mm/gup: page->hpage_pinned_refcount: exact pin counts for huge pages

For huge pages (and in fact, any compound page), the GUP_PIN_COUNTING_BIAS
scheme tends to overflow too easily, each tail page increments the head
page->_refcount by GUP_PIN_COUNTING_BIAS (1024).  That limits the number
of huge pages that can be pinned.

This patch removes that limitation, by using an exact form of pin counting
for compound pages of order > 1.  The "order > 1" is required because this
approach uses the 3rd struct page in the compound page, and order 1
compound pages only have two pages, so that won't work there.

A new struct page field, hpage_pinned_refcount, has been added, replacing
a padding field in the union (so no new space is used).

This enhancement also has a useful side effect: huge pages and compound
pages (of order > 1) do not suffer from the "potential false positives"
problem that is discussed in the page_dma_pinned() comment block.  That is
because these compound pages have extra space for tracking things, so they
get exact pin counts instead of overloading page->_refcount.

Documentation/core-api/pin_user_pages.rst is updated accordingly.

Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Link: http://lkml.kernel.org/r/20200211001536.1027652-8-jhubbard@nvidia.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 17 insertions, 23 deletions

diff --git a/Documentation/core-api/pin_user_pages.rst b/Documentation/core-api/pin_user_pages.rst
index 9829345428f8..7e5dd8b1b3f2 100644
--- a/Documentation/core-api/pin_user_pages.rst
+++ b/Documentation/core-api/pin_user_pages.rst
@@ -52,8 +52,22 @@ Which flags are set by each wrapper
 
 For these pin_user_pages*() functions, FOLL_PIN is OR'd in with whatever gup
 flags the caller provides. The caller is required to pass in a non-null struct
-pages* array, and the function then pin pages by incrementing each by a special
-value. For now, that value is +1, just like get_user_pages*().::
+pages* array, and the function then pins pages by incrementing each by a special
+value: GUP_PIN_COUNTING_BIAS.
+
+For huge pages (and in fact, any compound page of more than 2 pages), the
+GUP_PIN_COUNTING_BIAS scheme is not used. Instead, an exact form of pin counting
+is achieved, by using the 3rd struct page in the compound page. A new struct
+page field, hpage_pinned_refcount, has been added in order to support this.
+
+This approach for compound pages avoids the counting upper limit problems that
+are discussed below. Those limitations would have been aggravated severely by
+huge pages, because each tail page adds a refcount to the head page. And in
+fact, testing revealed that, without a separate hpage_pinned_refcount field,
+page overflows were seen in some huge page stress tests.
+
+This also means that huge pages and compound pages (of order > 1) do not suffer
+from the false positives problem that is mentioned below.::
 
  Function
  --------
@@ -99,27 +113,6 @@ pages:
 This also leads to limitations: there are only 31-10==21 bits available for a
 counter that increments 10 bits at a time.
 
-TODO: for 1GB and larger huge pages, this is cutting it close. That's because
-when pin_user_pages() follows such pages, it increments the head page by "1"
-(where "1" used to mean "+1" for get_user_pages(), but now means "+1024" for
-pin_user_pages()) for each tail page. So if you have a 1GB huge page:
-
-* There are 256K (18 bits) worth of 4 KB tail pages.
-* There are 21 bits available to count up via GUP_PIN_COUNTING_BIAS (that is,
-  10 bits at a time)
-* There are 21 - 18 == 3 bits available to count. Except that there aren't,
-  because you need to allow for a few normal get_page() calls on the head page,
-  as well. Fortunately, the approach of using addition, rather than "hard"
-  bitfields, within page->_refcount, allows for sharing these bits gracefully.
-  But we're still looking at about 8 references.
-
-This, however, is a missing feature more than anything else, because it's easily
-solved by addressing an obvious inefficiency in the original get_user_pages()
-approach of retrieving pages: stop treating all the pages as if they were
-PAGE_SIZE. Retrieve huge pages as huge pages. The callers need to be aware of
-this, so some work is required. Once that's in place, this limitation mostly
-disappears from view, because there will be ample refcounting range available.
-
 * Callers must specifically request "dma-pinned tracking of pages". In other
   words, just calling get_user_pages() will not suffice; a new set of functions,
   pin_user_page() and related, must be used.
@@ -228,5 +221,6 @@ References
 * `Some slow progress on get_user_pages() (Apr 2, 2019) <https://lwn.net/Articles/784574/>`_
 * `DMA and get_user_pages() (LPC: Dec 12, 2018) <https://lwn.net/Articles/774411/>`_
 * `The trouble with get_user_pages() (Apr 30, 2018) <https://lwn.net/Articles/753027/>`_
+* `LWN kernel index: get_user_pages() <https://lwn.net/Kernel/Index/#Memory_management-get_user_pages>`_
 
 John Hubbard, October, 2019