On 10/19/2024 3:32 AM, Dan Williams wrote:
> Alison Schofield wrote:
>>
>> + linux-cxl mailing list
>
> Thanks for forwarding...
>
>> On Fri, Oct 11, 2024 at 05:58:52PM +0800, Coly Li wrote:
>>> Hi list,
>>>
>>> Recently I have a report for a warning message from CXL subsystem,
>>> [ 48.142342] cxl_port port2: Couldn't locate the CXL.cache and CXL.mem
>>> capability array header.
>>> [ 48.144690] cxl_port port3: Couldn't locate the CXL.cache and CXL.mem
>>> capability array header.
>>> [ 48.144704] cxl_port port3: HDM decoder capability not found
>>> [ 48.144850] cxl_port port4: Couldn't locate the CXL.cache and CXL.mem
>>> capability array header.
>>> [ 48.144859] cxl_port port4: HDM decoder capability not found
>>> [ 48.170374] cxl_port port6: Couldn't locate the CXL.cache and CXL.mem
>>> capability array header.
>>> [ 48.172893] cxl_port port7: Couldn't locate the CXL.cache and CXL.mem
>>> capability array header.
>>> [ 48.174689] cxl_port port7: HDM decoder capability not found
>>> [ 48.175091] cxl_port port8: Couldn't locate the CXL.cache and CXL.mem
>>> capability array header.
>>> [ 48.175105] cxl_port port8: HDM decoder capability not found
>>>
>>> After checking the source code I realize this is not a real bug,
>>> just a warning message that expected device was not detected. But
>>> from the above warning information itself, users/customers are
>>> worried there is something wrong (IMHO indeed not).
>>>
>>> Is there any chance that we can improve the code logic that only
>>> printing out the warning message when it is really a problem to be
>>> noticed?
>
> There is a short term fix and a long term fix. The short term fix could
> be to just delete the warning message, or downgrade it to dev_dbg(), for
> now since it is more often a false positive than not. The long term fix,
> and the logic needed to resolve false-positive reports, is to flip the
> capability discovery until *after* it is clear that there is a
> downstream endpoint capable of CXL.cachemem.
>
> Without an endpoint there is no point in reporting that a potentially
> CXL capable port is missing cachemem registers.
>
> So, if you want to send a patch changing that warning to dev_dbg() for
> now I would support that.
>
I noticed the short term solution been merged, may I know if anyone is working
on the long term solution? If not, I can work on it.
Thanks
Ming
