On Wed, Nov 16, 2005 at 07:47:36PM -0800, Garrett D'Amore wrote:
> So, either bge is doing something bad and calling this with a zero
> length, or the bus_dma implementation is broken because a zero length is
> legal.

We had a similar case for scsipi recently, where userland ioctls with
borked arguments could cause the load of zero length bus_dmamaps. In that 
thread the conclusion was that 0 length is illegal and should be caught
by assertions, if possible.

So it looks like bge needs to be fixed.

Martin