projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'random_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso...
[deliverable/linux.git] / arch / x86 / mm / init_64.c
1 /*
2 * linux/arch/x86_64/mm/init.c
3 *
4 * Copyright (C) 1995 Linus Torvalds
5 * Copyright (C) 2000 Pavel Machek <pavel@ucw.cz>
6 * Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
7 */
8
9 #include <linux/signal.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/string.h>
14 #include <linux/types.h>
15 #include <linux/ptrace.h>
16 #include <linux/mman.h>
17 #include <linux/mm.h>
18 #include <linux/swap.h>
19 #include <linux/smp.h>
20 #include <linux/init.h>
21 #include <linux/initrd.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/proc_fs.h>
26 #include <linux/pci.h>
27 #include <linux/pfn.h>
28 #include <linux/poison.h>
29 #include <linux/dma-mapping.h>
30 #include <linux/module.h>
31 #include <linux/memory.h>
32 #include <linux/memory_hotplug.h>
33 #include <linux/memremap.h>
34 #include <linux/nmi.h>
35 #include <linux/gfp.h>
36 #include <linux/kcore.h>
37
38 #include <asm/processor.h>
39 #include <asm/bios_ebda.h>
40 #include <asm/uaccess.h>
41 #include <asm/pgtable.h>
42 #include <asm/pgalloc.h>
43 #include <asm/dma.h>
44 #include <asm/fixmap.h>
45 #include <asm/e820.h>
46 #include <asm/apic.h>
47 #include <asm/tlb.h>
48 #include <asm/mmu_context.h>
49 #include <asm/proto.h>
50 #include <asm/smp.h>
51 #include <asm/sections.h>
52 #include <asm/kdebug.h>
53 #include <asm/numa.h>
54 #include <asm/cacheflush.h>
55 #include <asm/init.h>
56 #include <asm/uv/uv.h>
57 #include <asm/setup.h>
58
59 #include "mm_internal.h"
60
61 #include "ident_map.c"
62
63 /*
64 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
65 * physical space so we can cache the place of the first one and move
66 * around without checking the pgd every time.
67 */
68
69 pteval_t __supported_pte_mask __read_mostly = ~0;
70 EXPORT_SYMBOL_GPL(__supported_pte_mask);
71
72 int force_personality32;
73
74 /*
75 * noexec32=on|off
76 * Control non executable heap for 32bit processes.
77 * To control the stack too use noexec=off
78 *
79 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
80 * off PROT_READ implies PROT_EXEC
81 */
82 static int __init nonx32_setup(char *str)
83 {
84 if (!strcmp(str, "on"))
85 force_personality32 &= ~READ_IMPLIES_EXEC;
86 else if (!strcmp(str, "off"))
87 force_personality32 |= READ_IMPLIES_EXEC;
88 return 1;
89 }
90 __setup("noexec32=", nonx32_setup);
91
92 /*
93 * When memory was added/removed make sure all the processes MM have
94 * suitable PGD entries in the local PGD level page.
95 */
96 void sync_global_pgds(unsigned long start, unsigned long end, int removed)
97 {
98 unsigned long address;
99
100 for (address = start; address <= end; address += PGDIR_SIZE) {
101 const pgd_t *pgd_ref = pgd_offset_k(address);
102 struct page *page;
103
104 /*
105 * When it is called after memory hot remove, pgd_none()
106 * returns true. In this case (removed == 1), we must clear
107 * the PGD entries in the local PGD level page.
108 */
109 if (pgd_none(*pgd_ref) && !removed)
110 continue;
111
112 spin_lock(&pgd_lock);
113 list_for_each_entry(page, &pgd_list, lru) {
114 pgd_t *pgd;
115 spinlock_t *pgt_lock;
116
117 pgd = (pgd_t *)page_address(page) + pgd_index(address);
118 /* the pgt_lock only for Xen */
119 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
120 spin_lock(pgt_lock);
121
122 if (!pgd_none(*pgd_ref) && !pgd_none(*pgd))
123 BUG_ON(pgd_page_vaddr(*pgd)
124 != pgd_page_vaddr(*pgd_ref));
125
126 if (removed) {
127 if (pgd_none(*pgd_ref) && !pgd_none(*pgd))
128 pgd_clear(pgd);
129 } else {
130 if (pgd_none(*pgd))
131 set_pgd(pgd, *pgd_ref);
132 }
133
134 spin_unlock(pgt_lock);
135 }
136 spin_unlock(&pgd_lock);
137 }
138 }
139
140 /*
141 * NOTE: This function is marked __ref because it calls __init function
142 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
143 */
144 static __ref void *spp_getpage(void)
145 {
146 void *ptr;
147
148 if (after_bootmem)
149 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
150 else
151 ptr = alloc_bootmem_pages(PAGE_SIZE);
152
153 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
154 panic("set_pte_phys: cannot allocate page data %s\n",
155 after_bootmem ? "after bootmem" : "");
156 }
157
158 pr_debug("spp_getpage %p\n", ptr);
159
160 return ptr;
161 }
162
163 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
164 {
165 if (pgd_none(*pgd)) {
166 pud_t *pud = (pud_t *)spp_getpage();
167 pgd_populate(&init_mm, pgd, pud);
168 if (pud != pud_offset(pgd, 0))
169 printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
170 pud, pud_offset(pgd, 0));
171 }
172 return pud_offset(pgd, vaddr);
173 }
174
175 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
176 {
177 if (pud_none(*pud)) {
178 pmd_t *pmd = (pmd_t *) spp_getpage();
179 pud_populate(&init_mm, pud, pmd);
180 if (pmd != pmd_offset(pud, 0))
181 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
182 pmd, pmd_offset(pud, 0));
183 }
184 return pmd_offset(pud, vaddr);
185 }
186
187 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
188 {
189 if (pmd_none(*pmd)) {
190 pte_t *pte = (pte_t *) spp_getpage();
191 pmd_populate_kernel(&init_mm, pmd, pte);
192 if (pte != pte_offset_kernel(pmd, 0))
193 printk(KERN_ERR "PAGETABLE BUG #02!\n");
194 }
195 return pte_offset_kernel(pmd, vaddr);
196 }
197
198 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
199 {
200 pud_t *pud;
201 pmd_t *pmd;
202 pte_t *pte;
203
204 pud = pud_page + pud_index(vaddr);
205 pmd = fill_pmd(pud, vaddr);
206 pte = fill_pte(pmd, vaddr);
207
208 set_pte(pte, new_pte);
209
210 /*
211 * It's enough to flush this one mapping.
212 * (PGE mappings get flushed as well)
213 */
214 __flush_tlb_one(vaddr);
215 }
216
217 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
218 {
219 pgd_t *pgd;
220 pud_t *pud_page;
221
222 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
223
224 pgd = pgd_offset_k(vaddr);
225 if (pgd_none(*pgd)) {
226 printk(KERN_ERR
227 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
228 return;
229 }
230 pud_page = (pud_t*)pgd_page_vaddr(*pgd);
231 set_pte_vaddr_pud(pud_page, vaddr, pteval);
232 }
233
234 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
235 {
236 pgd_t *pgd;
237 pud_t *pud;
238
239 pgd = pgd_offset_k(vaddr);
240 pud = fill_pud(pgd, vaddr);
241 return fill_pmd(pud, vaddr);
242 }
243
244 pte_t * __init populate_extra_pte(unsigned long vaddr)
245 {
246 pmd_t *pmd;
247
248 pmd = populate_extra_pmd(vaddr);
249 return fill_pte(pmd, vaddr);
250 }
251
252 /*
253 * Create large page table mappings for a range of physical addresses.
254 */
255 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
256 enum page_cache_mode cache)
257 {
258 pgd_t *pgd;
259 pud_t *pud;
260 pmd_t *pmd;
261 pgprot_t prot;
262
263 pgprot_val(prot) = pgprot_val(PAGE_KERNEL_LARGE) |
264 pgprot_val(pgprot_4k_2_large(cachemode2pgprot(cache)));
265 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
266 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
267 pgd = pgd_offset_k((unsigned long)__va(phys));
268 if (pgd_none(*pgd)) {
269 pud = (pud_t *) spp_getpage();
270 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
271 _PAGE_USER));
272 }
273 pud = pud_offset(pgd, (unsigned long)__va(phys));
274 if (pud_none(*pud)) {
275 pmd = (pmd_t *) spp_getpage();
276 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
277 _PAGE_USER));
278 }
279 pmd = pmd_offset(pud, phys);
280 BUG_ON(!pmd_none(*pmd));
281 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
282 }
283 }
284
285 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
286 {
287 __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_WB);
288 }
289
290 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
291 {
292 __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_UC);
293 }
294
295 /*
296 * The head.S code sets up the kernel high mapping:
297 *
298 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
299 *
300 * phys_base holds the negative offset to the kernel, which is added
301 * to the compile time generated pmds. This results in invalid pmds up
302 * to the point where we hit the physaddr 0 mapping.
303 *
304 * We limit the mappings to the region from _text to _brk_end. _brk_end
305 * is rounded up to the 2MB boundary. This catches the invalid pmds as
306 * well, as they are located before _text:
307 */
308 void __init cleanup_highmap(void)
309 {
310 unsigned long vaddr = __START_KERNEL_map;
311 unsigned long vaddr_end = __START_KERNEL_map + KERNEL_IMAGE_SIZE;
312 unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
313 pmd_t *pmd = level2_kernel_pgt;
314
315 /*
316 * Native path, max_pfn_mapped is not set yet.
317 * Xen has valid max_pfn_mapped set in
318 * arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
319 */
320 if (max_pfn_mapped)
321 vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
322
323 for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
324 if (pmd_none(*pmd))
325 continue;
326 if (vaddr < (unsigned long) _text || vaddr > end)
327 set_pmd(pmd, __pmd(0));
328 }
329 }
330
331 /*
332 * Create PTE level page table mapping for physical addresses.
333 * It returns the last physical address mapped.
334 */
335 static unsigned long __meminit
336 phys_pte_init(pte_t *pte_page, unsigned long paddr, unsigned long paddr_end,
337 pgprot_t prot)
338 {
339 unsigned long pages = 0, paddr_next;
340 unsigned long paddr_last = paddr_end;
341 pte_t *pte;
342 int i;
343
344 pte = pte_page + pte_index(paddr);
345 i = pte_index(paddr);
346
347 for (; i < PTRS_PER_PTE; i++, paddr = paddr_next, pte++) {
348 paddr_next = (paddr & PAGE_MASK) + PAGE_SIZE;
349 if (paddr >= paddr_end) {
350 if (!after_bootmem &&
351 !e820_any_mapped(paddr & PAGE_MASK, paddr_next,
352 E820_RAM) &&
353 !e820_any_mapped(paddr & PAGE_MASK, paddr_next,
354 E820_RESERVED_KERN))
355 set_pte(pte, __pte(0));
356 continue;
357 }
358
359 /*
360 * We will re-use the existing mapping.
361 * Xen for example has some special requirements, like mapping
362 * pagetable pages as RO. So assume someone who pre-setup
363 * these mappings are more intelligent.
364 */
365 if (!pte_none(*pte)) {
366 if (!after_bootmem)
367 pages++;
368 continue;
369 }
370
371 if (0)
372 pr_info(" pte=%p addr=%lx pte=%016lx\n", pte, paddr,
373 pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL).pte);
374 pages++;
375 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot));
376 paddr_last = (paddr & PAGE_MASK) + PAGE_SIZE;
377 }
378
379 update_page_count(PG_LEVEL_4K, pages);
380
381 return paddr_last;
382 }
383
384 /*
385 * Create PMD level page table mapping for physical addresses. The virtual
386 * and physical address have to be aligned at this level.
387 * It returns the last physical address mapped.
388 */
389 static unsigned long __meminit
390 phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
391 unsigned long page_size_mask, pgprot_t prot)
392 {
393 unsigned long pages = 0, paddr_next;
394 unsigned long paddr_last = paddr_end;
395
396 int i = pmd_index(paddr);
397
398 for (; i < PTRS_PER_PMD; i++, paddr = paddr_next) {
399 pmd_t *pmd = pmd_page + pmd_index(paddr);
400 pte_t *pte;
401 pgprot_t new_prot = prot;
402
403 paddr_next = (paddr & PMD_MASK) + PMD_SIZE;
404 if (paddr >= paddr_end) {
405 if (!after_bootmem &&
406 !e820_any_mapped(paddr & PMD_MASK, paddr_next,
407 E820_RAM) &&
408 !e820_any_mapped(paddr & PMD_MASK, paddr_next,
409 E820_RESERVED_KERN))
410 set_pmd(pmd, __pmd(0));
411 continue;
412 }
413
414 if (!pmd_none(*pmd)) {
415 if (!pmd_large(*pmd)) {
416 spin_lock(&init_mm.page_table_lock);
417 pte = (pte_t *)pmd_page_vaddr(*pmd);
418 paddr_last = phys_pte_init(pte, paddr,
419 paddr_end, prot);
420 spin_unlock(&init_mm.page_table_lock);
421 continue;
422 }
423 /*
424 * If we are ok with PG_LEVEL_2M mapping, then we will
425 * use the existing mapping,
426 *
427 * Otherwise, we will split the large page mapping but
428 * use the same existing protection bits except for
429 * large page, so that we don't violate Intel's TLB
430 * Application note (317080) which says, while changing
431 * the page sizes, new and old translations should
432 * not differ with respect to page frame and
433 * attributes.
434 */
435 if (page_size_mask & (1 << PG_LEVEL_2M)) {
436 if (!after_bootmem)
437 pages++;
438 paddr_last = paddr_next;
439 continue;
440 }
441 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
442 }
443
444 if (page_size_mask & (1<<PG_LEVEL_2M)) {
445 pages++;
446 spin_lock(&init_mm.page_table_lock);
447 set_pte((pte_t *)pmd,
448 pfn_pte((paddr & PMD_MASK) >> PAGE_SHIFT,
449 __pgprot(pgprot_val(prot) | _PAGE_PSE)));
450 spin_unlock(&init_mm.page_table_lock);
451 paddr_last = paddr_next;
452 continue;
453 }
454
455 pte = alloc_low_page();
456 paddr_last = phys_pte_init(pte, paddr, paddr_end, new_prot);
457
458 spin_lock(&init_mm.page_table_lock);
459 pmd_populate_kernel(&init_mm, pmd, pte);
460 spin_unlock(&init_mm.page_table_lock);
461 }
462 update_page_count(PG_LEVEL_2M, pages);
463 return paddr_last;
464 }
465
466 /*
467 * Create PUD level page table mapping for physical addresses. The virtual
468 * and physical address do not have to be aligned at this level. KASLR can
469 * randomize virtual addresses up to this level.
470 * It returns the last physical address mapped.
471 */
472 static unsigned long __meminit
473 phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
474 unsigned long page_size_mask)
475 {
476 unsigned long pages = 0, paddr_next;
477 unsigned long paddr_last = paddr_end;
478 unsigned long vaddr = (unsigned long)__va(paddr);
479 int i = pud_index(vaddr);
480
481 for (; i < PTRS_PER_PUD; i++, paddr = paddr_next) {
482 pud_t *pud;
483 pmd_t *pmd;
484 pgprot_t prot = PAGE_KERNEL;
485
486 vaddr = (unsigned long)__va(paddr);
487 pud = pud_page + pud_index(vaddr);
488 paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
489
490 if (paddr >= paddr_end) {
491 if (!after_bootmem &&
492 !e820_any_mapped(paddr & PUD_MASK, paddr_next,
493 E820_RAM) &&
494 !e820_any_mapped(paddr & PUD_MASK, paddr_next,
495 E820_RESERVED_KERN))
496 set_pud(pud, __pud(0));
497 continue;
498 }
499
500 if (!pud_none(*pud)) {
501 if (!pud_large(*pud)) {
502 pmd = pmd_offset(pud, 0);
503 paddr_last = phys_pmd_init(pmd, paddr,
504 paddr_end,
505 page_size_mask,
506 prot);
507 __flush_tlb_all();
508 continue;
509 }
510 /*
511 * If we are ok with PG_LEVEL_1G mapping, then we will
512 * use the existing mapping.
513 *
514 * Otherwise, we will split the gbpage mapping but use
515 * the same existing protection bits except for large
516 * page, so that we don't violate Intel's TLB
517 * Application note (317080) which says, while changing
518 * the page sizes, new and old translations should
519 * not differ with respect to page frame and
520 * attributes.
521 */
522 if (page_size_mask & (1 << PG_LEVEL_1G)) {
523 if (!after_bootmem)
524 pages++;
525 paddr_last = paddr_next;
526 continue;
527 }
528 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
529 }
530
531 if (page_size_mask & (1<<PG_LEVEL_1G)) {
532 pages++;
533 spin_lock(&init_mm.page_table_lock);
534 set_pte((pte_t *)pud,
535 pfn_pte((paddr & PUD_MASK) >> PAGE_SHIFT,
536 PAGE_KERNEL_LARGE));
537 spin_unlock(&init_mm.page_table_lock);
538 paddr_last = paddr_next;
539 continue;
540 }
541
542 pmd = alloc_low_page();
543 paddr_last = phys_pmd_init(pmd, paddr, paddr_end,
544 page_size_mask, prot);
545
546 spin_lock(&init_mm.page_table_lock);
547 pud_populate(&init_mm, pud, pmd);
548 spin_unlock(&init_mm.page_table_lock);
549 }
550 __flush_tlb_all();
551
552 update_page_count(PG_LEVEL_1G, pages);
553
554 return paddr_last;
555 }
556
557 /*
558 * Create page table mapping for the physical memory for specific physical
559 * addresses. The virtual and physical addresses have to be aligned on PMD level
560 * down. It returns the last physical address mapped.
561 */
562 unsigned long __meminit
563 kernel_physical_mapping_init(unsigned long paddr_start,
564 unsigned long paddr_end,
565 unsigned long page_size_mask)
566 {
567 bool pgd_changed = false;
568 unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next, paddr_last;
569
570 paddr_last = paddr_end;
571 vaddr = (unsigned long)__va(paddr_start);
572 vaddr_end = (unsigned long)__va(paddr_end);
573 vaddr_start = vaddr;
574
575 for (; vaddr < vaddr_end; vaddr = vaddr_next) {
576 pgd_t *pgd = pgd_offset_k(vaddr);
577 pud_t *pud;
578
579 vaddr_next = (vaddr & PGDIR_MASK) + PGDIR_SIZE;
580
581 if (pgd_val(*pgd)) {
582 pud = (pud_t *)pgd_page_vaddr(*pgd);
583 paddr_last = phys_pud_init(pud, __pa(vaddr),
584 __pa(vaddr_end),
585 page_size_mask);
586 continue;
587 }
588
589 pud = alloc_low_page();
590 paddr_last = phys_pud_init(pud, __pa(vaddr), __pa(vaddr_end),
591 page_size_mask);
592
593 spin_lock(&init_mm.page_table_lock);
594 pgd_populate(&init_mm, pgd, pud);
595 spin_unlock(&init_mm.page_table_lock);
596 pgd_changed = true;
597 }
598
599 if (pgd_changed)
600 sync_global_pgds(vaddr_start, vaddr_end - 1, 0);
601
602 __flush_tlb_all();
603
604 return paddr_last;
605 }
606
607 #ifndef CONFIG_NUMA
608 void __init initmem_init(void)
609 {
610 memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
611 }
612 #endif
613
614 void __init paging_init(void)
615 {
616 sparse_memory_present_with_active_regions(MAX_NUMNODES);
617 sparse_init();
618
619 /*
620 * clear the default setting with node 0
621 * note: don't use nodes_clear here, that is really clearing when
622 * numa support is not compiled in, and later node_set_state
623 * will not set it back.
624 */
625 node_clear_state(0, N_MEMORY);
626 if (N_MEMORY != N_NORMAL_MEMORY)
627 node_clear_state(0, N_NORMAL_MEMORY);
628
629 zone_sizes_init();
630 }
631
632 /*
633 * Memory hotplug specific functions
634 */
635 #ifdef CONFIG_MEMORY_HOTPLUG
636 /*
637 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
638 * updating.
639 */
640 static void update_end_of_memory_vars(u64 start, u64 size)
641 {
642 unsigned long end_pfn = PFN_UP(start + size);
643
644 if (end_pfn > max_pfn) {
645 max_pfn = end_pfn;
646 max_low_pfn = end_pfn;
647 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
648 }
649 }
650
651 /*
652 * Memory is added always to NORMAL zone. This means you will never get
653 * additional DMA/DMA32 memory.
654 */
655 int arch_add_memory(int nid, u64 start, u64 size, bool for_device)
656 {
657 struct pglist_data *pgdat = NODE_DATA(nid);
658 struct zone *zone = pgdat->node_zones +
659 zone_for_memory(nid, start, size, ZONE_NORMAL, for_device);
660 unsigned long start_pfn = start >> PAGE_SHIFT;
661 unsigned long nr_pages = size >> PAGE_SHIFT;
662 int ret;
663
664 init_memory_mapping(start, start + size);
665
666 ret = __add_pages(nid, zone, start_pfn, nr_pages);
667 WARN_ON_ONCE(ret);
668
669 /* update max_pfn, max_low_pfn and high_memory */
670 update_end_of_memory_vars(start, size);
671
672 return ret;
673 }
674 EXPORT_SYMBOL_GPL(arch_add_memory);
675
676 #define PAGE_INUSE 0xFD
677
678 static void __meminit free_pagetable(struct page *page, int order)
679 {
680 unsigned long magic;
681 unsigned int nr_pages = 1 << order;
682 struct vmem_altmap *altmap = to_vmem_altmap((unsigned long) page);
683
684 if (altmap) {
685 vmem_altmap_free(altmap, nr_pages);
686 return;
687 }
688
689 /* bootmem page has reserved flag */
690 if (PageReserved(page)) {
691 __ClearPageReserved(page);
692
693 magic = (unsigned long)page->lru.next;
694 if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
695 while (nr_pages--)
696 put_page_bootmem(page++);
697 } else
698 while (nr_pages--)
699 free_reserved_page(page++);
700 } else
701 free_pages((unsigned long)page_address(page), order);
702 }
703
704 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
705 {
706 pte_t *pte;
707 int i;
708
709 for (i = 0; i < PTRS_PER_PTE; i++) {
710 pte = pte_start + i;
711 if (!pte_none(*pte))
712 return;
713 }
714
715 /* free a pte talbe */
716 free_pagetable(pmd_page(*pmd), 0);
717 spin_lock(&init_mm.page_table_lock);
718 pmd_clear(pmd);
719 spin_unlock(&init_mm.page_table_lock);
720 }
721
722 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud)
723 {
724 pmd_t *pmd;
725 int i;
726
727 for (i = 0; i < PTRS_PER_PMD; i++) {
728 pmd = pmd_start + i;
729 if (!pmd_none(*pmd))
730 return;
731 }
732
733 /* free a pmd talbe */
734 free_pagetable(pud_page(*pud), 0);
735 spin_lock(&init_mm.page_table_lock);
736 pud_clear(pud);
737 spin_unlock(&init_mm.page_table_lock);
738 }
739
740 static void __meminit
741 remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end,
742 bool direct)
743 {
744 unsigned long next, pages = 0;
745 pte_t *pte;
746 void *page_addr;
747 phys_addr_t phys_addr;
748
749 pte = pte_start + pte_index(addr);
750 for (; addr < end; addr = next, pte++) {
751 next = (addr + PAGE_SIZE) & PAGE_MASK;
752 if (next > end)
753 next = end;
754
755 if (!pte_present(*pte))
756 continue;
757
758 /*
759 * We mapped [0,1G) memory as identity mapping when
760 * initializing, in arch/x86/kernel/head_64.S. These
761 * pagetables cannot be removed.
762 */
763 phys_addr = pte_val(*pte) + (addr & PAGE_MASK);
764 if (phys_addr < (phys_addr_t)0x40000000)
765 return;
766
767 if (PAGE_ALIGNED(addr) && PAGE_ALIGNED(next)) {
768 /*
769 * Do not free direct mapping pages since they were
770 * freed when offlining, or simplely not in use.
771 */
772 if (!direct)
773 free_pagetable(pte_page(*pte), 0);
774
775 spin_lock(&init_mm.page_table_lock);
776 pte_clear(&init_mm, addr, pte);
777 spin_unlock(&init_mm.page_table_lock);
778
779 /* For non-direct mapping, pages means nothing. */
780 pages++;
781 } else {
782 /*
783 * If we are here, we are freeing vmemmap pages since
784 * direct mapped memory ranges to be freed are aligned.
785 *
786 * If we are not removing the whole page, it means
787 * other page structs in this page are being used and
788 * we canot remove them. So fill the unused page_structs
789 * with 0xFD, and remove the page when it is wholly
790 * filled with 0xFD.
791 */
792 memset((void *)addr, PAGE_INUSE, next - addr);
793
794 page_addr = page_address(pte_page(*pte));
795 if (!memchr_inv(page_addr, PAGE_INUSE, PAGE_SIZE)) {
796 free_pagetable(pte_page(*pte), 0);
797
798 spin_lock(&init_mm.page_table_lock);
799 pte_clear(&init_mm, addr, pte);
800 spin_unlock(&init_mm.page_table_lock);
801 }
802 }
803 }
804
805 /* Call free_pte_table() in remove_pmd_table(). */
806 flush_tlb_all();
807 if (direct)
808 update_page_count(PG_LEVEL_4K, -pages);
809 }
810
811 static void __meminit
812 remove_pmd_table(pmd_t *pmd_start, unsigned long addr, unsigned long end,
813 bool direct)
814 {
815 unsigned long next, pages = 0;
816 pte_t *pte_base;
817 pmd_t *pmd;
818 void *page_addr;
819
820 pmd = pmd_start + pmd_index(addr);
821 for (; addr < end; addr = next, pmd++) {
822 next = pmd_addr_end(addr, end);
823
824 if (!pmd_present(*pmd))
825 continue;
826
827 if (pmd_large(*pmd)) {
828 if (IS_ALIGNED(addr, PMD_SIZE) &&
829 IS_ALIGNED(next, PMD_SIZE)) {
830 if (!direct)
831 free_pagetable(pmd_page(*pmd),
832 get_order(PMD_SIZE));
833
834 spin_lock(&init_mm.page_table_lock);
835 pmd_clear(pmd);
836 spin_unlock(&init_mm.page_table_lock);
837 pages++;
838 } else {
839 /* If here, we are freeing vmemmap pages. */
840 memset((void *)addr, PAGE_INUSE, next - addr);
841
842 page_addr = page_address(pmd_page(*pmd));
843 if (!memchr_inv(page_addr, PAGE_INUSE,
844 PMD_SIZE)) {
845 free_pagetable(pmd_page(*pmd),
846 get_order(PMD_SIZE));
847
848 spin_lock(&init_mm.page_table_lock);
849 pmd_clear(pmd);
850 spin_unlock(&init_mm.page_table_lock);
851 }
852 }
853
854 continue;
855 }
856
857 pte_base = (pte_t *)pmd_page_vaddr(*pmd);
858 remove_pte_table(pte_base, addr, next, direct);
859 free_pte_table(pte_base, pmd);
860 }
861
862 /* Call free_pmd_table() in remove_pud_table(). */
863 if (direct)
864 update_page_count(PG_LEVEL_2M, -pages);
865 }
866
867 static void __meminit
868 remove_pud_table(pud_t *pud_start, unsigned long addr, unsigned long end,
869 bool direct)
870 {
871 unsigned long next, pages = 0;
872 pmd_t *pmd_base;
873 pud_t *pud;
874 void *page_addr;
875
876 pud = pud_start + pud_index(addr);
877 for (; addr < end; addr = next, pud++) {
878 next = pud_addr_end(addr, end);
879
880 if (!pud_present(*pud))
881 continue;
882
883 if (pud_large(*pud)) {
884 if (IS_ALIGNED(addr, PUD_SIZE) &&
885 IS_ALIGNED(next, PUD_SIZE)) {
886 if (!direct)
887 free_pagetable(pud_page(*pud),
888 get_order(PUD_SIZE));
889
890 spin_lock(&init_mm.page_table_lock);
891 pud_clear(pud);
892 spin_unlock(&init_mm.page_table_lock);
893 pages++;
894 } else {
895 /* If here, we are freeing vmemmap pages. */
896 memset((void *)addr, PAGE_INUSE, next - addr);
897
898 page_addr = page_address(pud_page(*pud));
899 if (!memchr_inv(page_addr, PAGE_INUSE,
900 PUD_SIZE)) {
901 free_pagetable(pud_page(*pud),
902 get_order(PUD_SIZE));
903
904 spin_lock(&init_mm.page_table_lock);
905 pud_clear(pud);
906 spin_unlock(&init_mm.page_table_lock);
907 }
908 }
909
910 continue;
911 }
912
913 pmd_base = (pmd_t *)pud_page_vaddr(*pud);
914 remove_pmd_table(pmd_base, addr, next, direct);
915 free_pmd_table(pmd_base, pud);
916 }
917
918 if (direct)
919 update_page_count(PG_LEVEL_1G, -pages);
920 }
921
922 /* start and end are both virtual address. */
923 static void __meminit
924 remove_pagetable(unsigned long start, unsigned long end, bool direct)
925 {
926 unsigned long next;
927 unsigned long addr;
928 pgd_t *pgd;
929 pud_t *pud;
930
931 for (addr = start; addr < end; addr = next) {
932 next = pgd_addr_end(addr, end);
933
934 pgd = pgd_offset_k(addr);
935 if (!pgd_present(*pgd))
936 continue;
937
938 pud = (pud_t *)pgd_page_vaddr(*pgd);
939 remove_pud_table(pud, addr, next, direct);
940 }
941
942 flush_tlb_all();
943 }
944
945 void __ref vmemmap_free(unsigned long start, unsigned long end)
946 {
947 remove_pagetable(start, end, false);
948 }
949
950 #ifdef CONFIG_MEMORY_HOTREMOVE
951 static void __meminit
952 kernel_physical_mapping_remove(unsigned long start, unsigned long end)
953 {
954 start = (unsigned long)__va(start);
955 end = (unsigned long)__va(end);
956
957 remove_pagetable(start, end, true);
958 }
959
960 int __ref arch_remove_memory(u64 start, u64 size)
961 {
962 unsigned long start_pfn = start >> PAGE_SHIFT;
963 unsigned long nr_pages = size >> PAGE_SHIFT;
964 struct page *page = pfn_to_page(start_pfn);
965 struct vmem_altmap *altmap;
966 struct zone *zone;
967 int ret;
968
969 /* With altmap the first mapped page is offset from @start */
970 altmap = to_vmem_altmap((unsigned long) page);
971 if (altmap)
972 page += vmem_altmap_offset(altmap);
973 zone = page_zone(page);
974 ret = __remove_pages(zone, start_pfn, nr_pages);
975 WARN_ON_ONCE(ret);
976 kernel_physical_mapping_remove(start, start + size);
977
978 return ret;
979 }
980 #endif
981 #endif /* CONFIG_MEMORY_HOTPLUG */
982
983 static struct kcore_list kcore_vsyscall;
984
985 static void __init register_page_bootmem_info(void)
986 {
987 #ifdef CONFIG_NUMA
988 int i;
989
990 for_each_online_node(i)
991 register_page_bootmem_info_node(NODE_DATA(i));
992 #endif
993 }
994
995 void __init mem_init(void)
996 {
997 pci_iommu_alloc();
998
999 /* clear_bss() already clear the empty_zero_page */
1000
1001 register_page_bootmem_info();
1002
1003 /* this will put all memory onto the freelists */
1004 free_all_bootmem();
1005 after_bootmem = 1;
1006
1007 /* Register memory areas for /proc/kcore */
1008 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_ADDR,
1009 PAGE_SIZE, KCORE_OTHER);
1010
1011 mem_init_print_info(NULL);
1012 }
1013
1014 const int rodata_test_data = 0xC3;
1015 EXPORT_SYMBOL_GPL(rodata_test_data);
1016
1017 int kernel_set_to_readonly;
1018
1019 void set_kernel_text_rw(void)
1020 {
1021 unsigned long start = PFN_ALIGN(_text);
1022 unsigned long end = PFN_ALIGN(__stop___ex_table);
1023
1024 if (!kernel_set_to_readonly)
1025 return;
1026
1027 pr_debug("Set kernel text: %lx - %lx for read write\n",
1028 start, end);
1029
1030 /*
1031 * Make the kernel identity mapping for text RW. Kernel text
1032 * mapping will always be RO. Refer to the comment in
1033 * static_protections() in pageattr.c
1034 */
1035 set_memory_rw(start, (end - start) >> PAGE_SHIFT);
1036 }
1037
1038 void set_kernel_text_ro(void)
1039 {
1040 unsigned long start = PFN_ALIGN(_text);
1041 unsigned long end = PFN_ALIGN(__stop___ex_table);
1042
1043 if (!kernel_set_to_readonly)
1044 return;
1045
1046 pr_debug("Set kernel text: %lx - %lx for read only\n",
1047 start, end);
1048
1049 /*
1050 * Set the kernel identity mapping for text RO.
1051 */
1052 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1053 }
1054
1055 void mark_rodata_ro(void)
1056 {
1057 unsigned long start = PFN_ALIGN(_text);
1058 unsigned long rodata_start = PFN_ALIGN(__start_rodata);
1059 unsigned long end = (unsigned long) &__end_rodata_hpage_align;
1060 unsigned long text_end = PFN_ALIGN(&__stop___ex_table);
1061 unsigned long rodata_end = PFN_ALIGN(&__end_rodata);
1062 unsigned long all_end;
1063
1064 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1065 (end - start) >> 10);
1066 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1067
1068 kernel_set_to_readonly = 1;
1069
1070 /*
1071 * The rodata/data/bss/brk section (but not the kernel text!)
1072 * should also be not-executable.
1073 *
1074 * We align all_end to PMD_SIZE because the existing mapping
1075 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1076 * split the PMD and the reminder between _brk_end and the end
1077 * of the PMD will remain mapped executable.
1078 *
1079 * Any PMD which was setup after the one which covers _brk_end
1080 * has been zapped already via cleanup_highmem().
1081 */
1082 all_end = roundup((unsigned long)_brk_end, PMD_SIZE);
1083 set_memory_nx(text_end, (all_end - text_end) >> PAGE_SHIFT);
1084
1085 rodata_test();
1086
1087 #ifdef CONFIG_CPA_DEBUG
1088 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1089 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1090
1091 printk(KERN_INFO "Testing CPA: again\n");
1092 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1093 #endif
1094
1095 free_init_pages("unused kernel",
1096 (unsigned long) __va(__pa_symbol(text_end)),
1097 (unsigned long) __va(__pa_symbol(rodata_start)));
1098 free_init_pages("unused kernel",
1099 (unsigned long) __va(__pa_symbol(rodata_end)),
1100 (unsigned long) __va(__pa_symbol(_sdata)));
1101
1102 debug_checkwx();
1103 }
1104
1105 int kern_addr_valid(unsigned long addr)
1106 {
1107 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1108 pgd_t *pgd;
1109 pud_t *pud;
1110 pmd_t *pmd;
1111 pte_t *pte;
1112
1113 if (above != 0 && above != -1UL)
1114 return 0;
1115
1116 pgd = pgd_offset_k(addr);
1117 if (pgd_none(*pgd))
1118 return 0;
1119
1120 pud = pud_offset(pgd, addr);
1121 if (pud_none(*pud))
1122 return 0;
1123
1124 if (pud_large(*pud))
1125 return pfn_valid(pud_pfn(*pud));
1126
1127 pmd = pmd_offset(pud, addr);
1128 if (pmd_none(*pmd))
1129 return 0;
1130
1131 if (pmd_large(*pmd))
1132 return pfn_valid(pmd_pfn(*pmd));
1133
1134 pte = pte_offset_kernel(pmd, addr);
1135 if (pte_none(*pte))
1136 return 0;
1137
1138 return pfn_valid(pte_pfn(*pte));
1139 }
1140
1141 static unsigned long probe_memory_block_size(void)
1142 {
1143 unsigned long bz = MIN_MEMORY_BLOCK_SIZE;
1144
1145 /* if system is UV or has 64GB of RAM or more, use large blocks */
1146 if (is_uv_system() || ((max_pfn << PAGE_SHIFT) >= (64UL << 30)))
1147 bz = 2UL << 30; /* 2GB */
1148
1149 pr_info("x86/mm: Memory block size: %ldMB\n", bz >> 20);
1150
1151 return bz;
1152 }
1153
1154 static unsigned long memory_block_size_probed;
1155 unsigned long memory_block_size_bytes(void)
1156 {
1157 if (!memory_block_size_probed)
1158 memory_block_size_probed = probe_memory_block_size();
1159
1160 return memory_block_size_probed;
1161 }
1162
1163 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1164 /*
1165 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1166 */
1167 static long __meminitdata addr_start, addr_end;
1168 static void __meminitdata *p_start, *p_end;
1169 static int __meminitdata node_start;
1170
1171 static int __meminit vmemmap_populate_hugepages(unsigned long start,
1172 unsigned long end, int node, struct vmem_altmap *altmap)
1173 {
1174 unsigned long addr;
1175 unsigned long next;
1176 pgd_t *pgd;
1177 pud_t *pud;
1178 pmd_t *pmd;
1179
1180 for (addr = start; addr < end; addr = next) {
1181 next = pmd_addr_end(addr, end);
1182
1183 pgd = vmemmap_pgd_populate(addr, node);
1184 if (!pgd)
1185 return -ENOMEM;
1186
1187 pud = vmemmap_pud_populate(pgd, addr, node);
1188 if (!pud)
1189 return -ENOMEM;
1190
1191 pmd = pmd_offset(pud, addr);
1192 if (pmd_none(*pmd)) {
1193 void *p;
1194
1195 p = __vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
1196 if (p) {
1197 pte_t entry;
1198
1199 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1200 PAGE_KERNEL_LARGE);
1201 set_pmd(pmd, __pmd(pte_val(entry)));
1202
1203 /* check to see if we have contiguous blocks */
1204 if (p_end != p || node_start != node) {
1205 if (p_start)
1206 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1207 addr_start, addr_end-1, p_start, p_end-1, node_start);
1208 addr_start = addr;
1209 node_start = node;
1210 p_start = p;
1211 }
1212
1213 addr_end = addr + PMD_SIZE;
1214 p_end = p + PMD_SIZE;
1215 continue;
1216 } else if (altmap)
1217 return -ENOMEM; /* no fallback */
1218 } else if (pmd_large(*pmd)) {
1219 vmemmap_verify((pte_t *)pmd, node, addr, next);
1220 continue;
1221 }
1222 pr_warn_once("vmemmap: falling back to regular page backing\n");
1223 if (vmemmap_populate_basepages(addr, next, node))
1224 return -ENOMEM;
1225 }
1226 return 0;
1227 }
1228
1229 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
1230 {
1231 struct vmem_altmap *altmap = to_vmem_altmap(start);
1232 int err;
1233
1234 if (boot_cpu_has(X86_FEATURE_PSE))
1235 err = vmemmap_populate_hugepages(start, end, node, altmap);
1236 else if (altmap) {
1237 pr_err_once("%s: no cpu support for altmap allocations\n",
1238 __func__);
1239 err = -ENOMEM;
1240 } else
1241 err = vmemmap_populate_basepages(start, end, node);
1242 if (!err)
1243 sync_global_pgds(start, end - 1, 0);
1244 return err;
1245 }
1246
1247 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1248 void register_page_bootmem_memmap(unsigned long section_nr,
1249 struct page *start_page, unsigned long size)
1250 {
1251 unsigned long addr = (unsigned long)start_page;
1252 unsigned long end = (unsigned long)(start_page + size);
1253 unsigned long next;
1254 pgd_t *pgd;
1255 pud_t *pud;
1256 pmd_t *pmd;
1257 unsigned int nr_pages;
1258 struct page *page;
1259
1260 for (; addr < end; addr = next) {
1261 pte_t *pte = NULL;
1262
1263 pgd = pgd_offset_k(addr);
1264 if (pgd_none(*pgd)) {
1265 next = (addr + PAGE_SIZE) & PAGE_MASK;
1266 continue;
1267 }
1268 get_page_bootmem(section_nr, pgd_page(*pgd), MIX_SECTION_INFO);
1269
1270 pud = pud_offset(pgd, addr);
1271 if (pud_none(*pud)) {
1272 next = (addr + PAGE_SIZE) & PAGE_MASK;
1273 continue;
1274 }
1275 get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
1276
1277 if (!boot_cpu_has(X86_FEATURE_PSE)) {
1278 next = (addr + PAGE_SIZE) & PAGE_MASK;
1279 pmd = pmd_offset(pud, addr);
1280 if (pmd_none(*pmd))
1281 continue;
1282 get_page_bootmem(section_nr, pmd_page(*pmd),
1283 MIX_SECTION_INFO);
1284
1285 pte = pte_offset_kernel(pmd, addr);
1286 if (pte_none(*pte))
1287 continue;
1288 get_page_bootmem(section_nr, pte_page(*pte),
1289 SECTION_INFO);
1290 } else {
1291 next = pmd_addr_end(addr, end);
1292
1293 pmd = pmd_offset(pud, addr);
1294 if (pmd_none(*pmd))
1295 continue;
1296
1297 nr_pages = 1 << (get_order(PMD_SIZE));
1298 page = pmd_page(*pmd);
1299 while (nr_pages--)
1300 get_page_bootmem(section_nr, page++,
1301 SECTION_INFO);
1302 }
1303 }
1304 }
1305 #endif
1306
1307 void __meminit vmemmap_populate_print_last(void)
1308 {
1309 if (p_start) {
1310 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1311 addr_start, addr_end-1, p_start, p_end-1, node_start);
1312 p_start = NULL;
1313 p_end = NULL;
1314 node_start = 0;
1315 }
1316 }
1317 #endif
Linux kernel - Deliverables
This page took 0.089763 seconds and 5 git commands to generate.