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 <asm/uaccess.h>
40 #include <asm/pgtable.h>
41 #include <asm/pgalloc.h>
43 #include <asm/fixmap.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/cacheflush.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/removed 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
, int removed
)
97 unsigned long address
;
99 for (address
= start
; address
<= end
; address
+= PGDIR_SIZE
) {
100 const pgd_t
*pgd_ref
= pgd_offset_k(address
);
104 * When it is called after memory hot remove, pgd_none()
105 * returns true. In this case (removed == 1), we must clear
106 * the PGD entries in the local PGD level page.
108 if (pgd_none(*pgd_ref
) && !removed
)
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(address
);
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
)
123 != pgd_page_vaddr(*pgd_ref
));
126 if (pgd_none(*pgd_ref
) && !pgd_none(*pgd
))
130 set_pgd(pgd
, *pgd_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 pud_t
*fill_pud(pgd_t
*pgd
, unsigned long vaddr
)
164 if (pgd_none(*pgd
)) {
165 pud_t
*pud
= (pud_t
*)spp_getpage();
166 pgd_populate(&init_mm
, pgd
, pud
);
167 if (pud
!= pud_offset(pgd
, 0))
168 printk(KERN_ERR
"PAGETABLE BUG #00! %p <-> %p\n",
169 pud
, pud_offset(pgd
, 0));
171 return pud_offset(pgd
, vaddr
);
174 static pmd_t
*fill_pmd(pud_t
*pud
, unsigned long vaddr
)
176 if (pud_none(*pud
)) {
177 pmd_t
*pmd
= (pmd_t
*) spp_getpage();
178 pud_populate(&init_mm
, pud
, pmd
);
179 if (pmd
!= pmd_offset(pud
, 0))
180 printk(KERN_ERR
"PAGETABLE BUG #01! %p <-> %p\n",
181 pmd
, pmd_offset(pud
, 0));
183 return pmd_offset(pud
, vaddr
);
186 static pte_t
*fill_pte(pmd_t
*pmd
, unsigned long vaddr
)
188 if (pmd_none(*pmd
)) {
189 pte_t
*pte
= (pte_t
*) spp_getpage();
190 pmd_populate_kernel(&init_mm
, pmd
, pte
);
191 if (pte
!= pte_offset_kernel(pmd
, 0))
192 printk(KERN_ERR
"PAGETABLE BUG #02!\n");
194 return pte_offset_kernel(pmd
, vaddr
);
197 void set_pte_vaddr_pud(pud_t
*pud_page
, unsigned long vaddr
, pte_t new_pte
)
203 pud
= pud_page
+ pud_index(vaddr
);
204 pmd
= fill_pmd(pud
, vaddr
);
205 pte
= fill_pte(pmd
, vaddr
);
207 set_pte(pte
, new_pte
);
210 * It's enough to flush this one mapping.
211 * (PGE mappings get flushed as well)
213 __flush_tlb_one(vaddr
);
216 void set_pte_vaddr(unsigned long vaddr
, pte_t pteval
)
221 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr
, native_pte_val(pteval
));
223 pgd
= pgd_offset_k(vaddr
);
224 if (pgd_none(*pgd
)) {
226 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
229 pud_page
= (pud_t
*)pgd_page_vaddr(*pgd
);
230 set_pte_vaddr_pud(pud_page
, vaddr
, pteval
);
233 pmd_t
* __init
populate_extra_pmd(unsigned long vaddr
)
238 pgd
= pgd_offset_k(vaddr
);
239 pud
= fill_pud(pgd
, vaddr
);
240 return fill_pmd(pud
, vaddr
);
243 pte_t
* __init
populate_extra_pte(unsigned long vaddr
)
247 pmd
= populate_extra_pmd(vaddr
);
248 return fill_pte(pmd
, vaddr
);
252 * Create large page table mappings for a range of physical addresses.
254 static void __init
__init_extra_mapping(unsigned long phys
, unsigned long size
,
255 enum page_cache_mode cache
)
262 pgprot_val(prot
) = pgprot_val(PAGE_KERNEL_LARGE
) |
263 pgprot_val(pgprot_4k_2_large(cachemode2pgprot(cache
)));
264 BUG_ON((phys
& ~PMD_MASK
) || (size
& ~PMD_MASK
));
265 for (; size
; phys
+= PMD_SIZE
, size
-= PMD_SIZE
) {
266 pgd
= pgd_offset_k((unsigned long)__va(phys
));
267 if (pgd_none(*pgd
)) {
268 pud
= (pud_t
*) spp_getpage();
269 set_pgd(pgd
, __pgd(__pa(pud
) | _KERNPG_TABLE
|
272 pud
= pud_offset(pgd
, (unsigned long)__va(phys
));
273 if (pud_none(*pud
)) {
274 pmd
= (pmd_t
*) spp_getpage();
275 set_pud(pud
, __pud(__pa(pmd
) | _KERNPG_TABLE
|
278 pmd
= pmd_offset(pud
, phys
);
279 BUG_ON(!pmd_none(*pmd
));
280 set_pmd(pmd
, __pmd(phys
| pgprot_val(prot
)));
284 void __init
init_extra_mapping_wb(unsigned long phys
, unsigned long size
)
286 __init_extra_mapping(phys
, size
, _PAGE_CACHE_MODE_WB
);
289 void __init
init_extra_mapping_uc(unsigned long phys
, unsigned long size
)
291 __init_extra_mapping(phys
, size
, _PAGE_CACHE_MODE_UC
);
295 * The head.S code sets up the kernel high mapping:
297 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
299 * phys_base holds the negative offset to the kernel, which is added
300 * to the compile time generated pmds. This results in invalid pmds up
301 * to the point where we hit the physaddr 0 mapping.
303 * We limit the mappings to the region from _text to _brk_end. _brk_end
304 * is rounded up to the 2MB boundary. This catches the invalid pmds as
305 * well, as they are located before _text:
307 void __init
cleanup_highmap(void)
309 unsigned long vaddr
= __START_KERNEL_map
;
310 unsigned long vaddr_end
= __START_KERNEL_map
+ KERNEL_IMAGE_SIZE
;
311 unsigned long end
= roundup((unsigned long)_brk_end
, PMD_SIZE
) - 1;
312 pmd_t
*pmd
= level2_kernel_pgt
;
315 * Native path, max_pfn_mapped is not set yet.
316 * Xen has valid max_pfn_mapped set in
317 * arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
320 vaddr_end
= __START_KERNEL_map
+ (max_pfn_mapped
<< PAGE_SHIFT
);
322 for (; vaddr
+ PMD_SIZE
- 1 < vaddr_end
; pmd
++, vaddr
+= PMD_SIZE
) {
325 if (vaddr
< (unsigned long) _text
|| vaddr
> end
)
326 set_pmd(pmd
, __pmd(0));
331 * Create PTE level page table mapping for physical addresses.
332 * It returns the last physical address mapped.
334 static unsigned long __meminit
335 phys_pte_init(pte_t
*pte_page
, unsigned long paddr
, unsigned long paddr_end
,
338 unsigned long pages
= 0, paddr_next
;
339 unsigned long paddr_last
= paddr_end
;
343 pte
= pte_page
+ pte_index(paddr
);
344 i
= pte_index(paddr
);
346 for (; i
< PTRS_PER_PTE
; i
++, paddr
= paddr_next
, pte
++) {
347 paddr_next
= (paddr
& PAGE_MASK
) + PAGE_SIZE
;
348 if (paddr
>= paddr_end
) {
349 if (!after_bootmem
&&
350 !e820_any_mapped(paddr
& PAGE_MASK
, paddr_next
,
352 !e820_any_mapped(paddr
& PAGE_MASK
, paddr_next
,
354 set_pte(pte
, __pte(0));
359 * We will re-use the existing mapping.
360 * Xen for example has some special requirements, like mapping
361 * pagetable pages as RO. So assume someone who pre-setup
362 * these mappings are more intelligent.
364 if (!pte_none(*pte
)) {
371 pr_info(" pte=%p addr=%lx pte=%016lx\n", pte
, paddr
,
372 pfn_pte(paddr
>> PAGE_SHIFT
, PAGE_KERNEL
).pte
);
374 set_pte(pte
, pfn_pte(paddr
>> PAGE_SHIFT
, prot
));
375 paddr_last
= (paddr
& PAGE_MASK
) + PAGE_SIZE
;
378 update_page_count(PG_LEVEL_4K
, pages
);
384 * Create PMD level page table mapping for physical addresses. The virtual
385 * and physical address have to be aligned at this level.
386 * It returns the last physical address mapped.
388 static unsigned long __meminit
389 phys_pmd_init(pmd_t
*pmd_page
, unsigned long paddr
, unsigned long paddr_end
,
390 unsigned long page_size_mask
, pgprot_t prot
)
392 unsigned long pages
= 0, paddr_next
;
393 unsigned long paddr_last
= paddr_end
;
395 int i
= pmd_index(paddr
);
397 for (; i
< PTRS_PER_PMD
; i
++, paddr
= paddr_next
) {
398 pmd_t
*pmd
= pmd_page
+ pmd_index(paddr
);
400 pgprot_t new_prot
= prot
;
402 paddr_next
= (paddr
& PMD_MASK
) + PMD_SIZE
;
403 if (paddr
>= paddr_end
) {
404 if (!after_bootmem
&&
405 !e820_any_mapped(paddr
& PMD_MASK
, paddr_next
,
407 !e820_any_mapped(paddr
& PMD_MASK
, paddr_next
,
409 set_pmd(pmd
, __pmd(0));
413 if (!pmd_none(*pmd
)) {
414 if (!pmd_large(*pmd
)) {
415 spin_lock(&init_mm
.page_table_lock
);
416 pte
= (pte_t
*)pmd_page_vaddr(*pmd
);
417 paddr_last
= phys_pte_init(pte
, paddr
,
419 spin_unlock(&init_mm
.page_table_lock
);
423 * If we are ok with PG_LEVEL_2M mapping, then we will
424 * use the existing mapping,
426 * Otherwise, we will split the large page mapping but
427 * use the same existing protection bits except for
428 * large page, so that we don't violate Intel's TLB
429 * Application note (317080) which says, while changing
430 * the page sizes, new and old translations should
431 * not differ with respect to page frame and
434 if (page_size_mask
& (1 << PG_LEVEL_2M
)) {
437 paddr_last
= paddr_next
;
440 new_prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pmd
));
443 if (page_size_mask
& (1<<PG_LEVEL_2M
)) {
445 spin_lock(&init_mm
.page_table_lock
);
446 set_pte((pte_t
*)pmd
,
447 pfn_pte((paddr
& PMD_MASK
) >> PAGE_SHIFT
,
448 __pgprot(pgprot_val(prot
) | _PAGE_PSE
)));
449 spin_unlock(&init_mm
.page_table_lock
);
450 paddr_last
= paddr_next
;
454 pte
= alloc_low_page();
455 paddr_last
= phys_pte_init(pte
, paddr
, paddr_end
, new_prot
);
457 spin_lock(&init_mm
.page_table_lock
);
458 pmd_populate_kernel(&init_mm
, pmd
, pte
);
459 spin_unlock(&init_mm
.page_table_lock
);
461 update_page_count(PG_LEVEL_2M
, pages
);
466 * Create PUD level page table mapping for physical addresses. The virtual
467 * and physical address do not have to be aligned at this level. KASLR can
468 * randomize virtual addresses up to this level.
469 * It returns the last physical address mapped.
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
)
475 unsigned long pages
= 0, paddr_next
;
476 unsigned long paddr_last
= paddr_end
;
477 unsigned long vaddr
= (unsigned long)__va(paddr
);
478 int i
= pud_index(vaddr
);
480 for (; i
< PTRS_PER_PUD
; i
++, paddr
= paddr_next
) {
483 pgprot_t prot
= PAGE_KERNEL
;
485 vaddr
= (unsigned long)__va(paddr
);
486 pud
= pud_page
+ pud_index(vaddr
);
487 paddr_next
= (paddr
& PUD_MASK
) + PUD_SIZE
;
489 if (paddr
>= paddr_end
) {
490 if (!after_bootmem
&&
491 !e820_any_mapped(paddr
& PUD_MASK
, paddr_next
,
493 !e820_any_mapped(paddr
& PUD_MASK
, paddr_next
,
495 set_pud(pud
, __pud(0));
499 if (!pud_none(*pud
)) {
500 if (!pud_large(*pud
)) {
501 pmd
= pmd_offset(pud
, 0);
502 paddr_last
= phys_pmd_init(pmd
, paddr
,
510 * If we are ok with PG_LEVEL_1G mapping, then we will
511 * use the existing mapping.
513 * Otherwise, we will split the gbpage mapping but use
514 * the same existing protection bits except for large
515 * page, so that we don't violate Intel's TLB
516 * Application note (317080) which says, while changing
517 * the page sizes, new and old translations should
518 * not differ with respect to page frame and
521 if (page_size_mask
& (1 << PG_LEVEL_1G
)) {
524 paddr_last
= paddr_next
;
527 prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pud
));
530 if (page_size_mask
& (1<<PG_LEVEL_1G
)) {
532 spin_lock(&init_mm
.page_table_lock
);
533 set_pte((pte_t
*)pud
,
534 pfn_pte((paddr
& PUD_MASK
) >> PAGE_SHIFT
,
536 spin_unlock(&init_mm
.page_table_lock
);
537 paddr_last
= paddr_next
;
541 pmd
= alloc_low_page();
542 paddr_last
= phys_pmd_init(pmd
, paddr
, paddr_end
,
543 page_size_mask
, prot
);
545 spin_lock(&init_mm
.page_table_lock
);
546 pud_populate(&init_mm
, pud
, pmd
);
547 spin_unlock(&init_mm
.page_table_lock
);
551 update_page_count(PG_LEVEL_1G
, pages
);
557 * Create page table mapping for the physical memory for specific physical
558 * addresses. The virtual and physical addresses have to be aligned on PMD level
559 * down. It returns the last physical address mapped.
561 unsigned long __meminit
562 kernel_physical_mapping_init(unsigned long paddr_start
,
563 unsigned long paddr_end
,
564 unsigned long page_size_mask
)
566 bool pgd_changed
= false;
567 unsigned long vaddr
, vaddr_start
, vaddr_end
, vaddr_next
, paddr_last
;
569 paddr_last
= paddr_end
;
570 vaddr
= (unsigned long)__va(paddr_start
);
571 vaddr_end
= (unsigned long)__va(paddr_end
);
574 for (; vaddr
< vaddr_end
; vaddr
= vaddr_next
) {
575 pgd_t
*pgd
= pgd_offset_k(vaddr
);
578 vaddr_next
= (vaddr
& PGDIR_MASK
) + PGDIR_SIZE
;
581 pud
= (pud_t
*)pgd_page_vaddr(*pgd
);
582 paddr_last
= phys_pud_init(pud
, __pa(vaddr
),
588 pud
= alloc_low_page();
589 paddr_last
= phys_pud_init(pud
, __pa(vaddr
), __pa(vaddr_end
),
592 spin_lock(&init_mm
.page_table_lock
);
593 pgd_populate(&init_mm
, pgd
, pud
);
594 spin_unlock(&init_mm
.page_table_lock
);
599 sync_global_pgds(vaddr_start
, vaddr_end
- 1, 0);
607 void __init
initmem_init(void)
609 memblock_set_node(0, (phys_addr_t
)ULLONG_MAX
, &memblock
.memory
, 0);
613 void __init
paging_init(void)
615 sparse_memory_present_with_active_regions(MAX_NUMNODES
);
619 * clear the default setting with node 0
620 * note: don't use nodes_clear here, that is really clearing when
621 * numa support is not compiled in, and later node_set_state
622 * will not set it back.
624 node_clear_state(0, N_MEMORY
);
625 if (N_MEMORY
!= N_NORMAL_MEMORY
)
626 node_clear_state(0, N_NORMAL_MEMORY
);
632 * Memory hotplug specific functions
634 #ifdef CONFIG_MEMORY_HOTPLUG
636 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
639 static void update_end_of_memory_vars(u64 start
, u64 size
)
641 unsigned long end_pfn
= PFN_UP(start
+ size
);
643 if (end_pfn
> max_pfn
) {
645 max_low_pfn
= end_pfn
;
646 high_memory
= (void *)__va(max_pfn
* PAGE_SIZE
- 1) + 1;
651 * Memory is added always to NORMAL zone. This means you will never get
652 * additional DMA/DMA32 memory.
654 int arch_add_memory(int nid
, u64 start
, u64 size
, bool for_device
)
656 struct pglist_data
*pgdat
= NODE_DATA(nid
);
657 struct zone
*zone
= pgdat
->node_zones
+
658 zone_for_memory(nid
, start
, size
, ZONE_NORMAL
, for_device
);
659 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
660 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
663 init_memory_mapping(start
, start
+ size
);
665 ret
= __add_pages(nid
, zone
, start_pfn
, nr_pages
);
668 /* update max_pfn, max_low_pfn and high_memory */
669 update_end_of_memory_vars(start
, size
);
673 EXPORT_SYMBOL_GPL(arch_add_memory
);
675 #define PAGE_INUSE 0xFD
677 static void __meminit
free_pagetable(struct page
*page
, int order
)
680 unsigned int nr_pages
= 1 << order
;
681 struct vmem_altmap
*altmap
= to_vmem_altmap((unsigned long) page
);
684 vmem_altmap_free(altmap
, nr_pages
);
688 /* bootmem page has reserved flag */
689 if (PageReserved(page
)) {
690 __ClearPageReserved(page
);
692 magic
= (unsigned long)page
->lru
.next
;
693 if (magic
== SECTION_INFO
|| magic
== MIX_SECTION_INFO
) {
695 put_page_bootmem(page
++);
698 free_reserved_page(page
++);
700 free_pages((unsigned long)page_address(page
), order
);
703 static void __meminit
free_pte_table(pte_t
*pte_start
, pmd_t
*pmd
)
708 for (i
= 0; i
< PTRS_PER_PTE
; i
++) {
714 /* free a pte talbe */
715 free_pagetable(pmd_page(*pmd
), 0);
716 spin_lock(&init_mm
.page_table_lock
);
718 spin_unlock(&init_mm
.page_table_lock
);
721 static void __meminit
free_pmd_table(pmd_t
*pmd_start
, pud_t
*pud
)
726 for (i
= 0; i
< PTRS_PER_PMD
; i
++) {
732 /* free a pmd talbe */
733 free_pagetable(pud_page(*pud
), 0);
734 spin_lock(&init_mm
.page_table_lock
);
736 spin_unlock(&init_mm
.page_table_lock
);
739 static void __meminit
740 remove_pte_table(pte_t
*pte_start
, unsigned long addr
, unsigned long end
,
743 unsigned long next
, pages
= 0;
746 phys_addr_t phys_addr
;
748 pte
= pte_start
+ pte_index(addr
);
749 for (; addr
< end
; addr
= next
, pte
++) {
750 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
754 if (!pte_present(*pte
))
758 * We mapped [0,1G) memory as identity mapping when
759 * initializing, in arch/x86/kernel/head_64.S. These
760 * pagetables cannot be removed.
762 phys_addr
= pte_val(*pte
) + (addr
& PAGE_MASK
);
763 if (phys_addr
< (phys_addr_t
)0x40000000)
766 if (PAGE_ALIGNED(addr
) && PAGE_ALIGNED(next
)) {
768 * Do not free direct mapping pages since they were
769 * freed when offlining, or simplely not in use.
772 free_pagetable(pte_page(*pte
), 0);
774 spin_lock(&init_mm
.page_table_lock
);
775 pte_clear(&init_mm
, addr
, pte
);
776 spin_unlock(&init_mm
.page_table_lock
);
778 /* For non-direct mapping, pages means nothing. */
782 * If we are here, we are freeing vmemmap pages since
783 * direct mapped memory ranges to be freed are aligned.
785 * If we are not removing the whole page, it means
786 * other page structs in this page are being used and
787 * we canot remove them. So fill the unused page_structs
788 * with 0xFD, and remove the page when it is wholly
791 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
793 page_addr
= page_address(pte_page(*pte
));
794 if (!memchr_inv(page_addr
, PAGE_INUSE
, PAGE_SIZE
)) {
795 free_pagetable(pte_page(*pte
), 0);
797 spin_lock(&init_mm
.page_table_lock
);
798 pte_clear(&init_mm
, addr
, pte
);
799 spin_unlock(&init_mm
.page_table_lock
);
804 /* Call free_pte_table() in remove_pmd_table(). */
807 update_page_count(PG_LEVEL_4K
, -pages
);
810 static void __meminit
811 remove_pmd_table(pmd_t
*pmd_start
, unsigned long addr
, unsigned long end
,
814 unsigned long next
, pages
= 0;
819 pmd
= pmd_start
+ pmd_index(addr
);
820 for (; addr
< end
; addr
= next
, pmd
++) {
821 next
= pmd_addr_end(addr
, end
);
823 if (!pmd_present(*pmd
))
826 if (pmd_large(*pmd
)) {
827 if (IS_ALIGNED(addr
, PMD_SIZE
) &&
828 IS_ALIGNED(next
, PMD_SIZE
)) {
830 free_pagetable(pmd_page(*pmd
),
831 get_order(PMD_SIZE
));
833 spin_lock(&init_mm
.page_table_lock
);
835 spin_unlock(&init_mm
.page_table_lock
);
838 /* If here, we are freeing vmemmap pages. */
839 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
841 page_addr
= page_address(pmd_page(*pmd
));
842 if (!memchr_inv(page_addr
, PAGE_INUSE
,
844 free_pagetable(pmd_page(*pmd
),
845 get_order(PMD_SIZE
));
847 spin_lock(&init_mm
.page_table_lock
);
849 spin_unlock(&init_mm
.page_table_lock
);
856 pte_base
= (pte_t
*)pmd_page_vaddr(*pmd
);
857 remove_pte_table(pte_base
, addr
, next
, direct
);
858 free_pte_table(pte_base
, pmd
);
861 /* Call free_pmd_table() in remove_pud_table(). */
863 update_page_count(PG_LEVEL_2M
, -pages
);
866 static void __meminit
867 remove_pud_table(pud_t
*pud_start
, unsigned long addr
, unsigned long end
,
870 unsigned long next
, pages
= 0;
875 pud
= pud_start
+ pud_index(addr
);
876 for (; addr
< end
; addr
= next
, pud
++) {
877 next
= pud_addr_end(addr
, end
);
879 if (!pud_present(*pud
))
882 if (pud_large(*pud
)) {
883 if (IS_ALIGNED(addr
, PUD_SIZE
) &&
884 IS_ALIGNED(next
, PUD_SIZE
)) {
886 free_pagetable(pud_page(*pud
),
887 get_order(PUD_SIZE
));
889 spin_lock(&init_mm
.page_table_lock
);
891 spin_unlock(&init_mm
.page_table_lock
);
894 /* If here, we are freeing vmemmap pages. */
895 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
897 page_addr
= page_address(pud_page(*pud
));
898 if (!memchr_inv(page_addr
, PAGE_INUSE
,
900 free_pagetable(pud_page(*pud
),
901 get_order(PUD_SIZE
));
903 spin_lock(&init_mm
.page_table_lock
);
905 spin_unlock(&init_mm
.page_table_lock
);
912 pmd_base
= (pmd_t
*)pud_page_vaddr(*pud
);
913 remove_pmd_table(pmd_base
, addr
, next
, direct
);
914 free_pmd_table(pmd_base
, pud
);
918 update_page_count(PG_LEVEL_1G
, -pages
);
921 /* start and end are both virtual address. */
922 static void __meminit
923 remove_pagetable(unsigned long start
, unsigned long end
, bool direct
)
930 for (addr
= start
; addr
< end
; addr
= next
) {
931 next
= pgd_addr_end(addr
, end
);
933 pgd
= pgd_offset_k(addr
);
934 if (!pgd_present(*pgd
))
937 pud
= (pud_t
*)pgd_page_vaddr(*pgd
);
938 remove_pud_table(pud
, addr
, next
, direct
);
944 void __ref
vmemmap_free(unsigned long start
, unsigned long end
)
946 remove_pagetable(start
, end
, false);
949 #ifdef CONFIG_MEMORY_HOTREMOVE
950 static void __meminit
951 kernel_physical_mapping_remove(unsigned long start
, unsigned long end
)
953 start
= (unsigned long)__va(start
);
954 end
= (unsigned long)__va(end
);
956 remove_pagetable(start
, end
, true);
959 int __ref
arch_remove_memory(u64 start
, u64 size
)
961 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
962 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
963 struct page
*page
= pfn_to_page(start_pfn
);
964 struct vmem_altmap
*altmap
;
968 /* With altmap the first mapped page is offset from @start */
969 altmap
= to_vmem_altmap((unsigned long) page
);
971 page
+= vmem_altmap_offset(altmap
);
972 zone
= page_zone(page
);
973 ret
= __remove_pages(zone
, start_pfn
, nr_pages
);
975 kernel_physical_mapping_remove(start
, start
+ size
);
980 #endif /* CONFIG_MEMORY_HOTPLUG */
982 static struct kcore_list kcore_vsyscall
;
984 static void __init
register_page_bootmem_info(void)
989 for_each_online_node(i
)
990 register_page_bootmem_info_node(NODE_DATA(i
));
994 void __init
mem_init(void)
998 /* clear_bss() already clear the empty_zero_page */
1000 register_page_bootmem_info();
1002 /* this will put all memory onto the freelists */
1006 /* Register memory areas for /proc/kcore */
1007 kclist_add(&kcore_vsyscall
, (void *)VSYSCALL_ADDR
,
1008 PAGE_SIZE
, KCORE_OTHER
);
1010 mem_init_print_info(NULL
);
1013 const int rodata_test_data
= 0xC3;
1014 EXPORT_SYMBOL_GPL(rodata_test_data
);
1016 int kernel_set_to_readonly
;
1018 void set_kernel_text_rw(void)
1020 unsigned long start
= PFN_ALIGN(_text
);
1021 unsigned long end
= PFN_ALIGN(__stop___ex_table
);
1023 if (!kernel_set_to_readonly
)
1026 pr_debug("Set kernel text: %lx - %lx for read write\n",
1030 * Make the kernel identity mapping for text RW. Kernel text
1031 * mapping will always be RO. Refer to the comment in
1032 * static_protections() in pageattr.c
1034 set_memory_rw(start
, (end
- start
) >> PAGE_SHIFT
);
1037 void set_kernel_text_ro(void)
1039 unsigned long start
= PFN_ALIGN(_text
);
1040 unsigned long end
= PFN_ALIGN(__stop___ex_table
);
1042 if (!kernel_set_to_readonly
)
1045 pr_debug("Set kernel text: %lx - %lx for read only\n",
1049 * Set the kernel identity mapping for text RO.
1051 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
1054 void mark_rodata_ro(void)
1056 unsigned long start
= PFN_ALIGN(_text
);
1057 unsigned long rodata_start
= PFN_ALIGN(__start_rodata
);
1058 unsigned long end
= (unsigned long) &__end_rodata_hpage_align
;
1059 unsigned long text_end
= PFN_ALIGN(&__stop___ex_table
);
1060 unsigned long rodata_end
= PFN_ALIGN(&__end_rodata
);
1061 unsigned long all_end
;
1063 printk(KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
1064 (end
- start
) >> 10);
1065 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
1067 kernel_set_to_readonly
= 1;
1070 * The rodata/data/bss/brk section (but not the kernel text!)
1071 * should also be not-executable.
1073 * We align all_end to PMD_SIZE because the existing mapping
1074 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1075 * split the PMD and the reminder between _brk_end and the end
1076 * of the PMD will remain mapped executable.
1078 * Any PMD which was setup after the one which covers _brk_end
1079 * has been zapped already via cleanup_highmem().
1081 all_end
= roundup((unsigned long)_brk_end
, PMD_SIZE
);
1082 set_memory_nx(text_end
, (all_end
- text_end
) >> PAGE_SHIFT
);
1086 #ifdef CONFIG_CPA_DEBUG
1087 printk(KERN_INFO
"Testing CPA: undo %lx-%lx\n", start
, end
);
1088 set_memory_rw(start
, (end
-start
) >> PAGE_SHIFT
);
1090 printk(KERN_INFO
"Testing CPA: again\n");
1091 set_memory_ro(start
, (end
-start
) >> PAGE_SHIFT
);
1094 free_init_pages("unused kernel",
1095 (unsigned long) __va(__pa_symbol(text_end
)),
1096 (unsigned long) __va(__pa_symbol(rodata_start
)));
1097 free_init_pages("unused kernel",
1098 (unsigned long) __va(__pa_symbol(rodata_end
)),
1099 (unsigned long) __va(__pa_symbol(_sdata
)));
1104 int kern_addr_valid(unsigned long addr
)
1106 unsigned long above
= ((long)addr
) >> __VIRTUAL_MASK_SHIFT
;
1112 if (above
!= 0 && above
!= -1UL)
1115 pgd
= pgd_offset_k(addr
);
1119 pud
= pud_offset(pgd
, addr
);
1123 if (pud_large(*pud
))
1124 return pfn_valid(pud_pfn(*pud
));
1126 pmd
= pmd_offset(pud
, addr
);
1130 if (pmd_large(*pmd
))
1131 return pfn_valid(pmd_pfn(*pmd
));
1133 pte
= pte_offset_kernel(pmd
, addr
);
1137 return pfn_valid(pte_pfn(*pte
));
1140 static unsigned long probe_memory_block_size(void)
1142 unsigned long bz
= MIN_MEMORY_BLOCK_SIZE
;
1144 /* if system is UV or has 64GB of RAM or more, use large blocks */
1145 if (is_uv_system() || ((max_pfn
<< PAGE_SHIFT
) >= (64UL << 30)))
1146 bz
= 2UL << 30; /* 2GB */
1148 pr_info("x86/mm: Memory block size: %ldMB\n", bz
>> 20);
1153 static unsigned long memory_block_size_probed
;
1154 unsigned long memory_block_size_bytes(void)
1156 if (!memory_block_size_probed
)
1157 memory_block_size_probed
= probe_memory_block_size();
1159 return memory_block_size_probed
;
1162 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1164 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1166 static long __meminitdata addr_start
, addr_end
;
1167 static void __meminitdata
*p_start
, *p_end
;
1168 static int __meminitdata node_start
;
1170 static int __meminit
vmemmap_populate_hugepages(unsigned long start
,
1171 unsigned long end
, int node
, struct vmem_altmap
*altmap
)
1179 for (addr
= start
; addr
< end
; addr
= next
) {
1180 next
= pmd_addr_end(addr
, end
);
1182 pgd
= vmemmap_pgd_populate(addr
, node
);
1186 pud
= vmemmap_pud_populate(pgd
, addr
, node
);
1190 pmd
= pmd_offset(pud
, addr
);
1191 if (pmd_none(*pmd
)) {
1194 p
= __vmemmap_alloc_block_buf(PMD_SIZE
, node
, altmap
);
1198 entry
= pfn_pte(__pa(p
) >> PAGE_SHIFT
,
1200 set_pmd(pmd
, __pmd(pte_val(entry
)));
1202 /* check to see if we have contiguous blocks */
1203 if (p_end
!= p
|| node_start
!= node
) {
1205 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1206 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);
1212 addr_end
= addr
+ PMD_SIZE
;
1213 p_end
= p
+ PMD_SIZE
;
1216 return -ENOMEM
; /* no fallback */
1217 } else if (pmd_large(*pmd
)) {
1218 vmemmap_verify((pte_t
*)pmd
, node
, addr
, next
);
1221 pr_warn_once("vmemmap: falling back to regular page backing\n");
1222 if (vmemmap_populate_basepages(addr
, next
, node
))
1228 int __meminit
vmemmap_populate(unsigned long start
, unsigned long end
, int node
)
1230 struct vmem_altmap
*altmap
= to_vmem_altmap(start
);
1233 if (boot_cpu_has(X86_FEATURE_PSE
))
1234 err
= vmemmap_populate_hugepages(start
, end
, node
, altmap
);
1236 pr_err_once("%s: no cpu support for altmap allocations\n",
1240 err
= vmemmap_populate_basepages(start
, end
, node
);
1242 sync_global_pgds(start
, end
- 1, 0);
1246 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1247 void register_page_bootmem_memmap(unsigned long section_nr
,
1248 struct page
*start_page
, unsigned long size
)
1250 unsigned long addr
= (unsigned long)start_page
;
1251 unsigned long end
= (unsigned long)(start_page
+ size
);
1256 unsigned int nr_pages
;
1259 for (; addr
< end
; addr
= next
) {
1262 pgd
= pgd_offset_k(addr
);
1263 if (pgd_none(*pgd
)) {
1264 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1267 get_page_bootmem(section_nr
, pgd_page(*pgd
), MIX_SECTION_INFO
);
1269 pud
= pud_offset(pgd
, addr
);
1270 if (pud_none(*pud
)) {
1271 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1274 get_page_bootmem(section_nr
, pud_page(*pud
), MIX_SECTION_INFO
);
1276 if (!boot_cpu_has(X86_FEATURE_PSE
)) {
1277 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1278 pmd
= pmd_offset(pud
, addr
);
1281 get_page_bootmem(section_nr
, pmd_page(*pmd
),
1284 pte
= pte_offset_kernel(pmd
, addr
);
1287 get_page_bootmem(section_nr
, pte_page(*pte
),
1290 next
= pmd_addr_end(addr
, end
);
1292 pmd
= pmd_offset(pud
, addr
);
1296 nr_pages
= 1 << (get_order(PMD_SIZE
));
1297 page
= pmd_page(*pmd
);
1299 get_page_bootmem(section_nr
, page
++,
1306 void __meminit
vmemmap_populate_print_last(void)
1309 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1310 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);