This commit adds tests in test_sound_thread_failure to test that
process_control() returns Err() in the following four conditions:
* control request with a single descriptor
* control request in which first descriptor is write-only
* control request in which second descriptor is read-only
* control request with less than three descriptors for control requests
that require three, e.g., PcmInfo, ChmapInfo
Signed-off-by: Matias Ezequiel Vara Larsen <mvaralar@redhat.com>
Previous commit added categories to Cargo.toml, but the only value was
`multimedia::audio`.
Thanks @stefano-garzarella for noticing I missed this.
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
Add self to CODEOWNERS to contribute with reviewing, merging other
PRs and maintaining the vhost-device-sound crate.
Also remove Mathieu Poirier, who is unavailable for maintainer duties at
this time.
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
A subtraction between unsigned integers is made, which by default panics
on overflow. However, we don't really need to know the difference, only
that it is not zero or less.
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
in this context, epoll listener can be already registered via other TX
event, so let it try epoll_modify first to avoid 'silent' failure which
possibly drops packets.
Signed-off-by: Jeongik Cha <jeongik@google.com>
instead of consume() fn name, use read_output() fn which
is more appropriate and in sync with the write_input() fn.
Signed-off-by: Dorinda Bassey <dbassey@redhat.com>
We have the following circular references found by Li Zebin:
VhostUserBackend ==> VhostUserVsockThread ==> VringEpollHandler
In addition to causing a resource leak, this causes also an error
after we merged commit 38caab2 ("vsock: Don't allow duplicate CIDs").
When the VM reboot or shutdown, the application exits with the
following error:
[ERROR vhost_device_vsock] Could not create backend:
CID already in use by another vsock device
This happened because we have these circular references and
VhostUserVsockThread::drop() is never invoked. So, we don't remove
the cid from the map.
Let's fix this problem by simply removing the reference to
VringEpollHandler from VhostUserVsockThread. In fact, we do not
need to keep the reference for the lifetime of VhostUserVsockThread,
as we only need to add the handlers once.
Let's also rename the fields to follow the current VhostUserDaemon
API.
Closes#438
Reported-by: Li Zebin <cutelizebin@gmail.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Add a method to return the length of the descriptor memory region and
make `data_descriptor` field private.
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
We use input/output direction values in many places, yet we lose Rust's
exhaustive static checks when matching for a stream's direction. This
commits adds a Direction enum and From/Into implementations for audio
backend respective direction types.
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
This new version contains alsa and pipewire libraries to build
vhost-device-sound audio backends.
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
- Add some documentation comments
- Rename TxState to IoState in preparation for RX implementation
- Fix inline rust code not being escaped properly in doc comments
- Rename some fields to make their function clearer
- Cleanup some error log messages
- Replace an old TODO comment about queue size with an explanation
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
The latency field specifies how long it will take until the device
finishes playing the I/O message's buffers.
Since the I/O message is dropped as soon as the last bytes are copied
over to the host's internal buffers, assume the message is dropped
almost after the host has started playing this buffer.
This solves the issue of a guest app exiting before the host has done
playing audio because it kept sending new buffers as fast as possible
without waiting (latency).
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
The host's period and buffer sizes were not being calculated correctly.
This resulted in timing mismatches.
This commit calculates the correct sizes by taking the guest's
parameters into account.
Impossible cases when matching with stream parameters will lead to
unreachable!().
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
Use of consume() method was incorrect. This commit fixes the buffer
bounds used with playback in ALSA.
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
Queued buffers were played even after the guest issued PCM_STOP.
This commit:
- makes the ALSA worker threads respect the stopped status instead
- send a start signal to worker threads on PCM_START to handle
pre-buffering
- reduce the time of playback callbacks holding the stream lock
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
When handling controlq messages, handle StreamWithIdNotFound the same
way on all methods:
- if we are passed a &mut ControlMessage, set it to
VIRTIO_SND_S_BAD_MSG.
- if we are passed a stream_id, return StreamWithIdNotFound.
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
Handle some state transition errors more gracefully by retrying ALSA
operations if possible. Also rewrite the PCM_RELEASE handler to make it
more clear.
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
If a PCM_RELEASE control request comes from the guest and the state
transition is invalid by spec standards, release the buffers anyway.
This doesn't change the current behavior.
When dealing with an unrelated vhost-user issue, the state transitions
became invalid because of unexpected errors in the virto-snd driver.
Making this change made the issue obvious while debugging. Plus, the
spec does not state rejecting a PCM lifecycle control request if the
transition is not possible. (The possible state transitions are
non-normative statements.)
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
Add v4l2 wrapper functions to be able
to mock (and test) v4l2r library responses
without the need of having a underlying video device.
Signed-off-by: Albert Esteve <aesteve@redhat.com>
Complete the README file for the video crate
with the help information for the CLI, and a
working example to run and test the device.
Signed-off-by: Albert Esteve <aesteve@redhat.com>
Add test for all modules, with dev-dependencies
(including rstest [1] for parametrized tests), and
infrastructure.
[1] - https://docs.rs/rstest/latest/rstest/
Signed-off-by: Albert Esteve <aesteve@redhat.com>
Add new v4l2-decoder backend, that
uses v4l2r [1] for interactions with the
video device in the host. Specialized for
Linux systems.
[1] - https://github.com/Gnurou/v4l2r
Signed-off-by: Albert Esteve <aesteve@redhat.com>
Initial skeleton for virtio-video crate.
This crate is based on the v3 of the virtio-video
specs patch[1].
It has a big part of the infrastructure
required, although not all commands are implemented,
and does not have any backend available.
Includes support for async responses to the driver
(through VIDEO_EVENT) to QueueResource messages.
[1] -
https://lists.oasis-open.org/archives/virtio-dev/202002/msg00002.html
Related: #364
Signed-off-by: Albert Esteve <aesteve@redhat.com>
Let's use custom pipelines to run CI tests in the nested worskpace.
We have included all the tests we usually run in the main workspace
(rust-vmm-ci/.buildkite/test_description.json), except for "commit-format"
which also covers commits for staging crates.
Let's add a `staging\coverage_config_x86_64.json` for testing coverage of
crates in staging.
All staging tests have `soft_fail = true` to prevent failures in staging
from affecting the CI of the main workspace.
Closes#478
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Lets be super clear and explicit about what it takes to include a
backend in the repository as well as the minimum requirements for
being included in staging.
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Having all the workspace crates under the crates/ directory is
unnecessary. Rust documentation itself recommends all crates to be in
the root directory:
https://doc.rust-lang.org/book/ch14-03-cargo-workspaces.html#creating-the-second-package-in-the-workspace
I paste the text content here, in case the online page ever changes or
becomes unavailable:
## Creating the Second Package in the Workspace
Next, let’s create another member package in the workspace and call it add_one. Change the top-level Cargo.toml to specify the add_one path in the members list:
Filename: Cargo.toml
[workspace]
members = [
"adder",
"add_one",
]
Then generate a new library crate named add_one:
$ cargo new add_one --lib
Created library `add_one` package
Your add directory should now have these directories and files:
├── Cargo.lock
├── Cargo.toml
├── add_one
│ ├── Cargo.toml
│ └── src
│ └── lib.rs
├── adder
│ ├── Cargo.toml
│ └── src
│ └── main.rs
└── target
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
Add a new workspace for devices that have no final specification merged
into the virtio spec.
Closes#459
Signed-off-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
With the refactoring before this commit, `PhysDevice` is `Sync` and
`Send` automatically. So let's drop the unsafe impls to prevent future
type changes to result in unsound behaviour (again).
Fixes: #442
Signed-off-by: Erik Schilling <erik.schilling@linaro.org>
RwLock allows multiple concurrent readers. libgpiod, however, does not
offer thread-safety for any function - even if they look like innocent
getters. Therefore, libgpiod objects do not implement `Sync`.
Now, `PhysDevice` is incorrectly marked as `Sync`, giving multiple threads
access to thw RwLock - resulting in unsafe behaviour.
We fix this by just turning the `RwLock` into a `Mutex`.
Issue: #442
Signed-off-by: Erik Schilling <erik.schilling@linaro.org>
