On 1/9/20 2:07 PM, John Hubbard wrote:

On 1/7/20 2:45 PM, John Hubbard wrote:

quoted

Hi,

The "track FOLL_PIN pages" would have been the very next patch, but it is
not included here because I'm still debugging a bug report from Leon.
Let's get all of the prerequisite work (it's been reviewed) into the tree
so that future reviews are easier. It's clear that any fixes that are
required to the tracking patch, won't affect these patches here.

This implements an API naming change (put_user_page*() -->
unpin_user_page*()), and also adds FOLL_PIN page support, up to
*but not including* actually tracking FOLL_PIN pages. It extends
the FOLL_PIN support to a few select subsystems. More subsystems will
be added in follow up work.

Hi Andrew and all,

To clarify: I'm hoping that this series can go into 5.6.

Meanwhile, I'm working on tracking down and solving the problem that Leon
reported, in the "track FOLL_PIN pages" patch, and that patch is not part of
this series.

Hi Andrew and all,

Any thoughts on this?

As for the not-included-yet tracking patch, my local testing still suggests the
need to allow for larger refcounts of huge pages (in other words, I can write a test
to pin huge pages many times, and overflow with the same backtrace that Leon has
reported).

The second struct page (I recall Jan suggested) can hold those, so I'm going to proceed
with that approach, while waiting to see if Leon has any more test data for me.

Again, I think this series is worth getting out of the way, in the meantime.


thanks,
-- 
John Hubbard
NVIDIA