On 2020-02-01 at 11:28 Mike Kravetz wrote: >On 1/30/20 5:33 PM, Li Xinhai wrote: >> MPOL_MF_STRICT is used in mbind() for purposes: >> (1) MPOL_MF_STRICT is set alone without MPOL_MF_MOVE or MPOL_MF_MOVE_ALL, >> to check if there is misplaced page and return -EIO; >> (2) MPOL_MF_STRICT is set with MPOL_MF_MOVE or MPOL_MF_MOVE_ALL, to check >> if there is misplaced page which is failed to isolate, or page is >> success on isolate but failed to move, and return -EIO. >> >> For non hugepage mapping, (1) and (2) are implemented as expectation. >> For hugepage mapping, (1) is not implemented. And in (2), the part about >> failed to isolate and report -EIO is not implemented. >> >> This patch implements the missed parts for hugepage mapping. Benefits >> with it applied: >> - User space can apply same code logic to handle mbind() on hugepage and >> non hugepage mapping; >> - Reliably using MPOL_MF_STRICT alone to check whether there is misplaced >> page or not when bind policy on address range, especially for address >> range which contains both hugepage and non hugepage mapping. >> >> Analysis of potential impact on existing users: >> - For users who using MPOL_MF_STRICT alone, since mbind() don't report >> reliable answer about misplaced page, their existing code have to >> utilize other facilities, e.g. numa_maps of proc, to check misplaced >> page. After this patch applied, that dedicated checking will still work >> without been impacted; > >I do not agree with this characterization of impact to existing users. If >someone uses MPOL_MF_STRICT alone with hugetlb pages today, they will never >get EIO. After this patch, they will get EIO if the hugetlb pages do not >follow the policy. > Thanks, your point is correct on this specific case. Try to summarize cases that have been used as setting MPOL_MF_STRICT alone: - If user has not noticed that STRICT been ignored, and implemented code branch for failure, then that branch will now be reachable when page is misplaced. This should not be a problem (maybe bug in that user's branch would be revealed now). - If user has noticed that STRICT been ignored, and prepared to check misplaced pages by dedicated code (this case is that described in changelog, no impaction). - (*) If user has noticed that STRICT been ignored, and assumed mbind() would never return failure for misplaced pages (e.g., assert on it). Then, those users will be impacted, although their code logic seems weird. - if users didn't define code logic to handle failure on misplaced pages, no matter they noticed that STRICT been ignored or not, they can't be impacted. In my mind, (*) part is the only case impact to users. >Yes, this is the desired behavior after the change with updated documentation. >However, it is a potential change for existing users. I honestly do not >believe anyone will notice. But, it is a change in behavior. > >> - For users who using MPOL_MF_STRICT with MPOL_MF_MOVE or >> MPOL_MF_MOVE_ALL, the semantic about some pages could not be moved will >> not be changed by this patch, because failed to isolate and failed to >> move have same effects to users, so their existing code will not be >> impacted. >> >> In mbind man page, the note about 'MPOL_MF_STRICT is ignored on huge page >> mappings' can be removed after this patch is applied. >> >> Signed-off-by: Li Xinhai <lixinhai.lxh@xxxxxxxxx> >> Cc: Michal Hocko <mhocko@xxxxxxxx> >> Cc: Mike Kravetz <mike.kravetz@xxxxxxxxxx> >> Cc: Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx> >> Cc: linux-man <linux-man@xxxxxxxxxxxxxxx> >> --- >> Link to relevant discussion: >> https://lore.kernel.org/linux-mm/1578993378-10860-2-git-send-email-lixinhai.lxh@xxxxxxxxx/ >> > >Thanks for doing this and the commit message. I do not see any issues with >the code. Thanks for inspecting the code. > >I believe removing the special case for hugetlb pages in mbind is a good >thing. It is unfortunate that it will cause a change in behavior. My >belief is that this change will go unnoticed. Providing consistent >behavior that matches the (updated) documentation is better that maintaining >the current functionality into the future. > >-- >Mike Kravetz
