On Thu, 1 Sep 2011 17:24:17 +0200
Andrea Arcangeli [off-list ref] wrote:

From: Andrea Arcangeli <redacted>

Michel while working on the working set estimation code, noticed that calling
get_page_unless_zero() on a random pfn_to_page(random_pfn) wasn't safe, if the
pfn ended up being a tail page of a transparent hugepage under splitting by
__split_huge_page_refcount(). He then found the problem could also
theoretically materialize with page_cache_get_speculative() during the
speculative radix tree lookups that uses get_page_unless_zero() in SMP if the
radix tree page is freed and reallocated and get_user_pages is called on it
before page_cache_get_speculative has a chance to call get_page_unless_zero().

So the best way to fix the problem is to keep page_tail->_count zero at all
times. This will guarantee that get_page_unless_zero() can never succeed on any
tail page. page_tail->_mapcount is guaranteed zero and is unused for all tail
pages of a compound page, so we can simply account the tail page references
there and transfer them to tail_page->_count in __split_huge_page_refcount() (in
addition to the head_page->_mapcount).

While debugging this s/_count/_mapcount/ change I also noticed get_page is
called by direct-io.c on pages returned by get_user_pages. That wasn't entirely
safe because the two atomic_inc in get_page weren't atomic. As opposed other
get_user_page users like secondary-MMU page fault to establish the shadow
pagetables would never call any superflous get_page after get_user_page
returns. It's safer to make get_page universally safe for tail pages and to use
get_page_foll() within follow_page (inside get_user_pages()). get_page_foll()
is safe to do the refcounting for tail pages without taking any locks because
it is run within PT lock protected critical sections (PT lock for pte and
page_table_lock for pmd_trans_huge). The standard get_page() as invoked by
direct-io instead will now take the compound_lock but still only for tail
pages. The direct-io paths are usually I/O bound and the compound_lock is per
THP so very finegrined, so there's no risk of scalability issues with it. A
simple direct-io benchmarks with all lockdep prove locking and spinlock
debugging infrastructure enabled shows identical performance and no overhead.
So it's worth it. Ideally direct-io should stop calling get_page() on pages
returned by get_user_pages(). The spinlock in get_page() is already optimized
away for no-THP builds but doing get_page() on tail pages returned by GUP is
generally a rare operation and usually only run in I/O paths.

This new refcounting on page_tail->_mapcount in addition to avoiding new RCU
critical sections will also allow the working set estimation code to work
without any further complexity associated to the tail page refcounting
with THP.

The patch overall takes the x86_64 allmodconfig text size of
arch/x86/mm/gup.o, mm/huge_memory.o, mm/memory.o and mm/swap.o from a
total of 85059 bytes up to 85973.  That's quite a lot of bloat for a
pretty small patch.

I'm suspecting that most of this is due to the new inlined
get_page_foll(), which is large enough to squish an elephant.  Could
you please take a look at reducing this impact?

...

+/*
+ * The atomic page->_mapcount, starts from -1: so that transitions
+ * both from it and to it can be tracked, using atomic_inc_and_test
+ * and atomic_add_negative(-1).
+ */
+static inline void reset_page_mapcount(struct page *page)

I think we should have originally named this page_mapcount_reset() This
is extra unimportant as it's a static symbol.

...

 static inline void get_page(struct page *page)
 {
+	if (unlikely(PageTail(page)))
+		if (likely(__get_page_tail(page)))
+			return;

OK so we still have approximately one test-n-branch in the non-debug
get_page().

 	/*
 	 * Getting a normal page or the head of a compound page
-	 * requires to already have an elevated page->_count. Only if
-	 * we're getting a tail page, the elevated page->_count is
-	 * required only in the head page, so for tail pages the
-	 * bugcheck only verifies that the page->_count isn't
-	 * negative.
+	 * requires to already have an elevated page->_count.
 	 */
-	VM_BUG_ON(atomic_read(&page->_count) < !PageTail(page));
+	VM_BUG_ON(atomic_read(&page->_count) <= 0);

I wonder how many people enable VM_BUG_ON().  We're pretty profligate
with those things in hot paths.

 	atomic_inc(&page->_count);
-	/*
-	 * Getting a tail page will elevate both the head and tail
-	 * page->_count(s).
-	 */
-	if (unlikely(PageTail(page))) {
-		/*
-		 * This is safe only because
-		 * __split_huge_page_refcount can't run under
-		 * get_page().
-		 */
-		VM_BUG_ON(atomic_read(&page->first_page->_count) <= 0);
-		atomic_inc(&page->first_page->_count);
-	}
 }
 
 static inline struct page *virt_to_head_page(const void *x)

...

+int __get_page_tail(struct page *page)
+{
+	/*
+	 * This takes care of get_page() if run on a tail page
+	 * returned by one of the get_user_pages/follow_page variants.
+	 * get_user_pages/follow_page itself doesn't need the compound
+	 * lock because it runs __get_page_tail_foll() under the
+	 * proper PT lock that already serializes against
+	 * split_huge_page().
+	 */
+	unsigned long flags;
+	int got = 0;

Could be a bool if you like that sort of thing..

+	struct page *page_head = compound_trans_head(page);

Missing newline here

+	if (likely(page != page_head && get_page_unless_zero(page_head))) {
+		/*
+		 * page_head wasn't a dangling pointer but it
+		 * may not be a head page anymore by the time
+		 * we obtain the lock. That is ok as long as it
+		 * can't be freed from under us.
+		 */
+		flags = compound_lock_irqsave(page_head);
+		/* here __split_huge_page_refcount won't run anymore */
+		if (likely(PageTail(page))) {
+			__get_page_tail_foll(page, false);
+			got = 1;
+		}
+		compound_unlock_irqrestore(page_head, flags);
+		if (unlikely(!got))
+			put_page(page_head);
+	}
+	return got;
+}
+EXPORT_SYMBOL(__get_page_tail);

Ordinarily I'd squeak about a global, exported-to-modules function
which is undocumented.  But this one is internal to get_page(), so it's
less necessary.

Still, documenting at least the return value (the "why" rather than the
"what") would make get_page() more understandable.

...

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>