[ofa-general] [RFC 0/1] libmthca: CQ/DMA race on Altix

[Roland Dreier]

 >  dma_map_sg(struct device *dev, struct scatterlist *sg, int nents,
 > -	   enum dma_data_direction direction)
 > +	   enum dma_data_direction direction, int coherent)

"coherent" seems like the wrong name here... really the property being
asked for is "flush other in-flight DMAs" or something like that (I
don't know precisely what setting the magic bit in the DMA address
does on Altix).

Also maybe it would make more sense to fold this into the existing
direction parameter somehow, so that most of the kernel can stay
unchanged (because as far as I know, Altix is the only platform that
has this extra quirk of allowing DMAs to pass each other).

 - R.