Currently quite a few tests fail with this patch. I've added a ticket to track the failures: https://bugs.llvm.org/show_bug.cgi?id=32146 . We may want to hold of enabling -verify-machineinstr until all the failures are resolved.
If at all possible I'd prefer to get some triage done on a majority of the bugs first before avoiding the problem in this way - my concern is that XFAIL tagging will be used just to ignore lurking problems that will bite back later on.
@qcolombet 's analyses in PR27481 covers quite a few cases and I think similar fixes should cover many of the remaining x86 failures, and I wouldn't be surprised if other targets are very similar.
Well intuitively I would expect this to be a bigger issue, mostly because you have to get target maintainers on board to fix their targets.
Because of that I was thinking more in the lines of adding a bool TargetSubtargetInfo::isMachineVerifierClean() and adding a return false implementation to all the broken targets. That way we could fix the targets one by one.