The only way I have ever been able to exercise this is by
using the scsi-debug fake host adapter, and by then running all of the
scsi code in user mode under gdb. Crude, but before I was able to do
something like this, the abort/reset code was completely useless.
That certainly explains a few things. I can't claim to really understand all
the permutations of reset/abort handling that code attempts. I am going to
attempt to prototype some changes to scsi.c togerther with my BusLogic driver
along the lines I discussed in the long message I sent recently. Hopefully, I
will be able to come up with a 100% stable solution. I'm rather leery of
making wholesale changes to code I don't really understand completely right
before a release. Once there is *a* solution, we can try to figure out how
best to generalize it and where the boundary between the mid level code and
driver code ought to be.
Yeah, I saw it, but I have been kind of busy.
It looks like linux-scsi is just very slow. It took over 1.5 days for the
message to come back to me.
Leonard