On Thu, Apr 23, 2020 at 09:59:55PM +0200, Uladzislau Rezki wrote: > b) Double argument(with rcu_head) > This case we consider as it gets called from atomic context even though > it can be not. Why we consider such case as atomic: we just assume that. > The reason is to keep it simple, because it is not possible to detect whether > a current context is attomic or not(for all type of kernels), i mean the one > that calls kfree_rcu(). > > In this case we do not have synchronize_rcu() option. Instead we have an > object with rcu_head inside. If an allocation gets failed we just make > use of rcu_head inside the object, regular queuing. > > In this case we do not need to hard in order to obtain memory. Therefore > my question was to Johannes what is best way here. Since we decided to > minimize reclaiming, whereas GFP_NOWAIT wakes up kswapd if no memory. > GFP_ATOMIC also is not good, because for (b) we do not need to waste > it. Waking kswapd is fine, because it's a shared facility that doesn't just reclaim on your behalf but on behalf of a central goal: to get the freelist back to the watermarks. If they're low, somebody will sooner or later kick kswapd anyway to do exactly that. So unless you ask kswapd for a high order page that is unlikely to be needed by anybody else, you're only doing the inevitable.
