blob: bd0276d5f66b17f6b6c78e503b91972c7677e599 [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 */
Christoph Hellwiga8fc3572017-12-29 08:53:58 +010077void * __meminit vmemmap_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/**
Christoph Hellwigeb804532017-12-29 08:53:59 +0100110 * altmap_alloc_block_buf - allocate pages from the device page map
111 * @altmap: device page map
112 * @size: size (in bytes) of the allocation
Dan Williams4b94ffd2016-01-15 16:56:22 -0800113 *
Christoph Hellwigeb804532017-12-29 08:53:59 +0100114 * Allocations are aligned to the size of the request.
Dan Williams4b94ffd2016-01-15 16:56:22 -0800115 */
Christoph Hellwiga8fc3572017-12-29 08:53:58 +0100116void * __meminit altmap_alloc_block_buf(unsigned long size,
Dan Williams4b94ffd2016-01-15 16:56:22 -0800117 struct vmem_altmap *altmap)
118{
Christoph Hellwigeb804532017-12-29 08:53:59 +0100119 unsigned long pfn, nr_pfns, nr_align;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800120
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;
125 }
126
Christoph Hellwigeb804532017-12-29 08:53:59 +0100127 pfn = vmem_altmap_next_pfn(altmap);
Dan Williams4b94ffd2016-01-15 16:56:22 -0800128 nr_pfns = size >> PAGE_SHIFT;
Christoph Hellwigeb804532017-12-29 08:53:59 +0100129 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;
133
134 altmap->alloc += nr_pfns;
135 altmap->align += nr_align;
136 pfn += nr_align;
137
Dan Williams4b94ffd2016-01-15 16:56:22 -0800138 pr_debug("%s: pfn: %#lx alloc: %ld align: %ld nr: %#lx\n",
139 __func__, pfn, altmap->alloc, altmap->align, nr_pfns);
Christoph Hellwigeb804532017-12-29 08:53:59 +0100140 return __va(__pfn_to_phys(pfn));
Dan Williams4b94ffd2016-01-15 16:56:22 -0800141}
142
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700143void __meminit vmemmap_verify(pte_t *pte, int node,
144 unsigned long start, unsigned long end)
145{
146 unsigned long pfn = pte_pfn(*pte);
147 int actual_node = early_pfn_to_nid(pfn);
148
David Rientjesb41ad142008-11-06 12:53:31 -0800149 if (node_distance(actual_node, node) > LOCAL_DISTANCE)
Joe Perches11705322016-03-17 14:19:50 -0700150 pr_warn("[%lx-%lx] potential offnode page_structs\n",
151 start, end - 1);
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700152}
153
Andy Whitcroft29c71112007-10-16 01:24:14 -0700154pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700155{
Andy Whitcroft29c71112007-10-16 01:24:14 -0700156 pte_t *pte = pte_offset_kernel(pmd, addr);
157 if (pte_none(*pte)) {
158 pte_t entry;
Christoph Hellwiga8fc3572017-12-29 08:53:58 +0100159 void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700160 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000161 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700162 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
163 set_pte_at(&init_mm, addr, pte, entry);
164 }
165 return pte;
166}
167
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800168static void * __meminit vmemmap_alloc_block_zero(unsigned long size, int node)
169{
170 void *p = vmemmap_alloc_block(size, node);
171
172 if (!p)
173 return NULL;
174 memset(p, 0, size);
175
176 return p;
177}
178
Andy Whitcroft29c71112007-10-16 01:24:14 -0700179pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
180{
181 pmd_t *pmd = pmd_offset(pud, addr);
182 if (pmd_none(*pmd)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800183 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700184 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000185 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700186 pmd_populate_kernel(&init_mm, pmd, p);
187 }
188 return pmd;
189}
190
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300191pud_t * __meminit vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node)
Andy Whitcroft29c71112007-10-16 01:24:14 -0700192{
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300193 pud_t *pud = pud_offset(p4d, addr);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700194 if (pud_none(*pud)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800195 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700196 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000197 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700198 pud_populate(&init_mm, pud, p);
199 }
200 return pud;
201}
202
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300203p4d_t * __meminit vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node)
204{
205 p4d_t *p4d = p4d_offset(pgd, addr);
206 if (p4d_none(*p4d)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800207 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300208 if (!p)
209 return NULL;
210 p4d_populate(&init_mm, p4d, p);
211 }
212 return p4d;
213}
214
Andy Whitcroft29c71112007-10-16 01:24:14 -0700215pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
216{
217 pgd_t *pgd = pgd_offset_k(addr);
218 if (pgd_none(*pgd)) {
Pavel Tatashinf7f99102017-11-15 17:36:44 -0800219 void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700220 if (!p)
Al Viro9dce07f2008-03-29 03:07:28 +0000221 return NULL;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700222 pgd_populate(&init_mm, pgd, p);
223 }
224 return pgd;
225}
226
Johannes Weiner0aad8182013-04-29 15:07:50 -0700227int __meminit vmemmap_populate_basepages(unsigned long start,
228 unsigned long end, int node)
Andy Whitcroft29c71112007-10-16 01:24:14 -0700229{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700230 unsigned long addr = start;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700231 pgd_t *pgd;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300232 p4d_t *p4d;
Andy Whitcroft29c71112007-10-16 01:24:14 -0700233 pud_t *pud;
234 pmd_t *pmd;
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700235 pte_t *pte;
236
Andy Whitcroft29c71112007-10-16 01:24:14 -0700237 for (; addr < end; addr += PAGE_SIZE) {
238 pgd = vmemmap_pgd_populate(addr, node);
239 if (!pgd)
240 return -ENOMEM;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300241 p4d = vmemmap_p4d_populate(pgd, addr, node);
242 if (!p4d)
243 return -ENOMEM;
244 pud = vmemmap_pud_populate(p4d, addr, node);
Andy Whitcroft29c71112007-10-16 01:24:14 -0700245 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);
254 }
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700255
256 return 0;
257}
258
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100259struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid,
260 struct vmem_altmap *altmap)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700261{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700262 unsigned long start;
263 unsigned long end;
264 struct page *map;
265
266 map = pfn_to_page(pnum * PAGES_PER_SECTION);
267 start = (unsigned long)map;
268 end = (unsigned long)(map + PAGES_PER_SECTION);
269
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100270 if (vmemmap_populate(start, end, nid, altmap))
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700271 return NULL;
272
273 return map;
274}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800275
276void __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)
280{
281 unsigned long pnum;
282 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
283 void *vmemmap_buf_start;
284
285 size = ALIGN(size, PMD_SIZE);
286 vmemmap_buf_start = __earlyonly_bootmem_alloc(nodeid, size * map_count,
287 PMD_SIZE, __pa(MAX_DMA_ADDRESS));
288
289 if (vmemmap_buf_start) {
290 vmemmap_buf = vmemmap_buf_start;
291 vmemmap_buf_end = vmemmap_buf_start + size * map_count;
292 }
293
294 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
295 struct mem_section *ms;
296
297 if (!present_section_nr(pnum))
298 continue;
299
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100300 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid, NULL);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800301 if (map_map[pnum])
302 continue;
303 ms = __nr_to_section(pnum);
Joe Perches11705322016-03-17 14:19:50 -0700304 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
Joe Perches756a0252016-03-17 14:19:47 -0700305 __func__);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800306 ms->section_mem_map = 0;
307 }
308
309 if (vmemmap_buf_start) {
310 /* need to free left buf */
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800311 memblock_free_early(__pa(vmemmap_buf),
312 vmemmap_buf_end - vmemmap_buf);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800313 vmemmap_buf = NULL;
314 vmemmap_buf_end = NULL;
315 }
316}