Subject: Re: Request for review on scsipi changes
To: Manuel Bouyer <>
From: Matthew Jacob <>
List: tech-kern
Date: 09/17/2002 12:26:03
On Tue, 17 Sep 2002, Manuel Bouyer wrote:

> On Tue, Sep 17, 2002 at 12:09:59PM -0700, Matthew Jacob wrote:
> > 
> > 
> > BTW-I'd like to request that we have a flag that allows you to skip bus
> > resets- they're meaningless for Fibre Channel (well- not meaningless,
> > actually, but you really don't want to do them), and in multi-initiator
> > environments you really shouldn't be issue bus resets.
> bus resets are not issued by scsipi, it's up to the driver to decide to do
> them or not. Fibre Channel adapters can chose to skip them.

No- sorry- that's really not quite the right thing to do. You may want
to issue them at a later time- for something other than booting cleanup.

Right now isp/FC *is* ignoring this- but it shouldn't be done this

> For SCSI adapters I really think they should be there, it's the only way to
> get the busses back to a sane state when something goes wrong.

No, that's also generally wrong for normal operating devices. You should
be able to negotiate back to async/narrow.

Granted, not all is well behaved. That's why I stated this as a 'flag'-
multi-initiator environments and FC can set this flag. Or, for that
matter, those who don't want to wait a generally pointless 2 seconds to

> And it looks like solaris 2.5 is doing it as well.

I believe it was disabled as of Solaris 2.6. All OBPs later than 3.0 no
longer issue bus resets.

> Other adapters on the bus should be able to cope with the bus reset, and
> restart commands as appropriate.

Wrong.Wrong.Wrong.Wrong.Wrong.Wrong.Wrong.Wrong.Wrong.Wrong.Wrong.Wrong.Wrong. :-)

Think 'stateful' devices like 'tape'. I'm sorry- I booted a device on
the fabric and wiped out the end of a 9 hour critical tape backup
section. Oops!

Think 'stateful' commands like 'format' or 'fixate'.