On Mon, Dec 07, 2020 at 09:12:34PM +0000, Peter Maydell wrote: > On Mon, 7 Dec 2020 at 21:06, Claudio Fontana <cfont...@suse.de> wrote: > > As I understand it, for the purpose of code separation, > > we could: > > > > 1) skip do_interrupt move to the separate tcg_ops structure, wait until > > KVM/ARM uses another approach (if ever) > > 2) do the move, and just call arm_cpu_do_interrupt() directly, since for > > KVM64 it is the only one that can be assigned to cc->do_interrupt(). > > > > Which way would you suggest? > > So what's the intention here? To put tcg-only methods in their > own struct so that on a KVM-only QEMU they're compiled out > and attempts to use them are a compile error? In that case > I guess if Arm really is the only user of do_interrupt outside > TCG then we should do what this patch does and do a direct call.
Oh, I thought you were arguing that CPUClass.do_interrupt is not TCG_specific. If you agree with doing what this patch does, the plan sounds good to me. -- Eduardo