On 03.04.19 10:34, Oscar Salvador wrote: > On Wed, Apr 03, 2019 at 10:12:32AM +0200, Michal Hocko wrote: >> What does prevent calling somebody arch_add_memory for a range spanning >> multiple memblocks from a driver directly. In other words aren't you >> making assumptions about a future usage based on the qemu usecase? > > Well, right now they cannot as it is not exported. > But if we want to do it in the future, then yes, I would have to > be more careful because I made the assumption that hot-add/hot-remove > are working with the same granularity, which is the case right now. > > Given said this, I think that something like you said before, giving > the option to the caller to specify whether it wants vmemmaps per the > whole hot-added range or per memblock is a reasonable thing to do. > That way, there will not be a problem working with different granularities > in hot-add/hot-remove operations and we would be on safe side. There might still be an issue if the person adding memory might be somebody else removing memory. I am not yet sure if we should even allow add_memory/remove_memory with different granularity. But as I noted, ACPI and powernv. -- Thanks, David / dhildenb
