Re: [PATCH v3 1/3] xen/x86: add nmi continuation framework

[Jan Beulich <jbeulich@xxxxxxxx>]

On 05.11.2020 11:22, Jürgen Groß wrote:
> On 20.10.20 15:33, Jan Beulich wrote:
>> On 16.10.2020 10:53, Juergen Gross wrote:
>>> Actions in NMI context are rather limited as e.g. locking is rather
>>> fragile.
>>>
>>> Add a generic framework to continue processing in normal interrupt
>>> context after leaving NMI processing.
>>>
>>> This is done by a high priority interrupt vector triggered via a
>>> self IPI from NMI context, which will then call the continuation
>>> function specified during NMI handling.
>>
>> I'm concerned by there being just a single handler allowed, when
>> the series already introduces two uses. A single NMI instance
>> may signal multiple things in one go. At the very least we then
>> need a priority, such that SERR could override oprofile.
> 
> A different approach could be not to introduce a generic interface,
> but to explicitly call the continuation handlers in the interrupt
> handler.
> 
> Instead of a function pointer, a parameter pointer and a busy
> indicator (probably another function pointer) per cpu, we'd need for
> now only a parameter value per cpu (for the oprofile case) and a
> global flag (for the SERR case).
> 
> The downside would be having to add additional fields for other
> use cases, but for now I think this could be the better way,
> especially as this would remove the theoretical case of multiple
> issues overwriting one another.

Yes, perhaps less abstraction is the better approach here, for now
at least. Perhaps give Andrew and Roger a little bit of time to
object before going down that route.

Jan