On Tue, 11 Apr 2006, Ludovic Court�s wrote:
Just in case it is of interest to you, I got the same sort of problem
with an SS20 equipped with 2 SuperSPARCs, 2 hard disks and a 2.6 kernel:
http://article.gmane.org/gmane.linux.ports.sparc/5979
I haven't had time to dig into it any further so far.
Ok, I did some digging. Since people also see this bug on Ultra1, it
must be esp-specific. Most recent change to esp.c dates to February
2006, all other changes are over a year old, so it's pretty unlikely
that they are the cause of this problem. The recent change [0], on the
other hand, looks pretty suspicious to me, since it changes the locking
order. I've built a kernel with this change reverted, and was not able
to crash the machine by running two concurrent loops of
unpacking/removing the kernel tree for an hour. I'd appreciate if you
could test this kernel, it is available at [1]. By the way, this problem
has been also reported to kernel's bugzilla
as bug 6344 [2].