Ok,
so i did some stress testing on different file-systems this weekend
trying to track down this "bug". As far as I can tell, it only happens
when using XFS, because ext3 and reiser work fine(the patches for jfs
and xfs don't place nice, so i'll have to build an extra kernel for
that).
What i did was, make a 200 meg /dev/urandom file, then mount it via
loopback with twofish, then made the necessary file systems on it. And
I untarr'd kernel sources in there, so there would be a lot of meta-data
generation.
Anyway, so it consistantly crashes with xfs, at different points in the
untar'ing. My friend showed me how to play with Sysrq key, so after it
crashed i looked at things through there... and it seems to stop working
in the kupdated kernel thread. So i read a bit about it in my handy
dandy understanding the linux kernel book, and it "flushes old dirty
buffers from memory", which i believe has something to do with
filesystem block data in memory. So, that's what i have found so far,
any suggestions on where to go from here? i'm reading up on doing more
in depth kernel debugging, but very new to it ;-)
dave
|
