LGTM, just one minor comment.

Oh, thanks for noticing! Those are leftovers from when we used the port as the opaque handler.

LGTM, just a minor comment.

I've been thinking a little bit about this, and AFAIK from the gntdev you need to send a signal to a specific event channel port that's provided from an ioctl. I don't think you need refcounting for this, can you simply add something like:

Just some initial thoughts, I will try to do a more in-deep review tomorrow.

Thanks, the code looks good, but it's a complicated device and there are no comments almost anywhere describing what's going on. Please add comments to the major functions (as described in style(9).

Oh, forgot to ask, but since I'm not going to MFC r303076 could you please MFC this to stable/11?

I cannot test this ATM (I might be able to do it tomorrow), but it looks fine, and AFAICT it should also solve the issue I was seeing. Thanks for taking care of this (and sorry for messing it up)!

In D7196#150652, @dumbbell wrote:

Or the reverse: vesa_late_load(flags) calls vesa_load()? Just to avoid code duplication.

But my suggestion could be wrong: vesa_load() checks vesa_init_done and nullifies vesa_adp. I don't know the code enough :(

In D7196#150638, @dumbbell wrote:

I don't know the interaction between vga and vesa here and how all those functions are called and in which order.

I suppose vesa_load() is not called at all in the code path you hit. But can't you use it as vga_sub_configure?

Ping? Sorry for the hurry, but I would like to get this into 11.

Fix comment.

I'm currently running a tinderbox with it and will commit unless anything comes up.

Fix comments.

Ping? I have this and a couple more revisions (D6656 and D6611) that I would like to get committed before the code freeze (although I consider them bug-fixes, so they could also go in during the freeze).

Committed as rS301268.

Committed as rP416310.

Split xn_kick_rings from xn_connect and leave the set_state call at the same
place. Also remove the taskqueue kick.

Committed as rS301197

Committed as rS301198

Add missing dependencies to the xen-tools package, as reported by poudriere
testport:

Disable bash and xgettext configure tests, they are not used on FreeBSD
(just as they where before).

Remove leftover debug printf

Use m_copydata instead of mtodo in order to get the TCP header. Using mtodo
is not correct because it's possible that the TCP header is not in the first
mbuf.

Add more checks to make sure the output interface supports the TSO chain.