Do we need to handle EXCP_MOE in the kernel?

Can you split out enabling MOPS & using it into separate changes? There is a risk one of the mem*_early changes is missing & it would be easier to revert a change to add them rather than everything that depends on MOPS.

In D54943#1256825, @andrew wrote:

Can you split out enabling MOPS & using it into separate changes? There is a risk one of the mem*_early changes is missing & it would be easier to revert a change to add them rather than everything that depends on MOPS.

mem* changes now in https://reviews.freebsd.org/D55051

It looks like there is a race between identify_cpu_sysinit setting mops_supported and late_ifunc_resolve that calls the ifunc resolvers that may use it. They are both run at SI_SUB_CONFIGURE, SI_ORDER_ANY so could be run in either order.

The easiest option might be to move the former to SI_ORDER_ANY - 1.

D55103 moves it earlier, along with other cleanups.

Do we need mops_supported? It looks like we could check elf_hwcap2 in the resolver functions.

We could also use get_kernel_reg to get the sanitised register, however it's probably safest to only do so after all CPUs have started as there may be a case where not all CPUs support MOPS, or have errata that means we can't use it.