Linux Audio

Check our new training course

Loading...
v5.14.15
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * Free some vmemmap pages of HugeTLB
  4 *
  5 * Copyright (c) 2020, Bytedance. All rights reserved.
  6 *
  7 *     Author: Muchun Song <songmuchun@bytedance.com>
  8 *
  9 * The struct page structures (page structs) are used to describe a physical
 10 * page frame. By default, there is a one-to-one mapping from a page frame to
 11 * it's corresponding page struct.
 12 *
 13 * HugeTLB pages consist of multiple base page size pages and is supported by
 14 * many architectures. See hugetlbpage.rst in the Documentation directory for
 15 * more details. On the x86-64 architecture, HugeTLB pages of size 2MB and 1GB
 16 * are currently supported. Since the base page size on x86 is 4KB, a 2MB
 17 * HugeTLB page consists of 512 base pages and a 1GB HugeTLB page consists of
 18 * 4096 base pages. For each base page, there is a corresponding page struct.
 19 *
 20 * Within the HugeTLB subsystem, only the first 4 page structs are used to
 21 * contain unique information about a HugeTLB page. __NR_USED_SUBPAGE provides
 22 * this upper limit. The only 'useful' information in the remaining page structs
 23 * is the compound_head field, and this field is the same for all tail pages.
 24 *
 25 * By removing redundant page structs for HugeTLB pages, memory can be returned
 26 * to the buddy allocator for other uses.
 27 *
 28 * Different architectures support different HugeTLB pages. For example, the
 29 * following table is the HugeTLB page size supported by x86 and arm64
 30 * architectures. Because arm64 supports 4k, 16k, and 64k base pages and
 31 * supports contiguous entries, so it supports many kinds of sizes of HugeTLB
 32 * page.
 33 *
 34 * +--------------+-----------+-----------------------------------------------+
 35 * | Architecture | Page Size |                HugeTLB Page Size              |
 36 * +--------------+-----------+-----------+-----------+-----------+-----------+
 37 * |    x86-64    |    4KB    |    2MB    |    1GB    |           |           |
 38 * +--------------+-----------+-----------+-----------+-----------+-----------+
 39 * |              |    4KB    |   64KB    |    2MB    |    32MB   |    1GB    |
 40 * |              +-----------+-----------+-----------+-----------+-----------+
 41 * |    arm64     |   16KB    |    2MB    |   32MB    |     1GB   |           |
 42 * |              +-----------+-----------+-----------+-----------+-----------+
 43 * |              |   64KB    |    2MB    |  512MB    |    16GB   |           |
 44 * +--------------+-----------+-----------+-----------+-----------+-----------+
 45 *
 46 * When the system boot up, every HugeTLB page has more than one struct page
 47 * structs which size is (unit: pages):
 48 *
 49 *    struct_size = HugeTLB_Size / PAGE_SIZE * sizeof(struct page) / PAGE_SIZE
 50 *
 51 * Where HugeTLB_Size is the size of the HugeTLB page. We know that the size
 52 * of the HugeTLB page is always n times PAGE_SIZE. So we can get the following
 53 * relationship.
 54 *
 55 *    HugeTLB_Size = n * PAGE_SIZE
 56 *
 57 * Then,
 58 *
 59 *    struct_size = n * PAGE_SIZE / PAGE_SIZE * sizeof(struct page) / PAGE_SIZE
 60 *                = n * sizeof(struct page) / PAGE_SIZE
 61 *
 62 * We can use huge mapping at the pud/pmd level for the HugeTLB page.
 63 *
 64 * For the HugeTLB page of the pmd level mapping, then
 65 *
 66 *    struct_size = n * sizeof(struct page) / PAGE_SIZE
 67 *                = PAGE_SIZE / sizeof(pte_t) * sizeof(struct page) / PAGE_SIZE
 68 *                = sizeof(struct page) / sizeof(pte_t)
 69 *                = 64 / 8
 70 *                = 8 (pages)
 71 *
 72 * Where n is how many pte entries which one page can contains. So the value of
 73 * n is (PAGE_SIZE / sizeof(pte_t)).
 74 *
 75 * This optimization only supports 64-bit system, so the value of sizeof(pte_t)
 76 * is 8. And this optimization also applicable only when the size of struct page
 77 * is a power of two. In most cases, the size of struct page is 64 bytes (e.g.
 78 * x86-64 and arm64). So if we use pmd level mapping for a HugeTLB page, the
 79 * size of struct page structs of it is 8 page frames which size depends on the
 80 * size of the base page.
 81 *
 82 * For the HugeTLB page of the pud level mapping, then
 83 *
 84 *    struct_size = PAGE_SIZE / sizeof(pmd_t) * struct_size(pmd)
 85 *                = PAGE_SIZE / 8 * 8 (pages)
 86 *                = PAGE_SIZE (pages)
 87 *
 88 * Where the struct_size(pmd) is the size of the struct page structs of a
 89 * HugeTLB page of the pmd level mapping.
 90 *
 91 * E.g.: A 2MB HugeTLB page on x86_64 consists in 8 page frames while 1GB
 92 * HugeTLB page consists in 4096.
 93 *
 94 * Next, we take the pmd level mapping of the HugeTLB page as an example to
 95 * show the internal implementation of this optimization. There are 8 pages
 96 * struct page structs associated with a HugeTLB page which is pmd mapped.
 97 *
 98 * Here is how things look before optimization.
 99 *
100 *    HugeTLB                  struct pages(8 pages)         page frame(8 pages)
101 * +-----------+ ---virt_to_page---> +-----------+   mapping to   +-----------+
102 * |           |                     |     0     | -------------> |     0     |
103 * |           |                     +-----------+                +-----------+
104 * |           |                     |     1     | -------------> |     1     |
105 * |           |                     +-----------+                +-----------+
106 * |           |                     |     2     | -------------> |     2     |
107 * |           |                     +-----------+                +-----------+
108 * |           |                     |     3     | -------------> |     3     |
109 * |           |                     +-----------+                +-----------+
110 * |           |                     |     4     | -------------> |     4     |
111 * |    PMD    |                     +-----------+                +-----------+
112 * |   level   |                     |     5     | -------------> |     5     |
113 * |  mapping  |                     +-----------+                +-----------+
114 * |           |                     |     6     | -------------> |     6     |
115 * |           |                     +-----------+                +-----------+
116 * |           |                     |     7     | -------------> |     7     |
117 * |           |                     +-----------+                +-----------+
118 * |           |
119 * |           |
120 * |           |
121 * +-----------+
122 *
123 * The value of page->compound_head is the same for all tail pages. The first
124 * page of page structs (page 0) associated with the HugeTLB page contains the 4
125 * page structs necessary to describe the HugeTLB. The only use of the remaining
126 * pages of page structs (page 1 to page 7) is to point to page->compound_head.
127 * Therefore, we can remap pages 2 to 7 to page 1. Only 2 pages of page structs
128 * will be used for each HugeTLB page. This will allow us to free the remaining
129 * 6 pages to the buddy allocator.
130 *
131 * Here is how things look after remapping.
132 *
133 *    HugeTLB                  struct pages(8 pages)         page frame(8 pages)
134 * +-----------+ ---virt_to_page---> +-----------+   mapping to   +-----------+
135 * |           |                     |     0     | -------------> |     0     |
136 * |           |                     +-----------+                +-----------+
137 * |           |                     |     1     | -------------> |     1     |
138 * |           |                     +-----------+                +-----------+
139 * |           |                     |     2     | ----------------^ ^ ^ ^ ^ ^
140 * |           |                     +-----------+                   | | | | |
141 * |           |                     |     3     | ------------------+ | | | |
142 * |           |                     +-----------+                     | | | |
143 * |           |                     |     4     | --------------------+ | | |
144 * |    PMD    |                     +-----------+                       | | |
145 * |   level   |                     |     5     | ----------------------+ | |
146 * |  mapping  |                     +-----------+                         | |
147 * |           |                     |     6     | ------------------------+ |
148 * |           |                     +-----------+                           |
149 * |           |                     |     7     | --------------------------+
150 * |           |                     +-----------+
151 * |           |
152 * |           |
153 * |           |
154 * +-----------+
155 *
156 * When a HugeTLB is freed to the buddy system, we should allocate 6 pages for
157 * vmemmap pages and restore the previous mapping relationship.
158 *
159 * For the HugeTLB page of the pud level mapping. It is similar to the former.
160 * We also can use this approach to free (PAGE_SIZE - 2) vmemmap pages.
161 *
162 * Apart from the HugeTLB page of the pmd/pud level mapping, some architectures
163 * (e.g. aarch64) provides a contiguous bit in the translation table entries
164 * that hints to the MMU to indicate that it is one of a contiguous set of
165 * entries that can be cached in a single TLB entry.
166 *
167 * The contiguous bit is used to increase the mapping size at the pmd and pte
168 * (last) level. So this type of HugeTLB page can be optimized only when its
169 * size of the struct page structs is greater than 2 pages.
170 */
171#define pr_fmt(fmt)	"HugeTLB: " fmt
172
 
 
 
 
 
173#include "hugetlb_vmemmap.h"
174
175/*
176 * There are a lot of struct page structures associated with each HugeTLB page.
177 * For tail pages, the value of compound_head is the same. So we can reuse first
178 * page of tail page structures. We map the virtual addresses of the remaining
179 * pages of tail page structures to the first tail page struct, and then free
180 * these page frames. Therefore, we need to reserve two pages as vmemmap areas.
 
 
 
181 */
182#define RESERVE_VMEMMAP_NR		2U
183#define RESERVE_VMEMMAP_SIZE		(RESERVE_VMEMMAP_NR << PAGE_SHIFT)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
184
185bool hugetlb_free_vmemmap_enabled = IS_ENABLED(CONFIG_HUGETLB_PAGE_FREE_VMEMMAP_DEFAULT_ON);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
186
187static int __init early_hugetlb_free_vmemmap_param(char *buf)
 
 
 
188{
189	/* We cannot optimize if a "struct page" crosses page boundaries. */
190	if ((!is_power_of_2(sizeof(struct page)))) {
191		pr_warn("cannot free vmemmap pages because \"struct page\" crosses page boundaries\n");
 
 
 
 
192		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
193	}
 
194
195	if (!buf)
196		return -EINVAL;
 
 
 
 
197
198	if (!strcmp(buf, "on"))
199		hugetlb_free_vmemmap_enabled = true;
200	else if (!strcmp(buf, "off"))
201		hugetlb_free_vmemmap_enabled = false;
202	else
203		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
204
205	return 0;
206}
207early_param("hugetlb_free_vmemmap", early_hugetlb_free_vmemmap_param);
208
209static inline unsigned long free_vmemmap_pages_size_per_hpage(struct hstate *h)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
210{
211	return (unsigned long)free_vmemmap_pages_per_hpage(h) << PAGE_SHIFT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
212}
213
214/*
215 * Previously discarded vmemmap pages will be allocated and remapping
216 * after this function returns zero.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
217 */
218int alloc_huge_page_vmemmap(struct hstate *h, struct page *head)
 
219{
220	int ret;
221	unsigned long vmemmap_addr = (unsigned long)head;
222	unsigned long vmemmap_end, vmemmap_reuse;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
223
224	if (!HPageVmemmapOptimized(head))
225		return 0;
226
227	vmemmap_addr += RESERVE_VMEMMAP_SIZE;
228	vmemmap_end = vmemmap_addr + free_vmemmap_pages_size_per_hpage(h);
229	vmemmap_reuse = vmemmap_addr - PAGE_SIZE;
 
230	/*
231	 * The pages which the vmemmap virtual address range [@vmemmap_addr,
232	 * @vmemmap_end) are mapped to are freed to the buddy allocator, and
233	 * the range is mapped to the page which @vmemmap_reuse is mapped to.
234	 * When a HugeTLB page is freed to the buddy allocator, previously
235	 * discarded vmemmap pages must be allocated and remapping.
236	 */
237	ret = vmemmap_remap_alloc(vmemmap_addr, vmemmap_end, vmemmap_reuse,
238				  GFP_KERNEL | __GFP_NORETRY | __GFP_THISNODE);
239
240	if (!ret)
241		ClearHPageVmemmapOptimized(head);
 
 
242
243	return ret;
244}
245
246void free_huge_page_vmemmap(struct hstate *h, struct page *head)
 
247{
248	unsigned long vmemmap_addr = (unsigned long)head;
249	unsigned long vmemmap_end, vmemmap_reuse;
250
251	if (!free_vmemmap_pages_per_hpage(h))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
252		return;
253
254	vmemmap_addr += RESERVE_VMEMMAP_SIZE;
255	vmemmap_end = vmemmap_addr + free_vmemmap_pages_size_per_hpage(h);
256	vmemmap_reuse = vmemmap_addr - PAGE_SIZE;
 
 
257
258	/*
259	 * Remap the vmemmap virtual address range [@vmemmap_addr, @vmemmap_end)
260	 * to the page which @vmemmap_reuse is mapped to, then free the pages
261	 * which the range [@vmemmap_addr, @vmemmap_end] is mapped to.
262	 */
263	if (!vmemmap_remap_free(vmemmap_addr, vmemmap_end, vmemmap_reuse))
 
 
264		SetHPageVmemmapOptimized(head);
265}
266
267void __init hugetlb_vmemmap_init(struct hstate *h)
268{
269	unsigned int nr_pages = pages_per_huge_page(h);
270	unsigned int vmemmap_pages;
 
 
 
 
 
 
271
272	/*
273	 * There are only (RESERVE_VMEMMAP_SIZE / sizeof(struct page)) struct
274	 * page structs that can be used when CONFIG_HUGETLB_PAGE_FREE_VMEMMAP,
275	 * so add a BUILD_BUG_ON to catch invalid usage of the tail struct page.
276	 */
277	BUILD_BUG_ON(__NR_USED_SUBPAGE >=
278		     RESERVE_VMEMMAP_SIZE / sizeof(struct page));
279
280	if (!hugetlb_free_vmemmap_enabled)
281		return;
282
283	vmemmap_pages = (nr_pages * sizeof(struct page)) >> PAGE_SHIFT;
284	/*
285	 * The head page and the first tail page are not to be freed to buddy
286	 * allocator, the other pages will map to the first tail page, so they
287	 * can be freed.
288	 *
289	 * Could RESERVE_VMEMMAP_NR be greater than @vmemmap_pages? It is true
290	 * on some architectures (e.g. aarch64). See Documentation/arm64/
291	 * hugetlbpage.rst for more details.
292	 */
293	if (likely(vmemmap_pages > RESERVE_VMEMMAP_NR))
294		h->nr_free_vmemmap_pages = vmemmap_pages - RESERVE_VMEMMAP_NR;
295
296	pr_info("can free %d vmemmap pages for %s\n", h->nr_free_vmemmap_pages,
297		h->name);
298}
v6.2
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * HugeTLB Vmemmap Optimization (HVO)
  4 *
  5 * Copyright (c) 2020, ByteDance. All rights reserved.
  6 *
  7 *     Author: Muchun Song <songmuchun@bytedance.com>
  8 *
  9 * See Documentation/mm/vmemmap_dedup.rst
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 10 */
 11#define pr_fmt(fmt)	"HugeTLB: " fmt
 12
 13#include <linux/pgtable.h>
 14#include <linux/moduleparam.h>
 15#include <linux/bootmem_info.h>
 16#include <asm/pgalloc.h>
 17#include <asm/tlbflush.h>
 18#include "hugetlb_vmemmap.h"
 19
 20/**
 21 * struct vmemmap_remap_walk - walk vmemmap page table
 22 *
 23 * @remap_pte:		called for each lowest-level entry (PTE).
 24 * @nr_walked:		the number of walked pte.
 25 * @reuse_page:		the page which is reused for the tail vmemmap pages.
 26 * @reuse_addr:		the virtual address of the @reuse_page page.
 27 * @vmemmap_pages:	the list head of the vmemmap pages that can be freed
 28 *			or is mapped from.
 29 */
 30struct vmemmap_remap_walk {
 31	void			(*remap_pte)(pte_t *pte, unsigned long addr,
 32					     struct vmemmap_remap_walk *walk);
 33	unsigned long		nr_walked;
 34	struct page		*reuse_page;
 35	unsigned long		reuse_addr;
 36	struct list_head	*vmemmap_pages;
 37};
 38
 39static int __split_vmemmap_huge_pmd(pmd_t *pmd, unsigned long start)
 40{
 41	pmd_t __pmd;
 42	int i;
 43	unsigned long addr = start;
 44	struct page *page = pmd_page(*pmd);
 45	pte_t *pgtable = pte_alloc_one_kernel(&init_mm);
 46
 47	if (!pgtable)
 48		return -ENOMEM;
 49
 50	pmd_populate_kernel(&init_mm, &__pmd, pgtable);
 51
 52	for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE) {
 53		pte_t entry, *pte;
 54		pgprot_t pgprot = PAGE_KERNEL;
 55
 56		entry = mk_pte(page + i, pgprot);
 57		pte = pte_offset_kernel(&__pmd, addr);
 58		set_pte_at(&init_mm, addr, pte, entry);
 59	}
 60
 61	spin_lock(&init_mm.page_table_lock);
 62	if (likely(pmd_leaf(*pmd))) {
 63		/*
 64		 * Higher order allocations from buddy allocator must be able to
 65		 * be treated as indepdenent small pages (as they can be freed
 66		 * individually).
 67		 */
 68		if (!PageReserved(page))
 69			split_page(page, get_order(PMD_SIZE));
 70
 71		/* Make pte visible before pmd. See comment in pmd_install(). */
 72		smp_wmb();
 73		pmd_populate_kernel(&init_mm, pmd, pgtable);
 74		flush_tlb_kernel_range(start, start + PMD_SIZE);
 75	} else {
 76		pte_free_kernel(&init_mm, pgtable);
 77	}
 78	spin_unlock(&init_mm.page_table_lock);
 79
 80	return 0;
 81}
 82
 83static int split_vmemmap_huge_pmd(pmd_t *pmd, unsigned long start)
 84{
 85	int leaf;
 86
 87	spin_lock(&init_mm.page_table_lock);
 88	leaf = pmd_leaf(*pmd);
 89	spin_unlock(&init_mm.page_table_lock);
 90
 91	if (!leaf)
 92		return 0;
 93
 94	return __split_vmemmap_huge_pmd(pmd, start);
 95}
 96
 97static void vmemmap_pte_range(pmd_t *pmd, unsigned long addr,
 98			      unsigned long end,
 99			      struct vmemmap_remap_walk *walk)
100{
101	pte_t *pte = pte_offset_kernel(pmd, addr);
102
103	/*
104	 * The reuse_page is found 'first' in table walk before we start
105	 * remapping (which is calling @walk->remap_pte).
106	 */
107	if (!walk->reuse_page) {
108		walk->reuse_page = pte_page(*pte);
109		/*
110		 * Because the reuse address is part of the range that we are
111		 * walking, skip the reuse address range.
112		 */
113		addr += PAGE_SIZE;
114		pte++;
115		walk->nr_walked++;
116	}
117
118	for (; addr != end; addr += PAGE_SIZE, pte++) {
119		walk->remap_pte(pte, addr, walk);
120		walk->nr_walked++;
121	}
122}
123
124static int vmemmap_pmd_range(pud_t *pud, unsigned long addr,
125			     unsigned long end,
126			     struct vmemmap_remap_walk *walk)
127{
128	pmd_t *pmd;
129	unsigned long next;
130
131	pmd = pmd_offset(pud, addr);
132	do {
133		int ret;
134
135		ret = split_vmemmap_huge_pmd(pmd, addr & PMD_MASK);
136		if (ret)
137			return ret;
138
139		next = pmd_addr_end(addr, end);
140		vmemmap_pte_range(pmd, addr, next, walk);
141	} while (pmd++, addr = next, addr != end);
142
143	return 0;
144}
145
146static int vmemmap_pud_range(p4d_t *p4d, unsigned long addr,
147			     unsigned long end,
148			     struct vmemmap_remap_walk *walk)
149{
150	pud_t *pud;
151	unsigned long next;
152
153	pud = pud_offset(p4d, addr);
154	do {
155		int ret;
156
157		next = pud_addr_end(addr, end);
158		ret = vmemmap_pmd_range(pud, addr, next, walk);
159		if (ret)
160			return ret;
161	} while (pud++, addr = next, addr != end);
162
163	return 0;
164}
 
165
166static int vmemmap_p4d_range(pgd_t *pgd, unsigned long addr,
167			     unsigned long end,
168			     struct vmemmap_remap_walk *walk)
169{
170	p4d_t *p4d;
171	unsigned long next;
172
173	p4d = p4d_offset(pgd, addr);
174	do {
175		int ret;
176
177		next = p4d_addr_end(addr, end);
178		ret = vmemmap_pud_range(p4d, addr, next, walk);
179		if (ret)
180			return ret;
181	} while (p4d++, addr = next, addr != end);
182
183	return 0;
184}
185
186static int vmemmap_remap_range(unsigned long start, unsigned long end,
187			       struct vmemmap_remap_walk *walk)
188{
189	unsigned long addr = start;
190	unsigned long next;
191	pgd_t *pgd;
192
193	VM_BUG_ON(!PAGE_ALIGNED(start));
194	VM_BUG_ON(!PAGE_ALIGNED(end));
195
196	pgd = pgd_offset_k(addr);
197	do {
198		int ret;
199
200		next = pgd_addr_end(addr, end);
201		ret = vmemmap_p4d_range(pgd, addr, next, walk);
202		if (ret)
203			return ret;
204	} while (pgd++, addr = next, addr != end);
205
206	flush_tlb_kernel_range(start, end);
207
208	return 0;
209}
210
211/*
212 * Free a vmemmap page. A vmemmap page can be allocated from the memblock
213 * allocator or buddy allocator. If the PG_reserved flag is set, it means
214 * that it allocated from the memblock allocator, just free it via the
215 * free_bootmem_page(). Otherwise, use __free_page().
216 */
217static inline void free_vmemmap_page(struct page *page)
218{
219	if (PageReserved(page))
220		free_bootmem_page(page);
221	else
222		__free_page(page);
223}
224
225/* Free a list of the vmemmap pages */
226static void free_vmemmap_page_list(struct list_head *list)
227{
228	struct page *page, *next;
229
230	list_for_each_entry_safe(page, next, list, lru)
231		free_vmemmap_page(page);
232}
233
234static void vmemmap_remap_pte(pte_t *pte, unsigned long addr,
235			      struct vmemmap_remap_walk *walk)
236{
237	/*
238	 * Remap the tail pages as read-only to catch illegal write operation
239	 * to the tail pages.
240	 */
241	pgprot_t pgprot = PAGE_KERNEL_RO;
242	struct page *page = pte_page(*pte);
243	pte_t entry;
244
245	/* Remapping the head page requires r/w */
246	if (unlikely(addr == walk->reuse_addr)) {
247		pgprot = PAGE_KERNEL;
248		list_del(&walk->reuse_page->lru);
249
250		/*
251		 * Makes sure that preceding stores to the page contents from
252		 * vmemmap_remap_free() become visible before the set_pte_at()
253		 * write.
254		 */
255		smp_wmb();
256	}
257
258	entry = mk_pte(walk->reuse_page, pgprot);
259	list_add_tail(&page->lru, walk->vmemmap_pages);
260	set_pte_at(&init_mm, addr, pte, entry);
261}
262
263/*
264 * How many struct page structs need to be reset. When we reuse the head
265 * struct page, the special metadata (e.g. page->flags or page->mapping)
266 * cannot copy to the tail struct page structs. The invalid value will be
267 * checked in the free_tail_pages_check(). In order to avoid the message
268 * of "corrupted mapping in tail page". We need to reset at least 3 (one
269 * head struct page struct and two tail struct page structs) struct page
270 * structs.
271 */
272#define NR_RESET_STRUCT_PAGE		3
273
274static inline void reset_struct_pages(struct page *start)
275{
276	struct page *from = start + NR_RESET_STRUCT_PAGE;
277
278	BUILD_BUG_ON(NR_RESET_STRUCT_PAGE * 2 > PAGE_SIZE / sizeof(struct page));
279	memcpy(start, from, sizeof(*from) * NR_RESET_STRUCT_PAGE);
280}
281
282static void vmemmap_restore_pte(pte_t *pte, unsigned long addr,
283				struct vmemmap_remap_walk *walk)
284{
285	pgprot_t pgprot = PAGE_KERNEL;
286	struct page *page;
287	void *to;
288
289	BUG_ON(pte_page(*pte) != walk->reuse_page);
290
291	page = list_first_entry(walk->vmemmap_pages, struct page, lru);
292	list_del(&page->lru);
293	to = page_to_virt(page);
294	copy_page(to, (void *)walk->reuse_addr);
295	reset_struct_pages(to);
296
297	/*
298	 * Makes sure that preceding stores to the page contents become visible
299	 * before the set_pte_at() write.
300	 */
301	smp_wmb();
302	set_pte_at(&init_mm, addr, pte, mk_pte(page, pgprot));
303}
304
305/**
306 * vmemmap_remap_free - remap the vmemmap virtual address range [@start, @end)
307 *			to the page which @reuse is mapped to, then free vmemmap
308 *			which the range are mapped to.
309 * @start:	start address of the vmemmap virtual address range that we want
310 *		to remap.
311 * @end:	end address of the vmemmap virtual address range that we want to
312 *		remap.
313 * @reuse:	reuse address.
314 *
315 * Return: %0 on success, negative error code otherwise.
316 */
317static int vmemmap_remap_free(unsigned long start, unsigned long end,
318			      unsigned long reuse)
319{
320	int ret;
321	LIST_HEAD(vmemmap_pages);
322	struct vmemmap_remap_walk walk = {
323		.remap_pte	= vmemmap_remap_pte,
324		.reuse_addr	= reuse,
325		.vmemmap_pages	= &vmemmap_pages,
326	};
327	int nid = page_to_nid((struct page *)start);
328	gfp_t gfp_mask = GFP_KERNEL | __GFP_THISNODE | __GFP_NORETRY |
329			__GFP_NOWARN;
330
331	/*
332	 * Allocate a new head vmemmap page to avoid breaking a contiguous
333	 * block of struct page memory when freeing it back to page allocator
334	 * in free_vmemmap_page_list(). This will allow the likely contiguous
335	 * struct page backing memory to be kept contiguous and allowing for
336	 * more allocations of hugepages. Fallback to the currently
337	 * mapped head page in case should it fail to allocate.
338	 */
339	walk.reuse_page = alloc_pages_node(nid, gfp_mask, 0);
340	if (walk.reuse_page) {
341		copy_page(page_to_virt(walk.reuse_page),
342			  (void *)walk.reuse_addr);
343		list_add(&walk.reuse_page->lru, &vmemmap_pages);
344	}
345
346	/*
347	 * In order to make remapping routine most efficient for the huge pages,
348	 * the routine of vmemmap page table walking has the following rules
349	 * (see more details from the vmemmap_pte_range()):
350	 *
351	 * - The range [@start, @end) and the range [@reuse, @reuse + PAGE_SIZE)
352	 *   should be continuous.
353	 * - The @reuse address is part of the range [@reuse, @end) that we are
354	 *   walking which is passed to vmemmap_remap_range().
355	 * - The @reuse address is the first in the complete range.
356	 *
357	 * So we need to make sure that @start and @reuse meet the above rules.
358	 */
359	BUG_ON(start - reuse != PAGE_SIZE);
360
361	mmap_read_lock(&init_mm);
362	ret = vmemmap_remap_range(reuse, end, &walk);
363	if (ret && walk.nr_walked) {
364		end = reuse + walk.nr_walked * PAGE_SIZE;
365		/*
366		 * vmemmap_pages contains pages from the previous
367		 * vmemmap_remap_range call which failed.  These
368		 * are pages which were removed from the vmemmap.
369		 * They will be restored in the following call.
370		 */
371		walk = (struct vmemmap_remap_walk) {
372			.remap_pte	= vmemmap_restore_pte,
373			.reuse_addr	= reuse,
374			.vmemmap_pages	= &vmemmap_pages,
375		};
376
377		vmemmap_remap_range(reuse, end, &walk);
378	}
379	mmap_read_unlock(&init_mm);
380
381	free_vmemmap_page_list(&vmemmap_pages);
382
383	return ret;
384}
385
386static int alloc_vmemmap_page_list(unsigned long start, unsigned long end,
387				   gfp_t gfp_mask, struct list_head *list)
388{
389	unsigned long nr_pages = (end - start) >> PAGE_SHIFT;
390	int nid = page_to_nid((struct page *)start);
391	struct page *page, *next;
392
393	while (nr_pages--) {
394		page = alloc_pages_node(nid, gfp_mask, 0);
395		if (!page)
396			goto out;
397		list_add_tail(&page->lru, list);
398	}
399
400	return 0;
401out:
402	list_for_each_entry_safe(page, next, list, lru)
403		__free_pages(page, 0);
404	return -ENOMEM;
405}
406
407/**
408 * vmemmap_remap_alloc - remap the vmemmap virtual address range [@start, end)
409 *			 to the page which is from the @vmemmap_pages
410 *			 respectively.
411 * @start:	start address of the vmemmap virtual address range that we want
412 *		to remap.
413 * @end:	end address of the vmemmap virtual address range that we want to
414 *		remap.
415 * @reuse:	reuse address.
416 * @gfp_mask:	GFP flag for allocating vmemmap pages.
417 *
418 * Return: %0 on success, negative error code otherwise.
419 */
420static int vmemmap_remap_alloc(unsigned long start, unsigned long end,
421			       unsigned long reuse, gfp_t gfp_mask)
422{
423	LIST_HEAD(vmemmap_pages);
424	struct vmemmap_remap_walk walk = {
425		.remap_pte	= vmemmap_restore_pte,
426		.reuse_addr	= reuse,
427		.vmemmap_pages	= &vmemmap_pages,
428	};
429
430	/* See the comment in the vmemmap_remap_free(). */
431	BUG_ON(start - reuse != PAGE_SIZE);
432
433	if (alloc_vmemmap_page_list(start, end, gfp_mask, &vmemmap_pages))
434		return -ENOMEM;
435
436	mmap_read_lock(&init_mm);
437	vmemmap_remap_range(reuse, end, &walk);
438	mmap_read_unlock(&init_mm);
439
440	return 0;
441}
442
443DEFINE_STATIC_KEY_FALSE(hugetlb_optimize_vmemmap_key);
444EXPORT_SYMBOL(hugetlb_optimize_vmemmap_key);
445
446static bool vmemmap_optimize_enabled = IS_ENABLED(CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP_DEFAULT_ON);
447core_param(hugetlb_free_vmemmap, vmemmap_optimize_enabled, bool, 0);
448
449/**
450 * hugetlb_vmemmap_restore - restore previously optimized (by
451 *			     hugetlb_vmemmap_optimize()) vmemmap pages which
452 *			     will be reallocated and remapped.
453 * @h:		struct hstate.
454 * @head:	the head page whose vmemmap pages will be restored.
455 *
456 * Return: %0 if @head's vmemmap pages have been reallocated and remapped,
457 * negative error code otherwise.
458 */
459int hugetlb_vmemmap_restore(const struct hstate *h, struct page *head)
460{
461	int ret;
462	unsigned long vmemmap_start = (unsigned long)head, vmemmap_end;
463	unsigned long vmemmap_reuse;
464
465	if (!HPageVmemmapOptimized(head))
466		return 0;
467
468	vmemmap_end	= vmemmap_start + hugetlb_vmemmap_size(h);
469	vmemmap_reuse	= vmemmap_start;
470	vmemmap_start	+= HUGETLB_VMEMMAP_RESERVE_SIZE;
471
472	/*
473	 * The pages which the vmemmap virtual address range [@vmemmap_start,
474	 * @vmemmap_end) are mapped to are freed to the buddy allocator, and
475	 * the range is mapped to the page which @vmemmap_reuse is mapped to.
476	 * When a HugeTLB page is freed to the buddy allocator, previously
477	 * discarded vmemmap pages must be allocated and remapping.
478	 */
479	ret = vmemmap_remap_alloc(vmemmap_start, vmemmap_end, vmemmap_reuse,
480				  GFP_KERNEL | __GFP_NORETRY | __GFP_THISNODE);
481	if (!ret) {
 
482		ClearHPageVmemmapOptimized(head);
483		static_branch_dec(&hugetlb_optimize_vmemmap_key);
484	}
485
486	return ret;
487}
488
489/* Return true iff a HugeTLB whose vmemmap should and can be optimized. */
490static bool vmemmap_should_optimize(const struct hstate *h, const struct page *head)
491{
492	if (!READ_ONCE(vmemmap_optimize_enabled))
493		return false;
494
495	if (!hugetlb_vmemmap_optimizable(h))
496		return false;
497
498	if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) {
499		pmd_t *pmdp, pmd;
500		struct page *vmemmap_page;
501		unsigned long vaddr = (unsigned long)head;
502
503		/*
504		 * Only the vmemmap page's vmemmap page can be self-hosted.
505		 * Walking the page tables to find the backing page of the
506		 * vmemmap page.
507		 */
508		pmdp = pmd_off_k(vaddr);
509		/*
510		 * The READ_ONCE() is used to stabilize *pmdp in a register or
511		 * on the stack so that it will stop changing under the code.
512		 * The only concurrent operation where it can be changed is
513		 * split_vmemmap_huge_pmd() (*pmdp will be stable after this
514		 * operation).
515		 */
516		pmd = READ_ONCE(*pmdp);
517		if (pmd_leaf(pmd))
518			vmemmap_page = pmd_page(pmd) + pte_index(vaddr);
519		else
520			vmemmap_page = pte_page(*pte_offset_kernel(pmdp, vaddr));
521		/*
522		 * Due to HugeTLB alignment requirements and the vmemmap pages
523		 * being at the start of the hotplugged memory region in
524		 * memory_hotplug.memmap_on_memory case. Checking any vmemmap
525		 * page's vmemmap page if it is marked as VmemmapSelfHosted is
526		 * sufficient.
527		 *
528		 * [                  hotplugged memory                  ]
529		 * [        section        ][...][        section        ]
530		 * [ vmemmap ][              usable memory               ]
531		 *   ^   |     |                                        |
532		 *   +---+     |                                        |
533		 *     ^       |                                        |
534		 *     +-------+                                        |
535		 *          ^                                           |
536		 *          +-------------------------------------------+
537		 */
538		if (PageVmemmapSelfHosted(vmemmap_page))
539			return false;
540	}
541
542	return true;
543}
544
545/**
546 * hugetlb_vmemmap_optimize - optimize @head page's vmemmap pages.
547 * @h:		struct hstate.
548 * @head:	the head page whose vmemmap pages will be optimized.
549 *
550 * This function only tries to optimize @head's vmemmap pages and does not
551 * guarantee that the optimization will succeed after it returns. The caller
552 * can use HPageVmemmapOptimized(@head) to detect if @head's vmemmap pages
553 * have been optimized.
554 */
555void hugetlb_vmemmap_optimize(const struct hstate *h, struct page *head)
556{
557	unsigned long vmemmap_start = (unsigned long)head, vmemmap_end;
558	unsigned long vmemmap_reuse;
559
560	if (!vmemmap_should_optimize(h, head))
561		return;
562
563	static_branch_inc(&hugetlb_optimize_vmemmap_key);
564
565	vmemmap_end	= vmemmap_start + hugetlb_vmemmap_size(h);
566	vmemmap_reuse	= vmemmap_start;
567	vmemmap_start	+= HUGETLB_VMEMMAP_RESERVE_SIZE;
568
569	/*
570	 * Remap the vmemmap virtual address range [@vmemmap_start, @vmemmap_end)
571	 * to the page which @vmemmap_reuse is mapped to, then free the pages
572	 * which the range [@vmemmap_start, @vmemmap_end] is mapped to.
573	 */
574	if (vmemmap_remap_free(vmemmap_start, vmemmap_end, vmemmap_reuse))
575		static_branch_dec(&hugetlb_optimize_vmemmap_key);
576	else
577		SetHPageVmemmapOptimized(head);
578}
579
580static struct ctl_table hugetlb_vmemmap_sysctls[] = {
581	{
582		.procname	= "hugetlb_optimize_vmemmap",
583		.data		= &vmemmap_optimize_enabled,
584		.maxlen		= sizeof(int),
585		.mode		= 0644,
586		.proc_handler	= proc_dobool,
587	},
588	{ }
589};
590
591static int __init hugetlb_vmemmap_init(void)
592{
593	/* HUGETLB_VMEMMAP_RESERVE_SIZE should cover all used struct pages */
594	BUILD_BUG_ON(__NR_USED_SUBPAGE * sizeof(struct page) > HUGETLB_VMEMMAP_RESERVE_SIZE);
 
 
 
595
596	if (IS_ENABLED(CONFIG_PROC_SYSCTL)) {
597		const struct hstate *h;
598
599		for_each_hstate(h) {
600			if (hugetlb_vmemmap_optimizable(h)) {
601				register_sysctl_init("vm", hugetlb_vmemmap_sysctls);
602				break;
603			}
604		}
605	}
606	return 0;
 
 
 
 
 
 
 
607}
608late_initcall(hugetlb_vmemmap_init);