swap_dup/swap_free errors with 2.4.20-pre10

I'm running 2.4.20-pre10 on a Dual PIII system with 2Gb of RAM and three 2Gb swap logical volumes.

It runs fine for a while, then I get lots of;

Oct 15 12:41:31 maeve kernel: swap_dup: Bad swap file entry 00000020
Oct 15 12:41:31 maeve kernel: swap_dup: Bad swap file entry 00000020
Oct 15 12:41:31 maeve kernel: swap_free: Bad swap file entry 00000020
Oct 15 12:41:31 maeve kernel: swap_free: Bad swap file entry 00000020
Oct 15 12:41:31 maeve kernel: swap_dup: Bad swap file entry 00000020
Oct 15 12:41:31 maeve kernel: swap_dup: Bad swap file entry 00000020
Oct 15 12:41:31 maeve kernel: swap_free: Bad swap file entry 00000020

The address is always 00000020. I've tried the machine without any swap space, and I get exactly the same error, so I'm assuming it's either bad RAM or a kernel issue. I ran memtest86 on it yesterday, and it didn't throw up any errors, but I'm going to swap the RAM out and see if that fixes it.

Thanks,
David

David Coulson                                  http://davidcoulson.net/
[email protected]                       http://journal.davidcoulson.net/

Any news on this one, David?

Hey Marcelo,

Sorry, I forgot to follow up - I spend most of yesterday morning trying to stabalise the thing and didn't end up posting my results. Basically, my board can only handle 1.5Gb of PC133 properly, even though it will try to use 2Gb if you put that much in it. Interestingly enough, it passed the memtest86 tests I ran on it the other night, so I'm not sure what's going on there. Tyan, the board manufacturer, confirmed that the system is only stable with 1.5Gb of PC133, which is somewhat disappointing, but I guess I'll have to live. Interestingly, it ran fine for about 8hrs before going funny the first time, then it would spit out swap_dup/free errors within 30 to 60mins five times in a row.

I had weird lockups under 2.4.20-pre9, where the system would behave oddly - Most commands would work, but 'ps' simply locked up and I couldn't Ctrl-C out of it. I've moved back to 2.4.19-ck7-rmap, which seems to be stable at the moment, although I may take another look at the 2.4.20-pre10 kernel sometime. As always, I didn't have a keyboard or monitor hooked up to it, so I couldn't do too much with sysrq, but I'll be ready if it does it again.

Any points about the above lock-ups would be useful - Since 'ps' locks and others don't (e.g. 'ls'), I think it's quite a specific issue, although I've so far been unable to track it down.

Thanks,
David

David Coulson                                  http://davidcoulson.net/
[email protected]                       http://journal.davidcoulson.net/

I've seen this bug multiple times. Basically, something is holding a mm_sem and not letting go. Anything that walks the process list hangs. Ultimately, this hangs anything that's remotely useful, and you have to crash the box.

I've seen it on my laptop several times, and it hung a UML server that we have. UML is frequently, but not always involved.

We got a sysrq t from the UML server. I posted to lkml about it, with no response. You can see that at

	http://marc.theaimsgroup.com/?l=linux-kernel&m=103351640614665&w=2

One factoid that I forgot to mention there is that when it happens on my laptop, the disk activity light is stuck on.

Jeff

Indeed. I experienced the problem every 24-36hrs around five times in a row last week. Pretty much every morning, I'd come in and the box was completly dead. It seems to be okay at the moment, but I'm not holding my breath.

My box doesn't have a light, but what I got from SNMP before the box died suggested that the system (e.g. the 'system' CPU usage MIB) was using a considerable amount of CPU time (>95%), so I'm not sure if it was swapping madly, or if something else was going on.

David

David Coulson                                  http://davidcoulson.net/
[email protected]                       http://journal.davidcoulson.net/