CX4 burned? - any way to recover?

[efschu3]

Last week, my colleague plugged a CX4 into a computer which wasnt actualy powerd off, it was in standby...

After that heroic move, the CX4 isnt been recognized neither by UEFI nor by OS...

Is there a way to bring some life back to this NIC?

 

[Freebsd1976]

try short the FNP jumper use paperclip , then boot to os see mst device in system . if it is then reflash firmware , otherwise cx4 card maybe bricked

 

[efschu3]

OK, cant find any pins on my card? You have a picture where to find the FNP Pins?

This does my card look like:

 

[altmind]

But PCIe is hot-pluggable. Not saying hoplug gonna work for everybody, but at least it should be safe.
I hot plugged pcie nic couple of times, got none fried.

 

[efschu3]

Just because the PCIe BUS is specified to be hotplugable, that does not mean, the card supports it (or the mainboard) This must be implemented by card manufactor and is not a must have.

 

[altmind]

> Not saying hoplug gonna work for everybody, but at least it should be safe.

 

[efschu3]

If mainboard has not implemented it, it can defenitly burn your device - btw. I dont think this was completly implemented on SandyBridge Plattform, at least for what happens in standby-hotplug. As I dont see any FNP around and bridging the wholes labeled JP2 (maybe these are the FNP?) didnt change the cards behaviour.

To sad to have a dead CX4

 

[Freebsd1976]

yes jp2 is fnp , then run mst status, is there have output?
if there has some output ,then it not brick

 

[efschu3]

Didnt check with mst, only lspci - will try again later.

By the way, if I plug the card Ingo a pcie slot which shares lanes with another slot, the device in the other slot isnt been recognized too ...

 

[edge]

But PCIe is hot-pluggable. Not saying hoplug gonna work for everybody, but at least it should be safe.
I hot plugged pcie nic couple of times, got none fried.

Pcie can be hot pluggable, but it is rarely implemented. HPE dropped hot plug in the mid 00's on the proliant line and relegated it to itanium systems. Hot plug went away for two reasons: the first and primary reason is economic - it adds about $20 a slot to implement which makes systems too expensive compared to non hot plug (you lose way too many rfp), the second reason is DC management: for the most part, DC staff didn't implement infrastructure so you could hot plug (wire tied cables without cable management arms, for example) and when they did, the tech didn't out of ignorance or an excess of caution. So, hot plug pcie inside the ISS server goes the way of the dodo.

 

[efschu3]

Didnt check with mst, only lspci - will try again later.

Doesnt work - but how could it, if kernel cant See any device xD

 

[WANg]

Doesnt work - but how could it, if kernel cant See any device xD

You sure it's the card and not the PCIe slot? Did you test it on another system?

 

[efschu3]

Ofc.

 

[WANg]

Ofc.

Well, in that case, your card is very likely fried, or at least, the fuse kicked in and turned off the card electrically.

Look at the card very carefully - see if you can find a feature (specifically an SMT fuse) that starts with the letter F (F2 on my ConnectX-3, might be F4 on the ConnectX-4). On each leg of the fuse there is usually a pin hole (that's the FNP mentioned by @Freebsd1976). Grab a multimeter, put it on continuity mode, and put one test lead on each side (to see if current is moving across the fuse). If it doesn't beep, then the fuse has blown. You'll likely need to borrow a magnifying glass to read the code off the fuse. If you can solder SMT parts, you might be able to look for a replacement fuse on mouser/digikey and swap it out. Or take advantage of the 2 pin-holes and solder in something that can act as a fuse.

If it beeps, well, you'll have to get a bench power supply, feed 12v into the PCIe power pin and troubleshoot the power plane with a multimeter. Depending on the time investment it might not be worth the hassle.