qla2xxx driver/firmware versions & thermal stability

In the recent past we ran qla2300 driver version 6.04.00 with qla2300F cards. The driver/card often died with the following messages (kernel 2.4.19-pre10aa4):

Jan 12 12:54:56 prodapp4 kernel: qla2x00: Status Entry invalid handle.
Jan 12 12:54:56 prodapp4 kernel: qla2x00: Performing ISP error recovery - ha= c493c07c.
...
Jan 12 12:55:22 prodapp4 kernel: qla2x00(2): ISP error recovery failed - board disabled

These failures seem to be temperature related. A slight upward shift in the computer room temperature (from very cold to not-quite-as-cold) *always* precipitated these errors. After upgrading all the machines to qla2340 boards we still have the same problem.

Now, I'm quite aware that this could be caused by poor cooling circulation (2U Intel Server chassis / STL2 board) but my basic tests with temperature probes indicate nothing excessive - in fact everything seems chilly. The room is kept at 18degC, the server cases are cold to touch, internal case temperature around the cards of about 22degC. I have detailed 5-minute temperature data going back for a year - all of it well inside the Qlogic operating specs.

My question is: has anyone else seen anything like this? Does the older Qlogic firmware tend to run hotter? Is the newer any better? At the moment we're on 6.04.00 as this is the release that is supposedly certified by Hitachi (storage vendor). I'd be happy to upgrade to later version if I knew it would cure the thermal sensitivity.

boot probe with 2340 and 6.04 driver:

qla2x00_set_info starts at address = c3ac0060
qla2x00: Found VID=1077 DID=2312 SSVID=1077 SSDID=100
scsi(2): Found a QLA2312 @ bus 1, device 0xa, irq 20, iobase 0x6400 scsi(2): Allocated 4096 SRB(s).
scsi(2): Configure NVRAM parameters...
scsi(2): 32 Bit PCI Addressing Enabled.
scsi(2): Verifying loaded RISC code...
scsi(2): Verifying chip...
scsi(2): Waiting for LIP to complete...
scsi(2): LOOP UP detected.
scsi(2): Topology - (F_Port), Host Loop address 0xffff
scsi2 : QLogic QLA2312 PCI to Fibre Channel Host Adapter: bus 1 device 10 irq 20

       Firmware version:  3.01.18, Driver version 6.04.00

Vendor: HITACHI Model: DF600F Rev: 0000
Type: Direct-Access ANSI SCSI revision: 03

Could you report your findings to QLogic tech-support (www.qlogic.com/support), they'll be able to assist in troubleshooting.

Regards,
Andrew Vasquez
QLogic Corporation

:) Been there recently. In a nutshell: "it shouldn't happen, but that's quite an old driver, perhaps you could try upgrading"

Forgive me for not taking the advice immediately, but if you knew how long I'd been looking at this, how much equipment isolated/replaced and how many versions of the qla driver I build/tried *before* 6.04...

When my colleague raised this on lkml 12-18months ago (before we knew it was heat related) he had an off-list reply from someone who had spent a long time chasing the same error - and traced it to heat sensitivity (within operating spec).

I originially concluded that it was perhaps a bad batch of cards - hence replacing all the 2300s with 2340s.

I may eventually replace the driver as I have almost run out of options - but I was hoping for something a little more scientific after my past driver upgrades.