On Mon, Oct 04, 2021 at 03:42:27PM +0200, Frederic Weisbecker wrote: > On Fri, Oct 01, 2021 at 06:51:08PM +0100, Valentin Schneider wrote: > > On 30/09/21 00:10, Frederic Weisbecker wrote: > > > The current condition to limit the number of callbacks executed in a > > > row checks the offloaded state of the rdp. Not only is it volatile > > > but it is also misleading: the rcu_core() may well be executing > > > callbacks concurrently with NOCB kthreads, and the offloaded state > > > would then be verified on both cases. As a result the limit would > > > spuriously not apply anymore on softirq while in the middle of > > > (de-)offloading process. > > > > > > Another issue with the condition is that rcu_is_callbacks_kthread() > > > doesn't check if we are actually running callbacks from rcuc itself or > > > from a softirq interrupting rcuc. > > > > > > > Doesn't rcutree.use_softirq imply rcuc is never woken, in which case > > RCU_SOFTIRQ can't interrupt rcuc (e.g. while run atop an IRQ exit)? > > I suppose during the (de)offload sequence we could have RCU_SOFTIRQ running > > atop the NOCB CB kthread, but that's not something > > rcu_is_callbacks_kthread() detects. > > Yes good point, I don't know if rcuc can be ever be interrupted by > irq_exit() -> do_softirq() -> rcu_core() itself after all. > > Paul can probably confirm your point? This is supposed to be prohibited by the fact that rcutree.use_softirq prevents the rcuc kthreads from being spawned in the first place. That said, I do not recall having thought through the de-offloading process with the rcuc kthreads in mind. There could easily be bugs in this case. (I -think- that it is OK because each rcuc kthread is confined to running on the corresponding CPU and because it disables BH. But it is the system's opinion that counts.) > > Also, why is rcu_is_callbacks_kthread() hardcoded to false for > > !CONFIG_RCU_BOOST? Isn't it relevant for do_rcu_batch() ratelimiting > > regardless (at least before your patches)? > > I believe rcuc is only used on CONFIG_RCU_BOOST? They are independent, though by default RT arranges for rcuc kthreads and CONFIG_RCU_BOOST. Now there are kthreads created for CONFIG_RCU_BOOST, but these are the rcub kthreads, and they are not per CPU, but rather per leaf rcu_node structure. Thanx, Paul
