On 2/25/20 5:24 AM, Peter Maydell wrote: > This check is supposed to be "did the user accidentally specify > some incompatible settings on their '-cpu,+this,-that' option?". > By making it check the actual ID register values, you're turning > it into also a check on "does the implementation specify sane > ID register values", which (a) is useful for TCG but ought to > be an assert and (b) we shouldn't be checking for KVM in case > the h/w is giving us dubious ID values.
Hmm. Because kvm64 unconditionally set VFP and NEON, you're right. It was only kvm32 that was examining id registers. The only consequence of kvm giving us dubious id values that I can see is if ADVSIMD is on, but FP is off, we won't migrate the register set. Do you want me to add a tcg_enabled check, or shall we just drop the patch? The existing test is good enough for just checking the command-line. r~
