Trouble with unreliably resetting Phis?

[Rich_E_]

Hi all,

I've got two 5110P Xeon Phis in a system, and they seem to unreliably misbehave.

I do a micctrl -rw, and mic0 and mic1 reset, but fail to come up, and log errors like:

mic1: Resetting (Post Code ��)
mic1: Transition from state resetting to reset failed
MIC 1 RESETFAIL postcode �� -1

At which point, both mic0 and mic1 are no longer reachable, and will require a cold power cycle in order to appear again. (If I attempt another -rw from this state, the kernel panics waiting on the PCIe device - I can provide a trace if desired.)

I'm on kernel 2.6.32-358.18.1.el6.x86_64 on SL 6.4. I'm using mpss_gold_update_3, and have tried uninstalling and reinstalling it at least twice.

Having logged the information from micinfo, the cards are running the following FW:

Flash Version : 2.1.02.0375
SMC Firmware Version : 1.7.4172
SMC Boot Loader Version : NotAvailable
uOS Version : 2.6.38.8-gefd324e
Device Serial Number : ADKC25204520

Am I doing something silly, or am I in a strange position? (I've tried flashing the MICs with newer versions, but since the reset command often results in them not coming back, it's a problem, and it doesn't seem to work.)

[Sumedh_N_Intel]

Hi,

I am not sure of what build of the MPSS you are using. There were some issues related to failed resets that were fixed in the Build 3 of MPSS update 3.You could possibly try updating to this MPSS release. 

Also, it seems that you are using older versions of the flash and smc and did not update them when you updated the mpss. This could be a part of your problem.