I have been having issues lately with the HiFn based crypto cards
locking up in 3.7 and 3.8.
They are usually fine but under some undefined load they lock up and
it seems rather random as to when it happens and how much load causes
it.
The cards are used to help out with a VPN between a few far flung
machines but they are all i386.
I've encountered this on two Soekris NET4501's and on a single Athlon
machine.
The only real clue is in the authlog where sshd reports:
sshd [####]: fatal: evp_crypt: EVP_Cipher failed
SSHD and isakmpd are both seeminly locked up but I can get into the
machine if I use the blowfish protocol which isn't supported on the
HiFn card thereby leading me to think there is a bug in the driver or
the card itself where it's not servicing an interrupt or is stuck
waiting for an interrupt which will never come.
The dmesg on the machines have the following line:
hifn0 at pci0 dev 13 function 0 "Hifn 7955/7954" rev 0x00: LZS 3DES
ARC4 MD5 SHA1 RNG AES PK, 32KB dram, irq 9
As well the cards in question are the VPN1401 (PCI) and VPN1411
(MiniPCI).
Since there is no kernel panic I'm sort of at a loss as to how to
track this down better.
As far as the kernels go, I am using 3.8_GENERIC on the Athlon and a
stripped (via flashdist) version of 3.8 on the NET4501's.
Again these lockups are always under some sort of load over the VPN
(VNC, file transfers ....) and are for the most part random.
Does anyone have any suggestions on how to track this down?
My current solution is just 'ssh somehost -c blowfish reboot' though
that is obviously far from optimal.
--
Sean
