Re: [PATCH 3/8] mm/memory_hotplug: add APIs for explicit online type control

["David Hildenbrand (Red Hat)" <david@kernel.org>]

>> For dax we know that user space will define the policy.
>>
> 
> Actually this may not always be true.  A driver spawning a dax on probe
> might also end up selecting the policy... eventually... maybe... I might
> be planning to add that glue between CXL and DAX so I can add some
> config similar to the system-default policy to avoid systems with
> multiple memory-devices being forced into the same policy
> 
> (e.g. CXL memory device can online auto in ZONE_MOVABLE, but the other
>   device can have its own policy).
> 
> There's a weird corner case for CXL auto-regions (BIOS configured
> everything but left the memory EFI_MEMORY_SP - so comes up as DAX).
> I'm trying to keep those systems working the same as they have been
> while the userland policy stuff catches up.  Early CXL patterns are :[

Right, but I don't want any other OOT kernel module to be able to make 
use of add_memory_driver_managed() to do arbitrary things, because we 
don't know if it's really user space setting the policy for that memory 
then.

So either restrict add_memory_driver_managed() to kmem+virtio_mem 
completely, or add another variant that will be kmem-only (or however 
that dax/cxl module is called).


-- 
Cheers

David