1/*
2 * Copyright 2007-2008 Paul Mackerras, IBM Corp.
3 *
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.
8 */
9
10#include <linux/errno.h>
11#include <linux/kernel.h>
12#include <linux/gfp.h>
13#include <linux/types.h>
14#include <linux/mm.h>
15#include <linux/hugetlb.h>
16
17#include <asm/pgtable.h>
18#include <asm/uaccess.h>
19#include <asm/tlbflush.h>
20
21/*
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.
25 */
26void subpage_prot_free(struct mm_struct *mm)
27{
28	struct subpage_prot_table *spt = &mm->context.spt;
29	unsigned long i, j, addr;
30	u32 **p;
31
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;
36		}
37	}
38	addr = 0;
39	for (i = 0; i < 2; ++i) {
40		p = spt->protptrs[i];
41		if (!p)
42			continue;
43		spt->protptrs[i] = NULL;
44		for (j = 0; j < SBP_L2_COUNT && addr < spt->maxaddr;
45		     ++j, addr += PAGE_SIZE)
46			if (p[j])
47				free_page((unsigned long)p[j]);
48		free_page((unsigned long)p);
49	}
50	spt->maxaddr = 0;
51}
52
53void subpage_prot_init_new_context(struct mm_struct *mm)
54{
55	struct subpage_prot_table *spt = &mm->context.spt;
56
57	memset(spt, 0, sizeof(*spt));
58}
59
60static void hpte_flush_range(struct mm_struct *mm, unsigned long addr,
61			     int npages)
62{
63	pgd_t *pgd;
64	pud_t *pud;
65	pmd_t *pmd;
66	pte_t *pte;
67	spinlock_t *ptl;
68
69	pgd = pgd_offset(mm, addr);
70	if (pgd_none(*pgd))
71		return;
72	pud = pud_offset(pgd, addr);
73	if (pud_none(*pud))
74		return;
75	pmd = pmd_offset(pud, addr);
76	if (pmd_none(*pmd))
77		return;
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);
82		addr += PAGE_SIZE;
83		++pte;
84	}
85	arch_leave_lazy_mmu_mode();
86	pte_unmap_unlock(pte - 1, ptl);
87}
88
89/*
90 * Clear the subpage protection map for an address range, allowing
91 * all accesses that are allowed by the pte permissions.
92 */
93static void subpage_prot_clear(unsigned long addr, unsigned long len)
94{
95	struct mm_struct *mm = current->mm;
96	struct subpage_prot_table *spt = &mm->context.spt;
97	u32 **spm, *spp;
98	unsigned long i;
99	size_t nw;
100	unsigned long next, limit;
101
102	down_write(&mm->mmap_sem);
103	limit = addr + len;
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) {
109			spm = spt->low_prot;
110		} else {
111			spm = spt->protptrs[addr >> SBP_L3_SHIFT];
112			if (!spm)
113				continue;
114		}
115		spp = spm[(addr >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
116		if (!spp)
117			continue;
118		spp += (addr >> PAGE_SHIFT) & (SBP_L1_COUNT - 1);
119
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;
124
125		memset(spp, 0, nw * sizeof(u32));
126
127		/* now flush any existing HPTEs for the range */
128		hpte_flush_range(mm, addr, nw);
129	}
130	up_write(&mm->mmap_sem);
131}
132
133#ifdef CONFIG_TRANSPARENT_HUGEPAGE
134static int subpage_walk_pmd_entry(pmd_t *pmd, unsigned long addr,
135				  unsigned long end, struct mm_walk *walk)
136{
137	struct vm_area_struct *vma = walk->vma;
138	split_huge_page_pmd(vma, addr, pmd);
139	return 0;
140}
141
142static void subpage_mark_vma_nohuge(struct mm_struct *mm, unsigned long addr,
143				    unsigned long len)
144{
145	struct vm_area_struct *vma;
146	struct mm_walk subpage_proto_walk = {
147		.mm = mm,
148		.pmd_entry = subpage_walk_pmd_entry,
149	};
150
151	/*
152	 * We don't try too hard, we just mark all the vma in that range
153	 * VM_NOHUGEPAGE and split them.
154	 */
155	vma = find_vma(mm, addr);
156	/*
157	 * If the range is in unmapped range, just return
158	 */
159	if (vma && ((addr + len) <= vma->vm_start))
160		return;
161
162	while (vma) {
163		if (vma->vm_start >= (addr + len))
164			break;
165		vma->vm_flags |= VM_NOHUGEPAGE;
166		walk_page_vma(vma, &subpage_proto_walk);
167		vma = vma->vm_next;
168	}
169}
170#else
171static void subpage_mark_vma_nohuge(struct mm_struct *mm, unsigned long addr,
172				    unsigned long len)
173{
174	return;
175}
176#endif
177
178/*
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
186 * write-protected.
187 */
188long sys_subpage_prot(unsigned long addr, unsigned long len, u32 __user *map)
189{
190	struct mm_struct *mm = current->mm;
191	struct subpage_prot_table *spt = &mm->context.spt;
192	u32 **spm, *spp;
193	unsigned long i;
194	size_t nw;
195	unsigned long next, limit;
196	int err;
197
198	/* Check parameters */
199	if ((addr & ~PAGE_MASK) || (len & ~PAGE_MASK) ||
200	    addr >= TASK_SIZE || len >= TASK_SIZE || addr + len > TASK_SIZE)
201		return -EINVAL;
202
203	if (is_hugepage_only_range(mm, addr, len))
204		return -EINVAL;
205
206	if (!map) {
207		/* Clear out the protection map for the address range */
208		subpage_prot_clear(addr, len);
209		return 0;
210	}
211
212	if (!access_ok(VERIFY_READ, map, (len >> PAGE_SHIFT) * sizeof(u32)))
213		return -EFAULT;
214
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);
219		err = -ENOMEM;
220		if (addr < 0x100000000UL) {
221			spm = spt->low_prot;
222		} else {
223			spm = spt->protptrs[addr >> SBP_L3_SHIFT];
224			if (!spm) {
225				spm = (u32 **)get_zeroed_page(GFP_KERNEL);
226				if (!spm)
227					goto out;
228				spt->protptrs[addr >> SBP_L3_SHIFT] = spm;
229			}
230		}
231		spm += (addr >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1);
232		spp = *spm;
233		if (!spp) {
234			spp = (u32 *)get_zeroed_page(GFP_KERNEL);
235			if (!spp)
236				goto out;
237			*spm = spp;
238		}
239		spp += (addr >> PAGE_SHIFT) & (SBP_L1_COUNT - 1);
240
241		local_irq_disable();
242		demote_segment_4k(mm, addr);
243		local_irq_enable();
244
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;
249
250		up_write(&mm->mmap_sem);
251		err = -EFAULT;
252		if (__copy_from_user(spp, map, nw * sizeof(u32)))
253			goto out2;
254		map += nw;
255		down_write(&mm->mmap_sem);
256
257		/* now flush any existing HPTEs for the range */
258		hpte_flush_range(mm, addr, nw);
259	}
260	if (limit > spt->maxaddr)
261		spt->maxaddr = limit;
262	err = 0;
263 out:
264	up_write(&mm->mmap_sem);
265 out2:
266	return err;
267}
268