Skip to content

Gpu socket final #5

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Draft
mtjhrc wants to merge 27 commits into
base: upstream
Choose a base branch
from
Draft

Gpu socket final #5

mtjhrc wants to merge 27 commits into from

Conversation

@mtjhrc
Copy link
Owner

@mtjhrc mtjhrc commented Apr 9, 2024 •
edited
Loading

Important changes:

  • Changed many doc comments

  • methods now take &self instead of &mut self (like I explained on Slack why)

  • Modified a lot doc comments, made all of them start with uppercase (some were lowercase some uppercase)

  • set_dmabuf_scanout needs to accept file descriptor inside an Option to be able to fulfill the requirement from spec:

    To disable a scanout, the dimensions width/height are set to 0, there is no file descriptor passed.

    (The file descriptor is passed in normal case when enabling a scanout, so it needs to be inside an Option.

  • update_dmabuf_scanout shouldn't accept the argument data: &[u8] according to spec

    Note: there is no data payload, since the scanout is shared thanks to DMABUF, that must have been set previously with VHOST_USER_GPU_DMABUF_SCANOUT.

  • renamed cursor_pos_update -> cursor_update
    (It also updates the cursor image not just the position and it is more in line with the name in the spec)

  • get_protocol_features shouldn't accept VhostUserU64 but only return it, spec says:

    request payload: N/A

  • impl<T: Req> VhostUserMsgValidator for VhostUserGpuMsgHeader<T> also checks the if header flags are valid now.

  • Added constant defintions for use with get_ /set_protocol_features call

@mtjhrc mtjhrc force-pushed the gpu-socket-final branch 7 times, most recently from 6563882 to 7cc0c02 Compare April 10, 2024 16:22
@MatiasVara @slp
vhost-user-backend: Prepare v0.15.0 release
bd6bf13 
This release changes the dependency for virtio-queue to 0.12.0.

Signed-off-by: Matias Ezequiel Vara Larsen <mvaralar@redhat.com>
@mtjhrc mtjhrc force-pushed the gpu-socket-final branch 20 times, most recently from d4703f6 to 427321e Compare April 17, 2024 11:42
dorindabassey
dorindabassey reviewed Apr 17, 2024
View reviewed changes
@mtjhrc mtjhrc force-pushed the gpu-socket-final branch from 6c85c1e to eea0e6d Compare April 17, 2024 17:56
@mtjhrc mtjhrc force-pushed the gpu-socket-final branch 4 times, most recently from 6a8c374 to 9c50abb Compare June 11, 2024 12:20
aesteve-rh and others added 8 commits June 19, 2024 13:11
@aesteve-rh @slp
vhost_user: separate non standard msgs
85e0097 
FS_* vhost-user backend request messages in
BackendReq enum did not make it into the
standard yet. Actually, currently there
are already SHARED_OBJECT_* messages that
use the message type IDs for FS_* types.

To fix this and avoid the issue for future
message IDs, let's move FS_* types further
away and prevent collision to occur.
This way we only need to change it once.

Signed-off-by: Albert Esteve <aesteve@redhat.com>
@aesteve-rh @slp
vhost_user: join frontend handler thread on tests
9323902 
Assertion fails on the frontend thread in tests
can go unnoticed.
To avoid that the main thread finishes before
and solve this issue, let's `join()` the thread
at the end of the test.

Signed-off-by: Albert Esteve <aesteve@redhat.com>
@aesteve-rh @slp
vhost_user: add shared object msg
76c347c 
Add SHARED_OBJECT_* vhost-user request
IDs to align with standard. Implement
backend functions to send these
requests.

Add SHARED_OBJECT protocol feature.

Add GET_SHARED_OBJECT to the frontend
requests (not implemented, this is just to
have the placeholder and align with current
standard).

Signed-off-by: Albert Esteve <aesteve@redhat.com>
@aesteve-rh @slp
vhost_user.backend_req: check shared_object flag
94be3df 
Check for VHOST_USER_PROTOCOL_F_SHARED_OBJECT flag
before allowing the backend to send SHARED_OBJECT_*
requests.

If it does have negotiated the protocol feature,
avoid sending the message and return error code
(i.e., !=0).

Signed-off-by: Albert Esteve <aesteve@redhat.com>
@aesteve-rh @slp
CHANGELOG.md: Add shared object feature
873a9b9 
Signed-off-by: Albert Esteve <aesteve@redhat.com>
@dependabot @stefano-garzarella
build(deps): update nix requirement from 0.28 to 0.29 in the vhost group
48081dd 
Updates the requirements on [nix](https://github.com/nix-rust/nix) to permit the latest version.

Updates `nix` to 0.29.0
- [Changelog](https://github.com/nix-rust/nix/blob/master/CHANGELOG.md)
- [Commits](nix-rust/nix@v0.28.0...v0.29.0)

---
updated-dependencies:
- dependency-name: nix
  dependency-type: direct:production
  dependency-group: vhost
...

Signed-off-by: dependabot[bot] <support@github.com>
@aesteve-rh @germag
vhost_user: Remove support for FS_* requests
569ef07 
Since these non-standard backend request
message types are confirmed not to be used
anymore, let's just remove them.

```
pub enum BackendReq {
    ...
    /// Virtio-fs draft: map file content into the window.
    FS_MAP = 100,
    /// Virtio-fs draft: unmap file content from the window.
    FS_UNMAP = 1001,
    /// Virtio-fs draft: sync file content.
    FS_SYNC = 1002,
    /// Virtio-fs draft: perform a read/write from an fd directly to GPA.
    FS_IO = 1003,
    ...
}
```

Closes: rust-vmm#213
Signed-off-by: Albert Esteve <aesteve@redhat.com>
@aesteve-rh @germag
CHANGELOG.md: Remove support for FS_* requests
6755c6c 
Signed-off-by: Albert Esteve <aesteve@redhat.com>
@mtjhrc mtjhrc force-pushed the gpu-socket-final branch from a84b32f to 76c9f38 Compare July 1, 2024 08:54
alyssais and others added 16 commits July 1, 2024 14:13
@alyssais @stefano-garzarella
vhost_user: ignore unknown features from backend
7554d06 
virtiofsd 1.11.0 added support for VHOST_USER_PROTOCOL_F_DEVICE_STATE.
Upgrading virtiofsd meant that the latest released version of Cloud
Hypervisor (39.0) was no longer able to communicate with it, because
rather than just ignoring the unsupported feature, it got an
unrecoverable "invalid message" error.  This demonstrates that it's
better for frontends to just ignore offers of unsupported features.
If the backend requires some feature, it'll get a chance to know that
when we send VHOST_USER_SET_PROTOCOL_FEATURES anyway.

Signed-off-by: Alyssa Ross <hi@alyssa.is>
@mtjhrc
vhost_user: Make Endpoint generic over Header not just Req
9db9688 
The Endpoint was previusly generic only over the Request.
This commit allows Endpoint to be used with a protocol with a slightly
different header such as the QEMU GPU protocol on the socket from
VHOST_USER_GPU_SET_SOCKET.
https://www.qemu.org/docs/master/interop/vhost-user-gpu.html

Signed-off-by: Matej Hrica <mhrica@redhat.com>
@mtjhrc
vhost_user: Move enum_value macro from message.rs to mod.rs
3e13255 
Move the enum_value macro into vhost_user module and make it public within
this module. This will be used in the next commit for defining gpu message
enum.

Signed-off-by: Matej Hrica <mhrica@redhat.com>
@mtjhrc
vhost_user: Add gpu_message protocol bits
3fca35a 
This commit adds the basic definitions of GPU commands on the socket
obtained from VHOST_USER_GPU_SET_SOCKET. This also introduces a new
feature flag `gpu-socket` in the vhost crate.

Signed-off-by: Matej Hrica <mhrica@redhat.com>
@mtjhrc
Forward VHOST_USER_GPU_SET_SOCKET to VhostUserBackend impls
abc7b62 
The VHOST_USER_GPU_SET_SOCKET is only handled when the feature gpu-socket
is enabled. This also introduces a GpuBackend for handling comunication
over the socket.
Signed-off-by: Matej Hrica <mhrica@redhat.com>
@mtjhrc
vhost_user: Support sending msg larger than MAX_MSG_SIZE
a02be30 
Introduce another constant MAX_MSG_SIZE that is part of the MsgHeader
trait. For now this is only used for sending messages.
Consider using the more specific trait constant everywhere.

The VHOST_USER_GPU_UPDATE and VHOST_USER_GPU_CURSOR_UPDATE contain image
data and are larger than the existing MAX_MSG_SIZE.

The existing MAX_MSG_SIZE wasn't really a limitation of the protocol,
just an implementation detail limitation in this crate.

Signed-off-by: Matej Hrica <mhrica@redhat.com>
@mtjhrc
vhost_user: Add GpuBackend::get_display_info
15a4fde 
Add a get_display_info() method and the related reply structs to send
VHOST_USER_GPU_GET_DISPLAY_INFO and receive the reply.

Signed-off-by: Dorinda Bassey <dbassey@redhat.com>
Signed-off-by: Matej Hrica <mhrica@redhat.com>
@mtjhrc
vhost_user: Add GpuBackend::get_edid
a7720ba 
Add a method and related structs to send VHOST_USER_GPU_GET_EDID and
receive the reply.

Signed-off-by: Dorinda Bassey <dbassey@redhat.com>
@mtjhrc
vhost_user: Add GpuBackend::set_scanout
e470b96 
Add a method set_scanout() that sends VHOST_USER_GPU_SCANOUT and doesn't
wait for a reply.

Signed-off-by: Dorinda Bassey <dbassey@redhat.com>
@mtjhrc
vhost_user: Add GpuBackend::update_scanout
7215dca 
Add a method and related struct to send VHOST_USER_GPU_UPDATE.

The data part of the message is not part of the struct like sugested by
the spec but a separate argument to update_scanout. This is necessary
because of limitations of having an unsized array inside of struct in Rust.
But this aproach seems preferable anyway, because it allows the consumer of
the crate to store the array in a diferent location than the struct.

Signed-off-by: Dorinda Bassey <dbassey@redhat.com>
Signed-off-by: Matej Hrica <mhrica@redhat.com>
@mtjhrc
vhost_user: Add GpuBackend::set_ and update_dmabuf_scanout
4d7bc1a 
Add methods and related structs to send the VHOST_USER_GPU_DMABUF_SCANOUT
and VHOST_USER_GPU_DMABUF_UPDATE update messages.

Signed-off-by: Matej Hrica <mhrica@redhat.com>
@mtjhrc
vhost_user: Add GpuBackend::get_ and set_protocol_features
65a0f2f 
Add get_protocol_features method to send
VHOST_USER_GPU_GET_PROTOCOL_FEATURES and receive the reply. Also add
a method to send SET_PROTOCOL_FEATURES.
Introduce VhostUserGpuProtocolFeatures bitmap that defines possible feature
flags.

Signed-off-by: Dorinda Bassey <dbassey@redhat.com>
@mtjhrc
vhost_user: Add GpuBackend::cursor_* message methods
91f9fb5 
Add methods to send cursor related messages: VHOST_USER_GPU_CURSOR_POS,
VHOST_USER_GPU_CURSOR_POS_HIDE and VHOST_USER_GPU_CURSOR_UPDATE.

VhostUserGpuCursorUpdate's `data` field is passed as a separate argument
into cursor_update method. The type is also an u8 array instead of u32
array like in the spec. Having the type be u8 array makes it easier to work
with the data without unsafe code for the consumer of this crate. Having it
be a separate argument is useful, because the user of the method doesn't
have to copy the array into a struct.

Signed-off-by: Dorinda Bassey <dbassey@redhat.com>
@mtjhrc
vhost_user: Add GpuBackend::set_dmabuf_scanout2
d71a010 
Add a method and related struct to send the VHOST_USER_GPU_DMABUF_SCANOUT2
message.

Signed-off-by: Matej Hrica <mhrica@redhat.com>
@mtjhrc
Modify changelogs mentioning upcoming gpu-socket feature
397b1d8 
Adds links to PR in CHANGELOG.md for both vhost and vhost-user-backend.

Signed-off-by: Matej Hrica <mhrica@redhat.com>
@mtjhrc
CI: Build and test with gpu-socket feature
b361166 
Compile and test with `gpu-socket` feature enabled.

Signed-off-by: Matej Hrica <mhrica@redhat.com>
@mtjhrc mtjhrc force-pushed the gpu-socket-final branch from 76c9f38 to b361166 Compare July 3, 2024 08:51
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

1 more reviewer

@dorindabassey dorindabassey dorindabassey left review comments

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

7 participants

@mtjhrc @dorindabassey @MatiasVara @eugpermar @aesteve-rh @alyssais