Re: [PATCH v21 1/4] mm: add VM_DROPPABLE for designating always lazily freeable mappings

["Jason A. Donenfeld" <Jason@zx2c4.com>]

Hi David,

On Mon, Jul 08, 2024 at 10:11:24AM +0200, David Hildenbrand wrote:
> The semantics are much more intuitive. No need for separate mmap flags.

Agreed.
 
> Likely we'll have to adjust mlock() as well. Also, I think we should 
> just bail out with hugetlb as well.

Ack.

> Further, maybe we want to disallow madvise() clearing these flags here, 
> just to be consistent.

Good thinking.

> As a side note, I'll raise that I am not a particular fan of the 
> "droppable" terminology, at least with the "read 0s" approach.
> 
>  From a user perspective, the memory might suddenly lose its state and 
> read as 0s just like volatile memory when it loses power. "dropping 
> pages" sounds more like an implementation detail.
> 
> Something like MAP_VOLATILE might be more intuitive (similar to the 
> proposed MADV_VOLATILE).
> 
> But naming is hard, just mentioning to share my thought :)

Naming is hard, but *renaming* is annoying. I like droppable simply
because that's what I've been calling it in my head. MAP_VOLATILE is
fine with me though, and seems reasonable enough. So I'll name it that,
and then please don't change your mind about it later so I won't have to
rename everything again. :)

Jason