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 void sync_global_pgds(unsigned long start
, unsigned long end
)
99 for (addr
= start
; addr
<= end
; addr
= ALIGN(addr
+ 1, PGDIR_SIZE
)) {
100 pgd_t
*pgd_ref
= pgd_offset_k(addr
);
101 const p4d_t
*p4d_ref
;
105 * With folded p4d, pgd_none() is always false, we need to
106 * handle synchonization on p4d level.
108 BUILD_BUG_ON(pgd_none(*pgd_ref
));
109 p4d_ref
= p4d_offset(pgd_ref
, addr
);
111 if (p4d_none(*p4d_ref
))
114 spin_lock(&pgd_lock
);
115 list_for_each_entry(page
, &pgd_list
, lru
) {
118 spinlock_t
*pgt_lock
;
120 pgd
= (pgd_t
*)page_address(page
) + pgd_index(addr
);
121 p4d
= p4d_offset(pgd
, addr
);
122 /* the pgt_lock only for Xen */
123 pgt_lock
= &pgd_page_get_mm(page
)->page_table_lock
;
126 if (!p4d_none(*p4d_ref
) && !p4d_none(*p4d
))
127 BUG_ON(p4d_page_vaddr(*p4d
)
128 != p4d_page_vaddr(*p4d_ref
));
131 set_p4d(p4d
, *p4d_ref
);
133 spin_unlock(pgt_lock
);
135 spin_unlock(&pgd_lock
);
140 * NOTE: This function is marked __ref because it calls __init function
141 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
143 static __ref
void *spp_getpage(void)
148 ptr
= (void *) get_zeroed_page(GFP_ATOMIC
| __GFP_NOTRACK
);
150 ptr
= alloc_bootmem_pages(PAGE_SIZE
);
152 if (!ptr
|| ((unsigned long)ptr
& ~PAGE_MASK
)) {
153 panic("set_pte_phys: cannot allocate page data %s\n",
154 after_bootmem
? "after bootmem" : "");
157 pr_debug("spp_getpage %p\n", ptr
);
162 static p4d_t
*fill_p4d(pgd_t
*pgd
, unsigned long vaddr
)
164 if (pgd_none(*pgd
)) {
165 p4d_t
*p4d
= (p4d_t
*)spp_getpage();
166 pgd_populate(&init_mm
, pgd
, p4d
);
167 if (p4d
!= p4d_offset(pgd
, 0))
168 printk(KERN_ERR
"PAGETABLE BUG #00! %p <-> %p\n",
169 p4d
, p4d_offset(pgd
, 0));
171 return p4d_offset(pgd
, vaddr
);
174 static pud_t
*fill_pud(p4d_t
*p4d
, unsigned long vaddr
)
176 if (p4d_none(*p4d
)) {
177 pud_t
*pud
= (pud_t
*)spp_getpage();
178 p4d_populate(&init_mm
, p4d
, pud
);
179 if (pud
!= pud_offset(p4d
, 0))
180 printk(KERN_ERR
"PAGETABLE BUG #01! %p <-> %p\n",
181 pud
, pud_offset(p4d
, 0));
183 return pud_offset(p4d
, vaddr
);
186 static pmd_t
*fill_pmd(pud_t
*pud
, unsigned long vaddr
)
188 if (pud_none(*pud
)) {
189 pmd_t
*pmd
= (pmd_t
*) spp_getpage();
190 pud_populate(&init_mm
, pud
, pmd
);
191 if (pmd
!= pmd_offset(pud
, 0))
192 printk(KERN_ERR
"PAGETABLE BUG #02! %p <-> %p\n",
193 pmd
, pmd_offset(pud
, 0));
195 return pmd_offset(pud
, vaddr
);
198 static pte_t
*fill_pte(pmd_t
*pmd
, unsigned long vaddr
)
200 if (pmd_none(*pmd
)) {
201 pte_t
*pte
= (pte_t
*) spp_getpage();
202 pmd_populate_kernel(&init_mm
, pmd
, pte
);
203 if (pte
!= pte_offset_kernel(pmd
, 0))
204 printk(KERN_ERR
"PAGETABLE BUG #03!\n");
206 return pte_offset_kernel(pmd
, vaddr
);
209 static void __set_pte_vaddr(pud_t
*pud
, unsigned long vaddr
, pte_t new_pte
)
211 pmd_t
*pmd
= fill_pmd(pud
, vaddr
);
212 pte_t
*pte
= fill_pte(pmd
, vaddr
);
214 set_pte(pte
, new_pte
);
217 * It's enough to flush this one mapping.
218 * (PGE mappings get flushed as well)
220 __flush_tlb_one(vaddr
);
223 void set_pte_vaddr_p4d(p4d_t
*p4d_page
, unsigned long vaddr
, pte_t new_pte
)
225 p4d_t
*p4d
= p4d_page
+ p4d_index(vaddr
);
226 pud_t
*pud
= fill_pud(p4d
, vaddr
);
228 __set_pte_vaddr(pud
, vaddr
, new_pte
);
231 void set_pte_vaddr_pud(pud_t
*pud_page
, unsigned long vaddr
, pte_t new_pte
)
233 pud_t
*pud
= pud_page
+ pud_index(vaddr
);
235 __set_pte_vaddr(pud
, vaddr
, new_pte
);
238 void set_pte_vaddr(unsigned long vaddr
, pte_t pteval
)
243 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr
, native_pte_val(pteval
));
245 pgd
= pgd_offset_k(vaddr
);
246 if (pgd_none(*pgd
)) {
248 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
252 p4d_page
= p4d_offset(pgd
, 0);
253 set_pte_vaddr_p4d(p4d_page
, vaddr
, pteval
);
256 pmd_t
* __init
populate_extra_pmd(unsigned long vaddr
)
262 pgd
= pgd_offset_k(vaddr
);
263 p4d
= fill_p4d(pgd
, vaddr
);
264 pud
= fill_pud(p4d
, vaddr
);
265 return fill_pmd(pud
, vaddr
);
268 pte_t
* __init
populate_extra_pte(unsigned long vaddr
)
272 pmd
= populate_extra_pmd(vaddr
);
273 return fill_pte(pmd
, vaddr
);
277 * Create large page table mappings for a range of physical addresses.
279 static void __init
__init_extra_mapping(unsigned long phys
, unsigned long size
,
280 enum page_cache_mode cache
)
288 pgprot_val(prot
) = pgprot_val(PAGE_KERNEL_LARGE
) |
289 pgprot_val(pgprot_4k_2_large(cachemode2pgprot(cache
)));
290 BUG_ON((phys
& ~PMD_MASK
) || (size
& ~PMD_MASK
));
291 for (; size
; phys
+= PMD_SIZE
, size
-= PMD_SIZE
) {
292 pgd
= pgd_offset_k((unsigned long)__va(phys
));
293 if (pgd_none(*pgd
)) {
294 p4d
= (p4d_t
*) spp_getpage();
295 set_pgd(pgd
, __pgd(__pa(p4d
) | _KERNPG_TABLE
|
298 p4d
= p4d_offset(pgd
, (unsigned long)__va(phys
));
299 if (p4d_none(*p4d
)) {
300 pud
= (pud_t
*) spp_getpage();
301 set_p4d(p4d
, __p4d(__pa(pud
) | _KERNPG_TABLE
|
304 pud
= pud_offset(p4d
, (unsigned long)__va(phys
));
305 if (pud_none(*pud
)) {
306 pmd
= (pmd_t
*) spp_getpage();
307 set_pud(pud
, __pud(__pa(pmd
) | _KERNPG_TABLE
|
310 pmd
= pmd_offset(pud
, phys
);
311 BUG_ON(!pmd_none(*pmd
));
312 set_pmd(pmd
, __pmd(phys
| pgprot_val(prot
)));
316 void __init
init_extra_mapping_wb(unsigned long phys
, unsigned long size
)
318 __init_extra_mapping(phys
, size
, _PAGE_CACHE_MODE_WB
);
321 void __init
init_extra_mapping_uc(unsigned long phys
, unsigned long size
)
323 __init_extra_mapping(phys
, size
, _PAGE_CACHE_MODE_UC
);
327 * The head.S code sets up the kernel high mapping:
329 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
331 * phys_base holds the negative offset to the kernel, which is added
332 * to the compile time generated pmds. This results in invalid pmds up
333 * to the point where we hit the physaddr 0 mapping.
335 * We limit the mappings to the region from _text to _brk_end. _brk_end
336 * is rounded up to the 2MB boundary. This catches the invalid pmds as
337 * well, as they are located before _text:
339 void __init
cleanup_highmap(void)
341 unsigned long vaddr
= __START_KERNEL_map
;
342 unsigned long vaddr_end
= __START_KERNEL_map
+ KERNEL_IMAGE_SIZE
;
343 unsigned long end
= roundup((unsigned long)_brk_end
, PMD_SIZE
) - 1;
344 pmd_t
*pmd
= level2_kernel_pgt
;
347 * Native path, max_pfn_mapped is not set yet.
348 * Xen has valid max_pfn_mapped set in
349 * arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
352 vaddr_end
= __START_KERNEL_map
+ (max_pfn_mapped
<< PAGE_SHIFT
);
354 for (; vaddr
+ PMD_SIZE
- 1 < vaddr_end
; pmd
++, vaddr
+= PMD_SIZE
) {
357 if (vaddr
< (unsigned long) _text
|| vaddr
> end
)
358 set_pmd(pmd
, __pmd(0));
363 * Create PTE level page table mapping for physical addresses.
364 * It returns the last physical address mapped.
366 static unsigned long __meminit
367 phys_pte_init(pte_t
*pte_page
, unsigned long paddr
, unsigned long paddr_end
,
370 unsigned long pages
= 0, paddr_next
;
371 unsigned long paddr_last
= paddr_end
;
375 pte
= pte_page
+ pte_index(paddr
);
376 i
= pte_index(paddr
);
378 for (; i
< PTRS_PER_PTE
; i
++, paddr
= paddr_next
, pte
++) {
379 paddr_next
= (paddr
& PAGE_MASK
) + PAGE_SIZE
;
380 if (paddr
>= paddr_end
) {
381 if (!after_bootmem
&&
382 !e820__mapped_any(paddr
& PAGE_MASK
, paddr_next
,
384 !e820__mapped_any(paddr
& PAGE_MASK
, paddr_next
,
385 E820_TYPE_RESERVED_KERN
))
386 set_pte(pte
, __pte(0));
391 * We will re-use the existing mapping.
392 * Xen for example has some special requirements, like mapping
393 * pagetable pages as RO. So assume someone who pre-setup
394 * these mappings are more intelligent.
396 if (!pte_none(*pte
)) {
403 pr_info(" pte=%p addr=%lx pte=%016lx\n", pte
, paddr
,
404 pfn_pte(paddr
>> PAGE_SHIFT
, PAGE_KERNEL
).pte
);
406 set_pte(pte
, pfn_pte(paddr
>> PAGE_SHIFT
, prot
));
407 paddr_last
= (paddr
& PAGE_MASK
) + PAGE_SIZE
;
410 update_page_count(PG_LEVEL_4K
, pages
);
416 * Create PMD level page table mapping for physical addresses. The virtual
417 * and physical address have to be aligned at this level.
418 * It returns the last physical address mapped.
420 static unsigned long __meminit
421 phys_pmd_init(pmd_t
*pmd_page
, unsigned long paddr
, unsigned long paddr_end
,
422 unsigned long page_size_mask
, pgprot_t prot
)
424 unsigned long pages
= 0, paddr_next
;
425 unsigned long paddr_last
= paddr_end
;
427 int i
= pmd_index(paddr
);
429 for (; i
< PTRS_PER_PMD
; i
++, paddr
= paddr_next
) {
430 pmd_t
*pmd
= pmd_page
+ pmd_index(paddr
);
432 pgprot_t new_prot
= prot
;
434 paddr_next
= (paddr
& PMD_MASK
) + PMD_SIZE
;
435 if (paddr
>= paddr_end
) {
436 if (!after_bootmem
&&
437 !e820__mapped_any(paddr
& PMD_MASK
, paddr_next
,
439 !e820__mapped_any(paddr
& PMD_MASK
, paddr_next
,
440 E820_TYPE_RESERVED_KERN
))
441 set_pmd(pmd
, __pmd(0));
445 if (!pmd_none(*pmd
)) {
446 if (!pmd_large(*pmd
)) {
447 spin_lock(&init_mm
.page_table_lock
);
448 pte
= (pte_t
*)pmd_page_vaddr(*pmd
);
449 paddr_last
= phys_pte_init(pte
, paddr
,
451 spin_unlock(&init_mm
.page_table_lock
);
455 * If we are ok with PG_LEVEL_2M mapping, then we will
456 * use the existing mapping,
458 * Otherwise, we will split the large page mapping but
459 * use the same existing protection bits except for
460 * large page, so that we don't violate Intel's TLB
461 * Application note (317080) which says, while changing
462 * the page sizes, new and old translations should
463 * not differ with respect to page frame and
466 if (page_size_mask
& (1 << PG_LEVEL_2M
)) {
469 paddr_last
= paddr_next
;
472 new_prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pmd
));
475 if (page_size_mask
& (1<<PG_LEVEL_2M
)) {
477 spin_lock(&init_mm
.page_table_lock
);
478 set_pte((pte_t
*)pmd
,
479 pfn_pte((paddr
& PMD_MASK
) >> PAGE_SHIFT
,
480 __pgprot(pgprot_val(prot
) | _PAGE_PSE
)));
481 spin_unlock(&init_mm
.page_table_lock
);
482 paddr_last
= paddr_next
;
486 pte
= alloc_low_page();
487 paddr_last
= phys_pte_init(pte
, paddr
, paddr_end
, new_prot
);
489 spin_lock(&init_mm
.page_table_lock
);
490 pmd_populate_kernel(&init_mm
, pmd
, pte
);
491 spin_unlock(&init_mm
.page_table_lock
);
493 update_page_count(PG_LEVEL_2M
, pages
);
498 * Create PUD level page table mapping for physical addresses. The virtual
499 * and physical address do not have to be aligned at this level. KASLR can
500 * randomize virtual addresses up to this level.
501 * It returns the last physical address mapped.
503 static unsigned long __meminit
504 phys_pud_init(pud_t
*pud_page
, unsigned long paddr
, unsigned long paddr_end
,
505 unsigned long page_size_mask
)
507 unsigned long pages
= 0, paddr_next
;
508 unsigned long paddr_last
= paddr_end
;
509 unsigned long vaddr
= (unsigned long)__va(paddr
);
510 int i
= pud_index(vaddr
);
512 for (; i
< PTRS_PER_PUD
; i
++, paddr
= paddr_next
) {
515 pgprot_t prot
= PAGE_KERNEL
;
517 vaddr
= (unsigned long)__va(paddr
);
518 pud
= pud_page
+ pud_index(vaddr
);
519 paddr_next
= (paddr
& PUD_MASK
) + PUD_SIZE
;
521 if (paddr
>= paddr_end
) {
522 if (!after_bootmem
&&
523 !e820__mapped_any(paddr
& PUD_MASK
, paddr_next
,
525 !e820__mapped_any(paddr
& PUD_MASK
, paddr_next
,
526 E820_TYPE_RESERVED_KERN
))
527 set_pud(pud
, __pud(0));
531 if (!pud_none(*pud
)) {
532 if (!pud_large(*pud
)) {
533 pmd
= pmd_offset(pud
, 0);
534 paddr_last
= phys_pmd_init(pmd
, paddr
,
542 * If we are ok with PG_LEVEL_1G mapping, then we will
543 * use the existing mapping.
545 * Otherwise, we will split the gbpage mapping but use
546 * the same existing protection bits except for large
547 * page, so that we don't violate Intel's TLB
548 * Application note (317080) which says, while changing
549 * the page sizes, new and old translations should
550 * not differ with respect to page frame and
553 if (page_size_mask
& (1 << PG_LEVEL_1G
)) {
556 paddr_last
= paddr_next
;
559 prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pud
));
562 if (page_size_mask
& (1<<PG_LEVEL_1G
)) {
564 spin_lock(&init_mm
.page_table_lock
);
565 set_pte((pte_t
*)pud
,
566 pfn_pte((paddr
& PUD_MASK
) >> PAGE_SHIFT
,
568 spin_unlock(&init_mm
.page_table_lock
);
569 paddr_last
= paddr_next
;
573 pmd
= alloc_low_page();
574 paddr_last
= phys_pmd_init(pmd
, paddr
, paddr_end
,
575 page_size_mask
, prot
);
577 spin_lock(&init_mm
.page_table_lock
);
578 pud_populate(&init_mm
, pud
, pmd
);
579 spin_unlock(&init_mm
.page_table_lock
);
583 update_page_count(PG_LEVEL_1G
, pages
);
589 * Create page table mapping for the physical memory for specific physical
590 * addresses. The virtual and physical addresses have to be aligned on PMD level
591 * down. It returns the last physical address mapped.
593 unsigned long __meminit
594 kernel_physical_mapping_init(unsigned long paddr_start
,
595 unsigned long paddr_end
,
596 unsigned long page_size_mask
)
598 bool pgd_changed
= false;
599 unsigned long vaddr
, vaddr_start
, vaddr_end
, vaddr_next
, paddr_last
;
601 paddr_last
= paddr_end
;
602 vaddr
= (unsigned long)__va(paddr_start
);
603 vaddr_end
= (unsigned long)__va(paddr_end
);
606 for (; vaddr
< vaddr_end
; vaddr
= vaddr_next
) {
607 pgd_t
*pgd
= pgd_offset_k(vaddr
);
611 vaddr_next
= (vaddr
& PGDIR_MASK
) + PGDIR_SIZE
;
613 BUILD_BUG_ON(pgd_none(*pgd
));
614 p4d
= p4d_offset(pgd
, vaddr
);
616 pud
= (pud_t
*)p4d_page_vaddr(*p4d
);
617 paddr_last
= phys_pud_init(pud
, __pa(vaddr
),
623 pud
= alloc_low_page();
624 paddr_last
= phys_pud_init(pud
, __pa(vaddr
), __pa(vaddr_end
),
627 spin_lock(&init_mm
.page_table_lock
);
628 p4d_populate(&init_mm
, p4d
, pud
);
629 spin_unlock(&init_mm
.page_table_lock
);
634 sync_global_pgds(vaddr_start
, vaddr_end
- 1);
642 void __init
initmem_init(void)
644 memblock_set_node(0, (phys_addr_t
)ULLONG_MAX
, &memblock
.memory
, 0);
648 void __init
paging_init(void)
650 sparse_memory_present_with_active_regions(MAX_NUMNODES
);
654 * clear the default setting with node 0
655 * note: don't use nodes_clear here, that is really clearing when
656 * numa support is not compiled in, and later node_set_state
657 * will not set it back.
659 node_clear_state(0, N_MEMORY
);
660 if (N_MEMORY
!= N_NORMAL_MEMORY
)
661 node_clear_state(0, N_NORMAL_MEMORY
);
667 * Memory hotplug specific functions
669 #ifdef CONFIG_MEMORY_HOTPLUG
671 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
674 static void update_end_of_memory_vars(u64 start
, u64 size
)
676 unsigned long end_pfn
= PFN_UP(start
+ size
);
678 if (end_pfn
> max_pfn
) {
680 max_low_pfn
= end_pfn
;
681 high_memory
= (void *)__va(max_pfn
* PAGE_SIZE
- 1) + 1;
686 * Memory is added always to NORMAL zone. This means you will never get
687 * additional DMA/DMA32 memory.
689 int arch_add_memory(int nid
, u64 start
, u64 size
, bool for_device
)
691 struct pglist_data
*pgdat
= NODE_DATA(nid
);
692 struct zone
*zone
= pgdat
->node_zones
+
693 zone_for_memory(nid
, start
, size
, ZONE_NORMAL
, for_device
);
694 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
695 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
698 init_memory_mapping(start
, start
+ size
);
700 ret
= __add_pages(nid
, zone
, start_pfn
, nr_pages
);
703 /* update max_pfn, max_low_pfn and high_memory */
704 update_end_of_memory_vars(start
, size
);
708 EXPORT_SYMBOL_GPL(arch_add_memory
);
710 #define PAGE_INUSE 0xFD
712 static void __meminit
free_pagetable(struct page
*page
, int order
)
715 unsigned int nr_pages
= 1 << order
;
716 struct vmem_altmap
*altmap
= to_vmem_altmap((unsigned long) page
);
719 vmem_altmap_free(altmap
, nr_pages
);
723 /* bootmem page has reserved flag */
724 if (PageReserved(page
)) {
725 __ClearPageReserved(page
);
727 magic
= (unsigned long)page
->freelist
;
728 if (magic
== SECTION_INFO
|| magic
== MIX_SECTION_INFO
) {
730 put_page_bootmem(page
++);
733 free_reserved_page(page
++);
735 free_pages((unsigned long)page_address(page
), order
);
738 static void __meminit
free_pte_table(pte_t
*pte_start
, pmd_t
*pmd
)
743 for (i
= 0; i
< PTRS_PER_PTE
; i
++) {
749 /* free a pte talbe */
750 free_pagetable(pmd_page(*pmd
), 0);
751 spin_lock(&init_mm
.page_table_lock
);
753 spin_unlock(&init_mm
.page_table_lock
);
756 static void __meminit
free_pmd_table(pmd_t
*pmd_start
, pud_t
*pud
)
761 for (i
= 0; i
< PTRS_PER_PMD
; i
++) {
767 /* free a pmd talbe */
768 free_pagetable(pud_page(*pud
), 0);
769 spin_lock(&init_mm
.page_table_lock
);
771 spin_unlock(&init_mm
.page_table_lock
);
774 static void __meminit
free_pud_table(pud_t
*pud_start
, p4d_t
*p4d
)
779 for (i
= 0; i
< PTRS_PER_PUD
; i
++) {
785 /* free a pud talbe */
786 free_pagetable(p4d_page(*p4d
), 0);
787 spin_lock(&init_mm
.page_table_lock
);
789 spin_unlock(&init_mm
.page_table_lock
);
792 static void __meminit
793 remove_pte_table(pte_t
*pte_start
, unsigned long addr
, unsigned long end
,
796 unsigned long next
, pages
= 0;
799 phys_addr_t phys_addr
;
801 pte
= pte_start
+ pte_index(addr
);
802 for (; addr
< end
; addr
= next
, pte
++) {
803 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
807 if (!pte_present(*pte
))
811 * We mapped [0,1G) memory as identity mapping when
812 * initializing, in arch/x86/kernel/head_64.S. These
813 * pagetables cannot be removed.
815 phys_addr
= pte_val(*pte
) + (addr
& PAGE_MASK
);
816 if (phys_addr
< (phys_addr_t
)0x40000000)
819 if (PAGE_ALIGNED(addr
) && PAGE_ALIGNED(next
)) {
821 * Do not free direct mapping pages since they were
822 * freed when offlining, or simplely not in use.
825 free_pagetable(pte_page(*pte
), 0);
827 spin_lock(&init_mm
.page_table_lock
);
828 pte_clear(&init_mm
, addr
, pte
);
829 spin_unlock(&init_mm
.page_table_lock
);
831 /* For non-direct mapping, pages means nothing. */
835 * If we are here, we are freeing vmemmap pages since
836 * direct mapped memory ranges to be freed are aligned.
838 * If we are not removing the whole page, it means
839 * other page structs in this page are being used and
840 * we canot remove them. So fill the unused page_structs
841 * with 0xFD, and remove the page when it is wholly
844 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
846 page_addr
= page_address(pte_page(*pte
));
847 if (!memchr_inv(page_addr
, PAGE_INUSE
, PAGE_SIZE
)) {
848 free_pagetable(pte_page(*pte
), 0);
850 spin_lock(&init_mm
.page_table_lock
);
851 pte_clear(&init_mm
, addr
, pte
);
852 spin_unlock(&init_mm
.page_table_lock
);
857 /* Call free_pte_table() in remove_pmd_table(). */
860 update_page_count(PG_LEVEL_4K
, -pages
);
863 static void __meminit
864 remove_pmd_table(pmd_t
*pmd_start
, unsigned long addr
, unsigned long end
,
867 unsigned long next
, pages
= 0;
872 pmd
= pmd_start
+ pmd_index(addr
);
873 for (; addr
< end
; addr
= next
, pmd
++) {
874 next
= pmd_addr_end(addr
, end
);
876 if (!pmd_present(*pmd
))
879 if (pmd_large(*pmd
)) {
880 if (IS_ALIGNED(addr
, PMD_SIZE
) &&
881 IS_ALIGNED(next
, PMD_SIZE
)) {
883 free_pagetable(pmd_page(*pmd
),
884 get_order(PMD_SIZE
));
886 spin_lock(&init_mm
.page_table_lock
);
888 spin_unlock(&init_mm
.page_table_lock
);
891 /* If here, we are freeing vmemmap pages. */
892 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
894 page_addr
= page_address(pmd_page(*pmd
));
895 if (!memchr_inv(page_addr
, PAGE_INUSE
,
897 free_pagetable(pmd_page(*pmd
),
898 get_order(PMD_SIZE
));
900 spin_lock(&init_mm
.page_table_lock
);
902 spin_unlock(&init_mm
.page_table_lock
);
909 pte_base
= (pte_t
*)pmd_page_vaddr(*pmd
);
910 remove_pte_table(pte_base
, addr
, next
, direct
);
911 free_pte_table(pte_base
, pmd
);
914 /* Call free_pmd_table() in remove_pud_table(). */
916 update_page_count(PG_LEVEL_2M
, -pages
);
919 static void __meminit
920 remove_pud_table(pud_t
*pud_start
, unsigned long addr
, unsigned long end
,
923 unsigned long next
, pages
= 0;
928 pud
= pud_start
+ pud_index(addr
);
929 for (; addr
< end
; addr
= next
, pud
++) {
930 next
= pud_addr_end(addr
, end
);
932 if (!pud_present(*pud
))
935 if (pud_large(*pud
)) {
936 if (IS_ALIGNED(addr
, PUD_SIZE
) &&
937 IS_ALIGNED(next
, PUD_SIZE
)) {
939 free_pagetable(pud_page(*pud
),
940 get_order(PUD_SIZE
));
942 spin_lock(&init_mm
.page_table_lock
);
944 spin_unlock(&init_mm
.page_table_lock
);
947 /* If here, we are freeing vmemmap pages. */
948 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
950 page_addr
= page_address(pud_page(*pud
));
951 if (!memchr_inv(page_addr
, PAGE_INUSE
,
953 free_pagetable(pud_page(*pud
),
954 get_order(PUD_SIZE
));
956 spin_lock(&init_mm
.page_table_lock
);
958 spin_unlock(&init_mm
.page_table_lock
);
965 pmd_base
= pmd_offset(pud
, 0);
966 remove_pmd_table(pmd_base
, addr
, next
, direct
);
967 free_pmd_table(pmd_base
, pud
);
971 update_page_count(PG_LEVEL_1G
, -pages
);
974 static void __meminit
975 remove_p4d_table(p4d_t
*p4d_start
, unsigned long addr
, unsigned long end
,
978 unsigned long next
, pages
= 0;
982 p4d
= p4d_start
+ p4d_index(addr
);
983 for (; addr
< end
; addr
= next
, p4d
++) {
984 next
= p4d_addr_end(addr
, end
);
986 if (!p4d_present(*p4d
))
989 BUILD_BUG_ON(p4d_large(*p4d
));
991 pud_base
= pud_offset(p4d
, 0);
992 remove_pud_table(pud_base
, addr
, next
, direct
);
994 * For 4-level page tables we do not want to free PUDs, but in the
995 * 5-level case we should free them. This code will have to change
996 * to adapt for boot-time switching between 4 and 5 level page tables.
998 if (CONFIG_PGTABLE_LEVELS
== 5)
999 free_pud_table(pud_base
, p4d
);
1003 update_page_count(PG_LEVEL_512G
, -pages
);
1006 /* start and end are both virtual address. */
1007 static void __meminit
1008 remove_pagetable(unsigned long start
, unsigned long end
, bool direct
)
1015 for (addr
= start
; addr
< end
; addr
= next
) {
1016 next
= pgd_addr_end(addr
, end
);
1018 pgd
= pgd_offset_k(addr
);
1019 if (!pgd_present(*pgd
))
1022 p4d
= p4d_offset(pgd
, 0);
1023 remove_p4d_table(p4d
, addr
, next
, direct
);
1029 void __ref
vmemmap_free(unsigned long start
, unsigned long end
)
1031 remove_pagetable(start
, end
, false);
1034 #ifdef CONFIG_MEMORY_HOTREMOVE
1035 static void __meminit
1036 kernel_physical_mapping_remove(unsigned long start
, unsigned long end
)
1038 start
= (unsigned long)__va(start
);
1039 end
= (unsigned long)__va(end
);
1041 remove_pagetable(start
, end
, true);
1044 int __ref
arch_remove_memory(u64 start
, u64 size
)
1046 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1047 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
1048 struct page
*page
= pfn_to_page(start_pfn
);
1049 struct vmem_altmap
*altmap
;
1053 /* With altmap the first mapped page is offset from @start */
1054 altmap
= to_vmem_altmap((unsigned long) page
);
1056 page
+= vmem_altmap_offset(altmap
);
1057 zone
= page_zone(page
);
1058 ret
= __remove_pages(zone
, start_pfn
, nr_pages
);
1060 kernel_physical_mapping_remove(start
, start
+ size
);
1065 #endif /* CONFIG_MEMORY_HOTPLUG */
1067 static struct kcore_list kcore_vsyscall
;
1069 static void __init
register_page_bootmem_info(void)
1074 for_each_online_node(i
)
1075 register_page_bootmem_info_node(NODE_DATA(i
));
1079 void __init
mem_init(void)
1083 /* clear_bss() already clear the empty_zero_page */
1085 register_page_bootmem_info();
1087 /* this will put all memory onto the freelists */
1091 /* Register memory areas for /proc/kcore */
1092 kclist_add(&kcore_vsyscall
, (void *)VSYSCALL_ADDR
,
1093 PAGE_SIZE
, KCORE_OTHER
);
1095 mem_init_print_info(NULL
);
1098 int kernel_set_to_readonly
;
1100 void set_kernel_text_rw(void)
1102 unsigned long start
= PFN_ALIGN(_text
);
1103 unsigned long end
= PFN_ALIGN(__stop___ex_table
);
1105 if (!kernel_set_to_readonly
)
1108 pr_debug("Set kernel text: %lx - %lx for read write\n",
1112 * Make the kernel identity mapping for text RW. Kernel text
1113 * mapping will always be RO. Refer to the comment in
1114 * static_protections() in pageattr.c
1116 set_memory_rw(start
, (end
- start
) >> PAGE_SHIFT
);
1119 void set_kernel_text_ro(void)
1121 unsigned long start
= PFN_ALIGN(_text
);
1122 unsigned long end
= PFN_ALIGN(__stop___ex_table
);
1124 if (!kernel_set_to_readonly
)
1127 pr_debug("Set kernel text: %lx - %lx for read only\n",
1131 * Set the kernel identity mapping for text RO.
1133 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
1136 void mark_rodata_ro(void)
1138 unsigned long start
= PFN_ALIGN(_text
);
1139 unsigned long rodata_start
= PFN_ALIGN(__start_rodata
);
1140 unsigned long end
= (unsigned long) &__end_rodata_hpage_align
;
1141 unsigned long text_end
= PFN_ALIGN(&__stop___ex_table
);
1142 unsigned long rodata_end
= PFN_ALIGN(&__end_rodata
);
1143 unsigned long all_end
;
1145 printk(KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
1146 (end
- start
) >> 10);
1147 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
1149 kernel_set_to_readonly
= 1;
1152 * The rodata/data/bss/brk section (but not the kernel text!)
1153 * should also be not-executable.
1155 * We align all_end to PMD_SIZE because the existing mapping
1156 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1157 * split the PMD and the reminder between _brk_end and the end
1158 * of the PMD will remain mapped executable.
1160 * Any PMD which was setup after the one which covers _brk_end
1161 * has been zapped already via cleanup_highmem().
1163 all_end
= roundup((unsigned long)_brk_end
, PMD_SIZE
);
1164 set_memory_nx(text_end
, (all_end
- text_end
) >> PAGE_SHIFT
);
1166 #ifdef CONFIG_CPA_DEBUG
1167 printk(KERN_INFO
"Testing CPA: undo %lx-%lx\n", start
, end
);
1168 set_memory_rw(start
, (end
-start
) >> PAGE_SHIFT
);
1170 printk(KERN_INFO
"Testing CPA: again\n");
1171 set_memory_ro(start
, (end
-start
) >> PAGE_SHIFT
);
1174 free_init_pages("unused kernel",
1175 (unsigned long) __va(__pa_symbol(text_end
)),
1176 (unsigned long) __va(__pa_symbol(rodata_start
)));
1177 free_init_pages("unused kernel",
1178 (unsigned long) __va(__pa_symbol(rodata_end
)),
1179 (unsigned long) __va(__pa_symbol(_sdata
)));
1184 int kern_addr_valid(unsigned long addr
)
1186 unsigned long above
= ((long)addr
) >> __VIRTUAL_MASK_SHIFT
;
1193 if (above
!= 0 && above
!= -1UL)
1196 pgd
= pgd_offset_k(addr
);
1200 p4d
= p4d_offset(pgd
, addr
);
1204 pud
= pud_offset(p4d
, addr
);
1208 if (pud_large(*pud
))
1209 return pfn_valid(pud_pfn(*pud
));
1211 pmd
= pmd_offset(pud
, addr
);
1215 if (pmd_large(*pmd
))
1216 return pfn_valid(pmd_pfn(*pmd
));
1218 pte
= pte_offset_kernel(pmd
, addr
);
1222 return pfn_valid(pte_pfn(*pte
));
1225 static unsigned long probe_memory_block_size(void)
1227 unsigned long bz
= MIN_MEMORY_BLOCK_SIZE
;
1229 /* if system is UV or has 64GB of RAM or more, use large blocks */
1230 if (is_uv_system() || ((max_pfn
<< PAGE_SHIFT
) >= (64UL << 30)))
1231 bz
= 2UL << 30; /* 2GB */
1233 pr_info("x86/mm: Memory block size: %ldMB\n", bz
>> 20);
1238 static unsigned long memory_block_size_probed
;
1239 unsigned long memory_block_size_bytes(void)
1241 if (!memory_block_size_probed
)
1242 memory_block_size_probed
= probe_memory_block_size();
1244 return memory_block_size_probed
;
1247 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1249 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1251 static long __meminitdata addr_start
, addr_end
;
1252 static void __meminitdata
*p_start
, *p_end
;
1253 static int __meminitdata node_start
;
1255 static int __meminit
vmemmap_populate_hugepages(unsigned long start
,
1256 unsigned long end
, int node
, struct vmem_altmap
*altmap
)
1265 for (addr
= start
; addr
< end
; addr
= next
) {
1266 next
= pmd_addr_end(addr
, end
);
1268 pgd
= vmemmap_pgd_populate(addr
, node
);
1272 p4d
= vmemmap_p4d_populate(pgd
, addr
, node
);
1276 pud
= vmemmap_pud_populate(p4d
, addr
, node
);
1280 pmd
= pmd_offset(pud
, addr
);
1281 if (pmd_none(*pmd
)) {
1284 p
= __vmemmap_alloc_block_buf(PMD_SIZE
, node
, altmap
);
1288 entry
= pfn_pte(__pa(p
) >> PAGE_SHIFT
,
1290 set_pmd(pmd
, __pmd(pte_val(entry
)));
1292 /* check to see if we have contiguous blocks */
1293 if (p_end
!= p
|| node_start
!= node
) {
1295 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1296 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);
1302 addr_end
= addr
+ PMD_SIZE
;
1303 p_end
= p
+ PMD_SIZE
;
1306 return -ENOMEM
; /* no fallback */
1307 } else if (pmd_large(*pmd
)) {
1308 vmemmap_verify((pte_t
*)pmd
, node
, addr
, next
);
1311 pr_warn_once("vmemmap: falling back to regular page backing\n");
1312 if (vmemmap_populate_basepages(addr
, next
, node
))
1318 int __meminit
vmemmap_populate(unsigned long start
, unsigned long end
, int node
)
1320 struct vmem_altmap
*altmap
= to_vmem_altmap(start
);
1323 if (boot_cpu_has(X86_FEATURE_PSE
))
1324 err
= vmemmap_populate_hugepages(start
, end
, node
, altmap
);
1326 pr_err_once("%s: no cpu support for altmap allocations\n",
1330 err
= vmemmap_populate_basepages(start
, end
, node
);
1332 sync_global_pgds(start
, end
- 1);
1336 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1337 void register_page_bootmem_memmap(unsigned long section_nr
,
1338 struct page
*start_page
, unsigned long size
)
1340 unsigned long addr
= (unsigned long)start_page
;
1341 unsigned long end
= (unsigned long)(start_page
+ size
);
1347 unsigned int nr_pages
;
1350 for (; addr
< end
; addr
= next
) {
1353 pgd
= pgd_offset_k(addr
);
1354 if (pgd_none(*pgd
)) {
1355 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1358 get_page_bootmem(section_nr
, pgd_page(*pgd
), MIX_SECTION_INFO
);
1360 p4d
= p4d_offset(pgd
, addr
);
1361 if (p4d_none(*p4d
)) {
1362 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1365 get_page_bootmem(section_nr
, p4d_page(*p4d
), MIX_SECTION_INFO
);
1367 pud
= pud_offset(p4d
, addr
);
1368 if (pud_none(*pud
)) {
1369 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1372 get_page_bootmem(section_nr
, pud_page(*pud
), MIX_SECTION_INFO
);
1374 if (!boot_cpu_has(X86_FEATURE_PSE
)) {
1375 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1376 pmd
= pmd_offset(pud
, addr
);
1379 get_page_bootmem(section_nr
, pmd_page(*pmd
),
1382 pte
= pte_offset_kernel(pmd
, addr
);
1385 get_page_bootmem(section_nr
, pte_page(*pte
),
1388 next
= pmd_addr_end(addr
, end
);
1390 pmd
= pmd_offset(pud
, addr
);
1394 nr_pages
= 1 << (get_order(PMD_SIZE
));
1395 page
= pmd_page(*pmd
);
1397 get_page_bootmem(section_nr
, page
++,
1404 void __meminit
vmemmap_populate_print_last(void)
1407 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1408 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);