From the console log: fault virtual address = 0x20^M fault code = supervisor read data, page not present^M ... [ thread pid 15852 tid 101888 ]^M Stopped at turnstile_broadcast+0x9c: movq 0x20(%rbx,%rax,1),%rcx^M db> where^M Tracing pid 15852 tid 101888 td 0xfffff801418fc000^M turnstile_broadcast() at turnstile_broadcast+0x9c/frame 0xfffffe1835ad14a0^M __mtx_unlock_sleep() at __mtx_unlock_sleep+0x60/frame 0xfffffe1835ad14d0^M cache_zap_locked_vnode_kl2() at cache_zap_locked_vnode_kl2+0x387/frame 0xfffffe1835ad1530^M cache_purge() at cache_purge+0x176/frame 0xfffffe1835ad15a0^M tmpfs_rename() at tmpfs_rename+0xeff/frame 0xfffffe1835ad1690^M VOP_RENAME_APV() at VOP_RENAME_APV+0x97/frame 0xfffffe1835ad16c0^M kern_renameat() at kern_renameat+0x4a8/frame 0xfffffe1835ad1920^M amd64_syscall() at amd64_syscall+0x51e/frame 0xfffffe1835ad1ab0^M Xfast_syscall() at Xfast_syscall+0xfb/frame 0xfffffe1835ad1ab0^M --- syscall (501, FreeBSD ELF64, sys_renameat), rip = 0x6fc42a, rsp = 0x7fffffffbd98, rbp = 0x7fffffffc740 ---^M ... Locked vnodes^M vnode 0xfffff80b9f074938: tag tmpfs, type VREG^M usecount 1, writecount 0, refcount 1 mountedhere 0^M flags (VI_ACTIVE)^M v_object 0xfffff812c9dc0a50 ref 1 pages 10 cleanbuf 0 dirtybuf 0^M lock type tmpfs: EXCL by thread 0xfffff801323ba500 (pid 94836, pkg-static, tid 101212)^M tag VT_TMPFS, tmpfs_node 0xfffff8152c481570, flags 0x0, links 1^M mode 0600, owner 0, group 0, size 40960, status 0x2^M ^M vnode 0xfffff80bc52d83b0: tag tmpfs, type VDIR^M usecount 4, writecount 0, refcount 6 mountedhere 0^M flags (VI_ACTIVE)^M lock type tmpfs: EXCL by thread 0xfffff801418fc000 (pid 15852, pkg-static, tid 101888)^M tag VT_TMPFS, tmpfs_node 0xfffff814ef18e658, flags 0x0, links 2^M mode 0755, owner 0, group 0, size 2048, status 0x0^M ^M I do have an alltrace but do not have a savecore.
Kernel revision: 12.0-CURRENT #0 r308208: Wed Nov 2 09:55:11 UTC 2016 Load is a ports exp run.
(There have been several changes in namecache since this bug was filed, including a bugfix: r308665. The stack didn't look completely identical to Don Lewis', though, so I am leaving it open.)
This most likely got fixed, nothing reproduces the problem to the best of my knowledge and there are no more reports.