Kernel trap on linux compat load

[John Baldwin]

On Friday 17 June 2005 05:56 pm, Eric Millbrandt wrote:
> John Baldwin wrote:
> >On Thursday 16 June 2005 08:21 pm, Eric Millbrandt wrote:
> >>John Baldwin wrote:
> >>>On Friday 10 June 2005 07:02 pm, Eric Millbrandt wrote:
> >>>>I've noticed that linux compatibility caused my system to crash since I
> >>>>upgraded from RELENG_4_9 (RELENG_4_10, RELENG_4_11, RELENG_4).  I
> >>>>upgraded using buildworld, I haven't had a change to test using a clean
> >>>>binary install.  Has anyone else seen this behavior?  It's easy to
> >>>>reproduce, the system traps right on linux compatibility load.  The
> >>>> only similar behavior I've found on google is an unanswered post
> >>>>http://lists.freebsd.org/pipermail/freebsd-alpha/2004-June/001555.html
> >>>>
> >>>>Jun  8 07:18:17 mongoloid /kernel: fatal kernel trap:
> >>>>Jun  8 07:18:17 mongoloid /kernel:
> >>>>Jun  8 07:18:17 mongoloid /kernel: trap entry = 0x4 (unaligned access
> >>>>fault) Jun  8 07:18:17 mongoloid /kernel: a0         =
> >>>> 0xfffffe0015285e34 Jun  8 07:18:17 mongoloid /kernel: a1         =
> >>>> 0x2d
> >>>>Jun  8 07:18:17 mongoloid /kernel: a2         = 0x1
> >>>>Jun  8 07:18:17 mongoloid /kernel: pc         = 0xfffffc0000386e68
> >>>>Jun  8 07:18:17 mongoloid /kernel: ra         = 0xfffffc0000386dd8
> >>>>Jun  8 07:18:17 mongoloid /kernel: curproc    = 0xfffffe001177e780
> >>>>Jun  8 07:18:17 mongoloid /kernel: pid = 173, comm = ldconfig
> >>>>Jun  8 07:18:17 mongoloid /kernel:
> >>>>Jun  8 07:18:17 mongoloid /kernel: panic: trap
> >>>>
> >>>>EEM
> >>>
> >>>Did you rebuild your kernel with buildkernel / installkernel?  I believe
> >>>that buildworld on 4.x still builds and installs new modules into
> >>>/modules including the linux and osf1 compat modules.  New modules
> >>> aren't guaranteed to work on older kernels, so if your kernel is still
> >>> 4.9, that is probably your problem.
> >>
> >>I used buildworld, buildkernel, installkernel, drop to single user,
> >>installworld, and mergemaster.  This is with linux_compat loading from
> >>rc.conf.  I misspoke earlier loading linux.ko works fine but running a
> >>linux binary, /compat/linux/sbin/ldconfig or sophie, causes the trap.
> >>On a side not osf1_compat works fine.
> >>
> >>Here are my stack traces after running /compat/linux/sbin/ldconfig...
> >>Debugger() at Debugger=0x2c
> >>panic() at panic+0x100
> >>trap() at trap+0x600
> >>XentUna() at XentUna+0x2c
> >>--- unaligned access fault (from ipl0) ---
> >>kernel_sysctl at kernel_sysctl+ox1a8
> >>347() at -0x1fffe4a63d0
> >>Here ddb traps on itself (memory management fault)
> >>
> >>This one is from running sophie
> >>Debugger() at Debugger=0x2c
> >>panic() at panic+0x100
> >>trap() at trap+0x600
> >>XentUna() at XentUna+0x2c
> >>--- unaligned access fault (from ipl0) ---
> >>kernel_sysctl at kernel_sysctl+ox1a8
> >>linux_newuname() at linux_newuname+0xd0
> >>syscall() at syscal+0x224
> >>XentSys at XentSys+0x5c
> >>--- syscall (339, Linux ELF, linux_newuname) ---
> >>--- user mode ---
> >
> >Ok, this is helpful.  Can you do 'gdb kernel.debug' and then type 'list
> >*kernel_sysctl+0x1a8'?  This will give us the source file/line that it
> >faulted at.  Thanks!
>
> Ok here is what I found.
>
> (gdb) list *kernel_sysctl+0x1a8
> 0xfffffc000038e9a8 is in kernel_sysctl
> (/usr/src/sys/kern/kern_sysctl.c:938).
> 933                     if (req.oldptr && req.oldidx > req.oldlen)
> 934                             *retval = req.oldlen;
> 935                     else
> 936                             *retval = req.oldidx;
> 937             }
> 938             return (error);
> 939     }
> 940
> 941     int
> 942     kernel_sysctlbyname(struct proc *p, char *name, void *old,
> size_t *oldlenp,

Ok, try this patch please:

Index: compat/linux/linux_misc.c
===================================================================
RCS file: /usr/cvs/src/sys/compat/linux/linux_misc.c,v
retrieving revision 1.85.2.11
diff -u -r1.85.2.11 linux_misc.c
--- compat/linux/linux_misc.c   23 Mar 2004 12:16:48 -0000      1.85.2.11
+++ compat/linux/linux_misc.c   20 Jun 2005 18:33:07 -0000
@@ -687,7 +687,8 @@
        struct l_new_utsname utsname;
        char *osrelease, *osname;
        int name[2];
-       int error, plen, olen;
+       int error;
+       size_t plen, olen;

 #ifdef DEBUG
        if (ldebug(newuname))


-- 
John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org