On Wed 17-11-21 15:39:30, Neil Brown wrote: > > __GFP_ATOMIC serves little purpose. > It's main effect is to set ALLOC_HARDER which adds a few little boosts to > increase the chance of an allocation succeeding, one of which is to > lower the water-mark at which it will succeed. > > It is *always* paired with __GFP_HIGH which sets ALLOC_HIGH which also > adjusts this watermark. It is probable that other users of __GFP_HIGH > should benefit from the other little bonuses that __GFP_ATOMIC gets. While I like to see __GFP_ATOMIC going away I am not really sure about this particular part. We have 3 ways to get to memory reserves. One of thme is directly controlable by __GFP_HIGH and two are internal to the allocator to handle different situations - ALLOC_OOM is to help the oom victim to make a fwd progress and ALLOC_HARDER should be for contexts which cannot rely on the memory reclaim to continue. What is the point of having ALLOC_HIGH and ALLOC_HARDER if you just add both of them for __GFP_HIGH? I think you should be instead really get back to pre d0164adc89f6b and allow ALLOC_HARDER for requests which have neither of the reclaim allowed. That would require tweaking GFP_ATOMIC as well I suspect and drop __GFP_KSWAPD_RECLAIM. Or do something else. > __GFP_ATOMIC also gives a warning if used with __GFP_DIRECT_RECLAIM. > There is little point to this. We already get a might_sleep() warning > if __GFP_DIRECT_RECLAIM is set. I believe the point of the warning was to stop any abuse of an additional memory reserves for context which can reclaim and to spare those to interrupt handlers - which usually use GFP_ATOMIC. A lack of any reports suggests this hasn't happened and the warning can be dropped. Would be worth a patch on its own with this explanation. > __GFP_ATOMIC allows the "watermark_boost" to be side-stepped. It is > probable that testing ALLOC_HARDER is a better fit here. This has been introduced by f80b08fc44536 but I have to say that I haven't understood why this couldn't check for __GFP_DIRECT_RECLAIM or one ALLOC_$FOO boosters rather than __GFP_ATOMIC. Again something for a separate patch. > __GFP_ATOMIC is used by tegra-smmu.c to check if the allocation might > sleep. This should test __GFP_DIRECT_RECLAIM instead. Willy has already proposed a better alternative. Thanks! -- Michal Hocko SUSE Labs
