On 26 October 2017 at 18:31, Jason A. Donenfeld <ja...@zx2c4.com> wrote: > On Thu, Oct 26, 2017 at 6:41 PM, Peter Maydell <peter.mayd...@linaro.org> > wrote: >> Those are primarily about migration compatibility -- you should >> be able to live migrate a guest from QEMU 2.6's virt-2.6 machine to >> QEMU 2.7's virt-2.6 machine. That's a stricter requirement than >> the compatibility we want to keep for non-versioned machines, >> where generally we don't want to break command lines but are OK >> to make changes if they don't break those existing uses. >> >> Command lines get baked into all sorts of scripts and so they're >> effectively part of QEMU's "ABI". People don't like it when we >> break their previously working flows :-) > > Fair enough. So in that case, secure=no means serials 1-4 are for > kernel. With secure=yes, serials 1,3-5 are for kernel, with serial 2 > being for trustzone. Seems like that's the best of all worlds.
Now that the 2.11 release is mostly out of the way, I've been working through some of my other todo list items so I've come back to this thread. My suggestion is that we should add a second non-secure UART to the virt board, so if you're using secure=no then you get serial 1 and 2, and for secure=yes serial 1 is the first NS uart, serial 2 is the secure uart and serial 3 is the 2nd NS uart. (I don't really want to add a 4th uart unless there's some good reason to -- maybe we would turn out to want it for the secure side, for instance.) I'll probably write a patch for this next week, unless you want to do it. thanks -- PMM
