On Fri 23-10-15 19:36:30, Tejun Heo wrote: > Hello, Michal. > > On Fri, Oct 23, 2015 at 10:33:16AM +0200, Michal Hocko wrote: > > Ohh, OK I can see wq_worker_sleeping now. I've missed your point in > > other email, sorry about that. But now I am wondering whether this > > is an intended behavior. The documentation says: > > This is. > > > WQ_MEM_RECLAIM > > > > All wq which might be used in the memory reclaim paths _MUST_ > > have this flag set. The wq is guaranteed to have at least one > > execution context regardless of memory pressure. > > > > Which doesn't seem to be true currently, right? Now I can see your patch > > It is true. > > > to introduce WQ_IMMEDIATE but I am wondering which WQ_MEM_RECLAIM users > > could do without WQ_IMMEDIATE? I mean all current workers might be > > looping in the page allocator and it seems possible that WQ_MEM_RECLAIM > > work items might be waiting behind them so they cannot help to relieve > > the memory pressure. This doesn't sound right to me. Or I am completely > > confused and still fail to understand what is WQ_MEM_RECLAIM supposed to > > be used for. > > It guarantees that there always is enough execution resource to > execute a work item from that workqueue. OK, strictly speaking the rescuer is there but it is kind of pointless if it doesn't fire up and do a work. > The problem here is not lack > of execution resource but concurrency management misunderstanding the > situation. And this sounds like a bug to me. > This also can be fixed by teaching concurrency management > to be a bit smarter - e.g. if a work item is burning a lot of CPU > cycles continuously or pool hasn't finished a work item over a certain > amount of time, automatically ignore the in-flight work item for the > purpose of concurrency management; however, this sort of inter-work > item busy waits are so extremely rare and undesirable that I'm not > sure the added complexity would be worthwhile. Don't we have some IO related paths which would suffer from the same problem. I haven't checked all the WQ_MEM_RECLAIM users but from the name I would expect they _do_ participate in the reclaim and so they should be able to make a progress. Now if your new IMMEDIATE flag will guarantee that then I would argue that it should be implicit for WQ_MEM_RECLAIM otherwise we always risk a similar situation. What would be a counter argument for doing that? -- Michal Hocko SUSE Labs -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>