The VMM guest memory was cloned (copied) everywhere the code needed to
have ownership of it. In order to clean the code, and in anticipation
for future support of modifying this guest memory instance at runtime,
it is important that every part of the code share the same instance.
Because VirtioDevice implementations need to have access to it from
different threads, that's why Arc must be used in this case.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
When there are no available descriptors in the queue (observed when the
network interface hasn't been brought up by the kernel) stop waiting for
notifications that the TAP fd should be read from.
This avoids a situation where the TAP device has data avaiable and wakes
up the virtio-net thread only for the virtio-net thread not read that
data as it has nowhere to put it.
When there are descriptors available in the queue then we resume waiting
for the epoll event on the TAP fd.
This bug demonstrated itself as 100% CPU usage for cloud-hypervisor
binary prior to the guest network interface being brought up. The
solution was inspired by the Firecracker virtio-net code.
Fixes: #208
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Latest clippy version complains about our existing code for the
following reasons:
- trait objects without an explicit `dyn` are deprecated
- `...` range patterns are deprecated
- lint `clippy::const_static_lifetime` has been renamed to
`clippy::redundant_static_lifetimes`
- unnecessary `unsafe` block
- unneeded return statement
All these issues have been fixed through this patch, and rustfmt has
been run to cleanup potential formatting errors due to those changes.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This patch enables the vhost-user protocol features to let the slave
initiates some request towards the master (VMM). It also takes care of
receiving the requests from the slave and take appropriate actions based
on the request type.
The way the flow works now are as follow:
- The VMM creates a region of memory that is made available to the
guest by exposing it through the virtio-fs PCI BAR 2.
- The virtio-fs device is created by the VMM, exposing some protocol
features bits to virtiofsd, letting it know that it can send some
request to the VMM through a dedicated socket.
- On behalf of the guest driver asking for reading or writing a file,
virtiofsd sends a request to the VMM, asking for a file descriptor to
be mapped into the shared memory region at a specific offset.
- The guest can directly read/write the file at the offset of the
memory region.
This implementation is more performant than the one using exclusively
the virtqueues. With the virtqueues, the content of the file needs to be
copied to the queues every time the guest is asking to access it.
With the shared memory region, the virtqueues become the control plane
where the libfuse commands are sent to virtiofsd. The data plane is
literally the whole memory region which does not need any extra copy of
the file content. The only penalty is the first time a file is accessed,
it needs to be mapped into the VMM virtual address space.
Another interesting case where this solution will not perform as well as
expected is when a file is larger than the region itself. This means the
file needs to be mapped in several times, but more than that this means
it needs to be remapped every time it's being accessed.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
When the cache_size parameter from virtio-fs device is not empty, the
VMM creates a dedicated memory region where the shared files will be
memory mapped by the virtio-fs device.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
In the context of shared memory regions, they could not be present for
most of the virtio devices. For this reason, we prefer dedicate a BAR
for the shared memory regions.
Another reason is that memory regions, if there are several, can be
allocated all at once as a contiguous region, which then can be used as
its own BAR. It would be more complicated to try to allocate the BAR 0
holding the regular information about the virtio-pci device along with
the shared memory regions.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This patch cleans up the VirtioDevice trait. Since some function are PCI
specific and since they are not even used, it makes sense to remove them
from the trait definition.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Based on the newly added SharedMemoryConfig capability to the virtio
specification, and based on the fact that it is not tied to the type of
transport (pci or mmio), we can create as part of the VirtioDevice trait
a new method that will provide the shared memory regions associated with
the device.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Based on the latest version of the virtio specification, the structure
virtio_pci_cap has been updated and a new structure virtio_pci_cap64 has
been introduced.
virtio_pci_cap now includes a field "id" that does not modify the
existing structure size since there was a 3 bytes reserved field
already there. The id is used in the context of shared memory regions
which need to be identified since there could be more than one of this
kind of capability.
virtio_pci_cap64 is a new structure that includes virtio_pci_cap and
extends it to allow 64 bits offsets and 64 bits region length. This is
used in the context of shared memory regions capability, as we might need
to describe regions of 4G or more, that could be placed at a 4G offset
or more in the associated BAR.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The length of the PCI capability as it is being calculated by the guest
was not accurate since it was not including the implicit 2 bytes offset.
The reason for this offset is that the structure itself does not contain
the capability ID (1 byte) and the next capability pointer (1 byte), but
the structure exposed through PCI config space does include those bytes.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The numbr of bytes read was being incorrectly increased by the potential
length of the end of the sliced data rather than the number of bytes
that was in the range. This caused a panic when the the network was
under load by using iperf.
It's important to note that in the Firecracker code base the function
that read_slice() returns the number of bytes read which is used to
increment this counter. The VM memory version however only returns the
empty unit "()".
Fixes: #166
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
One of the features of the virtio console device is its size can be
configured and updated. Our first iteration of the console device
implementation is lack of this feature. As a result, it had a
default fixed size which could not be changed. This commit implements
the console config feature and lets us change the console size from
the vmm side.
During the activation of the device, vmm reads the current terminal
size, sets the console configuration accordinly, and lets the driver
know about this configuration by sending an interrupt. Later, if
someone changes the terminal size, the vmm detects the corresponding
event, updates the configuration, and sends interrupt as before. As a
result, the console device driver, in the guest, updates the console
size.
Signed-off-by: A K M Fazla Mehrab <fazla.mehrab.akm@intel.com>
Relying on the newly added MSI-X helper, the interrupt callback checks
the interrupts are enabled on the device before to try triggering the
interrupt.
Fixes#156
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The structure of the vmm-sys-util crate has changed with lots of code
moving to submodules.
This change adjusts the use of the imported structs to reference the
submodules.
Fixes: #145
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Update all dependencies with "cargo upgrade" with the exception of
vmm-sys-utils which needs some extra porting work.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
The existing code taking care of the epoll loop was too restrictive as
it was propagating the error returned from the epoll_wait() syscall, no
matter what was the error. This causes the epoll loop to be broken,
leading to a non-functional virtio device.
This patch enforces the parsing of the returned error and prevent from
the error propagation in case it is EINTR, which stands for Interrupted.
In case the epoll loop is interrupted, it is appropriate to retry.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
As per the VIRTIO specification, every virtio device configuration can
be updated while the guest is running. The guest needs to be notified
when this happens, and it can be done in two different ways, depending
on the type of interrupt being used for those devices.
In case the device uses INTx, the allocated IRQ pin is shared between
queues and configuration updates. The way for the guest to differentiate
between an interrupt meant for a virtqueue or meant for a configuration
update is tied to the value of the ISR status field. This field is a
simple 32 bits bitmask where only bit 0 and 1 can be changed, the rest
is reserved.
In case the device uses MSI/MSI-X, the driver should allocate a
dedicated vector for configuration updates. This case is much simpler as
it only requires the device to send the appropriate MSI vector.
The cloud-hypervisor codebase was not supporting the update of a virtio
device configuration. This patch extends the existing VirtioInterrupt
closure to accept a type that can be Config or Queue, so that based on
this type, the closure implementation can make the right choice about
which interrupt pin or vector to trigger.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The virtio console device is a console for the communication between
the host and guest userspace. It has two parts: the device and the
driver. The console device is implemented here as a virtio-pci device
to the guest. On the other side, the guest OS expected to have a
character device driver which provides an interface to the userspace
applications.
The console device can have multiple ports where each port has one
transmit queue and one receive queue. The current implementation only
supports one port. For data IO communication, one or more empty
buffers are placed in the receive queue for incoming data, and
outgoing characters are placed in the transmit queue. Details spec
can be found from the following link.
https://docs.oasis-open.org/virtio/virtio/v1.1/csprd01/virtio-v1.1-csprd01.pdf#e7
Apart from the console, for the communication between guest and host,
the Cloud Hypervisor has a legacy serial device implemented. However,
the implementation of a console device lets us be independent of legacy
pin-based interrupts without losing the logs and access to the VM.
Signed-off-by: A K M Fazla Mehrab <fazla.mehrab.akm@intel.com>
There is alignment support for AddressAllocator but there are occations
that the alignment is known only when we call allocate(). One example
is PCI BAR which is natually aligned, means for which we have to align
the base address to its size.
Signed-off-by: Chao Peng <chao.p.peng@linux.intel.com>
This patch adds the support for both IO and Memory BARs by expecting
the function allocate_bars() to identify the type of each BAR.
Based on the type, register_mapping() insert the address range on the
appropriate bus (PIO or MMIO).
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This commit enhances the current msi-x code hosted in the pci crate
in order to be reused by the vfio crate. Specifically, it creates
several useful methods for the MsixCap structure that can simplify
the caller's code.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The type of interrupt_evt has changed along with the addition of an
msix_config member for the virtio device.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
When reading from or writing to a PCI BAR to handle a VM exit, we need
to have the BAR address itself to be able to support multiple BARs PCI
devices.
Fixes: #87
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
With the range base for the IO/MMIO vm exit address, a device with
multiple ranges has all the needed information for resolving which of
its range the exit is coming from
Fixes: #87
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
This commit introduces the implementation of the virtio-pmem device
based on the pending proposal of the virtio specification here:
https://lists.oasis-open.org/archives/virtio-dev/201903/msg00083.html
It is also based on the kernel patches coming along with the virtio
proposal: https://lkml.org/lkml/2019/6/12/624
And it is based off of the current crosvm implementation found in
devices/src/virtio/pmem.rs relying on commit
bb340d9a94d48514cbe310d05e1ce539aae31264
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The vhost-user-fs or virtio-fs device allows files and directories to
be shared between host and guest. This patch adds the implementation
of this device to the cloud-hypervisor device model.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
BusDevice includes two methods which are only for PCI devices, which should
be as members of PciDevice trait for a better clean high level APIs.
Signed-off-by: Jing Liu <jing2.liu@linux.intel.com>
When the KVM capability KVM_CAP_SIGNAL_MSI is not present, the VMM
falls back from MSI-X onto pin based interrupts. Unfortunately, this
was not working as expected because the VirtioPciDevice object was
always creating an MSI-X capability structure in the PCI configuration
space. This was causing the guest drivers to expect MSI-X interrupts
instead of the pin based generated ones.
This patch takes care of avoiding the creation of a dedicated MSI-X
capability structure when MSI is not supported by KVM.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
As mentioned in the PCI specification, the Function Mask from the
Message Control Register can be set to prevent a device from injecting
MSI-X messages. This supersedes the vector masking as it interacts at
the device level.
Here quoted from the specification:
For MSI and MSI-X, while a vector is masked, the function is prohibited
from sending the associated message, and the function must set the
associated Pending bit whenever the function would otherwise send the
message. When software unmasks a vector whose associated Pending bit is
set, the function must schedule sending the associated message, and
clear the Pending bit as soon as the message has been sent. Note that
clearing the MSI-X Function Mask bit may result in many messages
needing to be sent.
This commit implements the behavior described above by reorganizing
the way the PCI configuration space is being written. It is indeed
important to be able to catch a change in the Message Control
Register without having to implement it for every PciDevice
implementation. Instead, the PciConfiguration has been modified to
take care of handling any update made to this register.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
The current MSI-X implementation completely ignores the values found
in the Vector Control register related to a specific vector, and never
updates the Pending Bit Array.
According to the PCI specification, MSI-X vectors can be masked
through the Vector Control register on bit 0. If this bit is set,
the device should not inject any MSI message. When the device
runs into such situation, it must not inject the interrupt, but
instead it must update the bit corresponding to the vector number
in the Pending Bit Array.
Later on, if/when the Vector Control register is updated, and if
the bit 0 is flipped from 0 to 1, the device must look into the PBA
to find out if there was a pending interrupt for this specific
vector. If that's the case, an MSI message is injected and the
bit from the PBA is cleared.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
As mentioned in the PCI specification:
If a dedicated Base Address register is not feasible, it is
recommended that a function isolate the MSI-X structures from
the non-MSI-X structures with aligned 8 KB ranges rather than
the mandatory aligned 4 KB ranges.
That's why this patch ensures that each structure present on the
BAR is 8KiB aligned.
It also fixes the MSI-X table and PBA sizes so that they can support
up to 2048 vectors, as specified for MSI-X.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
In order to factorize the complexity brought by closures, this commit
merges IrqClosure and MsixClosure into a generic InterruptDelivery one.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
In order to allow virtio-pci devices to use MSI-X messages instead
of legacy pin based interrupts, this patch implements the MSI-X
support for cloud-hypervisor. The VMM code and virtio-pci bits have
been modified based on the "msix" module previously added to the pci
crate.
Fixes#12
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Because we cannot always assume the irq fd will be the way to send
an IRQ to the guest, this means we cannot make the assumption that
every virtio device implementation should expect an EventFd to
trigger an IRQ.
This commit organizes the code related to virtio devices so that it
now expects a Rust closure instead of a known EventFd. This lets the
caller decide what should be done whenever a device needs to trigger
an interrupt to the guest.
The closure will allow for other type of interrupt mechanism such as
MSI to be implemented. From the device perspective, it could be a
pin based interrupt or an MSI, it does not matter since the device
will simply call into the provided callback, passing the appropriate
Queue as a reference. This design keeps the device model generic.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Most of the code is taken from crosvm(bbd24c5) but is modified to
be adapted to the current VirtioDevice definition and epoll
implementation.
A new command option '--rng' is provided and it gives one the option
to override the entropy source which is /dev/urandom by default.
Signed-off-by: Chao Peng <chao.p.peng@linux.intel.com>
In order to provide connectivity through network interface between
host and guest, this patch introduces the virtio-net backend.
This code is based on Firecracker commit
d4a89cdc0bd2867f821e3678328dabad6dd8b767
It is a trimmed down version of the original files as it removes the
rate limiter support. It has been ported to support vm-memory crate
and the epoll handler has been modified in order to run a dedicated
epoll loop from the device itself. This epoll loop runs in its own
dedicated thread.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
As it is necessary to return the interrupt EventFD and the queue EventFD
to the transport layer upon reset the activate function has been
modified to clone these descriptors as well as the underlying disk
itself.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
If the driver triggers a reset by writing zero into the status register
then reset the underlying device if supported. A device reset also
requires resetting various aspects of the queue.
In order to be able to do a subsequent reactivate it is required to
reclaim certain resources (interrupt and queue EventFDs.) If a device
reset is requested by the driver but the underlying device does not
support it then generate an error as the driver would not be able to
configure it anyway.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Add the BSD and Apache license.
Make all crosvm references point to the BSD license.
Add the right copyrights and identifier to our VMM code.
Add Intel copyright to the vm-virtio and pci crates.
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
After the virtio-blk device support has been introduced in the
previous commit, the vmm need to rely on this new device to boot
from disk images instead of initrd built into the kernel.
In order to achieve the proper support of virtio-blk, this commit
had to handle a few things:
- Register an ioevent fd for each virtqueue. This important to be
notified from the virtio driver that something has been written
on the queue.
- Fix the retrieval of 64bits BAR address. This is needed to provide
the right address which need to be registered as the notification
address from the virtio driver.
- Fix the write_bar and read_bar functions. They were both assuming
to be provided with an address, from which they were trying to
find the associated offset. But the reality is that the offset is
directly provided by the Bus layer.
- Register a new virtio-blk device as a virtio-pci device from the
vm.rs code. When the VM is started, it expects a block device to
be created, using this block device as the VM rootfs.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This commit introduces the virtio-blk backend implementation, which is
the first device implementing the VirtioDevice trait.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Copied from crosvm 107edb3e with one main modification: VirtioPciDevice
implements BusDevice.
We need this modification because it is the only way for us to be able
to add a VirtioPciDevice to the MMIO bus. Bus insertion takes a
BusDevice. The fact that VirtioPciDevice implements PciDevice which
itself implements BusDevice does not mean that Rust will automatically
downcast a VirtioPciDevice into a BusDevice.
crosvm works around that issue by having the PCI, virtio and BusDevice
implementations in the same crate.
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
Copied from Firecracker 17a9089d for the queue implementation and from
crosvm 107edb3e for the device Trait. The device trait has some PCI
specific methods hence its crosvm origin.
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
