@forest So if KSM is a possible attack does that make each Argon2 type vulnerable? And probably most of the alternatives.

Guess the first page using knowledge of the salt and a bunch of candidate passwords. Since data isn't overwritten until the next pass that first page of memory remains unmodified for a relatively long time. You'd only need 4k bytes per side-channel password guess and only the amount of time it takes to fill it with password derived data.

That's assuming known salt, KSM for R/W pages, no secret key preprocessing of the password, and you prepare the attack in time for the account with a given salt, knowing when it will be successfully logged into.