[linux-2.6-block.git] / arch / powerpc / mm / subpage-prot.c

/*
 * Copyright 2007-2008 Paul Mackerras, IBM Corp.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version
 * 2 of the License, or (at your option) any later version.
 */

#include <linux/errno.h>
#include <linux/kernel.h>
#include <linux/gfp.h>
#include <linux/types.h>
#include <linux/mm.h>
#include <linux/hugetlb.h>
#include <linux/syscalls.h>

#include <asm/pgtable.h>
#include <linux/uaccess.h>

/*
 * Free all pages allocated for subpage protection maps and pointers.
 * Also makes sure that the subpage_prot_table structure is
 * reinitialized for the next user.
 */
void subpage_prot_free(struct mm_struct *mm)
{
	struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context);
	unsigned long i, j, addr;
	u32 **p;

	if (!spt)
		return;

	for (i = 0; i < 4; ++i) {
		if (spt->low_prot[i]) {
			free_page((unsigned long)spt->low_prot[i]);
			spt->low_prot[i] = NULL;
		}
	}
	addr = 0;
	for (i = 0; i < (TASK_SIZE_USER64 >> 43); ++i) {
		p = spt->protptrs[i];
		if (!p)
			continue;
		spt->protptrs[i] = NULL;
		for (j = 0; j < SBP_L2_COUNT && addr < spt->maxaddr;
		     ++j, addr += PAGE_SIZE)
			if (p[j])
				free_page((unsigned long)p[j]);
		free_page((unsigned long)p);
	}
	spt->maxaddr = 0;
	kfree(spt);
}

static void hpte_flush_range(struct mm_struct *mm, unsigned long addr,
			     int npages)
{
	pgd_t *pgd;
	pud_t *pud;
	pmd_t *pmd;
	pte_t *pte;
	spinlock_t *ptl;

	pgd = pgd_offset(mm, addr);
	if (pgd_none(*pgd))
		return;
	pud = pud_offset(pgd, addr);
	if (pud_none(*pud))
		return;
	pmd = pmd_offset(pud, addr);
	if (pmd_none(*pmd))
		return;
	pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
	arch_enter_lazy_mmu_mode();
	for (; npages > 0; --npages) {
		pte_update(mm, addr, pte, 0, 0, 0);
		addr += PAGE_SIZE;
		++pte;
	}
	arch_leave_lazy_mmu_mode();
	pte_unmap_unlock(pte - 1, ptl);
}

/*
 * Clear the subpage protection map for an address range, allowing
 * all accesses that are allowed by the pte permissions.
 */
static void subpage_prot_clear(unsigned long addr, unsigned long len)
{
	struct mm_struct *mm = current->mm;
	struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context);
	u32 **spm, *spp;
	unsigned long i;
	size_t nw;
	unsigned long next, limit;

	if (!spt)
		return ;

	down_write(&mm->mmap_sem);
	limit = addr + len;
	if (limit > spt->maxaddr)
		limit = spt->maxaddr;
	for (; addr < limit; addr = next) {
		next = pmd_addr_end(addr, limit);
		if (addr < 0x100000000UL) {
			spm = spt->low_prot;
		} else {
			spm = spt->protptrs[addr >> SBP_L3_SHIFT];
			if (!spm)
				continue;
		}
		spp = spm[(addr >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
		if (!spp)
			continue;
		spp += (addr >> PAGE_SHIFT) & (SBP_L1_COUNT - 1);

		i = (addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
		nw = PTRS_PER_PTE - i;
		if (addr + (nw << PAGE_SHIFT) > next)
			nw = (next - addr) >> PAGE_SHIFT;

		memset(spp, 0, nw * sizeof(u32));

		/* now flush any existing HPTEs for the range */
		hpte_flush_range(mm, addr, nw);
	}
	up_write(&mm->mmap_sem);
}

#ifdef CONFIG_TRANSPARENT_HUGEPAGE
static int subpage_walk_pmd_entry(pmd_t *pmd, unsigned long addr,
				  unsigned long end, struct mm_walk *walk)
{
	struct vm_area_struct *vma = walk->vma;
	split_huge_pmd(vma, pmd, addr);
	return 0;
}

static void subpage_mark_vma_nohuge(struct mm_struct *mm, unsigned long addr,
				    unsigned long len)
{
	struct vm_area_struct *vma;
	struct mm_walk subpage_proto_walk = {
		.mm = mm,
		.pmd_entry = subpage_walk_pmd_entry,
	};

	/*
	 * We don't try too hard, we just mark all the vma in that range
	 * VM_NOHUGEPAGE and split them.
	 */
	vma = find_vma(mm, addr);
	/*
	 * If the range is in unmapped range, just return
	 */
	if (vma && ((addr + len) <= vma->vm_start))
		return;

	while (vma) {
		if (vma->vm_start >= (addr + len))
			break;
		vma->vm_flags |= VM_NOHUGEPAGE;
		walk_page_vma(vma, &subpage_proto_walk);
		vma = vma->vm_next;
	}
}
#else
static void subpage_mark_vma_nohuge(struct mm_struct *mm, unsigned long addr,
				    unsigned long len)
{
	return;
}
#endif

/*
 * Copy in a subpage protection map for an address range.
 * The map has 2 bits per 4k subpage, so 32 bits per 64k page.
 * Each 2-bit field is 0 to allow any access, 1 to prevent writes,
 * 2 or 3 to prevent all accesses.
 * Note that the normal page protections also apply; the subpage
 * protection mechanism is an additional constraint, so putting 0
 * in a 2-bit field won't allow writes to a page that is otherwise
 * write-protected.
 */
SYSCALL_DEFINE3(subpage_prot, unsigned long, addr,
		unsigned long, len, u32 __user *, map)
{
	struct mm_struct *mm = current->mm;
	struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context);
	u32 **spm, *spp;
	unsigned long i;
	size_t nw;
	unsigned long next, limit;
	int err;

	if (radix_enabled())
		return -ENOENT;

	/* Check parameters */
	if ((addr & ~PAGE_MASK) || (len & ~PAGE_MASK) ||
	    addr >= mm->task_size || len >= mm->task_size ||
	    addr + len > mm->task_size)
		return -EINVAL;

	if (is_hugepage_only_range(mm, addr, len))
		return -EINVAL;

	if (!map) {
		/* Clear out the protection map for the address range */
		subpage_prot_clear(addr, len);
		return 0;
	}

	if (!access_ok(map, (len >> PAGE_SHIFT) * sizeof(u32)))
		return -EFAULT;

	down_write(&mm->mmap_sem);

	if (!spt) {
		/*
		 * Allocate subpage prot table if not already done.
		 * Do this with mmap_sem held
		 */
		spt = kzalloc(sizeof(struct subpage_prot_table), GFP_KERNEL);
		if (!spt) {
			err = -ENOMEM;
			goto out;
		}
		mm->context.hash_context->spt = spt;
	}

	subpage_mark_vma_nohuge(mm, addr, len);
	for (limit = addr + len; addr < limit; addr = next) {
		next = pmd_addr_end(addr, limit);
		err = -ENOMEM;
		if (addr < 0x100000000UL) {
			spm = spt->low_prot;
		} else {
			spm = spt->protptrs[addr >> SBP_L3_SHIFT];
			if (!spm) {
				spm = (u32 **)get_zeroed_page(GFP_KERNEL);
				if (!spm)
					goto out;
				spt->protptrs[addr >> SBP_L3_SHIFT] = spm;
			}
		}
		spm += (addr >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1);
		spp = *spm;
		if (!spp) {
			spp = (u32 *)get_zeroed_page(GFP_KERNEL);
			if (!spp)
				goto out;
			*spm = spp;
		}
		spp += (addr >> PAGE_SHIFT) & (SBP_L1_COUNT - 1);

		local_irq_disable();
		demote_segment_4k(mm, addr);
		local_irq_enable();

		i = (addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
		nw = PTRS_PER_PTE - i;
		if (addr + (nw << PAGE_SHIFT) > next)
			nw = (next - addr) >> PAGE_SHIFT;

		up_write(&mm->mmap_sem);
		if (__copy_from_user(spp, map, nw * sizeof(u32)))
			return -EFAULT;
		map += nw;
		down_write(&mm->mmap_sem);

		/* now flush any existing HPTEs for the range */
		hpte_flush_range(mm, addr, nw);
	}
	if (limit > spt->maxaddr)
		spt->maxaddr = limit;
	err = 0;
 out:
	up_write(&mm->mmap_sem);
	return err;
}
Commit	Line	Data
fa28237c PM	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>
fa28237c PM	13	#include <linux/types.h>
	14	#include <linux/mm.h>
	15	#include <linux/hugetlb.h>
3691d614	16	#include <linux/syscalls.h>
fa28237c PM	17
fa28237c PM	18	#include <asm/pgtable.h>
7c0f6ba6	19	#include <linux/uaccess.h>
fa28237c PM	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	*/
d28513bc	26	void subpage_prot_free(struct mm_struct *mm)
fa28237c	27	{
60458fba	28	struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context);
fa28237c PM	29	unsigned long i, j, addr;
	30	u32 **p;
	31
ef629cc5 AK	32	if (!spt)
	33	return;
	34
fa28237c PM	35	for (i = 0; i < 4; ++i) {
	36	if (spt->low_prot[i]) {
	37	free_page((unsigned long)spt->low_prot[i]);
	38	spt->low_prot[i] = NULL;
	39	}
	40	}
	41	addr = 0;
0da12a7a	42	for (i = 0; i < (TASK_SIZE_USER64 >> 43); ++i) {
fa28237c PM	43	p = spt->protptrs[i];
	44	if (!p)
	45	continue;
	46	spt->protptrs[i] = NULL;
	47	for (j = 0; j < SBP_L2_COUNT && addr < spt->maxaddr;
	48	++j, addr += PAGE_SIZE)
	49	if (p[j])
	50	free_page((unsigned long)p[j]);
	51	free_page((unsigned long)p);
	52	}
	53	spt->maxaddr = 0;
ef629cc5	54	kfree(spt);
d28513bc DG	55	}
d28513bc DG	56
fa28237c PM	57	static void hpte_flush_range(struct mm_struct *mm, unsigned long addr,
	58	int npages)
	59	{
	60	pgd_t *pgd;
	61	pud_t *pud;
	62	pmd_t *pmd;
	63	pte_t *pte;
	64	spinlock_t *ptl;
	65
	66	pgd = pgd_offset(mm, addr);
	67	if (pgd_none(*pgd))
	68	return;
	69	pud = pud_offset(pgd, addr);
	70	if (pud_none(*pud))
	71	return;
	72	pmd = pmd_offset(pud, addr);
	73	if (pmd_none(*pmd))
	74	return;
	75	pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
	76	arch_enter_lazy_mmu_mode();
	77	for (; npages > 0; --npages) {
88247e8d	78	pte_update(mm, addr, pte, 0, 0, 0);
fa28237c PM	79	addr += PAGE_SIZE;
	80	++pte;
	81	}
	82	arch_leave_lazy_mmu_mode();
	83	pte_unmap_unlock(pte - 1, ptl);
	84	}
	85
	86	/*
	87	* Clear the subpage protection map for an address range, allowing
	88	* all accesses that are allowed by the pte permissions.
	89	*/
	90	static void subpage_prot_clear(unsigned long addr, unsigned long len)
	91	{
	92	struct mm_struct *mm = current->mm;
60458fba	93	struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context);
fa28237c	94	u32 *spm, spp;
6b5e7229 JM	95	unsigned long i;
6b5e7229 JM	96	size_t nw;
fa28237c PM	97	unsigned long next, limit;
fa28237c PM	98
ef629cc5 AK	99	if (!spt)
	100	return ;
	101
fa28237c PM	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);
b0d436c7	108	if (addr < 0x100000000UL) {
fa28237c PM	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
d8e355a2 AK	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)
	136	{
1757bbd9	137	struct vm_area_struct *vma = walk->vma;
78ddc534	138	split_huge_pmd(vma, pmd, addr);
d8e355a2 AK	139	return 0;
	140	}
	141
	142	static 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;
1757bbd9	166	walk_page_vma(vma, &subpage_proto_walk);
d8e355a2 AK	167	vma = vma->vm_next;
	168	}
	169	}
	170	#else
	171	static void subpage_mark_vma_nohuge(struct mm_struct *mm, unsigned long addr,
	172	unsigned long len)
	173	{
	174	return;
	175	}
	176	#endif
	177
fa28237c PM	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	*/
3691d614 AV	188	SYSCALL_DEFINE3(subpage_prot, unsigned long, addr,
3691d614 AV	189	unsigned long, len, u32 __user *, map)
fa28237c PM	190	{
fa28237c PM	191	struct mm_struct *mm = current->mm;
60458fba	192	struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context);
fa28237c	193	u32 *spm, spp;
6b5e7229 JM	194	unsigned long i;
6b5e7229 JM	195	size_t nw;
fa28237c PM	196	unsigned long next, limit;
	197	int err;
	198
5b2b8071 AK	199	if (radix_enabled())
	200	return -ENOENT;
	201
fa28237c PM	202	/* Check parameters */
fa28237c PM	203	if ((addr & ~PAGE_MASK) \|\| (len & ~PAGE_MASK) \|\|
be77e999 AK	204	addr >= mm->task_size \|\| len >= mm->task_size \|\|
be77e999 AK	205	addr + len > mm->task_size)
fa28237c PM	206	return -EINVAL;
	207
	208	if (is_hugepage_only_range(mm, addr, len))
	209	return -EINVAL;
	210
	211	if (!map) {
	212	/* Clear out the protection map for the address range */
	213	subpage_prot_clear(addr, len);
	214	return 0;
	215	}
	216
96d4f267	217	if (!access_ok(map, (len >> PAGE_SHIFT) * sizeof(u32)))
fa28237c PM	218	return -EFAULT;
	219
	220	down_write(&mm->mmap_sem);
ef629cc5 AK	221
	222	if (!spt) {
	223	/*
	224	* Allocate subpage prot table if not already done.
	225	* Do this with mmap_sem held
	226	*/
	227	spt = kzalloc(sizeof(struct subpage_prot_table), GFP_KERNEL);
	228	if (!spt) {
	229	err = -ENOMEM;
	230	goto out;
	231	}
	232	mm->context.hash_context->spt = spt;
	233	}
	234
d8e355a2	235	subpage_mark_vma_nohuge(mm, addr, len);
fa28237c PM	236	for (limit = addr + len; addr < limit; addr = next) {
	237	next = pmd_addr_end(addr, limit);
	238	err = -ENOMEM;
b0d436c7	239	if (addr < 0x100000000UL) {
fa28237c PM	240	spm = spt->low_prot;
	241	} else {
	242	spm = spt->protptrs[addr >> SBP_L3_SHIFT];
	243	if (!spm) {
	244	spm = (u32 **)get_zeroed_page(GFP_KERNEL);
	245	if (!spm)
	246	goto out;
	247	spt->protptrs[addr >> SBP_L3_SHIFT] = spm;
	248	}
	249	}
	250	spm += (addr >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1);
	251	spp = *spm;
	252	if (!spp) {
	253	spp = (u32 *)get_zeroed_page(GFP_KERNEL);
	254	if (!spp)
	255	goto out;
	256	*spm = spp;
	257	}
	258	spp += (addr >> PAGE_SHIFT) & (SBP_L1_COUNT - 1);
	259
	260	local_irq_disable();
	261	demote_segment_4k(mm, addr);
	262	local_irq_enable();
	263
	264	i = (addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
	265	nw = PTRS_PER_PTE - i;
	266	if (addr + (nw << PAGE_SHIFT) > next)
	267	nw = (next - addr) >> PAGE_SHIFT;
	268
	269	up_write(&mm->mmap_sem);
fa28237c	270	if (__copy_from_user(spp, map, nw * sizeof(u32)))
a967f161	271	return -EFAULT;
fa28237c PM	272	map += nw;
	273	down_write(&mm->mmap_sem);
	274
	275	/* now flush any existing HPTEs for the range */
	276	hpte_flush_range(mm, addr, nw);
	277	}
	278	if (limit > spt->maxaddr)
	279	spt->maxaddr = limit;
	280	err = 0;
	281	out:
	282	up_write(&mm->mmap_sem);
fa28237c PM	283	return err;
fa28237c PM	284	}