Hi Michal, Thanks for the review and ACKs to 1/5 and 2/5 patches. On Fri, Aug 06, 2021 at 03:35:48PM +0200, Michal Hocko wrote: > On Tue 03-08-21 13:59:20, Feng Tang wrote: > > From: Ben Widawsky <ben.widawsky@xxxxxxxxx> > > > > Implement the missing huge page allocation functionality while obeying > > the preferred node semantics. This is similar to the implementation > > for general page allocation, as it uses a fallback mechanism to try > > multiple preferred nodes first, and then all other nodes. > > > > [akpm: fix compling issue when merging with other hugetlb patch] > > [Thanks to 0day bot for catching the missing #ifdef CONFIG_NUMA issue] > > Link: https://lore.kernel.org/r/20200630212517.308045-12-ben.widawsky@xxxxxxxxx > > Suggested-by: Michal Hocko <mhocko@xxxxxxxx> > > Signed-off-by: Ben Widawsky <ben.widawsky@xxxxxxxxx> > > Co-developed-by: Feng Tang <feng.tang@xxxxxxxxx> > > Signed-off-by: Feng Tang <feng.tang@xxxxxxxxx> > > ifdefery is just ugly as hell. One way to get rid of that would be to > provide a mpol_is_preferred_many() wrapper and hide the CONFIG_NUMA in > mempolicy.h. I haven't checked but this might help to remove some other > ifdefery as well. > > I especially dislike the label hidden in the ifdef. You can get rid of > that by checking the page for NULL. Yes, the 'ifdef's were annoying to me too, and thanks for the suggestions. Following is the revised patch upon the suggestion. Thanks, Feng -------8<--------------------- >From fc30718c40f02ba5ea73456af49173e66b5032c1 Mon Sep 17 00:00:00 2001 From: Ben Widawsky <ben.widawsky@xxxxxxxxx> Date: Thu, 5 Aug 2021 23:01:11 -0400 Subject: [PATCH] mm/hugetlb: add support for mempolicy MPOL_PREFERRED_MANY Implement the missing huge page allocation functionality while obeying the preferred node semantics. This is similar to the implementation for general page allocation, as it uses a fallback mechanism to try multiple preferred nodes first, and then all other nodes. To avoid adding too many "#ifdef CONFIG_NUMA" check, add a helper function in mempolicy.h to check whether a mempolicy is MPOL_PREFERRED_MANY. [akpm: fix compling issue when merging with other hugetlb patch] [Thanks to 0day bot for catching the !CONFIG_NUMA compiling issue] [Michal Hocko: suggest to remove the #ifdef CONFIG_NUMA check] Link: https://lore.kernel.org/r/20200630212517.308045-12-ben.widawsky@xxxxxxxxx Link: https://lkml.kernel.org/r/1627970362-61305-4-git-send-email-feng.tang@xxxxxxxxx Suggested-by: Michal Hocko <mhocko@xxxxxxxx> Signed-off-by: Ben Widawsky <ben.widawsky@xxxxxxxxx> Co-developed-by: Feng Tang <feng.tang@xxxxxxxxx> Signed-off-by: Feng Tang <feng.tang@xxxxxxxxx> -- include/linux/mempolicy.h | 12 ++++++++++++ mm/hugetlb.c | 28 ++++++++++++++++++++++++---- 2 files changed, 36 insertions(+), 4 deletions(-) diff --git a/include/linux/mempolicy.h b/include/linux/mempolicy.h index 0117e1e..60d5e6c 100644 --- a/include/linux/mempolicy.h +++ b/include/linux/mempolicy.h @@ -187,6 +187,12 @@ extern void mpol_put_task_policy(struct task_struct *); extern bool numa_demotion_enabled; +static inline bool mpol_is_preferred_many(struct mempolicy *pol) +{ + return (pol->mode == MPOL_PREFERRED_MANY); +} + + #else struct mempolicy {}; @@ -297,5 +303,11 @@ static inline nodemask_t *policy_nodemask_current(gfp_t gfp) } #define numa_demotion_enabled false + +static inline bool mpol_is_preferred_many(struct mempolicy *pol) +{ + return false; +} + #endif /* CONFIG_NUMA */ #endif diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 95714fb..75ea8bc 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -1145,7 +1145,7 @@ static struct page *dequeue_huge_page_vma(struct hstate *h, unsigned long address, int avoid_reserve, long chg) { - struct page *page; + struct page *page = NULL; struct mempolicy *mpol; gfp_t gfp_mask; nodemask_t *nodemask; @@ -1166,7 +1166,17 @@ static struct page *dequeue_huge_page_vma(struct hstate *h, gfp_mask = htlb_alloc_mask(h); nid = huge_node(vma, address, gfp_mask, &mpol, &nodemask); - page = dequeue_huge_page_nodemask(h, gfp_mask, nid, nodemask); + + if (mpol_is_preferred_many(mpol)) { + page = dequeue_huge_page_nodemask(h, gfp_mask, nid, nodemask); + + /* Fallback to all nodes if page==NULL */ + nodemask = NULL; + } + + if (!page) + page = dequeue_huge_page_nodemask(h, gfp_mask, nid, nodemask); + if (page && !avoid_reserve && vma_has_reserves(vma, chg)) { SetHPageRestoreReserve(page); h->resv_huge_pages--; @@ -2147,9 +2157,19 @@ struct page *alloc_buddy_huge_page_with_mpol(struct hstate *h, nodemask_t *nodemask; nid = huge_node(vma, addr, gfp_mask, &mpol, &nodemask); - page = alloc_surplus_huge_page(h, gfp_mask, nid, nodemask, false); - mpol_cond_put(mpol); + if (mpol_is_preferred_many(mpol)) { + gfp_t gfp = gfp_mask | __GFP_NOWARN; + gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL); + page = alloc_surplus_huge_page(h, gfp, nid, nodemask, false); + + /* Fallback to all nodes if page==NULL */ + nodemask = NULL; + } + + if (!page) + page = alloc_surplus_huge_page(h, gfp_mask, nid, nodemask, false); + mpol_cond_put(mpol); return page; } -- 2.7.4
