WITNESS crashes on boot

Some time ago, a diff circulated either here or on tech@
to test lock order reversals via witness(4). The diff is
already in cvs, but I kept option WITNESS in all MP kernels.

That has mostly resulted in something like

witness: acquiring duplicate lock of same type: "&uobj->vmobjlock"
1st uobjlk
2nd uobjlk
Starting stack trace...
witness_checkorder(fffffd81df11c8a0,9,0) at witness_checkorder+0x8ac
rw_enter(fffffd81df11c890,1) at rw_enter+0x68
uvm_obj_wire(fffffd821ce5bab0,0,40000,ffff80003378f738) at uvm_obj_wire+0x46
shmem_get_pages(ffff800001075280) at shmem_get_pages+0xb8
__i915_gem_object_get_pages(ffff800001075280) at __i915_gem_object_get_pages+0x6d
i915_gem_fault(ffff800001075280,ffff80003378fad0,10009b000,66431f53000,ffff80003378fa50,1,7dc736254d30337a,ffff80000011c078,ffff80000011c178) at i915_gem_fault+0x203
drm_fault(ffff80003378fad0,66431f53000,ffff80003378fa50,1,0,0,4687aa2e86d72b6e,ffff80003378fad0) at drm_fault+0x156
uvm_fault(fffffd81e1173180,66431f53000,0,2) at uvm_fault+0x179
upageflttrap(ffff80003378fc30,66431f53000) at upageflttrap+0x62
usertrap(ffff80003378fc30) at usertrap+0x129
recall_trap() at recall_trap+0x8
end of kernel
end trace frame: 0x7f7ffffddf40, count: 246
End of stack trace.

after the boot sequence (always a graphics driver, such as i915 here)
and besides that, everything was running fine - until recently: now
it's crashing in witness_checkorder(); please see the attached pictures
(sorry, this mahcine has no cereal).

This happens e.g. on all amd64 Thinkpads I run.
Have the previous warnings been upgraded to errors?
Wihout WITNESS, everything runs fine.


Jan