kern/59518: edgeroute 4 asserts every few months

[mrg%eterna23.net@localhost]

>Number:         59518
>Category:       kern
>Synopsis:       edgeroute 4 asserts every few months
>Confidential:   no
>Severity:       serious
>Priority:       high
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Tue Jul 08 06:15:00 +0000 2025
>Originator:     matthew green
>Release:        -current (late april) but i've seen this a few times for a couple of years
>Organization:
>Environment:
>Description:
my er4 was idling after being up for a bit over 2 months, and then crashed
with this assert:

[ 6074695.5771516] panic: kernel diagnostic assertion "pte == xpte" failed: file "/usr/src/sys/arch/mips/mips/pmap_machdep.c", line 880 pmap=0xffffffff807a8d08 va=0xc000000013980000 asid=0: TLB pte (0xc1b99b) != real pte (0xc1b99f/0xc1b99f) @ 0x9800000000127300
[ 6074695.5771516] cpu0: Begin traceback...
[ 6074695.5771516] pid 1833088 not found
[ 6074695.5771516] cpu0: End traceback...
[ 6074695.5771516] kernel: breakpoint trap
Stopped in pid 0.6 (system) at  netbsd:cpu_Debugger+0x4:        jr      ra
                bdslot: nop
0x98000000001bf8a0: cpu_Debugger+0x4 (0x2,0,0xffffffff80f8cc20,0xffffffff807a83c0) ra 0xffffffff80519e50 sz 0
0x98000000001bf8a0: vpanic+0x1a8 (0x2,0,0xffffffff80f8cc20,0xffffffff807a83c0) ra 0xffffffff8065e9a8 sz 64
0x98000000001bf8e0: kern_assert+0x48 (0x2,0xffffffff8069d838,0xffffffff806aa490,0xffffffff806aa030) ra 0xffffffff8025e430 sz 96
0x98000000001bf940: pmap_md_tlb_check_entry+0xd0 (0x2,0xffffffff8069d838,0xffffffff806aa490,0xffffffff806aa030) ra 0 sz 0
User-level: pid 0.6


i had a look at the PTE bits being complained about: 0xc1b99b vs 0xc1b99f
which is the mips3 dirty bit:

mips3_pte.h:95:#define      MIPS3_PG_D      0x00000004      /* Dirty */

this check is only run for UP kernels, i assume because another CPU could
make the PTE dirty bit be set, but it shouldn't be set differently in the
UP case, because it shouldn't happen.  i don't know mips here enough.

>How-To-Repeat:
allow edgerouter 4 to idle for months.

my kernel does not have LOCKDEBUG (not for the UP case), but it does have
DEBUG and DIAGNOSTIC enabled.
>Fix: