On Thu 18-10-18 19:11:47, Andrew Morton wrote: > On Wed, 26 Sep 2018 16:22:27 +0200 Michal Hocko <mhocko@xxxxxxxxxx> wrote: > > > > MPOL_PREFERRED is handled by policy_node() before we call __alloc_pages_nodemask. > > > __GFP_THISNODE is applied only when we are not using > > > __GFP_DIRECT_RECLAIM which is handled in alloc_hugepage_direct_gfpmask > > > now. > > > Lastly MPOL_BIND wasn't handled explicitly but in the end the removed > > > late check would remove __GFP_THISNODE for it as well. So in the end we > > > are doing the same thing unless I miss something > > > > Forgot to add. One notable exception would be that the previous code > > would allow to hit > > WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE)); > > in policy_node if the requested node (e.g. cpu local one) was outside of > > the mbind nodemask. This is not possible now. We haven't heard about any > > such warning yet so it is unlikely that it happens though. > > Perhaps a changelog addition is needed to cover the above? : THP allocation mode is quite complex and it depends on the defrag : mode. This complexity is hidden in alloc_hugepage_direct_gfpmask from a : large part currently. The NUMA special casing (namely __GFP_THISNODE) is : however independent and placed in alloc_pages_vma currently. This both : adds an unnecessary branch to all vma based page allocation requests and : it makes the code more complex unnecessarily as well. Not to mention : that e.g. shmem THP used to do the node reclaiming unconditionally : regardless of the defrag mode until recently. This was not only : unexpected behavior but it was also hardly a good default behavior and I : strongly suspect it was just a side effect of the code sharing more than : a deliberate decision which suggests that such a layering is wrong. : : Moreover the oriinal code allowed to trigger : WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE)); : in policy_node if the requested node (e.g. cpu local one) was outside of : the mbind nodemask. This is not possible now. We haven't heard about any : such warning yet so it is unlikely that it happens but still a signal of : a wrong code layering. : : Get rid of the thp special casing from alloc_pages_vma and move the logic : to alloc_hugepage_direct_gfpmask. __GFP_THISNODE is applied to : the resulting gfp mask only when the direct reclaim is not requested and : when there is no explicit numa binding to preserve the current logic. : : This allows for removing alloc_hugepage_vma as well. Better? > I assume that David's mbind() concern has gone away. Either I've misunderstood it or it was not really a real issue. -- Michal Hocko SUSE Labs
