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/sched.h>
9 #include <linux/interrupt.h>
10 #include <linux/seq_file.h>
11 #include <linux/debugfs.h>
12 #include <linux/pfn.h>
13 #include <linux/percpu.h>
14 #include <linux/gfp.h>
15 #include <linux/pci.h>
16 #include <linux/vmalloc.h>
19 #include <asm/processor.h>
20 #include <asm/tlbflush.h>
21 #include <asm/sections.h>
22 #include <asm/setup.h>
23 #include <asm/uaccess.h>
24 #include <asm/pgalloc.h>
25 #include <asm/proto.h>
29 * The current flushing context - we pass it instead of 5 arguments:
39 unsigned force_split
: 1;
45 * Serialize cpa() (for !DEBUG_PAGEALLOC which uses large identity mappings)
46 * using cpa_lock. So that we don't allow any other cpu, with stale large tlb
47 * entries change the page attribute in parallel to some other cpu
48 * splitting a large page entry along with changing the attribute.
50 static DEFINE_SPINLOCK(cpa_lock
);
52 #define CPA_FLUSHTLB 1
54 #define CPA_PAGES_ARRAY 4
57 static unsigned long direct_pages_count
[PG_LEVEL_NUM
];
59 void update_page_count(int level
, unsigned long pages
)
61 /* Protect against CPA */
63 direct_pages_count
[level
] += pages
;
64 spin_unlock(&pgd_lock
);
67 static void split_page_count(int level
)
69 direct_pages_count
[level
]--;
70 direct_pages_count
[level
- 1] += PTRS_PER_PTE
;
73 void arch_report_meminfo(struct seq_file
*m
)
75 seq_printf(m
, "DirectMap4k: %8lu kB\n",
76 direct_pages_count
[PG_LEVEL_4K
] << 2);
77 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
78 seq_printf(m
, "DirectMap2M: %8lu kB\n",
79 direct_pages_count
[PG_LEVEL_2M
] << 11);
81 seq_printf(m
, "DirectMap4M: %8lu kB\n",
82 direct_pages_count
[PG_LEVEL_2M
] << 12);
85 seq_printf(m
, "DirectMap1G: %8lu kB\n",
86 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_symbol(_text
) >> PAGE_SHIFT
;
99 static inline unsigned long highmap_end_pfn(void)
101 return __pa_symbol(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 * @vaddr: virtual start address
125 * @size: number of bytes to flush
127 * clflushopt is an unordered instruction which needs fencing with mfence or
128 * sfence to avoid ordering issues.
130 void clflush_cache_range(void *vaddr
, unsigned int size
)
132 unsigned long clflush_mask
= boot_cpu_data
.x86_clflush_size
- 1;
133 void *vend
= vaddr
+ size
;
138 for (p
= (void *)((unsigned long)vaddr
& ~clflush_mask
);
139 p
< vend
; p
+= boot_cpu_data
.x86_clflush_size
)
144 EXPORT_SYMBOL_GPL(clflush_cache_range
);
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 #ifdef CONFIG_PCI_BIOS
261 if (pcibios_enabled
&& within(pfn
, BIOS_BEGIN
>> PAGE_SHIFT
, BIOS_END
>> PAGE_SHIFT
))
262 pgprot_val(forbidden
) |= _PAGE_NX
;
266 * The kernel text needs to be executable for obvious reasons
267 * Does not cover __inittext since that is gone later on. On
268 * 64bit we do not enforce !NX on the low mapping
270 if (within(address
, (unsigned long)_text
, (unsigned long)_etext
))
271 pgprot_val(forbidden
) |= _PAGE_NX
;
274 * The .rodata section needs to be read-only. Using the pfn
275 * catches all aliases.
277 if (within(pfn
, __pa_symbol(__start_rodata
) >> PAGE_SHIFT
,
278 __pa_symbol(__end_rodata
) >> PAGE_SHIFT
))
279 pgprot_val(forbidden
) |= _PAGE_RW
;
281 #if defined(CONFIG_X86_64) && defined(CONFIG_DEBUG_RODATA)
283 * Once the kernel maps the text as RO (kernel_set_to_readonly is set),
284 * kernel text mappings for the large page aligned text, rodata sections
285 * will be always read-only. For the kernel identity mappings covering
286 * the holes caused by this alignment can be anything that user asks.
288 * This will preserve the large page mappings for kernel text/data
291 if (kernel_set_to_readonly
&&
292 within(address
, (unsigned long)_text
,
293 (unsigned long)__end_rodata_hpage_align
)) {
297 * Don't enforce the !RW mapping for the kernel text mapping,
298 * if the current mapping is already using small page mapping.
299 * No need to work hard to preserve large page mappings in this
302 * This also fixes the Linux Xen paravirt guest boot failure
303 * (because of unexpected read-only mappings for kernel identity
304 * mappings). In this paravirt guest case, the kernel text
305 * mapping and the kernel identity mapping share the same
306 * page-table pages. Thus we can't really use different
307 * protections for the kernel text and identity mappings. Also,
308 * these shared mappings are made of small page mappings.
309 * Thus this don't enforce !RW mapping for small page kernel
310 * text mapping logic will help Linux Xen parvirt guest boot
313 if (lookup_address(address
, &level
) && (level
!= PG_LEVEL_4K
))
314 pgprot_val(forbidden
) |= _PAGE_RW
;
318 prot
= __pgprot(pgprot_val(prot
) & ~pgprot_val(forbidden
));
324 * Lookup the page table entry for a virtual address in a specific pgd.
325 * Return a pointer to the entry and the level of the mapping.
327 pte_t
*lookup_address_in_pgd(pgd_t
*pgd
, unsigned long address
,
333 *level
= PG_LEVEL_NONE
;
338 pud
= pud_offset(pgd
, address
);
342 *level
= PG_LEVEL_1G
;
343 if (pud_large(*pud
) || !pud_present(*pud
))
346 pmd
= pmd_offset(pud
, address
);
350 *level
= PG_LEVEL_2M
;
351 if (pmd_large(*pmd
) || !pmd_present(*pmd
))
354 *level
= PG_LEVEL_4K
;
356 return pte_offset_kernel(pmd
, address
);
360 * Lookup the page table entry for a virtual address. Return a pointer
361 * to the entry and the level of the mapping.
363 * Note: We return pud and pmd either when the entry is marked large
364 * or when the present bit is not set. Otherwise we would return a
365 * pointer to a nonexisting mapping.
367 pte_t
*lookup_address(unsigned long address
, unsigned int *level
)
369 return lookup_address_in_pgd(pgd_offset_k(address
), address
, level
);
371 EXPORT_SYMBOL_GPL(lookup_address
);
373 static pte_t
*_lookup_address_cpa(struct cpa_data
*cpa
, unsigned long address
,
377 return lookup_address_in_pgd(cpa
->pgd
+ pgd_index(address
),
380 return lookup_address(address
, level
);
384 * Lookup the PMD entry for a virtual address. Return a pointer to the entry
385 * or NULL if not present.
387 pmd_t
*lookup_pmd_address(unsigned long address
)
392 pgd
= pgd_offset_k(address
);
396 pud
= pud_offset(pgd
, address
);
397 if (pud_none(*pud
) || pud_large(*pud
) || !pud_present(*pud
))
400 return pmd_offset(pud
, address
);
404 * This is necessary because __pa() does not work on some
405 * kinds of memory, like vmalloc() or the alloc_remap()
406 * areas on 32-bit NUMA systems. The percpu areas can
407 * end up in this kind of memory, for instance.
409 * This could be optimized, but it is only intended to be
410 * used at inititalization time, and keeping it
411 * unoptimized should increase the testing coverage for
412 * the more obscure platforms.
414 phys_addr_t
slow_virt_to_phys(void *__virt_addr
)
416 unsigned long virt_addr
= (unsigned long)__virt_addr
;
417 unsigned long phys_addr
, offset
;
421 pte
= lookup_address(virt_addr
, &level
);
426 phys_addr
= pud_pfn(*(pud_t
*)pte
) << PAGE_SHIFT
;
427 offset
= virt_addr
& ~PUD_PAGE_MASK
;
430 phys_addr
= pmd_pfn(*(pmd_t
*)pte
) << PAGE_SHIFT
;
431 offset
= virt_addr
& ~PMD_PAGE_MASK
;
434 phys_addr
= pte_pfn(*pte
) << PAGE_SHIFT
;
435 offset
= virt_addr
& ~PAGE_MASK
;
438 return (phys_addr_t
)(phys_addr
| offset
);
440 EXPORT_SYMBOL_GPL(slow_virt_to_phys
);
443 * Set the new pmd in all the pgds we know about:
445 static void __set_pmd_pte(pte_t
*kpte
, unsigned long address
, pte_t pte
)
448 set_pte_atomic(kpte
, pte
);
450 if (!SHARED_KERNEL_PMD
) {
453 list_for_each_entry(page
, &pgd_list
, lru
) {
458 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
459 pud
= pud_offset(pgd
, address
);
460 pmd
= pmd_offset(pud
, address
);
461 set_pte_atomic((pte_t
*)pmd
, pte
);
468 try_preserve_large_page(pte_t
*kpte
, unsigned long address
,
469 struct cpa_data
*cpa
)
471 unsigned long nextpage_addr
, numpages
, pmask
, psize
, addr
, pfn
, old_pfn
;
472 pte_t new_pte
, old_pte
, *tmp
;
473 pgprot_t old_prot
, new_prot
, req_prot
;
477 if (cpa
->force_split
)
480 spin_lock(&pgd_lock
);
482 * Check for races, another CPU might have split this page
485 tmp
= _lookup_address_cpa(cpa
, address
, &level
);
491 old_prot
= pmd_pgprot(*(pmd_t
*)kpte
);
492 old_pfn
= pmd_pfn(*(pmd_t
*)kpte
);
495 old_prot
= pud_pgprot(*(pud_t
*)kpte
);
496 old_pfn
= pud_pfn(*(pud_t
*)kpte
);
503 psize
= page_level_size(level
);
504 pmask
= page_level_mask(level
);
507 * Calculate the number of pages, which fit into this large
508 * page starting at address:
510 nextpage_addr
= (address
+ psize
) & pmask
;
511 numpages
= (nextpage_addr
- address
) >> PAGE_SHIFT
;
512 if (numpages
< cpa
->numpages
)
513 cpa
->numpages
= numpages
;
516 * We are safe now. Check whether the new pgprot is the same:
517 * Convert protection attributes to 4k-format, as cpa->mask* are set
521 req_prot
= pgprot_large_2_4k(old_prot
);
523 pgprot_val(req_prot
) &= ~pgprot_val(cpa
->mask_clr
);
524 pgprot_val(req_prot
) |= pgprot_val(cpa
->mask_set
);
527 * req_prot is in format of 4k pages. It must be converted to large
528 * page format: the caching mode includes the PAT bit located at
529 * different bit positions in the two formats.
531 req_prot
= pgprot_4k_2_large(req_prot
);
534 * Set the PSE and GLOBAL flags only if the PRESENT flag is
535 * set otherwise pmd_present/pmd_huge will return true even on
536 * a non present pmd. The canon_pgprot will clear _PAGE_GLOBAL
537 * for the ancient hardware that doesn't support it.
539 if (pgprot_val(req_prot
) & _PAGE_PRESENT
)
540 pgprot_val(req_prot
) |= _PAGE_PSE
| _PAGE_GLOBAL
;
542 pgprot_val(req_prot
) &= ~(_PAGE_PSE
| _PAGE_GLOBAL
);
544 req_prot
= canon_pgprot(req_prot
);
547 * old_pfn points to the large page base pfn. So we need
548 * to add the offset of the virtual address:
550 pfn
= old_pfn
+ ((address
& (psize
- 1)) >> PAGE_SHIFT
);
553 new_prot
= static_protections(req_prot
, address
, pfn
);
556 * We need to check the full range, whether
557 * static_protection() requires a different pgprot for one of
558 * the pages in the range we try to preserve:
560 addr
= address
& pmask
;
562 for (i
= 0; i
< (psize
>> PAGE_SHIFT
); i
++, addr
+= PAGE_SIZE
, pfn
++) {
563 pgprot_t chk_prot
= static_protections(req_prot
, addr
, pfn
);
565 if (pgprot_val(chk_prot
) != pgprot_val(new_prot
))
570 * If there are no changes, return. maxpages has been updated
573 if (pgprot_val(new_prot
) == pgprot_val(old_prot
)) {
579 * We need to change the attributes. Check, whether we can
580 * change the large page in one go. We request a split, when
581 * the address is not aligned and the number of pages is
582 * smaller than the number of pages in the large page. Note
583 * that we limited the number of possible pages already to
584 * the number of pages in the large page.
586 if (address
== (address
& pmask
) && cpa
->numpages
== (psize
>> PAGE_SHIFT
)) {
588 * The address is aligned and the number of pages
589 * covers the full page.
591 new_pte
= pfn_pte(old_pfn
, new_prot
);
592 __set_pmd_pte(kpte
, address
, new_pte
);
593 cpa
->flags
|= CPA_FLUSHTLB
;
598 spin_unlock(&pgd_lock
);
604 __split_large_page(struct cpa_data
*cpa
, pte_t
*kpte
, unsigned long address
,
607 pte_t
*pbase
= (pte_t
*)page_address(base
);
608 unsigned long ref_pfn
, pfn
, pfninc
= 1;
609 unsigned int i
, level
;
613 spin_lock(&pgd_lock
);
615 * Check for races, another CPU might have split this page
618 tmp
= _lookup_address_cpa(cpa
, address
, &level
);
620 spin_unlock(&pgd_lock
);
624 paravirt_alloc_pte(&init_mm
, page_to_pfn(base
));
628 ref_prot
= pmd_pgprot(*(pmd_t
*)kpte
);
629 /* clear PSE and promote PAT bit to correct position */
630 ref_prot
= pgprot_large_2_4k(ref_prot
);
631 ref_pfn
= pmd_pfn(*(pmd_t
*)kpte
);
635 ref_prot
= pud_pgprot(*(pud_t
*)kpte
);
636 ref_pfn
= pud_pfn(*(pud_t
*)kpte
);
637 pfninc
= PMD_PAGE_SIZE
>> PAGE_SHIFT
;
640 * Clear the PSE flags if the PRESENT flag is not set
641 * otherwise pmd_present/pmd_huge will return true
642 * even on a non present pmd.
644 if (!(pgprot_val(ref_prot
) & _PAGE_PRESENT
))
645 pgprot_val(ref_prot
) &= ~_PAGE_PSE
;
649 spin_unlock(&pgd_lock
);
654 * Set the GLOBAL flags only if the PRESENT flag is set
655 * otherwise pmd/pte_present will return true even on a non
656 * present pmd/pte. The canon_pgprot will clear _PAGE_GLOBAL
657 * for the ancient hardware that doesn't support it.
659 if (pgprot_val(ref_prot
) & _PAGE_PRESENT
)
660 pgprot_val(ref_prot
) |= _PAGE_GLOBAL
;
662 pgprot_val(ref_prot
) &= ~_PAGE_GLOBAL
;
665 * Get the target pfn from the original entry:
668 for (i
= 0; i
< PTRS_PER_PTE
; i
++, pfn
+= pfninc
)
669 set_pte(&pbase
[i
], pfn_pte(pfn
, canon_pgprot(ref_prot
)));
671 if (virt_addr_valid(address
)) {
672 unsigned long pfn
= PFN_DOWN(__pa(address
));
674 if (pfn_range_is_mapped(pfn
, pfn
+ 1))
675 split_page_count(level
);
679 * Install the new, split up pagetable.
681 * We use the standard kernel pagetable protections for the new
682 * pagetable protections, the actual ptes set above control the
683 * primary protection behavior:
685 __set_pmd_pte(kpte
, address
, mk_pte(base
, __pgprot(_KERNPG_TABLE
)));
688 * Intel Atom errata AAH41 workaround.
690 * The real fix should be in hw or in a microcode update, but
691 * we also probabilistically try to reduce the window of having
692 * a large TLB mixed with 4K TLBs while instruction fetches are
696 spin_unlock(&pgd_lock
);
701 static int split_large_page(struct cpa_data
*cpa
, pte_t
*kpte
,
702 unsigned long address
)
706 if (!debug_pagealloc
)
707 spin_unlock(&cpa_lock
);
708 base
= alloc_pages(GFP_KERNEL
| __GFP_NOTRACK
, 0);
709 if (!debug_pagealloc
)
710 spin_lock(&cpa_lock
);
714 if (__split_large_page(cpa
, kpte
, address
, base
))
720 static bool try_to_free_pte_page(pte_t
*pte
)
724 for (i
= 0; i
< PTRS_PER_PTE
; i
++)
725 if (!pte_none(pte
[i
]))
728 free_page((unsigned long)pte
);
732 static bool try_to_free_pmd_page(pmd_t
*pmd
)
736 for (i
= 0; i
< PTRS_PER_PMD
; i
++)
737 if (!pmd_none(pmd
[i
]))
740 free_page((unsigned long)pmd
);
744 static bool try_to_free_pud_page(pud_t
*pud
)
748 for (i
= 0; i
< PTRS_PER_PUD
; i
++)
749 if (!pud_none(pud
[i
]))
752 free_page((unsigned long)pud
);
756 static bool unmap_pte_range(pmd_t
*pmd
, unsigned long start
, unsigned long end
)
758 pte_t
*pte
= pte_offset_kernel(pmd
, start
);
760 while (start
< end
) {
761 set_pte(pte
, __pte(0));
767 if (try_to_free_pte_page((pte_t
*)pmd_page_vaddr(*pmd
))) {
774 static void __unmap_pmd_range(pud_t
*pud
, pmd_t
*pmd
,
775 unsigned long start
, unsigned long end
)
777 if (unmap_pte_range(pmd
, start
, end
))
778 if (try_to_free_pmd_page((pmd_t
*)pud_page_vaddr(*pud
)))
782 static void unmap_pmd_range(pud_t
*pud
, unsigned long start
, unsigned long end
)
784 pmd_t
*pmd
= pmd_offset(pud
, start
);
787 * Not on a 2MB page boundary?
789 if (start
& (PMD_SIZE
- 1)) {
790 unsigned long next_page
= (start
+ PMD_SIZE
) & PMD_MASK
;
791 unsigned long pre_end
= min_t(unsigned long, end
, next_page
);
793 __unmap_pmd_range(pud
, pmd
, start
, pre_end
);
800 * Try to unmap in 2M chunks.
802 while (end
- start
>= PMD_SIZE
) {
806 __unmap_pmd_range(pud
, pmd
, start
, start
+ PMD_SIZE
);
816 return __unmap_pmd_range(pud
, pmd
, start
, end
);
819 * Try again to free the PMD page if haven't succeeded above.
822 if (try_to_free_pmd_page((pmd_t
*)pud_page_vaddr(*pud
)))
826 static void unmap_pud_range(pgd_t
*pgd
, unsigned long start
, unsigned long end
)
828 pud_t
*pud
= pud_offset(pgd
, start
);
831 * Not on a GB page boundary?
833 if (start
& (PUD_SIZE
- 1)) {
834 unsigned long next_page
= (start
+ PUD_SIZE
) & PUD_MASK
;
835 unsigned long pre_end
= min_t(unsigned long, end
, next_page
);
837 unmap_pmd_range(pud
, start
, pre_end
);
844 * Try to unmap in 1G chunks?
846 while (end
- start
>= PUD_SIZE
) {
851 unmap_pmd_range(pud
, start
, start
+ PUD_SIZE
);
861 unmap_pmd_range(pud
, start
, end
);
864 * No need to try to free the PUD page because we'll free it in
865 * populate_pgd's error path
869 static void unmap_pgd_range(pgd_t
*root
, unsigned long addr
, unsigned long end
)
871 pgd_t
*pgd_entry
= root
+ pgd_index(addr
);
873 unmap_pud_range(pgd_entry
, addr
, end
);
875 if (try_to_free_pud_page((pud_t
*)pgd_page_vaddr(*pgd_entry
)))
876 pgd_clear(pgd_entry
);
879 static int alloc_pte_page(pmd_t
*pmd
)
881 pte_t
*pte
= (pte_t
*)get_zeroed_page(GFP_KERNEL
| __GFP_NOTRACK
);
885 set_pmd(pmd
, __pmd(__pa(pte
) | _KERNPG_TABLE
));
889 static int alloc_pmd_page(pud_t
*pud
)
891 pmd_t
*pmd
= (pmd_t
*)get_zeroed_page(GFP_KERNEL
| __GFP_NOTRACK
);
895 set_pud(pud
, __pud(__pa(pmd
) | _KERNPG_TABLE
));
899 static void populate_pte(struct cpa_data
*cpa
,
900 unsigned long start
, unsigned long end
,
901 unsigned num_pages
, pmd_t
*pmd
, pgprot_t pgprot
)
905 pte
= pte_offset_kernel(pmd
, start
);
907 while (num_pages
-- && start
< end
) {
909 /* deal with the NX bit */
910 if (!(pgprot_val(pgprot
) & _PAGE_NX
))
911 cpa
->pfn
&= ~_PAGE_NX
;
913 set_pte(pte
, pfn_pte(cpa
->pfn
>> PAGE_SHIFT
, pgprot
));
916 cpa
->pfn
+= PAGE_SIZE
;
921 static int populate_pmd(struct cpa_data
*cpa
,
922 unsigned long start
, unsigned long end
,
923 unsigned num_pages
, pud_t
*pud
, pgprot_t pgprot
)
925 unsigned int cur_pages
= 0;
930 * Not on a 2M boundary?
932 if (start
& (PMD_SIZE
- 1)) {
933 unsigned long pre_end
= start
+ (num_pages
<< PAGE_SHIFT
);
934 unsigned long next_page
= (start
+ PMD_SIZE
) & PMD_MASK
;
936 pre_end
= min_t(unsigned long, pre_end
, next_page
);
937 cur_pages
= (pre_end
- start
) >> PAGE_SHIFT
;
938 cur_pages
= min_t(unsigned int, num_pages
, cur_pages
);
943 pmd
= pmd_offset(pud
, start
);
945 if (alloc_pte_page(pmd
))
948 populate_pte(cpa
, start
, pre_end
, cur_pages
, pmd
, pgprot
);
954 * We mapped them all?
956 if (num_pages
== cur_pages
)
959 pmd_pgprot
= pgprot_4k_2_large(pgprot
);
961 while (end
- start
>= PMD_SIZE
) {
964 * We cannot use a 1G page so allocate a PMD page if needed.
967 if (alloc_pmd_page(pud
))
970 pmd
= pmd_offset(pud
, start
);
972 set_pmd(pmd
, __pmd(cpa
->pfn
| _PAGE_PSE
|
973 massage_pgprot(pmd_pgprot
)));
976 cpa
->pfn
+= PMD_SIZE
;
977 cur_pages
+= PMD_SIZE
>> PAGE_SHIFT
;
981 * Map trailing 4K pages.
984 pmd
= pmd_offset(pud
, start
);
986 if (alloc_pte_page(pmd
))
989 populate_pte(cpa
, start
, end
, num_pages
- cur_pages
,
995 static int populate_pud(struct cpa_data
*cpa
, unsigned long start
, pgd_t
*pgd
,
1001 pgprot_t pud_pgprot
;
1003 end
= start
+ (cpa
->numpages
<< PAGE_SHIFT
);
1006 * Not on a Gb page boundary? => map everything up to it with
1009 if (start
& (PUD_SIZE
- 1)) {
1010 unsigned long pre_end
;
1011 unsigned long next_page
= (start
+ PUD_SIZE
) & PUD_MASK
;
1013 pre_end
= min_t(unsigned long, end
, next_page
);
1014 cur_pages
= (pre_end
- start
) >> PAGE_SHIFT
;
1015 cur_pages
= min_t(int, (int)cpa
->numpages
, cur_pages
);
1017 pud
= pud_offset(pgd
, start
);
1023 if (alloc_pmd_page(pud
))
1026 cur_pages
= populate_pmd(cpa
, start
, pre_end
, cur_pages
,
1034 /* We mapped them all? */
1035 if (cpa
->numpages
== cur_pages
)
1038 pud
= pud_offset(pgd
, start
);
1039 pud_pgprot
= pgprot_4k_2_large(pgprot
);
1042 * Map everything starting from the Gb boundary, possibly with 1G pages
1044 while (end
- start
>= PUD_SIZE
) {
1045 set_pud(pud
, __pud(cpa
->pfn
| _PAGE_PSE
|
1046 massage_pgprot(pud_pgprot
)));
1049 cpa
->pfn
+= PUD_SIZE
;
1050 cur_pages
+= PUD_SIZE
>> PAGE_SHIFT
;
1054 /* Map trailing leftover */
1058 pud
= pud_offset(pgd
, start
);
1060 if (alloc_pmd_page(pud
))
1063 tmp
= populate_pmd(cpa
, start
, end
, cpa
->numpages
- cur_pages
,
1074 * Restrictions for kernel page table do not necessarily apply when mapping in
1077 static int populate_pgd(struct cpa_data
*cpa
, unsigned long addr
)
1079 pgprot_t pgprot
= __pgprot(_KERNPG_TABLE
);
1080 pud_t
*pud
= NULL
; /* shut up gcc */
1084 pgd_entry
= cpa
->pgd
+ pgd_index(addr
);
1087 * Allocate a PUD page and hand it down for mapping.
1089 if (pgd_none(*pgd_entry
)) {
1090 pud
= (pud_t
*)get_zeroed_page(GFP_KERNEL
| __GFP_NOTRACK
);
1094 set_pgd(pgd_entry
, __pgd(__pa(pud
) | _KERNPG_TABLE
));
1097 pgprot_val(pgprot
) &= ~pgprot_val(cpa
->mask_clr
);
1098 pgprot_val(pgprot
) |= pgprot_val(cpa
->mask_set
);
1100 ret
= populate_pud(cpa
, addr
, pgd_entry
, pgprot
);
1102 unmap_pgd_range(cpa
->pgd
, addr
,
1103 addr
+ (cpa
->numpages
<< PAGE_SHIFT
));
1107 cpa
->numpages
= ret
;
1111 static int __cpa_process_fault(struct cpa_data
*cpa
, unsigned long vaddr
,
1115 return populate_pgd(cpa
, vaddr
);
1118 * Ignore all non primary paths.
1124 * Ignore the NULL PTE for kernel identity mapping, as it is expected
1126 * Also set numpages to '1' indicating that we processed cpa req for
1127 * one virtual address page and its pfn. TBD: numpages can be set based
1128 * on the initial value and the level returned by lookup_address().
1130 if (within(vaddr
, PAGE_OFFSET
,
1131 PAGE_OFFSET
+ (max_pfn_mapped
<< PAGE_SHIFT
))) {
1133 cpa
->pfn
= __pa(vaddr
) >> PAGE_SHIFT
;
1136 WARN(1, KERN_WARNING
"CPA: called for zero pte. "
1137 "vaddr = %lx cpa->vaddr = %lx\n", vaddr
,
1144 static int __change_page_attr(struct cpa_data
*cpa
, int primary
)
1146 unsigned long address
;
1149 pte_t
*kpte
, old_pte
;
1151 if (cpa
->flags
& CPA_PAGES_ARRAY
) {
1152 struct page
*page
= cpa
->pages
[cpa
->curpage
];
1153 if (unlikely(PageHighMem(page
)))
1155 address
= (unsigned long)page_address(page
);
1156 } else if (cpa
->flags
& CPA_ARRAY
)
1157 address
= cpa
->vaddr
[cpa
->curpage
];
1159 address
= *cpa
->vaddr
;
1161 kpte
= _lookup_address_cpa(cpa
, address
, &level
);
1163 return __cpa_process_fault(cpa
, address
, primary
);
1166 if (!pte_val(old_pte
))
1167 return __cpa_process_fault(cpa
, address
, primary
);
1169 if (level
== PG_LEVEL_4K
) {
1171 pgprot_t new_prot
= pte_pgprot(old_pte
);
1172 unsigned long pfn
= pte_pfn(old_pte
);
1174 pgprot_val(new_prot
) &= ~pgprot_val(cpa
->mask_clr
);
1175 pgprot_val(new_prot
) |= pgprot_val(cpa
->mask_set
);
1177 new_prot
= static_protections(new_prot
, address
, pfn
);
1180 * Set the GLOBAL flags only if the PRESENT flag is
1181 * set otherwise pte_present will return true even on
1182 * a non present pte. The canon_pgprot will clear
1183 * _PAGE_GLOBAL for the ancient hardware that doesn't
1186 if (pgprot_val(new_prot
) & _PAGE_PRESENT
)
1187 pgprot_val(new_prot
) |= _PAGE_GLOBAL
;
1189 pgprot_val(new_prot
) &= ~_PAGE_GLOBAL
;
1192 * We need to keep the pfn from the existing PTE,
1193 * after all we're only going to change it's attributes
1194 * not the memory it points to
1196 new_pte
= pfn_pte(pfn
, canon_pgprot(new_prot
));
1199 * Do we really change anything ?
1201 if (pte_val(old_pte
) != pte_val(new_pte
)) {
1202 set_pte_atomic(kpte
, new_pte
);
1203 cpa
->flags
|= CPA_FLUSHTLB
;
1210 * Check, whether we can keep the large page intact
1211 * and just change the pte:
1213 do_split
= try_preserve_large_page(kpte
, address
, cpa
);
1215 * When the range fits into the existing large page,
1216 * return. cp->numpages and cpa->tlbflush have been updated in
1223 * We have to split the large page:
1225 err
= split_large_page(cpa
, kpte
, address
);
1228 * Do a global flush tlb after splitting the large page
1229 * and before we do the actual change page attribute in the PTE.
1231 * With out this, we violate the TLB application note, that says
1232 * "The TLBs may contain both ordinary and large-page
1233 * translations for a 4-KByte range of linear addresses. This
1234 * may occur if software modifies the paging structures so that
1235 * the page size used for the address range changes. If the two
1236 * translations differ with respect to page frame or attributes
1237 * (e.g., permissions), processor behavior is undefined and may
1238 * be implementation-specific."
1240 * We do this global tlb flush inside the cpa_lock, so that we
1241 * don't allow any other cpu, with stale tlb entries change the
1242 * page attribute in parallel, that also falls into the
1243 * just split large page entry.
1252 static int __change_page_attr_set_clr(struct cpa_data
*cpa
, int checkalias
);
1254 static int cpa_process_alias(struct cpa_data
*cpa
)
1256 struct cpa_data alias_cpa
;
1257 unsigned long laddr
= (unsigned long)__va(cpa
->pfn
<< PAGE_SHIFT
);
1258 unsigned long vaddr
;
1261 if (!pfn_range_is_mapped(cpa
->pfn
, cpa
->pfn
+ 1))
1265 * No need to redo, when the primary call touched the direct
1268 if (cpa
->flags
& CPA_PAGES_ARRAY
) {
1269 struct page
*page
= cpa
->pages
[cpa
->curpage
];
1270 if (unlikely(PageHighMem(page
)))
1272 vaddr
= (unsigned long)page_address(page
);
1273 } else if (cpa
->flags
& CPA_ARRAY
)
1274 vaddr
= cpa
->vaddr
[cpa
->curpage
];
1276 vaddr
= *cpa
->vaddr
;
1278 if (!(within(vaddr
, PAGE_OFFSET
,
1279 PAGE_OFFSET
+ (max_pfn_mapped
<< PAGE_SHIFT
)))) {
1282 alias_cpa
.vaddr
= &laddr
;
1283 alias_cpa
.flags
&= ~(CPA_PAGES_ARRAY
| CPA_ARRAY
);
1285 ret
= __change_page_attr_set_clr(&alias_cpa
, 0);
1290 #ifdef CONFIG_X86_64
1292 * If the primary call didn't touch the high mapping already
1293 * and the physical address is inside the kernel map, we need
1294 * to touch the high mapped kernel as well:
1296 if (!within(vaddr
, (unsigned long)_text
, _brk_end
) &&
1297 within(cpa
->pfn
, highmap_start_pfn(), highmap_end_pfn())) {
1298 unsigned long temp_cpa_vaddr
= (cpa
->pfn
<< PAGE_SHIFT
) +
1299 __START_KERNEL_map
- phys_base
;
1301 alias_cpa
.vaddr
= &temp_cpa_vaddr
;
1302 alias_cpa
.flags
&= ~(CPA_PAGES_ARRAY
| CPA_ARRAY
);
1305 * The high mapping range is imprecise, so ignore the
1308 __change_page_attr_set_clr(&alias_cpa
, 0);
1315 static int __change_page_attr_set_clr(struct cpa_data
*cpa
, int checkalias
)
1317 int ret
, numpages
= cpa
->numpages
;
1321 * Store the remaining nr of pages for the large page
1322 * preservation check.
1324 cpa
->numpages
= numpages
;
1325 /* for array changes, we can't use large page */
1326 if (cpa
->flags
& (CPA_ARRAY
| CPA_PAGES_ARRAY
))
1329 if (!debug_pagealloc
)
1330 spin_lock(&cpa_lock
);
1331 ret
= __change_page_attr(cpa
, checkalias
);
1332 if (!debug_pagealloc
)
1333 spin_unlock(&cpa_lock
);
1338 ret
= cpa_process_alias(cpa
);
1344 * Adjust the number of pages with the result of the
1345 * CPA operation. Either a large page has been
1346 * preserved or a single page update happened.
1348 BUG_ON(cpa
->numpages
> numpages
);
1349 numpages
-= cpa
->numpages
;
1350 if (cpa
->flags
& (CPA_PAGES_ARRAY
| CPA_ARRAY
))
1353 *cpa
->vaddr
+= cpa
->numpages
* PAGE_SIZE
;
1359 static int change_page_attr_set_clr(unsigned long *addr
, int numpages
,
1360 pgprot_t mask_set
, pgprot_t mask_clr
,
1361 int force_split
, int in_flag
,
1362 struct page
**pages
)
1364 struct cpa_data cpa
;
1365 int ret
, cache
, checkalias
;
1366 unsigned long baddr
= 0;
1368 memset(&cpa
, 0, sizeof(cpa
));
1371 * Check, if we are requested to change a not supported
1374 mask_set
= canon_pgprot(mask_set
);
1375 mask_clr
= canon_pgprot(mask_clr
);
1376 if (!pgprot_val(mask_set
) && !pgprot_val(mask_clr
) && !force_split
)
1379 /* Ensure we are PAGE_SIZE aligned */
1380 if (in_flag
& CPA_ARRAY
) {
1382 for (i
= 0; i
< numpages
; i
++) {
1383 if (addr
[i
] & ~PAGE_MASK
) {
1384 addr
[i
] &= PAGE_MASK
;
1388 } else if (!(in_flag
& CPA_PAGES_ARRAY
)) {
1390 * in_flag of CPA_PAGES_ARRAY implies it is aligned.
1391 * No need to cehck in that case
1393 if (*addr
& ~PAGE_MASK
) {
1396 * People should not be passing in unaligned addresses:
1401 * Save address for cache flush. *addr is modified in the call
1402 * to __change_page_attr_set_clr() below.
1407 /* Must avoid aliasing mappings in the highmem code */
1408 kmap_flush_unused();
1414 cpa
.numpages
= numpages
;
1415 cpa
.mask_set
= mask_set
;
1416 cpa
.mask_clr
= mask_clr
;
1419 cpa
.force_split
= force_split
;
1421 if (in_flag
& (CPA_ARRAY
| CPA_PAGES_ARRAY
))
1422 cpa
.flags
|= in_flag
;
1424 /* No alias checking for _NX bit modifications */
1425 checkalias
= (pgprot_val(mask_set
) | pgprot_val(mask_clr
)) != _PAGE_NX
;
1427 ret
= __change_page_attr_set_clr(&cpa
, checkalias
);
1430 * Check whether we really changed something:
1432 if (!(cpa
.flags
& CPA_FLUSHTLB
))
1436 * No need to flush, when we did not set any of the caching
1439 cache
= !!pgprot2cachemode(mask_set
);
1442 * On success we use CLFLUSH, when the CPU supports it to
1443 * avoid the WBINVD. If the CPU does not support it and in the
1444 * error case we fall back to cpa_flush_all (which uses
1447 if (!ret
&& cpu_has_clflush
) {
1448 if (cpa
.flags
& (CPA_PAGES_ARRAY
| CPA_ARRAY
)) {
1449 cpa_flush_array(addr
, numpages
, cache
,
1452 cpa_flush_range(baddr
, numpages
, cache
);
1454 cpa_flush_all(cache
);
1460 static inline int change_page_attr_set(unsigned long *addr
, int numpages
,
1461 pgprot_t mask
, int array
)
1463 return change_page_attr_set_clr(addr
, numpages
, mask
, __pgprot(0), 0,
1464 (array
? CPA_ARRAY
: 0), NULL
);
1467 static inline int change_page_attr_clear(unsigned long *addr
, int numpages
,
1468 pgprot_t mask
, int array
)
1470 return change_page_attr_set_clr(addr
, numpages
, __pgprot(0), mask
, 0,
1471 (array
? CPA_ARRAY
: 0), NULL
);
1474 static inline int cpa_set_pages_array(struct page
**pages
, int numpages
,
1477 return change_page_attr_set_clr(NULL
, numpages
, mask
, __pgprot(0), 0,
1478 CPA_PAGES_ARRAY
, pages
);
1481 static inline int cpa_clear_pages_array(struct page
**pages
, int numpages
,
1484 return change_page_attr_set_clr(NULL
, numpages
, __pgprot(0), mask
, 0,
1485 CPA_PAGES_ARRAY
, pages
);
1488 int _set_memory_uc(unsigned long addr
, int numpages
)
1491 * for now UC MINUS. see comments in ioremap_nocache()
1492 * If you really need strong UC use ioremap_uc(), but note
1493 * that you cannot override IO areas with set_memory_*() as
1494 * these helpers cannot work with IO memory.
1496 return change_page_attr_set(&addr
, numpages
,
1497 cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS
),
1501 int set_memory_uc(unsigned long addr
, int numpages
)
1506 * for now UC MINUS. see comments in ioremap_nocache()
1508 ret
= reserve_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
,
1509 _PAGE_CACHE_MODE_UC_MINUS
, NULL
);
1513 ret
= _set_memory_uc(addr
, numpages
);
1520 free_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
);
1524 EXPORT_SYMBOL(set_memory_uc
);
1526 static int _set_memory_array(unsigned long *addr
, int addrinarray
,
1527 enum page_cache_mode new_type
)
1529 enum page_cache_mode set_type
;
1533 for (i
= 0; i
< addrinarray
; i
++) {
1534 ret
= reserve_memtype(__pa(addr
[i
]), __pa(addr
[i
]) + PAGE_SIZE
,
1540 /* If WC, set to UC- first and then WC */
1541 set_type
= (new_type
== _PAGE_CACHE_MODE_WC
) ?
1542 _PAGE_CACHE_MODE_UC_MINUS
: new_type
;
1544 ret
= change_page_attr_set(addr
, addrinarray
,
1545 cachemode2pgprot(set_type
), 1);
1547 if (!ret
&& new_type
== _PAGE_CACHE_MODE_WC
)
1548 ret
= change_page_attr_set_clr(addr
, addrinarray
,
1550 _PAGE_CACHE_MODE_WC
),
1551 __pgprot(_PAGE_CACHE_MASK
),
1552 0, CPA_ARRAY
, NULL
);
1559 for (j
= 0; j
< i
; j
++)
1560 free_memtype(__pa(addr
[j
]), __pa(addr
[j
]) + PAGE_SIZE
);
1565 int set_memory_array_uc(unsigned long *addr
, int addrinarray
)
1567 return _set_memory_array(addr
, addrinarray
, _PAGE_CACHE_MODE_UC_MINUS
);
1569 EXPORT_SYMBOL(set_memory_array_uc
);
1571 int set_memory_array_wc(unsigned long *addr
, int addrinarray
)
1573 return _set_memory_array(addr
, addrinarray
, _PAGE_CACHE_MODE_WC
);
1575 EXPORT_SYMBOL(set_memory_array_wc
);
1577 int set_memory_array_wt(unsigned long *addr
, int addrinarray
)
1579 return _set_memory_array(addr
, addrinarray
, _PAGE_CACHE_MODE_WT
);
1581 EXPORT_SYMBOL_GPL(set_memory_array_wt
);
1583 int _set_memory_wc(unsigned long addr
, int numpages
)
1586 unsigned long addr_copy
= addr
;
1588 ret
= change_page_attr_set(&addr
, numpages
,
1589 cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS
),
1592 ret
= change_page_attr_set_clr(&addr_copy
, numpages
,
1594 _PAGE_CACHE_MODE_WC
),
1595 __pgprot(_PAGE_CACHE_MASK
),
1601 int set_memory_wc(unsigned long addr
, int numpages
)
1605 ret
= reserve_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
,
1606 _PAGE_CACHE_MODE_WC
, NULL
);
1610 ret
= _set_memory_wc(addr
, numpages
);
1612 free_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
);
1616 EXPORT_SYMBOL(set_memory_wc
);
1618 int _set_memory_wt(unsigned long addr
, int numpages
)
1620 return change_page_attr_set(&addr
, numpages
,
1621 cachemode2pgprot(_PAGE_CACHE_MODE_WT
), 0);
1624 int set_memory_wt(unsigned long addr
, int numpages
)
1628 ret
= reserve_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
,
1629 _PAGE_CACHE_MODE_WT
, NULL
);
1633 ret
= _set_memory_wt(addr
, numpages
);
1635 free_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
);
1639 EXPORT_SYMBOL_GPL(set_memory_wt
);
1641 int _set_memory_wb(unsigned long addr
, int numpages
)
1643 /* WB cache mode is hard wired to all cache attribute bits being 0 */
1644 return change_page_attr_clear(&addr
, numpages
,
1645 __pgprot(_PAGE_CACHE_MASK
), 0);
1648 int set_memory_wb(unsigned long addr
, int numpages
)
1652 ret
= _set_memory_wb(addr
, numpages
);
1656 free_memtype(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
);
1659 EXPORT_SYMBOL(set_memory_wb
);
1661 int set_memory_array_wb(unsigned long *addr
, int addrinarray
)
1666 /* WB cache mode is hard wired to all cache attribute bits being 0 */
1667 ret
= change_page_attr_clear(addr
, addrinarray
,
1668 __pgprot(_PAGE_CACHE_MASK
), 1);
1672 for (i
= 0; i
< addrinarray
; i
++)
1673 free_memtype(__pa(addr
[i
]), __pa(addr
[i
]) + PAGE_SIZE
);
1677 EXPORT_SYMBOL(set_memory_array_wb
);
1679 int set_memory_x(unsigned long addr
, int numpages
)
1681 if (!(__supported_pte_mask
& _PAGE_NX
))
1684 return change_page_attr_clear(&addr
, numpages
, __pgprot(_PAGE_NX
), 0);
1686 EXPORT_SYMBOL(set_memory_x
);
1688 int set_memory_nx(unsigned long addr
, int numpages
)
1690 if (!(__supported_pte_mask
& _PAGE_NX
))
1693 return change_page_attr_set(&addr
, numpages
, __pgprot(_PAGE_NX
), 0);
1695 EXPORT_SYMBOL(set_memory_nx
);
1697 int set_memory_ro(unsigned long addr
, int numpages
)
1699 return change_page_attr_clear(&addr
, numpages
, __pgprot(_PAGE_RW
), 0);
1702 int set_memory_rw(unsigned long addr
, int numpages
)
1704 return change_page_attr_set(&addr
, numpages
, __pgprot(_PAGE_RW
), 0);
1707 int set_memory_np(unsigned long addr
, int numpages
)
1709 return change_page_attr_clear(&addr
, numpages
, __pgprot(_PAGE_PRESENT
), 0);
1712 int set_memory_4k(unsigned long addr
, int numpages
)
1714 return change_page_attr_set_clr(&addr
, numpages
, __pgprot(0),
1715 __pgprot(0), 1, 0, NULL
);
1718 int set_pages_uc(struct page
*page
, int numpages
)
1720 unsigned long addr
= (unsigned long)page_address(page
);
1722 return set_memory_uc(addr
, numpages
);
1724 EXPORT_SYMBOL(set_pages_uc
);
1726 static int _set_pages_array(struct page
**pages
, int addrinarray
,
1727 enum page_cache_mode new_type
)
1729 unsigned long start
;
1731 enum page_cache_mode set_type
;
1736 for (i
= 0; i
< addrinarray
; i
++) {
1737 if (PageHighMem(pages
[i
]))
1739 start
= page_to_pfn(pages
[i
]) << PAGE_SHIFT
;
1740 end
= start
+ PAGE_SIZE
;
1741 if (reserve_memtype(start
, end
, new_type
, NULL
))
1745 /* If WC, set to UC- first and then WC */
1746 set_type
= (new_type
== _PAGE_CACHE_MODE_WC
) ?
1747 _PAGE_CACHE_MODE_UC_MINUS
: new_type
;
1749 ret
= cpa_set_pages_array(pages
, addrinarray
,
1750 cachemode2pgprot(set_type
));
1751 if (!ret
&& new_type
== _PAGE_CACHE_MODE_WC
)
1752 ret
= change_page_attr_set_clr(NULL
, addrinarray
,
1754 _PAGE_CACHE_MODE_WC
),
1755 __pgprot(_PAGE_CACHE_MASK
),
1756 0, CPA_PAGES_ARRAY
, pages
);
1759 return 0; /* Success */
1762 for (i
= 0; i
< free_idx
; i
++) {
1763 if (PageHighMem(pages
[i
]))
1765 start
= page_to_pfn(pages
[i
]) << PAGE_SHIFT
;
1766 end
= start
+ PAGE_SIZE
;
1767 free_memtype(start
, end
);
1772 int set_pages_array_uc(struct page
**pages
, int addrinarray
)
1774 return _set_pages_array(pages
, addrinarray
, _PAGE_CACHE_MODE_UC_MINUS
);
1776 EXPORT_SYMBOL(set_pages_array_uc
);
1778 int set_pages_array_wc(struct page
**pages
, int addrinarray
)
1780 return _set_pages_array(pages
, addrinarray
, _PAGE_CACHE_MODE_WC
);
1782 EXPORT_SYMBOL(set_pages_array_wc
);
1784 int set_pages_array_wt(struct page
**pages
, int addrinarray
)
1786 return _set_pages_array(pages
, addrinarray
, _PAGE_CACHE_MODE_WT
);
1788 EXPORT_SYMBOL_GPL(set_pages_array_wt
);
1790 int set_pages_wb(struct page
*page
, int numpages
)
1792 unsigned long addr
= (unsigned long)page_address(page
);
1794 return set_memory_wb(addr
, numpages
);
1796 EXPORT_SYMBOL(set_pages_wb
);
1798 int set_pages_array_wb(struct page
**pages
, int addrinarray
)
1801 unsigned long start
;
1805 /* WB cache mode is hard wired to all cache attribute bits being 0 */
1806 retval
= cpa_clear_pages_array(pages
, addrinarray
,
1807 __pgprot(_PAGE_CACHE_MASK
));
1811 for (i
= 0; i
< addrinarray
; i
++) {
1812 if (PageHighMem(pages
[i
]))
1814 start
= page_to_pfn(pages
[i
]) << PAGE_SHIFT
;
1815 end
= start
+ PAGE_SIZE
;
1816 free_memtype(start
, end
);
1821 EXPORT_SYMBOL(set_pages_array_wb
);
1823 int set_pages_x(struct page
*page
, int numpages
)
1825 unsigned long addr
= (unsigned long)page_address(page
);
1827 return set_memory_x(addr
, numpages
);
1829 EXPORT_SYMBOL(set_pages_x
);
1831 int set_pages_nx(struct page
*page
, int numpages
)
1833 unsigned long addr
= (unsigned long)page_address(page
);
1835 return set_memory_nx(addr
, numpages
);
1837 EXPORT_SYMBOL(set_pages_nx
);
1839 int set_pages_ro(struct page
*page
, int numpages
)
1841 unsigned long addr
= (unsigned long)page_address(page
);
1843 return set_memory_ro(addr
, numpages
);
1846 int set_pages_rw(struct page
*page
, int numpages
)
1848 unsigned long addr
= (unsigned long)page_address(page
);
1850 return set_memory_rw(addr
, numpages
);
1853 #ifdef CONFIG_DEBUG_PAGEALLOC
1855 static int __set_pages_p(struct page
*page
, int numpages
)
1857 unsigned long tempaddr
= (unsigned long) page_address(page
);
1858 struct cpa_data cpa
= { .vaddr
= &tempaddr
,
1860 .numpages
= numpages
,
1861 .mask_set
= __pgprot(_PAGE_PRESENT
| _PAGE_RW
),
1862 .mask_clr
= __pgprot(0),
1866 * No alias checking needed for setting present flag. otherwise,
1867 * we may need to break large pages for 64-bit kernel text
1868 * mappings (this adds to complexity if we want to do this from
1869 * atomic context especially). Let's keep it simple!
1871 return __change_page_attr_set_clr(&cpa
, 0);
1874 static int __set_pages_np(struct page
*page
, int numpages
)
1876 unsigned long tempaddr
= (unsigned long) page_address(page
);
1877 struct cpa_data cpa
= { .vaddr
= &tempaddr
,
1879 .numpages
= numpages
,
1880 .mask_set
= __pgprot(0),
1881 .mask_clr
= __pgprot(_PAGE_PRESENT
| _PAGE_RW
),
1885 * No alias checking needed for setting not present flag. otherwise,
1886 * we may need to break large pages for 64-bit kernel text
1887 * mappings (this adds to complexity if we want to do this from
1888 * atomic context especially). Let's keep it simple!
1890 return __change_page_attr_set_clr(&cpa
, 0);
1893 void __kernel_map_pages(struct page
*page
, int numpages
, int enable
)
1895 if (PageHighMem(page
))
1898 debug_check_no_locks_freed(page_address(page
),
1899 numpages
* PAGE_SIZE
);
1903 * The return value is ignored as the calls cannot fail.
1904 * Large pages for identity mappings are not used at boot time
1905 * and hence no memory allocations during large page split.
1908 __set_pages_p(page
, numpages
);
1910 __set_pages_np(page
, numpages
);
1913 * We should perform an IPI and flush all tlbs,
1914 * but that can deadlock->flush only current cpu:
1918 arch_flush_lazy_mmu_mode();
1921 #ifdef CONFIG_HIBERNATION
1923 bool kernel_page_present(struct page
*page
)
1928 if (PageHighMem(page
))
1931 pte
= lookup_address((unsigned long)page_address(page
), &level
);
1932 return (pte_val(*pte
) & _PAGE_PRESENT
);
1935 #endif /* CONFIG_HIBERNATION */
1937 #endif /* CONFIG_DEBUG_PAGEALLOC */
1939 int kernel_map_pages_in_pgd(pgd_t
*pgd
, u64 pfn
, unsigned long address
,
1940 unsigned numpages
, unsigned long page_flags
)
1942 int retval
= -EINVAL
;
1944 struct cpa_data cpa
= {
1948 .numpages
= numpages
,
1949 .mask_set
= __pgprot(0),
1950 .mask_clr
= __pgprot(0),
1954 if (!(__supported_pte_mask
& _PAGE_NX
))
1957 if (!(page_flags
& _PAGE_NX
))
1958 cpa
.mask_clr
= __pgprot(_PAGE_NX
);
1960 cpa
.mask_set
= __pgprot(_PAGE_PRESENT
| page_flags
);
1962 retval
= __change_page_attr_set_clr(&cpa
, 0);
1969 void kernel_unmap_pages_in_pgd(pgd_t
*root
, unsigned long address
,
1972 unmap_pgd_range(root
, address
, address
+ (numpages
<< PAGE_SHIFT
));
1976 * The testcases use internal knowledge of the implementation that shouldn't
1977 * be exposed to the rest of the kernel. Include these directly here.
1979 #ifdef CONFIG_CPA_DEBUG
1980 #include "pageattr-test.c"