[PATCH V2 fix 5/6] mm: hugetlb: add a new function to allocate a new gigantic page
From: Vlastimil Babka <hidden>
Date: 2016-11-28 14:17:33
Also in:
linux-mm
On 11/16/2016 07:55 AM, Huang Shijie wrote:
quoted hunk ↗ jump to hunk
There are three ways we can allocate a new gigantic page: 1. When the NUMA is not enabled, use alloc_gigantic_page() to get the gigantic page. 2. The NUMA is enabled, but the vma is NULL. There is no memory policy we can refer to. So create a @nodes_allowed, initialize it with init_nodemask_of_mempolicy() or init_nodemask_of_node(). Then use alloc_fresh_gigantic_page() to get the gigantic page. 3. The NUMA is enabled, and the vma is valid. We can follow the memory policy of the @vma. Get @nodes_allowed by huge_nodemask(), and use alloc_fresh_gigantic_page() to get the gigantic page. Signed-off-by: Huang Shijie <redacted> --- Since the huge_nodemask() is changed, we have to change this function a little. --- mm/hugetlb.c | 63 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 63 insertions(+)diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 6995087..c33bddc 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c@@ -1502,6 +1502,69 @@ int dissolve_free_huge_pages(unsigned long start_pfn, unsigned long end_pfn) /* * There are 3 ways this can get called: + * + * 1. When the NUMA is not enabled, use alloc_gigantic_page() to get + * the gigantic page. + * + * 2. The NUMA is enabled, but the vma is NULL. + * Create a @nodes_allowed, and use alloc_fresh_gigantic_page() to get + * the gigantic page. + * + * 3. The NUMA is enabled, and the vma is valid. + * Use the @vma's memory policy. + * Get @nodes_allowed by huge_nodemask(), and use alloc_fresh_gigantic_page() + * to get the gigantic page. + */ +static struct page *__hugetlb_alloc_gigantic_page(struct hstate *h, + struct vm_area_struct *vma, unsigned long addr, int nid) +{ + NODEMASK_ALLOC(nodemask_t, nodes_allowed, GFP_KERNEL | __GFP_NORETRY);
What if the allocation fails and nodes_allowed is NULL? It might work fine now, but it's rather fragile, so I'd rather see an explicit check. BTW same thing applies to __nr_hugepages_store_common().
+ struct page *page = NULL;
+
+ /* Not NUMA */
+ if (!IS_ENABLED(CONFIG_NUMA)) {
+ if (nid == NUMA_NO_NODE)
+ nid = numa_mem_id();
+
+ page = alloc_gigantic_page(nid, huge_page_order(h));
+ if (page)
+ prep_compound_gigantic_page(page, huge_page_order(h));
+
+ NODEMASK_FREE(nodes_allowed);
+ return page;
+ }
+
+ /* NUMA && !vma */
+ if (!vma) {
+ if (nid == NUMA_NO_NODE) {
+ if (!init_nodemask_of_mempolicy(nodes_allowed)) {
+ NODEMASK_FREE(nodes_allowed);
+ nodes_allowed = &node_states[N_MEMORY];
+ }
+ } else if (nodes_allowed) {
+ init_nodemask_of_node(nodes_allowed, nid);
+ } else {
+ nodes_allowed = &node_states[N_MEMORY];
+ }
+
+ page = alloc_fresh_gigantic_page(h, nodes_allowed, true);
+
+ if (nodes_allowed != &node_states[N_MEMORY])
+ NODEMASK_FREE(nodes_allowed);
+
+ return page;
+ }
+
+ /* NUMA && vma */
+ if (huge_nodemask(vma, addr, nodes_allowed))
+ page = alloc_fresh_gigantic_page(h, nodes_allowed, true);
+
+ NODEMASK_FREE(nodes_allowed);
+ return page;
+}
+
+/*
+ * There are 3 ways this can get called:
* 1. With vma+addr: we use the VMA's memory policy
* 2. With !vma, but nid=NUMA_NO_NODE: We try to allocate a huge
* page from any node, and let the buddy allocator itself figure