On Mon, Nov 03, 2014 at 10:26:02PM +0000, Rob Herring wrote: > Adding Leif... [...] > I'm just wondering does UEFI being used for the memory information > have any impact here as the DT would not have valid memory nodes > either? For UEFI, things should just work (TM). The dtb will be free of memory nodes (and memreserves), and the next kernel will discover the UEFI memory map via the linux,uefi-mmap-* properties as the first kernel did. We shouldn't need to pass any additional data in this case. > I'd assume reserved memory comes from UEFI (or both) in that > case? Yes. Any memory that should be reserved should be described as such in the UEFI memory map. One caveat is that reserved-memory nodes are preserved (because they can be referred to by phandle for CMA and such). There are few potential problems with this (we'll have to reconcile it with the UEFI memory map, prevent overlap when the kernel is relocated, etc). The reserved-memory nodes should only be used for CMA type carve outs rather than for protecting firmware and similar. > Perhaps we need to expose memory layout independent of DT, UEFI > or anything else. I can see that being useful for debugging, but I don't think that we require it. Thanks, Mark.
