ereport: DIMM failure

[hawkw]

PR #2174 adds code to the Cosmo SPD task to check for DIMM failures. After PR #2126 has merged, we should also generate an ereport for DIMM failures when one is detected.

[hawkw]

I presume there may be a comparable thing on Gimlet?

[labbott]

I presume there may be a comparable thing on Gimlet?

I think the answer might be no or much trickier than cosmo based on reading RFD 88. But we should check with hardware people to confirm.

[rmustacc]

I presume there may be a comparable thing on Gimlet?

Not quite. I think we need to be more specific here. Based on this being about #2174 this is specifically about DDR5 PMIC failures generally. I think this is especially true as I don't see the register or dies being connected to this pin, but need to dig a bit more into the reference designs.

In this case, this is more specifically a DIMM regulator failure. DIMM rails changed dramatically between DDR4 and DDR5. We basically have on-board rails that do generic step-down in DDR4 that are spread across a few devices. In DDR5 these all moved onto the device, hence the PMICs themselves. So if we're specific to power regulator failures, which is the main thing that can be detected then the manifestation is different in DDR4/Gimlet, though there is a dedicated hotswap per group of DIMMs in Cosmo that would want similar error handling.