Hi Edgar,
On 08/19/2018 09:28 PM, ed...@pettijohn-web.com wrote:
Thread 1 received signal SIGILL, Illegal instruction.
0xb66f3d50 in gcm_ghash_4bit_mmx () from /usr/lib/libcrypto.so.12
(gdb) bt
#0 0xb66f3d50 in gcm_ghash_4bit_mmx () from /usr/lib/libcrypto.so.12
#1 0x00000000 in ?? ()
very nice:)
now I don't remember if this CPU has MMX or not...
cpu0 at mainbus0
cpu0: Intel 586-class, 232MHz, id 0x581
cpu0: package 0, core 0, smt 0
It should though, according to:https://www.thinkwiki.org/wiki/Category:600
Riccardo
Looks like you found the issue.
There was a thread on tech I believe talking about something very similar not
too long ago. Might be worth looking through the archives.
I took the time to browse through the archives, going back a couple of
months, but couldn't find anything regarding this.
I wonder if there is a "solution", but also know exactly what the issue is.
At first I thought it was using MMX without CPU supporting it, but it is
not. Then I searched further and found this thread, which I ignored at
first because I have "genuine intel"
https://www.mail-archive.com/tech-userlevel@netbsd.org/msg01643.html
It really looks a crash in the same place and MMX using actually on SSE
which my cpu (and probably many others) do not have.
I tried the trick of setting
export OPENSSL_ia32cap=~0x800000
and it works.
Reading the discussion it appears to be an issue on how NetBSD builds
libcrypto and that it has been fixed by a commit by Christos.
I wonder if I can get a "new" libcrypto binary somehow, e.g. by a test
build, and replace mine? some update without updating the whole system
to unstable?
Riccardo
