Re: Patches for SCSI scanning

[Patrick Mansfield <patmans@us.ibm.com>]

On Tue, Apr 20, 2004 at 06:42:38PM +0200, Kurt Garloff wrote:
> Hi Patrick,
> 
> On Tue, Apr 20, 2004 at 09:26:51AM -0700, Patrick Mansfield wrote:
> > On Tue, Apr 20, 2004 at 09:38:00AM -0500, James Bottomley wrote:
> > > Yes, sounds reasonable.  We can do it all with the boot flags now.
> > 
> > max_scsi_luns should default to zero to avoid devices that should be black
> > listed as single lun.
> 
> I strongly disagree.

There were some posts in the past by either Alan or Doug L on this, I
can't find them. Redhat ships with CONFIG_SCSI_MULTI_LUN not set.

> The target should be to blacklist broken devices not working ones. 

I agree that is where we would like to be, but that can break current usage.

> Devices that have multiple LUNs are not, devices that have only one but
> report severals are.
> 
> With max_scsi_luns=1, we would need to "black"list every single multi lun
> device. I very much dislike that idea.

Not individually, just boot with max_scsi_luns=128 or whatever as is done
today for redhat, or boot with default_dev_flags=0x040 (BLIST_SPARSELUN).

> Fortunately, distributors don't do this.

Suse doesn't, Redhat does, I don't know about others.

> > So how can you remove those and maintain compatibilty? The removal should
> > wait for 2.7.
> 
> Expecting your multi lun devices to work with max_scsi_luns=1 is just
> wrong. max_scsi_luns should not default to 1.
> Anyways, BLIST_SPARSELUN will have the side-effect of also working
> around that misconfiguration.

Again that is not compatible with the current 2.6.x code, users with
non-black listed single lun devices will get failures if we default to
max_scsi_luns > 1.

-- Patrick Mansfield