Slightly concerned about Apollo Lake erratum -- expected degradation and early failure.

[Cheddoleum]

I'm running Linux on a Sophos SG115v3 as a core router (SOHO), and not clear on whether I've got a ticking time bomb. The notification about an update to this particular CPU (e3940) was issued March 2019, my unit's build date is June 2019.

It'd be nice to imagine that any unit built 2-3 months after the notification would have the fix but I'm not that hopeful; My guess they're just sourcing boards from the likes of Lanner and that chip was soldered down months earlier. The uncorrected silicon's stepping is D0, the corrected one is given as F1. Linux /proc/cpuinfo and similar tools just says "stepping 9", which is not directly translatable.

Anyway, any ideas on how to confirm which one I have would be appreciated (preferably one that doesn't involve booting into another OS!)

I can replace this router with something else, but it'd be nice to certain whether I need to.

 

[Jorge Perez]

Use HWinfo

 

[Cheddoleum]

Use it how? It produces a very long data dump from a number of sources: but if it translates the CPUID info in a way that maps to intel stepping numbers, it's not in a way I'm able to spot.

Edit: Ah. You're referring to the Windows/DOS utility HWiNFO, not linux "hwinfo".

 

[RolloZ170]

there are two E3940.
x5-E3940 cpuid 506C9 "CPU-Z stepping 9"
x5-E3940 cpuid 506CA "CPU-Z stepping A" E0,F1
don't know if D0 and F1 is "9" or F1 is "A"

 

[Cheddoleum]

Thanks. Interesting that they have two versions, but I can find documents from 2018 for both of those, so neither of them would seem to be response to this erratum.

 

[RolloZ170]

can you remove HEatsink a look at the printing ?

 

[mattventura]

I'm also interested in this. I just bought a couple atom boxes, and they have stepping 10 (0xA).

 

[Cheddoleum]

can you remove HEatsink a look at the printing ?

Perhaps, but that's more trouble than booting a DOS usb live stick and running HWiNFO, so I'll try that first. Either way I need to configure a cold spare router on another multi-nic box that can reach the same switches; fortunately there's a candidate that's always on anyway so I should be able to get to it soon.

 

[ericloewe]

Of course it's the damned LPC bus again, god did they mess up that block...

 

[Cheddoleum]

Hmm. It's still a little unclear, but Intel seems to have mostly walked the warning back about 6 months later. But it's a little unclear to what extent a real problem exists, whether the firmware update fixes it (if I even have that firmware update), and the e-series isn't mentioned as such.

Clearly, as HAL 9000 said, the best plan is "to put the unit back into service, and wait for it to fail". I mean that went well for everyone, right?

 

[ericloewe]

whether the firmware update fixes it

If it's anything like the C2000 issue - and it sure sounds like it - it wouldn't. Not unless they can patch the CPU's internal boot ROM to bypass the affected pin and always boot from SPI.

 

[RolloZ170]

Summary of Changes from D-0 to F-1 Stepping:
Intel identified an issue with the Low Pin Count (LPC), Real Time Clock (RTC), USB2.0, SD Card, eMMC, and SDIO interfaces on Intel® Atom™ Processor E3900 Series resulting in degradation of these signals at a rate higher than Intel's quality goals after multiple years in service.