blob: 4e49762599c87d1581823b50b7b667b150f06af7 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Christoph Lameter8f6aac42007-10-16 01:24:13 -07002/*
3 * Virtual Memory Map support
4 *
Christoph Lametercde53532008-07-04 09:59:22 -07005 * (C) 2007 sgi. Christoph Lameter.
Christoph Lameter8f6aac42007-10-16 01:24:13 -07006 *
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.
10 *
11 * However, virtual mappings need a page table and TLBs. Many Linux
12 * architectures already map their physical space using 1-1 mappings
Uwe Kleine-Königb5950762010-11-01 15:38:34 -040013 * via TLBs. For those arches the virtual memory map is essentially
Christoph Lameter8f6aac42007-10-16 01:24:13 -070014 * 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.
17 *
Andy Whitcroft29c71112007-10-16 01:24:14 -070018 * The architecture is expected to provide a vmemmap_populate() function
19 * to instantiate the mapping.
Christoph Lameter8f6aac42007-10-16 01:24:13 -070020 */
21#include <linux/mm.h>
22#include <linux/mmzone.h>
23#include <linux/bootmem.h>
Dan Williams4b94ffd2016-01-15 16:56:22 -080024#include <linux/memremap.h>
Christoph Lameter8f6aac42007-10-16 01:24:13 -070025#include <linux/highmem.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090026#include <linux/slab.h>
Christoph Lameter8f6aac42007-10-16 01:24:13 -070027#include <linux/spinlock.h>
28#include <linux/vmalloc.h>
Glauber de Oliveira Costa8bca44b2007-10-29 14:37:19 -070029#include <linux/sched.h>
Christoph Lameter8f6aac42007-10-16 01:24:13 -070030#include <asm/dma.h>
31#include <asm/pgalloc.h>
32#include <asm/pgtable.h>
33
34/*
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.
38 */
KAMEZAWA Hiroyukie0dc3a52007-11-28 16:21:57 -080039
Fabian Frederickbd721ea2016-08-02 14:03:33 -070040static void * __ref __earlyonly_bootmem_alloc(int node,
KAMEZAWA Hiroyukie0dc3a52007-11-28 16:21:57 -080041 unsigned long size,
42 unsigned long align,
43 unsigned long goal)
44{
Pavel Tatashinf7f99102017-11-15 17:36:44 -080045 return memblock_virt_alloc_try_nid_raw(size, align, goal,
Santosh Shilimkarbb016b82014-01-21 15:50:34 -080046 BOOTMEM_ALLOC_ACCESSIBLE, node);
KAMEZAWA Hiroyukie0dc3a52007-11-28 16:21:57 -080047}
48
Yinghai Lu9bdac912010-02-10 01:20:22 -080049static void *vmemmap_buf;
50static void *vmemmap_buf_end;
KAMEZAWA Hiroyukie0dc3a52007-11-28 16:21:57 -080051
Christoph Lameter8f6aac42007-10-16 01:24:13 -070052void * __meminit vmemmap_alloc_block(unsigned long size, int node)
53{
54 /* If the main allocator is up use that, fallback to bootmem. */
55 if (slab_is_available()) {
Shaohua Lif52407c2009-09-21 17:01:19 -070056 struct page *page;
57
Pavel Tatashinf7f99102017-11-15 17:36:44 -080058 page = alloc_pages_node(node, GFP_KERNEL | __GFP_RETRY_MAYFAIL,
59 get_order(size));
Christoph Lameter8f6aac42007-10-16 01:24:13 -070060 if (page)
61 return page_address(page);
62 return NULL;
63 } else
KAMEZAWA Hiroyukie0dc3a52007-11-28 16:21:57 -080064 return __earlyonly_bootmem_alloc(node, size, size,
Christoph Lameter8f6aac42007-10-16 01:24:13 -070065 __pa(MAX_DMA_ADDRESS));
66}
67
Yinghai Lu9bdac912010-02-10 01:20:22 -080068/* need to make sure size is all the same during early stage */
Dan Williams4b94ffd2016-01-15 16:56:22 -080069static void * __meminit alloc_block_buf(unsigned long size, int node)
Yinghai Lu9bdac912010-02-10 01:20:22 -080070{
71 void *ptr;
72
73 if (!vmemmap_buf)
74 return vmemmap_alloc_block(size, node);
75
76 /* take the from buf */
77 ptr = (void *)ALIGN((unsigned long)vmemmap_buf, size);
78 if (ptr + size > vmemmap_buf_end)
79 return vmemmap_alloc_block(size, node);
80
81 vmemmap_buf = ptr + size;
82
83 return ptr;
84}
85
Dan Williams4b94ffd2016-01-15 16:56:22 -080086static unsigned long __meminit vmem_altmap_next_pfn(struct vmem_altmap *altmap)
87{
88 return altmap->base_pfn + altmap->reserve + altmap->alloc
89 + altmap->align;
90}
91
92static unsigned long __meminit vmem_altmap_nr_free(struct vmem_altmap *altmap)
93{
94 unsigned long allocated = altmap->alloc + altmap->align;
95
96 if (altmap->free > allocated)
97 return altmap->free - allocated;
98 return 0;
99}
100
101/**
102 * vmem_altmap_alloc - allocate pages from the vmem_altmap reservation
103 * @altmap - reserved page pool for the allocation
104 * @nr_pfns - size (in pages) of the allocation
105 *
106 * Allocations are aligned to the size of the request
107 */
108static unsigned long __meminit vmem_altmap_alloc(struct vmem_altmap *altmap,
109 unsigned long nr_pfns)
110{
111 unsigned long pfn = vmem_altmap_next_pfn(altmap);
112 unsigned long nr_align;
113
114 nr_align = 1UL << find_first_bit(&nr_pfns, BITS_PER_LONG);
115 nr_align = ALIGN(pfn, nr_align) - pfn;
116
117 if (nr_pfns + nr_align > vmem_altmap_nr_free(altmap))
118 return ULONG_MAX;
119 altmap->alloc += nr_pfns;
120 altmap->align += nr_align;
121 return pfn + nr_align;
122}
123
124static void * __meminit altmap_alloc_block_buf(unsigned long size,
125 struct vmem_altmap *altmap)
126{
127 unsigned long pfn, nr_pfns;
128 void *ptr;
129
130 if (size & ~PAGE_MASK) {
131 pr_warn_once("%s: allocations must be multiple of PAGE_SIZE (%ld)\n",
132 __func__, size);
133 return NULL;
134 }
135
136 nr_pfns = size >> PAGE_SHIFT;
137 pfn = vmem_altmap_alloc(altmap, nr_pfns);
138 if (pfn < ULONG_MAX)
139 ptr = __va(__pfn_to_phys(pfn));
140 else
141 ptr = NULL;
142 pr_debug("%s: pfn: %#lx alloc: %ld align: %ld nr: %#lx\n",
143 __func__, pfn, altmap->alloc, altmap->align, nr_pfns);
144
145 return ptr;
146}
147
148/* need to make sure size is all the same during early stage */
149void * __meminit __vmemmap_alloc_block_buf(unsigned long size, int node,
150 struct vmem_altmap *altmap)
151{
152 if (altmap)
153 return altmap_alloc_block_buf(size, altmap);
154 return alloc_block_buf(size, node);
155}
156
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700157void __meminit vmemmap_verify(pte_t *pte, int node,
158 unsigned long start, unsigned long end)
159{
160 unsigned long pfn = pte_pfn(*pte);
161 int actual_node = early_pfn_to_nid(pfn);
162
David Rientjesb41ad142008-11-06 12:53:31 -0800163 if (node_distance(actual_node, node) > LOCAL_DISTANCE)
Joe Perches11705322016-03-17 14:19:50 -0700164 pr_warn("[%lx-%lx] potential offnode page_structs\n",
165 start, end - 1);
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700166}
167
Andy Whitcroft29c71112007-10-16 01:24:14 -0700168pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700169{
Andy Whitcroft29c71112007-10-16 01:24:14 -0700170 pte_t *pte = pte_offset_kernel(pmd, addr);
171 if (pte_none(*pte)) {
172 pte_t entry;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800173 void *p = alloc_block_buf(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700174 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000175 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700176 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
177 set_pte_at(&init_mm, addr, pte, entry);
178 }
179 return pte;
180}
181
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800182static void * __meminit vmemmap_alloc_block_zero(unsigned long size, int node)
183{
184 void *p = vmemmap_alloc_block(size, node);
185
186 if (!p)
187 return NULL;
188 memset(p, 0, size);
189
190 return p;
191}
192
Andy Whitcroft29c71112007-10-16 01:24:14 -0700193pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
194{
195 pmd_t *pmd = pmd_offset(pud, addr);
196 if (pmd_none(*pmd)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800197 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700198 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000199 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700200 pmd_populate_kernel(&init_mm, pmd, p);
201 }
202 return pmd;
203}
204
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300205pud_t * __meminit vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node)
Andy Whitcroft29c71112007-10-16 01:24:14 -0700206{
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300207 pud_t *pud = pud_offset(p4d, addr);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700208 if (pud_none(*pud)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800209 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700210 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000211 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700212 pud_populate(&init_mm, pud, p);
213 }
214 return pud;
215}
216
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300217p4d_t * __meminit vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node)
218{
219 p4d_t *p4d = p4d_offset(pgd, addr);
220 if (p4d_none(*p4d)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800221 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300222 if (!p)
223 return NULL;
224 p4d_populate(&init_mm, p4d, p);
225 }
226 return p4d;
227}
228
Andy Whitcroft29c71112007-10-16 01:24:14 -0700229pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
230{
231 pgd_t *pgd = pgd_offset_k(addr);
232 if (pgd_none(*pgd)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800233 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700234 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000235 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700236 pgd_populate(&init_mm, pgd, p);
237 }
238 return pgd;
239}
240
Johannes Weiner0aad8182013-04-29 15:07:50 -0700241int __meminit vmemmap_populate_basepages(unsigned long start,
242 unsigned long end, int node)
Andy Whitcroft29c71112007-10-16 01:24:14 -0700243{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700244 unsigned long addr = start;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700245 pgd_t *pgd;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300246 p4d_t *p4d;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700247 pud_t *pud;
248 pmd_t *pmd;
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700249 pte_t *pte;
250
Andy Whitcroft29c71112007-10-16 01:24:14 -0700251 for (; addr < end; addr += PAGE_SIZE) {
252 pgd = vmemmap_pgd_populate(addr, node);
253 if (!pgd)
254 return -ENOMEM;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300255 p4d = vmemmap_p4d_populate(pgd, addr, node);
256 if (!p4d)
257 return -ENOMEM;
258 pud = vmemmap_pud_populate(p4d, addr, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700259 if (!pud)
260 return -ENOMEM;
261 pmd = vmemmap_pmd_populate(pud, addr, node);
262 if (!pmd)
263 return -ENOMEM;
264 pte = vmemmap_pte_populate(pmd, addr, node);
265 if (!pte)
266 return -ENOMEM;
267 vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
268 }
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700269
270 return 0;
271}
272
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700273struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700274{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700275 unsigned long start;
276 unsigned long end;
277 struct page *map;
278
279 map = pfn_to_page(pnum * PAGES_PER_SECTION);
280 start = (unsigned long)map;
281 end = (unsigned long)(map + PAGES_PER_SECTION);
282
283 if (vmemmap_populate(start, end, nid))
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700284 return NULL;
285
286 return map;
287}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800288
289void __init sparse_mem_maps_populate_node(struct page **map_map,
290 unsigned long pnum_begin,
291 unsigned long pnum_end,
292 unsigned long map_count, int nodeid)
293{
294 unsigned long pnum;
295 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
296 void *vmemmap_buf_start;
297
298 size = ALIGN(size, PMD_SIZE);
299 vmemmap_buf_start = __earlyonly_bootmem_alloc(nodeid, size * map_count,
300 PMD_SIZE, __pa(MAX_DMA_ADDRESS));
301
302 if (vmemmap_buf_start) {
303 vmemmap_buf = vmemmap_buf_start;
304 vmemmap_buf_end = vmemmap_buf_start + size * map_count;
305 }
306
307 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
308 struct mem_section *ms;
309
310 if (!present_section_nr(pnum))
311 continue;
312
313 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
314 if (map_map[pnum])
315 continue;
316 ms = __nr_to_section(pnum);
Joe Perches11705322016-03-17 14:19:50 -0700317 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
Joe Perches756a0252016-03-17 14:19:47 -0700318 __func__);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800319 ms->section_mem_map = 0;
320 }
321
322 if (vmemmap_buf_start) {
323 /* need to free left buf */
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800324 memblock_free_early(__pa(vmemmap_buf),
325 vmemmap_buf_end - vmemmap_buf);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800326 vmemmap_buf = NULL;
327 vmemmap_buf_end = NULL;
328 }
329}