Code review comment for ~rafaeldtinoco/ubuntu/+source/qemu:lp1805256-bionic-refix

Generally ok with me, the patch seems small and reasonable now.
And going forward this was the right change, but since the issue only ever occurred/reproduced/matched on armhf/arm64 I wonder a bit.

I know that Paolo has gone to length to improve the relax the writes.
But for the SRU I think this would be even safer if you'd throw in "if arch = arm* as preprocessor macros to all of these changes.

That way the regression risk would reduce a lot while giving the fix to the arch where it is needed.

I'll mark it as "need info" as this is more starting a discussion than anything else.