]>
git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - arch/x86/mm/pageattr.c
2 * Copyright 2002 Andi Kleen, SuSE Labs.
3 * Thanks to Ben LaHaise for precious feedback.
5 #include <linux/highmem.h>
6 #include <linux/bootmem.h>
7 #include <linux/module.h>
8 #include <linux/sched.h>
9 #include <linux/slab.h>
11 #include <linux/interrupt.h>
12 #include <linux/seq_file.h>
13 #include <linux/debugfs.h>
16 #include <asm/processor.h>
17 #include <asm/tlbflush.h>
18 #include <asm/sections.h>
19 #include <asm/setup.h>
20 #include <asm/uaccess.h>
21 #include <asm/pgalloc.h>
22 #include <asm/proto.h>
26 * The current flushing context - we pass it instead of 5 arguments:
35 unsigned force_split
: 1;
41 * Serialize cpa() (for !DEBUG_PAGEALLOC which uses large identity mappings)
42 * using cpa_lock. So that we don't allow any other cpu, with stale large tlb
43 * entries change the page attribute in parallel to some other cpu
44 * splitting a large page entry along with changing the attribute.
46 static DEFINE_SPINLOCK(cpa_lock
);
48 #define CPA_FLUSHTLB 1
50 #define CPA_PAGES_ARRAY 4
53 static unsigned long direct_pages_count
[PG_LEVEL_NUM
];
55 void update_page_count(int level
, unsigned long pages
)
59 /* Protect against CPA */
60 spin_lock_irqsave(&pgd_lock
, flags
);
61 direct_pages_count
[level
] += pages
;
62 spin_unlock_irqrestore(&pgd_lock
, flags
);
65 static void split_page_count(int level
)
67 direct_pages_count
[level
]--;
68 direct_pages_count
[level
- 1] += PTRS_PER_PTE
;
71 void arch_report_meminfo(struct seq_file
*m
)
73 seq_printf(m
, "DirectMap4k: %8lu kB\n",
74 direct_pages_count
[PG_LEVEL_4K
] << 2);
75 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
76 seq_printf(m
, "DirectMap2M: %8lu kB\n",
77 direct_pages_count
[PG_LEVEL_2M
] << 11);
79 seq_printf(m
, "DirectMap4M: %8lu kB\n",
80 direct_pages_count
[PG_LEVEL_2M
] << 12);
84 seq_printf(m
, "DirectMap1G: %8lu kB\n",
85 direct_pages_count
[PG_LEVEL_1G
] << 20);
89 static inline void split_page_count(int level
) { }
94 static inline unsigned long highmap_start_pfn(void)
96 return __pa(_text
) >> PAGE_SHIFT
;
99 static inline unsigned long highmap_end_pfn(void)
101 return __pa(roundup(_brk_end
, PMD_SIZE
)) >> PAGE_SHIFT
;
106 #ifdef CONFIG_DEBUG_PAGEALLOC
107 # define debug_pagealloc 1
109 # define debug_pagealloc 0
113 within(unsigned long addr
, unsigned long start
, unsigned long end
)
115 return addr
>= start
&& addr
< end
;
123 * clflush_cache_range - flush a cache range with clflush
124 * @addr: virtual start address
125 * @size: number of bytes to flush
127 * clflush is an unordered instruction which needs fencing with mfence
128 * to avoid ordering issues.
130 void clflush_cache_range(void *vaddr
, unsigned int size
)
132 void *vend
= vaddr
+ size
- 1;
136 for (; vaddr
< vend
; vaddr
+= boot_cpu_data
.x86_clflush_size
)
139 * Flush any possible final partial cacheline:
146 static void __cpa_flush_all(void *arg
)
148 unsigned long cache
= (unsigned long)arg
;
151 * Flush all to work around Errata in early athlons regarding
152 * large page flushing.
156 if (cache
&& boot_cpu_data
.x86
>= 4)
160 static void cpa_flush_all(unsigned long cache
)
162 BUG_ON(irqs_disabled());
164 on_each_cpu(__cpa_flush_all
, (void *) cache
, 1);
167 static void __cpa_flush_range(void *arg
)
170 * We could optimize that further and do individual per page
171 * tlb invalidates for a low number of pages. Caveat: we must
172 * flush the high aliases on 64bit as well.
177 static void cpa_flush_range(unsigned long start
, int numpages
, int cache
)
179 unsigned int i
, level
;
182 BUG_ON(irqs_disabled());
183 WARN_ON(PAGE_ALIGN(start
) != start
);
185 on_each_cpu(__cpa_flush_range
, NULL
, 1);
191 * We only need to flush on one CPU,
192 * clflush is a MESI-coherent instruction that
193 * will cause all other CPUs to flush the same
196 for (i
= 0, addr
= start
; i
< numpages
; i
++, addr
+= PAGE_SIZE
) {
197 pte_t
*pte
= lookup_address(addr
, &level
);
200 * Only flush present addresses:
202 if (pte
&& (pte_val(*pte
) & _PAGE_PRESENT
))
203 clflush_cache_range((void *) addr
, PAGE_SIZE
);
207 static void cpa_flush_array(unsigned long *start
, int numpages
, int cache
,
208 int in_flags
, struct page
**pages
)
210 unsigned int i
, level
;
211 unsigned long do_wbinvd
= cache
&& numpages
>= 1024; /* 4M threshold */
213 BUG_ON(irqs_disabled());
215 on_each_cpu(__cpa_flush_all
, (void *) do_wbinvd
, 1);
217 if (!cache
|| do_wbinvd
)
221 * We only need to flush on one CPU,
222 * clflush is a MESI-coherent instruction that
223 * will cause all other CPUs to flush the same
226 for (i
= 0; i
< numpages
; i
++) {
230 if (in_flags
& CPA_PAGES_ARRAY
)
231 addr
= (unsigned long)page_address(pages
[i
]);
235 pte
= lookup_address(addr
, &level
);
238 * Only flush present addresses:
240 if (pte
&& (pte_val(*pte
) & _PAGE_PRESENT
))
241 clflush_cache_range((void *)addr
, PAGE_SIZE
);
246 * Certain areas of memory on x86 require very specific protection flags,
247 * for example the BIOS area or kernel text. Callers don't always get this
248 * right (again, ioremap() on BIOS memory is not uncommon) so this function
249 * checks and fixes these known static required protection bits.
251 static inline pgprot_t
static_protections(pgprot_t prot
, unsigned long address
,
254 pgprot_t forbidden
= __pgprot(0);
257 * The BIOS area between 640k and 1Mb needs to be executable for
258 * PCI BIOS based config access (CONFIG_PCI_GOBIOS) support.
260 if (within(pfn
, BIOS_BEGIN
>> PAGE_SHIFT
, BIOS_END
>> PAGE_SHIFT
))
261 pgprot_val(forbidden
) |= _PAGE_NX
;
264 * The kernel text needs to be executable for obvious reasons
265 * Does not cover __inittext since that is gone later on. On
266 * 64bit we do not enforce !NX on the low mapping
268 if (within(address
, (unsigned long)_text
, (unsigned long)_etext
))
269 pgprot_val(forbidden
) |= _PAGE_NX
;
272 * The .rodata section needs to be read-only. Using the pfn
273 * catches all aliases.
275 if (within(pfn
, __pa((unsigned long)__start_rodata
) >> PAGE_SHIFT
,
276 __pa((unsigned long)__end_rodata
) >> PAGE_SHIFT
))
277 pgprot_val(forbidden
) |= _PAGE_RW
;
279 prot
= __pgprot(pgprot_val(prot
) & ~pgprot_val(forbidden
));
285 * Lookup the page table entry for a virtual address. Return a pointer
286 * to the entry and the level of the mapping.
288 * Note: We return pud and pmd either when the entry is marked large
289 * or when the present bit is not set. Otherwise we would return a
290 * pointer to a nonexisting mapping.
292 pte_t
*lookup_address(unsigned long address
, unsigned int *level
)
294 pgd_t
*pgd
= pgd_offset_k(address
);
298 *level
= PG_LEVEL_NONE
;
303 pud
= pud_offset(pgd
, address
);
307 *level
= PG_LEVEL_1G
;
308 if (pud_large(*pud
) || !pud_present(*pud
))
311 pmd
= pmd_offset(pud
, address
);
315 *level
= PG_LEVEL_2M
;
316 if (pmd_large(*pmd
) || !pmd_present(*pmd
))
319 *level
= PG_LEVEL_4K
;
321 return pte_offset_kernel(pmd
, address
);
323 EXPORT_SYMBOL_GPL(lookup_address
);
326 * Set the new pmd in all the pgds we know about:
328 static void __set_pmd_pte(pte_t
*kpte
, unsigned long address
, pte_t pte
)
331 set_pte_atomic(kpte
, pte
);
333 if (!SHARED_KERNEL_PMD
) {
336 list_for_each_entry(page
, &pgd_list
, lru
) {
341 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
342 pud
= pud_offset(pgd
, address
);
343 pmd
= pmd_offset(pud
, address
);
344 set_pte_atomic((pte_t
*)pmd
, pte
);
351 try_preserve_large_page(pte_t
*kpte
, unsigned long address
,
352 struct cpa_data
*cpa
)
354 unsigned long nextpage_addr
, numpages
, pmask
, psize
, flags
, addr
, pfn
;
355 pte_t new_pte
, old_pte
, *tmp
;
356 pgprot_t old_prot
, new_prot
;
360 if (cpa
->force_split
)
363 spin_lock_irqsave(&pgd_lock
, flags
);
365 * Check for races, another CPU might have split this page
368 tmp
= lookup_address(address
, &level
);
374 psize
= PMD_PAGE_SIZE
;
375 pmask
= PMD_PAGE_MASK
;
379 psize
= PUD_PAGE_SIZE
;
380 pmask
= PUD_PAGE_MASK
;
389 * Calculate the number of pages, which fit into this large
390 * page starting at address:
392 nextpage_addr
= (address
+ psize
) & pmask
;
393 numpages
= (nextpage_addr
- address
) >> PAGE_SHIFT
;
394 if (numpages
< cpa
->numpages
)
395 cpa
->numpages
= numpages
;
398 * We are safe now. Check whether the new pgprot is the same:
401 old_prot
= new_prot
= pte_pgprot(old_pte
);
403 pgprot_val(new_prot
) &= ~pgprot_val(cpa
->mask_clr
);
404 pgprot_val(new_prot
) |= pgprot_val(cpa
->mask_set
);
407 * old_pte points to the large page base address. So we need
408 * to add the offset of the virtual address:
410 pfn
= pte_pfn(old_pte
) + ((address
& (psize
- 1)) >> PAGE_SHIFT
);
413 new_prot
= static_protections(new_prot
, address
, pfn
);
416 * We need to check the full range, whether
417 * static_protection() requires a different pgprot for one of
418 * the pages in the range we try to preserve:
420 addr
= address
+ PAGE_SIZE
;
422 for (i
= 1; i
< cpa
->numpages
; i
++, addr
+= PAGE_SIZE
, pfn
++) {
423 pgprot_t chk_prot
= static_protections(new_prot
, addr
, pfn
);
425 if (pgprot_val(chk_prot
) != pgprot_val(new_prot
))
430 * If there are no changes, return. maxpages has been updated
433 if (pgprot_val(new_prot
) == pgprot_val(old_prot
)) {
439 * We need to change the attributes. Check, whether we can
440 * change the large page in one go. We request a split, when
441 * the address is not aligned and the number of pages is
442 * smaller than the number of pages in the large page. Note
443 * that we limited the number of possible pages already to
444 * the number of pages in the large page.
446 if (address
== (nextpage_addr
- psize
) && cpa
->numpages
== numpages
) {
448 * The address is aligned and the number of pages
449 * covers the full page.
451 new_pte
= pfn_pte(pte_pfn(old_pte
), canon_pgprot(new_prot
));
452 __set_pmd_pte(kpte
, address
, new_pte
);
453 cpa
->flags
|= CPA_FLUSHTLB
;
458 spin_unlock_irqrestore(&pgd_lock
, flags
);
463 static int split_large_page(pte_t
*kpte
, unsigned long address
)
465 unsigned long flags
, pfn
, pfninc
= 1;
466 unsigned int i
, level
;
471 if (!debug_pagealloc
)
472 spin_unlock(&cpa_lock
);
473 base
= alloc_pages(GFP_KERNEL
| __GFP_NOTRACK
, 0);
474 if (!debug_pagealloc
)
475 spin_lock(&cpa_lock
);
479 spin_lock_irqsave(&pgd_lock
, flags
);
481 * Check for races, another CPU might have split this page
484 tmp
= lookup_address(address
, &level
);
488 pbase
= (pte_t
*)page_address(base
);
489 paravirt_alloc_pte(&init_mm
, page_to_pfn(base
));
490 ref_prot
= pte_pgprot(pte_clrhuge(*kpte
));
492 * If we ever want to utilize the PAT bit, we need to
493 * update this function to make sure it's converted from
494 * bit 12 to bit 7 when we cross from the 2MB level to
497 WARN_ON_ONCE(pgprot_val(ref_prot
) & _PAGE_PAT_LARGE
);
500 if (level
== PG_LEVEL_1G
) {
501 pfninc
= PMD_PAGE_SIZE
>> PAGE_SHIFT
;
502 pgprot_val(ref_prot
) |= _PAGE_PSE
;
507 * Get the target pfn from the original entry:
509 pfn
= pte_pfn(*kpte
);
510 for (i
= 0; i
< PTRS_PER_PTE
; i
++, pfn
+= pfninc
)
511 set_pte(&pbase
[i
], pfn_pte(pfn
, ref_prot
));
513 if (address
>= (unsigned long)__va(0) &&
514 address
< (unsigned long)__va(max_low_pfn_mapped
<< PAGE_SHIFT
))
515 split_page_count(level
);
518 if (address
>= (unsigned long)__va(1UL<<32) &&
519 address
< (unsigned long)__va(max_pfn_mapped
<< PAGE_SHIFT
))
520 split_page_count(level
);
524 * Install the new, split up pagetable.
526 * We use the standard kernel pagetable protections for the new
527 * pagetable protections, the actual ptes set above control the
528 * primary protection behavior:
530 __set_pmd_pte(kpte
, address
, mk_pte(base
, __pgprot(_KERNPG_TABLE
)));
533 * Intel Atom errata AAH41 workaround.
535 * The real fix should be in hw or in a microcode update, but
536 * we also probabilistically try to reduce the window of having
537 * a large TLB mixed with 4K TLBs while instruction fetches are
546 * If we dropped out via the lookup_address check under
547 * pgd_lock then stick the page back into the pool:
551 spin_unlock_irqrestore(&pgd_lock
, flags
);
556 static int __cpa_process_fault(struct cpa_data
*cpa
, unsigned long vaddr
,
560 * Ignore all non primary paths.
566 * Ignore the NULL PTE for kernel identity mapping, as it is expected
568 * Also set numpages to '1' indicating that we processed cpa req for
569 * one virtual address page and its pfn. TBD: numpages can be set based
570 * on the initial value and the level returned by lookup_address().
572 if (within(vaddr
, PAGE_OFFSET
,
573 PAGE_OFFSET
+ (max_pfn_mapped
<< PAGE_SHIFT
))) {
575 cpa
->pfn
= __pa(vaddr
) >> PAGE_SHIFT
;
578 WARN(1, KERN_WARNING
"CPA: called for zero pte. "
579 "vaddr = %lx cpa->vaddr = %lx\n", vaddr
,
586 static int __change_page_attr(struct cpa_data
*cpa
, int primary
)
588 unsigned long address
;
591 pte_t
*kpte
, old_pte
;
593 if (cpa
->flags
& CPA_PAGES_ARRAY
)
594 address
= (unsigned long)page_address(cpa
->pages
[cpa
->curpage
]);
595 else if (cpa
->flags
& CPA_ARRAY
)
596 address
= cpa
->vaddr
[cpa
->curpage
];
598 address
= *cpa
->vaddr
;
600 kpte
= lookup_address(address
, &level
);
602 return __cpa_process_fault(cpa
, address
, primary
);
605 if (!pte_val(old_pte
))
606 return __cpa_process_fault(cpa
, address
, primary
);
608 if (level
== PG_LEVEL_4K
) {
610 pgprot_t new_prot
= pte_pgprot(old_pte
);
611 unsigned long pfn
= pte_pfn(old_pte
);
613 pgprot_val(new_prot
) &= ~pgprot_val(cpa
->mask_clr
);
614 pgprot_val(new_prot
) |= pgprot_val(cpa
->mask_set
);
616 new_prot
= static_protections(new_prot
, address
, pfn
);
619 * We need to keep the pfn from the existing PTE,
620 * after all we're only going to change it's attributes
621 * not the memory it points to
623 new_pte
= pfn_pte(pfn
, canon_pgprot(new_prot
));
626 * Do we really change anything ?
628 if (pte_val(old_pte
) != pte_val(new_pte
)) {
629 set_pte_atomic(kpte
, new_pte
);
630 cpa
->flags
|= CPA_FLUSHTLB
;
637 * Check, whether we can keep the large page intact
638 * and just change the pte:
640 do_split
= try_preserve_large_page(kpte
, address
, cpa
);
642 * When the range fits into the existing large page,
643 * return. cp->numpages and cpa->tlbflush have been updated in
650 * We have to split the large page:
652 err
= split_large_page(kpte
, address
);
655 * Do a global flush tlb after splitting the large page
656 * and before we do the actual change page attribute in the PTE.
658 * With out this, we violate the TLB application note, that says
659 * "The TLBs may contain both ordinary and large-page
660 * translations for a 4-KByte range of linear addresses. This
661 * may occur if software modifies the paging structures so that
662 * the page size used for the address range changes. If the two
663 * translations differ with respect to page frame or attributes
664 * (e.g., permissions), processor behavior is undefined and may
665 * be implementation-specific."
667 * We do this global tlb flush inside the cpa_lock, so that we
668 * don't allow any other cpu, with stale tlb entries change the
669 * page attribute in parallel, that also falls into the
670 * just split large page entry.
679 static int __change_page_attr_set_clr(struct cpa_data
*cpa
, int checkalias
);
681 static int cpa_process_alias(struct cpa_data
*cpa
)
683 struct cpa_data alias_cpa
;
685 unsigned long temp_cpa_vaddr
, vaddr
;
687 if (cpa
->pfn
>= max_pfn_mapped
)
691 if (cpa
->pfn
>= max_low_pfn_mapped
&& cpa
->pfn
< (1UL<<(32-PAGE_SHIFT
)))
695 * No need to redo, when the primary call touched the direct
698 if (cpa
->flags
& CPA_PAGES_ARRAY
)
699 vaddr
= (unsigned long)page_address(cpa
->pages
[cpa
->curpage
]);
700 else if (cpa
->flags
& CPA_ARRAY
)
701 vaddr
= cpa
->vaddr
[cpa
->curpage
];
705 if (!(within(vaddr
, PAGE_OFFSET
,
706 PAGE_OFFSET
+ (max_pfn_mapped
<< PAGE_SHIFT
)))) {
709 temp_cpa_vaddr
= (unsigned long) __va(cpa
->pfn
<< PAGE_SHIFT
);
710 alias_cpa
.vaddr
= &temp_cpa_vaddr
;
711 alias_cpa
.flags
&= ~(CPA_PAGES_ARRAY
| CPA_ARRAY
);
714 ret
= __change_page_attr_set_clr(&alias_cpa
, 0);
721 * No need to redo, when the primary call touched the high
724 if (within(vaddr
, (unsigned long) _text
, _brk_end
))
728 * If the physical address is inside the kernel map, we need
729 * to touch the high mapped kernel as well:
731 if (!within(cpa
->pfn
, highmap_start_pfn(), highmap_end_pfn()))
735 temp_cpa_vaddr
= (cpa
->pfn
<< PAGE_SHIFT
) + __START_KERNEL_map
- phys_base
;
736 alias_cpa
.vaddr
= &temp_cpa_vaddr
;
737 alias_cpa
.flags
&= ~(CPA_PAGES_ARRAY
| CPA_ARRAY
);
740 * The high mapping range is imprecise, so ignore the return value.
742 __change_page_attr_set_clr(&alias_cpa
, 0);
747 static int __change_page_attr_set_clr(struct cpa_data
*cpa
, int checkalias
)
749 int ret
, numpages
= cpa
->numpages
;
753 * Store the remaining nr of pages for the large page
754 * preservation check.
756 cpa
->numpages
= numpages
;
757 /* for array changes, we can't use large page */
758 if (cpa
->flags
& (CPA_ARRAY
| CPA_PAGES_ARRAY
))
761 if (!debug_pagealloc
)
762 spin_lock(&cpa_lock
);
763 ret
= __change_page_attr(cpa
, checkalias
);
764 if (!debug_pagealloc
)
765 spin_unlock(&cpa_lock
);
770 ret
= cpa_process_alias(cpa
);
776 * Adjust the number of pages with the result of the
777 * CPA operation. Either a large page has been
778 * preserved or a single page update happened.
780 BUG_ON(cpa
->numpages
> numpages
);
781 numpages
-= cpa
->numpages
;
782 if (cpa
->flags
& (CPA_PAGES_ARRAY
| CPA_ARRAY
))
785 *cpa
->vaddr
+= cpa
->numpages
* PAGE_SIZE
;
791 static inline int cache_attr(pgprot_t attr
)
793 return pgprot_val(attr
) &
794 (_PAGE_PAT
| _PAGE_PAT_LARGE
| _PAGE_PWT
| _PAGE_PCD
);
797 static int change_page_attr_set_clr(unsigned long *addr
, int numpages
,
798 pgprot_t mask_set
, pgprot_t mask_clr
,
799 int force_split
, int in_flag
,
803 int ret
, cache
, checkalias
;
806 * Check, if we are requested to change a not supported
809 mask_set
= canon_pgprot(mask_set
);
810 mask_clr
= canon_pgprot(mask_clr
);
811 if (!pgprot_val(mask_set
) && !pgprot_val(mask_clr
) && !force_split
)
814 /* Ensure we are PAGE_SIZE aligned */
815 if (in_flag
& CPA_ARRAY
) {
817 for (i
= 0; i
< numpages
; i
++) {
818 if (addr
[i
] & ~PAGE_MASK
) {
819 addr
[i
] &= PAGE_MASK
;
823 } else if (!(in_flag
& CPA_PAGES_ARRAY
)) {
825 * in_flag of CPA_PAGES_ARRAY implies it is aligned.
826 * No need to cehck in that case
828 if (*addr
& ~PAGE_MASK
) {
831 * People should not be passing in unaligned addresses:
837 /* Must avoid aliasing mappings in the highmem code */
844 cpa
.numpages
= numpages
;
845 cpa
.mask_set
= mask_set
;
846 cpa
.mask_clr
= mask_clr
;
849 cpa
.force_split
= force_split
;
851 if (in_flag
& (CPA_ARRAY
| CPA_PAGES_ARRAY
))
852 cpa
.flags
|= in_flag
;
854 /* No alias checking for _NX bit modifications */
855 checkalias
= (pgprot_val(mask_set
) | pgprot_val(mask_clr
)) != _PAGE_NX
;
857 ret
= __change_page_attr_set_clr(&cpa
, checkalias
);
860 * Check whether we really changed something:
862 if (!(cpa
.flags
& CPA_FLUSHTLB
))
866 * No need to flush, when we did not set any of the caching
869 cache
= cache_attr(mask_set
);
872 * On success we use clflush, when the CPU supports it to
873 * avoid the wbindv. If the CPU does not support it and in the
874 * error case we fall back to cpa_flush_all (which uses
877 if (!ret
&& cpu_has_clflush
) {
878 if (cpa
.flags
& (CPA_PAGES_ARRAY
| CPA_ARRAY
)) {
879 cpa_flush_array(addr
, numpages
, cache
,
882 cpa_flush_range(*addr
, numpages
, cache
);
884 cpa_flush_all(cache
);
890 static inline int change_page_attr_set(unsigned long *addr
, int numpages
,
891 pgprot_t mask
, int array
)
893 return change_page_attr_set_clr(addr
, numpages
, mask
, __pgprot(0), 0,
894 (array
? CPA_ARRAY
: 0), NULL
);
897 static inline int change_page_attr_clear(unsigned long *addr
, int numpages
,
898 pgprot_t mask
, int array
)
900 return change_page_attr_set_clr(addr
, numpages
, __pgprot(0), mask
, 0,
901 (array
? CPA_ARRAY
: 0), NULL
);
904 static inline int cpa_set_pages_array(struct page
**pages
, int numpages
,
907 return change_page_attr_set_clr(NULL
, numpages
, mask
, __pgprot(0), 0,
908 CPA_PAGES_ARRAY
, pages
);
911 static inline int cpa_clear_pages_array(struct page
**pages
, int numpages
,
914 return change_page_attr_set_clr(NULL
, numpages
, __pgprot(0), mask
, 0,
915 CPA_PAGES_ARRAY
, pages
);
918 int _set_memory_uc(unsigned long addr
, int numpages
)
921 * for now UC MINUS. see comments in ioremap_nocache()
923 return change_page_attr_set(&addr
, numpages
,
924 __pgprot(_PAGE_CACHE_UC_MINUS
), 0);
927 int set_memory_uc(unsigned long addr
, int numpages
)
932 * for now UC MINUS. see comments in ioremap_nocache()
934 ret
= reserve_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
,
935 _PAGE_CACHE_UC_MINUS
, NULL
);
939 ret
= _set_memory_uc(addr
, numpages
);
946 free_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
);
950 EXPORT_SYMBOL(set_memory_uc
);
952 int set_memory_array_uc(unsigned long *addr
, int addrinarray
)
958 * for now UC MINUS. see comments in ioremap_nocache()
960 for (i
= 0; i
< addrinarray
; i
++) {
961 ret
= reserve_memtype(__pa(addr
[i
]), __pa(addr
[i
]) + PAGE_SIZE
,
962 _PAGE_CACHE_UC_MINUS
, NULL
);
967 ret
= change_page_attr_set(addr
, addrinarray
,
968 __pgprot(_PAGE_CACHE_UC_MINUS
), 1);
975 for (j
= 0; j
< i
; j
++)
976 free_memtype(__pa(addr
[j
]), __pa(addr
[j
]) + PAGE_SIZE
);
980 EXPORT_SYMBOL(set_memory_array_uc
);
982 int _set_memory_wc(unsigned long addr
, int numpages
)
985 ret
= change_page_attr_set(&addr
, numpages
,
986 __pgprot(_PAGE_CACHE_UC_MINUS
), 0);
989 ret
= change_page_attr_set(&addr
, numpages
,
990 __pgprot(_PAGE_CACHE_WC
), 0);
995 int set_memory_wc(unsigned long addr
, int numpages
)
1000 return set_memory_uc(addr
, numpages
);
1002 ret
= reserve_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
,
1003 _PAGE_CACHE_WC
, NULL
);
1007 ret
= _set_memory_wc(addr
, numpages
);
1014 free_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
);
1018 EXPORT_SYMBOL(set_memory_wc
);
1020 int _set_memory_wb(unsigned long addr
, int numpages
)
1022 return change_page_attr_clear(&addr
, numpages
,
1023 __pgprot(_PAGE_CACHE_MASK
), 0);
1026 int set_memory_wb(unsigned long addr
, int numpages
)
1030 ret
= _set_memory_wb(addr
, numpages
);
1034 free_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
);
1037 EXPORT_SYMBOL(set_memory_wb
);
1039 int set_memory_array_wb(unsigned long *addr
, int addrinarray
)
1044 ret
= change_page_attr_clear(addr
, addrinarray
,
1045 __pgprot(_PAGE_CACHE_MASK
), 1);
1049 for (i
= 0; i
< addrinarray
; i
++)
1050 free_memtype(__pa(addr
[i
]), __pa(addr
[i
]) + PAGE_SIZE
);
1054 EXPORT_SYMBOL(set_memory_array_wb
);
1056 int set_memory_x(unsigned long addr
, int numpages
)
1058 return change_page_attr_clear(&addr
, numpages
, __pgprot(_PAGE_NX
), 0);
1060 EXPORT_SYMBOL(set_memory_x
);
1062 int set_memory_nx(unsigned long addr
, int numpages
)
1064 return change_page_attr_set(&addr
, numpages
, __pgprot(_PAGE_NX
), 0);
1066 EXPORT_SYMBOL(set_memory_nx
);
1068 int set_memory_ro(unsigned long addr
, int numpages
)
1070 return change_page_attr_clear(&addr
, numpages
, __pgprot(_PAGE_RW
), 0);
1072 EXPORT_SYMBOL_GPL(set_memory_ro
);
1074 int set_memory_rw(unsigned long addr
, int numpages
)
1076 return change_page_attr_set(&addr
, numpages
, __pgprot(_PAGE_RW
), 0);
1078 EXPORT_SYMBOL_GPL(set_memory_rw
);
1080 int set_memory_np(unsigned long addr
, int numpages
)
1082 return change_page_attr_clear(&addr
, numpages
, __pgprot(_PAGE_PRESENT
), 0);
1085 int set_memory_4k(unsigned long addr
, int numpages
)
1087 return change_page_attr_set_clr(&addr
, numpages
, __pgprot(0),
1088 __pgprot(0), 1, 0, NULL
);
1091 int set_pages_uc(struct page
*page
, int numpages
)
1093 unsigned long addr
= (unsigned long)page_address(page
);
1095 return set_memory_uc(addr
, numpages
);
1097 EXPORT_SYMBOL(set_pages_uc
);
1099 int set_pages_array_uc(struct page
**pages
, int addrinarray
)
1101 unsigned long start
;
1106 for (i
= 0; i
< addrinarray
; i
++) {
1107 start
= (unsigned long)page_address(pages
[i
]);
1108 end
= start
+ PAGE_SIZE
;
1109 if (reserve_memtype(start
, end
, _PAGE_CACHE_UC_MINUS
, NULL
))
1113 if (cpa_set_pages_array(pages
, addrinarray
,
1114 __pgprot(_PAGE_CACHE_UC_MINUS
)) == 0) {
1115 return 0; /* Success */
1119 for (i
= 0; i
< free_idx
; i
++) {
1120 start
= (unsigned long)page_address(pages
[i
]);
1121 end
= start
+ PAGE_SIZE
;
1122 free_memtype(start
, end
);
1126 EXPORT_SYMBOL(set_pages_array_uc
);
1128 int set_pages_wb(struct page
*page
, int numpages
)
1130 unsigned long addr
= (unsigned long)page_address(page
);
1132 return set_memory_wb(addr
, numpages
);
1134 EXPORT_SYMBOL(set_pages_wb
);
1136 int set_pages_array_wb(struct page
**pages
, int addrinarray
)
1139 unsigned long start
;
1143 retval
= cpa_clear_pages_array(pages
, addrinarray
,
1144 __pgprot(_PAGE_CACHE_MASK
));
1148 for (i
= 0; i
< addrinarray
; i
++) {
1149 start
= (unsigned long)page_address(pages
[i
]);
1150 end
= start
+ PAGE_SIZE
;
1151 free_memtype(start
, end
);
1156 EXPORT_SYMBOL(set_pages_array_wb
);
1158 int set_pages_x(struct page
*page
, int numpages
)
1160 unsigned long addr
= (unsigned long)page_address(page
);
1162 return set_memory_x(addr
, numpages
);
1164 EXPORT_SYMBOL(set_pages_x
);
1166 int set_pages_nx(struct page
*page
, int numpages
)
1168 unsigned long addr
= (unsigned long)page_address(page
);
1170 return set_memory_nx(addr
, numpages
);
1172 EXPORT_SYMBOL(set_pages_nx
);
1174 int set_pages_ro(struct page
*page
, int numpages
)
1176 unsigned long addr
= (unsigned long)page_address(page
);
1178 return set_memory_ro(addr
, numpages
);
1181 int set_pages_rw(struct page
*page
, int numpages
)
1183 unsigned long addr
= (unsigned long)page_address(page
);
1185 return set_memory_rw(addr
, numpages
);
1188 #ifdef CONFIG_DEBUG_PAGEALLOC
1190 static int __set_pages_p(struct page
*page
, int numpages
)
1192 unsigned long tempaddr
= (unsigned long) page_address(page
);
1193 struct cpa_data cpa
= { .vaddr
= &tempaddr
,
1194 .numpages
= numpages
,
1195 .mask_set
= __pgprot(_PAGE_PRESENT
| _PAGE_RW
),
1196 .mask_clr
= __pgprot(0),
1200 * No alias checking needed for setting present flag. otherwise,
1201 * we may need to break large pages for 64-bit kernel text
1202 * mappings (this adds to complexity if we want to do this from
1203 * atomic context especially). Let's keep it simple!
1205 return __change_page_attr_set_clr(&cpa
, 0);
1208 static int __set_pages_np(struct page
*page
, int numpages
)
1210 unsigned long tempaddr
= (unsigned long) page_address(page
);
1211 struct cpa_data cpa
= { .vaddr
= &tempaddr
,
1212 .numpages
= numpages
,
1213 .mask_set
= __pgprot(0),
1214 .mask_clr
= __pgprot(_PAGE_PRESENT
| _PAGE_RW
),
1218 * No alias checking needed for setting not present flag. otherwise,
1219 * we may need to break large pages for 64-bit kernel text
1220 * mappings (this adds to complexity if we want to do this from
1221 * atomic context especially). Let's keep it simple!
1223 return __change_page_attr_set_clr(&cpa
, 0);
1226 void kernel_map_pages(struct page
*page
, int numpages
, int enable
)
1228 if (PageHighMem(page
))
1231 debug_check_no_locks_freed(page_address(page
),
1232 numpages
* PAGE_SIZE
);
1236 * If page allocator is not up yet then do not call c_p_a():
1238 if (!debug_pagealloc_enabled
)
1242 * The return value is ignored as the calls cannot fail.
1243 * Large pages for identity mappings are not used at boot time
1244 * and hence no memory allocations during large page split.
1247 __set_pages_p(page
, numpages
);
1249 __set_pages_np(page
, numpages
);
1252 * We should perform an IPI and flush all tlbs,
1253 * but that can deadlock->flush only current cpu:
1258 #ifdef CONFIG_HIBERNATION
1260 bool kernel_page_present(struct page
*page
)
1265 if (PageHighMem(page
))
1268 pte
= lookup_address((unsigned long)page_address(page
), &level
);
1269 return (pte_val(*pte
) & _PAGE_PRESENT
);
1272 #endif /* CONFIG_HIBERNATION */
1274 #endif /* CONFIG_DEBUG_PAGEALLOC */
1277 * The testcases use internal knowledge of the implementation that shouldn't
1278 * be exposed to the rest of the kernel. Include these directly here.
1280 #ifdef CONFIG_CPA_DEBUG
1281 #include "pageattr-test.c"