No reason. I'll change it to = before committing.

As an aside, the RISC-V designers have made an interesting architectural choice here. Hypothetically, we could speed up fork() on x86 by removing write permission at the upper levels of the page table, and sharing the lower-level page table pages until pmap_enter() tries to create a writeable mapping at a lower level. Only then would you allocate a new page table page, copying the old one, and removing write access from the other PTEs in that copied page table page. However, this way of expressing superpage mappings eliminates the possibility of such an optimization.