[openib-general] mthca and DDR not hidden

[Roland Dreier]

    Sean> Having an allocator routine might force users to perform
    Sean> data copies when sending data.

    Sean> Do all of the existing MAD implementations have routines to
    Sean> allocate MADs when sending data, and require those routines
    Sean> to be used?

Not Topspin's.

I think moving allocation into the MAD layer is probably a bad idea,
especially if the only motivation is to handle a DMA address of
0xffffffffffffffff for Tavor (I don't think that will ever happen in
practice).  The argument about IOMMU resources seems to be an argument
in favor of letting the consumer handle allocation.

 - R.