hugepage: add sysctl for hugepage alloc and mig

euleros inclusion
category: feature
feature: etmem
bugzilla: 48246

-------------------------------------------------

Add /proc/sys/kernel/hugepage_pmem_allocall switch. Set 1 to allowed all
memory in pmem could alloc for hugepage. Set 0(default) hugepage alloc is
limited by zone watermark as usual.
Add /proc/sys/kernel/hugepage_mig_noalloc switch. Set 1 to forbid new
hugepage alloc in hugepage migration when hugepage in dest node runs
out. Set 0(default) to allow hugepage alloc in hugepage migration as
usual.
Signed-off-by: NKemeng Shi <shikemeng@huawei.com>
Reviewed-by: Nlouhongxiang <louhongxiang@huawei.com>
Signed-off-by: NZheng Zengkai <zhengzengkai@huawei.com>

include/linux/hugetlb.h

@@ -159,6 +159,9 @@ struct address_space *hugetlb_page_mapping_lock_write(struct page *hpage);
 extern int sysctl_hugetlb_shm_group;
 extern struct list_head huge_boot_pages;
 
+extern int sysctl_hugetlb_mig_noalloc;
+extern int sysctl_hugetlb_pmem_allocall;
+
 /* arch callbacks */
 
 pte_t *huge_pte_alloc(struct mm_struct *mm,

kernel/sysctl.c

@@ -3129,6 +3129,24 @@ static struct ctl_table vm_table[] = {
 		.extra2		= SYSCTL_ONE,
 	},
 #endif
+	{
+		.procname       = "hugepage_mig_noalloc",
+		.data           = &sysctl_hugetlb_mig_noalloc,
+		.maxlen         = sizeof(sysctl_hugetlb_mig_noalloc),
+		.mode           = 0600,
+		.proc_handler   = proc_dointvec_minmax,
+		.extra1         = SYSCTL_ZERO,
+		.extra2         = SYSCTL_ONE,
+	},
+	{
+		.procname       = "hugepage_pmem_allocall",
+		.data           = &sysctl_hugetlb_pmem_allocall,
+		.maxlen         = sizeof(sysctl_hugetlb_pmem_allocall),
+		.mode           = 0600,
+		.proc_handler   = proc_dointvec_minmax,
+		.extra1         = SYSCTL_ZERO,
+		.extra2         = SYSCTL_ONE,
+	},
 	{ }
 };
 

mm/hugetlb.c

@@ -79,6 +79,9 @@ DEFINE_SPINLOCK(hugetlb_lock);
 static int num_fault_mutexes;
 struct mutex *hugetlb_fault_mutex_table ____cacheline_aligned_in_smp;
 
+int sysctl_hugetlb_mig_noalloc;
+int sysctl_hugetlb_pmem_allocall;
+
 static inline bool PageHugeFreed(struct page *head)
 {
 	return page_private(head + 4) == -1UL;
@@ -1736,6 +1739,8 @@ static int alloc_pool_huge_page(struct hstate *h, nodemask_t *nodes_allowed,
 	gfp_t gfp_mask = htlb_alloc_mask(h) | __GFP_THISNODE;
 
 	for_each_node_mask_to_alloc(h, nr_nodes, node, nodes_allowed) {
+		if (is_node_pmem(node) && sysctl_hugetlb_pmem_allocall)
+			gfp_mask |= __GFP_MEMALLOC;
 		page = alloc_fresh_huge_page(h, gfp_mask, node, nodes_allowed,
 						node_alloc_noretry);
 		if (page)
@@ -1980,7 +1985,7 @@ struct page *alloc_huge_page_nodemask(struct hstate *h, int preferred_nid,
 		struct page *page;
 
 		page = dequeue_huge_page_nodemask(h, gfp_mask, preferred_nid, nmask);
-		if (page) {
+		if (page || sysctl_hugetlb_mig_noalloc) {
 			spin_unlock(&hugetlb_lock);
 			return page;
 		}