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);
 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	int i, nw;
 99	unsigned long next, limit;
100
101	down_write(&mm->mmap_sem);
102	limit = addr + len;
103	if (limit > spt->maxaddr)
104		limit = spt->maxaddr;
105	for (; addr < limit; addr = next) {
106		next = pmd_addr_end(addr, limit);
107		if (addr < 0x100000000) {
108			spm = spt->low_prot;
109		} else {
110			spm = spt->protptrs[addr >> SBP_L3_SHIFT];
111			if (!spm)
112				continue;
113		}
114		spp = spm[(addr >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
115		if (!spp)
116			continue;
117		spp += (addr >> PAGE_SHIFT) & (SBP_L1_COUNT - 1);
118
119		i = (addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
120		nw = PTRS_PER_PTE - i;
121		if (addr + (nw << PAGE_SHIFT) > next)
122			nw = (next - addr) >> PAGE_SHIFT;
123
124		memset(spp, 0, nw * sizeof(u32));
125
126		/* now flush any existing HPTEs for the range */
127		hpte_flush_range(mm, addr, nw);
128	}
129	up_write(&mm->mmap_sem);
130}
131
132/*
133 * Copy in a subpage protection map for an address range.
134 * The map has 2 bits per 4k subpage, so 32 bits per 64k page.
135 * Each 2-bit field is 0 to allow any access, 1 to prevent writes,
136 * 2 or 3 to prevent all accesses.
137 * Note that the normal page protections also apply; the subpage
138 * protection mechanism is an additional constraint, so putting 0
139 * in a 2-bit field won't allow writes to a page that is otherwise
140 * write-protected.
141 */
142long sys_subpage_prot(unsigned long addr, unsigned long len, u32 __user *map)
143{
144	struct mm_struct *mm = current->mm;
145	struct subpage_prot_table *spt = &mm->context.spt;
146	u32 **spm, *spp;
147	int i, nw;
148	unsigned long next, limit;
149	int err;
150
151	/* Check parameters */
152	if ((addr & ~PAGE_MASK) || (len & ~PAGE_MASK) ||
153	    addr >= TASK_SIZE || len >= TASK_SIZE || addr + len > TASK_SIZE)
154		return -EINVAL;
155
156	if (is_hugepage_only_range(mm, addr, len))
157		return -EINVAL;
158
159	if (!map) {
160		/* Clear out the protection map for the address range */
161		subpage_prot_clear(addr, len);
162		return 0;
163	}
164
165	if (!access_ok(VERIFY_READ, map, (len >> PAGE_SHIFT) * sizeof(u32)))
166		return -EFAULT;
167
168	down_write(&mm->mmap_sem);
169	for (limit = addr + len; addr < limit; addr = next) {
170		next = pmd_addr_end(addr, limit);
171		err = -ENOMEM;
172		if (addr < 0x100000000) {
173			spm = spt->low_prot;
174		} else {
175			spm = spt->protptrs[addr >> SBP_L3_SHIFT];
176			if (!spm) {
177				spm = (u32 **)get_zeroed_page(GFP_KERNEL);
178				if (!spm)
179					goto out;
180				spt->protptrs[addr >> SBP_L3_SHIFT] = spm;
181			}
182		}
183		spm += (addr >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1);
184		spp = *spm;
185		if (!spp) {
186			spp = (u32 *)get_zeroed_page(GFP_KERNEL);
187			if (!spp)
188				goto out;
189			*spm = spp;
190		}
191		spp += (addr >> PAGE_SHIFT) & (SBP_L1_COUNT - 1);
192
193		local_irq_disable();
194		demote_segment_4k(mm, addr);
195		local_irq_enable();
196
197		i = (addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
198		nw = PTRS_PER_PTE - i;
199		if (addr + (nw << PAGE_SHIFT) > next)
200			nw = (next - addr) >> PAGE_SHIFT;
201
202		up_write(&mm->mmap_sem);
203		err = -EFAULT;
204		if (__copy_from_user(spp, map, nw * sizeof(u32)))
205			goto out2;
206		map += nw;
207		down_write(&mm->mmap_sem);
208
209		/* now flush any existing HPTEs for the range */
210		hpte_flush_range(mm, addr, nw);
211	}
212	if (limit > spt->maxaddr)
213		spt->maxaddr = limit;
214	err = 0;
215 out:
216	up_write(&mm->mmap_sem);
217 out2:
218	return err;
219}