On Monday 07 Dec 2020 at 13:40:52 (+0000), Will Deacon wrote:

Why not use the RESERVEDMEM_OF_DECLARE() interface for the hypervisor
memory? That way, the hypervisor memory can either be statically partitioned
as a carveout or allocated dynamically for us -- we wouldn't need to care.

Yup, I did consider that, but the actual amount of memory we need to
reserve for the hypervisor depends on things such as the size of struct
hyp_page, which depends on the kernel you're running (that is, it might
change over time). So, that really felt like something the kernel should
be doing, to keep the DT backward compatible, ... Or did you have
something more elaborate in mind?

Thanks,
Quentin