In message <20060113020701.GA5477@panix.com>Thor Lancelot Simon writes
>I have an amd64 system with 8GB of RAM.  I'm trying to run a memory tester,
>and find that even with the datasize, memoryuse, and memorylocked hard and
>soft limits set to unlimited, I can't mlock even 4GB.
>

Below is an ugly 30-second kludge I used yesterday to make progress on
another issue, for which this retardd mlock() limit was a roadblock.
Setting raw pageframe counts with no bounds-check is clearly too
fragile for normal use (let alone commit).  One should instead
allocate a new parameter number in uvm_param.h, add a switch-entry for
that new parameter to update_vm_updateminmax() and route any
userland-initiated sysctl changes via sysctl_vm_updateminmax() for
vetting.

Left as an exercise for the reader... 

Index: uvm_meter.c
===================================================================
RCS file: /cvsroot/src/sys/uvm/uvm_meter.c,v
retrieving revision 1.33
diff -w -u -r1.33 uvm_meter.c
--- uvm_meter.c	10 Oct 2004 09:57:31 -0000	1.33
+++ uvm_meter.c
@@ -411,6 +411,23 @@
 		       SYSCTL_DESCR("Whether try to zero pages in idle loop"),
 		       NULL, 0, &vm_page_zero_enable, 0,
 		       CTL_VM, CTL_CREATE, CTL_EOL);
+
+	/*XXX make wiredmax a percentage via sysctl_vm_updateminmax? */
+ 	sysctl_createv(clog, 0, NULL, NULL,
+		       CTLFLAG_PERMANENT|CTLFLAG_READONLY,
+		       CTLTYPE_INT, "npages",
+		       SYSCTL_DESCR(" count of physical "
+				    "pages managed via uvm"),
+		       NULL, 0, &uvmexp.npages, 0,
+		       CTL_VM, CTL_CREATE, CTL_EOL);
+
+ 	sysctl_createv(clog, 0, NULL, NULL,
+		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
+		       CTLTYPE_INT, "wiredmax",
+		       SYSCTL_DESCR("maximum count of physical "
+				    "pages lockable via mlock() and friends"),
+		       NULL, 0, &uvmexp.wiredmax, 0,
+		       CTL_VM, CTL_CREATE, CTL_EOL);
 }
 
 /*