One of us is *seriously* confusing the issue here.

   I don't think that SIGTRAP is appropriate under these circumstances.
   There is no (valid) possibility that the user code could turn on the
   trace bit for system level code.

I never suggested that there was, or that it should generate SIGTRAP
for a trap from kernel code.

   Do you mean after setting a breakpoint at its start address?

Yes.  Currently that works, but at the cost of DDB being *SEVERELY*
broken, such that a user program can caused a trap into DDB.  If one
did the obvious thing and simply ignored trace traps near the beginning
of _Xsyscall (which is, given the current code, really the only way to
fix it), it would not be possible to step through _Xsyscall.

One way or another, DDB absolutely must be fixed.