Use mio to replace Epoll by uran0sH · Pull Request #316 · rust-vmm/vhost

uran0sH · 2025-08-19T13:44:59Z

Summary of the PR

Epoll is linux-specific. So we use mio, which is a cross-platform event notification, to replace Epoll.

Requirements

Before submitting your PR, please make sure you addressed the following
requirements:

[ x ] All commits in this PR have Signed-Off-By trailers (with
git commit -s), and the commit message has max 60 characters for the
summary and max 75 characters for each description line.
[ x ] All added/changed functionality has a corresponding unit/integration
test.
[ x ] All added/changed public-facing functionality has entries in the "Upcoming
Release" section of CHANGELOG.md (if no such section exists, please create one).
[ x ] Any newly added unsafe code is properly documented.

germag · 2025-09-12T11:45:31Z

this LGTM

uran0sH · 2025-10-08T10:09:03Z

@stefano-garzarella @germag any problems about this?

name1e5s · 2026-02-27T16:27:20Z

Hi @uran0sH , I'm trying to build the vhost crate on macOS and came across this PR. It seems this PR hasn't been updated for a while, are you still working on it? If not, I'd be happy to take over and continue the work to help get it merged. Let me know what you think. Thanks!

uran0sH · 2026-02-27T16:46:35Z

Hi @uran0sH , I'm trying to build the vhost crate on macOS and came across this PR. It seems this PR hasn't been updated for a while, are you still working on it? If not, I'd be happy to take over and continue the work to help get it merged. Let me know what you think. Thanks!

Yeah I will update this pr as soon as possible

stefano-garzarella · 2026-03-05T12:26:40Z

@uran0sH thanks for restarting this work! Please, can you rebase this on current main.

FYI I'll be off from March 7 to 29 with limited internet access, so not sure I can review this before leaving, I'll do my best, but I'll leave to @germag and others handle it.

When register_event we pass in the value of u32/u64, so we don't need to convert it to u16 and then pass it to handle_event. Signed-off-by: Wenyu Huang <huangwenyuu@outlook.com>

We can safely use usize instead of u64, because normally we will not register a data that exceeds the size of usize Signed-off-by: Wenyu Huang <huangwenyuu@outlook.com>

Epoll is linux-specific. So we use mio, which is a cross-platform event notification, to replace Epoll. Signed-off-by: Wenyu Huang <huangwenyuu@outlook.com>

germag · 2026-03-18T11:29:28Z

Hi, this LGTM

@name1e5s can you try to build it again?

The `VhostBackend` and `VhostBackendMut` traits in `vhost/src/backend.rs` take `&EventFd` in `set_vring_call`, `set_vring_kick`, and `set_vring_err`. The implementations of these traits are all Linux-only consumers — kernel vhost drivers (`vhost_kern/`, gated on the `vhost-kern` feature) and the vhost-user frontend (gated on `vhost-user-frontend`). No macOS consumer ever calls these methods. Until now, the unconditional `use vmm_sys_util::eventfd::EventFd;` at the top of `backend.rs` forced every dependent crate (including `vhost-user-backend` consumers like virtiofsd) to compile against `vmm_sys_util::eventfd` even when nothing in the build path called these methods. On macOS that meant requiring an `eventfd` shim inside vmm-sys-util just for the type to resolve. Gate the import, the six trait method signatures, the corresponding methods in `impl<T: VhostBackendMut> VhostBackend for {RwLock,RefCell}<T>`, and the test mock + test on `#[cfg(target_os = "linux")]`. The trait becomes a smaller surface on macOS — exactly the surface that has working consumers there. Linux behaviour is unchanged. This unblocks dropping the macOS `eventfd` shim from vmm-sys-util (rust-vmm/vmm-sys-util#268), and combined with the mio rewrite of `vhost-user-backend`'s event loop (rust-vmm#316), removes the last hard dependency between the macOS build path and `vmm_sys_util::eventfd`. Tested: full virtiofsd → vhost-user-backend → vhost → vmm-sys-util stack builds and tests pass on aarch64-apple-darwin with no eventfd shim in vmm-sys-util. Signed-off-by: Chris Thomas <chris.thomas@antimatter-studios.com>

eric-14 · 2026-05-19T10:42:49Z

@uran0sH building vhost backend on my macOs still fails due to eventfd in vhost/backend.rs. I am porting Vhostbackend trait methods to eventconsumer and eventNotifier. Are you considering taking this approach ?

uran0sH · 2026-05-19T13:52:28Z

@uran0sH building vhost backend on my macOs still fails due to eventfd in vhost/backend.rs. I am porting Vhostbackend trait methods to eventconsumer and eventNotifier. Are you considering taking this approach ?

Yeah, but the changes still are in my own repo: uran0sH@54d673a, not submitted to the upstream. If you want to test, you can use this branch: https://github.com/uran0sH/vhost/tree/macos. And I also change the vhost-device. These changes in uran0sH/vhost-device#2

stefano-garzarella · 2026-05-19T13:57:41Z

@uran0sH building vhost backend on my macOs still fails due to eventfd in vhost/backend.rs. I am porting Vhostbackend trait methods to eventconsumer and eventNotifier. Are you considering taking this approach ?

Yeah, but the changes still are in my own repo: uran0sH@54d673a, not submitted to the upstream. If you want to test, you can use this branch: https://github.com/uran0sH/vhost/tree/macos. And I also change the vhost-device. These changes in uran0sH/vhost-device#2

@uran0sH what is missing to submit it here?

uran0sH · 2026-05-19T14:47:16Z

@uran0sH building vhost backend on my macOs still fails due to eventfd in vhost/backend.rs. I am porting Vhostbackend trait methods to eventconsumer and eventNotifier. Are you considering taking this approach ?

Yeah, but the changes still are in my own repo: uran0sH@54d673a, not submitted to the upstream. If you want to test, you can use this branch: https://github.com/uran0sH/vhost/tree/macos. And I also change the vhost-device. These changes in uran0sH/vhost-device#2

@uran0sH what is missing to submit it here?

I think nothing major is missing. I am not sure whether uran0sH@54d673a should be submitted in this pr or as a sperate one. What do you think?

stefano-garzarella · 2026-05-19T14:51:58Z

I think nothing major is missing. I am not sure whether uran0sH@54d673a should be submitted in this pr or as a sperate one. What do you think?

I think better a separate PR if they are not depending on each other.

eric-14 · 2026-05-19T15:29:31Z

@uran0sH building vhost backend on my macOs still fails due to eventfd in vhost/backend.rs. I am porting Vhostbackend trait methods to eventconsumer and eventNotifier. Are you considering taking this approach ?

Yeah, but the changes still are in my own repo: uran0sH@54d673a, not submitted to the upstream. If you want to test, you can use this branch: https://github.com/uran0sH/vhost/tree/macos. And I also change the vhost-device. These changes in uran0sH/vhost-device#2

@uran0sH what is missing to submit it here?

I think nothing major is missing. I am not sure whether uran0sH@54d673a should be submitted in this pr or as a sperate one. What do you think?

@uran0sH I found the following issue with https://github.com/uran0sH/vhost/tree/macos

I have fixed it on my branch https://github.com/eric-14/rust_vhost_/tree/experimental_mac

stefano-garzarella · 2026-06-19T10:13:05Z

@uran0sH talking with @eric-14 he would like to complete this PR since it's blocking some stuff. Is it okay for you if he will take your work (keeping your authorship of course) and complete it to upstream ASAP?

uran0sH · 2026-06-19T10:18:44Z

@uran0sH talking with @eric-14 he would like to complete this PR since it's blocking some stuff. Is it okay for you if he will take your work (keeping your authorship of course) and complete it to upstream ASAP?

It's okay for me.

germag · 2026-06-22T08:18:37Z

@eric-14 Thanks!

After looking at it (honestly, just skimming) again, there are some minor issues that should be fixed and a couple not so minor:

poller: Mutex<Poll> - Can the mutex be removed (I know VringPollHandler needs sync maybe that's the problem we need to fix idk yet)?, epoll (and the mio impl) is designed to work with a single thread.
println!("ignoring unknown event set:...) -> log::warn! (or debug!?)

Less important:

register_/unregister_listener api data is too general name, but we do: data <= self.backend.num_queues()
register_event() does ...map_err(std::io::Error::other)?; while unregister_event()... .map_err(|e| std::io::Error::other(format!("Failed to deregister fd {fd}: {e}")))?;
We need to check if nothing important is ignored inside event_to_event_set(), for example evt.is_error().

eric-14 · 2026-06-23T17:55:52Z

@germag Thank you for the latest review. I will make the necessary changes and ping you.

eric-14 · 2026-07-03T06:01:04Z

@germag, I have completed the changes. Do you prefer I create a new PR or commit to this PR?

stefano-garzarella · 2026-07-03T07:40:36Z

@germag, I have completed the changes. Do you prefer I create a new PR or commit to this PR?

I'd suggest to start a new PR, I'm not sure if you have rights to push on this.

uran0sH requested review from Ablu, epilys, eryugey, germag, jiangliu, sboeuf, slp and stefano-garzarella as code owners August 19, 2025 13:45

uran0sH mentioned this pull request Aug 19, 2025

Use mio to replace Epoll uran0sH/vhost#2

Closed

uran0sH force-pushed the mio branch 2 times, most recently from cbb223d to 138400e Compare August 19, 2025 14:15

germag reviewed Aug 19, 2025

View reviewed changes

Comment thread vhost-user-backend/src/event_loop.rs Outdated

Comment thread vhost-user-backend/src/backend.rs Outdated

uran0sH force-pushed the mio branch 3 times, most recently from 2b19107 to e53a34c Compare August 19, 2025 17:02

germag reviewed Aug 19, 2025

View reviewed changes

Comment thread vhost-user-backend/src/event_loop.rs Outdated

germag reviewed Aug 19, 2025

View reviewed changes

Comment thread vhost-user-backend/src/event_loop.rs Outdated

Comment thread vhost-user-backend/src/event_loop.rs Outdated

uran0sH force-pushed the mio branch from e53a34c to 2fed57d Compare August 20, 2025 15:19

germag reviewed Aug 26, 2025

View reviewed changes

Comment thread vhost-user-backend/src/event_loop.rs Outdated

Comment thread vhost-user-backend/src/event_loop.rs Outdated

Comment thread vhost-user-backend/src/event_loop.rs Outdated

Comment thread vhost-user-backend/src/event_loop.rs Outdated

Comment thread vhost-user-backend/src/event_loop.rs

uran0sH force-pushed the mio branch 3 times, most recently from be19e30 to 158ca1d Compare August 27, 2025 16:46

uran0sH force-pushed the mio branch 2 times, most recently from 44e7bb3 to 77ba627 Compare September 3, 2025 17:05

germag previously approved these changes Sep 12, 2025

View reviewed changes

Comment thread vhost-user-backend/src/handler.rs Outdated

uran0sH dismissed germag’s stale review via c13a265 September 14, 2025 15:14

uran0sH force-pushed the mio branch from 77ba627 to c13a265 Compare September 14, 2025 15:14

uran0sH force-pushed the mio branch from c13a265 to f96a9b8 Compare October 8, 2025 10:02

uran0sH force-pushed the mio branch 3 times, most recently from 6dde03e to fbc40ce Compare February 28, 2026 08:35

uran0sH requested a review from stefano-garzarella March 5, 2026 11:55

uran0sH added 3 commits March 5, 2026 16:16

Change device_event in handle_event() from u16 to usize

149ce44

When register_event we pass in the value of u32/u64, so we don't need to convert it to u16 and then pass it to handle_event. Signed-off-by: Wenyu Huang <huangwenyuu@outlook.com>

Change u64 to usize in un/register_event

41cdc99

We can safely use usize instead of u64, because normally we will not register a data that exceeds the size of usize Signed-off-by: Wenyu Huang <huangwenyuu@outlook.com>

Use mio to replace Epoll

6c4eed2

Epoll is linux-specific. So we use mio, which is a cross-platform event notification, to replace Epoll. Signed-off-by: Wenyu Huang <huangwenyuu@outlook.com>

uran0sH force-pushed the mio branch from fbc40ce to 6c4eed2 Compare March 5, 2026 16:16

stefano-garzarella mentioned this pull request Apr 21, 2026

macos: add epoll and eventfd compatibility shims rust-vmm/vmm-sys-util#266

Closed

4 tasks

germag marked this pull request as draft May 12, 2026 09:23

christhomas mentioned this pull request May 18, 2026

macos: add signal shim for downstream rust-vmm crates rust-vmm/vmm-sys-util#268

Closed

4 tasks

This was referenced Jul 3, 2026

Epoll to mio migration eric-14/rust_vhost_#1

Merged

epoll to mio migration #362

Open

Uh oh!

Conversation

uran0sH commented Aug 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary of the PR

Requirements

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

germag commented Sep 12, 2025

Uh oh!

uran0sH commented Oct 8, 2025

Uh oh!

name1e5s commented Feb 27, 2026

Uh oh!

uran0sH commented Feb 27, 2026

Uh oh!

stefano-garzarella commented Mar 5, 2026

Uh oh!

germag commented Mar 18, 2026

Uh oh!

eric-14 commented May 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

uran0sH commented May 19, 2026

Uh oh!

stefano-garzarella commented May 19, 2026

Uh oh!

uran0sH commented May 19, 2026

Uh oh!

stefano-garzarella commented May 19, 2026

Uh oh!

eric-14 commented May 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stefano-garzarella commented Jun 19, 2026

Uh oh!

uran0sH commented Jun 19, 2026

Uh oh!

germag commented Jun 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

eric-14 commented Jun 23, 2026

Uh oh!

eric-14 commented Jul 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stefano-garzarella commented Jul 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

uran0sH commented Aug 19, 2025 •

edited

Loading

eric-14 commented May 19, 2026 •

edited

Loading

eric-14 commented May 19, 2026 •

edited

Loading

germag commented Jun 22, 2026 •

edited

Loading

eric-14 commented Jul 3, 2026 •

edited

Loading