[openib-general] assumptions on page mapping (was High memory)

[Or Gerlitz]

Roland Dreier wrote:
> The right way to use the MR from get_dma_mr() is to use "bus
> addresses" from the DMA mapping API.  For highmem, the right way to
> get those addresses is with dma_map_sg() or dma_map_page().

Looking on the kernel x86_64 code, both dma_map_sg and dma_map_page seem 
to assume that the page is already mapped, since they call 
page_address(page).

Specifically is it safe in a SCSI LLD (eg SRP and iSER which is among 
other things such) to call dma_map_sg on a SG which comes with a SCSI 
command, so the SCSI Mid-Layer always makes sure the pages are mapped?

Or.