Sean Davis <dive-nb@endersgame.net> writes:

> Should I send-pr this?

Go ahead and send-pr it; the problem has been known for a while and I
haven't yet fixed it, but it's good to have a record. The problem is
that somehow LWPs are sticking around on the alllwp list after the
process has exited. Eventually uvm_swapout_threads() iterates over
alllwp and examines l->l_proc->p_flag of every lwp, and at some point
l->l_proc of the dead lwp ceases to point to valid memory (pool
reclamation?) and it goes boom.

        - Nathan

> #10 0xc01b7e5a in uvmpd_scan () at ../../../../uvm/uvm_pdaemon.c:753
> #11 0xc01b7826 in uvm_pageout (arg=0xe113d294)
>     at ../../../../uvm/uvm_pdaemon.c:253
> (gdb) eros#