There are more 4k boundaries than 16k boundaries. The issue is code compiled for 4k boundaries running on a 16k system.

I'm missing something here. Assuming there are pages at 0k, 16k, 32k etc - all of those pages are aligned on 4k boundaries as 4k > 16k. So code written with the assumption that its pages are 4k aligned should have that assumption met when running with 16k pages. It is still early here and I have only had one cup of coffee. Am I misunderstanding something really obvious?

x86 app might mmap 8kb, then munmap the second 4kb and expect that to work. But not possible on 16k pages.

ah ok, so it would not be pointer alignment inside the pages but instead the assumption that page +4k is a page.

I was a little confused by that in the article as well. It being a granularity issue makes more sense to me.