[ofa-general] Re: mthca use of dma_sync_single is bogus

[Roland Dreier <rdreier@cisco.com>]

 > One thought is that if you *do* move to dma_sync_single_range() then
 > lib/swiotlb.c still needs fixing. It's buggy in that
 > swiotlb_sync_single_range(dma_addr, offset) calls
 > swiotlb_sync_single(dma_addr+offset), and this will fail if the offset is
 > large enough that it ends up dereferencing a different slot index in
 > io_tlb_orig_addr.

Yes, I realized the same thing (our emails crossed).

 > So, I should be able to get my swiotlb workaround fixes accepted upstream as
 > a genuine bug fix. :-)

Yeah, seems so.

 > dma_sync_single_range() looks to me to be the right thing for you to be
 > using. But I'm not a DMA-API expert.

yes, I'll try to get confirmation from James Bottomley and/or Dave Miller
that it is the right thing to do (and also fix the documentation to
match what the kernel actually implements).

 - R.