Re: [PATCH 2/3] mm/gup: Introduce get_user_pages_fast_longterm()

[Ira Weiny <ira.weiny@intel.com>]

On Mon, Feb 11, 2019 at 01:13:56PM -0800, John Hubbard wrote:
> On 2/11/19 12:39 PM, Jason Gunthorpe wrote:
> > On Mon, Feb 11, 2019 at 12:16:42PM -0800, ira.weiny@intel.com wrote:
> >> From: Ira Weiny <ira.weiny@intel.com>
> [...]
> >> +static inline int get_user_pages_fast_longterm(unsigned long start, int nr_pages,
> >> +					       bool write, struct page **pages)
> >> +{
> >> +	return get_user_pages_fast(start, nr_pages, write, pages);
> >> +}
> >>  #endif /* CONFIG_FS_DAX */
> >>  
> >>  int get_user_pages_fast(unsigned long start, int nr_pages, int write,
> >> @@ -2615,6 +2622,7 @@ struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
> >>  #define FOLL_REMOTE	0x2000	/* we are working on non-current tsk/mm */
> >>  #define FOLL_COW	0x4000	/* internal GUP flag */
> >>  #define FOLL_ANON	0x8000	/* don't do file mappings */
> >> +#define FOLL_LONGTERM	0x10000	/* mapping is intended for a long term pin */
> > 
> > If we are adding a new flag, maybe we should get rid of the 'longterm'
> > entry points and just rely on the callers to pass the flag?
> > 
> > Jason
> > 
> 
> +1, I agree that the overall get_user_pages*() API family will be cleaner
> *without* get_user_pages_longterm*() calls. And this new flag makes that possible.
> So I'd like to see the "longerm" call replaced with just passing this flag. Maybe
> even as part of this patchset, but either way.

Yes I've thought about this as well.  I have a couple of different versions of
this series which I've been mulling over and this was one of the other
variations.  But see below...

> 
> Taking a moment to reflect on where I think this might go eventually (the notes
> below do not need to affect your patchset here, but this seems like a good place
> to mention this):
> 
> It seems to me that the longterm vs. short-term is of questionable value.

This is exactly why I did not post this before.  I've been waiting our other
discussions on how GUP pins are going to be handled to play out.  But with the
netdev thread today[1] it seems like we need to make sure we have a "safe" fast
variant for a while.  Introducing FOLL_LONGTERM seemed like the cleanest way to
do that even if we will not need the distinction in the future...  :-(

> It's actually better to just call get_user_pages(), and then if it really is
> long-term enough to matter internally, we'll see the pages marked as gup-pinned.
> If the gup pages are released before anyone (filesystem, that is) notices, then
> it must have been short term.
> 
> Doing it that way is self-maintaining. Of course, this assumes that we end up with
> a design that doesn't require being told, by the call sites, that a given gup
> call is intended for "long term" use. So I could be wrong about this direction, but
> let's please consider the possibility.

This is why I've been holding these patches.  I'm also not 100% sure if we will
need the longterm flag in the future.

This is also why I did not change the get_user_pages_longterm because we could
be ripping this all out by the end of the year...  (I hope. :-)

So while this does "pollute" the GUP family of calls I'm hoping it is not
forever.

Ira

[1] https://lkml.org/lkml/2019/2/11/1789

> 
> thanks,
> -- 
> John Hubbard
> NVIDIA