Is debian accidentally desupporting i386 processors?
Scenario: I have been running debian on a 386 processor. Yes - a real 386
- not a 486. Makes a great stand alone web server.
After using potato on an AMD K6 I decided to update the machine to the
current incarnation of potato. The machine has been running 2.0.27 kernel
and the 2.2.10.
snmpd and nfs-server will start running and then die with an illegal
instruction. The snmpd test is simple, install and run snmp-walk on the
server, watch it die.
If I compile snmpd and the nfs-server without optimization and or with -g,
everything is fine. When compiled with optimization, death
(asn_parse_length). The same executables do not suffer on the K6.
The default compiler is now egcs. I notice that the specs file installed
will use the 486 instruction set by default. (simple test will verify
this, compiling with -m386, -m486, or no arguments and see the code that
is produced.)
So, is Debian, by using the egcs compiler configured on a 486
inadvertantly producing code that might die on a 386?
Unfortunately, snmpd when compiled -m386 also dies - so I need to diagnose
that some more - it could be an egcs optimization bug for 386 specific
code.
Am I barking up the wrong tree - or might this be a problem?
Ezra Peisach
Reply to: