On Sat, 05 Oct 2013, 10:57 +0300, Konstantin Belousov wrote: > Since you have a will to play with this, please try to bisect the > stable to see which commit introduced the regression (you said that 9.1 > does not panic).
stable/9@r244654 is where the panic is introduced. So I am guessing that nullfs is an unanticipated necessary ingredient? I run jails on both systems and use nullfs to mount the host system's /usr /sbin /bin /lib /libexec read-only in the jails. The jails are managed with jail.conf(5) and the nullfs mounts are created and deleted with the jails via a per-jail mount.fstab directive in jail.conf. The panics I see (when ntpd or watchdogd exit during shutdown) happen after the jails have been removed (and their nullfs mounts deleted) FreeBSD 9.1-PRERELEASE #0 r244654: Mon Oct 7 20:48:44 AEST 2013 ns4: removed Stopping ntpd. Waiting for PIDS: 1900 panic: vm_page_unwire: page 0xfffffe0236f7eb60's wire count is zero cpuid = 5 KDB: stack backtrace: #0 0xffffffff804834a8 at kdb_backtrace+0x68 #1 0xffffffff8044a7fa at panic+0x21a #2 0xffffffff80679382 at vm_page_unwire+0x102 #3 0xffffffff806662c2 at vm_fault_unwire+0xd2 #4 0xffffffff8066e421 at vm_map_delete+0x171 #5 0xffffffff8066e69f at vm_map_remove+0x5f #6 0xffffffff80671969 at vmspace_exit+0xc9 #7 0xffffffff80413c7d at exit1+0x71d #8 0xffffffff80414c7e at sys_sys_exit+0xe #9 0xffffffff8069d0df at amd64_syscall+0x3bf #10 0xffffffff80687d67 at Xfast_syscall+0xf7 Additional information is in core.txt.31 in the same location indicated in my OP. -- John Marshall
pgpctZUOjHkc1.pgp
Description: PGP signature
