Re: [PATCH 1/3] mm/mlock.c: convert put_page() to put_user_page*()

From: Michal Hocko
Date: Fri Aug 09 2019 - 13:52:20 EST


On Fri 09-08-19 15:58:13, Jan Kara wrote:
> On Fri 09-08-19 10:23:07, Michal Hocko wrote:
> > On Fri 09-08-19 10:12:48, Vlastimil Babka wrote:
> > > On 8/9/19 12:59 AM, John Hubbard wrote:
> > > >>> That's true. However, I'm not sure munlocking is where the
> > > >>> put_user_page() machinery is intended to be used anyway? These are
> > > >>> short-term pins for struct page manipulation, not e.g. dirtying of page
> > > >>> contents. Reading commit fc1d8e7cca2d I don't think this case falls
> > > >>> within the reasoning there. Perhaps not all GUP users should be
> > > >>> converted to the planned separate GUP tracking, and instead we should
> > > >>> have a GUP/follow_page_mask() variant that keeps using get_page/put_page?
> > > >>>
> > > >>
> > > >> Interesting. So far, the approach has been to get all the gup callers to
> > > >> release via put_user_page(), but if we add in Jan's and Ira's vaddr_pin_pages()
> > > >> wrapper, then maybe we could leave some sites unconverted.
> > > >>
> > > >> However, in order to do so, we would have to change things so that we have
> > > >> one set of APIs (gup) that do *not* increment a pin count, and another set
> > > >> (vaddr_pin_pages) that do.
> > > >>
> > > >> Is that where we want to go...?
> > > >>
> > >
> > > We already have a FOLL_LONGTERM flag, isn't that somehow related? And if
> > > it's not exactly the same thing, perhaps a new gup flag to distinguish
> > > which kind of pinning to use?
> >
> > Agreed. This is a shiny example how forcing all existing gup users into
> > the new scheme is subotimal at best. Not the mention the overal
> > fragility mention elsewhere. I dislike the conversion even more now.
> >
> > Sorry if this was already discussed already but why the new pinning is
> > not bound to FOLL_LONGTERM (ideally hidden by an interface so that users
> > do not have to care about the flag) only?
>
> The new tracking cannot be bound to FOLL_LONGTERM. Anything that gets page
> reference and then touches page data (e.g. direct IO) needs the new kind of
> tracking so that filesystem knows someone is messing with the page data.
> So what John is trying to address is a different (although related) problem
> to someone pinning a page for a long time.

OK, I see. Thanks for the clarification.

> In principle, I'm not strongly opposed to a new FOLL flag to determine
> whether a pin or an ordinary page reference will be acquired at least as an
> internal implementation detail inside mm/gup.c. But I would really like to
> discourage new GUP users taking just page reference as the most clueless
> users (drivers) usually need a pin in the sense John implements. So in
> terms of API I'd strongly prefer to deprecate GUP as an API, provide
> vaddr_pin_pages() for drivers to get their buffer pages pinned and then for
> those few users who really know what they are doing (and who are not
> interested in page contents) we can have APIs like follow_page() to get a
> page reference from a virtual address.

Yes, going with a dedicated API sounds much better to me. Whether a
dedicated FOLL flag is used internally is not that important. I am also
for making the underlying gup to be really internal to the core kernel.

Thanks!
--
Michal Hocko
SUSE Labs