Re: [Qemu-devel] [PATCH] megasas: Update function megasys_scsi_uninit

[Hannes Reinecke <hare@suse.de>]

On 07/31/2012 05:45 PM, Paolo Bonzini wrote:
> Il 31/07/2012 17:15, Anthony Liguori ha scritto:
>> Andreas Färber <afaerber@suse.de> writes:
>>
>>> Am 31.07.2012 16:50, schrieb Paolo Bonzini:
>>>> Il 31/07/2012 16:46, Andreas Färber ha scritto:
>>>>>>> Why would megasas be in master but not compiled/linked?
>>>>>> Because Anthony objected to how it picks the initiator WWN.
>>>>> Ah, anything keeping us from fixing that? :)
>>>>
>>>> Exact knowledge of the requirements, basically. :)
>>>
>>> I.e. waiting on feedback about numeric constraints from Hannes? :)
>>> Or waiting on general direction from Anthony?
>>> How to make progress there for 1.2 / Soft Freeze?
>>
>> You cannot build guest visible state by casting pointers.
>>
>> I'm really disappointed that this hasn't been resolved yet.  If it's not
>> resolved before the soft freeze, I think we ought to revert the megasas
>> patches completely.
>
> Hannes, can you give a quick yes/no on the approach of the attached
> patch?  The address of the HBA is given by a wwn property or, in the
> lack of one, by an increasing index similar to the one used for MAC
> addresses.  The address of the disks is also given by a wwn property or,
> in the lack of one, by combining the HBA address with the target number.
>
Sorry for not answering earlier, but real life interfered
(customer escalations et al ...)

Anyway, Paolo, your approach is partially correct.
Ad 1, yes, we should be having a property for setting the SAS address of 
the HBA. So that's okay and can go in. Anyone concerned with uniqueness 
can then set it as appropriate.

Ad 2: no, the SAS address for the devices is _not_ the WWN.
The WWN is the ID of the LUN, whereas the SAS address is the ID of the 
Target. So to be correct we would need to generate unique SAS addressed 
per target which needs to be different from the WWN.
However, megasas does not need to present SAS addresses here; we can set 
the interface to PCI-E (instead of SAS) and just use the LUN number.
Sadly I've yet to figure out the code for PCIE.
(Doing it correctly involves staring a binary output and figuring the 
meaning of bits. Did I mention I don't have documentation for that beast?)
That's what's stalled the patch for now.

I see to have a patch cooked up tomorrow.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		      zSeries & Storage
hare@suse.de			      +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Markus Rex, HRB 16746 (AG Nürnberg)