Re: [RFC PATCH 2/2] mm/gup: introduce vaddr_pin_pages_remote()

[Jerome Glisse]

On Fri, Aug 16, 2019 at 06:13:55PM +0200, Jan Kara wrote:
> On Fri 16-08-19 11:52:20, Jerome Glisse wrote:
> > On Fri, Aug 16, 2019 at 05:44:04PM +0200, Jan Kara wrote:
> > > On Fri 16-08-19 10:47:21, Vlastimil Babka wrote:
> > > > On 8/15/19 3:35 PM, Jan Kara wrote:
> > > > >> 
> > > > >> So when the GUP user uses MMU notifiers to stop writing to pages whenever
> > > > >> they are writeprotected with page_mkclean(), they don't really need page
> > > > >> pin - their access is then fully equivalent to any other mmap userspace
> > > > >> access and filesystem knows how to deal with those. I forgot out this case
> > > > >> when I wrote the above sentence.
> > > > >> 
> > > > >> So to sum up there are three cases:
> > > > >> 1) DIO case - GUP references to pages serving as DIO buffers are needed for
> > > > >>    relatively short time, no special synchronization with page_mkclean() or
> > > > >>    munmap() => needs FOLL_PIN
> > > > >> 2) RDMA case - GUP references to pages serving as DMA buffers needed for a
> > > > >>    long time, no special synchronization with page_mkclean() or munmap()
> > > > >>    => needs FOLL_PIN | FOLL_LONGTERM
> > > > >>    This case has also a special case when the pages are actually DAX. Then
> > > > >>    the caller additionally needs file lease and additional file_pin
> > > > >>    structure is used for tracking this usage.
> > > > >> 3) ODP case - GUP references to pages serving as DMA buffers, MMU notifiers
> > > > >>    used to synchronize with page_mkclean() and munmap() => normal page
> > > > >>    references are fine.
> > > > 
> > > > IMHO the munlock lesson told us about another one, that's in the end equivalent
> > > > to 3)
> > > > 
> > > > 4) pinning for struct page manipulation only => normal page references
> > > > are fine
> > > 
> > > Right, it's good to have this for clarity.
> > > 
> > > > > I want to add that I'd like to convert users in cases 1) and 2) from using
> > > > > GUP to using differently named function. Users in case 3) can stay as they
> > > > > are for now although ultimately I'd like to denote such use cases in a
> > > > > special way as well...
> > > > 
> > > > So after 1/2/3 is renamed/specially denoted, only 4) keeps the current
> > > > interface?
> > > 
> > > Well, munlock() code doesn't even use GUP, just follow_page(). I'd wait to
> > > see what's left after handling cases 1), 2), and 3) to decide about the
> > > interface for the remainder.
> > > 
> > 
> > For 3 we do not need to take a reference at all :) So just forget about 3
> > it does not exist. For 3 the reference is the reference the CPU page table
> > has on the page and that's it. GUP is no longer involve in ODP or anything
> > like that.
> 
> Yes, I understand. But the fact is that GUP calls are currently still there
> e.g. in ODP code. If you can make the code work without taking a page
> reference at all, I'm only happy :)

Already in rdma next AFAIK so in 5.4 it will be gone :) i have been
removing all GUP users that do not need reference. Intel i915 driver
is a left over i will work some more with them to get rid of it too.

Cheers,
Jérôme