[Linux-PowerEdge] PERC H730 issues (extended LD support, span element misdetection)

Linda A. Walsh dell at tlinx.org
Tue Sep 15 22:21:26 CDT 2015


Stephen Dowdy wrote:
> I have a Poweredge T630 with PERC H730 Adapter running Debian Jessie (8.1)
> 
> ​    ​
> # lspci -mmnn | grep PERC
> ​    ​
> 03:00.0 "RAID bus controller [0104]" "LSI Logic / Symbios Logic [1000]" 
> "MegaRAID SAS-3 3108 [Invader] [005d]" -r02 "Dell [1028]" "PERC H730 
> Adapter [1f43]"
> ​It has 18 4TB SATA disks.​
> 
> 
> ​Using stock kernel (3.16.0-4-amd64) or backports kernel 
> (4.1.0-0.bpo.1-amd64/megaraid_sas=06.806.08.00-rc1) i see multiple 
> 'dmesg' statements of form:
> 
>     ​[    2.798743] megasas:span 0 rowDataSize 10
> 
> ​RHEL has a paid/subscriber page on this that is not very helpful.  
> Basically, the claim is that this message indicates that the kernel and 
> the PERC firmware disagree on the #elements in a span.  The numbers are 
> printed as %x, so kernel thinks SPAN 0 has 16 elements in it, whereas, 
> the system has 18 4TB SATA drives in a single RAID6 disk group with 4 
> Logical Volumes of various sizes.  It is unclear to me if there is any 
> potential catastrophic (data destructive) end-result from this 
> discrepency, or any performance related issues, or if it's just an 
> informational message.
----
	Just a random opinion, but the 'parity' disks (of which RAID6
has 2) aren't usually considered 'data' disks.  So if you have 18 disks
in all, most linux utils that talk about RAID size talk about the width
of the DATA SPAN -- because that's what is important to have lined up
with various volume and disk structures.  

	I.e. say you have a 64K stripe unit x 16 data stripes.  If
you had that in a RAID5, it would be 17 disks as it is 18 w/RAID6.
The system needs 64Kx16=1MB alignment which most modern partition
managers formatters will default to.  Example with xfs, you can specify
a file system size in terms of "swidth+sunit" or su+sw -- su=# bytes
on 1 stripe, while sw = a multiplier of the stripe unit, usually the same
number of data disks in a RAID device (paraphrase of mkfs.xfs manpage).

	I.e.   Between that... and if the disks are working I wouldn't
be too concerned.  The controller pretty much takes care of the layout
of the RAID on disk -- not the linux kernel (unless you use the software
MD RAID driver), which I wouldn't suggest if you have a dedicated card.


> 
> But that's not all ...
> 
>     perccli /c0 show events type=sincereboot filter=warning
>     ...
>     Event Description: Host driver needs to be upgraded to enable 
> extended LD support
>     ...
> 
> ​(this message appears in the 3.16 kernel, which is why i tried out the 
> 4.1 kernel, in which the message does not appear.   I can't find any 
> documentation that defines exactly what "Extended LD support" for a 
> MegaRAID card actually entails, but it doesn't fix the span #elements 
> mis-detection issue.
----
	If you go to the docs & downloads for LSI's stuff,
http://www.lsi.com/support/pages/download-results.aspx
 or the new site @
http://www.avagotech.com/support/download-search

The README on the new firmware has a notation about portability
issues between that firmware and firmware for a 
240_VD_Feature_Limitations_Known_Issues.  Seems like there are
2 versions of the firmware.  So just a W-A-G on my part, but
it might be referring to firmware upgrades to support 240 VD's?
But that *really* is a guess.

Best of luck!
remember -- RAID6 is not a backup solution! ;-)



More information about the Linux-PowerEdge mailing list