1 1
2 .. SPDX-License-Identifier: GPL-2.0 2 .. SPDX-License-Identifier: GPL-2.0
3 3
4 ========================================= 4 =========================================
5 A vmemmap diet for HugeTLB and Device DAX 5 A vmemmap diet for HugeTLB and Device DAX
6 ========================================= 6 =========================================
7 7
8 HugeTLB 8 HugeTLB
9 ======= 9 =======
10 10
11 This section is to explain how HugeTLB Vmemmap 11 This section is to explain how HugeTLB Vmemmap Optimization (HVO) works.
12 12
13 The ``struct page`` structures are used to des 13 The ``struct page`` structures are used to describe a physical page frame. By
14 default, there is a one-to-one mapping from a 14 default, there is a one-to-one mapping from a page frame to its corresponding
15 ``struct page``. 15 ``struct page``.
16 16
17 HugeTLB pages consist of multiple base page si 17 HugeTLB pages consist of multiple base page size pages and is supported by many
18 architectures. See Documentation/admin-guide/m 18 architectures. See Documentation/admin-guide/mm/hugetlbpage.rst for more
19 details. On the x86-64 architecture, HugeTLB p 19 details. On the x86-64 architecture, HugeTLB pages of size 2MB and 1GB are
20 currently supported. Since the base page size 20 currently supported. Since the base page size on x86 is 4KB, a 2MB HugeTLB page
21 consists of 512 base pages and a 1GB HugeTLB p 21 consists of 512 base pages and a 1GB HugeTLB page consists of 262144 base pages.
22 For each base page, there is a corresponding ` 22 For each base page, there is a corresponding ``struct page``.
23 23
24 Within the HugeTLB subsystem, only the first 4 24 Within the HugeTLB subsystem, only the first 4 ``struct page`` are used to
25 contain unique information about a HugeTLB pag 25 contain unique information about a HugeTLB page. ``__NR_USED_SUBPAGE`` provides
26 this upper limit. The only 'useful' informatio 26 this upper limit. The only 'useful' information in the remaining ``struct page``
27 is the compound_head field, and this field is 27 is the compound_head field, and this field is the same for all tail pages.
28 28
29 By removing redundant ``struct page`` for Huge 29 By removing redundant ``struct page`` for HugeTLB pages, memory can be returned
30 to the buddy allocator for other uses. 30 to the buddy allocator for other uses.
31 31
32 Different architectures support different Huge 32 Different architectures support different HugeTLB pages. For example, the
33 following table is the HugeTLB page size suppo 33 following table is the HugeTLB page size supported by x86 and arm64
34 architectures. Because arm64 supports 4k, 16k, 34 architectures. Because arm64 supports 4k, 16k, and 64k base pages and
35 supports contiguous entries, so it supports ma 35 supports contiguous entries, so it supports many kinds of sizes of HugeTLB
36 page. 36 page.
37 37
38 +--------------+-----------+------------------ 38 +--------------+-----------+-----------------------------------------------+
39 | Architecture | Page Size | Hu 39 | Architecture | Page Size | HugeTLB Page Size |
40 +--------------+-----------+-----------+------ 40 +--------------+-----------+-----------+-----------+-----------+-----------+
41 | x86-64 | 4KB | 2MB | 1G 41 | x86-64 | 4KB | 2MB | 1GB | | |
42 +--------------+-----------+-----------+------ 42 +--------------+-----------+-----------+-----------+-----------+-----------+
43 | | 4KB | 64KB | 2M 43 | | 4KB | 64KB | 2MB | 32MB | 1GB |
44 | +-----------+-----------+------ 44 | +-----------+-----------+-----------+-----------+-----------+
45 | arm64 | 16KB | 2MB | 32M 45 | arm64 | 16KB | 2MB | 32MB | 1GB | |
46 | +-----------+-----------+------ 46 | +-----------+-----------+-----------+-----------+-----------+
47 | | 64KB | 2MB | 512M 47 | | 64KB | 2MB | 512MB | 16GB | |
48 +--------------+-----------+-----------+------ 48 +--------------+-----------+-----------+-----------+-----------+-----------+
49 49
50 When the system boot up, every HugeTLB page ha 50 When the system boot up, every HugeTLB page has more than one ``struct page``
51 structs which size is (unit: pages):: 51 structs which size is (unit: pages)::
52 52
53 struct_size = HugeTLB_Size / PAGE_SIZE * si 53 struct_size = HugeTLB_Size / PAGE_SIZE * sizeof(struct page) / PAGE_SIZE
54 54
55 Where HugeTLB_Size is the size of the HugeTLB 55 Where HugeTLB_Size is the size of the HugeTLB page. We know that the size
56 of the HugeTLB page is always n times PAGE_SIZ 56 of the HugeTLB page is always n times PAGE_SIZE. So we can get the following
57 relationship:: 57 relationship::
58 58
59 HugeTLB_Size = n * PAGE_SIZE 59 HugeTLB_Size = n * PAGE_SIZE
60 60
61 Then:: 61 Then::
62 62
63 struct_size = n * PAGE_SIZE / PAGE_SIZE * s 63 struct_size = n * PAGE_SIZE / PAGE_SIZE * sizeof(struct page) / PAGE_SIZE
64 = n * sizeof(struct page) / PAG 64 = n * sizeof(struct page) / PAGE_SIZE
65 65
66 We can use huge mapping at the pud/pmd level f 66 We can use huge mapping at the pud/pmd level for the HugeTLB page.
67 67
68 For the HugeTLB page of the pmd level mapping, 68 For the HugeTLB page of the pmd level mapping, then::
69 69
70 struct_size = n * sizeof(struct page) / PAG 70 struct_size = n * sizeof(struct page) / PAGE_SIZE
71 = PAGE_SIZE / sizeof(pte_t) * s 71 = PAGE_SIZE / sizeof(pte_t) * sizeof(struct page) / PAGE_SIZE
72 = sizeof(struct page) / sizeof( 72 = sizeof(struct page) / sizeof(pte_t)
73 = 64 / 8 73 = 64 / 8
74 = 8 (pages) 74 = 8 (pages)
75 75
76 Where n is how many pte entries which one page 76 Where n is how many pte entries which one page can contains. So the value of
77 n is (PAGE_SIZE / sizeof(pte_t)). 77 n is (PAGE_SIZE / sizeof(pte_t)).
78 78
79 This optimization only supports 64-bit system, 79 This optimization only supports 64-bit system, so the value of sizeof(pte_t)
80 is 8. And this optimization also applicable on 80 is 8. And this optimization also applicable only when the size of ``struct page``
81 is a power of two. In most cases, the size of 81 is a power of two. In most cases, the size of ``struct page`` is 64 bytes (e.g.
82 x86-64 and arm64). So if we use pmd level mapp 82 x86-64 and arm64). So if we use pmd level mapping for a HugeTLB page, the
83 size of ``struct page`` structs of it is 8 pag 83 size of ``struct page`` structs of it is 8 page frames which size depends on the
84 size of the base page. 84 size of the base page.
85 85
86 For the HugeTLB page of the pud level mapping, 86 For the HugeTLB page of the pud level mapping, then::
87 87
88 struct_size = PAGE_SIZE / sizeof(pmd_t) * s 88 struct_size = PAGE_SIZE / sizeof(pmd_t) * struct_size(pmd)
89 = PAGE_SIZE / 8 * 8 (pages) 89 = PAGE_SIZE / 8 * 8 (pages)
90 = PAGE_SIZE (pages) 90 = PAGE_SIZE (pages)
91 91
92 Where the struct_size(pmd) is the size of the 92 Where the struct_size(pmd) is the size of the ``struct page`` structs of a
93 HugeTLB page of the pmd level mapping. 93 HugeTLB page of the pmd level mapping.
94 94
95 E.g.: A 2MB HugeTLB page on x86_64 consists in 95 E.g.: A 2MB HugeTLB page on x86_64 consists in 8 page frames while 1GB
96 HugeTLB page consists in 4096. 96 HugeTLB page consists in 4096.
97 97
98 Next, we take the pmd level mapping of the Hug 98 Next, we take the pmd level mapping of the HugeTLB page as an example to
99 show the internal implementation of this optim 99 show the internal implementation of this optimization. There are 8 pages
100 ``struct page`` structs associated with a Huge 100 ``struct page`` structs associated with a HugeTLB page which is pmd mapped.
101 101
102 Here is how things look before optimization:: 102 Here is how things look before optimization::
103 103
104 HugeTLB struct pages(8 pa 104 HugeTLB struct pages(8 pages) page frame(8 pages)
105 +-----------+ ---virt_to_page---> +---------- 105 +-----------+ ---virt_to_page---> +-----------+ mapping to +-----------+
106 | | | 0 106 | | | 0 | -------------> | 0 |
107 | | +---------- 107 | | +-----------+ +-----------+
108 | | | 1 108 | | | 1 | -------------> | 1 |
109 | | +---------- 109 | | +-----------+ +-----------+
110 | | | 2 110 | | | 2 | -------------> | 2 |
111 | | +---------- 111 | | +-----------+ +-----------+
112 | | | 3 112 | | | 3 | -------------> | 3 |
113 | | +---------- 113 | | +-----------+ +-----------+
114 | | | 4 114 | | | 4 | -------------> | 4 |
115 | PMD | +---------- 115 | PMD | +-----------+ +-----------+
116 | level | | 5 116 | level | | 5 | -------------> | 5 |
117 | mapping | +---------- 117 | mapping | +-----------+ +-----------+
118 | | | 6 118 | | | 6 | -------------> | 6 |
119 | | +---------- 119 | | +-----------+ +-----------+
120 | | | 7 120 | | | 7 | -------------> | 7 |
121 | | +---------- 121 | | +-----------+ +-----------+
122 | | 122 | |
123 | | 123 | |
124 | | 124 | |
125 +-----------+ 125 +-----------+
126 126
127 The value of page->compound_head is the same f 127 The value of page->compound_head is the same for all tail pages. The first
128 page of ``struct page`` (page 0) associated wi 128 page of ``struct page`` (page 0) associated with the HugeTLB page contains the 4
129 ``struct page`` necessary to describe the Huge 129 ``struct page`` necessary to describe the HugeTLB. The only use of the remaining
130 pages of ``struct page`` (page 1 to page 7) is 130 pages of ``struct page`` (page 1 to page 7) is to point to page->compound_head.
131 Therefore, we can remap pages 1 to 7 to page 0 131 Therefore, we can remap pages 1 to 7 to page 0. Only 1 page of ``struct page``
132 will be used for each HugeTLB page. This will 132 will be used for each HugeTLB page. This will allow us to free the remaining
133 7 pages to the buddy allocator. 133 7 pages to the buddy allocator.
134 134
135 Here is how things look after remapping:: 135 Here is how things look after remapping::
136 136
137 HugeTLB struct pages(8 pa 137 HugeTLB struct pages(8 pages) page frame(8 pages)
138 +-----------+ ---virt_to_page---> +---------- 138 +-----------+ ---virt_to_page---> +-----------+ mapping to +-----------+
139 | | | 0 139 | | | 0 | -------------> | 0 |
140 | | +---------- 140 | | +-----------+ +-----------+
141 | | | 1 141 | | | 1 | ---------------^ ^ ^ ^ ^ ^ ^
142 | | +---------- 142 | | +-----------+ | | | | | |
143 | | | 2 143 | | | 2 | -----------------+ | | | | |
144 | | +---------- 144 | | +-----------+ | | | | |
145 | | | 3 145 | | | 3 | -------------------+ | | | |
146 | | +---------- 146 | | +-----------+ | | | |
147 | | | 4 147 | | | 4 | ---------------------+ | | |
148 | PMD | +---------- 148 | PMD | +-----------+ | | |
149 | level | | 5 149 | level | | 5 | -----------------------+ | |
150 | mapping | +---------- 150 | mapping | +-----------+ | |
151 | | | 6 151 | | | 6 | -------------------------+ |
152 | | +---------- 152 | | +-----------+ |
153 | | | 7 153 | | | 7 | ---------------------------+
154 | | +---------- 154 | | +-----------+
155 | | 155 | |
156 | | 156 | |
157 | | 157 | |
158 +-----------+ 158 +-----------+
159 159
160 When a HugeTLB is freed to the buddy system, w 160 When a HugeTLB is freed to the buddy system, we should allocate 7 pages for
161 vmemmap pages and restore the previous mapping 161 vmemmap pages and restore the previous mapping relationship.
162 162
163 For the HugeTLB page of the pud level mapping. 163 For the HugeTLB page of the pud level mapping. It is similar to the former.
164 We also can use this approach to free (PAGE_SI 164 We also can use this approach to free (PAGE_SIZE - 1) vmemmap pages.
165 165
166 Apart from the HugeTLB page of the pmd/pud lev 166 Apart from the HugeTLB page of the pmd/pud level mapping, some architectures
167 (e.g. aarch64) provides a contiguous bit in th 167 (e.g. aarch64) provides a contiguous bit in the translation table entries
168 that hints to the MMU to indicate that it is o 168 that hints to the MMU to indicate that it is one of a contiguous set of
169 entries that can be cached in a single TLB ent 169 entries that can be cached in a single TLB entry.
170 170
171 The contiguous bit is used to increase the map 171 The contiguous bit is used to increase the mapping size at the pmd and pte
172 (last) level. So this type of HugeTLB page can 172 (last) level. So this type of HugeTLB page can be optimized only when its
173 size of the ``struct page`` structs is greater 173 size of the ``struct page`` structs is greater than **1** page.
174 174
175 Notice: The head vmemmap page is not freed to 175 Notice: The head vmemmap page is not freed to the buddy allocator and all
176 tail vmemmap pages are mapped to the head vmem 176 tail vmemmap pages are mapped to the head vmemmap page frame. So we can see
177 more than one ``struct page`` struct with ``PG 177 more than one ``struct page`` struct with ``PG_head`` (e.g. 8 per 2 MB HugeTLB
178 page) associated with each HugeTLB page. The ` 178 page) associated with each HugeTLB page. The ``compound_head()`` can handle
179 this correctly. There is only **one** head ``s 179 this correctly. There is only **one** head ``struct page``, the tail
180 ``struct page`` with ``PG_head`` are fake head 180 ``struct page`` with ``PG_head`` are fake head ``struct page``. We need an
181 approach to distinguish between those two diff 181 approach to distinguish between those two different types of ``struct page`` so
182 that ``compound_head()`` can return the real h 182 that ``compound_head()`` can return the real head ``struct page`` when the
183 parameter is the tail ``struct page`` but with !! 183 parameter is the tail ``struct page`` but with ``PG_head``. The following code
>> 184 snippet describes how to distinguish between real and fake head ``struct page``.
>> 185
>> 186 .. code-block:: c
>> 187
>> 188 if (test_bit(PG_head, &page->flags)) {
>> 189 unsigned long head = READ_ONCE(page[1].compound_head);
>> 190
>> 191 if (head & 1) {
>> 192 if (head == (unsigned long)page + 1)
>> 193 /* head struct page */
>> 194 else
>> 195 /* tail struct page */
>> 196 } else {
>> 197 /* head struct page */
>> 198 }
>> 199 }
>> 200
>> 201 We can safely access the field of the **page[1]** with ``PG_head`` because the
>> 202 page is a compound page composed with at least two contiguous pages.
>> 203 The implementation refers to ``page_fixed_fake_head()``.
184 204
185 Device DAX 205 Device DAX
186 ========== 206 ==========
187 207
188 The device-dax interface uses the same tail de 208 The device-dax interface uses the same tail deduplication technique explained
189 in the previous chapter, except when used with 209 in the previous chapter, except when used with the vmemmap in
190 the device (altmap). 210 the device (altmap).
191 211
192 The following page sizes are supported in DAX: 212 The following page sizes are supported in DAX: PAGE_SIZE (4K on x86_64),
193 PMD_SIZE (2M on x86_64) and PUD_SIZE (1G on x8 213 PMD_SIZE (2M on x86_64) and PUD_SIZE (1G on x86_64).
194 For powerpc equivalent details see Documentati 214 For powerpc equivalent details see Documentation/arch/powerpc/vmemmap_dedup.rst
195 215
196 The differences with HugeTLB are relatively mi 216 The differences with HugeTLB are relatively minor.
197 217
198 It only use 3 ``struct page`` for storing all 218 It only use 3 ``struct page`` for storing all information as opposed
199 to 4 on HugeTLB pages. 219 to 4 on HugeTLB pages.
200 220
201 There's no remapping of vmemmap given that dev 221 There's no remapping of vmemmap given that device-dax memory is not part of
202 System RAM ranges initialized at boot. Thus th 222 System RAM ranges initialized at boot. Thus the tail page deduplication
203 happens at a later stage when we populate the 223 happens at a later stage when we populate the sections. HugeTLB reuses the
204 the head vmemmap page representing, whereas de 224 the head vmemmap page representing, whereas device-dax reuses the tail
205 vmemmap page. This results in only half of the 225 vmemmap page. This results in only half of the savings compared to HugeTLB.
206 226
207 Deduplicated tail pages are not mapped read-on 227 Deduplicated tail pages are not mapped read-only.
208 228
209 Here's how things look like on device-dax afte 229 Here's how things look like on device-dax after the sections are populated::
210 230
211 +-----------+ ---virt_to_page---> +---------- 231 +-----------+ ---virt_to_page---> +-----------+ mapping to +-----------+
212 | | | 0 232 | | | 0 | -------------> | 0 |
213 | | +---------- 233 | | +-----------+ +-----------+
214 | | | 1 234 | | | 1 | -------------> | 1 |
215 | | +---------- 235 | | +-----------+ +-----------+
216 | | | 2 236 | | | 2 | ----------------^ ^ ^ ^ ^ ^
217 | | +---------- 237 | | +-----------+ | | | | |
218 | | | 3 238 | | | 3 | ------------------+ | | | |
219 | | +---------- 239 | | +-----------+ | | | |
220 | | | 4 240 | | | 4 | --------------------+ | | |
221 | PMD | +---------- 241 | PMD | +-----------+ | | |
222 | level | | 5 242 | level | | 5 | ----------------------+ | |
223 | mapping | +---------- 243 | mapping | +-----------+ | |
224 | | | 6 244 | | | 6 | ------------------------+ |
225 | | +---------- 245 | | +-----------+ |
226 | | | 7 246 | | | 7 | --------------------------+
227 | | +---------- 247 | | +-----------+
228 | | 248 | |
229 | | 249 | |
230 | | 250 | |
231 +-----------+ 251 +-----------+
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.