Afaik, sys/types.h should always be the first header to include.

stray newline change

So, you can only connect once to the vm? If the connection gets lost, you won't be able to reconnect?

Shouldn't we lock sc->mtx before writing those values?

Maybe be more verbose and log some kind of error message. This makes it easier to solve issue like when reusing an address.

The sort order looks still wrong according to style (https://man.freebsd.org/cgi/man.cgi?style(9)). sys includes should be first and the rest should be ordered alphabetical.

Note that FreeBSD ships a .clang-format (https://github.com/freebsd/freebsd-src/blob/120ceebab5d4947dfc80d7492effa034a925d43e/.clang-format. So, you can sort this correctly by using the clang-format tool (https://clang.llvm.org/docs/ClangFormat.html)

clang-format -i --lines=30:40 --style=file uart_emul.c

I try to use clang-format but I think neither the original including order nor the patch I have submitted follows the style(9). Should I overwrite the original including order in this circumstance?

Yes. In my knowledge base, the only way to detect the disconnection in TCP is by the return size for the data or just setting the timer.
I don't have any good idea to achieve it without largely change the code in codebase because the mevent don't support timeout.

You can change the original order but that should be posted as additional commit. Try to follow style(9) as much as possible without changing the original order.

I'm not sure but I don't think that we should set errno in user space.

You're lucky that bhyve has a bad habit and keeps asserts in release code. Just use an if statement to avoid that this call is skipped accidentally when bhyve stops keeping asserts in release code.

Sorry but do you have any good idea on this? Because as I know the only way to detect disconnection is rely on read size in nonblocking mode. Can I just pass the returned value as pointer in parameter and just return the return value of read?

Why can't you just check if the function returned -1?

As my experiment, the way to detect disconnection is when return size is zero. You can check the code in the link. This code segment is aim at handling the disconnection of TCP like what we do. So we have to let our ttyread to be possible to distinguish between 0 and -1 return case.

BTW, the function "pci_vtcon_port_add" in the file of the link also change the errno. If this is allowed, I think the original code is more elegant. I don't know if it is another bad habit. This is the first time I contribute on bhyve.

You're reading a char, so you should return a char.

You can ignore this comment as this function is always called with sc->mtx locked.