FreeBSD core dump

[beefdart]

Handle found. Using p9iw6fu2. To change, hit 'C' and wait 5 seconds...Illegal
instruction - core dumped

we (Computer Science House) have had this happen on 3 of our machines... Plus the other problem I wrote about before.

We have 30 FreeBSD servers that could be folding, just thought I would throw this other problem out there... other than time, no other data in error.log

thanks

[Brian the Fist]

I have the person who wrote the networking section of the NCBI toolkit looking at the problem so it should be fixed soon I hope. As for the core dump, what CPU are you running on when you get the Illegal instruction - that usually occurs when the executable has opcodes for a different CPU.

[beefdart]

dual pentium 300

amd 350

[Brian the Fist]

The code probably only works on P2 or equivalent and higher

[beefdart]

Well, we here at computer science house at rit would be happy to "fix" the code... email myself ryanl@csh.rit.edu or petef@freebsd.org

also if you would like us to make a solaris 7 port, we ould be happy to.

thanks

[alpha]

I have exactly the same problem on my K6-2 400 4.5-STABLE box. It was upgraded from 4.3-RELEASE.

It wouldn't work at first on my Athlon XP 1700 (also upgraded from 4.3 to 4.5), but for some reason it all of a sudden started working. I had been fiddling with various CPU kernel options not that long ago, so perhaps I enabled something that it liked, since Howard seems to think its a CPU thing?

If it matters I can go find out what kernel option it was that I was trying (think it may have been an option to enable SSE, so I don't think it will have affected the DF client).

Matt

[Brian the Fist]

Could very well be SSE. The BSD version is built with GCC and the following flags:

-O3 -funroll-loops -ffast-math -pthread -march=pentiumpro -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64

I suspect the -march=pentiumpro is the source of your grief. If this is really a problem for people, Ill remove this flag. I was just hoping to get a little extra "bang" and thought that most modern processors should be pentium-pro (aka Pentium-II aka 686) compatible.. I can build a version without this flag and you can tell me if it goes away.

[beefdart]

that would be great... any closer on the networking issue??

[alpha]

Originally posted by Brian the Fist
<snip>
I can build a version without this flag and you can tell me if it goes away.

That would be great, thanks for looking into this.

Matt

[alpha]

Originally posted by Brian the Fist
<snip>
I suspect the -march=pentiumpro is the source of your grief.
<snip>

I think you are correct. According to /usr/src/sys/i386/conf/LINT, a CPU defined as I586 is considered "Pentium" and a CPU defined as I686 is considered "Pentium Pro". Of course, my K6-2 400 is a fifth-generation CPU and was therefore defined as I586 in my kernel config.

If this is all accurate, I think it is safe to say that if your CPU isnt defined as I686 in your kernel config, the client wont work.

Matt