Rob Bradford

I think we established that v28.0 would be our first LTS release and what this means has been documented (thanks @weltling)

I analysed the before/after in terms of VM exits. The number of I/O exits doesn't vary across the patch but just using the kernel tracked time to init there is...

Interesting thing to note. With perf I don't seen anything different when using `perf diff` between the runs. However with an added patch (https://github.com/cloud-hypervisor/cloud-hypervisor/commit/028f8ea2962fc8e29602a48b42f238c35283b81b) that adds consumption for VM exit...

> @rbradford so you're saying that you don't see more VM exits but when you increase the time spent in the VMM for each VM exit, then you can observe...

I identified the cause as this change making multiple updates to the MSI-X table for the devices. Updating the table may result in an `ioctl()` being made to reprogram the...

In theory if we can get the new kernel to also minimise the number of MSI-X table updates it makes it will improve boot time further.

Given that nobody has proposed an implementation for a guest agent I guess there is no demand as such; further such an agent seems very specific to how Cloud Hypervisor...

> I would propose we update MSRV if: > > a.) A dependency requires it, b.) If we want to use a new feature and that MSRV is at least...

@liuw Any update on this?

Simply open a PR that tries `test_vfio` on an on-demand worker.