port-mips/53078: bogus mutex asserts in a MULTIPROCESSOR kernel

[coypu%sdf.org@localhost]

>Number:         53078
>Category:       port-mips
>Synopsis:       bogus mutex asserts in a MULTIPROCESSOR kernel
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    port-mips-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Wed Mar 07 23:30:00 +0000 2018
>Originator:     coypu
>Release:        NetBSD 8.99.12
>Organization:
>Environment:
NetBSD  8.99.12 NetBSD 8.99.12 (ERLITE) #32: Thu Mar  8 01:07:10 IST 2018  fly@planets:/home/fly/mips64eb-gcc6-current/sys/arch/evbmips/compile/ERLITE evbmips

>Description:
if I add
  options MULTIPROCESSOR
  options LOCKDEBUG # Yes, this one is necessary too, and it's also a bug.

to the ERLITE kernel, I can boot it and reach userland.

but running atf tests, I get to things like:


    g_variable_syntax: [69.040000s] Passed.
    h_var_assign: [1.430000s] Passed.
    i_pipelines: Mutex error: mutex_vector_exit,742: not held by current CPU

lock address : 0x980000041ff1d600 type     :               spin
initialized  : 0xffffffff80242d08
shared holds :                  0 exclusive:                  1
shares wanted:                  0 exclusive:                  0
current cpu  :                  0 last held:                  1
current lwp  : 0x980000041b86baa0 last held: 0x980000041b86baa0
last locked* : 0xffffffff80242550 unlocked : 0xffffffff802436fc
owner field  : 000000000000000000 wait/spin:                0/1

xanic: Skekirnppeiln dg icargnasohs tiducm pas osne rtreiocun rs"civie- >cpain_imtc
 _cpoanuinct:  ==LO C-1KD" EBfUaGil: edM:u tefxi leer r"o/cr:u rrmuetntex/s_vrec/ctsyors/_ekexirnt,/7ke42r:n _nsyontc hh.elc"d , byli ncue rr67wi6 t miCP_sU
  tccpuh:0: c Bpeug0:in  cti_ramctex_baccouk.nt.. (
-2pi)d ! 2= 78-157 (48bl48o ckno wti tfoh unspdi
n-cmpuu0te: x Ehned ldtr)a
inFabauclkt.e.d .
k r mebidoo-ttirnagc.e.b.ac
; 
  aborting...

dump to dev 8,1 not possible
rebooting...

>How-To-Repeat:

>Fix:
Probably re-do the llsc mutex code.