summaryrefslogtreecommitdiff
path: root/sys
diff options
context:
space:
mode:
authorPer Fogelstrom <pefo@cvs.openbsd.org>1998-08-30 13:42:15 +0000
committerPer Fogelstrom <pefo@cvs.openbsd.org>1998-08-30 13:42:15 +0000
commita2c225fd21b562734f25815645ee93c9f2958aa6 (patch)
tree867dd1f9c6669578e48ac448332749f560aeafc3 /sys
parent5ef58ba424569c0766f9b4fba4308288f1449790 (diff)
OK, this is another fix for the infamous i386 "PTDI panic". This fix is the
same as is used in 'vslock()'. The situation when this occurs is when a process tries to read or write to another process vm space and the _segment_ referenced have not yet been mapped. A certain way of crashing a system is to debug a program larger than 4 Mb (yes, those critters exist!). Set a break in an address above 4 Mb and pmap_enter panics because it can't fault in a new ptd page since it is not the process itself that calls pmap_enter. This is a fix, not a solution, but at least a user can't crash the kernel.
Diffstat (limited to 'sys')
-rw-r--r--sys/miscfs/procfs/procfs_mem.c8
1 files changed, 6 insertions, 2 deletions
diff --git a/sys/miscfs/procfs/procfs_mem.c b/sys/miscfs/procfs/procfs_mem.c
index 733c8e02753..8953fe1b665 100644
--- a/sys/miscfs/procfs/procfs_mem.c
+++ b/sys/miscfs/procfs/procfs_mem.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: procfs_mem.c,v 1.7 1997/09/11 05:26:14 millert Exp $ */
+/* $OpenBSD: procfs_mem.c,v 1.8 1998/08/30 13:42:14 pefo Exp $ */
/* $NetBSD: procfs_mem.c,v 1.8 1996/02/09 22:40:50 christos Exp $ */
/*
@@ -157,9 +157,13 @@ procfs_rwmem(p, uio)
*/
if (!error && writing && object->shadow) {
m = vm_page_lookup(object, off);
- if (m == 0 || (m->flags & PG_COPYONWRITE))
+ if (m == 0 || (m->flags & PG_COPYONWRITE)) {
+#ifdef __i386__
+ pmap_prefault(map, uva, 4);
+#endif
error = vm_fault(map, pageno,
VM_PROT_WRITE, FALSE);
+ }
}
/* Find space in kernel_map for the page we're interested in */