Re: reproducible crash with amd64/SMP (with file system trouble)

["Steven M. Bellovin" <smb%cs.columbia.edu@localhost>]

Quoting "Steven M. Bellovin" <smb%cs.columbia.edu@localhost>:

> Running 5.99.6 from 10 January, my system crashes reproducibly when
> building apr from pkgsrc.  It has happened three different times.
> Twice, fsck gave up because of duplicate i-nodes, so I had to run it
> manually.  One of those times, fsck left things in an incorrect state:
> a directory with no files in it had a link count of 3.  I had to reboot
> into single user and rerun fsck to clear that one up.
>
> When I boot with -1, I can build the package with no trouble.
>
> The machine in question is a Thinkpad T61 with 3GB of RAM; per the
> subject line, it's running amd64.  I've been holding off running
> anything newer while Christos' time changes settle down, but if there
> have been other changes in the last week or so that may have fixed it I
> think I'll have to take my chances...
>
> Please cc me directly on any responses; I've had to rejigger my mail
> setup while my laptop is not fully functional, and I'm not sure I'll
> see this mailing list properly.
>
> --
>                 --Steve Bellovin, http://www.cs.columbia.edu/~smb
I should have added -- here's the crash traceback.  (Hand-copied,  
since I have no serial console on the box.)

uvm_fault(0xffff800004c1d0300, 0x0, 1) -> e
fatal page fault in supervisor mode
trap type 6 code 0 rip ffffffff8046750c cs 8 rflags 10286 cr2  10 cpl  
0 rsp ffff
800004c245880
kernel: page fault trap, code = 0
stopped in pid 29757.1 (expr) at         netbsd:hook_proc_run+0x1c:     movq

(%rbx),%rbx
db{1}> bt
book_proc_run() at netbsd:hook_proc_run+0x1c
execve1() at netbsd:execve1+0x86e
syscall() at netbsd:syscall+0xb6



--Steve Bellovin, http://www.cs.columbia.edu/~smb