Wei Liu

Closing this since it is not really a bug. Rather I think it is a configuration issue on the guest side.

This is good work. We can use the coverage data to further guide what tests to develop. Out of interest, why improvement over the existing coverage number after you fix...

Are you proposing to mark those APIs are unsafe? As you already noted, those are internal APIs, so the users are expected to know what they are doing.

Are you going to propose a patch? We probably only need to make `make_user_memory_region` unsafe and then hide all its fields.

> Hello! Is this task still relevant? And if the answer is "yes",can I be assigned to contribute? Yes. I've assigned this to you. Feel free to ask questions.

Cloud Hypervisor's block device is really simple. In this test setup you have, there is one thread that supports the disk. What it does is very minimal -- take requests...

I dug deeper into the io_uring code path in CH. I don't think it is a good idea to set the `ASYNC` flag. That forces the requests to go into...

I did a quick experiment. There is a way to cap the number of worker threads. Changing that doesn't make a huge difference.

Okay, it looks like with sufficient tuning, CH is on par with QEMU. I have been playing with some changes on my side. I think there is still come improvements...