Fix: rseq: arm branch to failure
[deliverable/linux.git] / mm / sparse.c
CommitLineData
d41dee36
AW
1/*
2 * sparse memory mappings.
3 */
d41dee36 4#include <linux/mm.h>
5a0e3ad6 5#include <linux/slab.h>
d41dee36
AW
6#include <linux/mmzone.h>
7#include <linux/bootmem.h>
3b32123d 8#include <linux/compiler.h>
0b0acbec 9#include <linux/highmem.h>
b95f1b31 10#include <linux/export.h>
28ae55c9 11#include <linux/spinlock.h>
0b0acbec 12#include <linux/vmalloc.h>
3b32123d 13
0c0a4a51 14#include "internal.h"
d41dee36 15#include <asm/dma.h>
8f6aac41
CL
16#include <asm/pgalloc.h>
17#include <asm/pgtable.h>
d41dee36
AW
18
19/*
20 * Permanent SPARSEMEM data:
21 *
22 * 1) mem_section - memory sections, mem_map's for valid memory
23 */
3e347261 24#ifdef CONFIG_SPARSEMEM_EXTREME
802f192e 25struct mem_section *mem_section[NR_SECTION_ROOTS]
22fc6ecc 26 ____cacheline_internodealigned_in_smp;
3e347261
BP
27#else
28struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]
22fc6ecc 29 ____cacheline_internodealigned_in_smp;
3e347261
BP
30#endif
31EXPORT_SYMBOL(mem_section);
32
89689ae7
CL
33#ifdef NODE_NOT_IN_PAGE_FLAGS
34/*
35 * If we did not store the node number in the page then we have to
36 * do a lookup in the section_to_node_table in order to find which
37 * node the page belongs to.
38 */
39#if MAX_NUMNODES <= 256
40static u8 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
41#else
42static u16 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
43#endif
44
33dd4e0e 45int page_to_nid(const struct page *page)
89689ae7
CL
46{
47 return section_to_node_table[page_to_section(page)];
48}
49EXPORT_SYMBOL(page_to_nid);
85770ffe
AW
50
51static void set_section_nid(unsigned long section_nr, int nid)
52{
53 section_to_node_table[section_nr] = nid;
54}
55#else /* !NODE_NOT_IN_PAGE_FLAGS */
56static inline void set_section_nid(unsigned long section_nr, int nid)
57{
58}
89689ae7
CL
59#endif
60
3e347261 61#ifdef CONFIG_SPARSEMEM_EXTREME
bd721ea7 62static noinline struct mem_section __ref *sparse_index_alloc(int nid)
28ae55c9
DH
63{
64 struct mem_section *section = NULL;
65 unsigned long array_size = SECTIONS_PER_ROOT *
66 sizeof(struct mem_section);
67
f52407ce
SL
68 if (slab_is_available()) {
69 if (node_state(nid, N_HIGH_MEMORY))
5b760e64 70 section = kzalloc_node(array_size, GFP_KERNEL, nid);
f52407ce 71 else
5b760e64
GS
72 section = kzalloc(array_size, GFP_KERNEL);
73 } else {
bb016b84 74 section = memblock_virt_alloc_node(array_size, nid);
5b760e64 75 }
28ae55c9
DH
76
77 return section;
3e347261 78}
802f192e 79
a3142c8e 80static int __meminit sparse_index_init(unsigned long section_nr, int nid)
802f192e 81{
28ae55c9
DH
82 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
83 struct mem_section *section;
802f192e
BP
84
85 if (mem_section[root])
28ae55c9 86 return -EEXIST;
3e347261 87
28ae55c9 88 section = sparse_index_alloc(nid);
af0cd5a7
WC
89 if (!section)
90 return -ENOMEM;
28ae55c9
DH
91
92 mem_section[root] = section;
c1c95183 93
9d1936cf 94 return 0;
28ae55c9
DH
95}
96#else /* !SPARSEMEM_EXTREME */
97static inline int sparse_index_init(unsigned long section_nr, int nid)
98{
99 return 0;
802f192e 100}
28ae55c9
DH
101#endif
102
91fd8b95 103#ifdef CONFIG_SPARSEMEM_EXTREME
4ca644d9
DH
104int __section_nr(struct mem_section* ms)
105{
106 unsigned long root_nr;
107 struct mem_section* root;
108
12783b00
MK
109 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
110 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
4ca644d9
DH
111 if (!root)
112 continue;
113
114 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
115 break;
116 }
117
db36a461
GS
118 VM_BUG_ON(root_nr == NR_SECTION_ROOTS);
119
4ca644d9
DH
120 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
121}
91fd8b95
ZC
122#else
123int __section_nr(struct mem_section* ms)
124{
125 return (int)(ms - mem_section[0]);
126}
127#endif
4ca644d9 128
30c253e6
AW
129/*
130 * During early boot, before section_mem_map is used for an actual
131 * mem_map, we use section_mem_map to store the section's NUMA
132 * node. This keeps us from having to use another data structure. The
133 * node information is cleared just before we store the real mem_map.
134 */
135static inline unsigned long sparse_encode_early_nid(int nid)
136{
137 return (nid << SECTION_NID_SHIFT);
138}
139
140static inline int sparse_early_nid(struct mem_section *section)
141{
142 return (section->section_mem_map >> SECTION_NID_SHIFT);
143}
144
2dbb51c4
MG
145/* Validate the physical addressing limitations of the model */
146void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
147 unsigned long *end_pfn)
d41dee36 148{
2dbb51c4 149 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
d41dee36 150
bead9a3a
IM
151 /*
152 * Sanity checks - do not allow an architecture to pass
153 * in larger pfns than the maximum scope of sparsemem:
154 */
2dbb51c4
MG
155 if (*start_pfn > max_sparsemem_pfn) {
156 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
157 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
158 *start_pfn, *end_pfn, max_sparsemem_pfn);
159 WARN_ON_ONCE(1);
160 *start_pfn = max_sparsemem_pfn;
161 *end_pfn = max_sparsemem_pfn;
ef161a98 162 } else if (*end_pfn > max_sparsemem_pfn) {
2dbb51c4
MG
163 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
164 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
165 *start_pfn, *end_pfn, max_sparsemem_pfn);
166 WARN_ON_ONCE(1);
167 *end_pfn = max_sparsemem_pfn;
168 }
169}
170
171/* Record a memory area against a node. */
172void __init memory_present(int nid, unsigned long start, unsigned long end)
173{
174 unsigned long pfn;
bead9a3a 175
d41dee36 176 start &= PAGE_SECTION_MASK;
2dbb51c4 177 mminit_validate_memmodel_limits(&start, &end);
d41dee36
AW
178 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
179 unsigned long section = pfn_to_section_nr(pfn);
802f192e
BP
180 struct mem_section *ms;
181
182 sparse_index_init(section, nid);
85770ffe 183 set_section_nid(section, nid);
802f192e
BP
184
185 ms = __nr_to_section(section);
186 if (!ms->section_mem_map)
30c253e6
AW
187 ms->section_mem_map = sparse_encode_early_nid(nid) |
188 SECTION_MARKED_PRESENT;
d41dee36
AW
189 }
190}
191
192/*
193 * Only used by the i386 NUMA architecures, but relatively
194 * generic code.
195 */
196unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn,
197 unsigned long end_pfn)
198{
199 unsigned long pfn;
200 unsigned long nr_pages = 0;
201
2dbb51c4 202 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
d41dee36
AW
203 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
204 if (nid != early_pfn_to_nid(pfn))
205 continue;
206
540557b9 207 if (pfn_present(pfn))
d41dee36
AW
208 nr_pages += PAGES_PER_SECTION;
209 }
210
211 return nr_pages * sizeof(struct page);
212}
213
29751f69
AW
214/*
215 * Subtle, we encode the real pfn into the mem_map such that
216 * the identity pfn - section_mem_map will return the actual
217 * physical page frame number.
218 */
219static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
220{
221 return (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
222}
223
224/*
ea01ea93 225 * Decode mem_map from the coded memmap
29751f69 226 */
29751f69
AW
227struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
228{
ea01ea93
BP
229 /* mask off the extra low bits of information */
230 coded_mem_map &= SECTION_MAP_MASK;
29751f69
AW
231 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
232}
233
a3142c8e 234static int __meminit sparse_init_one_section(struct mem_section *ms,
5c0e3066
MG
235 unsigned long pnum, struct page *mem_map,
236 unsigned long *pageblock_bitmap)
29751f69 237{
540557b9 238 if (!present_section(ms))
29751f69
AW
239 return -EINVAL;
240
30c253e6 241 ms->section_mem_map &= ~SECTION_MAP_MASK;
540557b9
AW
242 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
243 SECTION_HAS_MEM_MAP;
5c0e3066 244 ms->pageblock_flags = pageblock_bitmap;
29751f69
AW
245
246 return 1;
247}
248
04753278 249unsigned long usemap_size(void)
5c0e3066
MG
250{
251 unsigned long size_bytes;
252 size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8;
253 size_bytes = roundup(size_bytes, sizeof(unsigned long));
254 return size_bytes;
255}
256
257#ifdef CONFIG_MEMORY_HOTPLUG
258static unsigned long *__kmalloc_section_usemap(void)
259{
260 return kmalloc(usemap_size(), GFP_KERNEL);
261}
262#endif /* CONFIG_MEMORY_HOTPLUG */
263
48c90682
YG
264#ifdef CONFIG_MEMORY_HOTREMOVE
265static unsigned long * __init
a4322e1b 266sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
238305bb 267 unsigned long size)
48c90682 268{
99ab7b19
YL
269 unsigned long goal, limit;
270 unsigned long *p;
271 int nid;
48c90682
YG
272 /*
273 * A page may contain usemaps for other sections preventing the
274 * page being freed and making a section unremovable while
c800bcd5 275 * other sections referencing the usemap remain active. Similarly,
48c90682
YG
276 * a pgdat can prevent a section being removed. If section A
277 * contains a pgdat and section B contains the usemap, both
278 * sections become inter-dependent. This allocates usemaps
279 * from the same section as the pgdat where possible to avoid
280 * this problem.
281 */
07b4e2bc 282 goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
99ab7b19
YL
283 limit = goal + (1UL << PA_SECTION_SHIFT);
284 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
285again:
bb016b84
SS
286 p = memblock_virt_alloc_try_nid_nopanic(size,
287 SMP_CACHE_BYTES, goal, limit,
288 nid);
99ab7b19
YL
289 if (!p && limit) {
290 limit = 0;
291 goto again;
292 }
293 return p;
48c90682
YG
294}
295
296static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
297{
298 unsigned long usemap_snr, pgdat_snr;
299 static unsigned long old_usemap_snr = NR_MEM_SECTIONS;
300 static unsigned long old_pgdat_snr = NR_MEM_SECTIONS;
301 struct pglist_data *pgdat = NODE_DATA(nid);
302 int usemap_nid;
303
304 usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT);
305 pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
306 if (usemap_snr == pgdat_snr)
307 return;
308
309 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
310 /* skip redundant message */
311 return;
312
313 old_usemap_snr = usemap_snr;
314 old_pgdat_snr = pgdat_snr;
315
316 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
317 if (usemap_nid != nid) {
1170532b
JP
318 pr_info("node %d must be removed before remove section %ld\n",
319 nid, usemap_snr);
48c90682
YG
320 return;
321 }
322 /*
323 * There is a circular dependency.
324 * Some platforms allow un-removable section because they will just
325 * gather other removable sections for dynamic partitioning.
326 * Just notify un-removable section's number here.
327 */
1170532b
JP
328 pr_info("Section %ld and %ld (node %d) have a circular dependency on usemap and pgdat allocations\n",
329 usemap_snr, pgdat_snr, nid);
48c90682
YG
330}
331#else
332static unsigned long * __init
a4322e1b 333sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
238305bb 334 unsigned long size)
48c90682 335{
bb016b84 336 return memblock_virt_alloc_node_nopanic(size, pgdat->node_id);
48c90682
YG
337}
338
339static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
340{
341}
342#endif /* CONFIG_MEMORY_HOTREMOVE */
343
18732093 344static void __init sparse_early_usemaps_alloc_node(void *data,
a4322e1b
YL
345 unsigned long pnum_begin,
346 unsigned long pnum_end,
347 unsigned long usemap_count, int nodeid)
5c0e3066 348{
a4322e1b
YL
349 void *usemap;
350 unsigned long pnum;
18732093 351 unsigned long **usemap_map = (unsigned long **)data;
a4322e1b 352 int size = usemap_size();
5c0e3066 353
a4322e1b 354 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid),
238305bb 355 size * usemap_count);
f5bf18fa 356 if (!usemap) {
1170532b 357 pr_warn("%s: allocation failed\n", __func__);
238305bb 358 return;
48c90682
YG
359 }
360
f5bf18fa
NA
361 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
362 if (!present_section_nr(pnum))
363 continue;
364 usemap_map[pnum] = usemap;
365 usemap += size;
366 check_usemap_section_nr(nodeid, usemap_map[pnum]);
a4322e1b 367 }
5c0e3066
MG
368}
369
8f6aac41 370#ifndef CONFIG_SPARSEMEM_VMEMMAP
98f3cfc1 371struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
29751f69
AW
372{
373 struct page *map;
e48e67e0 374 unsigned long size;
29751f69
AW
375
376 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
377 if (map)
378 return map;
379
e48e67e0 380 size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
bb016b84
SS
381 map = memblock_virt_alloc_try_nid(size,
382 PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
383 BOOTMEM_ALLOC_ACCESSIBLE, nid);
8f6aac41
CL
384 return map;
385}
9bdac914
YL
386void __init sparse_mem_maps_populate_node(struct page **map_map,
387 unsigned long pnum_begin,
388 unsigned long pnum_end,
389 unsigned long map_count, int nodeid)
390{
391 void *map;
392 unsigned long pnum;
393 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
394
395 map = alloc_remap(nodeid, size * map_count);
396 if (map) {
397 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
398 if (!present_section_nr(pnum))
399 continue;
400 map_map[pnum] = map;
401 map += size;
402 }
403 return;
404 }
405
406 size = PAGE_ALIGN(size);
bb016b84
SS
407 map = memblock_virt_alloc_try_nid(size * map_count,
408 PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
409 BOOTMEM_ALLOC_ACCESSIBLE, nodeid);
9bdac914
YL
410 if (map) {
411 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
412 if (!present_section_nr(pnum))
413 continue;
414 map_map[pnum] = map;
415 map += size;
416 }
417 return;
418 }
419
420 /* fallback */
421 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
422 struct mem_section *ms;
423
424 if (!present_section_nr(pnum))
425 continue;
426 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
427 if (map_map[pnum])
428 continue;
429 ms = __nr_to_section(pnum);
1170532b 430 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
756a025f 431 __func__);
9bdac914
YL
432 ms->section_mem_map = 0;
433 }
434}
8f6aac41
CL
435#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
436
81d0d950 437#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
18732093 438static void __init sparse_early_mem_maps_alloc_node(void *data,
9bdac914
YL
439 unsigned long pnum_begin,
440 unsigned long pnum_end,
441 unsigned long map_count, int nodeid)
442{
18732093 443 struct page **map_map = (struct page **)data;
9bdac914
YL
444 sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
445 map_count, nodeid);
446}
81d0d950 447#else
9e5c6da7 448static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
8f6aac41
CL
449{
450 struct page *map;
451 struct mem_section *ms = __nr_to_section(pnum);
452 int nid = sparse_early_nid(ms);
453
98f3cfc1 454 map = sparse_mem_map_populate(pnum, nid);
29751f69
AW
455 if (map)
456 return map;
457
1170532b 458 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
756a025f 459 __func__);
802f192e 460 ms->section_mem_map = 0;
29751f69
AW
461 return NULL;
462}
9bdac914 463#endif
29751f69 464
3b32123d 465void __weak __meminit vmemmap_populate_print_last(void)
c2b91e2e
YL
466{
467}
a4322e1b 468
18732093
WL
469/**
470 * alloc_usemap_and_memmap - memory alloction for pageblock flags and vmemmap
471 * @map: usemap_map for pageblock flags or mmap_map for vmemmap
472 */
473static void __init alloc_usemap_and_memmap(void (*alloc_func)
474 (void *, unsigned long, unsigned long,
475 unsigned long, int), void *data)
476{
477 unsigned long pnum;
478 unsigned long map_count;
479 int nodeid_begin = 0;
480 unsigned long pnum_begin = 0;
481
482 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
483 struct mem_section *ms;
484
485 if (!present_section_nr(pnum))
486 continue;
487 ms = __nr_to_section(pnum);
488 nodeid_begin = sparse_early_nid(ms);
489 pnum_begin = pnum;
490 break;
491 }
492 map_count = 1;
493 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
494 struct mem_section *ms;
495 int nodeid;
496
497 if (!present_section_nr(pnum))
498 continue;
499 ms = __nr_to_section(pnum);
500 nodeid = sparse_early_nid(ms);
501 if (nodeid == nodeid_begin) {
502 map_count++;
503 continue;
504 }
505 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
506 alloc_func(data, pnum_begin, pnum,
507 map_count, nodeid_begin);
508 /* new start, update count etc*/
509 nodeid_begin = nodeid;
510 pnum_begin = pnum;
511 map_count = 1;
512 }
513 /* ok, last chunk */
514 alloc_func(data, pnum_begin, NR_MEM_SECTIONS,
515 map_count, nodeid_begin);
516}
517
193faea9
SR
518/*
519 * Allocate the accumulated non-linear sections, allocate a mem_map
520 * for each and record the physical to section mapping.
521 */
522void __init sparse_init(void)
523{
524 unsigned long pnum;
525 struct page *map;
5c0e3066 526 unsigned long *usemap;
e123dd3f 527 unsigned long **usemap_map;
81d0d950 528 int size;
81d0d950 529#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
81d0d950
YL
530 int size2;
531 struct page **map_map;
532#endif
e123dd3f 533
55878e88
CS
534 /* see include/linux/mmzone.h 'struct mem_section' definition */
535 BUILD_BUG_ON(!is_power_of_2(sizeof(struct mem_section)));
536
ca57df79
XQ
537 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
538 set_pageblock_order();
539
e123dd3f
YL
540 /*
541 * map is using big page (aka 2M in x86 64 bit)
542 * usemap is less one page (aka 24 bytes)
543 * so alloc 2M (with 2M align) and 24 bytes in turn will
544 * make next 2M slip to one more 2M later.
545 * then in big system, the memory will have a lot of holes...
25985edc 546 * here try to allocate 2M pages continuously.
e123dd3f
YL
547 *
548 * powerpc need to call sparse_init_one_section right after each
549 * sparse_early_mem_map_alloc, so allocate usemap_map at first.
550 */
551 size = sizeof(unsigned long *) * NR_MEM_SECTIONS;
bb016b84 552 usemap_map = memblock_virt_alloc(size, 0);
e123dd3f
YL
553 if (!usemap_map)
554 panic("can not allocate usemap_map\n");
18732093
WL
555 alloc_usemap_and_memmap(sparse_early_usemaps_alloc_node,
556 (void *)usemap_map);
193faea9 557
9bdac914
YL
558#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
559 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
bb016b84 560 map_map = memblock_virt_alloc(size2, 0);
9bdac914
YL
561 if (!map_map)
562 panic("can not allocate map_map\n");
18732093
WL
563 alloc_usemap_and_memmap(sparse_early_mem_maps_alloc_node,
564 (void *)map_map);
9bdac914
YL
565#endif
566
e123dd3f
YL
567 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
568 if (!present_section_nr(pnum))
193faea9 569 continue;
5c0e3066 570
e123dd3f 571 usemap = usemap_map[pnum];
5c0e3066
MG
572 if (!usemap)
573 continue;
574
9bdac914
YL
575#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
576 map = map_map[pnum];
577#else
e123dd3f 578 map = sparse_early_mem_map_alloc(pnum);
9bdac914 579#endif
e123dd3f
YL
580 if (!map)
581 continue;
582
5c0e3066
MG
583 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
584 usemap);
193faea9 585 }
e123dd3f 586
c2b91e2e
YL
587 vmemmap_populate_print_last();
588
9bdac914 589#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
bb016b84 590 memblock_free_early(__pa(map_map), size2);
9bdac914 591#endif
bb016b84 592 memblock_free_early(__pa(usemap_map), size);
193faea9
SR
593}
594
595#ifdef CONFIG_MEMORY_HOTPLUG
98f3cfc1 596#ifdef CONFIG_SPARSEMEM_VMEMMAP
85b35fea 597static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid)
98f3cfc1
YG
598{
599 /* This will make the necessary allocations eventually. */
600 return sparse_mem_map_populate(pnum, nid);
601}
85b35fea 602static void __kfree_section_memmap(struct page *memmap)
98f3cfc1 603{
0aad818b 604 unsigned long start = (unsigned long)memmap;
85b35fea 605 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
0aad818b
JW
606
607 vmemmap_free(start, end);
98f3cfc1 608}
4edd7cef 609#ifdef CONFIG_MEMORY_HOTREMOVE
81556b02 610static void free_map_bootmem(struct page *memmap)
0c0a4a51 611{
0aad818b 612 unsigned long start = (unsigned long)memmap;
81556b02 613 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
0aad818b
JW
614
615 vmemmap_free(start, end);
0c0a4a51 616}
4edd7cef 617#endif /* CONFIG_MEMORY_HOTREMOVE */
98f3cfc1 618#else
85b35fea 619static struct page *__kmalloc_section_memmap(void)
0b0acbec
DH
620{
621 struct page *page, *ret;
85b35fea 622 unsigned long memmap_size = sizeof(struct page) * PAGES_PER_SECTION;
0b0acbec 623
f2d0aa5b 624 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
0b0acbec
DH
625 if (page)
626 goto got_map_page;
627
628 ret = vmalloc(memmap_size);
629 if (ret)
630 goto got_map_ptr;
631
632 return NULL;
633got_map_page:
634 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
635got_map_ptr:
0b0acbec
DH
636
637 return ret;
638}
639
85b35fea 640static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid)
98f3cfc1 641{
85b35fea 642 return __kmalloc_section_memmap();
98f3cfc1
YG
643}
644
85b35fea 645static void __kfree_section_memmap(struct page *memmap)
0b0acbec 646{
9e2779fa 647 if (is_vmalloc_addr(memmap))
0b0acbec
DH
648 vfree(memmap);
649 else
650 free_pages((unsigned long)memmap,
85b35fea 651 get_order(sizeof(struct page) * PAGES_PER_SECTION));
0b0acbec 652}
0c0a4a51 653
4edd7cef 654#ifdef CONFIG_MEMORY_HOTREMOVE
81556b02 655static void free_map_bootmem(struct page *memmap)
0c0a4a51
YG
656{
657 unsigned long maps_section_nr, removing_section_nr, i;
81556b02 658 unsigned long magic, nr_pages;
ae64ffca 659 struct page *page = virt_to_page(memmap);
0c0a4a51 660
81556b02
ZY
661 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
662 >> PAGE_SHIFT;
663
0c0a4a51 664 for (i = 0; i < nr_pages; i++, page++) {
5f24ce5f 665 magic = (unsigned long) page->lru.next;
0c0a4a51
YG
666
667 BUG_ON(magic == NODE_INFO);
668
669 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
670 removing_section_nr = page->private;
671
672 /*
673 * When this function is called, the removing section is
674 * logical offlined state. This means all pages are isolated
675 * from page allocator. If removing section's memmap is placed
676 * on the same section, it must not be freed.
677 * If it is freed, page allocator may allocate it which will
678 * be removed physically soon.
679 */
680 if (maps_section_nr != removing_section_nr)
681 put_page_bootmem(page);
682 }
683}
4edd7cef 684#endif /* CONFIG_MEMORY_HOTREMOVE */
98f3cfc1 685#endif /* CONFIG_SPARSEMEM_VMEMMAP */
0b0acbec 686
29751f69
AW
687/*
688 * returns the number of sections whose mem_maps were properly
689 * set. If this is <=0, then that means that the passed-in
690 * map was not consumed and must be freed.
691 */
85b35fea 692int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn)
29751f69 693{
0b0acbec
DH
694 unsigned long section_nr = pfn_to_section_nr(start_pfn);
695 struct pglist_data *pgdat = zone->zone_pgdat;
696 struct mem_section *ms;
697 struct page *memmap;
5c0e3066 698 unsigned long *usemap;
0b0acbec
DH
699 unsigned long flags;
700 int ret;
29751f69 701
0b0acbec
DH
702 /*
703 * no locking for this, because it does its own
704 * plus, it does a kmalloc
705 */
bbd06825
WC
706 ret = sparse_index_init(section_nr, pgdat->node_id);
707 if (ret < 0 && ret != -EEXIST)
708 return ret;
85b35fea 709 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id);
bbd06825
WC
710 if (!memmap)
711 return -ENOMEM;
5c0e3066 712 usemap = __kmalloc_section_usemap();
bbd06825 713 if (!usemap) {
85b35fea 714 __kfree_section_memmap(memmap);
bbd06825
WC
715 return -ENOMEM;
716 }
0b0acbec
DH
717
718 pgdat_resize_lock(pgdat, &flags);
29751f69 719
0b0acbec
DH
720 ms = __pfn_to_section(start_pfn);
721 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
722 ret = -EEXIST;
723 goto out;
724 }
5c0e3066 725
85b35fea 726 memset(memmap, 0, sizeof(struct page) * PAGES_PER_SECTION);
3ac19f8e 727
29751f69
AW
728 ms->section_mem_map |= SECTION_MARKED_PRESENT;
729
5c0e3066 730 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
0b0acbec 731
0b0acbec
DH
732out:
733 pgdat_resize_unlock(pgdat, &flags);
bbd06825
WC
734 if (ret <= 0) {
735 kfree(usemap);
85b35fea 736 __kfree_section_memmap(memmap);
bbd06825 737 }
0b0acbec 738 return ret;
29751f69 739}
ea01ea93 740
f3deb687 741#ifdef CONFIG_MEMORY_HOTREMOVE
95a4774d
WC
742#ifdef CONFIG_MEMORY_FAILURE
743static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
744{
745 int i;
746
747 if (!memmap)
748 return;
749
4b94ffdc 750 for (i = 0; i < nr_pages; i++) {
95a4774d 751 if (PageHWPoison(&memmap[i])) {
293c07e3 752 atomic_long_sub(1, &num_poisoned_pages);
95a4774d
WC
753 ClearPageHWPoison(&memmap[i]);
754 }
755 }
756}
757#else
758static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
759{
760}
761#endif
762
4edd7cef
DR
763static void free_section_usemap(struct page *memmap, unsigned long *usemap)
764{
765 struct page *usemap_page;
4edd7cef
DR
766
767 if (!usemap)
768 return;
769
770 usemap_page = virt_to_page(usemap);
771 /*
772 * Check to see if allocation came from hot-plug-add
773 */
774 if (PageSlab(usemap_page) || PageCompound(usemap_page)) {
775 kfree(usemap);
776 if (memmap)
85b35fea 777 __kfree_section_memmap(memmap);
4edd7cef
DR
778 return;
779 }
780
781 /*
782 * The usemap came from bootmem. This is packed with other usemaps
783 * on the section which has pgdat at boot time. Just keep it as is now.
784 */
785
81556b02
ZY
786 if (memmap)
787 free_map_bootmem(memmap);
4edd7cef
DR
788}
789
4b94ffdc
DW
790void sparse_remove_one_section(struct zone *zone, struct mem_section *ms,
791 unsigned long map_offset)
ea01ea93
BP
792{
793 struct page *memmap = NULL;
cd099682
TC
794 unsigned long *usemap = NULL, flags;
795 struct pglist_data *pgdat = zone->zone_pgdat;
ea01ea93 796
cd099682 797 pgdat_resize_lock(pgdat, &flags);
ea01ea93
BP
798 if (ms->section_mem_map) {
799 usemap = ms->pageblock_flags;
800 memmap = sparse_decode_mem_map(ms->section_mem_map,
801 __section_nr(ms));
802 ms->section_mem_map = 0;
803 ms->pageblock_flags = NULL;
804 }
cd099682 805 pgdat_resize_unlock(pgdat, &flags);
ea01ea93 806
4b94ffdc
DW
807 clear_hwpoisoned_pages(memmap + map_offset,
808 PAGES_PER_SECTION - map_offset);
ea01ea93
BP
809 free_section_usemap(memmap, usemap);
810}
4edd7cef
DR
811#endif /* CONFIG_MEMORY_HOTREMOVE */
812#endif /* CONFIG_MEMORY_HOTPLUG */
This page took 0.815912 seconds and 5 git commands to generate.