OSDN Git Service

mm, hwpoison: fix condition in free hugetlb page path
authorNaoya Horiguchi <naoya.horiguchi@nec.com>
Sat, 25 Dec 2021 05:12:45 +0000 (21:12 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 25 Dec 2021 20:20:55 +0000 (12:20 -0800)
When a memory error hits a tail page of a free hugepage,
__page_handle_poison() is expected to be called to isolate the error in
4kB unit, but it's not called due to the outdated if-condition in
memory_failure_hugetlb().  This loses the chance to isolate the error in
the finer unit, so it's not optimal.  Drop the condition.

This "(p != head && TestSetPageHWPoison(head)" condition is based on the
old semantics of PageHWPoison on hugepage (where PG_hwpoison flag was
set on the subpage), so it's not necessray any more.  By getting to set
PG_hwpoison on head page for hugepages, concurrent error events on
different subpages in a single hugepage can be prevented by
TestSetPageHWPoison(head) at the beginning of memory_failure_hugetlb().
So dropping the condition should not reopen the race window originally
mentioned in commit b985194c8c0a ("hwpoison, hugetlb:
lock_page/unlock_page does not match for handling a free hugepage")

[naoya.horiguchi@linux.dev: fix "HardwareCorrupted" counter]
Link: https://lkml.kernel.org/r/20211220084851.GA1460264@u2004
Link: https://lkml.kernel.org/r/20211210110208.879740-1-naoya.horiguchi@linux.dev
Signed-off-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Reported-by: Fei Luo <luofei@unicloud.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: <stable@vger.kernel.org> [5.14+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memory-failure.c

index 07c875f..682828b 100644 (file)
@@ -1470,17 +1470,12 @@ static int memory_failure_hugetlb(unsigned long pfn, int flags)
        if (!(flags & MF_COUNT_INCREASED)) {
                res = get_hwpoison_page(p, flags);
                if (!res) {
-                       /*
-                        * Check "filter hit" and "race with other subpage."
-                        */
                        lock_page(head);
-                       if (PageHWPoison(head)) {
-                               if ((hwpoison_filter(p) && TestClearPageHWPoison(p))
-                                   || (p != head && TestSetPageHWPoison(head))) {
+                       if (hwpoison_filter(p)) {
+                               if (TestClearPageHWPoison(head))
                                        num_poisoned_pages_dec();
-                                       unlock_page(head);
-                                       return 0;
-                               }
+                               unlock_page(head);
+                               return 0;
                        }
                        unlock_page(head);
                        res = MF_FAILED;