On Thu 24-11-16 12:10:08, Mikulas Patocka wrote: > > > On Thu, 24 Nov 2016, Michal Hocko wrote: [...] > > Please note that even > > GFP_NOWAIT allocations will wake up kspwad which should clean up that > > The mempool is also using GFP_NOIO allocations - so do you claim that it > should not use GFP_NOIO too?
No, I am not claiming that. The last time I have asked the throttling didn't seem to serious enough to cause any problems. If the memory reclaim throttling is serious enough then let's measure and evaluate it. > You should provide a clear API that the block device drivers should use to > allocate memory - not to apply band aid to vm throttling problems as they > are being discovered. This is easier said than done, I am afraid. We have been using GFP_NOIO in mempool for years and there were no major complains. -- Michal Hocko SUSE Labs
