re: kern/59518: edgeroute 4 asserts every few months

[matthew green <mrg%eterna23.net@localhost>]

i noticed there is some info about the stack i missed before:

db> ps          
PID     LID S CPU     FLAGS       STRUCT LWP *               NAME WAIT
11131>11131 7   0     40000   980000004d534e00          makemandb
...
0    >    6 7   0       200   980000004eed7280          softser/0

which gave me two threads to inspect:

db> bt/a 980000004eed7280
pid 0.6 at 0x98000000001bc000
0x98000000001bfdc0: kernel_text+0xe4 (0,0,0,0) ra 0xffffffff804df90c sz 0
0x98000000001bfdc0: mi_switch+0x1e4 (0,0,0,0) ra 0xffffffff804dcf44 sz 96
0x98000000001bfe20: softint_dispatch+0xec (0,0,0,0) ra 0xffffffff80200234 sz 128
0x98000000001bfea0: softint_fast_dispatch+0x7c (0,0,0,0) ra 0 sz 32
User-level: pid 0.6
db> bt/a 980000004d534e00
pid 11131.11131 at 0x980000000289c000
0x980000000289f4c0: kernel_text+0xe4 (0,0,0,0) ra 0xffffffff8020025c sz 0
0x980000000289f4c0: softint_fast_dispatch+0xa4 (0x980000004eed7280,0,0,0) ra 0xffffffff8026f794 sz 32
0x980000000289f4e0: softint_process+0xfc (0x980000004eed7280,0,0,0) ra 0xffffffff802578ec sz 64
0x980000000289f520: mips64r2_kern_intr+0xec (0x8f00,0x4,0x10,0xffffffff807a83c0) ra 0xffffffff80258894 sz 352
0x980000000289f680: ras_mutex_spin_enter+0x18 (0x8f00,0x4,0x10,0xffffffff807a83c0) ra 0 sz 0
User-level: pid 0.6

so it seems that the makemandb process got interrupted by a fast
softint for serial, and i guess that something in this called
pmap_update(), which called pmap_md_tlb_check_entry() that asserts.


i wonder if this is doing something to set dirty leading to this
as a toctou bug?


.mrg.