Re: [Scst-devel] target<->initiator loop with QLogic

[Ming Z. <mi...@el...>]

On Sun, 2006-11-12 at 22:20 +0200, Dan Aloni wrote:
> On Sun, Nov 12, 2006 at 01:42:28PM -0500, Ming Zhang wrote:
> > On Sun, 2006-11-12 at 19:59 +0200, Dan Aloni wrote:
> > > On Sun, Nov 12, 2006 at 12:47:56PM -0500, Ming Zhang wrote:
> > > > i do not think the FC chip allow to run at both target mode and
> > > > initiator mode at same time. so u have to use 2 HBAs on each box.
> > > > 
> > > > Ming
> > > 
> > > As I understand it the Initiator implementation is mostly in the 
> > > driver and the firmware is unaware that a Target implementation
> > > was 'hooked' into it.
> > 
> > no, firmware is aware of this target mode. when target mode is enabled,
> > some mailbox commands are sent to firmware to enable the target mode.
> > target mode support is built in firmware as well.
> 
> Okay, that is very interesting. Did this support exist in all versions 
> of the qla23xx firmware so far, or was it added only recently?

should be in it from very beginning. 

> 
> > 
> > > 
> > > Are you suggesting that if I use qla2xxx as a target I can't use it
> > > as an initiator on the same port for any other target? IIRC there
> > 
> > i just have a quick check on ISP2300spec. It does not say target mode
> > and initiator mode can run at same time. but it does not say No neither.
> > also target mode bit and ini mode bit are different bits so possible you
> > are right, it can do so. just the driver code does not support this yet.
> 
> About the driver, I'm using the patched version which is supplied with 
> scst, so *that* version of the driver does support target mode. 

that version support target mode. and it is patched against ini mode
driver coming with kernel.

> 
> This problem with Fibre Channel can also be viewed from a different angle,
> suppose TCP/IP. In TCP/IP, if you want two computers to be both client
> and server, you need a minimum of two ports in each side in order to 
> support the two distinguished TCP/IP sessions. 
> 
> If the same restriction also applies to Fibre Channel SCSI, it means that 
> you must have two ports on each side in order to establish a BiDi-SCSI 
> setup. Perhaps it is a violation of the SCSI standard itself as it would
> have been a violation of the TCP/IP standard if we hacked a way to 
> create a TCP/IP session in a symmetrical fashion between two computers.

For SCSI it should be 2 independent sessions.

> 
> > > are already some implementations that combine target and initiator
> > > on the same port, notably LSI's implmentation.
> > 
> > lsi has ini support in kernel and a mptstm code to support target mode.
> > but i do not know if it can run at same time. i have no lsi spec, so not
> > idea.
> 
> About LSI, I have tested it, and it can run at the same time, although 
> there is a problem with the Bidi-SCSI setup that I've described - the
> problem is that there's a login-logout storm between the two ports, 
> because each direction takes up a 'session', and then you get two
> conflicting sessions for each direction. The firmware gets confused 
> and can even crash...
> 
> The problem with LSI is that you can't resolve that issue because most
> of the logic is in the firmware and the firmware is not opened. The
> advantage of QLogic is that it might be possible to implement this
> easily and this is why I wrote to you guys.
> 

qla has lot of stuff in firmware as well.