Very Computer

> ... does Linux support running as a SCSI target (rather than a host) at
> all?

> As a lot of my work involves computer preservation, it'd be quite handy to
> emulate an old SCSI disk / tape / CDROM using a modern Linux PC - when
> hooked up to an old system, the old system would just think it was
> talking to a standard SCSI device.

> I'm quite happy to write sg driver code, but I get the impression that
> there's no underlying support at all - the kernel always assumes it's the
> host rather than target :-(

> cheers

> Jules

>>The SCSI target does all the work. The master only starts it
>>up.

> Yep, but this is using the SCSI HBA as the target - with another HBA on
> the bus as the master. It's all within the SCSI spec AFAIK, but I'm not
> convinced that the Linux kernel drivers support target mode at all...

>>AFAIK, the interface chips used on PC SCSI cards do not handle the
>>target mode.

> Wow - that's pretty cheap and crappy if they don't :( I assumed the
> hardware side of it would always allow either (at least for
> reasonable-quality HBA's, forgetting cheap scanner interfaces etc.) and it
> was just the software that might be limited.

>>> The SCSI target does all the work. The master only starts it
>>> up.

>> Yep, but this is using the SCSI HBA as the target - with another HBA on
>> the bus as the master. It's all within the SCSI spec AFAIK, but I'm not
>> convinced that the Linux kernel drivers support target mode at all...

>>> AFAIK, the interface chips used on PC SCSI cards do not handle the
>>> target mode.

>> Wow - that's pretty cheap and crappy if they don't :( I assumed the
>> hardware side of it would always allow either (at least for
>> reasonable-quality HBA's, forgetting cheap scanner interfaces etc.)
>> and it
>> was just the software that might be limited.

> The target has to do more than the initiator.
> Please read the SCSI specification. After initiation,
> the transfer is controlled and timed by the target
> alone, the initiator just complies.