2.5: PCI scatter gather list change
Hi Jens,
I got 2.5 booting on ppc64 today and noticed that the behaviour of sglists has changed. Before we would get an array of sg->address, sg->length pairs, we now get sg->page, sg->offset, sg->length triplets (from blk_rq_map_sg).
I dont care too much either way, but if this change is here to stay I'll let the non intel maintainers know so they can fix up their pci dma code.
Anton
Quoted 11 lines · Anton Blanchard
Hi Jens,
I got 2.5 booting on ppc64 today and noticed that the behaviour of sglists has changed. Before we would get an array of
sg->address,sg->lengthpairs, we now getsg->page,sg->offset,sg->lengthtriplets (fromblk_rq_map_sg).I dont care too much either way, but if this change is here to stay I'll let the non intel maintainers know so they can fix up their pci dma code.
The change is here to stay, the ->address stuff will die in 2.5. Consider highmem pages which have no virtual mapping.
Signature · 1 line
Jens Axboe
From: Anton Blanchard <[email protected]>
Date: Sun, 2 Dec 2001 15:35:53 +1100
I dont care too much either way, but if this change is here to stay I'll let the non intel maintainers know so they can fix up their pci dma code.
Ummm, platforms need to support this format even in 2.4.x If PAGE is not NULL, "address" applies, else the PAGE+OFFSET pair applies. We'll kill "address" very soon in 2.5.x, but it has to stay around for compatibility in 2.4.x
Look at sparc64 and x86 in 2.4.x, they expect and handle it correctly already. IA-64 fixed up their stuff recently too.