Stock 2.6.4 with CONFIG_LBD=y and qla2xxx driver, reports wrong device size

Stock 2.6.4 with CONFIG_LBD=y and qla2xxx driver, reports wrong device size

[Terrence Martin]

I have a stock 2.6.4 kernel with CONFIG_LBD=y turned on in my kernel 
(confirmed via /proc/config.gz) but when the module is loaded I still get

SCSI device sdb: 191162368 512-byte hdwr sectors (97875 MB)
SCSI device sdb: drive cache: write through
 sdb:

I expect to get 4,287,645MB as that is what my Infortrend RAID array is 
set up for a partition size. Is there something, in addition to the LBD 
support in 2.6 that I need to do with the SCSI driver itself, in the 
case the qla2300 drivers that come with 2.6.4, to allow larger than 2TB 
block devices?

Platform is Intel Xeon dual with 2GB of RAM. Base OS is fedora core 1 
with changes to support 2.6 series kernels.

Smaller than 2TB file systems on the other hand work fine (aside from a 
strange performance issue where reads are not as fast as writes under 
both 2.4 and 2.6).

Thanks,

Terrence

Re: Stock 2.6.4 with CONFIG_LBD=y and qla2xxx driver, reports wrong device size

[James Bottomley]

On Wed, 2004-03-24 at 15:53, Terrence Martin wrote:
> I have a stock 2.6.4 kernel with CONFIG_LBD=y turned on in my kernel 
> (confirmed via /proc/config.gz) but when the module is loaded I still get
> 
> SCSI device sdb: 191162368 512-byte hdwr sectors (97875 MB)
> SCSI device sdb: drive cache: write through
>  sdb:
> 
> I expect to get 4,287,645MB as that is what my Infortrend RAID array is 
> set up for a partition size. Is there something, in addition to the LBD 
> support in 2.6 that I need to do with the SCSI driver itself, in the 
> case the qla2300 drivers that come with 2.6.4, to allow larger than 2TB 
> block devices?

This looks to be a failure of the device.  It's obviously reporting a
size smaller than it actually is for some reason.

The LDB code has been successfully tested with SCSI on >2TB arrays.

James

RE: Stock 2.6.4 with CONFIG_LBD=y and qla2xxx driver, reports wrongdevice size

[Guy]

I think you have reached a 4T limit somewhere.
If you take your expected size, 4287645M, subtract 4T you get a number very
close to your reported size!
 4287645M
-4194304M (2^20)
---------
=  93341M

If you can, make the device size 100,000M smaller.

The size is wrapping past zero, I think.
The limit could also be 2T or 1T.
You must try just under 4T, 2T and 1T until you find a size that works.

Once you determine the limit, split the array into equal parts smaller than
the limit.  Then create a RAID0 with the smaller parts.

I hope I made this clear!

This is just a guess!!!!!  But I think I am correct.

Guy


-----Original Message-----
From: linux-scsi-owner@vger.kernel.org
[mailto:linux-scsi-owner@vger.kernel.org] On Behalf Of James Bottomley
Sent: Wednesday, March 24, 2004 4:25 PM
To: Terrence Martin
Cc: SCSI Mailing List
Subject: Re: Stock 2.6.4 with CONFIG_LBD=y and qla2xxx driver, reports
wrongdevice size

On Wed, 2004-03-24 at 15:53, Terrence Martin wrote:
> I have a stock 2.6.4 kernel with CONFIG_LBD=y turned on in my kernel 
> (confirmed via /proc/config.gz) but when the module is loaded I still get
> 
> SCSI device sdb: 191162368 512-byte hdwr sectors (97875 MB)
> SCSI device sdb: drive cache: write through
>  sdb:
> 
> I expect to get 4,287,645MB as that is what my Infortrend RAID array is 
> set up for a partition size. Is there something, in addition to the LBD 
> support in 2.6 that I need to do with the SCSI driver itself, in the 
> case the qla2300 drivers that come with 2.6.4, to allow larger than 2TB 
> block devices?

This looks to be a failure of the device.  It's obviously reporting a
size smaller than it actually is for some reason.

The LDB code has been successfully tested with SCSI on >2TB arrays.

James


-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: Stock 2.6.4 with CONFIG_LBD=y and qla2xxx driver, reports wrongdevice size

[James Bottomley]

On Wed, 2004-03-24 at 16:53, Guy wrote:
> This is just a guess!!!!!  But I think I am correct.

The standards are very explicit about what READ CAPACITY(10) should
return for a >2TB device.  sd detects this and retries with a READ
CAPACITY(16).

There is no message in the log indicating this happened, so I think the
device reported an incorrect size to READ CAPACITY(10).  It may have
reported its size with bit 33 lopped off, but that's still a device
problem.

James

RE: Stock 2.6.4 with CONFIG_LBD=y and qla2xxx driver, reportswrongdevice size

[Guy]

Beyond my knowledge...  Wish I was you!  :)

But, if the device has a bug like "bit 33 lopped off".
If he splits the array into 2 luns/devices then it may be fine.
Then create the RAID0 with them.

Guy

-----Original Message-----
From: James Bottomley [mailto:James.Bottomley@SteelEye.com] 
Sent: Wednesday, March 24, 2004 5:01 PM
To: Guy
Cc: 'Terrence Martin'; 'SCSI Mailing List'
Subject: RE: Stock 2.6.4 with CONFIG_LBD=y and qla2xxx driver,
reportswrongdevice size

On Wed, 2004-03-24 at 16:53, Guy wrote:
> This is just a guess!!!!!  But I think I am correct.

The standards are very explicit about what READ CAPACITY(10) should
return for a >2TB device.  sd detects this and retries with a READ
CAPACITY(16).

There is no message in the log indicating this happened, so I think the
device reported an incorrect size to READ CAPACITY(10).  It may have
reported its size with bit 33 lopped off, but that's still a device
problem.

James

Re: Stock 2.6.4 with CONFIG_LBD=y and qla2xxx driver, reportswrongdevice size

[Terrence Martin]

Thanks Guy, I think I might just do that, create an MD device out of 3 
equal sized disks.

Where I may have gone wrong and gotten confused is that the Logical 
Drive limit is 64TB. However I have now read, and I assume it is still 
current, that each LUN is only 32bits limiting a LUN to 2TB. You then 
need some sort of additional RAID (hardware or software) presumably to 
map those LUNS to a single disk.

Linux RAID0 should handle this nicely I think.

Thanks

Terrence


Guy wrote:

>Beyond my knowledge...  Wish I was you!  :)
>
>But, if the device has a bug like "bit 33 lopped off".
>If he splits the array into 2 luns/devices then it may be fine.
>Then create the RAID0 with them.
>
>Guy
>
>-----Original Message-----
>From: James Bottomley [mailto:James.Bottomley@SteelEye.com] 
>Sent: Wednesday, March 24, 2004 5:01 PM
>To: Guy
>Cc: 'Terrence Martin'; 'SCSI Mailing List'
>Subject: RE: Stock 2.6.4 with CONFIG_LBD=y and qla2xxx driver,
>reportswrongdevice size
>
>On Wed, 2004-03-24 at 16:53, Guy wrote:
>  
>
>>This is just a guess!!!!!  But I think I am correct.
>>    
>>
>
>The standards are very explicit about what READ CAPACITY(10) should
>return for a >2TB device.  sd detects this and retries with a READ
>CAPACITY(16).
>
>There is no message in the log indicating this happened, so I think the
>device reported an incorrect size to READ CAPACITY(10).  It may have
>reported its size with bit 33 lopped off, but that's still a device
>problem.
>
>James
>
>
>
>  
>