Re: [PATCH 0/2] efi/cper, cxl: Decode CXL Protocol Errors CPER

[Jonathan Zhang (Infra)]

> On Oct 26, 2022, at 12:31 PM, Smita Koralahalli <Smita.KoralahalliChannabasappa@xxxxxxx> wrote:
> 
> On 10/25/2022 5:11 PM, Dan Williams wrote:
>> Smita Koralahalli wrote:
>>> Hi Dan,
>>> 
>>> On 10/21/2022 3:18 PM, Dan Williams wrote:
>>>> Hi Smita,
>>>> 
>>>> Smita Koralahalli wrote:
>>>>> This series adds decoding for the CXL Protocol Errors Common Platform
>>>>> Error Record.
>>>> Be sure to copy Ard Biesheuvel <ardb@xxxxxxxxxx>, added, on
>>>> drivers/firmware/efi/ patches.
>>>> 
>>>> Along those lines, drivers/cxl/ developers have an idea of what is
>>>> contained in the new CXL protocol error records and why Linux might want
>>>> to decode them, others from outside drivers/cxl/ might not. It always
>>>> helps to have a small summary of the benefit to end users of the
>>>> motivation to apply a patch set.
>>> Sure, will include in my v2.
>>> 
>>>>> Smita Koralahalli (2):
>>>>>    efi/cper, cxl: Decode CXL Protocol Error Section
>>>>>    efi/cper, cxl: Decode CXL Error Log
>>>>> 
>>>>>   drivers/firmware/efi/Makefile   |   2 +-
>>>>>   drivers/firmware/efi/cper.c     |   9 +++
>>>>>   drivers/firmware/efi/cper_cxl.c | 108 ++++++++++++++++++++++++++++++++
>>>>>   drivers/firmware/efi/cper_cxl.h |  58 +++++++++++++++++
>>>>>   include/linux/cxl_err.h         |  21 +++++++
>>>>>   5 files changed, 197 insertions(+), 1 deletion(-)
>>>> I notice no updates for the trace events in ghes_do_proc(), is that next
>>>> in your queue? That's ok to be a follow-on after v2.
>>> Sorry, if I haven't understood this right. Are you implying about the
>>> "handling"
>>> of cxl memory errors in ghes_do_proc() or is it just copying of CPER
>>> entries to
>>> tracepoints?
>> Right now ghes_do_proc() will let the CXL CPER records fall through to
>> log_non_standard_event(). Are you planning to add trace event decode
>> there for CPER_SEC_CXL_PROT_ERR records?
> 
> Thanks! Yeah its a good idea to add. I did not think about this before.
> I will send this as a separate patchset after v2.
> 
> I think with this cxl cper trace event support and Ira's patchset which traces
> specific event record types via Get Event Record, we can start the userspace
> handling probably in rasdaemon?
Yes, I think this makes sense. rasdaemon could aggregate data and provide user
with full picture:
* Memory errors from both processor attached memory and CXL memory.
* CXL protocol errors.
* CXL device errors.
Such errors may be handled either firmware first or OS first.

> 
>> 
>> I am not sure if the CXL CPER to trace record conversion belongs there,
>> or somewhere closer to trace_aer_event() invocations since the CXL
>> protocol errors are effectively an extenstion of PCI AER events.
> 
> Right, I will keep it simple in v1 and get the comments about the placement..
> 
> Thanks,
> Smita
> 
> 
>