On Thu, 2023-08-17 at 09:38 +0200, Jinpu Wang wrote: > > This adds cond_resched() to the first two loops: releasing non-SECS pages and > > releasing SECS pages which don't have children on remote vepc instance(s). > > There is a third round loop to release SECS pages which have children on remote > > vepc instance(s). I think it's better to add cond_resched() there too, although > > it's extremely unlikely it could trigger softlockup w/o cond_resched(). > > I also think the third one are extremely unlikely could cause softlockup. > And our tests do not hit it either. If we only consider your tests, then I believe you even don't need the second cond_resched()? I think we should just add cond_resched() to all the three loops. No harm, and we don't need to worry about this anymore.
