I have two questions regarding dynfpustate:
1 - first, is it safe to have a sleepable allocation right into
fputrap(), especially in case of memory pressure?
2 - following 1), why not kmem_alloc the pcb FPU within cpu_attach?
There are already allocations that happen there (in particular:
reserved pages for MMU), memory pressure is low at this time (during
boot, and a hotplugged CPU would not be usable until cpu_attach
return, no risk of having a working CPU but without FPU state
tracking). IMO it would make sense to do it there. The FPU is not
expected to change dynamically.