2 * linux/arch/x86_64/mm/init.c
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>
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>
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/memory.h>
31 #include <linux/memory_hotplug.h>
32 #include <linux/memremap.h>
33 #include <linux/nmi.h>
34 #include <linux/gfp.h>
35 #include <linux/kcore.h>
37 #include <asm/processor.h>
38 #include <asm/bios_ebda.h>
39 #include <linux/uaccess.h>
40 #include <asm/pgtable.h>
41 #include <asm/pgalloc.h>
43 #include <asm/fixmap.h>
44 #include <asm/e820/api.h>
47 #include <asm/mmu_context.h>
48 #include <asm/proto.h>
50 #include <asm/sections.h>
51 #include <asm/kdebug.h>
53 #include <asm/set_memory.h>
55 #include <asm/uv/uv.h>
56 #include <asm/setup.h>
58 #include "mm_internal.h"
60 #include "ident_map.c"
63 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
64 * physical space so we can cache the place of the first one and move
65 * around without checking the pgd every time.
68 pteval_t __supported_pte_mask __read_mostly
= ~0;
69 EXPORT_SYMBOL_GPL(__supported_pte_mask
);
71 int force_personality32
;
75 * Control non executable heap for 32bit processes.
76 * To control the stack too use noexec=off
78 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
79 * off PROT_READ implies PROT_EXEC
81 static int __init
nonx32_setup(char *str
)
83 if (!strcmp(str
, "on"))
84 force_personality32
&= ~READ_IMPLIES_EXEC
;
85 else if (!strcmp(str
, "off"))
86 force_personality32
|= READ_IMPLIES_EXEC
;
89 __setup("noexec32=", nonx32_setup
);
92 * When memory was added make sure all the processes MM have
93 * suitable PGD entries in the local PGD level page.
95 #ifdef CONFIG_X86_5LEVEL
96 void sync_global_pgds(unsigned long start
, unsigned long end
)
100 for (addr
= start
; addr
<= end
; addr
= ALIGN(addr
+ 1, PGDIR_SIZE
)) {
101 const pgd_t
*pgd_ref
= pgd_offset_k(addr
);
104 /* Check for overflow */
108 if (pgd_none(*pgd_ref
))
111 spin_lock(&pgd_lock
);
112 list_for_each_entry(page
, &pgd_list
, lru
) {
114 spinlock_t
*pgt_lock
;
116 pgd
= (pgd_t
*)page_address(page
) + pgd_index(addr
);
117 /* the pgt_lock only for Xen */
118 pgt_lock
= &pgd_page_get_mm(page
)->page_table_lock
;
121 if (!pgd_none(*pgd_ref
) && !pgd_none(*pgd
))
122 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
125 set_pgd(pgd
, *pgd_ref
);
127 spin_unlock(pgt_lock
);
129 spin_unlock(&pgd_lock
);
133 void sync_global_pgds(unsigned long start
, unsigned long end
)
137 for (addr
= start
; addr
<= end
; addr
= ALIGN(addr
+ 1, PGDIR_SIZE
)) {
138 pgd_t
*pgd_ref
= pgd_offset_k(addr
);
139 const p4d_t
*p4d_ref
;
143 * With folded p4d, pgd_none() is always false, we need to
144 * handle synchonization on p4d level.
146 BUILD_BUG_ON(pgd_none(*pgd_ref
));
147 p4d_ref
= p4d_offset(pgd_ref
, addr
);
149 if (p4d_none(*p4d_ref
))
152 spin_lock(&pgd_lock
);
153 list_for_each_entry(page
, &pgd_list
, lru
) {
156 spinlock_t
*pgt_lock
;
158 pgd
= (pgd_t
*)page_address(page
) + pgd_index(addr
);
159 p4d
= p4d_offset(pgd
, addr
);
160 /* the pgt_lock only for Xen */
161 pgt_lock
= &pgd_page_get_mm(page
)->page_table_lock
;
164 if (!p4d_none(*p4d_ref
) && !p4d_none(*p4d
))
165 BUG_ON(p4d_page_vaddr(*p4d
)
166 != p4d_page_vaddr(*p4d_ref
));
169 set_p4d(p4d
, *p4d_ref
);
171 spin_unlock(pgt_lock
);
173 spin_unlock(&pgd_lock
);
179 * NOTE: This function is marked __ref because it calls __init function
180 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
182 static __ref
void *spp_getpage(void)
187 ptr
= (void *) get_zeroed_page(GFP_ATOMIC
);
189 ptr
= alloc_bootmem_pages(PAGE_SIZE
);
191 if (!ptr
|| ((unsigned long)ptr
& ~PAGE_MASK
)) {
192 panic("set_pte_phys: cannot allocate page data %s\n",
193 after_bootmem
? "after bootmem" : "");
196 pr_debug("spp_getpage %p\n", ptr
);
201 static p4d_t
*fill_p4d(pgd_t
*pgd
, unsigned long vaddr
)
203 if (pgd_none(*pgd
)) {
204 p4d_t
*p4d
= (p4d_t
*)spp_getpage();
205 pgd_populate(&init_mm
, pgd
, p4d
);
206 if (p4d
!= p4d_offset(pgd
, 0))
207 printk(KERN_ERR
"PAGETABLE BUG #00! %p <-> %p\n",
208 p4d
, p4d_offset(pgd
, 0));
210 return p4d_offset(pgd
, vaddr
);
213 static pud_t
*fill_pud(p4d_t
*p4d
, unsigned long vaddr
)
215 if (p4d_none(*p4d
)) {
216 pud_t
*pud
= (pud_t
*)spp_getpage();
217 p4d_populate(&init_mm
, p4d
, pud
);
218 if (pud
!= pud_offset(p4d
, 0))
219 printk(KERN_ERR
"PAGETABLE BUG #01! %p <-> %p\n",
220 pud
, pud_offset(p4d
, 0));
222 return pud_offset(p4d
, vaddr
);
225 static pmd_t
*fill_pmd(pud_t
*pud
, unsigned long vaddr
)
227 if (pud_none(*pud
)) {
228 pmd_t
*pmd
= (pmd_t
*) spp_getpage();
229 pud_populate(&init_mm
, pud
, pmd
);
230 if (pmd
!= pmd_offset(pud
, 0))
231 printk(KERN_ERR
"PAGETABLE BUG #02! %p <-> %p\n",
232 pmd
, pmd_offset(pud
, 0));
234 return pmd_offset(pud
, vaddr
);
237 static pte_t
*fill_pte(pmd_t
*pmd
, unsigned long vaddr
)
239 if (pmd_none(*pmd
)) {
240 pte_t
*pte
= (pte_t
*) spp_getpage();
241 pmd_populate_kernel(&init_mm
, pmd
, pte
);
242 if (pte
!= pte_offset_kernel(pmd
, 0))
243 printk(KERN_ERR
"PAGETABLE BUG #03!\n");
245 return pte_offset_kernel(pmd
, vaddr
);
248 static void __set_pte_vaddr(pud_t
*pud
, unsigned long vaddr
, pte_t new_pte
)
250 pmd_t
*pmd
= fill_pmd(pud
, vaddr
);
251 pte_t
*pte
= fill_pte(pmd
, vaddr
);
253 set_pte(pte
, new_pte
);
256 * It's enough to flush this one mapping.
257 * (PGE mappings get flushed as well)
259 __flush_tlb_one(vaddr
);
262 void set_pte_vaddr_p4d(p4d_t
*p4d_page
, unsigned long vaddr
, pte_t new_pte
)
264 p4d_t
*p4d
= p4d_page
+ p4d_index(vaddr
);
265 pud_t
*pud
= fill_pud(p4d
, vaddr
);
267 __set_pte_vaddr(pud
, vaddr
, new_pte
);
270 void set_pte_vaddr_pud(pud_t
*pud_page
, unsigned long vaddr
, pte_t new_pte
)
272 pud_t
*pud
= pud_page
+ pud_index(vaddr
);
274 __set_pte_vaddr(pud
, vaddr
, new_pte
);
277 void set_pte_vaddr(unsigned long vaddr
, pte_t pteval
)
282 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr
, native_pte_val(pteval
));
284 pgd
= pgd_offset_k(vaddr
);
285 if (pgd_none(*pgd
)) {
287 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
291 p4d_page
= p4d_offset(pgd
, 0);
292 set_pte_vaddr_p4d(p4d_page
, vaddr
, pteval
);
295 pmd_t
* __init
populate_extra_pmd(unsigned long vaddr
)
301 pgd
= pgd_offset_k(vaddr
);
302 p4d
= fill_p4d(pgd
, vaddr
);
303 pud
= fill_pud(p4d
, vaddr
);
304 return fill_pmd(pud
, vaddr
);
307 pte_t
* __init
populate_extra_pte(unsigned long vaddr
)
311 pmd
= populate_extra_pmd(vaddr
);
312 return fill_pte(pmd
, vaddr
);
316 * Create large page table mappings for a range of physical addresses.
318 static void __init
__init_extra_mapping(unsigned long phys
, unsigned long size
,
319 enum page_cache_mode cache
)
327 pgprot_val(prot
) = pgprot_val(PAGE_KERNEL_LARGE
) |
328 pgprot_val(pgprot_4k_2_large(cachemode2pgprot(cache
)));
329 BUG_ON((phys
& ~PMD_MASK
) || (size
& ~PMD_MASK
));
330 for (; size
; phys
+= PMD_SIZE
, size
-= PMD_SIZE
) {
331 pgd
= pgd_offset_k((unsigned long)__va(phys
));
332 if (pgd_none(*pgd
)) {
333 p4d
= (p4d_t
*) spp_getpage();
334 set_pgd(pgd
, __pgd(__pa(p4d
) | _KERNPG_TABLE
|
337 p4d
= p4d_offset(pgd
, (unsigned long)__va(phys
));
338 if (p4d_none(*p4d
)) {
339 pud
= (pud_t
*) spp_getpage();
340 set_p4d(p4d
, __p4d(__pa(pud
) | _KERNPG_TABLE
|
343 pud
= pud_offset(p4d
, (unsigned long)__va(phys
));
344 if (pud_none(*pud
)) {
345 pmd
= (pmd_t
*) spp_getpage();
346 set_pud(pud
, __pud(__pa(pmd
) | _KERNPG_TABLE
|
349 pmd
= pmd_offset(pud
, phys
);
350 BUG_ON(!pmd_none(*pmd
));
351 set_pmd(pmd
, __pmd(phys
| pgprot_val(prot
)));
355 void __init
init_extra_mapping_wb(unsigned long phys
, unsigned long size
)
357 __init_extra_mapping(phys
, size
, _PAGE_CACHE_MODE_WB
);
360 void __init
init_extra_mapping_uc(unsigned long phys
, unsigned long size
)
362 __init_extra_mapping(phys
, size
, _PAGE_CACHE_MODE_UC
);
366 * The head.S code sets up the kernel high mapping:
368 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
370 * phys_base holds the negative offset to the kernel, which is added
371 * to the compile time generated pmds. This results in invalid pmds up
372 * to the point where we hit the physaddr 0 mapping.
374 * We limit the mappings to the region from _text to _brk_end. _brk_end
375 * is rounded up to the 2MB boundary. This catches the invalid pmds as
376 * well, as they are located before _text:
378 void __init
cleanup_highmap(void)
380 unsigned long vaddr
= __START_KERNEL_map
;
381 unsigned long vaddr_end
= __START_KERNEL_map
+ KERNEL_IMAGE_SIZE
;
382 unsigned long end
= roundup((unsigned long)_brk_end
, PMD_SIZE
) - 1;
383 pmd_t
*pmd
= level2_kernel_pgt
;
386 * Native path, max_pfn_mapped is not set yet.
387 * Xen has valid max_pfn_mapped set in
388 * arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
391 vaddr_end
= __START_KERNEL_map
+ (max_pfn_mapped
<< PAGE_SHIFT
);
393 for (; vaddr
+ PMD_SIZE
- 1 < vaddr_end
; pmd
++, vaddr
+= PMD_SIZE
) {
396 if (vaddr
< (unsigned long) _text
|| vaddr
> end
)
397 set_pmd(pmd
, __pmd(0));
402 * Create PTE level page table mapping for physical addresses.
403 * It returns the last physical address mapped.
405 static unsigned long __meminit
406 phys_pte_init(pte_t
*pte_page
, unsigned long paddr
, unsigned long paddr_end
,
409 unsigned long pages
= 0, paddr_next
;
410 unsigned long paddr_last
= paddr_end
;
414 pte
= pte_page
+ pte_index(paddr
);
415 i
= pte_index(paddr
);
417 for (; i
< PTRS_PER_PTE
; i
++, paddr
= paddr_next
, pte
++) {
418 paddr_next
= (paddr
& PAGE_MASK
) + PAGE_SIZE
;
419 if (paddr
>= paddr_end
) {
420 if (!after_bootmem
&&
421 !e820__mapped_any(paddr
& PAGE_MASK
, paddr_next
,
423 !e820__mapped_any(paddr
& PAGE_MASK
, paddr_next
,
424 E820_TYPE_RESERVED_KERN
))
425 set_pte(pte
, __pte(0));
430 * We will re-use the existing mapping.
431 * Xen for example has some special requirements, like mapping
432 * pagetable pages as RO. So assume someone who pre-setup
433 * these mappings are more intelligent.
435 if (!pte_none(*pte
)) {
442 pr_info(" pte=%p addr=%lx pte=%016lx\n", pte
, paddr
,
443 pfn_pte(paddr
>> PAGE_SHIFT
, PAGE_KERNEL
).pte
);
445 set_pte(pte
, pfn_pte(paddr
>> PAGE_SHIFT
, prot
));
446 paddr_last
= (paddr
& PAGE_MASK
) + PAGE_SIZE
;
449 update_page_count(PG_LEVEL_4K
, pages
);
455 * Create PMD level page table mapping for physical addresses. The virtual
456 * and physical address have to be aligned at this level.
457 * It returns the last physical address mapped.
459 static unsigned long __meminit
460 phys_pmd_init(pmd_t
*pmd_page
, unsigned long paddr
, unsigned long paddr_end
,
461 unsigned long page_size_mask
, pgprot_t prot
)
463 unsigned long pages
= 0, paddr_next
;
464 unsigned long paddr_last
= paddr_end
;
466 int i
= pmd_index(paddr
);
468 for (; i
< PTRS_PER_PMD
; i
++, paddr
= paddr_next
) {
469 pmd_t
*pmd
= pmd_page
+ pmd_index(paddr
);
471 pgprot_t new_prot
= prot
;
473 paddr_next
= (paddr
& PMD_MASK
) + PMD_SIZE
;
474 if (paddr
>= paddr_end
) {
475 if (!after_bootmem
&&
476 !e820__mapped_any(paddr
& PMD_MASK
, paddr_next
,
478 !e820__mapped_any(paddr
& PMD_MASK
, paddr_next
,
479 E820_TYPE_RESERVED_KERN
))
480 set_pmd(pmd
, __pmd(0));
484 if (!pmd_none(*pmd
)) {
485 if (!pmd_large(*pmd
)) {
486 spin_lock(&init_mm
.page_table_lock
);
487 pte
= (pte_t
*)pmd_page_vaddr(*pmd
);
488 paddr_last
= phys_pte_init(pte
, paddr
,
490 spin_unlock(&init_mm
.page_table_lock
);
494 * If we are ok with PG_LEVEL_2M mapping, then we will
495 * use the existing mapping,
497 * Otherwise, we will split the large page mapping but
498 * use the same existing protection bits except for
499 * large page, so that we don't violate Intel's TLB
500 * Application note (317080) which says, while changing
501 * the page sizes, new and old translations should
502 * not differ with respect to page frame and
505 if (page_size_mask
& (1 << PG_LEVEL_2M
)) {
508 paddr_last
= paddr_next
;
511 new_prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pmd
));
514 if (page_size_mask
& (1<<PG_LEVEL_2M
)) {
516 spin_lock(&init_mm
.page_table_lock
);
517 set_pte((pte_t
*)pmd
,
518 pfn_pte((paddr
& PMD_MASK
) >> PAGE_SHIFT
,
519 __pgprot(pgprot_val(prot
) | _PAGE_PSE
)));
520 spin_unlock(&init_mm
.page_table_lock
);
521 paddr_last
= paddr_next
;
525 pte
= alloc_low_page();
526 paddr_last
= phys_pte_init(pte
, paddr
, paddr_end
, new_prot
);
528 spin_lock(&init_mm
.page_table_lock
);
529 pmd_populate_kernel(&init_mm
, pmd
, pte
);
530 spin_unlock(&init_mm
.page_table_lock
);
532 update_page_count(PG_LEVEL_2M
, pages
);
537 * Create PUD level page table mapping for physical addresses. The virtual
538 * and physical address do not have to be aligned at this level. KASLR can
539 * randomize virtual addresses up to this level.
540 * It returns the last physical address mapped.
542 static unsigned long __meminit
543 phys_pud_init(pud_t
*pud_page
, unsigned long paddr
, unsigned long paddr_end
,
544 unsigned long page_size_mask
)
546 unsigned long pages
= 0, paddr_next
;
547 unsigned long paddr_last
= paddr_end
;
548 unsigned long vaddr
= (unsigned long)__va(paddr
);
549 int i
= pud_index(vaddr
);
551 for (; i
< PTRS_PER_PUD
; i
++, paddr
= paddr_next
) {
554 pgprot_t prot
= PAGE_KERNEL
;
556 vaddr
= (unsigned long)__va(paddr
);
557 pud
= pud_page
+ pud_index(vaddr
);
558 paddr_next
= (paddr
& PUD_MASK
) + PUD_SIZE
;
560 if (paddr
>= paddr_end
) {
561 if (!after_bootmem
&&
562 !e820__mapped_any(paddr
& PUD_MASK
, paddr_next
,
564 !e820__mapped_any(paddr
& PUD_MASK
, paddr_next
,
565 E820_TYPE_RESERVED_KERN
))
566 set_pud(pud
, __pud(0));
570 if (!pud_none(*pud
)) {
571 if (!pud_large(*pud
)) {
572 pmd
= pmd_offset(pud
, 0);
573 paddr_last
= phys_pmd_init(pmd
, paddr
,
581 * If we are ok with PG_LEVEL_1G mapping, then we will
582 * use the existing mapping.
584 * Otherwise, we will split the gbpage mapping but use
585 * the same existing protection bits except for large
586 * page, so that we don't violate Intel's TLB
587 * Application note (317080) which says, while changing
588 * the page sizes, new and old translations should
589 * not differ with respect to page frame and
592 if (page_size_mask
& (1 << PG_LEVEL_1G
)) {
595 paddr_last
= paddr_next
;
598 prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pud
));
601 if (page_size_mask
& (1<<PG_LEVEL_1G
)) {
603 spin_lock(&init_mm
.page_table_lock
);
604 set_pte((pte_t
*)pud
,
605 pfn_pte((paddr
& PUD_MASK
) >> PAGE_SHIFT
,
607 spin_unlock(&init_mm
.page_table_lock
);
608 paddr_last
= paddr_next
;
612 pmd
= alloc_low_page();
613 paddr_last
= phys_pmd_init(pmd
, paddr
, paddr_end
,
614 page_size_mask
, prot
);
616 spin_lock(&init_mm
.page_table_lock
);
617 pud_populate(&init_mm
, pud
, pmd
);
618 spin_unlock(&init_mm
.page_table_lock
);
622 update_page_count(PG_LEVEL_1G
, pages
);
627 static unsigned long __meminit
628 phys_p4d_init(p4d_t
*p4d_page
, unsigned long paddr
, unsigned long paddr_end
,
629 unsigned long page_size_mask
)
631 unsigned long paddr_next
, paddr_last
= paddr_end
;
632 unsigned long vaddr
= (unsigned long)__va(paddr
);
633 int i
= p4d_index(vaddr
);
635 if (!IS_ENABLED(CONFIG_X86_5LEVEL
))
636 return phys_pud_init((pud_t
*) p4d_page
, paddr
, paddr_end
, page_size_mask
);
638 for (; i
< PTRS_PER_P4D
; i
++, paddr
= paddr_next
) {
642 vaddr
= (unsigned long)__va(paddr
);
643 p4d
= p4d_page
+ p4d_index(vaddr
);
644 paddr_next
= (paddr
& P4D_MASK
) + P4D_SIZE
;
646 if (paddr
>= paddr_end
) {
647 if (!after_bootmem
&&
648 !e820__mapped_any(paddr
& P4D_MASK
, paddr_next
,
650 !e820__mapped_any(paddr
& P4D_MASK
, paddr_next
,
651 E820_TYPE_RESERVED_KERN
))
652 set_p4d(p4d
, __p4d(0));
656 if (!p4d_none(*p4d
)) {
657 pud
= pud_offset(p4d
, 0);
658 paddr_last
= phys_pud_init(pud
, paddr
,
665 pud
= alloc_low_page();
666 paddr_last
= phys_pud_init(pud
, paddr
, paddr_end
,
669 spin_lock(&init_mm
.page_table_lock
);
670 p4d_populate(&init_mm
, p4d
, pud
);
671 spin_unlock(&init_mm
.page_table_lock
);
679 * Create page table mapping for the physical memory for specific physical
680 * addresses. The virtual and physical addresses have to be aligned on PMD level
681 * down. It returns the last physical address mapped.
683 unsigned long __meminit
684 kernel_physical_mapping_init(unsigned long paddr_start
,
685 unsigned long paddr_end
,
686 unsigned long page_size_mask
)
688 bool pgd_changed
= false;
689 unsigned long vaddr
, vaddr_start
, vaddr_end
, vaddr_next
, paddr_last
;
691 paddr_last
= paddr_end
;
692 vaddr
= (unsigned long)__va(paddr_start
);
693 vaddr_end
= (unsigned long)__va(paddr_end
);
696 for (; vaddr
< vaddr_end
; vaddr
= vaddr_next
) {
697 pgd_t
*pgd
= pgd_offset_k(vaddr
);
700 vaddr_next
= (vaddr
& PGDIR_MASK
) + PGDIR_SIZE
;
703 p4d
= (p4d_t
*)pgd_page_vaddr(*pgd
);
704 paddr_last
= phys_p4d_init(p4d
, __pa(vaddr
),
710 p4d
= alloc_low_page();
711 paddr_last
= phys_p4d_init(p4d
, __pa(vaddr
), __pa(vaddr_end
),
714 spin_lock(&init_mm
.page_table_lock
);
715 if (IS_ENABLED(CONFIG_X86_5LEVEL
))
716 pgd_populate(&init_mm
, pgd
, p4d
);
718 p4d_populate(&init_mm
, p4d_offset(pgd
, vaddr
), (pud_t
*) p4d
);
719 spin_unlock(&init_mm
.page_table_lock
);
724 sync_global_pgds(vaddr_start
, vaddr_end
- 1);
732 void __init
initmem_init(void)
734 memblock_set_node(0, (phys_addr_t
)ULLONG_MAX
, &memblock
.memory
, 0);
738 void __init
paging_init(void)
740 sparse_memory_present_with_active_regions(MAX_NUMNODES
);
744 * clear the default setting with node 0
745 * note: don't use nodes_clear here, that is really clearing when
746 * numa support is not compiled in, and later node_set_state
747 * will not set it back.
749 node_clear_state(0, N_MEMORY
);
750 if (N_MEMORY
!= N_NORMAL_MEMORY
)
751 node_clear_state(0, N_NORMAL_MEMORY
);
757 * Memory hotplug specific functions
759 #ifdef CONFIG_MEMORY_HOTPLUG
761 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
764 static void update_end_of_memory_vars(u64 start
, u64 size
)
766 unsigned long end_pfn
= PFN_UP(start
+ size
);
768 if (end_pfn
> max_pfn
) {
770 max_low_pfn
= end_pfn
;
771 high_memory
= (void *)__va(max_pfn
* PAGE_SIZE
- 1) + 1;
775 int add_pages(int nid
, unsigned long start_pfn
,
776 unsigned long nr_pages
, bool want_memblock
)
780 ret
= __add_pages(nid
, start_pfn
, nr_pages
, want_memblock
);
783 /* update max_pfn, max_low_pfn and high_memory */
784 update_end_of_memory_vars(start_pfn
<< PAGE_SHIFT
,
785 nr_pages
<< PAGE_SHIFT
);
790 int arch_add_memory(int nid
, u64 start
, u64 size
, bool want_memblock
)
792 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
793 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
795 init_memory_mapping(start
, start
+ size
);
797 return add_pages(nid
, start_pfn
, nr_pages
, want_memblock
);
799 EXPORT_SYMBOL_GPL(arch_add_memory
);
801 #define PAGE_INUSE 0xFD
803 static void __meminit
free_pagetable(struct page
*page
, int order
)
806 unsigned int nr_pages
= 1 << order
;
807 struct vmem_altmap
*altmap
= to_vmem_altmap((unsigned long) page
);
810 vmem_altmap_free(altmap
, nr_pages
);
814 /* bootmem page has reserved flag */
815 if (PageReserved(page
)) {
816 __ClearPageReserved(page
);
818 magic
= (unsigned long)page
->freelist
;
819 if (magic
== SECTION_INFO
|| magic
== MIX_SECTION_INFO
) {
821 put_page_bootmem(page
++);
824 free_reserved_page(page
++);
826 free_pages((unsigned long)page_address(page
), order
);
829 static void __meminit
free_pte_table(pte_t
*pte_start
, pmd_t
*pmd
)
834 for (i
= 0; i
< PTRS_PER_PTE
; i
++) {
840 /* free a pte talbe */
841 free_pagetable(pmd_page(*pmd
), 0);
842 spin_lock(&init_mm
.page_table_lock
);
844 spin_unlock(&init_mm
.page_table_lock
);
847 static void __meminit
free_pmd_table(pmd_t
*pmd_start
, pud_t
*pud
)
852 for (i
= 0; i
< PTRS_PER_PMD
; i
++) {
858 /* free a pmd talbe */
859 free_pagetable(pud_page(*pud
), 0);
860 spin_lock(&init_mm
.page_table_lock
);
862 spin_unlock(&init_mm
.page_table_lock
);
865 static void __meminit
free_pud_table(pud_t
*pud_start
, p4d_t
*p4d
)
870 for (i
= 0; i
< PTRS_PER_PUD
; i
++) {
876 /* free a pud talbe */
877 free_pagetable(p4d_page(*p4d
), 0);
878 spin_lock(&init_mm
.page_table_lock
);
880 spin_unlock(&init_mm
.page_table_lock
);
883 static void __meminit
884 remove_pte_table(pte_t
*pte_start
, unsigned long addr
, unsigned long end
,
887 unsigned long next
, pages
= 0;
890 phys_addr_t phys_addr
;
892 pte
= pte_start
+ pte_index(addr
);
893 for (; addr
< end
; addr
= next
, pte
++) {
894 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
898 if (!pte_present(*pte
))
902 * We mapped [0,1G) memory as identity mapping when
903 * initializing, in arch/x86/kernel/head_64.S. These
904 * pagetables cannot be removed.
906 phys_addr
= pte_val(*pte
) + (addr
& PAGE_MASK
);
907 if (phys_addr
< (phys_addr_t
)0x40000000)
910 if (PAGE_ALIGNED(addr
) && PAGE_ALIGNED(next
)) {
912 * Do not free direct mapping pages since they were
913 * freed when offlining, or simplely not in use.
916 free_pagetable(pte_page(*pte
), 0);
918 spin_lock(&init_mm
.page_table_lock
);
919 pte_clear(&init_mm
, addr
, pte
);
920 spin_unlock(&init_mm
.page_table_lock
);
922 /* For non-direct mapping, pages means nothing. */
926 * If we are here, we are freeing vmemmap pages since
927 * direct mapped memory ranges to be freed are aligned.
929 * If we are not removing the whole page, it means
930 * other page structs in this page are being used and
931 * we canot remove them. So fill the unused page_structs
932 * with 0xFD, and remove the page when it is wholly
935 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
937 page_addr
= page_address(pte_page(*pte
));
938 if (!memchr_inv(page_addr
, PAGE_INUSE
, PAGE_SIZE
)) {
939 free_pagetable(pte_page(*pte
), 0);
941 spin_lock(&init_mm
.page_table_lock
);
942 pte_clear(&init_mm
, addr
, pte
);
943 spin_unlock(&init_mm
.page_table_lock
);
948 /* Call free_pte_table() in remove_pmd_table(). */
951 update_page_count(PG_LEVEL_4K
, -pages
);
954 static void __meminit
955 remove_pmd_table(pmd_t
*pmd_start
, unsigned long addr
, unsigned long end
,
958 unsigned long next
, pages
= 0;
963 pmd
= pmd_start
+ pmd_index(addr
);
964 for (; addr
< end
; addr
= next
, pmd
++) {
965 next
= pmd_addr_end(addr
, end
);
967 if (!pmd_present(*pmd
))
970 if (pmd_large(*pmd
)) {
971 if (IS_ALIGNED(addr
, PMD_SIZE
) &&
972 IS_ALIGNED(next
, PMD_SIZE
)) {
974 free_pagetable(pmd_page(*pmd
),
975 get_order(PMD_SIZE
));
977 spin_lock(&init_mm
.page_table_lock
);
979 spin_unlock(&init_mm
.page_table_lock
);
982 /* If here, we are freeing vmemmap pages. */
983 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
985 page_addr
= page_address(pmd_page(*pmd
));
986 if (!memchr_inv(page_addr
, PAGE_INUSE
,
988 free_pagetable(pmd_page(*pmd
),
989 get_order(PMD_SIZE
));
991 spin_lock(&init_mm
.page_table_lock
);
993 spin_unlock(&init_mm
.page_table_lock
);
1000 pte_base
= (pte_t
*)pmd_page_vaddr(*pmd
);
1001 remove_pte_table(pte_base
, addr
, next
, direct
);
1002 free_pte_table(pte_base
, pmd
);
1005 /* Call free_pmd_table() in remove_pud_table(). */
1007 update_page_count(PG_LEVEL_2M
, -pages
);
1010 static void __meminit
1011 remove_pud_table(pud_t
*pud_start
, unsigned long addr
, unsigned long end
,
1014 unsigned long next
, pages
= 0;
1019 pud
= pud_start
+ pud_index(addr
);
1020 for (; addr
< end
; addr
= next
, pud
++) {
1021 next
= pud_addr_end(addr
, end
);
1023 if (!pud_present(*pud
))
1026 if (pud_large(*pud
)) {
1027 if (IS_ALIGNED(addr
, PUD_SIZE
) &&
1028 IS_ALIGNED(next
, PUD_SIZE
)) {
1030 free_pagetable(pud_page(*pud
),
1031 get_order(PUD_SIZE
));
1033 spin_lock(&init_mm
.page_table_lock
);
1035 spin_unlock(&init_mm
.page_table_lock
);
1038 /* If here, we are freeing vmemmap pages. */
1039 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
1041 page_addr
= page_address(pud_page(*pud
));
1042 if (!memchr_inv(page_addr
, PAGE_INUSE
,
1044 free_pagetable(pud_page(*pud
),
1045 get_order(PUD_SIZE
));
1047 spin_lock(&init_mm
.page_table_lock
);
1049 spin_unlock(&init_mm
.page_table_lock
);
1056 pmd_base
= pmd_offset(pud
, 0);
1057 remove_pmd_table(pmd_base
, addr
, next
, direct
);
1058 free_pmd_table(pmd_base
, pud
);
1062 update_page_count(PG_LEVEL_1G
, -pages
);
1065 static void __meminit
1066 remove_p4d_table(p4d_t
*p4d_start
, unsigned long addr
, unsigned long end
,
1069 unsigned long next
, pages
= 0;
1073 p4d
= p4d_start
+ p4d_index(addr
);
1074 for (; addr
< end
; addr
= next
, p4d
++) {
1075 next
= p4d_addr_end(addr
, end
);
1077 if (!p4d_present(*p4d
))
1080 BUILD_BUG_ON(p4d_large(*p4d
));
1082 pud_base
= pud_offset(p4d
, 0);
1083 remove_pud_table(pud_base
, addr
, next
, direct
);
1085 * For 4-level page tables we do not want to free PUDs, but in the
1086 * 5-level case we should free them. This code will have to change
1087 * to adapt for boot-time switching between 4 and 5 level page tables.
1089 if (CONFIG_PGTABLE_LEVELS
== 5)
1090 free_pud_table(pud_base
, p4d
);
1094 update_page_count(PG_LEVEL_512G
, -pages
);
1097 /* start and end are both virtual address. */
1098 static void __meminit
1099 remove_pagetable(unsigned long start
, unsigned long end
, bool direct
)
1106 for (addr
= start
; addr
< end
; addr
= next
) {
1107 next
= pgd_addr_end(addr
, end
);
1109 pgd
= pgd_offset_k(addr
);
1110 if (!pgd_present(*pgd
))
1113 p4d
= p4d_offset(pgd
, 0);
1114 remove_p4d_table(p4d
, addr
, next
, direct
);
1120 void __ref
vmemmap_free(unsigned long start
, unsigned long end
)
1122 remove_pagetable(start
, end
, false);
1125 #ifdef CONFIG_MEMORY_HOTREMOVE
1126 static void __meminit
1127 kernel_physical_mapping_remove(unsigned long start
, unsigned long end
)
1129 start
= (unsigned long)__va(start
);
1130 end
= (unsigned long)__va(end
);
1132 remove_pagetable(start
, end
, true);
1135 int __ref
arch_remove_memory(u64 start
, u64 size
)
1137 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1138 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
1139 struct page
*page
= pfn_to_page(start_pfn
);
1140 struct vmem_altmap
*altmap
;
1144 /* With altmap the first mapped page is offset from @start */
1145 altmap
= to_vmem_altmap((unsigned long) page
);
1147 page
+= vmem_altmap_offset(altmap
);
1148 zone
= page_zone(page
);
1149 ret
= __remove_pages(zone
, start_pfn
, nr_pages
);
1151 kernel_physical_mapping_remove(start
, start
+ size
);
1156 #endif /* CONFIG_MEMORY_HOTPLUG */
1158 static struct kcore_list kcore_vsyscall
;
1160 static void __init
register_page_bootmem_info(void)
1165 for_each_online_node(i
)
1166 register_page_bootmem_info_node(NODE_DATA(i
));
1170 void __init
mem_init(void)
1174 /* clear_bss() already clear the empty_zero_page */
1176 /* this will put all memory onto the freelists */
1181 * Must be done after boot memory is put on freelist, because here we
1182 * might set fields in deferred struct pages that have not yet been
1183 * initialized, and free_all_bootmem() initializes all the reserved
1184 * deferred pages for us.
1186 register_page_bootmem_info();
1188 /* Register memory areas for /proc/kcore */
1189 kclist_add(&kcore_vsyscall
, (void *)VSYSCALL_ADDR
,
1190 PAGE_SIZE
, KCORE_OTHER
);
1192 mem_init_print_info(NULL
);
1195 int kernel_set_to_readonly
;
1197 void set_kernel_text_rw(void)
1199 unsigned long start
= PFN_ALIGN(_text
);
1200 unsigned long end
= PFN_ALIGN(__stop___ex_table
);
1202 if (!kernel_set_to_readonly
)
1205 pr_debug("Set kernel text: %lx - %lx for read write\n",
1209 * Make the kernel identity mapping for text RW. Kernel text
1210 * mapping will always be RO. Refer to the comment in
1211 * static_protections() in pageattr.c
1213 set_memory_rw(start
, (end
- start
) >> PAGE_SHIFT
);
1216 void set_kernel_text_ro(void)
1218 unsigned long start
= PFN_ALIGN(_text
);
1219 unsigned long end
= PFN_ALIGN(__stop___ex_table
);
1221 if (!kernel_set_to_readonly
)
1224 pr_debug("Set kernel text: %lx - %lx for read only\n",
1228 * Set the kernel identity mapping for text RO.
1230 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
1233 void mark_rodata_ro(void)
1235 unsigned long start
= PFN_ALIGN(_text
);
1236 unsigned long rodata_start
= PFN_ALIGN(__start_rodata
);
1237 unsigned long end
= (unsigned long) &__end_rodata_hpage_align
;
1238 unsigned long text_end
= PFN_ALIGN(&__stop___ex_table
);
1239 unsigned long rodata_end
= PFN_ALIGN(&__end_rodata
);
1240 unsigned long all_end
;
1242 printk(KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
1243 (end
- start
) >> 10);
1244 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
1246 kernel_set_to_readonly
= 1;
1249 * The rodata/data/bss/brk section (but not the kernel text!)
1250 * should also be not-executable.
1252 * We align all_end to PMD_SIZE because the existing mapping
1253 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1254 * split the PMD and the reminder between _brk_end and the end
1255 * of the PMD will remain mapped executable.
1257 * Any PMD which was setup after the one which covers _brk_end
1258 * has been zapped already via cleanup_highmem().
1260 all_end
= roundup((unsigned long)_brk_end
, PMD_SIZE
);
1261 set_memory_nx(text_end
, (all_end
- text_end
) >> PAGE_SHIFT
);
1263 #ifdef CONFIG_CPA_DEBUG
1264 printk(KERN_INFO
"Testing CPA: undo %lx-%lx\n", start
, end
);
1265 set_memory_rw(start
, (end
-start
) >> PAGE_SHIFT
);
1267 printk(KERN_INFO
"Testing CPA: again\n");
1268 set_memory_ro(start
, (end
-start
) >> PAGE_SHIFT
);
1271 free_init_pages("unused kernel",
1272 (unsigned long) __va(__pa_symbol(text_end
)),
1273 (unsigned long) __va(__pa_symbol(rodata_start
)));
1274 free_init_pages("unused kernel",
1275 (unsigned long) __va(__pa_symbol(rodata_end
)),
1276 (unsigned long) __va(__pa_symbol(_sdata
)));
1281 int kern_addr_valid(unsigned long addr
)
1283 unsigned long above
= ((long)addr
) >> __VIRTUAL_MASK_SHIFT
;
1290 if (above
!= 0 && above
!= -1UL)
1293 pgd
= pgd_offset_k(addr
);
1297 p4d
= p4d_offset(pgd
, addr
);
1301 pud
= pud_offset(p4d
, addr
);
1305 if (pud_large(*pud
))
1306 return pfn_valid(pud_pfn(*pud
));
1308 pmd
= pmd_offset(pud
, addr
);
1312 if (pmd_large(*pmd
))
1313 return pfn_valid(pmd_pfn(*pmd
));
1315 pte
= pte_offset_kernel(pmd
, addr
);
1319 return pfn_valid(pte_pfn(*pte
));
1322 static unsigned long probe_memory_block_size(void)
1324 unsigned long bz
= MIN_MEMORY_BLOCK_SIZE
;
1326 /* if system is UV or has 64GB of RAM or more, use large blocks */
1327 if (is_uv_system() || ((max_pfn
<< PAGE_SHIFT
) >= (64UL << 30)))
1328 bz
= 2UL << 30; /* 2GB */
1330 pr_info("x86/mm: Memory block size: %ldMB\n", bz
>> 20);
1335 static unsigned long memory_block_size_probed
;
1336 unsigned long memory_block_size_bytes(void)
1338 if (!memory_block_size_probed
)
1339 memory_block_size_probed
= probe_memory_block_size();
1341 return memory_block_size_probed
;
1344 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1346 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1348 static long __meminitdata addr_start
, addr_end
;
1349 static void __meminitdata
*p_start
, *p_end
;
1350 static int __meminitdata node_start
;
1352 static int __meminit
vmemmap_populate_hugepages(unsigned long start
,
1353 unsigned long end
, int node
, struct vmem_altmap
*altmap
)
1362 for (addr
= start
; addr
< end
; addr
= next
) {
1363 next
= pmd_addr_end(addr
, end
);
1365 pgd
= vmemmap_pgd_populate(addr
, node
);
1369 p4d
= vmemmap_p4d_populate(pgd
, addr
, node
);
1373 pud
= vmemmap_pud_populate(p4d
, addr
, node
);
1377 pmd
= pmd_offset(pud
, addr
);
1378 if (pmd_none(*pmd
)) {
1381 p
= __vmemmap_alloc_block_buf(PMD_SIZE
, node
, altmap
);
1385 entry
= pfn_pte(__pa(p
) >> PAGE_SHIFT
,
1387 set_pmd(pmd
, __pmd(pte_val(entry
)));
1389 /* check to see if we have contiguous blocks */
1390 if (p_end
!= p
|| node_start
!= node
) {
1392 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1393 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);
1399 addr_end
= addr
+ PMD_SIZE
;
1400 p_end
= p
+ PMD_SIZE
;
1403 return -ENOMEM
; /* no fallback */
1404 } else if (pmd_large(*pmd
)) {
1405 vmemmap_verify((pte_t
*)pmd
, node
, addr
, next
);
1408 if (vmemmap_populate_basepages(addr
, next
, node
))
1414 int __meminit
vmemmap_populate(unsigned long start
, unsigned long end
, int node
)
1416 struct vmem_altmap
*altmap
= to_vmem_altmap(start
);
1419 if (boot_cpu_has(X86_FEATURE_PSE
))
1420 err
= vmemmap_populate_hugepages(start
, end
, node
, altmap
);
1422 pr_err_once("%s: no cpu support for altmap allocations\n",
1426 err
= vmemmap_populate_basepages(start
, end
, node
);
1428 sync_global_pgds(start
, end
- 1);
1432 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1433 void register_page_bootmem_memmap(unsigned long section_nr
,
1434 struct page
*start_page
, unsigned long nr_pages
)
1436 unsigned long addr
= (unsigned long)start_page
;
1437 unsigned long end
= (unsigned long)(start_page
+ nr_pages
);
1443 unsigned int nr_pmd_pages
;
1446 for (; addr
< end
; addr
= next
) {
1449 pgd
= pgd_offset_k(addr
);
1450 if (pgd_none(*pgd
)) {
1451 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1454 get_page_bootmem(section_nr
, pgd_page(*pgd
), MIX_SECTION_INFO
);
1456 p4d
= p4d_offset(pgd
, addr
);
1457 if (p4d_none(*p4d
)) {
1458 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1461 get_page_bootmem(section_nr
, p4d_page(*p4d
), MIX_SECTION_INFO
);
1463 pud
= pud_offset(p4d
, addr
);
1464 if (pud_none(*pud
)) {
1465 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1468 get_page_bootmem(section_nr
, pud_page(*pud
), MIX_SECTION_INFO
);
1470 if (!boot_cpu_has(X86_FEATURE_PSE
)) {
1471 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1472 pmd
= pmd_offset(pud
, addr
);
1475 get_page_bootmem(section_nr
, pmd_page(*pmd
),
1478 pte
= pte_offset_kernel(pmd
, addr
);
1481 get_page_bootmem(section_nr
, pte_page(*pte
),
1484 next
= pmd_addr_end(addr
, end
);
1486 pmd
= pmd_offset(pud
, addr
);
1490 nr_pmd_pages
= 1 << get_order(PMD_SIZE
);
1491 page
= pmd_page(*pmd
);
1492 while (nr_pmd_pages
--)
1493 get_page_bootmem(section_nr
, page
++,
1500 void __meminit
vmemmap_populate_print_last(void)
1503 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1504 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);