Hi, Myron,
Thanks for your work and analysis.
> This leads me to believe that the mapping for the error status register,
> which is put in place by ghes_new(), is silently failing. The other
> possibility is that the mapping was un-mapped at some point.
> My current guess is that the mapping is failing due to the GAR in
> question residing within ACPI's NVS.
The ACPI's NVS works for ioremap when used by ACPI itself. So I guess it may be unmapped some place. One way to check that is to print out current mappings when we do mapping and unmapping in atomicio.c. But now I am in travelling, so I can only write the debug patch after returning to office in about 1 week. Do you have time to do that?
Best Regards,
Huang Ying
��.n��������+%������w��{.n�����{�����ܨ}���Ơz�j:+v�����w����ޙ��&�)ߡ�a����z�ޗ���ݢj��w�f
