> > I think there should be GFP_ATOMIC used, because it has more chance to > > return memory then GFP_NOWAIT. I see that Michal has same view on it. > > I don't think so because GFP_ATOMIC implies GFP_NOWAIT. I am Ok with keeping > the GFP_ATOMIC as it is btw. Paul mentioned he prefers this. I agree with > that as well. > GFP_ATOMIC can access to reserved memory whereas GFP_NOWAIT is not eligible to do so. So there is difference between them :) > > > > > > Yes, the benefit of the trace/warning is that the user can switch to a > > > non-headless API and avoid the synchronize_rcu(), that would help them get > > > faster kfree_rcu() performance instead of having silent slowdowns. > > > > > Agree. What about just adding WARN_ON_ONCE()? I am just thinking if it > > could be harmful or not. > > You mean WARN_ON_ONCE() before the synchronize_rcu() right? We could do that. > Paul mentioned to me he prefers if this new warning can be turned off with a > boot parameter since some future user may prefer no warning. I also agree. > Yes, we can add it before doing synchronize_rcu(). WARN_ON_ONCE() will emit only once the warning. I think that would be enough to pay an attention. > > If we add this then we can keep your __GFP_NOWARN flag with no additional GFP > flag changes. > We can also add __GFP_RETRY_MAYFAIL to GFP_ATOMIC to make it more tight. Basically your patch can be modified just adding that. > > > It also tells us whether the headless API is worth it in the long run, I > > > think it is worth it because we will likely never hit the synchronize_rcu() > > > failsafe. But if we hit it a lot, at least it wont happen silently. > > > > > Agree. > > > > > Paul was concerned about following scenario with hitting synchronize_rcu(): > > > 1. Consider a system under memory pressure. > > > 2. Consider some other subsystem X depending on another system Y which uses > > > kfree_rcu(). If Y doesn't complete the operation in time, X accumulates > > > more memory. > > > 3. Since kfree_rcu() on Y hits synchronize_rcu() a lot, it slows it down. > > > This causes X to further allocate memory, further causing a chain > > > reaction. > > > Paul, please correct me if I'm wrong. > > > > > I see your point and agree that in theory it can happen. So, we should > > make it more tight when it comes to rcu_head attachment logic. > > Right. Per discussion with Paul, we discussed that it is better if we > pre-allocate N number of array blocks per-CPU and use it for the cache. > Default for N being 1 and tunable with a boot parameter. I agree with this. > As discussed before, we can make use of memory pool API for such purpose. But i am not sure if it should be one pool per CPU or one pool per NR_CPUS, that would contain NR_CPUS * N pre-allocated blocks. > In current code, we have 1 cache page per CPU, but this is allocated only on > the first kvfree_rcu() request. So we could change this behavior as well to > make it pre-allocated. > > Does this all sound good to you? > I think that makes sense :) -- Vlad Rezki
