On 3/28/21 9:27 PM, Richard Henderson wrote: > On 3/26/21 1:35 PM, Claudio Fontana wrote: >> Here a new version of the series that enables kvm-only builds. >> >> The goal here is to enable the KVM-only build, but there is >> some additional cleanup too. >> >> In this iteration I mostly fixed existing issues, and added an attempt >> to put more content in cpu-sve. More splitting still required >> for both cpu-sve and cpu-auth. >> >> Comments welcome, thanks, > > My number 1 takeaway is that I don't think we should bother trying so hard to > compile out aarch64-specific code. The number of ifdefs seems to be quite > high > in the end. >
Understand - it turns out to be useful in practice though to go through some effort to improve the distinction between aarch64-specific code , aarch32-specific code, and code that is common to both, even though we might later decide to just have basically TARGET_AARCH64 be always true (ie removing it) in a future series. > The cleanup that I think would be better would be to *remove* the 32-bit-only > qemu-system-arm and build it all into a single binary. That would reduce the > number of build combinations and could in turn simplify or speed up CI. Still, I think it is interesting for the reader of the code, and in trying to debug issues, to be mindful about which code is 32bit-only, which code is 64-bit only, which is common. Not that this series fixes everything in this regard. > > We probably cannot remove qemu-arm for 32-bit linux-user binaries. But my > guess is the amount of aarch64-specific code that would leak in is pretty > minimal. The bulk of the difference is in the set of tcg helpers, which are > segregated already. > > The tcg/kvm split is much more interesting, and that part of the patch set > looks pretty good. > > > r~ > Thanks, I'll send a new version momentarily, we can continue from there. Ciao, Claudio
