Subject: kern/18958: uvm_loan users use kernel_object-backed VM ranges
To: None <gnats-bugs@gnats.netbsd.org>
From: None <thorpej@shagadelic.org>
List: netbsd-bugs
Date: 11/06/2002 22:09:17
>Number: 18958
>Category: kern
>Synopsis: uvm_loan users use kernel_object-backed VM ranges
>Confidential: no
>Severity: non-critical
>Priority: low
>Responsible: kern-bug-people
>State: open
>Class: change-request
>Submitter-Id: net
>Arrival-Date: Wed Nov 06 22:10:03 PST 2002
>Closed-Date:
>Last-Modified:
>Originator: Jason R Thorpe
>Release: NetBSD 1.6I
>Organization:
Wasabi Systems, Inc.
>Environment:
System: NetBSD yeah-baby.shagadelic.org 1.6I NetBSD 1.6I (YEAH-BABY-XP) #11: Sun Sep 29 17:42:25 PDT 2002 thorpej@yeah-baby.shagadelic.org:/u1/netbsd/src/sys/arch/i386/compile/YEAH-BABY-XP i386
Architecture: i386
Machine: i386
>Description:
Users of uvm_loan() tend to use uvm_km_valloc() or
uvm_km_valloc_wait() to allocate the KVA at which
loaned pages will be mapped.
This is not a problem with basic functionality, but can
cause interesting behavior in the face of other bugs, since
a write fault will cause UVM to COW'd (overwriting the k-mapping
with a non-k-mapping), which will cause a panic later, at some
unrelated time.
This was diagnosed by Chuq Silvers.
>How-To-Repeat:
See the audit trail of kern/18809.
>Fix:
Provide a uvm_km_*alloc() routine which creates a backed-
by-nothing mapping?
>Release-Note:
>Audit-Trail:
>Unformatted: