On Tue, 27 Feb 2024 at 14:46, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> For example, a purely "local lock" model would be to just lock all
> pages in order as you write them, and not unlock the previous page
> until you've locked the next one.
Thinking more about this, maybe that doesn't really guarantee much.
The final state would be that even with concurrent overlapping writes,
one writer did its overlapping write fully and you'd never have mixed
ABAB kind of results, but you could still have concurrent readers see
the two writes progressing concurrently.
Of course, since readers aren't serialized as-is, I'm not sure if
"readers can see intermediate state" is anything new or relevant
anyway.
Maybe the worry isn't worth it.
Linus
