[host] adds init-paging feature

[danbugs]

This is an extra feature flag that allows disabling our paging setup (when removed). This is needed to further enable custom guests and should not impact any of our current workloads. init-paging is on by default.

[ludfjig]

Nice! Have you considered a runtime config vs cargo feature?

[simongdavies]

A couple of suggestions

[danbugs]

Nice! Have you considered a runtime config vs cargo feature?

Yes, but for this use-case I think a cargo feature is preferred. Users that are disabling paging will most probably be wanting run a specific type of binary (e.g., Nanvix) where you either want paging or not, so I don't think they'll want to mix-and-match different configs there at runtime. Plus, having to bubble that config option all the way to setup_initial_sregs would be a bit messy, imo.

[ludfjig]

Nice! Have you considered a runtime config vs cargo feature?

Yes, but for this use-case I think a cargo feature is preferred. Users that are disabling paging will most probably be wanting run a specific type of binary (e.g., Nanvix) where you either want paging or not, so I don't think they'll want to mix-and-match different configs there at runtime. Plus, having to bubble that config option all the way to setup_initial_sregs would be a bit messy, imo.

I think I personally would prefer a runtime config (opt out), to avoid a large feature matrix.

[danbugs]

Nice! Have you considered a runtime config vs cargo feature?

Yes, but for this use-case I think a cargo feature is preferred. Users that are disabling paging will most probably be wanting run a specific type of binary (e.g., Nanvix) where you either want paging or not, so I don't think they'll want to mix-and-match different configs there at runtime. Plus, having to bubble that config option all the way to setup_initial_sregs would be a bit messy, imo.

I think I personally would prefer a runtime config (opt out), to avoid a large feature matrix.

But then you'd just have a larger runtime config matrix? Not sure if there's an 100% right option here. Are you ok w/ proceeding w/ a feature at first?

cc: @simongdavies ?*

[simongdavies]

Nice! Have you considered a runtime config vs cargo feature?

Yes, but for this use-case I think a cargo feature is preferred. Users that are disabling paging will most probably be wanting run a specific type of binary (e.g., Nanvix) where you either want paging or not, so I don't think they'll want to mix-and-match different configs there at runtime. Plus, having to bubble that config option all the way to setup_initial_sregs would be a bit messy, imo.

I think I personally would prefer a runtime config (opt out), to avoid a large feature matrix.

I don't think it makes sense to have this as a runtime config, since this is something that's either required statically or not, I understand the concern about proliferation of features but this one really should be a feature not a configuration option IMO