[src/trunk]: src/sys/uvm If a pager fault method returns ENOMEM but some memo...

[chs <chs%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/f71dba0130a4
branches:  trunk
changeset: 450658:f71dba0130a4
user:      chs <chs%NetBSD.org@localhost>
date:      Sun Apr 21 15:27:59 2019 +0000

description:
If a pager fault method returns ENOMEM but some memory appears to be reclaimable,
wake up the pagedaemon and retry the fault.  This fixes the problems with Xorg
being killed with an "out of swap" message due to a transient memory shortage.

diffstat:

 sys/uvm/uvm_fault.c |  14 ++++++++++++--
 1 files changed, 12 insertions(+), 2 deletions(-)

diffs (35 lines):

diff -r 4db0ae8058c6 -r f71dba0130a4 sys/uvm/uvm_fault.c
--- a/sys/uvm/uvm_fault.c       Sun Apr 21 14:14:14 2019 +0000
+++ b/sys/uvm/uvm_fault.c       Sun Apr 21 15:27:59 2019 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: uvm_fault.c,v 1.204 2018/05/08 19:33:57 christos Exp $ */
+/*     $NetBSD: uvm_fault.c,v 1.205 2019/04/21 15:27:59 chs Exp $      */
 
 /*
  * Copyright (c) 1997 Charles D. Cranor and Washington University.
@@ -32,7 +32,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: uvm_fault.c,v 1.204 2018/05/08 19:33:57 christos Exp $");
+__KERNEL_RCSID(0, "$NetBSD: uvm_fault.c,v 1.205 2019/04/21 15:27:59 chs Exp $");
 
 #include "opt_uvmhist.h"
 
@@ -901,6 +901,16 @@
                                 * object fault routine responsible for
                                 * pmap_update().
                                 */
+
+                               /*
+                                * Wake up the pagedaemon if the fault method
+                                * failed for lack of memory but some can be
+                                * reclaimed.
+                                */
+                               if (error == ENOMEM && uvm_reclaimable()) {
+                                       uvm_wait("pgo_fault");
+                                       error = ERESTART;
+                               }
                        } else {
                                error = uvm_fault_lower(&ufi, &flt, pages);
                        }