There are release schedules for a reason. If someone want to add new features or fixes for old bugs, there is a time for that. -rc4 isn't it. I am sure reasonable exceptions can be made, but it seems like Linus feels the DRM people are abusing the process and just outright ignoring the schedule, and I don't think that is okay. The whole point of an RC cycle is to test things, and you can't test things properly if they keep changing.
RCs are in fact the time and place for all sorts of bug fixes, not just those which concern code added by that particular merge window. However, in this case it seems the kexec fix was deemed inappropriate due to its size, risk of introducing other bugs or simply because it did not or would not get enough testing before the release.
Or at least that's my guess, since the LKML thread itself isn't particularly enlightening and might give the wrong impression. Have a look at the actual patch and Dave's 2nd reply after the merge request, where he says that approach might not be the best anyway.
I have a question, any one know what that radeon kms fix was for?
kexec. Which basically loads a new kernel into memory and executes it. The whole idea is so somewhat nutty as lots of devices have dma in progress. So the fix was to disable certain GPU features that use dma early in the driver init process so that any previously active operations from before kexec will hopefully be stopped before the driver gets a chance to set up the hardware again.