I usually do the releases on a Sunday afternoon, but occasionally cut the merge window short by a day just to keep people on their toes, and make sure people learn not to send in last-minute pull requests. No gaming the merge window to the last day. This is one such release.
To be fair, the reason I did it a day early this time around is less to stop people from trying to time their pull requests, and mostly because this has been a pretty big merge window, and not hugely enjoyable. I ended up stopping doing pulls twice during the merge window just because I was chasing down some random problem. That tends to turn my busy merge window time from "busy" to "somewhat stressful".
But hey, it's all good now, and while 4.9 looks to be a big release and we had a couple of hiccups, on the whole things look normal. The big new thing is the greybus addition, which Greg swears is actually getting used. But the bulk of the changes by far is actually a lot of small details under the hood, as usual.
My own favorite "small detail under the hood" happens to be Andy Lutomirski's new virtually mapped kernel stack allocations. They make it easier to find and recover from stack overflows, but the effort also cleaned up some code, and added a kernel stack mapping cache to avoid any performance downsides. Al has also been working on some vfs and uaccess cleanups (particularly a goo splice model cleanup) that I follow. But realistically, what _I_ consider cool small details is just my own personal thing, there's things all over.
The virtual stack mapping also happens to mean that people who try to do DMA from temporary buffers on the stack ("Don't do it!") now really need to change their evil ways. So there is some fallout from this, and I expect a couple of drivers to need minor fixes. But it's all for a good cause, really (and it isn't all that common, because doing DMA from the stack really has never been a good idea, and is generally not even workable in most situations).
But there really is a lot of other things going on, and the shortlog that I do for other releases is much too big during rc1. So as usual, I'm appending my "mergelog" instead, which gives a very high-level view of what I merged and from whom. And as usual, I want to point out that the person I merge from is not necessarily the person who did the work: we had 1500 people involved in this release, only the top-level maintainers get credited in my mergelog.