Regression caused by "arm64: Reduce PT size estimation complexity"

[Shawn Guo]

Hi folks,

While commit 94d30f476fe1 ("arm64: Reduce PT size estimation complexity")
speeds up the PT size calculation, it seems the result size becomes
unnecessarily bigger than before.  For example, on my device, the
original count_required_pts() returns 6 for get_page_table_size(),
while count_ranges() returns 14 now.  That said, the PT size returned by
get_page_table_size() eventually becomes 64 KiB bigger than before.
This is too much for my device which sets up PT on SRAM, and causing
a regression here.

Could you guys take a look and see whether we can get PT size
calculation speedy *and* compact? :)

Thanks,
Shawn