out of swap on shutdown

[Patrick Welche <prlw1%cam.ac.uk@localhost>]

On shutdown, on machines with a large tmpfs which has been thrashed, I
see:

UVM: pid 1029.1 (master), uid 0 killed: out of swap
UVM: pid 701.1 (qmgr), uid 12 killed: out of swap

with what appears as a hang. In the past, I have left a box in this
state overnight with no change.

(I think about a year ago this would panic instead.) The difference is that
today I just happen to have a serial console plugged in. Box running
yesterday's -current/amd64. 

Sending breaks shows that things aren't actually stuck:

fatal breakpoint trap in supervisor mode
trap type 1 code 0 rip ffffffff8024fa2d cs 8 rflags 202 cr2 7f7ff6c3c11a ilevel 
8 rsp fffffe8043a05c50
curlwp 0xfffffe807fbb5020 pid 0.7 lowest kstack 0xfffffe8043a032c0
Stopped in pid 0.7 (system) at  netbsd:breakpoint+0x5:  leave
breakpoint() at netbsd:breakpoint+0x5
comintr() at netbsd:comintr+0x524
Xintr_ioapic_edge9() at netbsd:Xintr_ioapic_edge9+0xea
--- interrupt ---
bus_space_read_4() at netbsd:bus_space_read_4+0xa
mi_switch() at netbsd:mi_switch+0x1e
sleepq_block() at netbsd:sleepq_block+0xa2
cv_wait() at netbsd:cv_wait+0x9a
xc_thread() at netbsd:xc_thread+0x52

Selection of processes:
PID    LID S CPU     FLAGS       STRUCT LWP *               NAME WAIT
999      1 3   1         0   fffffe807d16bb20            swapctl plpg
0       51 3   0       200   fffffe8043ee09a0            raidio0 plpg
0       50 3   1       200   fffffe8043f215a0              raid0 rfnodeq
1        1 3   1        80   fffffe8043f21180               init wait
0    >  17 7   1       201   fffffe8043a478e0             idle/1
0    >   7 3   0       200   fffffe807fbb5020            xcall/0 xcall
0        1 3   1       200   ffffffff80ccb980            swapper uvm

--- interrupt ---
uvmpdpol_pagedeactivate() at netbsd:uvmpdpol_pagedeactivate+0x22
uvmpdpol_balancequeue() at netbsd:uvmpdpol_balancequeue+0x87
uvm_pageout() at netbsd:uvm_pageout+0x3f8

--- interrupt ---
_atomic_cas_64() at netbsd:_atomic_cas_64+0x8
uvmpd_trylockowner() at netbsd:uvmpd_trylockowner+0x1e
uvm_pageout() at netbsd:uvm_pageout+0x1b9

--- interrupt ---
sleepq_remove() at netbsd:sleepq_remove+0x84
sleepq_wake() at netbsd:sleepq_wake+0x43
uvm_pageout() at netbsd:uvm_pageout+0x43b

--- interrupt ---
uvm_pageactivate() at netbsd:uvm_pageactivate

--- interrupt ---
mutex_tryenter() at netbsd:mutex_tryenter+0x80
uvmpd_trylockowner() at netbsd:uvmpd_trylockowner+0x1e
uvmpdpol_selectvictim() at netbsd:uvmpdpol_selectvictim+0x43
uvm_pageout() at netbsd:uvm_pageout+0x19d


It looks as though swapctl wants memory?
Thoughts on how to make a more useful bug report?

Cheers,

Patrick