1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright 2002 Andi Kleen, SuSE Labs.
4 * Thanks to Ben LaHaise for precious feedback.
6 #include <linux/highmem.h>
7 #include <linux/memblock.h>
8 #include <linux/sched.h>
10 #include <linux/interrupt.h>
11 #include <linux/seq_file.h>
12 #include <linux/debugfs.h>
13 #include <linux/pfn.h>
14 #include <linux/percpu.h>
15 #include <linux/gfp.h>
16 #include <linux/pci.h>
17 #include <linux/vmalloc.h>
18 #include <linux/libnvdimm.h>
20 #include <asm/e820/api.h>
21 #include <asm/processor.h>
22 #include <asm/tlbflush.h>
23 #include <asm/sections.h>
24 #include <asm/setup.h>
25 #include <linux/uaccess.h>
26 #include <asm/pgalloc.h>
27 #include <asm/proto.h>
28 #include <asm/memtype.h>
29 #include <asm/set_memory.h>
31 #include "../mm_internal.h"
34 * The current flushing context - we pass it instead of 5 arguments:
41 unsigned long numpages
;
42 unsigned long curpage
;
45 unsigned int force_split
: 1,
46 force_static_prot
: 1;
56 static const int cpa_warn_level
= CPA_PROTECT
;
59 * Serialize cpa() (for !DEBUG_PAGEALLOC which uses large identity mappings)
60 * using cpa_lock. So that we don't allow any other cpu, with stale large tlb
61 * entries change the page attribute in parallel to some other cpu
62 * splitting a large page entry along with changing the attribute.
64 static DEFINE_SPINLOCK(cpa_lock
);
66 #define CPA_FLUSHTLB 1
68 #define CPA_PAGES_ARRAY 4
69 #define CPA_NO_CHECK_ALIAS 8 /* Do not search for aliases */
72 static unsigned long direct_pages_count
[PG_LEVEL_NUM
];
74 void update_page_count(int level
, unsigned long pages
)
76 /* Protect against CPA */
78 direct_pages_count
[level
] += pages
;
79 spin_unlock(&pgd_lock
);
82 static void split_page_count(int level
)
84 if (direct_pages_count
[level
] == 0)
87 direct_pages_count
[level
]--;
88 direct_pages_count
[level
- 1] += PTRS_PER_PTE
;
91 void arch_report_meminfo(struct seq_file
*m
)
93 seq_printf(m
, "DirectMap4k: %8lu kB\n",
94 direct_pages_count
[PG_LEVEL_4K
] << 2);
95 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
96 seq_printf(m
, "DirectMap2M: %8lu kB\n",
97 direct_pages_count
[PG_LEVEL_2M
] << 11);
99 seq_printf(m
, "DirectMap4M: %8lu kB\n",
100 direct_pages_count
[PG_LEVEL_2M
] << 12);
103 seq_printf(m
, "DirectMap1G: %8lu kB\n",
104 direct_pages_count
[PG_LEVEL_1G
] << 20);
107 static inline void split_page_count(int level
) { }
110 #ifdef CONFIG_X86_CPA_STATISTICS
112 static unsigned long cpa_1g_checked
;
113 static unsigned long cpa_1g_sameprot
;
114 static unsigned long cpa_1g_preserved
;
115 static unsigned long cpa_2m_checked
;
116 static unsigned long cpa_2m_sameprot
;
117 static unsigned long cpa_2m_preserved
;
118 static unsigned long cpa_4k_install
;
120 static inline void cpa_inc_1g_checked(void)
125 static inline void cpa_inc_2m_checked(void)
130 static inline void cpa_inc_4k_install(void)
135 static inline void cpa_inc_lp_sameprot(int level
)
137 if (level
== PG_LEVEL_1G
)
143 static inline void cpa_inc_lp_preserved(int level
)
145 if (level
== PG_LEVEL_1G
)
151 static int cpastats_show(struct seq_file
*m
, void *p
)
153 seq_printf(m
, "1G pages checked: %16lu\n", cpa_1g_checked
);
154 seq_printf(m
, "1G pages sameprot: %16lu\n", cpa_1g_sameprot
);
155 seq_printf(m
, "1G pages preserved: %16lu\n", cpa_1g_preserved
);
156 seq_printf(m
, "2M pages checked: %16lu\n", cpa_2m_checked
);
157 seq_printf(m
, "2M pages sameprot: %16lu\n", cpa_2m_sameprot
);
158 seq_printf(m
, "2M pages preserved: %16lu\n", cpa_2m_preserved
);
159 seq_printf(m
, "4K pages set-checked: %16lu\n", cpa_4k_install
);
163 static int cpastats_open(struct inode
*inode
, struct file
*file
)
165 return single_open(file
, cpastats_show
, NULL
);
168 static const struct file_operations cpastats_fops
= {
169 .open
= cpastats_open
,
172 .release
= single_release
,
175 static int __init
cpa_stats_init(void)
177 debugfs_create_file("cpa_stats", S_IRUSR
, arch_debugfs_dir
, NULL
,
181 late_initcall(cpa_stats_init
);
183 static inline void cpa_inc_1g_checked(void) { }
184 static inline void cpa_inc_2m_checked(void) { }
185 static inline void cpa_inc_4k_install(void) { }
186 static inline void cpa_inc_lp_sameprot(int level
) { }
187 static inline void cpa_inc_lp_preserved(int level
) { }
192 within(unsigned long addr
, unsigned long start
, unsigned long end
)
194 return addr
>= start
&& addr
< end
;
198 within_inclusive(unsigned long addr
, unsigned long start
, unsigned long end
)
200 return addr
>= start
&& addr
<= end
;
205 static inline unsigned long highmap_start_pfn(void)
207 return __pa_symbol(_text
) >> PAGE_SHIFT
;
210 static inline unsigned long highmap_end_pfn(void)
212 /* Do not reference physical address outside the kernel. */
213 return __pa_symbol(roundup(_brk_end
, PMD_SIZE
) - 1) >> PAGE_SHIFT
;
216 static bool __cpa_pfn_in_highmap(unsigned long pfn
)
219 * Kernel text has an alias mapping at a high address, known
222 return within_inclusive(pfn
, highmap_start_pfn(), highmap_end_pfn());
227 static bool __cpa_pfn_in_highmap(unsigned long pfn
)
229 /* There is no highmap on 32-bit */
236 * See set_mce_nospec().
238 * Machine check recovery code needs to change cache mode of poisoned pages to
239 * UC to avoid speculative access logging another error. But passing the
240 * address of the 1:1 mapping to set_memory_uc() is a fine way to encourage a
241 * speculative access. So we cheat and flip the top bit of the address. This
242 * works fine for the code that updates the page tables. But at the end of the
243 * process we need to flush the TLB and cache and the non-canonical address
244 * causes a #GP fault when used by the INVLPG and CLFLUSH instructions.
246 * But in the common case we already have a canonical address. This code
247 * will fix the top bit if needed and is a no-op otherwise.
249 static inline unsigned long fix_addr(unsigned long addr
)
252 return (long)(addr
<< 1) >> 1;
258 static unsigned long __cpa_addr(struct cpa_data
*cpa
, unsigned long idx
)
260 if (cpa
->flags
& CPA_PAGES_ARRAY
) {
261 struct page
*page
= cpa
->pages
[idx
];
263 if (unlikely(PageHighMem(page
)))
266 return (unsigned long)page_address(page
);
269 if (cpa
->flags
& CPA_ARRAY
)
270 return cpa
->vaddr
[idx
];
272 return *cpa
->vaddr
+ idx
* PAGE_SIZE
;
279 static void clflush_cache_range_opt(void *vaddr
, unsigned int size
)
281 const unsigned long clflush_size
= boot_cpu_data
.x86_clflush_size
;
282 void *p
= (void *)((unsigned long)vaddr
& ~(clflush_size
- 1));
283 void *vend
= vaddr
+ size
;
288 for (; p
< vend
; p
+= clflush_size
)
293 * clflush_cache_range - flush a cache range with clflush
294 * @vaddr: virtual start address
295 * @size: number of bytes to flush
297 * CLFLUSHOPT is an unordered instruction which needs fencing with MFENCE or
298 * SFENCE to avoid ordering issues.
300 void clflush_cache_range(void *vaddr
, unsigned int size
)
303 clflush_cache_range_opt(vaddr
, size
);
306 EXPORT_SYMBOL_GPL(clflush_cache_range
);
308 #ifdef CONFIG_ARCH_HAS_PMEM_API
309 void arch_invalidate_pmem(void *addr
, size_t size
)
311 clflush_cache_range(addr
, size
);
313 EXPORT_SYMBOL_GPL(arch_invalidate_pmem
);
316 static void __cpa_flush_all(void *arg
)
318 unsigned long cache
= (unsigned long)arg
;
321 * Flush all to work around Errata in early athlons regarding
322 * large page flushing.
326 if (cache
&& boot_cpu_data
.x86
>= 4)
330 static void cpa_flush_all(unsigned long cache
)
332 BUG_ON(irqs_disabled() && !early_boot_irqs_disabled
);
334 on_each_cpu(__cpa_flush_all
, (void *) cache
, 1);
337 static void __cpa_flush_tlb(void *data
)
339 struct cpa_data
*cpa
= data
;
342 for (i
= 0; i
< cpa
->numpages
; i
++)
343 __flush_tlb_one_kernel(fix_addr(__cpa_addr(cpa
, i
)));
346 static void cpa_flush(struct cpa_data
*data
, int cache
)
348 struct cpa_data
*cpa
= data
;
351 BUG_ON(irqs_disabled() && !early_boot_irqs_disabled
);
353 if (cache
&& !static_cpu_has(X86_FEATURE_CLFLUSH
)) {
354 cpa_flush_all(cache
);
358 if (cpa
->numpages
<= tlb_single_page_flush_ceiling
)
359 on_each_cpu(__cpa_flush_tlb
, cpa
, 1);
367 for (i
= 0; i
< cpa
->numpages
; i
++) {
368 unsigned long addr
= __cpa_addr(cpa
, i
);
371 pte_t
*pte
= lookup_address(addr
, &level
);
374 * Only flush present addresses:
376 if (pte
&& (pte_val(*pte
) & _PAGE_PRESENT
))
377 clflush_cache_range_opt((void *)fix_addr(addr
), PAGE_SIZE
);
382 static bool overlaps(unsigned long r1_start
, unsigned long r1_end
,
383 unsigned long r2_start
, unsigned long r2_end
)
385 return (r1_start
<= r2_end
&& r1_end
>= r2_start
) ||
386 (r2_start
<= r1_end
&& r2_end
>= r1_start
);
389 #ifdef CONFIG_PCI_BIOS
391 * The BIOS area between 640k and 1Mb needs to be executable for PCI BIOS
392 * based config access (CONFIG_PCI_GOBIOS) support.
394 #define BIOS_PFN PFN_DOWN(BIOS_BEGIN)
395 #define BIOS_PFN_END PFN_DOWN(BIOS_END - 1)
397 static pgprotval_t
protect_pci_bios(unsigned long spfn
, unsigned long epfn
)
399 if (pcibios_enabled
&& overlaps(spfn
, epfn
, BIOS_PFN
, BIOS_PFN_END
))
404 static pgprotval_t
protect_pci_bios(unsigned long spfn
, unsigned long epfn
)
411 * The .rodata section needs to be read-only. Using the pfn catches all
412 * aliases. This also includes __ro_after_init, so do not enforce until
413 * kernel_set_to_readonly is true.
415 static pgprotval_t
protect_rodata(unsigned long spfn
, unsigned long epfn
)
417 unsigned long epfn_ro
, spfn_ro
= PFN_DOWN(__pa_symbol(__start_rodata
));
420 * Note: __end_rodata is at page aligned and not inclusive, so
421 * subtract 1 to get the last enforced PFN in the rodata area.
423 epfn_ro
= PFN_DOWN(__pa_symbol(__end_rodata
)) - 1;
425 if (kernel_set_to_readonly
&& overlaps(spfn
, epfn
, spfn_ro
, epfn_ro
))
431 * Protect kernel text against becoming non executable by forbidding
432 * _PAGE_NX. This protects only the high kernel mapping (_text -> _etext)
433 * out of which the kernel actually executes. Do not protect the low
436 * This does not cover __inittext since that is gone after boot.
438 static pgprotval_t
protect_kernel_text(unsigned long start
, unsigned long end
)
440 unsigned long t_end
= (unsigned long)_etext
- 1;
441 unsigned long t_start
= (unsigned long)_text
;
443 if (overlaps(start
, end
, t_start
, t_end
))
448 #if defined(CONFIG_X86_64)
450 * Once the kernel maps the text as RO (kernel_set_to_readonly is set),
451 * kernel text mappings for the large page aligned text, rodata sections
452 * will be always read-only. For the kernel identity mappings covering the
453 * holes caused by this alignment can be anything that user asks.
455 * This will preserve the large page mappings for kernel text/data at no
458 static pgprotval_t
protect_kernel_text_ro(unsigned long start
,
461 unsigned long t_end
= (unsigned long)__end_rodata_hpage_align
- 1;
462 unsigned long t_start
= (unsigned long)_text
;
465 if (!kernel_set_to_readonly
|| !overlaps(start
, end
, t_start
, t_end
))
468 * Don't enforce the !RW mapping for the kernel text mapping, if
469 * the current mapping is already using small page mapping. No
470 * need to work hard to preserve large page mappings in this case.
472 * This also fixes the Linux Xen paravirt guest boot failure caused
473 * by unexpected read-only mappings for kernel identity
474 * mappings. In this paravirt guest case, the kernel text mapping
475 * and the kernel identity mapping share the same page-table pages,
476 * so the protections for kernel text and identity mappings have to
479 if (lookup_address(start
, &level
) && (level
!= PG_LEVEL_4K
))
484 static pgprotval_t
protect_kernel_text_ro(unsigned long start
,
491 static inline bool conflicts(pgprot_t prot
, pgprotval_t val
)
493 return (pgprot_val(prot
) & ~val
) != pgprot_val(prot
);
496 static inline void check_conflict(int warnlvl
, pgprot_t prot
, pgprotval_t val
,
497 unsigned long start
, unsigned long end
,
498 unsigned long pfn
, const char *txt
)
500 static const char *lvltxt
[] = {
501 [CPA_CONFLICT
] = "conflict",
502 [CPA_PROTECT
] = "protect",
503 [CPA_DETECT
] = "detect",
506 if (warnlvl
> cpa_warn_level
|| !conflicts(prot
, val
))
509 pr_warn("CPA %8s %10s: 0x%016lx - 0x%016lx PFN %lx req %016llx prevent %016llx\n",
510 lvltxt
[warnlvl
], txt
, start
, end
, pfn
, (unsigned long long)pgprot_val(prot
),
511 (unsigned long long)val
);
515 * Certain areas of memory on x86 require very specific protection flags,
516 * for example the BIOS area or kernel text. Callers don't always get this
517 * right (again, ioremap() on BIOS memory is not uncommon) so this function
518 * checks and fixes these known static required protection bits.
520 static inline pgprot_t
static_protections(pgprot_t prot
, unsigned long start
,
521 unsigned long pfn
, unsigned long npg
,
522 unsigned long lpsize
, int warnlvl
)
524 pgprotval_t forbidden
, res
;
528 * There is no point in checking RW/NX conflicts when the requested
529 * mapping is setting the page !PRESENT.
531 if (!(pgprot_val(prot
) & _PAGE_PRESENT
))
534 /* Operate on the virtual address */
535 end
= start
+ npg
* PAGE_SIZE
- 1;
537 res
= protect_kernel_text(start
, end
);
538 check_conflict(warnlvl
, prot
, res
, start
, end
, pfn
, "Text NX");
542 * Special case to preserve a large page. If the change spawns the
543 * full large page mapping then there is no point to split it
544 * up. Happens with ftrace and is going to be removed once ftrace
545 * switched to text_poke().
547 if (lpsize
!= (npg
* PAGE_SIZE
) || (start
& (lpsize
- 1))) {
548 res
= protect_kernel_text_ro(start
, end
);
549 check_conflict(warnlvl
, prot
, res
, start
, end
, pfn
, "Text RO");
553 /* Check the PFN directly */
554 res
= protect_pci_bios(pfn
, pfn
+ npg
- 1);
555 check_conflict(warnlvl
, prot
, res
, start
, end
, pfn
, "PCIBIOS NX");
558 res
= protect_rodata(pfn
, pfn
+ npg
- 1);
559 check_conflict(warnlvl
, prot
, res
, start
, end
, pfn
, "Rodata RO");
562 return __pgprot(pgprot_val(prot
) & ~forbidden
);
566 * Lookup the page table entry for a virtual address in a specific pgd.
567 * Return a pointer to the entry and the level of the mapping.
569 pte_t
*lookup_address_in_pgd(pgd_t
*pgd
, unsigned long address
,
576 *level
= PG_LEVEL_NONE
;
581 p4d
= p4d_offset(pgd
, address
);
585 *level
= PG_LEVEL_512G
;
586 if (p4d_large(*p4d
) || !p4d_present(*p4d
))
589 pud
= pud_offset(p4d
, address
);
593 *level
= PG_LEVEL_1G
;
594 if (pud_large(*pud
) || !pud_present(*pud
))
597 pmd
= pmd_offset(pud
, address
);
601 *level
= PG_LEVEL_2M
;
602 if (pmd_large(*pmd
) || !pmd_present(*pmd
))
605 *level
= PG_LEVEL_4K
;
607 return pte_offset_kernel(pmd
, address
);
611 * Lookup the page table entry for a virtual address. Return a pointer
612 * to the entry and the level of the mapping.
614 * Note: We return pud and pmd either when the entry is marked large
615 * or when the present bit is not set. Otherwise we would return a
616 * pointer to a nonexisting mapping.
618 pte_t
*lookup_address(unsigned long address
, unsigned int *level
)
620 return lookup_address_in_pgd(pgd_offset_k(address
), address
, level
);
622 EXPORT_SYMBOL_GPL(lookup_address
);
625 * Lookup the page table entry for a virtual address in a given mm. Return a
626 * pointer to the entry and the level of the mapping.
628 pte_t
*lookup_address_in_mm(struct mm_struct
*mm
, unsigned long address
,
631 return lookup_address_in_pgd(pgd_offset(mm
, address
), address
, level
);
633 EXPORT_SYMBOL_GPL(lookup_address_in_mm
);
635 static pte_t
*_lookup_address_cpa(struct cpa_data
*cpa
, unsigned long address
,
639 return lookup_address_in_pgd(cpa
->pgd
+ pgd_index(address
),
642 return lookup_address(address
, level
);
646 * Lookup the PMD entry for a virtual address. Return a pointer to the entry
647 * or NULL if not present.
649 pmd_t
*lookup_pmd_address(unsigned long address
)
655 pgd
= pgd_offset_k(address
);
659 p4d
= p4d_offset(pgd
, address
);
660 if (p4d_none(*p4d
) || p4d_large(*p4d
) || !p4d_present(*p4d
))
663 pud
= pud_offset(p4d
, address
);
664 if (pud_none(*pud
) || pud_large(*pud
) || !pud_present(*pud
))
667 return pmd_offset(pud
, address
);
671 * This is necessary because __pa() does not work on some
672 * kinds of memory, like vmalloc() or the alloc_remap()
673 * areas on 32-bit NUMA systems. The percpu areas can
674 * end up in this kind of memory, for instance.
676 * This could be optimized, but it is only intended to be
677 * used at inititalization time, and keeping it
678 * unoptimized should increase the testing coverage for
679 * the more obscure platforms.
681 phys_addr_t
slow_virt_to_phys(void *__virt_addr
)
683 unsigned long virt_addr
= (unsigned long)__virt_addr
;
684 phys_addr_t phys_addr
;
685 unsigned long offset
;
689 pte
= lookup_address(virt_addr
, &level
);
693 * pXX_pfn() returns unsigned long, which must be cast to phys_addr_t
694 * before being left-shifted PAGE_SHIFT bits -- this trick is to
695 * make 32-PAE kernel work correctly.
699 phys_addr
= (phys_addr_t
)pud_pfn(*(pud_t
*)pte
) << PAGE_SHIFT
;
700 offset
= virt_addr
& ~PUD_PAGE_MASK
;
703 phys_addr
= (phys_addr_t
)pmd_pfn(*(pmd_t
*)pte
) << PAGE_SHIFT
;
704 offset
= virt_addr
& ~PMD_PAGE_MASK
;
707 phys_addr
= (phys_addr_t
)pte_pfn(*pte
) << PAGE_SHIFT
;
708 offset
= virt_addr
& ~PAGE_MASK
;
711 return (phys_addr_t
)(phys_addr
| offset
);
713 EXPORT_SYMBOL_GPL(slow_virt_to_phys
);
716 * Set the new pmd in all the pgds we know about:
718 static void __set_pmd_pte(pte_t
*kpte
, unsigned long address
, pte_t pte
)
721 set_pte_atomic(kpte
, pte
);
723 if (!SHARED_KERNEL_PMD
) {
726 list_for_each_entry(page
, &pgd_list
, lru
) {
732 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
733 p4d
= p4d_offset(pgd
, address
);
734 pud
= pud_offset(p4d
, address
);
735 pmd
= pmd_offset(pud
, address
);
736 set_pte_atomic((pte_t
*)pmd
, pte
);
742 static pgprot_t
pgprot_clear_protnone_bits(pgprot_t prot
)
745 * _PAGE_GLOBAL means "global page" for present PTEs.
746 * But, it is also used to indicate _PAGE_PROTNONE
747 * for non-present PTEs.
749 * This ensures that a _PAGE_GLOBAL PTE going from
750 * present to non-present is not confused as
753 if (!(pgprot_val(prot
) & _PAGE_PRESENT
))
754 pgprot_val(prot
) &= ~_PAGE_GLOBAL
;
759 static int __should_split_large_page(pte_t
*kpte
, unsigned long address
,
760 struct cpa_data
*cpa
)
762 unsigned long numpages
, pmask
, psize
, lpaddr
, pfn
, old_pfn
;
763 pgprot_t old_prot
, new_prot
, req_prot
, chk_prot
;
768 * Check for races, another CPU might have split this page
771 tmp
= _lookup_address_cpa(cpa
, address
, &level
);
777 old_prot
= pmd_pgprot(*(pmd_t
*)kpte
);
778 old_pfn
= pmd_pfn(*(pmd_t
*)kpte
);
779 cpa_inc_2m_checked();
782 old_prot
= pud_pgprot(*(pud_t
*)kpte
);
783 old_pfn
= pud_pfn(*(pud_t
*)kpte
);
784 cpa_inc_1g_checked();
790 psize
= page_level_size(level
);
791 pmask
= page_level_mask(level
);
794 * Calculate the number of pages, which fit into this large
795 * page starting at address:
797 lpaddr
= (address
+ psize
) & pmask
;
798 numpages
= (lpaddr
- address
) >> PAGE_SHIFT
;
799 if (numpages
< cpa
->numpages
)
800 cpa
->numpages
= numpages
;
803 * We are safe now. Check whether the new pgprot is the same:
804 * Convert protection attributes to 4k-format, as cpa->mask* are set
808 /* Clear PSE (aka _PAGE_PAT) and move PAT bit to correct position */
809 req_prot
= pgprot_large_2_4k(old_prot
);
811 pgprot_val(req_prot
) &= ~pgprot_val(cpa
->mask_clr
);
812 pgprot_val(req_prot
) |= pgprot_val(cpa
->mask_set
);
815 * req_prot is in format of 4k pages. It must be converted to large
816 * page format: the caching mode includes the PAT bit located at
817 * different bit positions in the two formats.
819 req_prot
= pgprot_4k_2_large(req_prot
);
820 req_prot
= pgprot_clear_protnone_bits(req_prot
);
821 if (pgprot_val(req_prot
) & _PAGE_PRESENT
)
822 pgprot_val(req_prot
) |= _PAGE_PSE
;
825 * old_pfn points to the large page base pfn. So we need to add the
826 * offset of the virtual address:
828 pfn
= old_pfn
+ ((address
& (psize
- 1)) >> PAGE_SHIFT
);
832 * Calculate the large page base address and the number of 4K pages
835 lpaddr
= address
& pmask
;
836 numpages
= psize
>> PAGE_SHIFT
;
839 * Sanity check that the existing mapping is correct versus the static
840 * protections. static_protections() guards against !PRESENT, so no
841 * extra conditional required here.
843 chk_prot
= static_protections(old_prot
, lpaddr
, old_pfn
, numpages
,
844 psize
, CPA_CONFLICT
);
846 if (WARN_ON_ONCE(pgprot_val(chk_prot
) != pgprot_val(old_prot
))) {
848 * Split the large page and tell the split code to
849 * enforce static protections.
851 cpa
->force_static_prot
= 1;
856 * Optimization: If the requested pgprot is the same as the current
857 * pgprot, then the large page can be preserved and no updates are
858 * required independent of alignment and length of the requested
859 * range. The above already established that the current pgprot is
860 * correct, which in consequence makes the requested pgprot correct
861 * as well if it is the same. The static protection scan below will
862 * not come to a different conclusion.
864 if (pgprot_val(req_prot
) == pgprot_val(old_prot
)) {
865 cpa_inc_lp_sameprot(level
);
870 * If the requested range does not cover the full page, split it up
872 if (address
!= lpaddr
|| cpa
->numpages
!= numpages
)
876 * Check whether the requested pgprot is conflicting with a static
877 * protection requirement in the large page.
879 new_prot
= static_protections(req_prot
, lpaddr
, old_pfn
, numpages
,
883 * If there is a conflict, split the large page.
885 * There used to be a 4k wise evaluation trying really hard to
886 * preserve the large pages, but experimentation has shown, that this
887 * does not help at all. There might be corner cases which would
888 * preserve one large page occasionally, but it's really not worth the
889 * extra code and cycles for the common case.
891 if (pgprot_val(req_prot
) != pgprot_val(new_prot
))
894 /* All checks passed. Update the large page mapping. */
895 new_pte
= pfn_pte(old_pfn
, new_prot
);
896 __set_pmd_pte(kpte
, address
, new_pte
);
897 cpa
->flags
|= CPA_FLUSHTLB
;
898 cpa_inc_lp_preserved(level
);
902 static int should_split_large_page(pte_t
*kpte
, unsigned long address
,
903 struct cpa_data
*cpa
)
907 if (cpa
->force_split
)
910 spin_lock(&pgd_lock
);
911 do_split
= __should_split_large_page(kpte
, address
, cpa
);
912 spin_unlock(&pgd_lock
);
917 static void split_set_pte(struct cpa_data
*cpa
, pte_t
*pte
, unsigned long pfn
,
918 pgprot_t ref_prot
, unsigned long address
,
921 unsigned int npg
= PFN_DOWN(size
);
925 * If should_split_large_page() discovered an inconsistent mapping,
926 * remove the invalid protection in the split mapping.
928 if (!cpa
->force_static_prot
)
931 /* Hand in lpsize = 0 to enforce the protection mechanism */
932 prot
= static_protections(ref_prot
, address
, pfn
, npg
, 0, CPA_PROTECT
);
934 if (pgprot_val(prot
) == pgprot_val(ref_prot
))
938 * If this is splitting a PMD, fix it up. PUD splits cannot be
939 * fixed trivially as that would require to rescan the newly
940 * installed PMD mappings after returning from split_large_page()
941 * so an eventual further split can allocate the necessary PTE
942 * pages. Warn for now and revisit it in case this actually
945 if (size
== PAGE_SIZE
)
948 pr_warn_once("CPA: Cannot fixup static protections for PUD split\n");
950 set_pte(pte
, pfn_pte(pfn
, ref_prot
));
954 __split_large_page(struct cpa_data
*cpa
, pte_t
*kpte
, unsigned long address
,
957 unsigned long lpaddr
, lpinc
, ref_pfn
, pfn
, pfninc
= 1;
958 pte_t
*pbase
= (pte_t
*)page_address(base
);
959 unsigned int i
, level
;
963 spin_lock(&pgd_lock
);
965 * Check for races, another CPU might have split this page
968 tmp
= _lookup_address_cpa(cpa
, address
, &level
);
970 spin_unlock(&pgd_lock
);
974 paravirt_alloc_pte(&init_mm
, page_to_pfn(base
));
978 ref_prot
= pmd_pgprot(*(pmd_t
*)kpte
);
980 * Clear PSE (aka _PAGE_PAT) and move
981 * PAT bit to correct position.
983 ref_prot
= pgprot_large_2_4k(ref_prot
);
984 ref_pfn
= pmd_pfn(*(pmd_t
*)kpte
);
985 lpaddr
= address
& PMD_MASK
;
990 ref_prot
= pud_pgprot(*(pud_t
*)kpte
);
991 ref_pfn
= pud_pfn(*(pud_t
*)kpte
);
992 pfninc
= PMD_PAGE_SIZE
>> PAGE_SHIFT
;
993 lpaddr
= address
& PUD_MASK
;
996 * Clear the PSE flags if the PRESENT flag is not set
997 * otherwise pmd_present/pmd_huge will return true
998 * even on a non present pmd.
1000 if (!(pgprot_val(ref_prot
) & _PAGE_PRESENT
))
1001 pgprot_val(ref_prot
) &= ~_PAGE_PSE
;
1005 spin_unlock(&pgd_lock
);
1009 ref_prot
= pgprot_clear_protnone_bits(ref_prot
);
1012 * Get the target pfn from the original entry:
1015 for (i
= 0; i
< PTRS_PER_PTE
; i
++, pfn
+= pfninc
, lpaddr
+= lpinc
)
1016 split_set_pte(cpa
, pbase
+ i
, pfn
, ref_prot
, lpaddr
, lpinc
);
1018 if (virt_addr_valid(address
)) {
1019 unsigned long pfn
= PFN_DOWN(__pa(address
));
1021 if (pfn_range_is_mapped(pfn
, pfn
+ 1))
1022 split_page_count(level
);
1026 * Install the new, split up pagetable.
1028 * We use the standard kernel pagetable protections for the new
1029 * pagetable protections, the actual ptes set above control the
1030 * primary protection behavior:
1032 __set_pmd_pte(kpte
, address
, mk_pte(base
, __pgprot(_KERNPG_TABLE
)));
1035 * Do a global flush tlb after splitting the large page
1036 * and before we do the actual change page attribute in the PTE.
1038 * Without this, we violate the TLB application note, that says:
1039 * "The TLBs may contain both ordinary and large-page
1040 * translations for a 4-KByte range of linear addresses. This
1041 * may occur if software modifies the paging structures so that
1042 * the page size used for the address range changes. If the two
1043 * translations differ with respect to page frame or attributes
1044 * (e.g., permissions), processor behavior is undefined and may
1045 * be implementation-specific."
1047 * We do this global tlb flush inside the cpa_lock, so that we
1048 * don't allow any other cpu, with stale tlb entries change the
1049 * page attribute in parallel, that also falls into the
1050 * just split large page entry.
1053 spin_unlock(&pgd_lock
);
1058 static int split_large_page(struct cpa_data
*cpa
, pte_t
*kpte
,
1059 unsigned long address
)
1063 if (!debug_pagealloc_enabled())
1064 spin_unlock(&cpa_lock
);
1065 base
= alloc_pages(GFP_KERNEL
, 0);
1066 if (!debug_pagealloc_enabled())
1067 spin_lock(&cpa_lock
);
1071 if (__split_large_page(cpa
, kpte
, address
, base
))
1077 static bool try_to_free_pte_page(pte_t
*pte
)
1081 for (i
= 0; i
< PTRS_PER_PTE
; i
++)
1082 if (!pte_none(pte
[i
]))
1085 free_page((unsigned long)pte
);
1089 static bool try_to_free_pmd_page(pmd_t
*pmd
)
1093 for (i
= 0; i
< PTRS_PER_PMD
; i
++)
1094 if (!pmd_none(pmd
[i
]))
1097 free_page((unsigned long)pmd
);
1101 static bool unmap_pte_range(pmd_t
*pmd
, unsigned long start
, unsigned long end
)
1103 pte_t
*pte
= pte_offset_kernel(pmd
, start
);
1105 while (start
< end
) {
1106 set_pte(pte
, __pte(0));
1112 if (try_to_free_pte_page((pte_t
*)pmd_page_vaddr(*pmd
))) {
1119 static void __unmap_pmd_range(pud_t
*pud
, pmd_t
*pmd
,
1120 unsigned long start
, unsigned long end
)
1122 if (unmap_pte_range(pmd
, start
, end
))
1123 if (try_to_free_pmd_page((pmd_t
*)pud_page_vaddr(*pud
)))
1127 static void unmap_pmd_range(pud_t
*pud
, unsigned long start
, unsigned long end
)
1129 pmd_t
*pmd
= pmd_offset(pud
, start
);
1132 * Not on a 2MB page boundary?
1134 if (start
& (PMD_SIZE
- 1)) {
1135 unsigned long next_page
= (start
+ PMD_SIZE
) & PMD_MASK
;
1136 unsigned long pre_end
= min_t(unsigned long, end
, next_page
);
1138 __unmap_pmd_range(pud
, pmd
, start
, pre_end
);
1145 * Try to unmap in 2M chunks.
1147 while (end
- start
>= PMD_SIZE
) {
1148 if (pmd_large(*pmd
))
1151 __unmap_pmd_range(pud
, pmd
, start
, start
+ PMD_SIZE
);
1161 return __unmap_pmd_range(pud
, pmd
, start
, end
);
1164 * Try again to free the PMD page if haven't succeeded above.
1166 if (!pud_none(*pud
))
1167 if (try_to_free_pmd_page((pmd_t
*)pud_page_vaddr(*pud
)))
1171 static void unmap_pud_range(p4d_t
*p4d
, unsigned long start
, unsigned long end
)
1173 pud_t
*pud
= pud_offset(p4d
, start
);
1176 * Not on a GB page boundary?
1178 if (start
& (PUD_SIZE
- 1)) {
1179 unsigned long next_page
= (start
+ PUD_SIZE
) & PUD_MASK
;
1180 unsigned long pre_end
= min_t(unsigned long, end
, next_page
);
1182 unmap_pmd_range(pud
, start
, pre_end
);
1189 * Try to unmap in 1G chunks?
1191 while (end
- start
>= PUD_SIZE
) {
1193 if (pud_large(*pud
))
1196 unmap_pmd_range(pud
, start
, start
+ PUD_SIZE
);
1206 unmap_pmd_range(pud
, start
, end
);
1209 * No need to try to free the PUD page because we'll free it in
1210 * populate_pgd's error path
1214 static int alloc_pte_page(pmd_t
*pmd
)
1216 pte_t
*pte
= (pte_t
*)get_zeroed_page(GFP_KERNEL
);
1220 set_pmd(pmd
, __pmd(__pa(pte
) | _KERNPG_TABLE
));
1224 static int alloc_pmd_page(pud_t
*pud
)
1226 pmd_t
*pmd
= (pmd_t
*)get_zeroed_page(GFP_KERNEL
);
1230 set_pud(pud
, __pud(__pa(pmd
) | _KERNPG_TABLE
));
1234 static void populate_pte(struct cpa_data
*cpa
,
1235 unsigned long start
, unsigned long end
,
1236 unsigned num_pages
, pmd_t
*pmd
, pgprot_t pgprot
)
1240 pte
= pte_offset_kernel(pmd
, start
);
1242 pgprot
= pgprot_clear_protnone_bits(pgprot
);
1244 while (num_pages
-- && start
< end
) {
1245 set_pte(pte
, pfn_pte(cpa
->pfn
, pgprot
));
1253 static long populate_pmd(struct cpa_data
*cpa
,
1254 unsigned long start
, unsigned long end
,
1255 unsigned num_pages
, pud_t
*pud
, pgprot_t pgprot
)
1259 pgprot_t pmd_pgprot
;
1262 * Not on a 2M boundary?
1264 if (start
& (PMD_SIZE
- 1)) {
1265 unsigned long pre_end
= start
+ (num_pages
<< PAGE_SHIFT
);
1266 unsigned long next_page
= (start
+ PMD_SIZE
) & PMD_MASK
;
1268 pre_end
= min_t(unsigned long, pre_end
, next_page
);
1269 cur_pages
= (pre_end
- start
) >> PAGE_SHIFT
;
1270 cur_pages
= min_t(unsigned int, num_pages
, cur_pages
);
1275 pmd
= pmd_offset(pud
, start
);
1277 if (alloc_pte_page(pmd
))
1280 populate_pte(cpa
, start
, pre_end
, cur_pages
, pmd
, pgprot
);
1286 * We mapped them all?
1288 if (num_pages
== cur_pages
)
1291 pmd_pgprot
= pgprot_4k_2_large(pgprot
);
1293 while (end
- start
>= PMD_SIZE
) {
1296 * We cannot use a 1G page so allocate a PMD page if needed.
1299 if (alloc_pmd_page(pud
))
1302 pmd
= pmd_offset(pud
, start
);
1304 set_pmd(pmd
, pmd_mkhuge(pfn_pmd(cpa
->pfn
,
1305 canon_pgprot(pmd_pgprot
))));
1308 cpa
->pfn
+= PMD_SIZE
>> PAGE_SHIFT
;
1309 cur_pages
+= PMD_SIZE
>> PAGE_SHIFT
;
1313 * Map trailing 4K pages.
1316 pmd
= pmd_offset(pud
, start
);
1318 if (alloc_pte_page(pmd
))
1321 populate_pte(cpa
, start
, end
, num_pages
- cur_pages
,
1327 static int populate_pud(struct cpa_data
*cpa
, unsigned long start
, p4d_t
*p4d
,
1333 pgprot_t pud_pgprot
;
1335 end
= start
+ (cpa
->numpages
<< PAGE_SHIFT
);
1338 * Not on a Gb page boundary? => map everything up to it with
1341 if (start
& (PUD_SIZE
- 1)) {
1342 unsigned long pre_end
;
1343 unsigned long next_page
= (start
+ PUD_SIZE
) & PUD_MASK
;
1345 pre_end
= min_t(unsigned long, end
, next_page
);
1346 cur_pages
= (pre_end
- start
) >> PAGE_SHIFT
;
1347 cur_pages
= min_t(int, (int)cpa
->numpages
, cur_pages
);
1349 pud
= pud_offset(p4d
, start
);
1355 if (alloc_pmd_page(pud
))
1358 cur_pages
= populate_pmd(cpa
, start
, pre_end
, cur_pages
,
1366 /* We mapped them all? */
1367 if (cpa
->numpages
== cur_pages
)
1370 pud
= pud_offset(p4d
, start
);
1371 pud_pgprot
= pgprot_4k_2_large(pgprot
);
1374 * Map everything starting from the Gb boundary, possibly with 1G pages
1376 while (boot_cpu_has(X86_FEATURE_GBPAGES
) && end
- start
>= PUD_SIZE
) {
1377 set_pud(pud
, pud_mkhuge(pfn_pud(cpa
->pfn
,
1378 canon_pgprot(pud_pgprot
))));
1381 cpa
->pfn
+= PUD_SIZE
>> PAGE_SHIFT
;
1382 cur_pages
+= PUD_SIZE
>> PAGE_SHIFT
;
1386 /* Map trailing leftover */
1390 pud
= pud_offset(p4d
, start
);
1392 if (alloc_pmd_page(pud
))
1395 tmp
= populate_pmd(cpa
, start
, end
, cpa
->numpages
- cur_pages
,
1406 * Restrictions for kernel page table do not necessarily apply when mapping in
1409 static int populate_pgd(struct cpa_data
*cpa
, unsigned long addr
)
1411 pgprot_t pgprot
= __pgprot(_KERNPG_TABLE
);
1412 pud_t
*pud
= NULL
; /* shut up gcc */
1417 pgd_entry
= cpa
->pgd
+ pgd_index(addr
);
1419 if (pgd_none(*pgd_entry
)) {
1420 p4d
= (p4d_t
*)get_zeroed_page(GFP_KERNEL
);
1424 set_pgd(pgd_entry
, __pgd(__pa(p4d
) | _KERNPG_TABLE
));
1428 * Allocate a PUD page and hand it down for mapping.
1430 p4d
= p4d_offset(pgd_entry
, addr
);
1431 if (p4d_none(*p4d
)) {
1432 pud
= (pud_t
*)get_zeroed_page(GFP_KERNEL
);
1436 set_p4d(p4d
, __p4d(__pa(pud
) | _KERNPG_TABLE
));
1439 pgprot_val(pgprot
) &= ~pgprot_val(cpa
->mask_clr
);
1440 pgprot_val(pgprot
) |= pgprot_val(cpa
->mask_set
);
1442 ret
= populate_pud(cpa
, addr
, p4d
, pgprot
);
1445 * Leave the PUD page in place in case some other CPU or thread
1446 * already found it, but remove any useless entries we just
1449 unmap_pud_range(p4d
, addr
,
1450 addr
+ (cpa
->numpages
<< PAGE_SHIFT
));
1454 cpa
->numpages
= ret
;
1458 static int __cpa_process_fault(struct cpa_data
*cpa
, unsigned long vaddr
,
1463 * Right now, we only execute this code path when mapping
1464 * the EFI virtual memory map regions, no other users
1465 * provide a ->pgd value. This may change in the future.
1467 return populate_pgd(cpa
, vaddr
);
1471 * Ignore all non primary paths.
1479 * Ignore the NULL PTE for kernel identity mapping, as it is expected
1481 * Also set numpages to '1' indicating that we processed cpa req for
1482 * one virtual address page and its pfn. TBD: numpages can be set based
1483 * on the initial value and the level returned by lookup_address().
1485 if (within(vaddr
, PAGE_OFFSET
,
1486 PAGE_OFFSET
+ (max_pfn_mapped
<< PAGE_SHIFT
))) {
1488 cpa
->pfn
= __pa(vaddr
) >> PAGE_SHIFT
;
1491 } else if (__cpa_pfn_in_highmap(cpa
->pfn
)) {
1492 /* Faults in the highmap are OK, so do not warn: */
1495 WARN(1, KERN_WARNING
"CPA: called for zero pte. "
1496 "vaddr = %lx cpa->vaddr = %lx\n", vaddr
,
1503 static int __change_page_attr(struct cpa_data
*cpa
, int primary
)
1505 unsigned long address
;
1508 pte_t
*kpte
, old_pte
;
1510 address
= __cpa_addr(cpa
, cpa
->curpage
);
1512 kpte
= _lookup_address_cpa(cpa
, address
, &level
);
1514 return __cpa_process_fault(cpa
, address
, primary
);
1517 if (pte_none(old_pte
))
1518 return __cpa_process_fault(cpa
, address
, primary
);
1520 if (level
== PG_LEVEL_4K
) {
1522 pgprot_t new_prot
= pte_pgprot(old_pte
);
1523 unsigned long pfn
= pte_pfn(old_pte
);
1525 pgprot_val(new_prot
) &= ~pgprot_val(cpa
->mask_clr
);
1526 pgprot_val(new_prot
) |= pgprot_val(cpa
->mask_set
);
1528 cpa_inc_4k_install();
1529 /* Hand in lpsize = 0 to enforce the protection mechanism */
1530 new_prot
= static_protections(new_prot
, address
, pfn
, 1, 0,
1533 new_prot
= pgprot_clear_protnone_bits(new_prot
);
1536 * We need to keep the pfn from the existing PTE,
1537 * after all we're only going to change it's attributes
1538 * not the memory it points to
1540 new_pte
= pfn_pte(pfn
, new_prot
);
1543 * Do we really change anything ?
1545 if (pte_val(old_pte
) != pte_val(new_pte
)) {
1546 set_pte_atomic(kpte
, new_pte
);
1547 cpa
->flags
|= CPA_FLUSHTLB
;
1554 * Check, whether we can keep the large page intact
1555 * and just change the pte:
1557 do_split
= should_split_large_page(kpte
, address
, cpa
);
1559 * When the range fits into the existing large page,
1560 * return. cp->numpages and cpa->tlbflush have been updated in
1567 * We have to split the large page:
1569 err
= split_large_page(cpa
, kpte
, address
);
1576 static int __change_page_attr_set_clr(struct cpa_data
*cpa
, int checkalias
);
1578 static int cpa_process_alias(struct cpa_data
*cpa
)
1580 struct cpa_data alias_cpa
;
1581 unsigned long laddr
= (unsigned long)__va(cpa
->pfn
<< PAGE_SHIFT
);
1582 unsigned long vaddr
;
1585 if (!pfn_range_is_mapped(cpa
->pfn
, cpa
->pfn
+ 1))
1589 * No need to redo, when the primary call touched the direct
1592 vaddr
= __cpa_addr(cpa
, cpa
->curpage
);
1593 if (!(within(vaddr
, PAGE_OFFSET
,
1594 PAGE_OFFSET
+ (max_pfn_mapped
<< PAGE_SHIFT
)))) {
1597 alias_cpa
.vaddr
= &laddr
;
1598 alias_cpa
.flags
&= ~(CPA_PAGES_ARRAY
| CPA_ARRAY
);
1599 alias_cpa
.curpage
= 0;
1601 ret
= __change_page_attr_set_clr(&alias_cpa
, 0);
1606 #ifdef CONFIG_X86_64
1608 * If the primary call didn't touch the high mapping already
1609 * and the physical address is inside the kernel map, we need
1610 * to touch the high mapped kernel as well:
1612 if (!within(vaddr
, (unsigned long)_text
, _brk_end
) &&
1613 __cpa_pfn_in_highmap(cpa
->pfn
)) {
1614 unsigned long temp_cpa_vaddr
= (cpa
->pfn
<< PAGE_SHIFT
) +
1615 __START_KERNEL_map
- phys_base
;
1617 alias_cpa
.vaddr
= &temp_cpa_vaddr
;
1618 alias_cpa
.flags
&= ~(CPA_PAGES_ARRAY
| CPA_ARRAY
);
1619 alias_cpa
.curpage
= 0;
1622 * The high mapping range is imprecise, so ignore the
1625 __change_page_attr_set_clr(&alias_cpa
, 0);
1632 static int __change_page_attr_set_clr(struct cpa_data
*cpa
, int checkalias
)
1634 unsigned long numpages
= cpa
->numpages
;
1635 unsigned long rempages
= numpages
;
1640 * Store the remaining nr of pages for the large page
1641 * preservation check.
1643 cpa
->numpages
= rempages
;
1644 /* for array changes, we can't use large page */
1645 if (cpa
->flags
& (CPA_ARRAY
| CPA_PAGES_ARRAY
))
1648 if (!debug_pagealloc_enabled())
1649 spin_lock(&cpa_lock
);
1650 ret
= __change_page_attr(cpa
, checkalias
);
1651 if (!debug_pagealloc_enabled())
1652 spin_unlock(&cpa_lock
);
1657 ret
= cpa_process_alias(cpa
);
1663 * Adjust the number of pages with the result of the
1664 * CPA operation. Either a large page has been
1665 * preserved or a single page update happened.
1667 BUG_ON(cpa
->numpages
> rempages
|| !cpa
->numpages
);
1668 rempages
-= cpa
->numpages
;
1669 cpa
->curpage
+= cpa
->numpages
;
1673 /* Restore the original numpages */
1674 cpa
->numpages
= numpages
;
1678 static int change_page_attr_set_clr(unsigned long *addr
, int numpages
,
1679 pgprot_t mask_set
, pgprot_t mask_clr
,
1680 int force_split
, int in_flag
,
1681 struct page
**pages
)
1683 struct cpa_data cpa
;
1684 int ret
, cache
, checkalias
;
1686 memset(&cpa
, 0, sizeof(cpa
));
1689 * Check, if we are requested to set a not supported
1690 * feature. Clearing non-supported features is OK.
1692 mask_set
= canon_pgprot(mask_set
);
1694 if (!pgprot_val(mask_set
) && !pgprot_val(mask_clr
) && !force_split
)
1697 /* Ensure we are PAGE_SIZE aligned */
1698 if (in_flag
& CPA_ARRAY
) {
1700 for (i
= 0; i
< numpages
; i
++) {
1701 if (addr
[i
] & ~PAGE_MASK
) {
1702 addr
[i
] &= PAGE_MASK
;
1706 } else if (!(in_flag
& CPA_PAGES_ARRAY
)) {
1708 * in_flag of CPA_PAGES_ARRAY implies it is aligned.
1709 * No need to check in that case
1711 if (*addr
& ~PAGE_MASK
) {
1714 * People should not be passing in unaligned addresses:
1720 /* Must avoid aliasing mappings in the highmem code */
1721 kmap_flush_unused();
1727 cpa
.numpages
= numpages
;
1728 cpa
.mask_set
= mask_set
;
1729 cpa
.mask_clr
= mask_clr
;
1732 cpa
.force_split
= force_split
;
1734 if (in_flag
& (CPA_ARRAY
| CPA_PAGES_ARRAY
))
1735 cpa
.flags
|= in_flag
;
1737 /* No alias checking for _NX bit modifications */
1738 checkalias
= (pgprot_val(mask_set
) | pgprot_val(mask_clr
)) != _PAGE_NX
;
1739 /* Has caller explicitly disabled alias checking? */
1740 if (in_flag
& CPA_NO_CHECK_ALIAS
)
1743 ret
= __change_page_attr_set_clr(&cpa
, checkalias
);
1746 * Check whether we really changed something:
1748 if (!(cpa
.flags
& CPA_FLUSHTLB
))
1752 * No need to flush, when we did not set any of the caching
1755 cache
= !!pgprot2cachemode(mask_set
);
1758 * On error; flush everything to be sure.
1761 cpa_flush_all(cache
);
1765 cpa_flush(&cpa
, cache
);
1770 static inline int change_page_attr_set(unsigned long *addr
, int numpages
,
1771 pgprot_t mask
, int array
)
1773 return change_page_attr_set_clr(addr
, numpages
, mask
, __pgprot(0), 0,
1774 (array
? CPA_ARRAY
: 0), NULL
);
1777 static inline int change_page_attr_clear(unsigned long *addr
, int numpages
,
1778 pgprot_t mask
, int array
)
1780 return change_page_attr_set_clr(addr
, numpages
, __pgprot(0), mask
, 0,
1781 (array
? CPA_ARRAY
: 0), NULL
);
1784 static inline int cpa_set_pages_array(struct page
**pages
, int numpages
,
1787 return change_page_attr_set_clr(NULL
, numpages
, mask
, __pgprot(0), 0,
1788 CPA_PAGES_ARRAY
, pages
);
1791 static inline int cpa_clear_pages_array(struct page
**pages
, int numpages
,
1794 return change_page_attr_set_clr(NULL
, numpages
, __pgprot(0), mask
, 0,
1795 CPA_PAGES_ARRAY
, pages
);
1799 * _set_memory_prot is an internal helper for callers that have been passed
1800 * a pgprot_t value from upper layers and a reservation has already been taken.
1801 * If you want to set the pgprot to a specific page protocol, use the
1802 * set_memory_xx() functions.
1804 int __set_memory_prot(unsigned long addr
, int numpages
, pgprot_t prot
)
1806 return change_page_attr_set_clr(&addr
, numpages
, prot
,
1807 __pgprot(~pgprot_val(prot
)), 0, 0,
1811 int _set_memory_uc(unsigned long addr
, int numpages
)
1814 * for now UC MINUS. see comments in ioremap()
1815 * If you really need strong UC use ioremap_uc(), but note
1816 * that you cannot override IO areas with set_memory_*() as
1817 * these helpers cannot work with IO memory.
1819 return change_page_attr_set(&addr
, numpages
,
1820 cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS
),
1824 int set_memory_uc(unsigned long addr
, int numpages
)
1829 * for now UC MINUS. see comments in ioremap()
1831 ret
= memtype_reserve(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
,
1832 _PAGE_CACHE_MODE_UC_MINUS
, NULL
);
1836 ret
= _set_memory_uc(addr
, numpages
);
1843 memtype_free(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
);
1847 EXPORT_SYMBOL(set_memory_uc
);
1849 int _set_memory_wc(unsigned long addr
, int numpages
)
1853 ret
= change_page_attr_set(&addr
, numpages
,
1854 cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS
),
1857 ret
= change_page_attr_set_clr(&addr
, numpages
,
1858 cachemode2pgprot(_PAGE_CACHE_MODE_WC
),
1859 __pgprot(_PAGE_CACHE_MASK
),
1865 int set_memory_wc(unsigned long addr
, int numpages
)
1869 ret
= memtype_reserve(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
,
1870 _PAGE_CACHE_MODE_WC
, NULL
);
1874 ret
= _set_memory_wc(addr
, numpages
);
1876 memtype_free(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
);
1880 EXPORT_SYMBOL(set_memory_wc
);
1882 int _set_memory_wt(unsigned long addr
, int numpages
)
1884 return change_page_attr_set(&addr
, numpages
,
1885 cachemode2pgprot(_PAGE_CACHE_MODE_WT
), 0);
1888 int _set_memory_wb(unsigned long addr
, int numpages
)
1890 /* WB cache mode is hard wired to all cache attribute bits being 0 */
1891 return change_page_attr_clear(&addr
, numpages
,
1892 __pgprot(_PAGE_CACHE_MASK
), 0);
1895 int set_memory_wb(unsigned long addr
, int numpages
)
1899 ret
= _set_memory_wb(addr
, numpages
);
1903 memtype_free(__pa(addr
), __pa(addr
) + numpages
* PAGE_SIZE
);
1906 EXPORT_SYMBOL(set_memory_wb
);
1908 int set_memory_x(unsigned long addr
, int numpages
)
1910 if (!(__supported_pte_mask
& _PAGE_NX
))
1913 return change_page_attr_clear(&addr
, numpages
, __pgprot(_PAGE_NX
), 0);
1916 int set_memory_nx(unsigned long addr
, int numpages
)
1918 if (!(__supported_pte_mask
& _PAGE_NX
))
1921 return change_page_attr_set(&addr
, numpages
, __pgprot(_PAGE_NX
), 0);
1924 int set_memory_ro(unsigned long addr
, int numpages
)
1926 return change_page_attr_clear(&addr
, numpages
, __pgprot(_PAGE_RW
), 0);
1929 int set_memory_rw(unsigned long addr
, int numpages
)
1931 return change_page_attr_set(&addr
, numpages
, __pgprot(_PAGE_RW
), 0);
1934 int set_memory_np(unsigned long addr
, int numpages
)
1936 return change_page_attr_clear(&addr
, numpages
, __pgprot(_PAGE_PRESENT
), 0);
1939 int set_memory_np_noalias(unsigned long addr
, int numpages
)
1941 int cpa_flags
= CPA_NO_CHECK_ALIAS
;
1943 return change_page_attr_set_clr(&addr
, numpages
, __pgprot(0),
1944 __pgprot(_PAGE_PRESENT
), 0,
1948 int set_memory_4k(unsigned long addr
, int numpages
)
1950 return change_page_attr_set_clr(&addr
, numpages
, __pgprot(0),
1951 __pgprot(0), 1, 0, NULL
);
1954 int set_memory_nonglobal(unsigned long addr
, int numpages
)
1956 return change_page_attr_clear(&addr
, numpages
,
1957 __pgprot(_PAGE_GLOBAL
), 0);
1960 int set_memory_global(unsigned long addr
, int numpages
)
1962 return change_page_attr_set(&addr
, numpages
,
1963 __pgprot(_PAGE_GLOBAL
), 0);
1966 static int __set_memory_enc_dec(unsigned long addr
, int numpages
, bool enc
)
1968 struct cpa_data cpa
;
1971 /* Nothing to do if memory encryption is not active */
1972 if (!mem_encrypt_active())
1975 /* Should not be working on unaligned addresses */
1976 if (WARN_ONCE(addr
& ~PAGE_MASK
, "misaligned address: %#lx\n", addr
))
1979 memset(&cpa
, 0, sizeof(cpa
));
1981 cpa
.numpages
= numpages
;
1982 cpa
.mask_set
= enc
? __pgprot(_PAGE_ENC
) : __pgprot(0);
1983 cpa
.mask_clr
= enc
? __pgprot(0) : __pgprot(_PAGE_ENC
);
1984 cpa
.pgd
= init_mm
.pgd
;
1986 /* Must avoid aliasing mappings in the highmem code */
1987 kmap_flush_unused();
1991 * Before changing the encryption attribute, we need to flush caches.
1995 ret
= __change_page_attr_set_clr(&cpa
, 1);
1998 * After changing the encryption attribute, we need to flush TLBs again
1999 * in case any speculative TLB caching occurred (but no need to flush
2000 * caches again). We could just use cpa_flush_all(), but in case TLB
2001 * flushing gets optimized in the cpa_flush() path use the same logic
2009 int set_memory_encrypted(unsigned long addr
, int numpages
)
2011 return __set_memory_enc_dec(addr
, numpages
, true);
2013 EXPORT_SYMBOL_GPL(set_memory_encrypted
);
2015 int set_memory_decrypted(unsigned long addr
, int numpages
)
2017 return __set_memory_enc_dec(addr
, numpages
, false);
2019 EXPORT_SYMBOL_GPL(set_memory_decrypted
);
2021 int set_pages_uc(struct page
*page
, int numpages
)
2023 unsigned long addr
= (unsigned long)page_address(page
);
2025 return set_memory_uc(addr
, numpages
);
2027 EXPORT_SYMBOL(set_pages_uc
);
2029 static int _set_pages_array(struct page
**pages
, int numpages
,
2030 enum page_cache_mode new_type
)
2032 unsigned long start
;
2034 enum page_cache_mode set_type
;
2039 for (i
= 0; i
< numpages
; i
++) {
2040 if (PageHighMem(pages
[i
]))
2042 start
= page_to_pfn(pages
[i
]) << PAGE_SHIFT
;
2043 end
= start
+ PAGE_SIZE
;
2044 if (memtype_reserve(start
, end
, new_type
, NULL
))
2048 /* If WC, set to UC- first and then WC */
2049 set_type
= (new_type
== _PAGE_CACHE_MODE_WC
) ?
2050 _PAGE_CACHE_MODE_UC_MINUS
: new_type
;
2052 ret
= cpa_set_pages_array(pages
, numpages
,
2053 cachemode2pgprot(set_type
));
2054 if (!ret
&& new_type
== _PAGE_CACHE_MODE_WC
)
2055 ret
= change_page_attr_set_clr(NULL
, numpages
,
2057 _PAGE_CACHE_MODE_WC
),
2058 __pgprot(_PAGE_CACHE_MASK
),
2059 0, CPA_PAGES_ARRAY
, pages
);
2062 return 0; /* Success */
2065 for (i
= 0; i
< free_idx
; i
++) {
2066 if (PageHighMem(pages
[i
]))
2068 start
= page_to_pfn(pages
[i
]) << PAGE_SHIFT
;
2069 end
= start
+ PAGE_SIZE
;
2070 memtype_free(start
, end
);
2075 int set_pages_array_uc(struct page
**pages
, int numpages
)
2077 return _set_pages_array(pages
, numpages
, _PAGE_CACHE_MODE_UC_MINUS
);
2079 EXPORT_SYMBOL(set_pages_array_uc
);
2081 int set_pages_array_wc(struct page
**pages
, int numpages
)
2083 return _set_pages_array(pages
, numpages
, _PAGE_CACHE_MODE_WC
);
2085 EXPORT_SYMBOL(set_pages_array_wc
);
2087 int set_pages_array_wt(struct page
**pages
, int numpages
)
2089 return _set_pages_array(pages
, numpages
, _PAGE_CACHE_MODE_WT
);
2091 EXPORT_SYMBOL_GPL(set_pages_array_wt
);
2093 int set_pages_wb(struct page
*page
, int numpages
)
2095 unsigned long addr
= (unsigned long)page_address(page
);
2097 return set_memory_wb(addr
, numpages
);
2099 EXPORT_SYMBOL(set_pages_wb
);
2101 int set_pages_array_wb(struct page
**pages
, int numpages
)
2104 unsigned long start
;
2108 /* WB cache mode is hard wired to all cache attribute bits being 0 */
2109 retval
= cpa_clear_pages_array(pages
, numpages
,
2110 __pgprot(_PAGE_CACHE_MASK
));
2114 for (i
= 0; i
< numpages
; i
++) {
2115 if (PageHighMem(pages
[i
]))
2117 start
= page_to_pfn(pages
[i
]) << PAGE_SHIFT
;
2118 end
= start
+ PAGE_SIZE
;
2119 memtype_free(start
, end
);
2124 EXPORT_SYMBOL(set_pages_array_wb
);
2126 int set_pages_ro(struct page
*page
, int numpages
)
2128 unsigned long addr
= (unsigned long)page_address(page
);
2130 return set_memory_ro(addr
, numpages
);
2133 int set_pages_rw(struct page
*page
, int numpages
)
2135 unsigned long addr
= (unsigned long)page_address(page
);
2137 return set_memory_rw(addr
, numpages
);
2140 static int __set_pages_p(struct page
*page
, int numpages
)
2142 unsigned long tempaddr
= (unsigned long) page_address(page
);
2143 struct cpa_data cpa
= { .vaddr
= &tempaddr
,
2145 .numpages
= numpages
,
2146 .mask_set
= __pgprot(_PAGE_PRESENT
| _PAGE_RW
),
2147 .mask_clr
= __pgprot(0),
2151 * No alias checking needed for setting present flag. otherwise,
2152 * we may need to break large pages for 64-bit kernel text
2153 * mappings (this adds to complexity if we want to do this from
2154 * atomic context especially). Let's keep it simple!
2156 return __change_page_attr_set_clr(&cpa
, 0);
2159 static int __set_pages_np(struct page
*page
, int numpages
)
2161 unsigned long tempaddr
= (unsigned long) page_address(page
);
2162 struct cpa_data cpa
= { .vaddr
= &tempaddr
,
2164 .numpages
= numpages
,
2165 .mask_set
= __pgprot(0),
2166 .mask_clr
= __pgprot(_PAGE_PRESENT
| _PAGE_RW
),
2170 * No alias checking needed for setting not present flag. otherwise,
2171 * we may need to break large pages for 64-bit kernel text
2172 * mappings (this adds to complexity if we want to do this from
2173 * atomic context especially). Let's keep it simple!
2175 return __change_page_attr_set_clr(&cpa
, 0);
2178 int set_direct_map_invalid_noflush(struct page
*page
)
2180 return __set_pages_np(page
, 1);
2183 int set_direct_map_default_noflush(struct page
*page
)
2185 return __set_pages_p(page
, 1);
2188 void __kernel_map_pages(struct page
*page
, int numpages
, int enable
)
2190 if (PageHighMem(page
))
2193 debug_check_no_locks_freed(page_address(page
),
2194 numpages
* PAGE_SIZE
);
2198 * The return value is ignored as the calls cannot fail.
2199 * Large pages for identity mappings are not used at boot time
2200 * and hence no memory allocations during large page split.
2203 __set_pages_p(page
, numpages
);
2205 __set_pages_np(page
, numpages
);
2208 * We should perform an IPI and flush all tlbs,
2209 * but that can deadlock->flush only current cpu.
2210 * Preemption needs to be disabled around __flush_tlb_all() due to
2211 * CR3 reload in __native_flush_tlb().
2217 arch_flush_lazy_mmu_mode();
2220 #ifdef CONFIG_HIBERNATION
2221 bool kernel_page_present(struct page
*page
)
2226 if (PageHighMem(page
))
2229 pte
= lookup_address((unsigned long)page_address(page
), &level
);
2230 return (pte_val(*pte
) & _PAGE_PRESENT
);
2232 #endif /* CONFIG_HIBERNATION */
2234 int __init
kernel_map_pages_in_pgd(pgd_t
*pgd
, u64 pfn
, unsigned long address
,
2235 unsigned numpages
, unsigned long page_flags
)
2237 int retval
= -EINVAL
;
2239 struct cpa_data cpa
= {
2243 .numpages
= numpages
,
2244 .mask_set
= __pgprot(0),
2245 .mask_clr
= __pgprot(~page_flags
& (_PAGE_NX
|_PAGE_RW
)),
2249 WARN_ONCE(num_online_cpus() > 1, "Don't call after initializing SMP");
2251 if (!(__supported_pte_mask
& _PAGE_NX
))
2254 if (!(page_flags
& _PAGE_ENC
))
2255 cpa
.mask_clr
= pgprot_encrypted(cpa
.mask_clr
);
2257 cpa
.mask_set
= __pgprot(_PAGE_PRESENT
| page_flags
);
2259 retval
= __change_page_attr_set_clr(&cpa
, 0);
2267 * __flush_tlb_all() flushes mappings only on current CPU and hence this
2268 * function shouldn't be used in an SMP environment. Presently, it's used only
2269 * during boot (way before smp_init()) by EFI subsystem and hence is ok.
2271 int __init
kernel_unmap_pages_in_pgd(pgd_t
*pgd
, unsigned long address
,
2272 unsigned long numpages
)
2277 * The typical sequence for unmapping is to find a pte through
2278 * lookup_address_in_pgd() (ideally, it should never return NULL because
2279 * the address is already mapped) and change it's protections. As pfn is
2280 * the *target* of a mapping, it's not useful while unmapping.
2282 struct cpa_data cpa
= {
2286 .numpages
= numpages
,
2287 .mask_set
= __pgprot(0),
2288 .mask_clr
= __pgprot(_PAGE_PRESENT
| _PAGE_RW
),
2292 WARN_ONCE(num_online_cpus() > 1, "Don't call after initializing SMP");
2294 retval
= __change_page_attr_set_clr(&cpa
, 0);
2301 * The testcases use internal knowledge of the implementation that shouldn't
2302 * be exposed to the rest of the kernel. Include these directly here.
2304 #ifdef CONFIG_CPA_DEBUG
2305 #include "cpa-test.c"