blob: 376dcf05a39c8c958f0a537230d316b645abdf86 [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()) {
Michal Hockofcdaf8422017-11-15 17:38:56 -080056 gfp_t gfp_mask = GFP_KERNEL|__GFP_RETRY_MAYFAIL|__GFP_NOWARN;
57 int order = get_order(size);
58 static bool warned;
Shaohua Lif52407c2009-09-21 17:01:19 -070059 struct page *page;
60
Michal Hockofcdaf8422017-11-15 17:38:56 -080061 page = alloc_pages_node(node, gfp_mask, order);
Christoph Lameter8f6aac42007-10-16 01:24:13 -070062 if (page)
63 return page_address(page);
Michal Hockofcdaf8422017-11-15 17:38:56 -080064
65 if (!warned) {
66 warn_alloc(gfp_mask & ~__GFP_NOWARN, NULL,
67 "vmemmap alloc failure: order:%u", order);
68 warned = true;
69 }
Christoph Lameter8f6aac42007-10-16 01:24:13 -070070 return NULL;
71 } else
KAMEZAWA Hiroyukie0dc3a52007-11-28 16:21:57 -080072 return __earlyonly_bootmem_alloc(node, size, size,
Christoph Lameter8f6aac42007-10-16 01:24:13 -070073 __pa(MAX_DMA_ADDRESS));
74}
75
Yinghai Lu9bdac912010-02-10 01:20:22 -080076/* need to make sure size is all the same during early stage */
Dan Williams4b94ffd2016-01-15 16:56:22 -080077static void * __meminit alloc_block_buf(unsigned long size, int node)
Yinghai Lu9bdac912010-02-10 01:20:22 -080078{
79 void *ptr;
80
81 if (!vmemmap_buf)
82 return vmemmap_alloc_block(size, node);
83
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);
88
89 vmemmap_buf = ptr + size;
90
91 return ptr;
92}
93
Dan Williams4b94ffd2016-01-15 16:56:22 -080094static unsigned long __meminit vmem_altmap_next_pfn(struct vmem_altmap *altmap)
95{
96 return altmap->base_pfn + altmap->reserve + altmap->alloc
97 + altmap->align;
98}
99
100static unsigned long __meminit vmem_altmap_nr_free(struct vmem_altmap *altmap)
101{
102 unsigned long allocated = altmap->alloc + altmap->align;
103
104 if (altmap->free > allocated)
105 return altmap->free - allocated;
106 return 0;
107}
108
109/**
110 * vmem_altmap_alloc - allocate pages from the vmem_altmap reservation
111 * @altmap - reserved page pool for the allocation
112 * @nr_pfns - size (in pages) of the allocation
113 *
114 * Allocations are aligned to the size of the request
115 */
116static unsigned long __meminit vmem_altmap_alloc(struct vmem_altmap *altmap,
117 unsigned long nr_pfns)
118{
119 unsigned long pfn = vmem_altmap_next_pfn(altmap);
120 unsigned long nr_align;
121
122 nr_align = 1UL << find_first_bit(&nr_pfns, BITS_PER_LONG);
123 nr_align = ALIGN(pfn, nr_align) - pfn;
124
125 if (nr_pfns + nr_align > vmem_altmap_nr_free(altmap))
126 return ULONG_MAX;
127 altmap->alloc += nr_pfns;
128 altmap->align += nr_align;
129 return pfn + nr_align;
130}
131
132static void * __meminit altmap_alloc_block_buf(unsigned long size,
133 struct vmem_altmap *altmap)
134{
135 unsigned long pfn, nr_pfns;
136 void *ptr;
137
138 if (size & ~PAGE_MASK) {
139 pr_warn_once("%s: allocations must be multiple of PAGE_SIZE (%ld)\n",
140 __func__, size);
141 return NULL;
142 }
143
144 nr_pfns = size >> PAGE_SHIFT;
145 pfn = vmem_altmap_alloc(altmap, nr_pfns);
146 if (pfn < ULONG_MAX)
147 ptr = __va(__pfn_to_phys(pfn));
148 else
149 ptr = NULL;
150 pr_debug("%s: pfn: %#lx alloc: %ld align: %ld nr: %#lx\n",
151 __func__, pfn, altmap->alloc, altmap->align, nr_pfns);
152
153 return ptr;
154}
155
156/* need to make sure size is all the same during early stage */
157void * __meminit __vmemmap_alloc_block_buf(unsigned long size, int node,
158 struct vmem_altmap *altmap)
159{
160 if (altmap)
161 return altmap_alloc_block_buf(size, altmap);
162 return alloc_block_buf(size, node);
163}
164
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700165void __meminit vmemmap_verify(pte_t *pte, int node,
166 unsigned long start, unsigned long end)
167{
168 unsigned long pfn = pte_pfn(*pte);
169 int actual_node = early_pfn_to_nid(pfn);
170
David Rientjesb41ad142008-11-06 12:53:31 -0800171 if (node_distance(actual_node, node) > LOCAL_DISTANCE)
Joe Perches11705322016-03-17 14:19:50 -0700172 pr_warn("[%lx-%lx] potential offnode page_structs\n",
173 start, end - 1);
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700174}
175
Andy Whitcroft29c71112007-10-16 01:24:14 -0700176pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700177{
Andy Whitcroft29c71112007-10-16 01:24:14 -0700178 pte_t *pte = pte_offset_kernel(pmd, addr);
179 if (pte_none(*pte)) {
180 pte_t entry;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800181 void *p = alloc_block_buf(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700182 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000183 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700184 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
185 set_pte_at(&init_mm, addr, pte, entry);
186 }
187 return pte;
188}
189
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800190static void * __meminit vmemmap_alloc_block_zero(unsigned long size, int node)
191{
192 void *p = vmemmap_alloc_block(size, node);
193
194 if (!p)
195 return NULL;
196 memset(p, 0, size);
197
198 return p;
199}
200
Andy Whitcroft29c71112007-10-16 01:24:14 -0700201pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
202{
203 pmd_t *pmd = pmd_offset(pud, addr);
204 if (pmd_none(*pmd)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800205 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700206 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000207 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700208 pmd_populate_kernel(&init_mm, pmd, p);
209 }
210 return pmd;
211}
212
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300213pud_t * __meminit vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node)
Andy Whitcroft29c71112007-10-16 01:24:14 -0700214{
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300215 pud_t *pud = pud_offset(p4d, addr);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700216 if (pud_none(*pud)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800217 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700218 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000219 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700220 pud_populate(&init_mm, pud, p);
221 }
222 return pud;
223}
224
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300225p4d_t * __meminit vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node)
226{
227 p4d_t *p4d = p4d_offset(pgd, addr);
228 if (p4d_none(*p4d)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800229 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300230 if (!p)
231 return NULL;
232 p4d_populate(&init_mm, p4d, p);
233 }
234 return p4d;
235}
236
Andy Whitcroft29c71112007-10-16 01:24:14 -0700237pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
238{
239 pgd_t *pgd = pgd_offset_k(addr);
240 if (pgd_none(*pgd)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800241 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700242 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000243 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700244 pgd_populate(&init_mm, pgd, p);
245 }
246 return pgd;
247}
248
Johannes Weiner0aad8182013-04-29 15:07:50 -0700249int __meminit vmemmap_populate_basepages(unsigned long start,
250 unsigned long end, int node)
Andy Whitcroft29c71112007-10-16 01:24:14 -0700251{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700252 unsigned long addr = start;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700253 pgd_t *pgd;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300254 p4d_t *p4d;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700255 pud_t *pud;
256 pmd_t *pmd;
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700257 pte_t *pte;
258
Andy Whitcroft29c71112007-10-16 01:24:14 -0700259 for (; addr < end; addr += PAGE_SIZE) {
260 pgd = vmemmap_pgd_populate(addr, node);
261 if (!pgd)
262 return -ENOMEM;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300263 p4d = vmemmap_p4d_populate(pgd, addr, node);
264 if (!p4d)
265 return -ENOMEM;
266 pud = vmemmap_pud_populate(p4d, addr, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700267 if (!pud)
268 return -ENOMEM;
269 pmd = vmemmap_pmd_populate(pud, addr, node);
270 if (!pmd)
271 return -ENOMEM;
272 pte = vmemmap_pte_populate(pmd, addr, node);
273 if (!pte)
274 return -ENOMEM;
275 vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
276 }
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700277
278 return 0;
279}
280
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100281struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid,
282 struct vmem_altmap *altmap)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700283{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700284 unsigned long start;
285 unsigned long end;
286 struct page *map;
287
288 map = pfn_to_page(pnum * PAGES_PER_SECTION);
289 start = (unsigned long)map;
290 end = (unsigned long)(map + PAGES_PER_SECTION);
291
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100292 if (vmemmap_populate(start, end, nid, altmap))
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700293 return NULL;
294
295 return map;
296}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800297
298void __init sparse_mem_maps_populate_node(struct page **map_map,
299 unsigned long pnum_begin,
300 unsigned long pnum_end,
301 unsigned long map_count, int nodeid)
302{
303 unsigned long pnum;
304 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
305 void *vmemmap_buf_start;
306
307 size = ALIGN(size, PMD_SIZE);
308 vmemmap_buf_start = __earlyonly_bootmem_alloc(nodeid, size * map_count,
309 PMD_SIZE, __pa(MAX_DMA_ADDRESS));
310
311 if (vmemmap_buf_start) {
312 vmemmap_buf = vmemmap_buf_start;
313 vmemmap_buf_end = vmemmap_buf_start + size * map_count;
314 }
315
316 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
317 struct mem_section *ms;
318
319 if (!present_section_nr(pnum))
320 continue;
321
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100322 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid, NULL);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800323 if (map_map[pnum])
324 continue;
325 ms = __nr_to_section(pnum);
Joe Perches11705322016-03-17 14:19:50 -0700326 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
Joe Perches756a0252016-03-17 14:19:47 -0700327 __func__);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800328 ms->section_mem_map = 0;
329 }
330
331 if (vmemmap_buf_start) {
332 /* need to free left buf */
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800333 memblock_free_early(__pa(vmemmap_buf),
334 vmemmap_buf_end - vmemmap_buf);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800335 vmemmap_buf = NULL;
336 vmemmap_buf_end = NULL;
337 }
338}