I used find_get_entries() to retrieve pagecache pages associated with all opened files. For each retrieved page, a put_page() invocation will be needed to revert the page reference count to its previous state. find_get_entries() seems to be racy. In some scenarios, the returned struct page * element contains garbage (perhaps because the page associated with an index has been evicted). Therefore, a put_page() call on such a page will cause kernel panic. How can I avoid this situation?
It seems that the shadow entries of recently evicted pages, etc. (recognized by calling xa_is_value()) will also be returned in the entries parameter. They can be identified and ignored by checking their LSB (i.e., if (unsigned long)entries[i] & 1 is not zero, the element is not a valid page).