Re: [PATCH] x86/PoD: defer nested P2M flushes

[Jan Beulich <jbeulich@suse.com>]

On 19.10.2021 12:39, Roger Pau Monné wrote:
> On Tue, Oct 19, 2021 at 10:19:57AM +0200, Jan Beulich wrote:
>> On 19.10.2021 10:17, Jan Beulich wrote:
>>> On 19.10.2021 10:09, Roger Pau Monné wrote:
>>>> On Mon, Oct 11, 2021 at 10:17:08AM +0200, Jan Beulich wrote:
>>>>> @@ -1229,8 +1242,9 @@ p2m_pod_demand_populate(struct p2m_domai
>>>>>          __trace_var(TRC_MEM_POD_POPULATE, 0, sizeof(t), &t);
>>>>>      }
>>>>>  
>>>>> -    pod_unlock(p2m);
>>>>> +    pod_unlock_and_flush(p2m);
>>>>>      return true;
>>>>> +
>>>>>  out_of_memory:
>>>>>      pod_unlock(p2m);
>>>>
>>>> Don't you need to set defer_nested_flush = false in the out_of_memory
>>>> label? (as you don't call pod_unlock_and_flush that would do it)
>>>
>>> Yes of course - thanks for spotting. I had pod_unlock_and_flush() here
>>> too initially, and when switching back I forgot to convert rather than
>>> just delete that.
>>
>> Oh, wait, that was on purpose: There's no point clearing the flag
>> when the next thing we do is invoke domain_crash(). If it wasn't
>> that way, I don't think I could avoid using pod_unlock_and_flush()
>> here as well.
> 
> Oh, I see. We would need to be careful if that domain crash is ever
> removed.

Well, I can change the call there as well. Doing so would seem preferable
over adding a respective comment there. I didn't do so originally because
as it stands this would be meaningless code churn. Let me know what you
think.

Jan