2 * Copyright 2002 Andi Kleen, SuSE Labs.
3 * Thanks to Ben LaHaise for precious feedback.
6 #include <linux/highmem.h>
7 #include <linux/module.h>
8 #include <linux/sched.h>
9 #include <linux/slab.h>
12 void clflush_cache_range(void *addr
, int size
)
16 for (i
= 0; i
< size
; i
+= boot_cpu_data
.x86_clflush_size
)
20 #include <asm/processor.h>
21 #include <asm/tlbflush.h>
22 #include <asm/sections.h>
23 #include <asm/uaccess.h>
24 #include <asm/pgalloc.h>
26 pte_t
*lookup_address(unsigned long address
, int *level
)
28 pgd_t
*pgd
= pgd_offset_k(address
);
34 pud
= pud_offset(pgd
, address
);
37 pmd
= pmd_offset(pud
, address
);
45 return pte_offset_kernel(pmd
, address
);
48 static void __set_pmd_pte(pte_t
*kpte
, unsigned long address
, pte_t pte
)
51 set_pte_atomic(kpte
, pte
);
53 if (SHARED_KERNEL_PMD
)
58 for (page
= pgd_list
; page
; page
= (struct page
*)page
->index
) {
63 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
64 pud
= pud_offset(pgd
, address
);
65 pmd
= pmd_offset(pud
, address
);
66 set_pte_atomic((pte_t
*)pmd
, pte
);
72 static int split_large_page(pte_t
*kpte
, unsigned long address
)
74 pgprot_t ref_prot
= pte_pgprot(pte_clrhuge(*kpte
));
75 gfp_t gfp_flags
= GFP_KERNEL
;
82 #ifdef CONFIG_DEBUG_PAGEALLOC
83 gfp_flags
= GFP_ATOMIC
;
85 base
= alloc_pages(gfp_flags
, 0);
89 spin_lock_irqsave(&pgd_lock
, flags
);
91 * Check for races, another CPU might have split this page
94 tmp
= lookup_address(address
, &level
);
100 address
= __pa(address
);
101 addr
= address
& LARGE_PAGE_MASK
;
102 pbase
= (pte_t
*)page_address(base
);
104 paravirt_alloc_pt(&init_mm
, page_to_pfn(base
));
107 for (i
= 0; i
< PTRS_PER_PTE
; i
++, addr
+= PAGE_SIZE
)
108 set_pte(&pbase
[i
], pfn_pte(addr
>> PAGE_SHIFT
, ref_prot
));
111 * Install the new, split up pagetable:
113 __set_pmd_pte(kpte
, address
, mk_pte(base
, ref_prot
));
117 spin_unlock_irqrestore(&pgd_lock
, flags
);
120 __free_pages(base
, 0);
126 __change_page_attr(unsigned long address
, struct page
*page
, pgprot_t prot
)
128 struct page
*kpte_page
;
132 BUG_ON(PageHighMem(page
));
135 kpte
= lookup_address(address
, &level
);
139 kpte_page
= virt_to_page(kpte
);
140 BUG_ON(PageLRU(kpte_page
));
141 BUG_ON(PageCompound(kpte_page
));
144 * Better fail early if someone sets the kernel text to NX.
145 * Does not cover __inittext
147 BUG_ON(address
>= (unsigned long)&_text
&&
148 address
< (unsigned long)&_etext
&&
149 (pgprot_val(prot
) & _PAGE_NX
));
152 set_pte_atomic(kpte
, mk_pte(page
, canon_pgprot(prot
)));
154 err
= split_large_page(kpte
, address
);
162 * change_page_attr_addr - Change page table attributes in linear mapping
163 * @address: Virtual address in linear mapping.
164 * @numpages: Number of pages to change
165 * @prot: New page table attribute (PAGE_*)
167 * Change page attributes of a page in the direct mapping. This is a variant
168 * of change_page_attr() that also works on memory holes that do not have
169 * mem_map entry (pfn_valid() is false).
171 * See change_page_attr() documentation for more details.
174 int change_page_attr_addr(unsigned long address
, int numpages
, pgprot_t prot
)
176 int err
= 0, kernel_map
= 0, i
;
179 if (address
>= __START_KERNEL_map
&&
180 address
< __START_KERNEL_map
+ KERNEL_TEXT_SIZE
) {
182 address
= (unsigned long)__va(__pa(address
));
187 for (i
= 0; i
< numpages
; i
++, address
+= PAGE_SIZE
) {
188 unsigned long pfn
= __pa(address
) >> PAGE_SHIFT
;
190 if (!kernel_map
|| pte_present(pfn_pte(0, prot
))) {
191 err
= __change_page_attr(address
, pfn_to_page(pfn
), prot
);
197 * Handle kernel mapping too which aliases part of
200 if (__pa(address
) < KERNEL_TEXT_SIZE
) {
204 addr2
= __START_KERNEL_map
+ __pa(address
);
205 /* Make sure the kernel mappings stay executable */
206 prot2
= pte_pgprot(pte_mkexec(pfn_pte(0, prot
)));
207 err
= __change_page_attr(addr2
, pfn_to_page(pfn
), prot2
);
216 * change_page_attr - Change page table attributes in the linear mapping.
217 * @page: First page to change
218 * @numpages: Number of pages to change
219 * @prot: New protection/caching type (PAGE_*)
221 * Returns 0 on success, otherwise a negated errno.
223 * This should be used when a page is mapped with a different caching policy
224 * than write-back somewhere - some CPUs do not like it when mappings with
225 * different caching policies exist. This changes the page attributes of the
226 * in kernel linear mapping too.
228 * Caller must call global_flush_tlb() later to make the changes active.
230 * The caller needs to ensure that there are no conflicting mappings elsewhere
231 * (e.g. in user space) * This function only deals with the kernel linear map.
233 * For MMIO areas without mem_map use change_page_attr_addr() instead.
235 int change_page_attr(struct page
*page
, int numpages
, pgprot_t prot
)
237 unsigned long addr
= (unsigned long)page_address(page
);
239 return change_page_attr_addr(addr
, numpages
, prot
);
241 EXPORT_SYMBOL(change_page_attr
);
243 static void flush_kernel_map(void *arg
)
246 * Flush all to work around Errata in early athlons regarding
247 * large page flushing.
251 if (boot_cpu_data
.x86_model
>= 4)
255 void global_flush_tlb(void)
257 BUG_ON(irqs_disabled());
259 on_each_cpu(flush_kernel_map
, NULL
, 1, 1);
261 EXPORT_SYMBOL(global_flush_tlb
);
263 #ifdef CONFIG_DEBUG_PAGEALLOC
264 void kernel_map_pages(struct page
*page
, int numpages
, int enable
)
266 if (PageHighMem(page
))
269 debug_check_no_locks_freed(page_address(page
),
270 numpages
* PAGE_SIZE
);
274 * If page allocator is not up yet then do not call c_p_a():
276 if (!debug_pagealloc_enabled
)
280 * the return value is ignored - the calls cannot fail,
281 * large pages are disabled at boot time.
283 change_page_attr(page
, numpages
, enable
? PAGE_KERNEL
: __pgprot(0));
286 * we should perform an IPI and flush all tlbs,
287 * but that can deadlock->flush only current cpu.