Re: [PATCH v2 3/3] ibmvscsi: Allow to configure maximum LUN

[Brian King <brking@linux.vnet.ibm.com>]

On 11/04/2015 07:02 AM, Hannes Reinecke wrote:
> On 11/04/2015 12:46 PM, Laurent Vivier wrote:
>>
>>
>> On 04/11/2015 12:16, Hannes Reinecke wrote:
>>> On 11/04/2015 11:20 AM, Laurent Vivier wrote:
>>>> QEMU allows until 32 LUNs.
>>>>
>>>> Signed-off-by: Laurent Vivier <lvivier@redhat.com>
>>>> ---
>>>>  drivers/scsi/ibmvscsi/ibmvscsi.c | 5 ++++-
>>>>  1 file changed, 4 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/scsi/ibmvscsi/ibmvscsi.c b/drivers/scsi/ibmvscsi/ibmvscsi.c
>>>> index 04de287..4480d3e 100644
>>>> --- a/drivers/scsi/ibmvscsi/ibmvscsi.c
>>>> +++ b/drivers/scsi/ibmvscsi/ibmvscsi.c
>>>> @@ -84,6 +84,7 @@
>>>>   */
>>>>  static int max_id = 64;
>>>>  static int max_channel = 3;
>>>> +static int max_lun = 8;
>>>>  static int init_timeout = 300;
>>>>  static int login_timeout = 60;
>>>>  static int info_timeout = 30;
>>>> @@ -117,6 +118,8 @@ module_param_named(fast_fail, fast_fail, int, S_IRUGO | S_IWUSR);
>>>>  MODULE_PARM_DESC(fast_fail, "Enable fast fail. [Default=1]");
>>>>  module_param_named(client_reserve, client_reserve, int, S_IRUGO );
>>>>  MODULE_PARM_DESC(client_reserve, "Attempt client managed reserve/release");
>>>> +module_param(max_lun, int, S_IRUGO);
>>>> +MODULE_PARM_DESC(max_lun, "Maximum LUN value [Default=8]");
>>>>  
>>>>  static void ibmvscsi_handle_crq(struct viosrp_crq *crq,
>>>>  				struct ibmvscsi_host_data *hostdata);
>>>> @@ -2289,7 +2292,7 @@ static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
>>>>  		goto init_pool_failed;
>>>>  	}
>>>>  
>>>> -	host->max_lun = 8;
>>>> +	host->max_lun = max_lun;
>>>>  	host->max_id = max_id;
>>>>  	host->max_channel = max_channel;
>>>>  	host->max_cmd_len = 16;
>>>>
>>> Please, don't do this.
>>>
>>> 'max_lun' should only be set if the HBA / transport has some hard
>>> limitations on the number of bytes it can use.
>>> Otherwise the scanning algorithm in scsi_scan.c should do the
>>> correct thing, independent on the 'max_lun' setting.
>>
>> So you are saying we can remove the line ?
>>
> Ho-hum. In principle you could, as ibmvscsi is using SRP internally,
> which does support 64 bit LUNs.
> 
> However, due to some weird design decisions there is
> the function 'lun_from_dev()', which mangles the incoming LUN number
> into something ... else.
> Which leaves only 4 bits free for the actual LUN number, requiring
> you to use max_lun = 16.
> 
> Personally I would just do away with that function and use the
> incoming LUN numbers as is.

I pulled out my copy of SAM, what lun_from_dev is doing is translating
the incoming bus / id / LUN to a LUN in the logical unit addressing format,
as defined in 4.6.9 of SAM-4.

--------------------------------------
|Bit | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
--------------------------------------
| n  | (10b) |       Target          |
--------------------------------------
| n+1|   Bus     |        LUN        |
--------------------------------------

So this means, in the current implementation, we have 6 bits for target (max=63),
3 bits for bus (max=7), and 5 bits for LUN (max=31). 

It might not be a bad idea to enforce these limits on the module parameters. Otherwise
we'll have a mess when we run through lun_from_dev...

As far as eliminating lun_from_dev and just passing the LUN through, I don't think we
can do that. The LUN list returned by Report LUNs to a VSCSI disk assumes we are
doing the translation above.

-Brian

-- 
Brian King
Power Linux I/O
IBM Linux Technology Center