Re: [PATCH v2 2/5] mm: introduce external memory hinting API

[Minchan Kim <minchan-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

On Mon, Jan 20, 2020 at 02:24:05PM +0100, Michal Hocko wrote:
> On Mon 20-01-20 15:39:35, Kirill A. Shutemov wrote:
> > On Mon, Jan 20, 2020 at 12:27:22PM +0100, Michal Hocko wrote:
> > > On Mon 20-01-20 13:24:35, Kirill Tkhai wrote:
> [...]
> > > > Even two threads on common memory need a synchronization
> > > > to manage mappings in a sane way. Managing memory from two processes
> > > > is the same in principle, and the only difference is that another level
> > > > of synchronization is required.
> > > 
> > > Well, not really. The operation might simply attempt to perform an
> > > operation on a specific memory area and get a failure if it doesn't
> > > reference the same object anymore. What I think we need is some form of
> > > a handle to operate on. In the past we have discussed several
> > > directions. I was proposing /proc/self/map_anon/ (analogous to
> > > map_files) where you could inspect anonymous memory and get a file
> > > handle for it. madvise would then operate on the fd and then there
> > > shouldn't be a real problem to revalidate that the object is still
> > > valid. But there was no general enthusiasm about that approach. There
> > > are likely some land mines on the way.
> > 
> > Converting anon memory to file-backed is bad idea and going to backfire.
> 
> I didn't mean to convert. I meant to expose that information via proc
> the same way we do for file backed mappings. That shouldn't really
> require to re-design the way how anonymous vma work IMO. But I haven't
> tried that so there might be many gotchas there.
> 
> There are obvious things to think about though. Such fd cannot be sent
> to other processes (SCM stuff), mmap of the file would have to be
> disallowed and many others I am not aware of. I am not even pushing this
> direction because I am not convinced about how viable it is myself. But
> it would sound like a nice extension of the existing mechanism we have
> and a file based madvise sounds attractive to me as well because we
> already have that.

I am not a fan of fd based approach but I already reserved last argument
of the API as extendable field so we could use the field as "fd" when we
really need that kinds of fine-grained synchronization model if it's not
enough with SGISTOP, freezer and so.