On Fri, Nov 22, 2024, Matthew Wilcox wrote: > On Fri, Nov 22, 2024 at 09:29:39AM -0800, Elliot Berman wrote: > > A few near-term features are coming to guest_memfd which make sense to > > create a built-in library. > > You haven't created a library, you've created a middle-layer. This > file primarily consists of functions which redispatch to a function > pointer. I think you'd be better off creating a library! That is, > have the consumers register their own address_space_operations and > have functions in this library which provide useful implementations. Or, go even further and move the overwhelming majority of guest_memfd code into mm/. But I suspect that will be practically infeasible, in which case I think I agree with Matthew. Either way, the Gunyah usage of the library/middle-layer needs to be included in the series. It's impossible to review the code split and APIs without seeing both KVM's usage and Gunyah's usage. E.g. this series allows __guestmem_invalidate_begin() to fail, without any explanation as to while that's legal or desirable.
