If nobody else is going to fix this I can try

Just in case - I'm working with this version of RISC-V tools:

> riscv64-unknown-elf-objcopy --version
GNU objcopy (SiFive Binutils 2.32.0-2019.08.0) 2.32
Copyright (C) 2019 Free Software Foundation, Inc.
This program is free software; you may redistribute it under the terms of
the GNU General Public License version 3 or (at your option) any later version.
This program has absolutely no warranty.

This is very alike good and bad cases:

fail4.hex,fail4.txt

I think I know what's going on

Small sections that is not aligned to 256-byte pages are written properly only if they are not overlapping the page boundaries - if it's longer than end of the page then it's corrupting beginning of that page and then verification fails

The same thing with reading

looks like page reading can't overlap page boundaries as well

or can?...

no - reading is good 🙂

only writing must be done inside of the 256-byte page

Updated my pull-request with complete fix