On Thu, 31 Oct 2024 at 17:31, Michael Tokarev <m...@tls.msk.ru> wrote: > What I see is that the whole thing - plus one more commit right > before the mentioned refactoring, ea8618382aba, "target/arm: Correct > comments about M-profile FPSCR" - applies cleanly to 9.0 and 8.2. > For 7.2 though, this is definitely too much because there, tcg code > has different rules for the temps, - in particular, it frees all > temps explicitly, so new code wont fit there anyway, resulting in > leaks and requiring major re-review. > > Being niche, I'd not bother with that even in 8.2, but the fact > this refactoring applies nicely.. :) > > It's probably the first time when the difference between many > changes but tested in master, vs smaller possible change, is quite > significant. I prefer for it to just work ;)) > > 8.2 is in ubuntu noble (LTS), so it might be with us a bit longer > than some other series, but in ubuntu they build with KVM support, > so the bug isn't there.
Note that the bug is not in "TCG + KVM" or "TCG only" configs, but specifically "KVM without TCG" (i.e you configured with '--disable-tcg' on an aarch64 host). -- PMM
