Walking between MTD and flashpage, MTD is, for lack of a clearer word, confusing (at least to me); I think that's mostly due to terminology not sufficiently explained, motivated or common.

For comparison, flash is pretty clear in that a page is something that can be erased, and writes happen in alignments and sizes of FLASHPAGE_WRITE_BLOCK_ALIGNMENT / FLASHPAGE_WRITE_BLOCK_SIZE (which I tend to mentally simplify to the maximum thereof -- that is safe, and while they do diverge on some architectures, I don't quite see how I'd use a 2-byte write in a 4-byte alignment in practice, like STM32 seems to allow for some chips).

But MTD introduces sectors and pages, and while sectors are implicitly the erasure units, pages seem to have no inherent semantics at all other than allowing addressing based on them in what vaguely reminds me of the CHS convention of the early ages. This is exacerbated by disconnect from MTD terminology as used on Linux (where I'd assume the term is borrowed from), which deliberately does not talk of pages and sectors but of eraseblocks.

Also odd (I'm leaning toward "missing feature" but with the above I'm not sure) is that while mtd_write says that some devices might enforce alignments, I don't see how the application can get that information. (It's tempting to make a conservative estimate of 8, but with LPC23xx having a FLASHPAGE_WRITE_BLOCK_SIZE of 256, conservative may need very conservative).

I'd like to make a doc PR to enhance things, but right now I don't have a sufficiently consistent mental image to do that. Could you help clarify this?

Pinging @vincent-d and and @AurelienGONCE as the original authors, @jnohlgard as reviewer and @benpicco who added the page addressed operations.