i386/79169 as USB issue?

[John Pettitt]

Bug i386/79169 is beginning to look like a USB issue -
http://www.freebsd.org/cgi/query-pr.cgi?pr=i386/79169

I rebuild the kernel with PREEMPTION defined and presto the bug no
longer triggers.  I'm guessing someplace there is a condition where the
USB driver sleeps with a lock and the second USB disk causes a fatal
embrace because the IRQ thread can't run without preemption.

Reproduce the bug with a striped USB disk set and a loop running dbench
25 - it should fail in less than two hours if it's going to.