On Mon, Apr 24, 2017 at 10:42:17AM +0200, Michal Hocko wrote: > On Sat 22-04-17 10:10:34, Stanislaw Gruszka wrote: > [...] > > > This whole special casing > > > of debug_guardpage_minorder is just too strange to me. We do have a rate > > > limit to not flood the log. > > > > I added this check to skip warning if buddy allocator fail, what I > > considered likely scenario taking the conditions. The check remove > > warning completely, rate limit only limit the speed warnings shows in > > logs. > > Yes and this is what I argue against. The feature limits the amount of > _usable_ memory and as such it changes the behavior of the allocator > which can lead to all sorts of problems (including high memory pressure, > stalls, OOM etc.). The warning is there to help debug all those > problems and removing it just changes that behavior in an unexpected > way. This is just wrong thing to do IMHO. Even worse so when it > motivates to make other code in the allocator more complicated. Allocation problems when using debug_guardpage_minorder should not be motivation to any mm change. This option is debug only (as name should suggest already). It purpose is to debug drivers/code that corrupt memory at random places, it is expected it will cause allocations problems. > If there > is really a problem logs flooded by the allocation failures while using > the guard page we should address it by a more strict ratelimiting. Ok, make sense. Stanislaw -- 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>
