2 * Copyright 2007-2008 Paul Mackerras, IBM Corp.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
10 #include <linux/errno.h>
11 #include <linux/kernel.h>
12 #include <linux/gfp.h>
13 #include <linux/types.h>
15 #include <linux/hugetlb.h>
17 #include <asm/pgtable.h>
18 #include <asm/uaccess.h>
19 #include <asm/tlbflush.h>
22 * Free all pages allocated for subpage protection maps and pointers.
23 * Also makes sure that the subpage_prot_table structure is
24 * reinitialized for the next user.
26 void subpage_prot_free(struct mm_struct *mm)
28 struct subpage_prot_table *spt = &mm->context.spt;
29 unsigned long i, j, addr;
32 for (i = 0; i < 4; ++i) {
33 if (spt->low_prot[i]) {
34 free_page((unsigned long)spt->low_prot[i]);
35 spt->low_prot[i] = NULL;
39 for (i = 0; i < 2; ++i) {
43 spt->protptrs[i] = NULL;
44 for (j = 0; j < SBP_L2_COUNT && addr < spt->maxaddr;
45 ++j, addr += PAGE_SIZE)
47 free_page((unsigned long)p[j]);
48 free_page((unsigned long)p);
53 void subpage_prot_init_new_context(struct mm_struct *mm)
55 struct subpage_prot_table *spt = &mm->context.spt;
57 memset(spt, 0, sizeof(*spt));
60 static void hpte_flush_range(struct mm_struct *mm, unsigned long addr,
69 pgd = pgd_offset(mm, addr);
72 pud = pud_offset(pgd, addr);
75 pmd = pmd_offset(pud, addr);
78 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
79 arch_enter_lazy_mmu_mode();
80 for (; npages > 0; --npages) {
81 pte_update(mm, addr, pte, 0, 0, 0);
85 arch_leave_lazy_mmu_mode();
86 pte_unmap_unlock(pte - 1, ptl);
90 * Clear the subpage protection map for an address range, allowing
91 * all accesses that are allowed by the pte permissions.
93 static void subpage_prot_clear(unsigned long addr, unsigned long len)
95 struct mm_struct *mm = current->mm;
96 struct subpage_prot_table *spt = &mm->context.spt;
100 unsigned long next, limit;
102 down_write(&mm->mmap_sem);
104 if (limit > spt->maxaddr)
105 limit = spt->maxaddr;
106 for (; addr < limit; addr = next) {
107 next = pmd_addr_end(addr, limit);
108 if (addr < 0x100000000UL) {
111 spm = spt->protptrs[addr >> SBP_L3_SHIFT];
115 spp = spm[(addr >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
118 spp += (addr >> PAGE_SHIFT) & (SBP_L1_COUNT - 1);
120 i = (addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
121 nw = PTRS_PER_PTE - i;
122 if (addr + (nw << PAGE_SHIFT) > next)
123 nw = (next - addr) >> PAGE_SHIFT;
125 memset(spp, 0, nw * sizeof(u32));
127 /* now flush any existing HPTEs for the range */
128 hpte_flush_range(mm, addr, nw);
130 up_write(&mm->mmap_sem);
133 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
134 static int subpage_walk_pmd_entry(pmd_t *pmd, unsigned long addr,
135 unsigned long end, struct mm_walk *walk)
137 struct vm_area_struct *vma = walk->vma;
138 split_huge_page_pmd(vma, addr, pmd);
142 static void subpage_mark_vma_nohuge(struct mm_struct *mm, unsigned long addr,
145 struct vm_area_struct *vma;
146 struct mm_walk subpage_proto_walk = {
148 .pmd_entry = subpage_walk_pmd_entry,
152 * We don't try too hard, we just mark all the vma in that range
153 * VM_NOHUGEPAGE and split them.
155 vma = find_vma(mm, addr);
157 * If the range is in unmapped range, just return
159 if (vma && ((addr + len) <= vma->vm_start))
163 if (vma->vm_start >= (addr + len))
165 vma->vm_flags |= VM_NOHUGEPAGE;
166 walk_page_vma(vma, &subpage_proto_walk);
171 static void subpage_mark_vma_nohuge(struct mm_struct *mm, unsigned long addr,
179 * Copy in a subpage protection map for an address range.
180 * The map has 2 bits per 4k subpage, so 32 bits per 64k page.
181 * Each 2-bit field is 0 to allow any access, 1 to prevent writes,
182 * 2 or 3 to prevent all accesses.
183 * Note that the normal page protections also apply; the subpage
184 * protection mechanism is an additional constraint, so putting 0
185 * in a 2-bit field won't allow writes to a page that is otherwise
188 long sys_subpage_prot(unsigned long addr, unsigned long len, u32 __user *map)
190 struct mm_struct *mm = current->mm;
191 struct subpage_prot_table *spt = &mm->context.spt;
195 unsigned long next, limit;
198 /* Check parameters */
199 if ((addr & ~PAGE_MASK) || (len & ~PAGE_MASK) ||
200 addr >= TASK_SIZE || len >= TASK_SIZE || addr + len > TASK_SIZE)
203 if (is_hugepage_only_range(mm, addr, len))
207 /* Clear out the protection map for the address range */
208 subpage_prot_clear(addr, len);
212 if (!access_ok(VERIFY_READ, map, (len >> PAGE_SHIFT) * sizeof(u32)))
215 down_write(&mm->mmap_sem);
216 subpage_mark_vma_nohuge(mm, addr, len);
217 for (limit = addr + len; addr < limit; addr = next) {
218 next = pmd_addr_end(addr, limit);
220 if (addr < 0x100000000UL) {
223 spm = spt->protptrs[addr >> SBP_L3_SHIFT];
225 spm = (u32 **)get_zeroed_page(GFP_KERNEL);
228 spt->protptrs[addr >> SBP_L3_SHIFT] = spm;
231 spm += (addr >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1);
234 spp = (u32 *)get_zeroed_page(GFP_KERNEL);
239 spp += (addr >> PAGE_SHIFT) & (SBP_L1_COUNT - 1);
242 demote_segment_4k(mm, addr);
245 i = (addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
246 nw = PTRS_PER_PTE - i;
247 if (addr + (nw << PAGE_SHIFT) > next)
248 nw = (next - addr) >> PAGE_SHIFT;
250 up_write(&mm->mmap_sem);
252 if (__copy_from_user(spp, map, nw * sizeof(u32)))
255 down_write(&mm->mmap_sem);
257 /* now flush any existing HPTEs for the range */
258 hpte_flush_range(mm, addr, nw);
260 if (limit > spt->maxaddr)
261 spt->maxaddr = limit;
264 up_write(&mm->mmap_sem);