EV4400 + RedHat3 -- LUNs only show up after an hp_rescan

[Bob_Vance]

(((((((
Brand new EVA400 with C700 Blade chassis w BL460c.
EVA4400 XCS updated
. from 0901003000
. to 0903003000
Firmware updated on the Blades using latest Firmware CD, 8.20.
Installed RedHat AS3u9 x86.
NOT using SecurePath
))))))


When I do a
# fdisk -l | grep sd

no FibreChannel LUNs show up.

NOTE: the device *name* actually exists, /dev/sda
it's just that 'fdisk' does not see it as a scsi device.


I have installed RedHat many times, x86 & x86_64, and all versions 3,4,5, on Proliant DL360s connected to MSA1000 & 1500 (and a couple years back, on an EVA4000) and never saw this problem.

((((
As a test of AS3u9, I just installed AS3u9 on a Proliant DL360 connected to a MSA1500 and it works fine, as expected.
))))

Doing an

# hp_rescan -a

fixes the situation and 'fdisk' will then see the SCSI devices.

BUT, a reboot breaks it again.

I added an 'hp_rescan -a' to '/etc/rc.sysinit', but that is obviously not a good solution.


ANY ideas why the OS would not be seeing/scanning the FC SCSI devices, by itself, without the hp_rescan ????


********
NOTES
********
1) VMware ESX 3.5.0 has no problem seeing LUNs after reboot

2) Win2k3 (the EVA Mgt server) also has no problems
seeing LUNs after reboot.

3) a 'cat /proc/scsi/'
DOES show the LUNs, so the driver sees them.

dsconnect ## cat /proc/scsi/qla2300/0
QLogic PCI to Fibre Channel Host Adapter for QMH2462:
Firmware version: 4.00.23, Driver version 7.07.05.02-fo
...
SCSI LUN Information:
(Id:Lun) * - indicates lun is not registered with the OS.
( 0: 0): Total reqs 1, Pending reqs 0, flags 0x2*, 0:0:83,
( 0: 1): Total reqs 0, Pending reqs 0, flags 0x0*, 0:0:83,


4) I was using the included Qlogic driver, v7.07.04b6.
But, using latest HP RH3 driver, 7.07.05.02,
made no difference, with or without failover option.


tks
bv

[Uwe Zessin · Accepted solution]

Bob,
from what I can see, RHE3 is not supported on the EVA4400 - minimum is RHE4 U6 and you need the 8.02.11 driver for the QMH2462.

[Bob_Vance]

Thanks, Uwe.

The customer is constrained by one of his Application vendors to RH3.
I see that the SAN Reference Design Guide shows that RH3 is supported for the 4000 but *not* for the 4400.

I would be surprised if that's the actual problem, though, since hp_rescan fixes it, and RH3 was supported and worked for the 4000 -- but I have seen far stranger things.

There is another app, to be installed on other Blades, that needs RH4, so I will try that tomorrow to see how that goes.

BTW, the QMH2462 driver page only has 8.01.07.25.
Where do I find 8.02.11 ?



bv

[Uwe Zessin]

The EVA4400 provides new device inquiry strings which the driver must recognize for a proper failover.

> 7.07.05.02-fo

You did apparently configure the QLogic built-in failover. Maybe (I'm just guessing) the code does not work properly during initial scans.

About the 8.02.11 driver -- I've just quoted what is in the latest compatibility list.
I have not checked whether it is available for download. Unfortunately, the publishing of new versions in the compatibility lists and the software kits is not good synchronized - I have seen this in other areas (MSA1500, MSA2012i), too :-(

[Bob_Vance]

We installed RedHat EL5 on another blade and it worked.
We are using HP driver 8.02.11-fo .

We are leaving the 'hp_rescan' modification to 'sysinit' in place for Redhat AS3.

tks
bv

[Thomas Callahan]

hp_rescan -a is simply telling the scsi_host bus to rescan.

But the devices are not found at initial bootup from init 1 to init 3/6, yet are found after another execution of the scsi_host rescan.

It sounds like the process during normal boot when scsi_devices are scanned is not actually executing the scsi-add-singledevice scsi midlayer commands.

Have you tried a different kernel? Is this by chance a custom compiled kernel?

[Bob_Vance]

Yeah, I know.

I checked the kernel config (xconfig) and the SCSI config looked fine.

I cannot explain it.

tks
bv

[Bob_Vance]

Oh, yeah. There is no other script that has "midlayer" in it, besides the
"/opt/hp/hp_fibreutils/probe-luns"
script.

The only other files were source & header files in the
/usr/src/linux/drivers
tree.

bv