On Tue, Dec 17, 2019 at 07:39:43AM -0700, Jens Axboe wrote:

+static void buffered_put_page(struct page *page, bool clear_mapping)
+{
+	if (clear_mapping)
+		page->mapping = NULL;
+	put_page(page);
+}

I'm not a huge fan of the variable name 'clear_mapping'.  It describes
what it does rather than why it does it.  So maybe 'drop_immediate'?
Or 'uncached'?

 		if (!page) {
 			if (iocb->ki_flags & IOCB_NOWAIT)
 				goto would_block;
+			/* UNCACHED implies no read-ahead */
+			if (iocb->ki_flags & IOCB_UNCACHED) {
+				did_dio_begin = true;
+				/* block truncate for UNCACHED reads */
+				inode_dio_begin(inode);

I think this needs to be:

				if (!did_dio_begin)
					inode_dio_begin(inode);
				did_dio_begin = true;

otherwise inode->i_dio_count is going to be increased once per uncached
page.  Do you have a test in your test-suite that does I/O to more than
one page at a time?