1.. SPDX-License-Identifier: GPL-2.0 2 3========================================= 4A vmemmap diet for HugeTLB and Device DAX 5========================================= 6 7HugeTLB 8======= 9 10This section is to explain how HugeTLB Vmemmap Optimization (HVO) works. 11 12The ``struct page`` structures are used to describe a physical page frame. By 13default, there is a one-to-one mapping from a page frame to it's corresponding 14``struct page``. 15 16HugeTLB pages consist of multiple base page size pages and is supported by many 17architectures. See Documentation/admin-guide/mm/hugetlbpage.rst for more 18details. On the x86-64 architecture, HugeTLB pages of size 2MB and 1GB are 19currently supported. Since the base page size on x86 is 4KB, a 2MB HugeTLB page 20consists of 512 base pages and a 1GB HugeTLB page consists of 4096 base pages. 21For each base page, there is a corresponding ``struct page``. 22 23Within the HugeTLB subsystem, only the first 4 ``struct page`` are used to 24contain unique information about a HugeTLB page. ``__NR_USED_SUBPAGE`` provides 25this upper limit. The only 'useful' information in the remaining ``struct page`` 26is the compound_head field, and this field is the same for all tail pages. 27 28By removing redundant ``struct page`` for HugeTLB pages, memory can be returned 29to the buddy allocator for other uses. 30 31Different architectures support different HugeTLB pages. For example, the 32following table is the HugeTLB page size supported by x86 and arm64 33architectures. Because arm64 supports 4k, 16k, and 64k base pages and 34supports contiguous entries, so it supports many kinds of sizes of HugeTLB 35page. 36 37+--------------+-----------+-----------------------------------------------+ 38| Architecture | Page Size | HugeTLB Page Size | 39+--------------+-----------+-----------+-----------+-----------+-----------+ 40| x86-64 | 4KB | 2MB | 1GB | | | 41+--------------+-----------+-----------+-----------+-----------+-----------+ 42| | 4KB | 64KB | 2MB | 32MB | 1GB | 43| +-----------+-----------+-----------+-----------+-----------+ 44| arm64 | 16KB | 2MB | 32MB | 1GB | | 45| +-----------+-----------+-----------+-----------+-----------+ 46| | 64KB | 2MB | 512MB | 16GB | | 47+--------------+-----------+-----------+-----------+-----------+-----------+ 48 49When the system boot up, every HugeTLB page has more than one ``struct page`` 50structs which size is (unit: pages):: 51 52 struct_size = HugeTLB_Size / PAGE_SIZE * sizeof(struct page) / PAGE_SIZE 53 54Where HugeTLB_Size is the size of the HugeTLB page. We know that the size 55of the HugeTLB page is always n times PAGE_SIZE. So we can get the following 56relationship:: 57 58 HugeTLB_Size = n * PAGE_SIZE 59 60Then:: 61 62 struct_size = n * PAGE_SIZE / PAGE_SIZE * sizeof(struct page) / PAGE_SIZE 63 = n * sizeof(struct page) / PAGE_SIZE 64 65We can use huge mapping at the pud/pmd level for the HugeTLB page. 66 67For the HugeTLB page of the pmd level mapping, then:: 68 69 struct_size = n * sizeof(struct page) / PAGE_SIZE 70 = PAGE_SIZE / sizeof(pte_t) * sizeof(struct page) / PAGE_SIZE 71 = sizeof(struct page) / sizeof(pte_t) 72 = 64 / 8 73 = 8 (pages) 74 75Where n is how many pte entries which one page can contains. So the value of 76n is (PAGE_SIZE / sizeof(pte_t)). 77 78This optimization only supports 64-bit system, so the value of sizeof(pte_t) 79is 8. And this optimization also applicable only when the size of ``struct page`` 80is a power of two. In most cases, the size of ``struct page`` is 64 bytes (e.g. 81x86-64 and arm64). So if we use pmd level mapping for a HugeTLB page, the 82size of ``struct page`` structs of it is 8 page frames which size depends on the 83size of the base page. 84 85For the HugeTLB page of the pud level mapping, then:: 86 87 struct_size = PAGE_SIZE / sizeof(pmd_t) * struct_size(pmd) 88 = PAGE_SIZE / 8 * 8 (pages) 89 = PAGE_SIZE (pages) 90 91Where the struct_size(pmd) is the size of the ``struct page`` structs of a 92HugeTLB page of the pmd level mapping. 93 94E.g.: A 2MB HugeTLB page on x86_64 consists in 8 page frames while 1GB 95HugeTLB page consists in 4096. 96 97Next, we take the pmd level mapping of the HugeTLB page as an example to 98show the internal implementation of this optimization. There are 8 pages 99``struct page`` structs associated with a HugeTLB page which is pmd mapped. 100 101Here is how things look before optimization:: 102 103 HugeTLB struct pages(8 pages) page frame(8 pages) 104 +-----------+ ---virt_to_page---> +-----------+ mapping to +-----------+ 105 | | | 0 | -------------> | 0 | 106 | | +-----------+ +-----------+ 107 | | | 1 | -------------> | 1 | 108 | | +-----------+ +-----------+ 109 | | | 2 | -------------> | 2 | 110 | | +-----------+ +-----------+ 111 | | | 3 | -------------> | 3 | 112 | | +-----------+ +-----------+ 113 | | | 4 | -------------> | 4 | 114 | PMD | +-----------+ +-----------+ 115 | level | | 5 | -------------> | 5 | 116 | mapping | +-----------+ +-----------+ 117 | | | 6 | -------------> | 6 | 118 | | +-----------+ +-----------+ 119 | | | 7 | -------------> | 7 | 120 | | +-----------+ +-----------+ 121 | | 122 | | 123 | | 124 +-----------+ 125 126The value of page->compound_head is the same for all tail pages. The first 127page of ``struct page`` (page 0) associated with the HugeTLB page contains the 4 128``struct page`` necessary to describe the HugeTLB. The only use of the remaining 129pages of ``struct page`` (page 1 to page 7) is to point to page->compound_head. 130Therefore, we can remap pages 1 to 7 to page 0. Only 1 page of ``struct page`` 131will be used for each HugeTLB page. This will allow us to free the remaining 1327 pages to the buddy allocator. 133 134Here is how things look after remapping:: 135 136 HugeTLB struct pages(8 pages) page frame(8 pages) 137 +-----------+ ---virt_to_page---> +-----------+ mapping to +-----------+ 138 | | | 0 | -------------> | 0 | 139 | | +-----------+ +-----------+ 140 | | | 1 | ---------------^ ^ ^ ^ ^ ^ ^ 141 | | +-----------+ | | | | | | 142 | | | 2 | -----------------+ | | | | | 143 | | +-----------+ | | | | | 144 | | | 3 | -------------------+ | | | | 145 | | +-----------+ | | | | 146 | | | 4 | ---------------------+ | | | 147 | PMD | +-----------+ | | | 148 | level | | 5 | -----------------------+ | | 149 | mapping | +-----------+ | | 150 | | | 6 | -------------------------+ | 151 | | +-----------+ | 152 | | | 7 | ---------------------------+ 153 | | +-----------+ 154 | | 155 | | 156 | | 157 +-----------+ 158 159When a HugeTLB is freed to the buddy system, we should allocate 7 pages for 160vmemmap pages and restore the previous mapping relationship. 161 162For the HugeTLB page of the pud level mapping. It is similar to the former. 163We also can use this approach to free (PAGE_SIZE - 1) vmemmap pages. 164 165Apart from the HugeTLB page of the pmd/pud level mapping, some architectures 166(e.g. aarch64) provides a contiguous bit in the translation table entries 167that hints to the MMU to indicate that it is one of a contiguous set of 168entries that can be cached in a single TLB entry. 169 170The contiguous bit is used to increase the mapping size at the pmd and pte 171(last) level. So this type of HugeTLB page can be optimized only when its 172size of the ``struct page`` structs is greater than **1** page. 173 174Notice: The head vmemmap page is not freed to the buddy allocator and all 175tail vmemmap pages are mapped to the head vmemmap page frame. So we can see 176more than one ``struct page`` struct with ``PG_head`` (e.g. 8 per 2 MB HugeTLB 177page) associated with each HugeTLB page. The ``compound_head()`` can handle 178this correctly. There is only **one** head ``struct page``, the tail 179``struct page`` with ``PG_head`` are fake head ``struct page``. We need an 180approach to distinguish between those two different types of ``struct page`` so 181that ``compound_head()`` can return the real head ``struct page`` when the 182parameter is the tail ``struct page`` but with ``PG_head``. The following code 183snippet describes how to distinguish between real and fake head ``struct page``. 184 185.. code-block:: c 186 187 if (test_bit(PG_head, &page->flags)) { 188 unsigned long head = READ_ONCE(page[1].compound_head); 189 190 if (head & 1) { 191 if (head == (unsigned long)page + 1) 192 /* head struct page */ 193 else 194 /* tail struct page */ 195 } else { 196 /* head struct page */ 197 } 198 } 199 200We can safely access the field of the **page[1]** with ``PG_head`` because the 201page is a compound page composed with at least two contiguous pages. 202The implementation refers to ``page_fixed_fake_head()``. 203 204Device DAX 205========== 206 207The device-dax interface uses the same tail deduplication technique explained 208in the previous chapter, except when used with the vmemmap in 209the device (altmap). 210 211The following page sizes are supported in DAX: PAGE_SIZE (4K on x86_64), 212PMD_SIZE (2M on x86_64) and PUD_SIZE (1G on x86_64). 213For powerpc equivalent details see Documentation/powerpc/vmemmap_dedup.rst 214 215The differences with HugeTLB are relatively minor. 216 217It only use 3 ``struct page`` for storing all information as opposed 218to 4 on HugeTLB pages. 219 220There's no remapping of vmemmap given that device-dax memory is not part of 221System RAM ranges initialized at boot. Thus the tail page deduplication 222happens at a later stage when we populate the sections. HugeTLB reuses the 223the head vmemmap page representing, whereas device-dax reuses the tail 224vmemmap page. This results in only half of the savings compared to HugeTLB. 225 226Deduplicated tail pages are not mapped read-only. 227 228Here's how things look like on device-dax after the sections are populated:: 229 230 +-----------+ ---virt_to_page---> +-----------+ mapping to +-----------+ 231 | | | 0 | -------------> | 0 | 232 | | +-----------+ +-----------+ 233 | | | 1 | -------------> | 1 | 234 | | +-----------+ +-----------+ 235 | | | 2 | ----------------^ ^ ^ ^ ^ ^ 236 | | +-----------+ | | | | | 237 | | | 3 | ------------------+ | | | | 238 | | +-----------+ | | | | 239 | | | 4 | --------------------+ | | | 240 | PMD | +-----------+ | | | 241 | level | | 5 | ----------------------+ | | 242 | mapping | +-----------+ | | 243 | | | 6 | ------------------------+ | 244 | | +-----------+ | 245 | | | 7 | --------------------------+ 246 | | +-----------+ 247 | | 248 | | 249 | | 250 +-----------+ 251