1 // SPDX-License-Identifier: GPL-2.0
3 * Virtual Memory Map support
5 * (C) 2007 sgi. Christoph Lameter.
7 * Virtual memory maps allow VM primitives pfn_to_page, page_to_pfn,
8 * virt_to_page, page_address() to be implemented as a base offset
9 * calculation without memory access.
11 * However, virtual mappings need a page table and TLBs. Many Linux
12 * architectures already map their physical space using 1-1 mappings
13 * via TLBs. For those arches the virtual memory map is essentially
14 * for free if we use the same page size as the 1-1 mappings. In that
15 * case the overhead consists of a few additional pages that are
16 * allocated to create a view of memory for vmemmap.
18 * The architecture is expected to provide a vmemmap_populate() function
19 * to instantiate the mapping.
22 #include <linux/mmzone.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/memremap.h>
26 #include <linux/highmem.h>
27 #include <linux/slab.h>
28 #include <linux/spinlock.h>
29 #include <linux/vmalloc.h>
30 #include <linux/sched.h>
32 #include <asm/pgalloc.h>
33 #include <asm/pgtable.h>
36 * Allocate a block of memory to be used to back the virtual memory map
37 * or to back the page tables that are used to create the mapping.
38 * Uses the main allocators if they are available, else bootmem.
41 static void * __ref __earlyonly_bootmem_alloc(int node,
46 return memblock_alloc_try_nid_raw(size, align, goal,
47 MEMBLOCK_ALLOC_ACCESSIBLE, node);
50 void * __meminit vmemmap_alloc_block(unsigned long size, int node)
52 /* If the main allocator is up use that, fallback to bootmem. */
53 if (slab_is_available()) {
54 gfp_t gfp_mask = GFP_KERNEL|__GFP_RETRY_MAYFAIL|__GFP_NOWARN;
55 int order = get_order(size);
59 page = alloc_pages_node(node, gfp_mask, order);
61 return page_address(page);
64 warn_alloc(gfp_mask & ~__GFP_NOWARN, NULL,
65 "vmemmap alloc failure: order:%u", order);
70 return __earlyonly_bootmem_alloc(node, size, size,
71 __pa(MAX_DMA_ADDRESS));
74 /* need to make sure size is all the same during early stage */
75 void * __meminit vmemmap_alloc_block_buf(unsigned long size, int node)
77 void *ptr = sparse_buffer_alloc(size);
80 ptr = vmemmap_alloc_block(size, node);
84 static unsigned long __meminit vmem_altmap_next_pfn(struct vmem_altmap *altmap)
86 return altmap->base_pfn + altmap->reserve + altmap->alloc
90 static unsigned long __meminit vmem_altmap_nr_free(struct vmem_altmap *altmap)
92 unsigned long allocated = altmap->alloc + altmap->align;
94 if (altmap->free > allocated)
95 return altmap->free - allocated;
100 * altmap_alloc_block_buf - allocate pages from the device page map
101 * @altmap: device page map
102 * @size: size (in bytes) of the allocation
104 * Allocations are aligned to the size of the request.
106 void * __meminit altmap_alloc_block_buf(unsigned long size,
107 struct vmem_altmap *altmap)
109 unsigned long pfn, nr_pfns, nr_align;
111 if (size & ~PAGE_MASK) {
112 pr_warn_once("%s: allocations must be multiple of PAGE_SIZE (%ld)\n",
117 pfn = vmem_altmap_next_pfn(altmap);
118 nr_pfns = size >> PAGE_SHIFT;
119 nr_align = 1UL << find_first_bit(&nr_pfns, BITS_PER_LONG);
120 nr_align = ALIGN(pfn, nr_align) - pfn;
121 if (nr_pfns + nr_align > vmem_altmap_nr_free(altmap))
124 altmap->alloc += nr_pfns;
125 altmap->align += nr_align;
128 pr_debug("%s: pfn: %#lx alloc: %ld align: %ld nr: %#lx\n",
129 __func__, pfn, altmap->alloc, altmap->align, nr_pfns);
130 return __va(__pfn_to_phys(pfn));
133 void __meminit vmemmap_verify(pte_t *pte, int node,
134 unsigned long start, unsigned long end)
136 unsigned long pfn = pte_pfn(*pte);
137 int actual_node = early_pfn_to_nid(pfn);
139 if (node_distance(actual_node, node) > LOCAL_DISTANCE)
140 pr_warn("[%lx-%lx] potential offnode page_structs\n",
144 pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
146 pte_t *pte = pte_offset_kernel(pmd, addr);
147 if (pte_none(*pte)) {
149 void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
152 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
153 set_pte_at(&init_mm, addr, pte, entry);
158 static void * __meminit vmemmap_alloc_block_zero(unsigned long size, int node)
160 void *p = vmemmap_alloc_block(size, node);
169 pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
171 pmd_t *pmd = pmd_offset(pud, addr);
172 if (pmd_none(*pmd)) {
173 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
176 pmd_populate_kernel(&init_mm, pmd, p);
181 pud_t * __meminit vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node)
183 pud_t *pud = pud_offset(p4d, addr);
184 if (pud_none(*pud)) {
185 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
188 pud_populate(&init_mm, pud, p);
193 p4d_t * __meminit vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node)
195 p4d_t *p4d = p4d_offset(pgd, addr);
196 if (p4d_none(*p4d)) {
197 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
200 p4d_populate(&init_mm, p4d, p);
205 pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
207 pgd_t *pgd = pgd_offset_k(addr);
208 if (pgd_none(*pgd)) {
209 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
212 pgd_populate(&init_mm, pgd, p);
217 int __meminit vmemmap_populate_basepages(unsigned long start,
218 unsigned long end, int node)
220 unsigned long addr = start;
227 for (; addr < end; addr += PAGE_SIZE) {
228 pgd = vmemmap_pgd_populate(addr, node);
231 p4d = vmemmap_p4d_populate(pgd, addr, node);
234 pud = vmemmap_pud_populate(p4d, addr, node);
237 pmd = vmemmap_pmd_populate(pud, addr, node);
240 pte = vmemmap_pte_populate(pmd, addr, node);
243 vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
249 struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid,
250 struct vmem_altmap *altmap)
256 map = pfn_to_page(pnum * PAGES_PER_SECTION);
257 start = (unsigned long)map;
258 end = (unsigned long)(map + PAGES_PER_SECTION);
260 if (vmemmap_populate(start, end, nid, altmap))