Commit Graph

526 Commits

Author SHA1 Message Date
Cathy Zhang
8c33eb3069 src: Add integration test for vhost-user-net backend
An integration test relying on the new vhost-user-net backend now
replaces the previous test using the QEMU test backend. This allows
us to avoid building the QEMU backend, and we now really exercise the
vhost-user-net implementation as it is used for the ssh communication
in this test.

Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 13:06:00 -07:00
Cathy Zhang
f6d1a9d9b8 src: Add vhost-user-net backend
Create vhost-user-net backend with Tap interface, to offload network
transaction from cloud-hypervisor. The goal is to provide flexibility
about the backend being in use, but also more security as it will allow
users to isolate the backend with different security profiles since it
will run as a dedicated process on the host.

Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 13:06:00 -07:00
Cathy Zhang
d724511a91 vm-virtio: Add set_protocol_features in vhost-user-net
While implement vhost-user-net backend with Tap interface, it keeps
failed to enable the tx vring, since there is a checking in
slave_req_handler.rs to require acked_protocol_features to be setup
as a pre-requirement, which is filled by set_protocol_features call.
Add this call in vhost-user-net device implementation to address the issue.

Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
2019-09-30 13:06:00 -07:00
Sebastien Boeuf
9ff42060e0 vhost_user_backend: Fix clippy issues
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 13:06:00 -07:00
Cathy Zhang
db151819f1 Cargo.toml: Add workspace config changes
Remove workspace from vhost_user_backend/Cargo.toml to have
vhost-user-backend compiled in cloud-hypervisor. Add workspace in
Cargo.toml to have vhost-user-backend consumed by vhost-user-net.

Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
2019-09-30 13:06:00 -07:00
Rob Bradford
9356af80c6 arch: Mark the PCI MMCONFIG region as reserved in the E820 tables
The PCI Express Firmware specification says that the region may
be included in the E820 tables (but it must always be in the ACPI
tables.)

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-30 18:00:31 +01:00
Rob Bradford
6a4a931b9e arch: acpi: Reserve the PCI MMCONFIG region
The PCI Express Firmware spec says that the region to be used for PCI
MMCONFIG should be reserved as part of the motherboard's resources in
the ACPI tables.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-30 18:00:31 +01:00
Rob Bradford
038f198129 arch: acpi: Fix off-by-one error in size of PCI device region
When comparing offsets it is necessry to increment by one to give the
appropriate size.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-30 18:00:31 +01:00
Rob Bradford
833a3d456c pci, vmm: Expose the PCI bus for configuration via MMIO
Refactor the PCI datastructures to move the device ownership to a PciBus
struct. This PciBus struct can then be used by both a PciConfigIo and
PciConfigMmio in order to expose the configuration space via both IO
port and also via MMIO for PCI MMCONFIG.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-30 18:00:31 +01:00
Rob Bradford
c0ca3b6b8e arch: acpi, layout: Correctly calculate and expose PCI MMCONFIG area
The PCI MMCONFIG area must be below 4GiB and must not be part of the
device space. Shrink the device area and put the PCI MMCONFIG region
above it.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-30 18:00:31 +01:00
Rob Bradford
a9eb352aea arch: acpi: Patch the 32-bit PCI device area in the APCI table
Patch the table with the currently used constants. This will be relevant
when we want to adjust the size of the PCI device area to accomodate the
PCI MMCONFIG region.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-30 18:00:31 +01:00
Sebastien Boeuf
09392f0530 vhost_user_backend: Fix remaining issues
This commit fixes all the remaining issues that were found as part of
the integration with vhost-user-net.

It fixes the way to notify that a vring is used, by using the proper
EventFd.

It removes the process_queue() function from the trait, since the
complexity it was introducing was leading to deadlocks with mutexes.

It moves the register/unregister functions for registering custom events
from the backend, from the VringEpollHandler to the VringWorker. This
allows for a lot of simplification and solve a deadlock issue.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Cathy Zhang
527dd68ce1 vhost_user_backend: Update vmm_va_to_gpa with adding offset
The original logic does not has any problem without offset, since the
current offset is zero. However, if offset is not zero, while convert
vmm address to backend process address, it needs to consider the
offset.

Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
2019-09-30 09:26:11 -07:00
Cathy Zhang
4a1af7f63c vhost-user-backend: Correct error handling in run
The error handling here to trigger break epoll seems not correct,
epoll will be ended once one event is handled, no matter successfully
or failed. Fix it.

Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
2019-09-30 09:26:11 -07:00
Cathy Zhang
c4309515c9 vhost-user-backend: Remove one checking from set_features
The vhost-user protocol does not indicate set_features could not
be issued more than once, the checking is not needed at all, and
prevent communication between master and slave. Remove it to
fix the issue.

Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
2019-09-30 09:26:11 -07:00
Sebastien Boeuf
347611b0c7 vhost_user_backend: Pass a backend that can be modified
This patch modifies the library so that a consumer can update the
backend after it's been passed to the daemon.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Sebastien Boeuf
f14ab872ec vhost_user_backend: Give access to the EpollVringHandler
By letting the consumer of this crate getting access to the vring
handler, we will be able to let it perform several actions without
producing a deadlock.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Sebastien Boeuf
cfc8c39446 vhost_user_backend: Provide some default trait implementations
We cannot expect every backend to support GET_CONFIG and SET_CONFIG
commands. That's why this patch adds some default implementations for
the trait VhostUserBackend regarding both get_config() and set_config()
functions.

Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Sebastien Boeuf
1aab372a06 vhost_user_backend: Make the backend a server
The code needs to initialize a listener to accept connection from the
VMM being the client in this case.

Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Sebastien Boeuf
c1b26b1fab vhost_user_backend: Don't process disabled queues
Every time an event is triggered, it needs to be read, but only based on
the status of the vring (enabled or not) will decide if the queue needs
to be processed.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Sebastien Boeuf
d80ac43ef1 vhost_user_backend: Remove useless started field
The Queue structure already contains a field "ready" that can be used to
track the status of the vrings.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Sebastien Boeuf
5f076923d3 vhost_user_backend: Allow for proper error propagation
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Sebastien Boeuf
36de390caf vhost_user_backend: Make some trait functions as mutable
Let's be realistic, the trait VhostUserBackend will need to have mutable
self for some functions like handle_event, process_queue and set_config,
which is the reason why this commit needs to introduce a RwLock on the
backend instance that was passed around as a simple Arc.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Sebastien Boeuf
d4f7f73bc8 vhost_user_backend: Move to a per-queue RwLock
Instead of locking every queues whenever something needs to be updated,
this patch modifies the code design to lock each Vring independently.
This allows for much finer granularity, and will allow multiple queues
to be handled at the same time.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Sebastien Boeuf
4ed81894aa vhost_user_backend: Replace Mutex with RwLock when possible
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Sebastien Boeuf
2e2cad91ae vhost_user_backend: Add new crate
The purpose of this new crate is to provide a common library to all
vhost-user backend implementations. The more is handled by this library,
the less duplication will need to happen in each vhost-user daemon.

This crate relies a lot on vhost_rs, vm-memory and vm-virtio crates.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
2019-09-30 09:26:11 -07:00
Rob Bradford
b5ee9212c1 vmm, devices: Use APIC address constant
In order to avoid introducing a dependency on arch in the devices crate
pass the constant in to the IOAPIC device creation.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 11:48:30 -07:00
Rob Bradford
162791b571 vmm, arch: Use IOAPIC constants from layout in DeviceManager
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 11:48:30 -07:00
Rob Bradford
8207b2e97d arch: Move addresses for GDT and IDT tables to layout module
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 11:48:30 -07:00
Rob Bradford
a0455167d0 vmm: Use layout constant for kernel command line
Remove the unnecessary field on CmdlineConfig and switch to using the
common offset.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 11:48:30 -07:00
Rob Bradford
1bc47507b7 arch: Move initial page table addresses to layout module
These are part of RAM and are used as the initial page table entries for
booting the OS and firmware (identity mapping.)

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 11:48:30 -07:00
Rob Bradford
5ba61f6d5e arch: Move address of MPTABLE to layout module
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 11:48:30 -07:00
Rob Bradford
6d6e290000 arch: Move APIC and IOAPIC addresses into layout
Move the addresses used for the APIC and IOAPIC into our new memory
layout module.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 11:48:30 -07:00
Rob Bradford
0e7a1fc923 arch, vmm: Start documenting major regions of RAM and reserved memory
Using the existing layout module start documenting the major regions of
RAM and those areas that are reserved. Some of the constants have also
been renamed to be more consistent and some functions that returned
constant variables have been replaced.

Future commits will move more constants into this file to make it the
canonical source of information about the memory layout.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 08:55:47 -07:00
Rob Bradford
f63cb85f93 net_util: Implement fmt::Display for MacAddr
Updated clippy does not like the declaration of a "to_string()" function
and instead requires fmt::Display to be implemented.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 08:05:56 -07:00
Rob Bradford
ff1cb11946 arch: Use if-let notation
To keep updated clippy happy.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 08:05:56 -07:00
Rob Bradford
df2516f229 vhost_rs: Add clippy override
clippy wants us to use "if let Some(fd) = fd {}" but to combine that
with && is an experimental feature.

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 08:05:56 -07:00
Rob Bradford
2ae3919181 vm-virtio: Fix formatting
With the 1.38.0 toolchain rustfmt is even stricter about formatting now

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
2019-09-27 08:05:56 -07:00
Samuel Ortiz
8188074300 main: Start the VMM thread
We now start the main VMM thread, which will be listening for VM and IPC
related events.
In order to start the configured VM, we no longer directly call the VM
API but we use the IPC instead, to first create and then start a VM.

Fixes: #303

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00
Samuel Ortiz
e235c6de4f vmm: Add VM creation and startup helpers
Based on the newly defined Cloud Hypervisor IPC, those helpers send
VmCreate and VmStart requests respectively. This will be used by the
main thread to create and start a VM based on the CLI parameters.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00
Samuel Ortiz
151f96e454 vmm: Add a VMM thread startup routine
This starts the main, single VMM thread, which:

1. Creates the VMM instance
2. Starts the VMM control loop
3. Manages the VMM control loop exits for handling resets and shutdowns.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00
Samuel Ortiz
2f1ff23066 vmm: (Re-)Introduce a VMM structure
Unlike the Vmm structure we removed with commit bdfd1a3f, this new one
is really meant to represent the VM monitoring/management object.
For that, we implement a control loop that will replace the one that's
currently embedded within the Vm structure itself.
This will allow us to decouple the VM lifecycle management from the VM
object itself, by having a constantly running VMM control loop.

Besides the VM specific events (exit, reset, stdin for now), the VMM
control loop also handles all the Cloud Hypervisor IPC requests.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00
Samuel Ortiz
4671a5831f vmm: Move the EpollContext implementation to lib
The VMM thread and control loop will be the sole consumer of the
EpollContext and EpollDispatch API, so let's move it to lib.rs.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00
Samuel Ortiz
03ab6839c1 vmm: Introduce Cloud Hypervisor IPC
Cloud Hypervisor IPC is a simple, mpsc based protocol for threads to
send command to the furture VMM thread. This patch adds the API
definition for that IPC, which will be used by both the main thread
to e.g. start a new VM based on the CLI arguments and the future HTTP
server to relay external requests received from a local Unix domain
socket.
We are moving it to its own "api" module because this is where the
external API (HTTP based) will also be implemented.

The VMM thread will be listening for IPC requests from an mpsc receiver,
process them and send a response back through another mpsc channel.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00
Samuel Ortiz
6710a39b5a vmm: Pass the exit and reset fds to the vm creation method
As we're going to move the control loop to the VMM thread, the exit and
reset EventFds are no longer going to be owned by the VM.
We pass a copy of them when creating the Vm instead.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00
Samuel Ortiz
feb1c33084 vmm: Add a VM config getter
We will need it from the VMM thread, when trying to reboot a VM.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00
Samuel Ortiz
47167a658e vmm: Add a VM console handling method
In order to handle the VM STDIN stream from a separate VMM thread
without having to export the DeviceManager, we simply add a console
handling method to the Vm structure.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00
Samuel Ortiz
ea7abc6c80 vmm: Add a VM stop method
In order to transfer the control loop to a separate VMM thread, we want
to shrink the VM control loop to a bare minimum.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00
Samuel Ortiz
e6ef9ece2c vmm: Move the tty setting to the VM start routine
We want to shrink the control loop to a bare minimal.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00
Samuel Ortiz
2e9d815701 vmm: Use a reference counted VmConfig when creating a new VM
Once passed to the VM creation routine, a VmConfig structure is
immutable. We can simply carry a Arc of it instead of a reference.
This also allows us to remove any lifetime bound from our VM.

Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2019-09-26 16:21:14 +02:00