Re: "invalid agent type: 1" in acpi/ghes, cper: Recognize and cache CXL Protocol errors

[Marc Herbert]

On 2025-07-28 09:25, Koralahalli Channabasappa, Smita wrote:

> On 7/22/2025 12:24 PM, Marc Herbert wrote:

>>    The code below triggers the error "invalid agent type: 1" in Intel
>> validation (internal issue 15018133056)
>>
>> It's not clear to anyone we asked why you did not include RCH_DP in
>> the `switch (prot_err->agent_type)` in cxl_cper_post_prot_err() below.
>>
>> I can see how RCH_DP is special in cxl_cper_PRINT_prot_err() and I can
>> even understand (despite my near-zero CPER knowledge) some of the
>> special cases there. But in cxl_cper_post_prot_err() here, it's not
>> clear why RCH_DP would be rejected. Could this be an oversight? If not,
>> a comment with a short explanation would not hurt.
>>
> 
> You're right. RCH_DP was excluded because it doesn’t report a valid
> SBDF in the CPER record. Instead, it provides only the RCRB base
> address.
> 
> I haven't thoroughly investigated whether SBDF can be reliably derived
> from the RCRB base. There might be a platform-specific mechanism for
> that, but at the time, it seemed non-trivial to implement. Introducing
> additional infrastructure solely to support RCH_DP felt like it was
> adding more complexity.
> 
> I agree that a brief comment explaining this rationale would help. I'm
> okay if you plan to include a fixup for this along with the one for
> the device serial number.

If I understood you correctly, I think a different error message
would be much better than a comment. Like this?

--- a/drivers/acpi/apei/ghes.c
+++ b/drivers/acpi/apei/ghes.c
@@ -730,6 +730,9 @@ static void cxl_cper_post_prot_err(struct cxl_cper_sec_prot_err *prot_err,
 		memcpy(&wd.ras_cap, cap_start, sizeof(wd.ras_cap));
 		wd.severity = cper_severity_to_aer(severity);
 		break;
+	case RCH_DP:
+		pr_err_ratelimited("CXL CPER agent type unsupported: RCH_DP\n");
+		return;
 	default:
 		pr_err_ratelimited("CXL CPER invalid agent type: %d\n",
 				   prot_err->agent_type);