NC510C 10G looses connection

[Michael Kochiashvili]

We have two HP ProLiant DL580 G5 servers running RHEL5 ( Kernel: 2.6.18-8.el5) connected to switch with HP NC510C 10G Ethernet Adapter.
With driver nx_nic-4.0.516 server works few hours and looses connection - no traffic, no ping, but switch shows that port is up and connection established, from server side we see that link is up.
We trie to turn off tcp segmentation offload but no effect
Server logs:
Dec 2 00:02:04 tea kernel: NETDEV WATCHDOG: eth3: transmit timed out
Dec 2 00:02:24 tea last message repeated 2 times
Dec 2 00:02:24 tea kernel: nx_nic[eth3]: Load stored FW
Dec 2 00:02:26 tea kernel: nx_nic: File FW[nxromimg.bin] version[3.4.339:0]
Dec 2 00:02:32 tea kernel: nx_nic HP NC510C 10G Ethernet Adapter Board S/N MY91MK0161 Chip id 0x1
Dec 2 00:02:32 tea kernel: nx_nic[eth3]: NIC Link is up

We trie driver version nx_nic-4.0.406 and connection loose after 15 sec.

Board info:
Board Ser# : MY91MK0086
Chip Rev : C1
Board Type : 0xf (HP NC510C 10G Ethernet Adapter)
Firmware ver.: 3.4.339
Driver ver. : 4.0.516


Second DL580 server have no traffic and we don't see connection looses.

[Michal Kapalka (mikap)]

hi,

i found only this link regarding this nic on ITRC:

http://forums11.itrc.hp.com/service/forums/questionanswer.do?threadId=1255941

mikap

[Ivan Ferreira]

I solved this kind of problem by appending noapic to the kernel boot parameters in the /boot/grub/grub.conf file.

[Michael Kochiashvili]

Thank you Michal and Ivan,
Im trying now to turn checsums off. Next step will be appending NOAPIC to kernrl boot parameters

[rick jones]

Keep in mind that if you disable CKO (ChecKsum Offload) you will also disable TSO (Large Send Offload) which means you will have a NIC that will consume more CPU per KB transferred than a typical 1 GbE NIC (which tends to have CKO, but not TSO). If you have plenty of idle CPU, or were already sending mostly << MTU sized packets this may be a don't care, but otherwise, you may not like the effect on performance - at least when the NIC doesn't freeze on you :(

I think the driver you have will use the firmware it has rather than what is on the NIC, but it might be worthwhile making sure the NIC(s) are updated to the current firmware on the off chance the issues relate to things done based on the FW on the card before the FW with the driver overrules it.

[Michael Kochiashvili]

Turning off checksums and adding noapic to kernel boot gives no effect.

I trie to update FirmWare from 3.4.336 to 3.4.339 with CP011470.scexe script

Responce was:
Update firmware on eth3 : ROM 3.4.336 to 3.4.339
MAC 00:**:**:**:**:1C Upgrade SUCCESS
NIC firmware update complete.

but NIC information shows me old FirmWare version

MAC PCI-ID NIC
00:**:**:**:**:1C 4040-0002-103C-7048 HP NC510C PCIe 10 Gigabit Serve Adapter
(Installed) (Available) Interface WOL PXE NCSI
Image Version Image Version eth3 NS ON NS
---------------------------------------------------------------------------
ROM 3.4.336 ROM 3.4.339

So.... have no idea how to get NIC to work
Any idea?