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/module.h>
31 #include <linux/memory.h>
32 #include <linux/memory_hotplug.h>
33 #include <linux/memremap.h>
34 #include <linux/nmi.h>
35 #include <linux/gfp.h>
36 #include <linux/kcore.h>
38 #include <asm/processor.h>
39 #include <asm/bios_ebda.h>
40 #include <asm/uaccess.h>
41 #include <asm/pgtable.h>
42 #include <asm/pgalloc.h>
44 #include <asm/fixmap.h>
48 #include <asm/mmu_context.h>
49 #include <asm/proto.h>
51 #include <asm/sections.h>
52 #include <asm/kdebug.h>
54 #include <asm/cacheflush.h>
56 #include <asm/uv/uv.h>
57 #include <asm/setup.h>
59 #include "mm_internal.h"
61 static void ident_pmd_init(unsigned long pmd_flag
, pmd_t
*pmd_page
,
62 unsigned long addr
, unsigned long end
)
65 for (; addr
< end
; addr
+= PMD_SIZE
) {
66 pmd_t
*pmd
= pmd_page
+ pmd_index(addr
);
68 if (!pmd_present(*pmd
))
69 set_pmd(pmd
, __pmd(addr
| pmd_flag
));
72 static int ident_pud_init(struct x86_mapping_info
*info
, pud_t
*pud_page
,
73 unsigned long addr
, unsigned long end
)
77 for (; addr
< end
; addr
= next
) {
78 pud_t
*pud
= pud_page
+ pud_index(addr
);
81 next
= (addr
& PUD_MASK
) + PUD_SIZE
;
85 if (pud_present(*pud
)) {
86 pmd
= pmd_offset(pud
, 0);
87 ident_pmd_init(info
->pmd_flag
, pmd
, addr
, next
);
90 pmd
= (pmd_t
*)info
->alloc_pgt_page(info
->context
);
93 ident_pmd_init(info
->pmd_flag
, pmd
, addr
, next
);
94 set_pud(pud
, __pud(__pa(pmd
) | _KERNPG_TABLE
));
100 int kernel_ident_mapping_init(struct x86_mapping_info
*info
, pgd_t
*pgd_page
,
101 unsigned long addr
, unsigned long end
)
105 int off
= info
->kernel_mapping
? pgd_index(__PAGE_OFFSET
) : 0;
107 for (; addr
< end
; addr
= next
) {
108 pgd_t
*pgd
= pgd_page
+ pgd_index(addr
) + off
;
111 next
= (addr
& PGDIR_MASK
) + PGDIR_SIZE
;
115 if (pgd_present(*pgd
)) {
116 pud
= pud_offset(pgd
, 0);
117 result
= ident_pud_init(info
, pud
, addr
, next
);
123 pud
= (pud_t
*)info
->alloc_pgt_page(info
->context
);
126 result
= ident_pud_init(info
, pud
, addr
, next
);
129 set_pgd(pgd
, __pgd(__pa(pud
) | _KERNPG_TABLE
));
136 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
137 * physical space so we can cache the place of the first one and move
138 * around without checking the pgd every time.
141 pteval_t __supported_pte_mask __read_mostly
= ~0;
142 EXPORT_SYMBOL_GPL(__supported_pte_mask
);
144 int force_personality32
;
148 * Control non executable heap for 32bit processes.
149 * To control the stack too use noexec=off
151 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
152 * off PROT_READ implies PROT_EXEC
154 static int __init
nonx32_setup(char *str
)
156 if (!strcmp(str
, "on"))
157 force_personality32
&= ~READ_IMPLIES_EXEC
;
158 else if (!strcmp(str
, "off"))
159 force_personality32
|= READ_IMPLIES_EXEC
;
162 __setup("noexec32=", nonx32_setup
);
165 * When memory was added/removed make sure all the processes MM have
166 * suitable PGD entries in the local PGD level page.
168 void sync_global_pgds(unsigned long start
, unsigned long end
, int removed
)
170 unsigned long address
;
172 for (address
= start
; address
<= end
; address
+= PGDIR_SIZE
) {
173 const pgd_t
*pgd_ref
= pgd_offset_k(address
);
177 * When it is called after memory hot remove, pgd_none()
178 * returns true. In this case (removed == 1), we must clear
179 * the PGD entries in the local PGD level page.
181 if (pgd_none(*pgd_ref
) && !removed
)
184 spin_lock(&pgd_lock
);
185 list_for_each_entry(page
, &pgd_list
, lru
) {
187 spinlock_t
*pgt_lock
;
189 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
190 /* the pgt_lock only for Xen */
191 pgt_lock
= &pgd_page_get_mm(page
)->page_table_lock
;
194 if (!pgd_none(*pgd_ref
) && !pgd_none(*pgd
))
195 BUG_ON(pgd_page_vaddr(*pgd
)
196 != pgd_page_vaddr(*pgd_ref
));
199 if (pgd_none(*pgd_ref
) && !pgd_none(*pgd
))
203 set_pgd(pgd
, *pgd_ref
);
206 spin_unlock(pgt_lock
);
208 spin_unlock(&pgd_lock
);
213 * NOTE: This function is marked __ref because it calls __init function
214 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
216 static __ref
void *spp_getpage(void)
221 ptr
= (void *) get_zeroed_page(GFP_ATOMIC
| __GFP_NOTRACK
);
223 ptr
= alloc_bootmem_pages(PAGE_SIZE
);
225 if (!ptr
|| ((unsigned long)ptr
& ~PAGE_MASK
)) {
226 panic("set_pte_phys: cannot allocate page data %s\n",
227 after_bootmem
? "after bootmem" : "");
230 pr_debug("spp_getpage %p\n", ptr
);
235 static pud_t
*fill_pud(pgd_t
*pgd
, unsigned long vaddr
)
237 if (pgd_none(*pgd
)) {
238 pud_t
*pud
= (pud_t
*)spp_getpage();
239 pgd_populate(&init_mm
, pgd
, pud
);
240 if (pud
!= pud_offset(pgd
, 0))
241 printk(KERN_ERR
"PAGETABLE BUG #00! %p <-> %p\n",
242 pud
, pud_offset(pgd
, 0));
244 return pud_offset(pgd
, vaddr
);
247 static pmd_t
*fill_pmd(pud_t
*pud
, unsigned long vaddr
)
249 if (pud_none(*pud
)) {
250 pmd_t
*pmd
= (pmd_t
*) spp_getpage();
251 pud_populate(&init_mm
, pud
, pmd
);
252 if (pmd
!= pmd_offset(pud
, 0))
253 printk(KERN_ERR
"PAGETABLE BUG #01! %p <-> %p\n",
254 pmd
, pmd_offset(pud
, 0));
256 return pmd_offset(pud
, vaddr
);
259 static pte_t
*fill_pte(pmd_t
*pmd
, unsigned long vaddr
)
261 if (pmd_none(*pmd
)) {
262 pte_t
*pte
= (pte_t
*) spp_getpage();
263 pmd_populate_kernel(&init_mm
, pmd
, pte
);
264 if (pte
!= pte_offset_kernel(pmd
, 0))
265 printk(KERN_ERR
"PAGETABLE BUG #02!\n");
267 return pte_offset_kernel(pmd
, vaddr
);
270 void set_pte_vaddr_pud(pud_t
*pud_page
, unsigned long vaddr
, pte_t new_pte
)
276 pud
= pud_page
+ pud_index(vaddr
);
277 pmd
= fill_pmd(pud
, vaddr
);
278 pte
= fill_pte(pmd
, vaddr
);
280 set_pte(pte
, new_pte
);
283 * It's enough to flush this one mapping.
284 * (PGE mappings get flushed as well)
286 __flush_tlb_one(vaddr
);
289 void set_pte_vaddr(unsigned long vaddr
, pte_t pteval
)
294 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr
, native_pte_val(pteval
));
296 pgd
= pgd_offset_k(vaddr
);
297 if (pgd_none(*pgd
)) {
299 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
302 pud_page
= (pud_t
*)pgd_page_vaddr(*pgd
);
303 set_pte_vaddr_pud(pud_page
, vaddr
, pteval
);
306 pmd_t
* __init
populate_extra_pmd(unsigned long vaddr
)
311 pgd
= pgd_offset_k(vaddr
);
312 pud
= fill_pud(pgd
, vaddr
);
313 return fill_pmd(pud
, vaddr
);
316 pte_t
* __init
populate_extra_pte(unsigned long vaddr
)
320 pmd
= populate_extra_pmd(vaddr
);
321 return fill_pte(pmd
, vaddr
);
325 * Create large page table mappings for a range of physical addresses.
327 static void __init
__init_extra_mapping(unsigned long phys
, unsigned long size
,
328 enum page_cache_mode cache
)
335 pgprot_val(prot
) = pgprot_val(PAGE_KERNEL_LARGE
) |
336 pgprot_val(pgprot_4k_2_large(cachemode2pgprot(cache
)));
337 BUG_ON((phys
& ~PMD_MASK
) || (size
& ~PMD_MASK
));
338 for (; size
; phys
+= PMD_SIZE
, size
-= PMD_SIZE
) {
339 pgd
= pgd_offset_k((unsigned long)__va(phys
));
340 if (pgd_none(*pgd
)) {
341 pud
= (pud_t
*) spp_getpage();
342 set_pgd(pgd
, __pgd(__pa(pud
) | _KERNPG_TABLE
|
345 pud
= pud_offset(pgd
, (unsigned long)__va(phys
));
346 if (pud_none(*pud
)) {
347 pmd
= (pmd_t
*) spp_getpage();
348 set_pud(pud
, __pud(__pa(pmd
) | _KERNPG_TABLE
|
351 pmd
= pmd_offset(pud
, phys
);
352 BUG_ON(!pmd_none(*pmd
));
353 set_pmd(pmd
, __pmd(phys
| pgprot_val(prot
)));
357 void __init
init_extra_mapping_wb(unsigned long phys
, unsigned long size
)
359 __init_extra_mapping(phys
, size
, _PAGE_CACHE_MODE_WB
);
362 void __init
init_extra_mapping_uc(unsigned long phys
, unsigned long size
)
364 __init_extra_mapping(phys
, size
, _PAGE_CACHE_MODE_UC
);
368 * The head.S code sets up the kernel high mapping:
370 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
372 * phys_base holds the negative offset to the kernel, which is added
373 * to the compile time generated pmds. This results in invalid pmds up
374 * to the point where we hit the physaddr 0 mapping.
376 * We limit the mappings to the region from _text to _brk_end. _brk_end
377 * is rounded up to the 2MB boundary. This catches the invalid pmds as
378 * well, as they are located before _text:
380 void __init
cleanup_highmap(void)
382 unsigned long vaddr
= __START_KERNEL_map
;
383 unsigned long vaddr_end
= __START_KERNEL_map
+ KERNEL_IMAGE_SIZE
;
384 unsigned long end
= roundup((unsigned long)_brk_end
, PMD_SIZE
) - 1;
385 pmd_t
*pmd
= level2_kernel_pgt
;
388 * Native path, max_pfn_mapped is not set yet.
389 * Xen has valid max_pfn_mapped set in
390 * arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
393 vaddr_end
= __START_KERNEL_map
+ (max_pfn_mapped
<< PAGE_SHIFT
);
395 for (; vaddr
+ PMD_SIZE
- 1 < vaddr_end
; pmd
++, vaddr
+= PMD_SIZE
) {
398 if (vaddr
< (unsigned long) _text
|| vaddr
> end
)
399 set_pmd(pmd
, __pmd(0));
403 static unsigned long __meminit
404 phys_pte_init(pte_t
*pte_page
, unsigned long addr
, unsigned long end
,
407 unsigned long pages
= 0, next
;
408 unsigned long last_map_addr
= end
;
411 pte_t
*pte
= pte_page
+ pte_index(addr
);
413 for (i
= pte_index(addr
); i
< PTRS_PER_PTE
; i
++, addr
= next
, pte
++) {
414 next
= (addr
& PAGE_MASK
) + PAGE_SIZE
;
416 if (!after_bootmem
&&
417 !e820_any_mapped(addr
& PAGE_MASK
, next
, E820_RAM
) &&
418 !e820_any_mapped(addr
& PAGE_MASK
, next
, E820_RESERVED_KERN
))
419 set_pte(pte
, __pte(0));
424 * We will re-use the existing mapping.
425 * Xen for example has some special requirements, like mapping
426 * pagetable pages as RO. So assume someone who pre-setup
427 * these mappings are more intelligent.
436 printk(" pte=%p addr=%lx pte=%016lx\n",
437 pte
, addr
, pfn_pte(addr
>> PAGE_SHIFT
, PAGE_KERNEL
).pte
);
439 set_pte(pte
, pfn_pte(addr
>> PAGE_SHIFT
, prot
));
440 last_map_addr
= (addr
& PAGE_MASK
) + PAGE_SIZE
;
443 update_page_count(PG_LEVEL_4K
, pages
);
445 return last_map_addr
;
448 static unsigned long __meminit
449 phys_pmd_init(pmd_t
*pmd_page
, unsigned long address
, unsigned long end
,
450 unsigned long page_size_mask
, pgprot_t prot
)
452 unsigned long pages
= 0, next
;
453 unsigned long last_map_addr
= end
;
455 int i
= pmd_index(address
);
457 for (; i
< PTRS_PER_PMD
; i
++, address
= next
) {
458 pmd_t
*pmd
= pmd_page
+ pmd_index(address
);
460 pgprot_t new_prot
= prot
;
462 next
= (address
& PMD_MASK
) + PMD_SIZE
;
463 if (address
>= end
) {
464 if (!after_bootmem
&&
465 !e820_any_mapped(address
& PMD_MASK
, next
, E820_RAM
) &&
466 !e820_any_mapped(address
& PMD_MASK
, next
, E820_RESERVED_KERN
))
467 set_pmd(pmd
, __pmd(0));
472 if (!pmd_large(*pmd
)) {
473 spin_lock(&init_mm
.page_table_lock
);
474 pte
= (pte_t
*)pmd_page_vaddr(*pmd
);
475 last_map_addr
= phys_pte_init(pte
, address
,
477 spin_unlock(&init_mm
.page_table_lock
);
481 * If we are ok with PG_LEVEL_2M mapping, then we will
482 * use the existing mapping,
484 * Otherwise, we will split the large page mapping but
485 * use the same existing protection bits except for
486 * large page, so that we don't violate Intel's TLB
487 * Application note (317080) which says, while changing
488 * the page sizes, new and old translations should
489 * not differ with respect to page frame and
492 if (page_size_mask
& (1 << PG_LEVEL_2M
)) {
495 last_map_addr
= next
;
498 new_prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pmd
));
501 if (page_size_mask
& (1<<PG_LEVEL_2M
)) {
503 spin_lock(&init_mm
.page_table_lock
);
504 set_pte((pte_t
*)pmd
,
505 pfn_pte((address
& PMD_MASK
) >> PAGE_SHIFT
,
506 __pgprot(pgprot_val(prot
) | _PAGE_PSE
)));
507 spin_unlock(&init_mm
.page_table_lock
);
508 last_map_addr
= next
;
512 pte
= alloc_low_page();
513 last_map_addr
= phys_pte_init(pte
, address
, end
, new_prot
);
515 spin_lock(&init_mm
.page_table_lock
);
516 pmd_populate_kernel(&init_mm
, pmd
, pte
);
517 spin_unlock(&init_mm
.page_table_lock
);
519 update_page_count(PG_LEVEL_2M
, pages
);
520 return last_map_addr
;
523 static unsigned long __meminit
524 phys_pud_init(pud_t
*pud_page
, unsigned long addr
, unsigned long end
,
525 unsigned long page_size_mask
)
527 unsigned long pages
= 0, next
;
528 unsigned long last_map_addr
= end
;
529 int i
= pud_index(addr
);
531 for (; i
< PTRS_PER_PUD
; i
++, addr
= next
) {
532 pud_t
*pud
= pud_page
+ pud_index(addr
);
534 pgprot_t prot
= PAGE_KERNEL
;
536 next
= (addr
& PUD_MASK
) + PUD_SIZE
;
538 if (!after_bootmem
&&
539 !e820_any_mapped(addr
& PUD_MASK
, next
, E820_RAM
) &&
540 !e820_any_mapped(addr
& PUD_MASK
, next
, E820_RESERVED_KERN
))
541 set_pud(pud
, __pud(0));
546 if (!pud_large(*pud
)) {
547 pmd
= pmd_offset(pud
, 0);
548 last_map_addr
= phys_pmd_init(pmd
, addr
, end
,
549 page_size_mask
, prot
);
554 * If we are ok with PG_LEVEL_1G mapping, then we will
555 * use the existing mapping.
557 * Otherwise, we will split the gbpage mapping but use
558 * the same existing protection bits except for large
559 * page, so that we don't violate Intel's TLB
560 * Application note (317080) which says, while changing
561 * the page sizes, new and old translations should
562 * not differ with respect to page frame and
565 if (page_size_mask
& (1 << PG_LEVEL_1G
)) {
568 last_map_addr
= next
;
571 prot
= pte_pgprot(pte_clrhuge(*(pte_t
*)pud
));
574 if (page_size_mask
& (1<<PG_LEVEL_1G
)) {
576 spin_lock(&init_mm
.page_table_lock
);
577 set_pte((pte_t
*)pud
,
578 pfn_pte((addr
& PUD_MASK
) >> PAGE_SHIFT
,
580 spin_unlock(&init_mm
.page_table_lock
);
581 last_map_addr
= next
;
585 pmd
= alloc_low_page();
586 last_map_addr
= phys_pmd_init(pmd
, addr
, end
, page_size_mask
,
589 spin_lock(&init_mm
.page_table_lock
);
590 pud_populate(&init_mm
, pud
, pmd
);
591 spin_unlock(&init_mm
.page_table_lock
);
595 update_page_count(PG_LEVEL_1G
, pages
);
597 return last_map_addr
;
600 unsigned long __meminit
601 kernel_physical_mapping_init(unsigned long start
,
603 unsigned long page_size_mask
)
605 bool pgd_changed
= false;
606 unsigned long next
, last_map_addr
= end
;
609 start
= (unsigned long)__va(start
);
610 end
= (unsigned long)__va(end
);
613 for (; start
< end
; start
= next
) {
614 pgd_t
*pgd
= pgd_offset_k(start
);
617 next
= (start
& PGDIR_MASK
) + PGDIR_SIZE
;
620 pud
= (pud_t
*)pgd_page_vaddr(*pgd
);
621 last_map_addr
= phys_pud_init(pud
, __pa(start
),
622 __pa(end
), page_size_mask
);
626 pud
= alloc_low_page();
627 last_map_addr
= phys_pud_init(pud
, __pa(start
), __pa(end
),
630 spin_lock(&init_mm
.page_table_lock
);
631 pgd_populate(&init_mm
, pgd
, pud
);
632 spin_unlock(&init_mm
.page_table_lock
);
637 sync_global_pgds(addr
, end
- 1, 0);
641 return last_map_addr
;
645 void __init
initmem_init(void)
647 memblock_set_node(0, (phys_addr_t
)ULLONG_MAX
, &memblock
.memory
, 0);
651 void __init
paging_init(void)
653 sparse_memory_present_with_active_regions(MAX_NUMNODES
);
657 * clear the default setting with node 0
658 * note: don't use nodes_clear here, that is really clearing when
659 * numa support is not compiled in, and later node_set_state
660 * will not set it back.
662 node_clear_state(0, N_MEMORY
);
663 if (N_MEMORY
!= N_NORMAL_MEMORY
)
664 node_clear_state(0, N_NORMAL_MEMORY
);
670 * Memory hotplug specific functions
672 #ifdef CONFIG_MEMORY_HOTPLUG
674 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
677 static void update_end_of_memory_vars(u64 start
, u64 size
)
679 unsigned long end_pfn
= PFN_UP(start
+ size
);
681 if (end_pfn
> max_pfn
) {
683 max_low_pfn
= end_pfn
;
684 high_memory
= (void *)__va(max_pfn
* PAGE_SIZE
- 1) + 1;
689 * Memory is added always to NORMAL zone. This means you will never get
690 * additional DMA/DMA32 memory.
692 int arch_add_memory(int nid
, u64 start
, u64 size
, bool for_device
)
694 struct pglist_data
*pgdat
= NODE_DATA(nid
);
695 struct zone
*zone
= pgdat
->node_zones
+
696 zone_for_memory(nid
, start
, size
, ZONE_NORMAL
, for_device
);
697 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
698 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
701 init_memory_mapping(start
, start
+ size
);
703 ret
= __add_pages(nid
, zone
, start_pfn
, nr_pages
);
706 /* update max_pfn, max_low_pfn and high_memory */
707 update_end_of_memory_vars(start
, size
);
711 EXPORT_SYMBOL_GPL(arch_add_memory
);
713 #define PAGE_INUSE 0xFD
715 static void __meminit
free_pagetable(struct page
*page
, int order
)
718 unsigned int nr_pages
= 1 << order
;
719 struct vmem_altmap
*altmap
= to_vmem_altmap((unsigned long) page
);
722 vmem_altmap_free(altmap
, nr_pages
);
726 /* bootmem page has reserved flag */
727 if (PageReserved(page
)) {
728 __ClearPageReserved(page
);
730 magic
= (unsigned long)page
->lru
.next
;
731 if (magic
== SECTION_INFO
|| magic
== MIX_SECTION_INFO
) {
733 put_page_bootmem(page
++);
736 free_reserved_page(page
++);
738 free_pages((unsigned long)page_address(page
), order
);
741 static void __meminit
free_pte_table(pte_t
*pte_start
, pmd_t
*pmd
)
746 for (i
= 0; i
< PTRS_PER_PTE
; i
++) {
752 /* free a pte talbe */
753 free_pagetable(pmd_page(*pmd
), 0);
754 spin_lock(&init_mm
.page_table_lock
);
756 spin_unlock(&init_mm
.page_table_lock
);
759 static void __meminit
free_pmd_table(pmd_t
*pmd_start
, pud_t
*pud
)
764 for (i
= 0; i
< PTRS_PER_PMD
; i
++) {
770 /* free a pmd talbe */
771 free_pagetable(pud_page(*pud
), 0);
772 spin_lock(&init_mm
.page_table_lock
);
774 spin_unlock(&init_mm
.page_table_lock
);
777 /* Return true if pgd is changed, otherwise return false. */
778 static bool __meminit
free_pud_table(pud_t
*pud_start
, pgd_t
*pgd
)
783 for (i
= 0; i
< PTRS_PER_PUD
; i
++) {
789 /* free a pud table */
790 free_pagetable(pgd_page(*pgd
), 0);
791 spin_lock(&init_mm
.page_table_lock
);
793 spin_unlock(&init_mm
.page_table_lock
);
798 static void __meminit
799 remove_pte_table(pte_t
*pte_start
, unsigned long addr
, unsigned long end
,
802 unsigned long next
, pages
= 0;
805 phys_addr_t phys_addr
;
807 pte
= pte_start
+ pte_index(addr
);
808 for (; addr
< end
; addr
= next
, pte
++) {
809 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
813 if (!pte_present(*pte
))
817 * We mapped [0,1G) memory as identity mapping when
818 * initializing, in arch/x86/kernel/head_64.S. These
819 * pagetables cannot be removed.
821 phys_addr
= pte_val(*pte
) + (addr
& PAGE_MASK
);
822 if (phys_addr
< (phys_addr_t
)0x40000000)
825 if (PAGE_ALIGNED(addr
) && PAGE_ALIGNED(next
)) {
827 * Do not free direct mapping pages since they were
828 * freed when offlining, or simplely not in use.
831 free_pagetable(pte_page(*pte
), 0);
833 spin_lock(&init_mm
.page_table_lock
);
834 pte_clear(&init_mm
, addr
, pte
);
835 spin_unlock(&init_mm
.page_table_lock
);
837 /* For non-direct mapping, pages means nothing. */
841 * If we are here, we are freeing vmemmap pages since
842 * direct mapped memory ranges to be freed are aligned.
844 * If we are not removing the whole page, it means
845 * other page structs in this page are being used and
846 * we canot remove them. So fill the unused page_structs
847 * with 0xFD, and remove the page when it is wholly
850 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
852 page_addr
= page_address(pte_page(*pte
));
853 if (!memchr_inv(page_addr
, PAGE_INUSE
, PAGE_SIZE
)) {
854 free_pagetable(pte_page(*pte
), 0);
856 spin_lock(&init_mm
.page_table_lock
);
857 pte_clear(&init_mm
, addr
, pte
);
858 spin_unlock(&init_mm
.page_table_lock
);
863 /* Call free_pte_table() in remove_pmd_table(). */
866 update_page_count(PG_LEVEL_4K
, -pages
);
869 static void __meminit
870 remove_pmd_table(pmd_t
*pmd_start
, unsigned long addr
, unsigned long end
,
873 unsigned long next
, pages
= 0;
878 pmd
= pmd_start
+ pmd_index(addr
);
879 for (; addr
< end
; addr
= next
, pmd
++) {
880 next
= pmd_addr_end(addr
, end
);
882 if (!pmd_present(*pmd
))
885 if (pmd_large(*pmd
)) {
886 if (IS_ALIGNED(addr
, PMD_SIZE
) &&
887 IS_ALIGNED(next
, PMD_SIZE
)) {
889 free_pagetable(pmd_page(*pmd
),
890 get_order(PMD_SIZE
));
892 spin_lock(&init_mm
.page_table_lock
);
894 spin_unlock(&init_mm
.page_table_lock
);
897 /* If here, we are freeing vmemmap pages. */
898 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
900 page_addr
= page_address(pmd_page(*pmd
));
901 if (!memchr_inv(page_addr
, PAGE_INUSE
,
903 free_pagetable(pmd_page(*pmd
),
904 get_order(PMD_SIZE
));
906 spin_lock(&init_mm
.page_table_lock
);
908 spin_unlock(&init_mm
.page_table_lock
);
915 pte_base
= (pte_t
*)pmd_page_vaddr(*pmd
);
916 remove_pte_table(pte_base
, addr
, next
, direct
);
917 free_pte_table(pte_base
, pmd
);
920 /* Call free_pmd_table() in remove_pud_table(). */
922 update_page_count(PG_LEVEL_2M
, -pages
);
925 static void __meminit
926 remove_pud_table(pud_t
*pud_start
, unsigned long addr
, unsigned long end
,
929 unsigned long next
, pages
= 0;
934 pud
= pud_start
+ pud_index(addr
);
935 for (; addr
< end
; addr
= next
, pud
++) {
936 next
= pud_addr_end(addr
, end
);
938 if (!pud_present(*pud
))
941 if (pud_large(*pud
)) {
942 if (IS_ALIGNED(addr
, PUD_SIZE
) &&
943 IS_ALIGNED(next
, PUD_SIZE
)) {
945 free_pagetable(pud_page(*pud
),
946 get_order(PUD_SIZE
));
948 spin_lock(&init_mm
.page_table_lock
);
950 spin_unlock(&init_mm
.page_table_lock
);
953 /* If here, we are freeing vmemmap pages. */
954 memset((void *)addr
, PAGE_INUSE
, next
- addr
);
956 page_addr
= page_address(pud_page(*pud
));
957 if (!memchr_inv(page_addr
, PAGE_INUSE
,
959 free_pagetable(pud_page(*pud
),
960 get_order(PUD_SIZE
));
962 spin_lock(&init_mm
.page_table_lock
);
964 spin_unlock(&init_mm
.page_table_lock
);
971 pmd_base
= (pmd_t
*)pud_page_vaddr(*pud
);
972 remove_pmd_table(pmd_base
, addr
, next
, direct
);
973 free_pmd_table(pmd_base
, pud
);
977 update_page_count(PG_LEVEL_1G
, -pages
);
980 /* start and end are both virtual address. */
981 static void __meminit
982 remove_pagetable(unsigned long start
, unsigned long end
, bool direct
)
988 bool pgd_changed
= false;
990 for (addr
= start
; addr
< end
; addr
= next
) {
991 next
= pgd_addr_end(addr
, end
);
993 pgd
= pgd_offset_k(addr
);
994 if (!pgd_present(*pgd
))
997 pud
= (pud_t
*)pgd_page_vaddr(*pgd
);
998 remove_pud_table(pud
, addr
, next
, direct
);
999 if (free_pud_table(pud
, pgd
))
1004 sync_global_pgds(start
, end
- 1, 1);
1009 void __ref
vmemmap_free(unsigned long start
, unsigned long end
)
1011 remove_pagetable(start
, end
, false);
1014 #ifdef CONFIG_MEMORY_HOTREMOVE
1015 static void __meminit
1016 kernel_physical_mapping_remove(unsigned long start
, unsigned long end
)
1018 start
= (unsigned long)__va(start
);
1019 end
= (unsigned long)__va(end
);
1021 remove_pagetable(start
, end
, true);
1024 int __ref
arch_remove_memory(u64 start
, u64 size
)
1026 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1027 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
1028 struct page
*page
= pfn_to_page(start_pfn
);
1029 struct vmem_altmap
*altmap
;
1033 /* With altmap the first mapped page is offset from @start */
1034 altmap
= to_vmem_altmap((unsigned long) page
);
1036 page
+= vmem_altmap_offset(altmap
);
1037 zone
= page_zone(page
);
1038 ret
= __remove_pages(zone
, start_pfn
, nr_pages
);
1040 kernel_physical_mapping_remove(start
, start
+ size
);
1045 #endif /* CONFIG_MEMORY_HOTPLUG */
1047 static struct kcore_list kcore_vsyscall
;
1049 static void __init
register_page_bootmem_info(void)
1054 for_each_online_node(i
)
1055 register_page_bootmem_info_node(NODE_DATA(i
));
1059 void __init
mem_init(void)
1063 /* clear_bss() already clear the empty_zero_page */
1065 register_page_bootmem_info();
1067 /* this will put all memory onto the freelists */
1071 /* Register memory areas for /proc/kcore */
1072 kclist_add(&kcore_vsyscall
, (void *)VSYSCALL_ADDR
,
1073 PAGE_SIZE
, KCORE_OTHER
);
1075 mem_init_print_info(NULL
);
1078 const int rodata_test_data
= 0xC3;
1079 EXPORT_SYMBOL_GPL(rodata_test_data
);
1081 int kernel_set_to_readonly
;
1083 void set_kernel_text_rw(void)
1085 unsigned long start
= PFN_ALIGN(_text
);
1086 unsigned long end
= PFN_ALIGN(__stop___ex_table
);
1088 if (!kernel_set_to_readonly
)
1091 pr_debug("Set kernel text: %lx - %lx for read write\n",
1095 * Make the kernel identity mapping for text RW. Kernel text
1096 * mapping will always be RO. Refer to the comment in
1097 * static_protections() in pageattr.c
1099 set_memory_rw(start
, (end
- start
) >> PAGE_SHIFT
);
1102 void set_kernel_text_ro(void)
1104 unsigned long start
= PFN_ALIGN(_text
);
1105 unsigned long end
= PFN_ALIGN(__stop___ex_table
);
1107 if (!kernel_set_to_readonly
)
1110 pr_debug("Set kernel text: %lx - %lx for read only\n",
1114 * Set the kernel identity mapping for text RO.
1116 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
1119 void mark_rodata_ro(void)
1121 unsigned long start
= PFN_ALIGN(_text
);
1122 unsigned long rodata_start
= PFN_ALIGN(__start_rodata
);
1123 unsigned long end
= (unsigned long) &__end_rodata_hpage_align
;
1124 unsigned long text_end
= PFN_ALIGN(&__stop___ex_table
);
1125 unsigned long rodata_end
= PFN_ALIGN(&__end_rodata
);
1126 unsigned long all_end
;
1128 printk(KERN_INFO
"Write protecting the kernel read-only data: %luk\n",
1129 (end
- start
) >> 10);
1130 set_memory_ro(start
, (end
- start
) >> PAGE_SHIFT
);
1132 kernel_set_to_readonly
= 1;
1135 * The rodata/data/bss/brk section (but not the kernel text!)
1136 * should also be not-executable.
1138 * We align all_end to PMD_SIZE because the existing mapping
1139 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1140 * split the PMD and the reminder between _brk_end and the end
1141 * of the PMD will remain mapped executable.
1143 * Any PMD which was setup after the one which covers _brk_end
1144 * has been zapped already via cleanup_highmem().
1146 all_end
= roundup((unsigned long)_brk_end
, PMD_SIZE
);
1147 set_memory_nx(text_end
, (all_end
- text_end
) >> PAGE_SHIFT
);
1151 #ifdef CONFIG_CPA_DEBUG
1152 printk(KERN_INFO
"Testing CPA: undo %lx-%lx\n", start
, end
);
1153 set_memory_rw(start
, (end
-start
) >> PAGE_SHIFT
);
1155 printk(KERN_INFO
"Testing CPA: again\n");
1156 set_memory_ro(start
, (end
-start
) >> PAGE_SHIFT
);
1159 free_init_pages("unused kernel",
1160 (unsigned long) __va(__pa_symbol(text_end
)),
1161 (unsigned long) __va(__pa_symbol(rodata_start
)));
1162 free_init_pages("unused kernel",
1163 (unsigned long) __va(__pa_symbol(rodata_end
)),
1164 (unsigned long) __va(__pa_symbol(_sdata
)));
1169 int kern_addr_valid(unsigned long addr
)
1171 unsigned long above
= ((long)addr
) >> __VIRTUAL_MASK_SHIFT
;
1177 if (above
!= 0 && above
!= -1UL)
1180 pgd
= pgd_offset_k(addr
);
1184 pud
= pud_offset(pgd
, addr
);
1188 if (pud_large(*pud
))
1189 return pfn_valid(pud_pfn(*pud
));
1191 pmd
= pmd_offset(pud
, addr
);
1195 if (pmd_large(*pmd
))
1196 return pfn_valid(pmd_pfn(*pmd
));
1198 pte
= pte_offset_kernel(pmd
, addr
);
1202 return pfn_valid(pte_pfn(*pte
));
1205 static unsigned long probe_memory_block_size(void)
1207 unsigned long bz
= MIN_MEMORY_BLOCK_SIZE
;
1209 /* if system is UV or has 64GB of RAM or more, use large blocks */
1210 if (is_uv_system() || ((max_pfn
<< PAGE_SHIFT
) >= (64UL << 30)))
1211 bz
= 2UL << 30; /* 2GB */
1213 pr_info("x86/mm: Memory block size: %ldMB\n", bz
>> 20);
1218 static unsigned long memory_block_size_probed
;
1219 unsigned long memory_block_size_bytes(void)
1221 if (!memory_block_size_probed
)
1222 memory_block_size_probed
= probe_memory_block_size();
1224 return memory_block_size_probed
;
1227 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1229 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1231 static long __meminitdata addr_start
, addr_end
;
1232 static void __meminitdata
*p_start
, *p_end
;
1233 static int __meminitdata node_start
;
1235 static int __meminit
vmemmap_populate_hugepages(unsigned long start
,
1236 unsigned long end
, int node
, struct vmem_altmap
*altmap
)
1244 for (addr
= start
; addr
< end
; addr
= next
) {
1245 next
= pmd_addr_end(addr
, end
);
1247 pgd
= vmemmap_pgd_populate(addr
, node
);
1251 pud
= vmemmap_pud_populate(pgd
, addr
, node
);
1255 pmd
= pmd_offset(pud
, addr
);
1256 if (pmd_none(*pmd
)) {
1259 p
= __vmemmap_alloc_block_buf(PMD_SIZE
, node
, altmap
);
1263 entry
= pfn_pte(__pa(p
) >> PAGE_SHIFT
,
1265 set_pmd(pmd
, __pmd(pte_val(entry
)));
1267 /* check to see if we have contiguous blocks */
1268 if (p_end
!= p
|| node_start
!= node
) {
1270 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1271 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);
1277 addr_end
= addr
+ PMD_SIZE
;
1278 p_end
= p
+ PMD_SIZE
;
1281 return -ENOMEM
; /* no fallback */
1282 } else if (pmd_large(*pmd
)) {
1283 vmemmap_verify((pte_t
*)pmd
, node
, addr
, next
);
1286 pr_warn_once("vmemmap: falling back to regular page backing\n");
1287 if (vmemmap_populate_basepages(addr
, next
, node
))
1293 int __meminit
vmemmap_populate(unsigned long start
, unsigned long end
, int node
)
1295 struct vmem_altmap
*altmap
= to_vmem_altmap(start
);
1299 err
= vmemmap_populate_hugepages(start
, end
, node
, altmap
);
1301 pr_err_once("%s: no cpu support for altmap allocations\n",
1305 err
= vmemmap_populate_basepages(start
, end
, node
);
1307 sync_global_pgds(start
, end
- 1, 0);
1311 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1312 void register_page_bootmem_memmap(unsigned long section_nr
,
1313 struct page
*start_page
, unsigned long size
)
1315 unsigned long addr
= (unsigned long)start_page
;
1316 unsigned long end
= (unsigned long)(start_page
+ size
);
1321 unsigned int nr_pages
;
1324 for (; addr
< end
; addr
= next
) {
1327 pgd
= pgd_offset_k(addr
);
1328 if (pgd_none(*pgd
)) {
1329 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1332 get_page_bootmem(section_nr
, pgd_page(*pgd
), MIX_SECTION_INFO
);
1334 pud
= pud_offset(pgd
, addr
);
1335 if (pud_none(*pud
)) {
1336 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1339 get_page_bootmem(section_nr
, pud_page(*pud
), MIX_SECTION_INFO
);
1342 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
1343 pmd
= pmd_offset(pud
, addr
);
1346 get_page_bootmem(section_nr
, pmd_page(*pmd
),
1349 pte
= pte_offset_kernel(pmd
, addr
);
1352 get_page_bootmem(section_nr
, pte_page(*pte
),
1355 next
= pmd_addr_end(addr
, end
);
1357 pmd
= pmd_offset(pud
, addr
);
1361 nr_pages
= 1 << (get_order(PMD_SIZE
));
1362 page
= pmd_page(*pmd
);
1364 get_page_bootmem(section_nr
, page
++,
1371 void __meminit
vmemmap_populate_print_last(void)
1374 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1375 addr_start
, addr_end
-1, p_start
, p_end
-1, node_start
);