Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[cris-mirror.git] / mm / sparse-vmemmap.c
blobbd0276d5f66b17f6b6c78e503b91972c7677e599
1 // SPDX-License-Identifier: GPL-2.0
2 /*
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.
21 #include <linux/mm.h>
22 #include <linux/mmzone.h>
23 #include <linux/bootmem.h>
24 #include <linux/memremap.h>
25 #include <linux/highmem.h>
26 #include <linux/slab.h>
27 #include <linux/spinlock.h>
28 #include <linux/vmalloc.h>
29 #include <linux/sched.h>
30 #include <asm/dma.h>
31 #include <asm/pgalloc.h>
32 #include <asm/pgtable.h>
35 * Allocate a block of memory to be used to back the virtual memory map
36 * or to back the page tables that are used to create the mapping.
37 * Uses the main allocators if they are available, else bootmem.
40 static void * __ref __earlyonly_bootmem_alloc(int node,
41 unsigned long size,
42 unsigned long align,
43 unsigned long goal)
45 return memblock_virt_alloc_try_nid_raw(size, align, goal,
46 BOOTMEM_ALLOC_ACCESSIBLE, node);
49 static void *vmemmap_buf;
50 static void *vmemmap_buf_end;
52 void * __meminit vmemmap_alloc_block(unsigned long size, int node)
54 /* If the main allocator is up use that, fallback to bootmem. */
55 if (slab_is_available()) {
56 gfp_t gfp_mask = GFP_KERNEL|__GFP_RETRY_MAYFAIL|__GFP_NOWARN;
57 int order = get_order(size);
58 static bool warned;
59 struct page *page;
61 page = alloc_pages_node(node, gfp_mask, order);
62 if (page)
63 return page_address(page);
65 if (!warned) {
66 warn_alloc(gfp_mask & ~__GFP_NOWARN, NULL,
67 "vmemmap alloc failure: order:%u", order);
68 warned = true;
70 return NULL;
71 } else
72 return __earlyonly_bootmem_alloc(node, size, size,
73 __pa(MAX_DMA_ADDRESS));
76 /* need to make sure size is all the same during early stage */
77 void * __meminit vmemmap_alloc_block_buf(unsigned long size, int node)
79 void *ptr;
81 if (!vmemmap_buf)
82 return vmemmap_alloc_block(size, node);
84 /* take the from buf */
85 ptr = (void *)ALIGN((unsigned long)vmemmap_buf, size);
86 if (ptr + size > vmemmap_buf_end)
87 return vmemmap_alloc_block(size, node);
89 vmemmap_buf = ptr + size;
91 return ptr;
94 static unsigned long __meminit vmem_altmap_next_pfn(struct vmem_altmap *altmap)
96 return altmap->base_pfn + altmap->reserve + altmap->alloc
97 + altmap->align;
100 static unsigned long __meminit vmem_altmap_nr_free(struct vmem_altmap *altmap)
102 unsigned long allocated = altmap->alloc + altmap->align;
104 if (altmap->free > allocated)
105 return altmap->free - allocated;
106 return 0;
110 * altmap_alloc_block_buf - allocate pages from the device page map
111 * @altmap: device page map
112 * @size: size (in bytes) of the allocation
114 * Allocations are aligned to the size of the request.
116 void * __meminit altmap_alloc_block_buf(unsigned long size,
117 struct vmem_altmap *altmap)
119 unsigned long pfn, nr_pfns, nr_align;
121 if (size & ~PAGE_MASK) {
122 pr_warn_once("%s: allocations must be multiple of PAGE_SIZE (%ld)\n",
123 __func__, size);
124 return NULL;
127 pfn = vmem_altmap_next_pfn(altmap);
128 nr_pfns = size >> PAGE_SHIFT;
129 nr_align = 1UL << find_first_bit(&nr_pfns, BITS_PER_LONG);
130 nr_align = ALIGN(pfn, nr_align) - pfn;
131 if (nr_pfns + nr_align > vmem_altmap_nr_free(altmap))
132 return NULL;
134 altmap->alloc += nr_pfns;
135 altmap->align += nr_align;
136 pfn += nr_align;
138 pr_debug("%s: pfn: %#lx alloc: %ld align: %ld nr: %#lx\n",
139 __func__, pfn, altmap->alloc, altmap->align, nr_pfns);
140 return __va(__pfn_to_phys(pfn));
143 void __meminit vmemmap_verify(pte_t *pte, int node,
144 unsigned long start, unsigned long end)
146 unsigned long pfn = pte_pfn(*pte);
147 int actual_node = early_pfn_to_nid(pfn);
149 if (node_distance(actual_node, node) > LOCAL_DISTANCE)
150 pr_warn("[%lx-%lx] potential offnode page_structs\n",
151 start, end - 1);
154 pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
156 pte_t *pte = pte_offset_kernel(pmd, addr);
157 if (pte_none(*pte)) {
158 pte_t entry;
159 void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
160 if (!p)
161 return NULL;
162 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
163 set_pte_at(&init_mm, addr, pte, entry);
165 return pte;
168 static void * __meminit vmemmap_alloc_block_zero(unsigned long size, int node)
170 void *p = vmemmap_alloc_block(size, node);
172 if (!p)
173 return NULL;
174 memset(p, 0, size);
176 return p;
179 pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
181 pmd_t *pmd = pmd_offset(pud, addr);
182 if (pmd_none(*pmd)) {
183 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
184 if (!p)
185 return NULL;
186 pmd_populate_kernel(&init_mm, pmd, p);
188 return pmd;
191 pud_t * __meminit vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node)
193 pud_t *pud = pud_offset(p4d, addr);
194 if (pud_none(*pud)) {
195 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
196 if (!p)
197 return NULL;
198 pud_populate(&init_mm, pud, p);
200 return pud;
203 p4d_t * __meminit vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node)
205 p4d_t *p4d = p4d_offset(pgd, addr);
206 if (p4d_none(*p4d)) {
207 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
208 if (!p)
209 return NULL;
210 p4d_populate(&init_mm, p4d, p);
212 return p4d;
215 pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
217 pgd_t *pgd = pgd_offset_k(addr);
218 if (pgd_none(*pgd)) {
219 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
220 if (!p)
221 return NULL;
222 pgd_populate(&init_mm, pgd, p);
224 return pgd;
227 int __meminit vmemmap_populate_basepages(unsigned long start,
228 unsigned long end, int node)
230 unsigned long addr = start;
231 pgd_t *pgd;
232 p4d_t *p4d;
233 pud_t *pud;
234 pmd_t *pmd;
235 pte_t *pte;
237 for (; addr < end; addr += PAGE_SIZE) {
238 pgd = vmemmap_pgd_populate(addr, node);
239 if (!pgd)
240 return -ENOMEM;
241 p4d = vmemmap_p4d_populate(pgd, addr, node);
242 if (!p4d)
243 return -ENOMEM;
244 pud = vmemmap_pud_populate(p4d, addr, node);
245 if (!pud)
246 return -ENOMEM;
247 pmd = vmemmap_pmd_populate(pud, addr, node);
248 if (!pmd)
249 return -ENOMEM;
250 pte = vmemmap_pte_populate(pmd, addr, node);
251 if (!pte)
252 return -ENOMEM;
253 vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
256 return 0;
259 struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid,
260 struct vmem_altmap *altmap)
262 unsigned long start;
263 unsigned long end;
264 struct page *map;
266 map = pfn_to_page(pnum * PAGES_PER_SECTION);
267 start = (unsigned long)map;
268 end = (unsigned long)(map + PAGES_PER_SECTION);
270 if (vmemmap_populate(start, end, nid, altmap))
271 return NULL;
273 return map;
276 void __init sparse_mem_maps_populate_node(struct page **map_map,
277 unsigned long pnum_begin,
278 unsigned long pnum_end,
279 unsigned long map_count, int nodeid)
281 unsigned long pnum;
282 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
283 void *vmemmap_buf_start;
285 size = ALIGN(size, PMD_SIZE);
286 vmemmap_buf_start = __earlyonly_bootmem_alloc(nodeid, size * map_count,
287 PMD_SIZE, __pa(MAX_DMA_ADDRESS));
289 if (vmemmap_buf_start) {
290 vmemmap_buf = vmemmap_buf_start;
291 vmemmap_buf_end = vmemmap_buf_start + size * map_count;
294 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
295 struct mem_section *ms;
297 if (!present_section_nr(pnum))
298 continue;
300 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid, NULL);
301 if (map_map[pnum])
302 continue;
303 ms = __nr_to_section(pnum);
304 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
305 __func__);
306 ms->section_mem_map = 0;
309 if (vmemmap_buf_start) {
310 /* need to free left buf */
311 memblock_free_early(__pa(vmemmap_buf),
312 vmemmap_buf_end - vmemmap_buf);
313 vmemmap_buf = NULL;
314 vmemmap_buf_end = NULL;