Re: [RFC PATCH 1/2] xen/unpopulated-alloc: Introduce helpers for DMA allocations

[Roger Pau Monné]

On Tue, Jun 14, 2022 at 05:45:53PM -0700, Stefano Stabellini wrote:
> Basically, if we allocate (and free) page-by-page it leads to more
> efficient resource utilization but it is slower. If we allocate larger
> contiguous chunks it is faster but it leads to less efficient resource
> utilization.
> 
> Given that on both x86 and ARM the unpopulated memory resource is
> arbitrarily large, I don't think we need to worry about resource
> utilization. It is not backed by real memory. The only limitation is the
> address space size which is very large.

Well, unpopulated memory will consume memory to populate the related
metadata (ie: struct page and realted tracking information) for the
unpopulated region, so it's not completely free.

Thanks, Roger.