On Mon, Sep 2, 2024 at 8:20 PM Sergey Senozhatsky <senozhatsky@xxxxxxxxxxxx> wrote: > > On (24/08/27 16:19), Vishal Moola wrote: > > > > Or should we aim to leave zsmalloc as it is currently implemented? > > > > Is this really an option? Yup. That would be similar to your initial suggestion: zpdescs would be a wrapper around struct page and we "chain zpdesc-s together to form a zspage". Although, we may as well aim for an improved implementation of zsmalloc (if at all possible) since implementing the zpdesc wrapper requires us to modify a good chunk of code anyways. That way zsmalloc gets direct wins for its users on top of shrinking struct page as Matthew described before. I believe you would be a better judge of whether or not any improvements are feasible (and worth the effort).
