s390: fix gmap_ipte_notifier vs. software dirty pages
authorChristian Borntraeger <borntraeger@de.ibm.com>
Fri, 17 May 2013 12:41:38 +0000 (14:41 +0200)
committerGleb Natapov <gleb@redhat.com>
Tue, 21 May 2013 08:55:38 +0000 (11:55 +0300)
On heavy paging load some guest cpus started to loop in gmap_ipte_notify.
This was visible as stalled cpus inside the guest. The gmap_ipte_notifier
tries to map a user page and then made sure that the pte is valid and
writable. Turns out that with the software change bit tracking the pte
can become read-only (and only software writable) if the page is clean.
Since we loop in this code, the page would stay clean and, therefore,
be never writable again.
Let us just use fixup_user_fault, that guarantees to call handle_mm_fault.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
arch/s390/mm/pgtable.c

index 5ca75683c654bf93b60d11d78a063a1e3329ad66..1e0c438dbd6519f06502a94f8ceb74c621263ce5 100644 (file)
@@ -677,8 +677,7 @@ int gmap_ipte_notify(struct gmap *gmap, unsigned long start, unsigned long len)
                        break;
                }
                /* Get the page mapped */
-               if (get_user_pages(current, gmap->mm, addr, 1, 1, 0,
-                                  NULL, NULL) != 1) {
+               if (fixup_user_fault(current, gmap->mm, addr, FAULT_FLAG_WRITE)) {
                        rc = -EFAULT;
                        break;
                }