1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2014, The Linux Foundation. All rights reserved.
5 #include <linux/kernel.h>
7 #include <linux/module.h>
8 #include <linux/sched.h>
9 #include <linux/vmalloc.h>
11 #include <asm/cacheflush.h>
12 #include <asm/set_memory.h>
13 #include <asm/tlbflush.h>
15 struct page_change_data {
20 bool rodata_full __ro_after_init = IS_ENABLED(CONFIG_RODATA_FULL_DEFAULT_ENABLED);
22 bool can_set_direct_map(void)
24 return rodata_full || debug_pagealloc_enabled();
27 static int change_page_range(pte_t *ptep, unsigned long addr, void *data)
29 struct page_change_data *cdata = data;
30 pte_t pte = READ_ONCE(*ptep);
32 pte = clear_pte_bit(pte, cdata->clear_mask);
33 pte = set_pte_bit(pte, cdata->set_mask);
40 * This function assumes that the range is mapped with PAGE_SIZE pages.
42 static int __change_memory_common(unsigned long start, unsigned long size,
43 pgprot_t set_mask, pgprot_t clear_mask)
45 struct page_change_data data;
48 data.set_mask = set_mask;
49 data.clear_mask = clear_mask;
51 ret = apply_to_page_range(&init_mm, start, size, change_page_range,
54 flush_tlb_kernel_range(start, start + size);
58 static int change_memory_common(unsigned long addr, int numpages,
59 pgprot_t set_mask, pgprot_t clear_mask)
61 unsigned long start = addr;
62 unsigned long size = PAGE_SIZE * numpages;
63 unsigned long end = start + size;
64 struct vm_struct *area;
67 if (!PAGE_ALIGNED(addr)) {
74 * Kernel VA mappings are always live, and splitting live section
75 * mappings into page mappings may cause TLB conflicts. This means
76 * we have to ensure that changing the permission bits of the range
77 * we are operating on does not result in such splitting.
79 * Let's restrict ourselves to mappings created by vmalloc (or vmap).
80 * Those are guaranteed to consist entirely of page mappings, and
81 * splitting is never needed.
83 * So check whether the [addr, addr + size) interval is entirely
84 * covered by precisely one VM area that has the VM_ALLOC flag set.
86 area = find_vm_area((void *)addr);
88 end > (unsigned long)area->addr + area->size ||
89 !(area->flags & VM_ALLOC))
96 * If we are manipulating read-only permissions, apply the same
97 * change to the linear mapping of the pages that back this VM area.
99 if (rodata_full && (pgprot_val(set_mask) == PTE_RDONLY ||
100 pgprot_val(clear_mask) == PTE_RDONLY)) {
101 for (i = 0; i < area->nr_pages; i++) {
102 __change_memory_common((u64)page_address(area->pages[i]),
103 PAGE_SIZE, set_mask, clear_mask);
108 * Get rid of potentially aliasing lazily unmapped vm areas that may
109 * have permissions set that deviate from the ones we are setting here.
113 return __change_memory_common(start, size, set_mask, clear_mask);
116 int set_memory_ro(unsigned long addr, int numpages)
118 return change_memory_common(addr, numpages,
119 __pgprot(PTE_RDONLY),
120 __pgprot(PTE_WRITE));
123 int set_memory_rw(unsigned long addr, int numpages)
125 return change_memory_common(addr, numpages,
127 __pgprot(PTE_RDONLY));
130 int set_memory_nx(unsigned long addr, int numpages)
132 return change_memory_common(addr, numpages,
134 __pgprot(PTE_MAYBE_GP));
137 int set_memory_x(unsigned long addr, int numpages)
139 return change_memory_common(addr, numpages,
140 __pgprot(PTE_MAYBE_GP),
144 int set_memory_valid(unsigned long addr, int numpages, int enable)
147 return __change_memory_common(addr, PAGE_SIZE * numpages,
151 return __change_memory_common(addr, PAGE_SIZE * numpages,
153 __pgprot(PTE_VALID));
156 int set_direct_map_invalid_noflush(struct page *page)
158 struct page_change_data data = {
159 .set_mask = __pgprot(0),
160 .clear_mask = __pgprot(PTE_VALID),
163 if (!can_set_direct_map())
166 return apply_to_page_range(&init_mm,
167 (unsigned long)page_address(page),
168 PAGE_SIZE, change_page_range, &data);
171 int set_direct_map_default_noflush(struct page *page)
173 struct page_change_data data = {
174 .set_mask = __pgprot(PTE_VALID | PTE_WRITE),
175 .clear_mask = __pgprot(PTE_RDONLY),
178 if (!can_set_direct_map())
181 return apply_to_page_range(&init_mm,
182 (unsigned long)page_address(page),
183 PAGE_SIZE, change_page_range, &data);
186 #ifdef CONFIG_DEBUG_PAGEALLOC
187 void __kernel_map_pages(struct page *page, int numpages, int enable)
189 if (!can_set_direct_map())
192 set_memory_valid((unsigned long)page_address(page), numpages, enable);
194 #endif /* CONFIG_DEBUG_PAGEALLOC */
197 * This function is used to determine if a linear map page has been marked as
198 * not-valid. Walk the page table and check the PTE_VALID bit. This is based
199 * on kern_addr_valid(), which almost does what we need.
201 * Because this is only called on the kernel linear map, p?d_sect() implies
202 * p?d_present(). When debug_pagealloc is enabled, sections mappings are
205 bool kernel_page_present(struct page *page)
212 unsigned long addr = (unsigned long)page_address(page);
214 if (!can_set_direct_map())
217 pgdp = pgd_offset_k(addr);
218 if (pgd_none(READ_ONCE(*pgdp)))
221 p4dp = p4d_offset(pgdp, addr);
222 if (p4d_none(READ_ONCE(*p4dp)))
225 pudp = pud_offset(p4dp, addr);
226 pud = READ_ONCE(*pudp);
232 pmdp = pmd_offset(pudp, addr);
233 pmd = READ_ONCE(*pmdp);
239 ptep = pte_offset_kernel(pmdp, addr);
240 return pte_valid(READ_ONCE(*ptep));