[SPAM] Sculpt scenario and block devices

[Josef Söntgen]

Hello Tomasz,

thanks for the feedback!

* Tomasz Gajewski <tomga at ...300...> [2018-03-04 12:09:56 +0100]:
> I can see block devices reported in /report/drivers/block_devices but I
> have two disks so driver_manager is not marking any of my disks as
> default.
> 
> Both disks are partitioned and I could give one partition exclusively to
> Genode to continue playing with it but not whole disk.
> 
> […]
> 
> I think there are two possibilities to continue.
> 
> First is to change Sculpt configuration during preparing image to match
> my configuration.

I have a similar setup (two disks, both partitioned) and I would
recommend to choose this option for now as you only need to change the
fs subinit to match your setup, i.e. point 'part_blk' to the proper
Block session and 'vfs' to the partition.

> Second is to change implementation of some component (?driver_manager?0
> to be flexible enough to automatically handle all my devices: disks
> (ahci), partitions on disks (part_blk on ahci devices), usb drives
> (usb), partitions on usb drives and expose them. Later it should be
> possible to just select a block device which user would like to user for
> Genode.
> 
> […]
> 
> I think that for each disk device (either ahci or usb) something should
> start part_blk for it, check if it properly detected partitions and
> expose them.

I agree, having some kind of probing mechanism in place would be nice
and is certainly something we want to address in future Sculpt
iterations. It stands to reason where to put such a probing mechanism,
though. I am not sure if the drivers subsystem is the proper place.
Although it would be nice to handle Block sessions transparently for the
other subsystems, it would increase the complexity of the mostly static
drivers subsystem (that is, if we ignore USB providing more than its Usb
session). The current layering of Sculpt subsystems considering
complexity and dynamic composition is first the drivers, than run-time
subsystem and at the very top the deploy mechanism…

> Addtionally there should be a configuration option that could be easily
> set in some config that would make a specific block device a default one
> (instead of policy that marks it as default if there is only one
> device).

… and at the moment, to make things simple, we rely on a somewhat static
configuration to point the deploy run-time to the depot location. Apart
from that, however, it is perfectly fine to make all decisions in the
run-time subsystem and, depending on the use case, even above that.

For example, on one hand I want to change the partition table on a disk.
That involves getting access to the whole Block device (“stop any
running 'part_blk'”) to make the change and afterwards re-reading the
partition table (“start 'part_blk'”). We already have a mechanism in
place to do just that: changing the configuration of the run-time
subsystem. On the other hand, if I just want to change the partition
table on a freshly inserted USB disk, I can do that from within the
deploy config — no need to change the run-time's configuration.

> I think you already have some vision how you plan to implement that part
> of system. Do you plan to extend driver manager implementation?

It really boils down to which services are necessary at which layer for
the rest of the system to work and what we need to do to use them
conveniently. If this requires extending the driver manager, we will
probably do that (although there are other developers that are more
qualified to comment on that).


Regards,
Josef

-- 
Josef Söntgen
Genode Labs

http://www.genode-labs.com/ · http://genode.org/