On 03/04/2020 17:18, Stefano Stabellini wrote:
On Fri, 3 Apr 2020, Marc Zyngier wrote:
> Doing what my patch here does might be OK until one of these guests
start to rely on ISACTIVER to be accurate. So far I have not seen any examples of it, but I agree it could happen, hence, it is risky.
I am only going to answer to this. This is not about *accuracy* but deadlock in your guest. I actually wrote a long e-mail on this thread explaining the possible deadlock.
It is not because you can't reproduce the deadlock that the dealock is not there. When are you going to stop dimissing real bug in your implementation?
Cheers, -- Julien Grall
