Originally posted by: SystemAdmin

How can I force AIX 5.3 OS to crash while running to test HACMP???

Originally posted by: nagger

Very funny but why would you want to do that?
Do you expect IBM to have such a dangerous command lying about?

halt -q
does a good job but it is not not a crash.

Unplugging the network and disk cables would probably works well but make sure you do not wear out your cables/connectors.

A power off button or yanking out the power cord is effective but gives AIX a few milliseconds or for a logical partition an immediate shutdown does the same.

I have heard of people deliberately corrupting the kernel memory via /dev/kmem but that is not recommended - all bets are off for recovering from this - i.e. you might do it for fun in a workshop or a system about to be reinstalled but never on a production system.

Originally posted by: n5red

When I took the HACMP class, they had us do 'echo "hello" >/dev/kmem'

But that was on AIX 4. I haven't tried it on AIX 5.

• Matt

Originally posted by: VirtualGreg

forgot about this thread - I was going to suggest the very same idea. That's a pretty good way to generate a protection exception crash. 8^)

Originally posted by: nagger

You might use this approach on a workshop or test system but please
DO NOT DO THIS FOR PRODUCTION MACHINE TESTING.

Writing to /dev/kmem like this corrupts your interrupt handler vectors (pointers to functions).

At this point ALL BETS ARE OFF.
1) If you are very lucky this will hang or halt AIX.
2) If you are very unluck it could thoretically corrupt every single file and filesystem before it goes down and HACMP is NOT going to be able to correct this - you will be reaching for your latest backup tape.

Originally posted by: VirtualGreg

I don't think so. This is an attempt to write to /dev/kmem. The point is you cannot!

Granted, it's been a while since I did AIX kernel support, but the idea here is that you are attempting to write at the beginning of /dev/kmem which is protected. You cannot write there, thus nothing can get corrupted. The resulting crash code tells you that you have generated a protection exception. In other words, you tried to write where you cannot write, do you do not write, you crash.

I recall a kernel debugging course where we had to debug a protection exception cause by a device driver that simply tried to write to address 0x0 or something like that - been a few years. 8^)