port-sparc64/58936: Building lang/swi-prolog-lite on NetBSD 10.1/sparc64 crashes the machine

[als%thangorodrim.ch@localhost]

>Number:         58936
>Category:       port-sparc64
>Synopsis:       Building lang/swi-prolog-lite on NetBSD 10.1/sparc64 crashes the machine
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    port-sparc64-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Thu Dec 26 16:25:00 +0000 2024
>Originator:     Alexander Schreiber
>Release:        NetBSD 10.1
>Organization:
not much
>Environment:
Sun Ultra-45 running NetBSD 10.1 sparc64 - unfortunately, the mainboard died a day after this test.
>Description:
Trying to build lang/swi-prolog-lite from pkgsrc on NetBSD/sparc64 crashes the system on some machine types. This was first reported by John Klos in https://mail-index.netbsd.org/port-sparc64/2024/12/18/msg003292.html and replicated by me on a Sun Ultra-45 (single CPU system).

The last output on the ssh session that initiated "make package" was:

 [ 81%] Generating lib/clpfdlib.tex

and then the connection eventually dropped.

Console output on the graphic console was:

login: [  1524.6727578] panic: fault type 104 for invalid va 9097b2684d812b0
[ 1524.6727578] cpu0: Begin traceback...
[ 1524.6727578] cpu0: End traceback...
Stopped in pid 9003.9003 (swipl) at     netbsd:cpu_Debugger+0x4:        nop
db{0}> bt
panic(1996a28, 68, 9097b2684d812b0, 0 , 59, 59) at netbsd:panic:+0x20
data_access_fault(3727172e0, 68, 1018178, 9097b2684d812b0, 9097b2684d80396, 118018) at netbsd:data_access_fault+0x508
?(9097b2684d80396, 10cafa800, 4fe, 9097b278877ab96, 10184c, 0) at 1010700
copyin_proc(0, 9097b2684d80396, 10cafa800, 500, e0048000,0) at netbsd:copyin_proc+0x24
proc_getauxv(10bc2a440, 372717678, 372717658, 10cafa800, 500, 9097b2684d80396) at netbsd:proc_getauxv+0x74
real_coredump_elf64(10bc34140, 372717910, 1c60a48, a0, 1, 10bc2a440) at netbsd:real_coredump_elf64+0x19c
coredump_elf64(10bc34140, 372717910, 106e8ee80, 0, 4e, 1c70f00) at netbsd:coredump_elf64+0x2c
coredump(10bc34140, 0, 1ccbc00, 0, 109068800, 10bc2a440) at netbsd:coredump+0x4ac
sigexit(10bc34140, b, 180000, 10bc2a440, 106e70100, 1c71080) at netbsd:sigexit+0x27c
postsig(b, 0, 10bc344b8, 372717b70, 10bc2a440, 10bc34140) at netbsd:postsig+0x198
lwp_userret(10bc34140, 1000000, 20000, 100000, 91a002, 10bc2a440) at netbsd:lwp_userret+0x1ac
data_access_fault(372717ed0, 30, 40566bd0, fffffffffffff2b0, ffffffffffffe000, 10bc2a440) at netbsd:data_access_fault+0x470
?(ffffffffffffa0c8, 40787b08, ffffffffffffa0c8, 40788110, 5310d, 0) at 1010700
db{0}>

There was an swipl.core file, but it was empty (ulimit -c was unlimited)

This was also tried on two Sun V100 machines, were it merely resulted in the swipl binary exiting with sig11 at the same place, also leaving behind an empty core file.

So I suspect this is CPU model specific, because:
 - Sun V100: UltraSPARC-IIe
 - Sun Ultra-45: UltraSPARC IIIi
 - Sun Fire v450: UltraSPARC IIIi

I'm currently unable to help with any tests, as my Ultra-45 mainboard died two days after this test and I'm currently hunting for a replacement board. Once I have a functioning Ultra-45 again, I'm available for testing.

Kernel on the Ultra-45 was a lightly modified GENERIC (enabled igx, but that didn't pick up the Sun 10G card and enabled INSECURE to be able to run X). Kernel on the V100s was GENERIC with a lot of unneeded drivers yanked and DIAGNOSTIC enabled.

The Ultra-45 was running with NFS root at the time, as the firmware refuses to boot from the NetBSD install on the internal disk.


>How-To-Repeat:
Grab a SUN machine with UltraSPARC IIIi CPU (replicated on Sun Ultra-45 and Fire v450) and try to build lang/swi-prolog-lite from pkgsrc. Observe the machine crashing.

>Fix: