Whilst (again) browsing the 3c59x.c sources, I note that when using DMA
bus mastering transfers, a different interrupt is used to signal TxDone.
As the TxDone interrupt FAILING IN HARDWARE is a potential cause of the
3c59x driver problems, perhaps the DMADone interrupt is more reliable? I'm
willing to tolerate a minor performance drop, to avoid using my
"dodgy-but-stops-the-hangs" hacked 3c59x.c driver.
Chris.