On 5-Mar-09, at 2:55 PM, Andrew Doran wrote:
Sure. I can buy "make it explode in the developers face"... I can't buy "let the DMA engine write to a bogus non-zero address if the guy is sloppy"...I have no idea how you made that connection.
yeah. That was clear in my head but not in my email... I just tracked down a bug in local code where a dmamap entry failed to bus_dmamap_load and hence my SG engine trampled on an mbuf that had since become re-used and was in the IP stack... Someone wasn't checking the result of a dmamap_load. So then we got talking here about error handling in dmamap_create...
Sorry.