On Fri, Aug 9, 2024 at 3:33 AM Peter Chubb via Devel <devel@sel4.systems> wrote:
>
> This kind of thing is doable, or rather will be doable, by swapping out
> modules.  Although the _architecture_ is static, not all the modules
> need to be runnable at any one time.  In LionsOS one will be able to
> instantiate template PDs that can run a range of different payloads
> loaded at run-time, to allow a degree of dynamism.
>
> We have a student working on the hotplug problem at the moment.  But
> it's unclear, in the general case, what should happen when something
> appears in, or disappears from, the system.  Utimately it's going to
> be up to the system designer to design something that works for each
> system.
>
> For example, if you remove a storage device (SDHC card, or USB stick).
> Your alternatives are:
>   -- discard all in-flight requests for the device, unmount any
>   filesystems etc.
>   -- Mark the device as 'gone'; but if there are any in-flight requests,
>   or if any new requests get queued, ask the GUI to pop up something
>   asking for reinsertion of the device (like AmigaDOS did)
>   -- something else I haven't thought of...
>
> In whatever case, you need system policy to decide what to do, and
> code to implement it.
>
> I expect the handheld device _will_ eventually be a target for a
> LionsOS system, but there's a fair bit of work to do.
>
> In the near term, there will have to be more device drivers and
> virtualisers in the sDDF; more higher-level components in LionsOS
> itself; and the template PDs that are currently being worked on here
> by an undergraduate student.
>

Even with this limited form of dynamism it's still going to be rather
limiting. I'd think it would be limited to be more or less like a
software secure enclave type of thing, or a fixed set of VMs for
different security levels. Applications are still likely going to be
run in Linux VMs with all of the issues of Linux. Running Linux VMs
and a relatively closed set of secure applications isn't going to
magically fix all of Linux's issues. IMO a better approach for
general-purpose OSes is to forgo verification and go with something
more akin to QNX architecturally (of course, with a centralized
security model rather than QNX's security model that requires origin
servers to verify permissions). Even a non-verified QNX-like system is
going to be far more secure, stable, and extensible than just about
any kind of Linux system.
_______________________________________________
Devel mailing list -- devel@sel4.systems
To unsubscribe send an email to devel-leave@sel4.systems

Reply via email to