Re: [PATCH 2.6.13 5/14] sas-class: sas_discover.c Discover process (end devices)

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

On Mon, Sep 12, 2005 at 12:21:20PM -0500, James Bottomley wrote:
> On Mon, 2005-09-12 at 09:45 -0700, Patrick Mansfield wrote:
> > On Mon, Sep 12, 2005 at 09:57:21AM -0500, James Bottomley wrote:
> > > be free to increase it if necessary.  Note: you do actually need either
> > > an array with more than two levels of nesting actually to need the
> > > increase and no-one actually seems to have one of these yet.
> > 
> > That is not correct, I posted before on this, the address method is in the
> > high bits of the 8 byte LUN and tells how to "interpret" the LUN value.
> > You can't convert from an int to 8 byte LUN (without any other
> > information) and set these bits. See SAM-4 in (or near) section 4.9.7.
> > 
> > So some storage devices that want to use addressing methods other than 00b
> > don't because we do not have 8 byte LUN support in linux, and then we have
> > other problems because of this.
> 
> Well, as long as we represent the u32 (or u64) as
> 
> scsilun[1] | (scsilun[0] << 8) | (scsilun[3] << 16) | (scsilun[2] << 24)
> 
> I think we cover all 2 level lun bases, don't we (obviously we ignore
> levels 3 and 4 [and 6 and 8 byte extended luns])?

hmmm ... yes, I'm wrong, it works (or should) in the existing code.
I don't know what I was thinking.

Though we still have problems with scsi_report_lun_scan code like:

                } else if (lun > sdev->host->max_lun) {

max_lun just has to be large, at least greater than 0xc001 (49153), maybe
even 0xffffffff, correct?

But then some sequential scanning could take a while. Maybe the above
check is not needed.

lpfc has max_luns set to 256, with max limited to 32768, I don't know how
it could be working OK here. (Has James S or anyone tested this?)

Probably similar for qlogic and maybe other non-SPI transports (though
current qlogic max_luns is 65535, gets us to ff ff 00 00 00 00 00 00).

> That representation works transparently for type 00b which is what SPI
> and other legacy expects, since our lun variable is equal to the actual
> numeric lun.  Although SAM allows type 01b for arrays with < 256 LUNs it
> does strongly suggest you use type 00b which hopefully will cover us for
> a while longer...
> 
> fc already uses int_to_scsilun and 8 byte LUN addressing, so it will
> work even in the 01b case (the numbers that the mid-layer prints will
> look odd, but at least the driver will work).

OK.

-- Patrick Mansfield