Invalid opcode in ramzswap.c:1314 (v0.6) about compcache

[GoogleCodeExporter]

This could be related to Issue #42, as this only occurs when backing swap is 
used, but 
I get a kernel BUG report in syslog, and this was not reported in #42.

Original comment by [email protected] on 2 Nov 2009 at 10:37

[GoogleCodeExporter]

Are you sure you have pulled latest hg version? for me, hg log shows:
[ngupta@vflare compcache]$ hg log | head

changeset:   91:65a5cd123d0e
tag:         tip
user:        Nitin Gupta <[email protected]>
date:        Thu Oct 15 04:36:55 2009 +0530
summary:     Correct permissions for code files

changeset:   90:ce3d45b1cb38
parent:      83:abafa311d458
user:        ngupta@vmFedora11
date:        Thu Oct 15 04:24:33 2009 +0530

also make sure you are in "default" branch.

---

btw, I could not yet understand how it could happen on compcache-0.6 either. I 
tried
same thing with latest hg version on Fedora-x64 but could not repro the issue. 
I will
find some 32-bit machine and maybe I will be luckier.

Maybe on 32-bit system, disksize value is overflowing. Is your backing swap 
device
>=4G? Also, how much RAM do you have? Anyway, I should be changing it to 64-bit.

Original comment by [email protected] on 2 Nov 2009 at 1:55

[GoogleCodeExporter]

I found the trigger of the bug. /dev/sda2 isn't a swap partition, it is an 
"Extended" 
partition which contains my real swap partition (/dev/sda5). swapon handles 
this more 
gracefully, giving the following error:
    swapon: /dev/sda2: read swap header failed: Invalid argument
presumably rzscontrol should also handle this more gracefully.

If this helps "cat /dev/sda2 | wc -c" reports that /dev/sda2 is 1024 characters 
long.

FYI, with regards to hg, I just followed the instruction at  
   http://code.google.com/p/compcache/source/checkout
That is, I ran 
   hg clone https://compcache.googlecode.com/hg/ compcache

Original comment by [email protected] on 2 Nov 2009 at 3:02

[GoogleCodeExporter]

> That is, I ran 
>    hg clone https://compcache.googlecode.com/hg/ compcache


To stay uptodate you need to run 'hg pull' to get latest changes.

---
In general, if you give non-swap partition as backing swap, ramzswap does not 
cause a
BUG() and you will get a graceful failure from swapon /dev/ramzswap0. However, 
in
this case -- i.e. with extended partition -- we get backing swap of zero
(i_size_read() returns 0). This fails ramzswap logic. I will fix this case.

Thanks,
Nitin

Original comment by [email protected] on 3 Nov 2009 at 1:54

[GoogleCodeExporter]

"hg pull" reports no new changes.

Original comment by [email protected] on 3 Nov 2009 at 3:24

[GoogleCodeExporter]

Fixed in repository. hg pull should now get you the fix. You should now get an 
error
in kernel log when you try to do --init with extended partition as backing 
swap. Thanks.

Nitin

Original comment by [email protected] on 12 Nov 2009 at 1:40

• Changed state: Fixed