Re: kern/45936: amd64 resets during startup

[Marc Balmer <mbalmer%NetBSD.org@localhost>]

Am 07.02.12 11:55, schrieb Marc Balmer:
> The following reply was made to PR kern/45936; it has been noted by GNATS.
> 
> From: Marc Balmer <mbalmer%NetBSD.org@localhost>
> To: gnats-bugs%NetBSD.org@localhost
> Cc: 
> Subject: Re: kern/45936: amd64 resets during startup
> Date: Tue, 07 Feb 2012 11:50:27 +0100
> 
>  Am 06.02.12 14:25, schrieb mbalmer%NetBSD.org@localhost:
>  >> Number:         45936
>  >> Category:       kern
>  >> Synopsis:       amd64 resets during startup
>  >> Confidential:   no
>  >> Severity:       critical
>  >> Priority:       high
>  >> Responsible:    kern-bug-people
>  >> State:          open
>  >> Class:          sw-bug
>  >> Submitter-Id:   net
>  >> Arrival-Date:   Mon Feb 06 13:25:00 +0000 2012
>  >> Originator:     Marc Balmer
>  >> Release:        5.99.64
>  >> Organization:
>  > TNF
>  >> Environment:
>  > amd64
>  >> Description:
>  > Since the kvm changes, this machines resets during startup.  Kernel boots, 
> /etc/rc begins to run, and towards end the machine spontaneously resets.
>  >> How-To-Repeat:
>  > Starzup the machine.
>  >> Fix:
>  > Revert the lvm changes. (5.99.56 works fine)
>  
>  After doing a binary search and going back in time, testing kernels from
>  older sources, I can now confirm that not the kmem changes cause this,
>  but the direct map changes r1.40 of sys/arch/amd64/include/types.h:
>  
>  revision 1.40
>  date: 2011/12/04 16:24:13;  author: chs;  state: Exp;  lines: +8 -2
>  map all of physical memory using large pages.
>  ported from openbsd years ago by Murray Armfield,
>  updated for changes since then by me.
>  
>  The following diff, which is obviously *not* a solution, prevents the
>  machine from resetting:
>  
>  Index: types.h
>  ===================================================================
>  RCS file: /cvsroot/src/sys/arch/amd64/include/types.h,v
>  retrieving revision 1.41
>  diff -u -r1.41 types.h
>  --- types.h     21 Jan 2012 16:48:56 -0000      1.41
>  +++ types.h     7 Feb 2012 10:49:50 -0000
>  @@ -96,12 +96,14 @@
>  
>   #include "opt_xen.h"
>   #if defined(__x86_64__) && !defined(XEN)
>  +#if 0
>   #define        __HAVE_DIRECT_MAP 1
>   #define        __HAVE_MM_MD_DIRECT_MAPPED_IO
>   #define        __HAVE_MM_MD_DIRECT_MAPPED_PHYS
>   #define        __HAVE_CPU_UAREA_ROUTINES
>   #endif
>   #endif
>  +#endif
>  
>   #else  /*      !__x86_64__     */
>  
>  

This PR is probably a duplicate of PR 45892, or at least it's the same
symptoms.