Anyone know why I might get different behavior when I run some
workload after m5 checkpoint, and run the same workload after resuming
from that checkpoint?

I do something like the following...
start gem5:
$> build/X86/gem5.fast
/home/gedare/work/gem5/gem5/configs/example/fs.py
--checkpoint-dir=/home/gedare/work/gem5/builds/hwds/m5out/1core-Java-7
--kernel=/home/gedare/work/linux/linux-2.6.28.4-gem5/vmlinux-2.6.28.4-8
--disk-image=/dist/m5/system/disks/linux-x86.img

in m5term terminal:
<boot  linux>
Script from M5 readfile is empty, starting bash shell...
(none) / # m5 checkpoint
(none) / # . setenv
(none) / # java -classpath * -Xms256M -Xmx256M -Djava.library.path=.
MyJavaApplication
<output as expected, MyJavaApplication works, yay)

Then I do it again...
$> build/X86/gem5.fast
/home/gedare/work/gem5/gem5/configs/example/fs.py
--checkpoint-dir=/home/gedare/work/gem5/builds/hwds/m5out/1core-Java-7
-r 1

in m5term terminal:
==== m5 slave terminal: Terminal 0 ====
(none) / # . setenv
(none) / # java -classpath * -Xms256M -Xmx256M -Djava.library.path=.
MyJavaApplication
<different output, with a runtime crash NoClassDefFoundError>

One thought I had is the disk size might be off, perhaps some
filesystem corruption occurs while restoring from the checkpointed
state.

Anyone have other ideas? FWIW I'm using a modfied gem5-devel from back
in October. It was working fine about 2 weeks ago, and I had to get it
working again, and something broke. I almost wonder if there is some
updated software on my host OS that gets used by gem5 to deal with
checkpoints that could cause problems? Does gem5 use something like
qemu-img to read/write the COW images?

-Gedare
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to