HWPOISON, hugetlb: isolate corrupted hugepage

If error hugepage is not in-use, we can fully recovery from error
by dequeuing it from freelist, so return RECOVERY.
Otherwise whether or not we can recovery depends on user processes,
so return DELAYED.

Dependency:
  "HWPOISON, hugetlb: enable error handling path for hugepage"

Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Andi Kleen <ak@linux.intel.com>

3 files changed, 38 insertions, 8 deletions

diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index e688fd89354..f479700df61 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -43,6 +43,7 @@ int hugetlb_reserve_pages(struct inode *inode, long from, long to,
 						struct vm_area_struct *vma,
 						int acctflags);
 void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed);
+void __isolate_hwpoisoned_huge_page(struct page *page);
 
 extern unsigned long hugepages_treat_as_movable;
 extern const unsigned long hugetlb_zero, hugetlb_infinity;
@@ -100,6 +101,7 @@ static inline void hugetlb_report_meminfo(struct seq_file *m)
 #define hugetlb_free_pgd_range(tlb, addr, end, floor, ceiling) ({BUG(); 0; })
 #define hugetlb_fault(mm, vma, addr, flags)	({ BUG(); 0; })
 #define huge_pte_offset(mm, address)	0
+#define __isolate_hwpoisoned_huge_page(page)	0
 
 #define hugetlb_change_protection(vma, address, end, newprot)
 
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index aa3c5173937..8c163f64cf1 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -2825,3 +2825,19 @@ void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed)
 	hugetlb_put_quota(inode->i_mapping, (chg - freed));
 	hugetlb_acct_memory(h, -(chg - freed));
 }
+
+/*
+ * This function is called from memory failure code.
+ * Assume the caller holds page lock of the head page.
+ */
+void __isolate_hwpoisoned_huge_page(struct page *hpage)
+{
+	struct hstate *h = page_hstate(hpage);
+	int nid = page_to_nid(hpage);
+
+	spin_lock(&hugetlb_lock);
+	list_del(&hpage->lru);
+	h->free_huge_pages--;
+	h->free_huge_pages_node[nid]--;
+	spin_unlock(&hugetlb_lock);
+}
diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 473f15a3356..d0b420aba72 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -690,17 +690,29 @@ static int me_swapcache_clean(struct page *p, unsigned long pfn)
 /*
  * Huge pages. Needs work.
  * Issues:
- * No rmap support so we cannot find the original mapper. In theory could walk
- * all MMs and look for the mappings, but that would be non atomic and racy.
- * Need rmap for hugepages for this. Alternatively we could employ a heuristic,
- * like just walking the current process and hoping it has it mapped (that
- * should be usually true for the common "shared database cache" case)
- * Should handle free huge pages and dequeue them too, but this needs to
- * handle huge page accounting correctly.
+ * - Error on hugepage is contained in hugepage unit (not in raw page unit.)
+ *   To narrow down kill region to one page, we need to break up pmd.
+ * - To support soft-offlining for hugepage, we need to support hugepage
+ *   migration.
  */
 static int me_huge_page(struct page *p, unsigned long pfn)
 {
-	return FAILED;
+	struct page *hpage = compound_head(p);
+	/*
+	 * We can safely recover from error on free or reserved (i.e.
+	 * not in-use) hugepage by dequeuing it from freelist.
+	 * To check whether a hugepage is in-use or not, we can't use
+	 * page->lru because it can be used in other hugepage operations,
+	 * such as __unmap_hugepage_range() and gather_surplus_pages().
+	 * So instead we use page_mapping() and PageAnon().
+	 * We assume that this function is called with page lock held,
+	 * so there is no race between isolation and mapping/unmapping.
+	 */
+	if (!(page_mapping(hpage) || PageAnon(hpage))) {
+		__isolate_hwpoisoned_huge_page(hpage);
+		return RECOVERED;
+	}
+	return DELAYED;
 }
 
 /*