(By the way, "coda_inactive: 0x.... usecount 1" messages at least
  somtimes repeat with repeating accesses, not only the first time as I
  wrongly believed)

These are annoying, but I think the rigth fix is just to remove the
printf.  I have commited that to NetBSD trunk, so when 6.0 comes out
they'll be toned down.


Are you running with DIAGNOSTIC?


What provokes the crash?

  linux binary reading a directory in coda

or

  linux binary in coda being run, not trying to read data from coda

or

  you have to do both?



The big issue with coda and NetBSD now is that getcwd() seems not to
work right.