On Tue, Sep 16, 2025 at 02:48:55PM +0200, Pratyush Yadav wrote: > On Mon, Sep 15 2025, Mike Rapoport wrote: > > > On Tue, Sep 09, 2025 at 04:33:27PM +0200, Pratyush Yadav wrote: > >> Hi Mike, > >> > >> Also, I am not a fan of using kho_restore_page() directly. I think the > >> vmalloc preservation is a layer above core KHO, and it should use the > >> public KHO APIs. It really doesn't need to poke into internal APIs. If > >> any of the public APIs are insufficient, we should add new ones. > >> > >> I don't suppose I'd insist on it, but something to consider since you > >> are likely going to do another revision anyway. > > > > I think vmalloc is as basic as folio. At some point we probably converge to > > kho_preserve(void *) that will choose the right internal handler. like > > folio, vmalloc, kmalloc etc. > > Sure, but do you need to use the internal APIs? Because doing this way > would miss some improvements for the public APIs. See my patch for > adding more sanity checking to kho_restore_folio() for example: > https://lore.kernel.org/linux-mm/[email protected]/ > > vmalloc preservation would miss this improvement since it uses the > internal API, even though it will clearly benefit from it.
The core restore API is kho_restore_page() and the improvements should land there, IMO. Then whatever uses that core API will benefit from them. -- Sincerely yours, Mike.
