
implementing a situation without an IOAPIC (no serial or GED device).
This change does not add support no-IOAPIC mode as the original code did
not either.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-06 12:34:06 +01:00
Rob Bradford
afea6a10a2 vmm: Stop initialising kernel based IOAPIC/PIC
Now that we require the modern capabilities we can stop creating a
kernel base irqchip.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-06 12:34:06 +01:00
Rob Bradford
9b1cb9621f vmm: Remove pin based interrupt setup for virtio devices
With MSI now required remove pin based interrupt support from all the
virtio PCI device setup.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-06 12:34:06 +01:00
Rob Bradford
72fb687e3f vmm: Check for required capabilities
We now require CAP_SIGNAL_MSI, CAP_TSC_DEADLINE_TIMER and
CAP_SPLIT_IRQCHIP.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-06 12:34:06 +01:00
Rob Bradford
f98b16f308 vmm: Update the configuration to preserve hot-plug CPUs after reboot
Update the configuration after a resize to ensure that after a reboot
the added vCPUs are preserved.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-05 16:39:19 +00:00
Rob Bradford
1722708612 vmm: Switch to storing VmConfig inside an Arc<Mutex<>>
This permits the runtime reconfiguration of the VM.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-05 16:39:19 +00:00
Qiu Wenbo
e1af17d93a vmm: Restore tty to canonical mode when SIGTERM or SIGINT received
The tty mode remains raw mode when cloud-hypervisor is terminted by
SIGTERM or SIGINT. The terminal is unusable due to echoing is
disabled which is really annoying.

Signed-off-by: Qiu Wenbo <qiuwenbo@phytium.com.cn>
2019-12-05 01:29:26 -08:00
Qiu Wenbo
5208ff86c8 vmm: Detect and handle AMD SME (Secure Memory Encryption)
Some physical address bits may become reserved in page table when SME
is enabled on AMD platform. Guest will trigger a reserved bit
violation page fault in this case due to write these reserved bits to 1
in page table. We need reduce the reserved bits to get the right
physical address range.

Signed-off-by: Qiu Wenbo <qiuwenbo@phytium.com.cn>
2019-12-04 14:46:44 +00:00
Rob Bradford
48bf141364 vmm: Trigger a hotplug device notification when resizing
When adjusting the number of vCPUs generate a hotplug notification.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-02 13:49:04 +00:00
Rob Bradford
ae9359c859 vmm: acpi: Create the CPU entries in the DSDT for all vCPUs
CPU entries need to be created for every potential vCPU in the system.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-02 13:49:04 +00:00
Rob Bradford
791ca3388f vmm: device_manager: Add ability to notify via GED device
Add ability to notify via the GED device that there is some new hotplug
activity. This will be used by the CpuManager (and later DeviceManager
itself) to notify of new hotplug activity.

Currently it has a hardcoded IRQ of 5 as the ACPI tables also need to
refer to this IRQ and the IRQ allocation does not permit the allocation
of specific IRQs.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-02 13:49:04 +00:00
Rob Bradford
86339b4cb4 vmm: Add HTTP API to resize the VM
Currently only increasing the number of vCPUs is supported but in the
future it will be extended.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-02 13:49:04 +00:00
Rob Bradford
1bbe48b24c vmm: acpi: Mark non-boot vCPUs as disabled in the MADT table
The MADT table contains the details of all the potential vCPUs and
whether they are present at boot (as indicated by the flags field.)

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-02 13:49:04 +00:00
Rob Bradford
82bc07cce4 vmm: Add boot and max vCPU handling to command line parser
Also retain support (with a warning for the old behaviour.)

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-02 13:49:04 +00:00
Rob Bradford
7543e00a07 vmm: Use new CpuManager accessor to get boot vCPUs
When initialising the ACPI tables and configuring the VM use the new
accessor on the CpuManager to get the number of boot vCPUs.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-02 13:49:04 +00:00
Rob Bradford
df0907845a vmm: cpu: Introduce concept of maximum vs boot vCPUs in CpuManager
For now the max vCPUs is the same as the boot vCPUs.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-12-02 13:49:04 +00:00
Samuel Ortiz
0f21781fbe cargo: Bump the kvm and vmm-sys-util crates
Since the kvm crates now depend on vmm-sys-util, the bump must be
atomic.
The kvm-bindings and ioctls 0.2.0 and 0.4.0 crates come with a few API
changes, one of them being the use of a kvm_ioctls specific error type.
Porting our code to that type makes for a fairly large diff stat.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-11-29 17:48:02 +00:00
Sebastien Boeuf
e4e8062dda vmm: Mark guest RAM pages as mergeable
In case the VM is started with the flag "--memory mergeable=on", it
means the user expects the guest RAM pages to be marked as mergeable.
This commit relies on the madvise(MADV_MERGEABLE) system call to inform
the host kernel about these pages.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-11-22 15:28:10 +00:00
Rob Bradford
1da0ff395d vmm: cpu: Add the CpuManager onto the IO bus
This allows the kernel (via ACPI based controls) to query and control
the CPU state.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-11-21 09:17:15 -08:00
Rob Bradford
1ac1231292 vmm: Encase CpuManager within an Arc<Mutex<>>
This is necessary to be able to add the CpuManager onto the IO bus.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-11-21 09:17:15 -08:00
Rob Bradford
6958ec4922 vmm: Move CPU management code to its own module
Move CpuManager, Vcpu and related functionality to its own module (and
file) inside the VMM crate

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-11-11 15:46:24 +00:00
Rob Bradford
a1a5fe0c93 vmm: Split CPU management into it's own struct
Pull details of vCPU management (booting, pausing, resuming, shutdown)
into it's own structure. This will ultimately enable this to be moved to
its own file and encapsulate all the vCPU handling for the VMM.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-11-08 11:59:21 +01:00
Rob Bradford
0319a4a09a arch: vmm: Move ACPI tables creation to vmm crate
Remove ACPI table creation from arch crate to the vmm crate simplifying
arch::configure_system()

GuestAddress(0) is used to mean no RSDP table rather than adding
complexity with a conditional argument or an Option type as it will
evaluate to a zero value which would be the default anyway.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-11-07 14:02:27 +00:00
Sebastien Boeuf
587a420429 cargo: Update to the latest kvm-ioctls version
We need to rely on the latest kvm-ioctls version to benefit from the
recent addition of unregister_ioevent(), allowing us to detach a
previously registered eventfd to a PIO or MMIO guest address.

Because of this update, we had to modify the current constraint we had
on the vmm-sys-util crate, using ">= 0.1.1" instead of being strictly
tied to "0.2.0".

Once the dependency conflict resolved, this commit took care of fixing
build issues caused by recent modification of kvm-ioctls relying on
EventFd reference instead of RawFd.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-10-31 09:30:59 +01:00
Sebastien Boeuf
8746c16593 vmm: Create AddressManager to own SystemAllocator
In order to reuse the SystemAllocator later at runtime, it is moved into
the new structure AddressManager. The goal is to have a hold onto the
SystemAllocator and both IO and MMIO buses so that we can use them
later.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-10-29 16:48:02 +01:00
Sebastien Boeuf
1870eb4295 devices: Lock the BtreeMap inside to avoid deadlocks
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-10-29 16:48:02 +01:00
Jose Carlos Venegas Munoz
78e2f7a99a api: http: handle cpu according to openapi
openapi definition defines an object for cpus not an integer

Signed-off-by: Jose Carlos Venegas Munoz <jose.carlos.venegas.munoz@intel.com>
2019-10-17 07:39:56 +02:00
Samuel Ortiz
2a0ba7aef8 vmm: vm: Add state validation unit test
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-14 06:35:36 +02:00
Samuel Ortiz
097b30669f vmm: vm: Verify that state transitions are valid
We should return an explicit error when the transition from on VM state
to another is invalid.
The valid_transition() routine for the VmState enum essentially
describes the VM state machine.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-14 06:35:36 +02:00
Samuel Ortiz
d2d3abb13c vmm: Rename Booted vm state to Running
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-10 17:13:44 -07:00
Samuel Ortiz
dbbd04a4cf vmm: Implement VM resume
To resume a VM, we unpark all its vCPU threads.

Fixes: #333

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-10 17:13:44 -07:00
Samuel Ortiz
4ac0cb9cff vmm: Implement VM pause
In order to pause a VM, we signal all the vCPU threads to get them out
of vmx non-root. Once out, the vCPU thread will check for a an atomic
pause boolean. If it's set to true, then the thread will park until
being resumed.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-10 17:13:44 -07:00
Samuel Ortiz
1298b508bf vmm: Manage the exit and reset behaviours from the control loop
So that we don't need to forward an ExitBehaviour up to the VMM thread.
This simplifies the control loop and the VMM thread even further.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-08 18:03:27 -07:00
Sebastien Boeuf
b918220b49 vmm: Support virtio-pci devices attached to a virtual IOMMU
This commit is the glue between the virtio-pci devices attached to the
vIOMMU, and the IORT ACPI table exposing them to the guest as sitting
behind this vIOMMU.

An important thing is the trait implementation provided to the virtio
vrings for each device attached to the vIOMMU, as they need to perform
proper address translation before they can access the buffers.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-10-07 10:12:07 +02:00
Sebastien Boeuf
03352f45f9 arch: Create ACPI IORT table
The virtual IOMMU exposed through virtio-iommu device has a dependency
on ACPI. It needs to expose the device ID of the virtio-iommu device,
and all the other devices attached to this virtual IOMMU. The IDs are
expressed from a PCI bus perspective, based on segment, bus, device and
function.

The guest relies on the topology description provided by the IORT table
to attach devices to the virtio-iommu device.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-10-07 10:12:07 +02:00
Samuel Ortiz
f9daf2e247 vmm: Factorize the vm boot and shutdown code
So that the API handling state machine is cleaner and easier to read.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-04 09:36:33 +02:00
Samuel Ortiz
43b3642955 vmm: Clean Error handling up
We used to have errors definitions spread across vmm, vm, api,
and http.

We now have a cleaner separation: All API routines only return an
ApiResult. All VM operations, including the VMM wrappers, return a
VmResult. This makes it easier to carry errors up to the HTTP caller.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-04 09:36:33 +02:00
Samuel Ortiz
27af983ec9 vmm: Track the VM state
We will expose it through the api/v1/vm.info endpoint.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-04 09:36:33 +02:00
Samuel Ortiz
46cde1a38e vmm: Rename the VM start and stop operations to boot and shutdown
To match the OpenAPI description. And also to map the real life
terminology.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-04 09:36:33 +02:00
Samuel Ortiz
f2de4d0315 vmm: config: Make the cmdline config serializable
The linux_loader crate Cmdline struct is not serializable.
Instead of forcing the upstream create to carry a serde dependency, we
simply use a String for the passed command line and build the actual
CmdLine when we need it (in vm::new()).
Also, the cmdline offset is not a configuration knob, so we remove it.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-04 09:36:33 +02:00
Samuel Ortiz
b14fd37db9 vmm: Make --kernel optional
The kernel path was the only mandatory command line option.
With the addition of the --api-socket option, we can run without a
kernel path and get it later through the API.

Since we can end up with VM configurations that are no longer valid by
default, we need to provide a validation check for it. For now, if the
kernel path is not defined, the VM configuration is invalid.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-10-04 09:36:33 +02:00