[PATCH 1/2] mm/vmalloc: acquire init_mm read lock on huge vmap promotion

[Lorenzo Stoakes]

As per Sashiko, we need to account for x86's slightly bat... interesting
ptdump of non-init_mm page tables.

The fix is to take a nested init_mm lock in ptdump_walk_pgd(), which is
safe as nothing depends on an init_mm lock prior to acquiring an arbitrary
mm's lock.

I think this could potentially be sent as an entirely separate patch, as it
doesn't change what this protects against, nor the x86 CPA patch, both of
which stand on their own.

Cheers, Lorenzo