On 14/08/2018 12:02, Peter Maydell wrote: > On 13 August 2018 at 10:47, Stefan Hajnoczi <stefa...@gmail.com> wrote: >> It's an assumption that on-board devices cannot be hot unplugged and >> that the machine type stays alive until QEMU terminates. >> >> Making this assumption saves 1 call to qemu_chr_fe_set_handlers(). >> The cost is that we cannot safely stop the system-on-chip because its >> devices don't clean up properly. >> >> Since cleanup is so trivial here I think it's worthwhile. > I would be more in favour of adding an unrealize function to > random always-present-never-unpluggable devices if we had > better documentation of exactly what the QOM lifecycle is > and what needs to be done in init/realize/unrealize/etc. > As it is, I don't know myself and therefore can't review > whether devices with unrealize methods get it right. And > obviously the code is completely untestable because it can > never run...
I agree. Don't bother with unrealize unless the device hot-unpluggable. What matters is that init/finalize are done right but that is tested by device-introspection-test. Paolo
