On 3-Mar-09, at 4:41 PM, Christoph Egger wrote:
*or* we change bus_dma(9) API to define the map pointer to be NULL when bus_dmamap_create() fails and fix the bus_dmamap_create() implementation across all ports. IMO, letting the drivers take care to set the map pointer to NULL is more error prone since it will surely be often forgotten by developers when changing or writing/porting a driver. So personally, I would prefer the bus_dma(9) API change since this is a "fix-this-once-and-for-all task". Comments?
I vigorously agree. I just encountered exactly this scenario last week...