projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
x86/mm: Update physical mapping variable names
[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_val(*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_val(*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 have to be aligned at this level.
469 * It returns the last physical address mapped.
470 */
471 static unsigned long __meminit
472 phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
473 unsigned long page_size_mask)
474 {
475 unsigned long pages = 0, paddr_next;
476 unsigned long paddr_last = paddr_end;
477 int i = pud_index(paddr);
478
479 for (; i < PTRS_PER_PUD; i++, paddr = paddr_next) {
480 pud_t *pud = pud_page + pud_index(paddr);
481 pmd_t *pmd;
482 pgprot_t prot = PAGE_KERNEL;
483
484 paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
485 if (paddr >= paddr_end) {
486 if (!after_bootmem &&
487 !e820_any_mapped(paddr & PUD_MASK, paddr_next,
488 E820_RAM) &&
489 !e820_any_mapped(paddr & PUD_MASK, paddr_next,
490 E820_RESERVED_KERN))
491 set_pud(pud, __pud(0));
492 continue;
493 }
494
495 if (pud_val(*pud)) {
496 if (!pud_large(*pud)) {
497 pmd = pmd_offset(pud, 0);
498 paddr_last = phys_pmd_init(pmd, paddr,
499 paddr_end,
500 page_size_mask,
501 prot);
502 __flush_tlb_all();
503 continue;
504 }
505 /*
506 * If we are ok with PG_LEVEL_1G mapping, then we will
507 * use the existing mapping.
508 *
509 * Otherwise, we will split the gbpage mapping but use
510 * the same existing protection bits except for large
511 * page, so that we don't violate Intel's TLB
512 * Application note (317080) which says, while changing
513 * the page sizes, new and old translations should
514 * not differ with respect to page frame and
515 * attributes.
516 */
517 if (page_size_mask & (1 << PG_LEVEL_1G)) {
518 if (!after_bootmem)
519 pages++;
520 paddr_last = paddr_next;
521 continue;
522 }
523 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
524 }
525
526 if (page_size_mask & (1<<PG_LEVEL_1G)) {
527 pages++;
528 spin_lock(&init_mm.page_table_lock);
529 set_pte((pte_t *)pud,
530 pfn_pte((paddr & PUD_MASK) >> PAGE_SHIFT,
531 PAGE_KERNEL_LARGE));
532 spin_unlock(&init_mm.page_table_lock);
533 paddr_last = paddr_next;
534 continue;
535 }
536
537 pmd = alloc_low_page();
538 paddr_last = phys_pmd_init(pmd, paddr, paddr_end,
539 page_size_mask, prot);
540
541 spin_lock(&init_mm.page_table_lock);
542 pud_populate(&init_mm, pud, pmd);
543 spin_unlock(&init_mm.page_table_lock);
544 }
545 __flush_tlb_all();
546
547 update_page_count(PG_LEVEL_1G, pages);
548
549 return paddr_last;
550 }
551
552 /*
553 * Create page table mapping for the physical memory for specific physical
554 * addresses. The virtual and physical addresses have to be aligned on PUD level
555 * down. It returns the last physical address mapped.
556 */
557 unsigned long __meminit
558 kernel_physical_mapping_init(unsigned long paddr_start,
559 unsigned long paddr_end,
560 unsigned long page_size_mask)
561 {
562 bool pgd_changed = false;
563 unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next, paddr_last;
564
565 paddr_last = paddr_end;
566 vaddr = (unsigned long)__va(paddr_start);
567 vaddr_end = (unsigned long)__va(paddr_end);
568 vaddr_start = vaddr;
569
570 for (; vaddr < vaddr_end; vaddr = vaddr_next) {
571 pgd_t *pgd = pgd_offset_k(vaddr);
572 pud_t *pud;
573
574 vaddr_next = (vaddr & PGDIR_MASK) + PGDIR_SIZE;
575
576 if (pgd_val(*pgd)) {
577 pud = (pud_t *)pgd_page_vaddr(*pgd);
578 paddr_last = phys_pud_init(pud, __pa(vaddr),
579 __pa(vaddr_end),
580 page_size_mask);
581 continue;
582 }
583
584 pud = alloc_low_page();
585 paddr_last = phys_pud_init(pud, __pa(vaddr), __pa(vaddr_end),
586 page_size_mask);
587
588 spin_lock(&init_mm.page_table_lock);
589 pgd_populate(&init_mm, pgd, pud);
590 spin_unlock(&init_mm.page_table_lock);
591 pgd_changed = true;
592 }
593
594 if (pgd_changed)
595 sync_global_pgds(vaddr_start, vaddr_end - 1, 0);
596
597 __flush_tlb_all();
598
599 return paddr_last;
600 }
601
602 #ifndef CONFIG_NUMA
603 void __init initmem_init(void)
604 {
605 memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
606 }
607 #endif
608
609 void __init paging_init(void)
610 {
611 sparse_memory_present_with_active_regions(MAX_NUMNODES);
612 sparse_init();
613
614 /*
615 * clear the default setting with node 0
616 * note: don't use nodes_clear here, that is really clearing when
617 * numa support is not compiled in, and later node_set_state
618 * will not set it back.
619 */
620 node_clear_state(0, N_MEMORY);
621 if (N_MEMORY != N_NORMAL_MEMORY)
622 node_clear_state(0, N_NORMAL_MEMORY);
623
624 zone_sizes_init();
625 }
626
627 /*
628 * Memory hotplug specific functions
629 */
630 #ifdef CONFIG_MEMORY_HOTPLUG
631 /*
632 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
633 * updating.
634 */
635 static void update_end_of_memory_vars(u64 start, u64 size)
636 {
637 unsigned long end_pfn = PFN_UP(start + size);
638
639 if (end_pfn > max_pfn) {
640 max_pfn = end_pfn;
641 max_low_pfn = end_pfn;
642 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
643 }
644 }
645
646 /*
647 * Memory is added always to NORMAL zone. This means you will never get
648 * additional DMA/DMA32 memory.
649 */
650 int arch_add_memory(int nid, u64 start, u64 size, bool for_device)
651 {
652 struct pglist_data *pgdat = NODE_DATA(nid);
653 struct zone *zone = pgdat->node_zones +
654 zone_for_memory(nid, start, size, ZONE_NORMAL, for_device);
655 unsigned long start_pfn = start >> PAGE_SHIFT;
656 unsigned long nr_pages = size >> PAGE_SHIFT;
657 int ret;
658
659 init_memory_mapping(start, start + size);
660
661 ret = __add_pages(nid, zone, start_pfn, nr_pages);
662 WARN_ON_ONCE(ret);
663
664 /* update max_pfn, max_low_pfn and high_memory */
665 update_end_of_memory_vars(start, size);
666
667 return ret;
668 }
669 EXPORT_SYMBOL_GPL(arch_add_memory);
670
671 #define PAGE_INUSE 0xFD
672
673 static void __meminit free_pagetable(struct page *page, int order)
674 {
675 unsigned long magic;
676 unsigned int nr_pages = 1 << order;
677 struct vmem_altmap *altmap = to_vmem_altmap((unsigned long) page);
678
679 if (altmap) {
680 vmem_altmap_free(altmap, nr_pages);
681 return;
682 }
683
684 /* bootmem page has reserved flag */
685 if (PageReserved(page)) {
686 __ClearPageReserved(page);
687
688 magic = (unsigned long)page->lru.next;
689 if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
690 while (nr_pages--)
691 put_page_bootmem(page++);
692 } else
693 while (nr_pages--)
694 free_reserved_page(page++);
695 } else
696 free_pages((unsigned long)page_address(page), order);
697 }
698
699 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
700 {
701 pte_t *pte;
702 int i;
703
704 for (i = 0; i < PTRS_PER_PTE; i++) {
705 pte = pte_start + i;
706 if (pte_val(*pte))
707 return;
708 }
709
710 /* free a pte talbe */
711 free_pagetable(pmd_page(*pmd), 0);
712 spin_lock(&init_mm.page_table_lock);
713 pmd_clear(pmd);
714 spin_unlock(&init_mm.page_table_lock);
715 }
716
717 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud)
718 {
719 pmd_t *pmd;
720 int i;
721
722 for (i = 0; i < PTRS_PER_PMD; i++) {
723 pmd = pmd_start + i;
724 if (pmd_val(*pmd))
725 return;
726 }
727
728 /* free a pmd talbe */
729 free_pagetable(pud_page(*pud), 0);
730 spin_lock(&init_mm.page_table_lock);
731 pud_clear(pud);
732 spin_unlock(&init_mm.page_table_lock);
733 }
734
735 /* Return true if pgd is changed, otherwise return false. */
736 static bool __meminit free_pud_table(pud_t *pud_start, pgd_t *pgd)
737 {
738 pud_t *pud;
739 int i;
740
741 for (i = 0; i < PTRS_PER_PUD; i++) {
742 pud = pud_start + i;
743 if (pud_val(*pud))
744 return false;
745 }
746
747 /* free a pud table */
748 free_pagetable(pgd_page(*pgd), 0);
749 spin_lock(&init_mm.page_table_lock);
750 pgd_clear(pgd);
751 spin_unlock(&init_mm.page_table_lock);
752
753 return true;
754 }
755
756 static void __meminit
757 remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end,
758 bool direct)
759 {
760 unsigned long next, pages = 0;
761 pte_t *pte;
762 void *page_addr;
763 phys_addr_t phys_addr;
764
765 pte = pte_start + pte_index(addr);
766 for (; addr < end; addr = next, pte++) {
767 next = (addr + PAGE_SIZE) & PAGE_MASK;
768 if (next > end)
769 next = end;
770
771 if (!pte_present(*pte))
772 continue;
773
774 /*
775 * We mapped [0,1G) memory as identity mapping when
776 * initializing, in arch/x86/kernel/head_64.S. These
777 * pagetables cannot be removed.
778 */
779 phys_addr = pte_val(*pte) + (addr & PAGE_MASK);
780 if (phys_addr < (phys_addr_t)0x40000000)
781 return;
782
783 if (PAGE_ALIGNED(addr) && PAGE_ALIGNED(next)) {
784 /*
785 * Do not free direct mapping pages since they were
786 * freed when offlining, or simplely not in use.
787 */
788 if (!direct)
789 free_pagetable(pte_page(*pte), 0);
790
791 spin_lock(&init_mm.page_table_lock);
792 pte_clear(&init_mm, addr, pte);
793 spin_unlock(&init_mm.page_table_lock);
794
795 /* For non-direct mapping, pages means nothing. */
796 pages++;
797 } else {
798 /*
799 * If we are here, we are freeing vmemmap pages since
800 * direct mapped memory ranges to be freed are aligned.
801 *
802 * If we are not removing the whole page, it means
803 * other page structs in this page are being used and
804 * we canot remove them. So fill the unused page_structs
805 * with 0xFD, and remove the page when it is wholly
806 * filled with 0xFD.
807 */
808 memset((void *)addr, PAGE_INUSE, next - addr);
809
810 page_addr = page_address(pte_page(*pte));
811 if (!memchr_inv(page_addr, PAGE_INUSE, PAGE_SIZE)) {
812 free_pagetable(pte_page(*pte), 0);
813
814 spin_lock(&init_mm.page_table_lock);
815 pte_clear(&init_mm, addr, pte);
816 spin_unlock(&init_mm.page_table_lock);
817 }
818 }
819 }
820
821 /* Call free_pte_table() in remove_pmd_table(). */
822 flush_tlb_all();
823 if (direct)
824 update_page_count(PG_LEVEL_4K, -pages);
825 }
826
827 static void __meminit
828 remove_pmd_table(pmd_t *pmd_start, unsigned long addr, unsigned long end,
829 bool direct)
830 {
831 unsigned long next, pages = 0;
832 pte_t *pte_base;
833 pmd_t *pmd;
834 void *page_addr;
835
836 pmd = pmd_start + pmd_index(addr);
837 for (; addr < end; addr = next, pmd++) {
838 next = pmd_addr_end(addr, end);
839
840 if (!pmd_present(*pmd))
841 continue;
842
843 if (pmd_large(*pmd)) {
844 if (IS_ALIGNED(addr, PMD_SIZE) &&
845 IS_ALIGNED(next, PMD_SIZE)) {
846 if (!direct)
847 free_pagetable(pmd_page(*pmd),
848 get_order(PMD_SIZE));
849
850 spin_lock(&init_mm.page_table_lock);
851 pmd_clear(pmd);
852 spin_unlock(&init_mm.page_table_lock);
853 pages++;
854 } else {
855 /* If here, we are freeing vmemmap pages. */
856 memset((void *)addr, PAGE_INUSE, next - addr);
857
858 page_addr = page_address(pmd_page(*pmd));
859 if (!memchr_inv(page_addr, PAGE_INUSE,
860 PMD_SIZE)) {
861 free_pagetable(pmd_page(*pmd),
862 get_order(PMD_SIZE));
863
864 spin_lock(&init_mm.page_table_lock);
865 pmd_clear(pmd);
866 spin_unlock(&init_mm.page_table_lock);
867 }
868 }
869
870 continue;
871 }
872
873 pte_base = (pte_t *)pmd_page_vaddr(*pmd);
874 remove_pte_table(pte_base, addr, next, direct);
875 free_pte_table(pte_base, pmd);
876 }
877
878 /* Call free_pmd_table() in remove_pud_table(). */
879 if (direct)
880 update_page_count(PG_LEVEL_2M, -pages);
881 }
882
883 static void __meminit
884 remove_pud_table(pud_t *pud_start, unsigned long addr, unsigned long end,
885 bool direct)
886 {
887 unsigned long next, pages = 0;
888 pmd_t *pmd_base;
889 pud_t *pud;
890 void *page_addr;
891
892 pud = pud_start + pud_index(addr);
893 for (; addr < end; addr = next, pud++) {
894 next = pud_addr_end(addr, end);
895
896 if (!pud_present(*pud))
897 continue;
898
899 if (pud_large(*pud)) {
900 if (IS_ALIGNED(addr, PUD_SIZE) &&
901 IS_ALIGNED(next, PUD_SIZE)) {
902 if (!direct)
903 free_pagetable(pud_page(*pud),
904 get_order(PUD_SIZE));
905
906 spin_lock(&init_mm.page_table_lock);
907 pud_clear(pud);
908 spin_unlock(&init_mm.page_table_lock);
909 pages++;
910 } else {
911 /* If here, we are freeing vmemmap pages. */
912 memset((void *)addr, PAGE_INUSE, next - addr);
913
914 page_addr = page_address(pud_page(*pud));
915 if (!memchr_inv(page_addr, PAGE_INUSE,
916 PUD_SIZE)) {
917 free_pagetable(pud_page(*pud),
918 get_order(PUD_SIZE));
919
920 spin_lock(&init_mm.page_table_lock);
921 pud_clear(pud);
922 spin_unlock(&init_mm.page_table_lock);
923 }
924 }
925
926 continue;
927 }
928
929 pmd_base = (pmd_t *)pud_page_vaddr(*pud);
930 remove_pmd_table(pmd_base, addr, next, direct);
931 free_pmd_table(pmd_base, pud);
932 }
933
934 if (direct)
935 update_page_count(PG_LEVEL_1G, -pages);
936 }
937
938 /* start and end are both virtual address. */
939 static void __meminit
940 remove_pagetable(unsigned long start, unsigned long end, bool direct)
941 {
942 unsigned long next;
943 unsigned long addr;
944 pgd_t *pgd;
945 pud_t *pud;
946 bool pgd_changed = false;
947
948 for (addr = start; addr < end; addr = next) {
949 next = pgd_addr_end(addr, end);
950
951 pgd = pgd_offset_k(addr);
952 if (!pgd_present(*pgd))
953 continue;
954
955 pud = (pud_t *)pgd_page_vaddr(*pgd);
956 remove_pud_table(pud, addr, next, direct);
957 if (free_pud_table(pud, pgd))
958 pgd_changed = true;
959 }
960
961 if (pgd_changed)
962 sync_global_pgds(start, end - 1, 1);
963
964 flush_tlb_all();
965 }
966
967 void __ref vmemmap_free(unsigned long start, unsigned long end)
968 {
969 remove_pagetable(start, end, false);
970 }
971
972 #ifdef CONFIG_MEMORY_HOTREMOVE
973 static void __meminit
974 kernel_physical_mapping_remove(unsigned long start, unsigned long end)
975 {
976 start = (unsigned long)__va(start);
977 end = (unsigned long)__va(end);
978
979 remove_pagetable(start, end, true);
980 }
981
982 int __ref arch_remove_memory(u64 start, u64 size)
983 {
984 unsigned long start_pfn = start >> PAGE_SHIFT;
985 unsigned long nr_pages = size >> PAGE_SHIFT;
986 struct page *page = pfn_to_page(start_pfn);
987 struct vmem_altmap *altmap;
988 struct zone *zone;
989 int ret;
990
991 /* With altmap the first mapped page is offset from @start */
992 altmap = to_vmem_altmap((unsigned long) page);
993 if (altmap)
994 page += vmem_altmap_offset(altmap);
995 zone = page_zone(page);
996 ret = __remove_pages(zone, start_pfn, nr_pages);
997 WARN_ON_ONCE(ret);
998 kernel_physical_mapping_remove(start, start + size);
999
1000 return ret;
1001 }
1002 #endif
1003 #endif /* CONFIG_MEMORY_HOTPLUG */
1004
1005 static struct kcore_list kcore_vsyscall;
1006
1007 static void __init register_page_bootmem_info(void)
1008 {
1009 #ifdef CONFIG_NUMA
1010 int i;
1011
1012 for_each_online_node(i)
1013 register_page_bootmem_info_node(NODE_DATA(i));
1014 #endif
1015 }
1016
1017 void __init mem_init(void)
1018 {
1019 pci_iommu_alloc();
1020
1021 /* clear_bss() already clear the empty_zero_page */
1022
1023 register_page_bootmem_info();
1024
1025 /* this will put all memory onto the freelists */
1026 free_all_bootmem();
1027 after_bootmem = 1;
1028
1029 /* Register memory areas for /proc/kcore */
1030 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_ADDR,
1031 PAGE_SIZE, KCORE_OTHER);
1032
1033 mem_init_print_info(NULL);
1034 }
1035
1036 const int rodata_test_data = 0xC3;
1037 EXPORT_SYMBOL_GPL(rodata_test_data);
1038
1039 int kernel_set_to_readonly;
1040
1041 void set_kernel_text_rw(void)
1042 {
1043 unsigned long start = PFN_ALIGN(_text);
1044 unsigned long end = PFN_ALIGN(__stop___ex_table);
1045
1046 if (!kernel_set_to_readonly)
1047 return;
1048
1049 pr_debug("Set kernel text: %lx - %lx for read write\n",
1050 start, end);
1051
1052 /*
1053 * Make the kernel identity mapping for text RW. Kernel text
1054 * mapping will always be RO. Refer to the comment in
1055 * static_protections() in pageattr.c
1056 */
1057 set_memory_rw(start, (end - start) >> PAGE_SHIFT);
1058 }
1059
1060 void set_kernel_text_ro(void)
1061 {
1062 unsigned long start = PFN_ALIGN(_text);
1063 unsigned long end = PFN_ALIGN(__stop___ex_table);
1064
1065 if (!kernel_set_to_readonly)
1066 return;
1067
1068 pr_debug("Set kernel text: %lx - %lx for read only\n",
1069 start, end);
1070
1071 /*
1072 * Set the kernel identity mapping for text RO.
1073 */
1074 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1075 }
1076
1077 void mark_rodata_ro(void)
1078 {
1079 unsigned long start = PFN_ALIGN(_text);
1080 unsigned long rodata_start = PFN_ALIGN(__start_rodata);
1081 unsigned long end = (unsigned long) &__end_rodata_hpage_align;
1082 unsigned long text_end = PFN_ALIGN(&__stop___ex_table);
1083 unsigned long rodata_end = PFN_ALIGN(&__end_rodata);
1084 unsigned long all_end;
1085
1086 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1087 (end - start) >> 10);
1088 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1089
1090 kernel_set_to_readonly = 1;
1091
1092 /*
1093 * The rodata/data/bss/brk section (but not the kernel text!)
1094 * should also be not-executable.
1095 *
1096 * We align all_end to PMD_SIZE because the existing mapping
1097 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1098 * split the PMD and the reminder between _brk_end and the end
1099 * of the PMD will remain mapped executable.
1100 *
1101 * Any PMD which was setup after the one which covers _brk_end
1102 * has been zapped already via cleanup_highmem().
1103 */
1104 all_end = roundup((unsigned long)_brk_end, PMD_SIZE);
1105 set_memory_nx(text_end, (all_end - text_end) >> PAGE_SHIFT);
1106
1107 rodata_test();
1108
1109 #ifdef CONFIG_CPA_DEBUG
1110 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1111 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1112
1113 printk(KERN_INFO "Testing CPA: again\n");
1114 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1115 #endif
1116
1117 free_init_pages("unused kernel",
1118 (unsigned long) __va(__pa_symbol(text_end)),
1119 (unsigned long) __va(__pa_symbol(rodata_start)));
1120 free_init_pages("unused kernel",
1121 (unsigned long) __va(__pa_symbol(rodata_end)),
1122 (unsigned long) __va(__pa_symbol(_sdata)));
1123
1124 debug_checkwx();
1125 }
1126
1127 int kern_addr_valid(unsigned long addr)
1128 {
1129 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1130 pgd_t *pgd;
1131 pud_t *pud;
1132 pmd_t *pmd;
1133 pte_t *pte;
1134
1135 if (above != 0 && above != -1UL)
1136 return 0;
1137
1138 pgd = pgd_offset_k(addr);
1139 if (pgd_none(*pgd))
1140 return 0;
1141
1142 pud = pud_offset(pgd, addr);
1143 if (pud_none(*pud))
1144 return 0;
1145
1146 if (pud_large(*pud))
1147 return pfn_valid(pud_pfn(*pud));
1148
1149 pmd = pmd_offset(pud, addr);
1150 if (pmd_none(*pmd))
1151 return 0;
1152
1153 if (pmd_large(*pmd))
1154 return pfn_valid(pmd_pfn(*pmd));
1155
1156 pte = pte_offset_kernel(pmd, addr);
1157 if (pte_none(*pte))
1158 return 0;
1159
1160 return pfn_valid(pte_pfn(*pte));
1161 }
1162
1163 static unsigned long probe_memory_block_size(void)
1164 {
1165 unsigned long bz = MIN_MEMORY_BLOCK_SIZE;
1166
1167 /* if system is UV or has 64GB of RAM or more, use large blocks */
1168 if (is_uv_system() || ((max_pfn << PAGE_SHIFT) >= (64UL << 30)))
1169 bz = 2UL << 30; /* 2GB */
1170
1171 pr_info("x86/mm: Memory block size: %ldMB\n", bz >> 20);
1172
1173 return bz;
1174 }
1175
1176 static unsigned long memory_block_size_probed;
1177 unsigned long memory_block_size_bytes(void)
1178 {
1179 if (!memory_block_size_probed)
1180 memory_block_size_probed = probe_memory_block_size();
1181
1182 return memory_block_size_probed;
1183 }
1184
1185 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1186 /*
1187 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1188 */
1189 static long __meminitdata addr_start, addr_end;
1190 static void __meminitdata *p_start, *p_end;
1191 static int __meminitdata node_start;
1192
1193 static int __meminit vmemmap_populate_hugepages(unsigned long start,
1194 unsigned long end, int node, struct vmem_altmap *altmap)
1195 {
1196 unsigned long addr;
1197 unsigned long next;
1198 pgd_t *pgd;
1199 pud_t *pud;
1200 pmd_t *pmd;
1201
1202 for (addr = start; addr < end; addr = next) {
1203 next = pmd_addr_end(addr, end);
1204
1205 pgd = vmemmap_pgd_populate(addr, node);
1206 if (!pgd)
1207 return -ENOMEM;
1208
1209 pud = vmemmap_pud_populate(pgd, addr, node);
1210 if (!pud)
1211 return -ENOMEM;
1212
1213 pmd = pmd_offset(pud, addr);
1214 if (pmd_none(*pmd)) {
1215 void *p;
1216
1217 p = __vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
1218 if (p) {
1219 pte_t entry;
1220
1221 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1222 PAGE_KERNEL_LARGE);
1223 set_pmd(pmd, __pmd(pte_val(entry)));
1224
1225 /* check to see if we have contiguous blocks */
1226 if (p_end != p || node_start != node) {
1227 if (p_start)
1228 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1229 addr_start, addr_end-1, p_start, p_end-1, node_start);
1230 addr_start = addr;
1231 node_start = node;
1232 p_start = p;
1233 }
1234
1235 addr_end = addr + PMD_SIZE;
1236 p_end = p + PMD_SIZE;
1237 continue;
1238 } else if (altmap)
1239 return -ENOMEM; /* no fallback */
1240 } else if (pmd_large(*pmd)) {
1241 vmemmap_verify((pte_t *)pmd, node, addr, next);
1242 continue;
1243 }
1244 pr_warn_once("vmemmap: falling back to regular page backing\n");
1245 if (vmemmap_populate_basepages(addr, next, node))
1246 return -ENOMEM;
1247 }
1248 return 0;
1249 }
1250
1251 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
1252 {
1253 struct vmem_altmap *altmap = to_vmem_altmap(start);
1254 int err;
1255
1256 if (boot_cpu_has(X86_FEATURE_PSE))
1257 err = vmemmap_populate_hugepages(start, end, node, altmap);
1258 else if (altmap) {
1259 pr_err_once("%s: no cpu support for altmap allocations\n",
1260 __func__);
1261 err = -ENOMEM;
1262 } else
1263 err = vmemmap_populate_basepages(start, end, node);
1264 if (!err)
1265 sync_global_pgds(start, end - 1, 0);
1266 return err;
1267 }
1268
1269 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1270 void register_page_bootmem_memmap(unsigned long section_nr,
1271 struct page *start_page, unsigned long size)
1272 {
1273 unsigned long addr = (unsigned long)start_page;
1274 unsigned long end = (unsigned long)(start_page + size);
1275 unsigned long next;
1276 pgd_t *pgd;
1277 pud_t *pud;
1278 pmd_t *pmd;
1279 unsigned int nr_pages;
1280 struct page *page;
1281
1282 for (; addr < end; addr = next) {
1283 pte_t *pte = NULL;
1284
1285 pgd = pgd_offset_k(addr);
1286 if (pgd_none(*pgd)) {
1287 next = (addr + PAGE_SIZE) & PAGE_MASK;
1288 continue;
1289 }
1290 get_page_bootmem(section_nr, pgd_page(*pgd), MIX_SECTION_INFO);
1291
1292 pud = pud_offset(pgd, addr);
1293 if (pud_none(*pud)) {
1294 next = (addr + PAGE_SIZE) & PAGE_MASK;
1295 continue;
1296 }
1297 get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
1298
1299 if (!boot_cpu_has(X86_FEATURE_PSE)) {
1300 next = (addr + PAGE_SIZE) & PAGE_MASK;
1301 pmd = pmd_offset(pud, addr);
1302 if (pmd_none(*pmd))
1303 continue;
1304 get_page_bootmem(section_nr, pmd_page(*pmd),
1305 MIX_SECTION_INFO);
1306
1307 pte = pte_offset_kernel(pmd, addr);
1308 if (pte_none(*pte))
1309 continue;
1310 get_page_bootmem(section_nr, pte_page(*pte),
1311 SECTION_INFO);
1312 } else {
1313 next = pmd_addr_end(addr, end);
1314
1315 pmd = pmd_offset(pud, addr);
1316 if (pmd_none(*pmd))
1317 continue;
1318
1319 nr_pages = 1 << (get_order(PMD_SIZE));
1320 page = pmd_page(*pmd);
1321 while (nr_pages--)
1322 get_page_bootmem(section_nr, page++,
1323 SECTION_INFO);
1324 }
1325 }
1326 }
1327 #endif
1328
1329 void __meminit vmemmap_populate_print_last(void)
1330 {
1331 if (p_start) {
1332 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1333 addr_start, addr_end-1, p_start, p_end-1, node_start);
1334 p_start = NULL;
1335 p_end = NULL;
1336 node_start = 0;
1337 }
1338 }
1339 #endif
Linux kernel - Deliverables
This page took 0.059084 seconds and 6 git commands to generate.