[ofw] "livefish" mode

[Leonid Keller]

I guess, one can fix behaviour of query_interface routine so as it will
support the work in "livefish" mode.
But it is worth to discuss whether we need this mode at all.
Historically, Linux always provided (and provides now) to the customers
two drivers - mthca and mst.
The latter was intended for burning FW and other maintainance/debugging
works.
These drivers are working in parallel.
Windows PnP Manager didn't allow to have two drivers for one device, so
I've added mst functionality into mthca driver (and in mlx4 also).
In due time we came to the conclusion that it is more comfortable to
have a standalone maintainance tools package with its own driver.
This package we are providing now from our site.
The "livefish" mode is used now only as a special mode, into which the
low-level drivers pass when the fail to start up.
One can run vstat and see the warning, that driver failed to start out
of (possibly) incorrest FW.
It is good especially at install time, because this behaviour allows for
all other drivers of the stack to get installed and loaded.
Otherwise the installation ended up with low-level driver failed to load
and other drivers not installed.
Is this enough worthy for keeping this mode ?

> -----Original Message-----
> From: Sean Hefty [mailto:sean.hefty at intel.com] 
> Sent: Thursday, June 26, 2008 9:13 PM
> To: Leonid Keller; ofw at lists.openfabrics.org
> Subject: RE: [ofw] [PATCH 7/9] mthca/pnp: respond to query interface
> 
> >Why do you remove "livefish" functionality ?
> >It's needed in case of incorrect FW.
> 
> I had removed it to avoid crashing when a query interface 
> arrived at the HCA after a PnP device add event.  From what I 
> could determine, livefish was used in this context simply to 
> avoid setting specific interface functions.
>  
> I did eventually end up changing winverbs to delay its query 
> interface call until it received a PnP D0 power event, so it 
> may be possible to add it back.  The interface wasn't usable 
> until then anyway.  I guess we can trust kernel clients not 
> to generate a query interface call until the D0 power event.
> 
> >You can add the same stuff in mlx4\kernel\hca\drv.c
> 
> Thanks - that saves me some searching.
> 
> - Sean
> 
> 
>