[deliverable/linux.git] / arch / s390 / lib / uaccess_pt.c

/*
 *  arch/s390/lib/uaccess_pt.c
 *
 *  User access functions based on page table walks for enhanced
 *  system layout without hardware support.
 *
 *    Copyright IBM Corp. 2006
 *    Author(s): Gerald Schaefer (gerald.schaefer@de.ibm.com)
 */

#include <linux/errno.h>
#include <linux/hardirq.h>
#include <linux/mm.h>
#include <asm/uaccess.h>
#include <asm/futex.h>
#include "uaccess.h"

static int __handle_fault(struct mm_struct *mm, unsigned long address,
			  int write_access)
{
	struct vm_area_struct *vma;
	int ret = -EFAULT;
	int fault;

	if (in_atomic())
		return ret;
	down_read(&mm->mmap_sem);
	vma = find_vma(mm, address);
	if (unlikely(!vma))
		goto out;
	if (unlikely(vma->vm_start > address)) {
		if (!(vma->vm_flags & VM_GROWSDOWN))
			goto out;
		if (expand_stack(vma, address))
			goto out;
	}

	if (!write_access) {
		/* page not present, check vm flags */
		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
			goto out;
	} else {
		if (!(vma->vm_flags & VM_WRITE))
			goto out;
	}

survive:
	fault = handle_mm_fault(mm, vma, address, write_access);
	if (unlikely(fault & VM_FAULT_ERROR)) {
		if (fault & VM_FAULT_OOM)
			goto out_of_memory;
		else if (fault & VM_FAULT_SIGBUS)
			goto out_sigbus;
		BUG();
	}
	if (fault & VM_FAULT_MAJOR)
		current->maj_flt++;
	else
		current->min_flt++;
	ret = 0;
out:
	up_read(&mm->mmap_sem);
	return ret;

out_of_memory:
	up_read(&mm->mmap_sem);
	if (is_global_init(current)) {
		yield();
		down_read(&mm->mmap_sem);
		goto survive;
	}
	printk("VM: killing process %s\n", current->comm);
	return ret;

out_sigbus:
	up_read(&mm->mmap_sem);
	current->thread.prot_addr = address;
	current->thread.trap_no = 0x11;
	force_sig(SIGBUS, current);
	return ret;
}

static size_t __user_copy_pt(unsigned long uaddr, void *kptr,
			     size_t n, int write_user)
{
	struct mm_struct *mm = current->mm;
	unsigned long offset, pfn, done, size;
	pgd_t *pgd;
	pmd_t *pmd;
	pte_t *pte;
	void *from, *to;

	done = 0;
retry:
	spin_lock(&mm->page_table_lock);
	do {
		pgd = pgd_offset(mm, uaddr);
		if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
			goto fault;

		pmd = pmd_offset(pgd, uaddr);
		if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
			goto fault;

		pte = pte_offset_map(pmd, uaddr);
		if (!pte || !pte_present(*pte) ||
		    (write_user && !pte_write(*pte)))
			goto fault;

		pfn = pte_pfn(*pte);
		if (!pfn_valid(pfn))
			goto out;

		offset = uaddr & (PAGE_SIZE - 1);
		size = min(n - done, PAGE_SIZE - offset);
		if (write_user) {
			to = (void *)((pfn << PAGE_SHIFT) + offset);
			from = kptr + done;
		} else {
			from = (void *)((pfn << PAGE_SHIFT) + offset);
			to = kptr + done;
		}
		memcpy(to, from, size);
		done += size;
		uaddr += size;
	} while (done < n);
out:
	spin_unlock(&mm->page_table_lock);
	return n - done;
fault:
	spin_unlock(&mm->page_table_lock);
	if (__handle_fault(mm, uaddr, write_user))
		return n - done;
	goto retry;
}

/*
 * Do DAT for user address by page table walk, return kernel address.
 * This function needs to be called with current->mm->page_table_lock held.
 */
static unsigned long __dat_user_addr(unsigned long uaddr)
{
	struct mm_struct *mm = current->mm;
	unsigned long pfn, ret;
	pgd_t *pgd;
	pmd_t *pmd;
	pte_t *pte;
	int rc;

	ret = 0;
retry:
	pgd = pgd_offset(mm, uaddr);
	if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
		goto fault;

	pmd = pmd_offset(pgd, uaddr);
	if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
		goto fault;

	pte = pte_offset_map(pmd, uaddr);
	if (!pte || !pte_present(*pte))
		goto fault;

	pfn = pte_pfn(*pte);
	if (!pfn_valid(pfn))
		goto out;

	ret = (pfn << PAGE_SHIFT) + (uaddr & (PAGE_SIZE - 1));
out:
	return ret;
fault:
	spin_unlock(&mm->page_table_lock);
	rc = __handle_fault(mm, uaddr, 0);
	spin_lock(&mm->page_table_lock);
	if (rc)
		goto out;
	goto retry;
}

size_t copy_from_user_pt(size_t n, const void __user *from, void *to)
{
	size_t rc;

	if (segment_eq(get_fs(), KERNEL_DS)) {
		memcpy(to, (void __kernel __force *) from, n);
		return 0;
	}
	rc = __user_copy_pt((unsigned long) from, to, n, 0);
	if (unlikely(rc))
		memset(to + n - rc, 0, rc);
	return rc;
}

size_t copy_to_user_pt(size_t n, void __user *to, const void *from)
{
	if (segment_eq(get_fs(), KERNEL_DS)) {
		memcpy((void __kernel __force *) to, from, n);
		return 0;
	}
	return __user_copy_pt((unsigned long) to, (void *) from, n, 1);
}

static size_t clear_user_pt(size_t n, void __user *to)
{
	long done, size, ret;

	if (segment_eq(get_fs(), KERNEL_DS)) {
		memset((void __kernel __force *) to, 0, n);
		return 0;
	}
	done = 0;
	do {
		if (n - done > PAGE_SIZE)
			size = PAGE_SIZE;
		else
			size = n - done;
		ret = __user_copy_pt((unsigned long) to + done,
				      &empty_zero_page, size, 1);
		done += size;
		if (ret)
			return ret + n - done;
	} while (done < n);
	return 0;
}

static size_t strnlen_user_pt(size_t count, const char __user *src)
{
	char *addr;
	unsigned long uaddr = (unsigned long) src;
	struct mm_struct *mm = current->mm;
	unsigned long offset, pfn, done, len;
	pgd_t *pgd;
	pmd_t *pmd;
	pte_t *pte;
	size_t len_str;

	if (segment_eq(get_fs(), KERNEL_DS))
		return strnlen((const char __kernel __force *) src, count) + 1;
	done = 0;
retry:
	spin_lock(&mm->page_table_lock);
	do {
		pgd = pgd_offset(mm, uaddr);
		if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
			goto fault;

		pmd = pmd_offset(pgd, uaddr);
		if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
			goto fault;

		pte = pte_offset_map(pmd, uaddr);
		if (!pte || !pte_present(*pte))
			goto fault;

		pfn = pte_pfn(*pte);
		if (!pfn_valid(pfn)) {
			done = -1;
			goto out;
		}

		offset = uaddr & (PAGE_SIZE-1);
		addr = (char *)(pfn << PAGE_SHIFT) + offset;
		len = min(count - done, PAGE_SIZE - offset);
		len_str = strnlen(addr, len);
		done += len_str;
		uaddr += len_str;
	} while ((len_str == len) && (done < count));
out:
	spin_unlock(&mm->page_table_lock);
	return done + 1;
fault:
	spin_unlock(&mm->page_table_lock);
	if (__handle_fault(mm, uaddr, 0)) {
		return 0;
	}
	goto retry;
}

static size_t strncpy_from_user_pt(size_t count, const char __user *src,
				   char *dst)
{
	size_t n = strnlen_user_pt(count, src);

	if (!n)
		return -EFAULT;
	if (n > count)
		n = count;
	if (segment_eq(get_fs(), KERNEL_DS)) {
		memcpy(dst, (const char __kernel __force *) src, n);
		if (dst[n-1] == '\0')
			return n-1;
		else
			return n;
	}
	if (__user_copy_pt((unsigned long) src, dst, n, 0))
		return -EFAULT;
	if (dst[n-1] == '\0')
		return n-1;
	else
		return n;
}

static size_t copy_in_user_pt(size_t n, void __user *to,
			      const void __user *from)
{
	struct mm_struct *mm = current->mm;
	unsigned long offset_from, offset_to, offset_max, pfn_from, pfn_to,
		      uaddr, done, size;
	unsigned long uaddr_from = (unsigned long) from;
	unsigned long uaddr_to = (unsigned long) to;
	pgd_t *pgd_from, *pgd_to;
	pmd_t *pmd_from, *pmd_to;
	pte_t *pte_from, *pte_to;
	int write_user;

	done = 0;
retry:
	spin_lock(&mm->page_table_lock);
	do {
		pgd_from = pgd_offset(mm, uaddr_from);
		if (pgd_none(*pgd_from) || unlikely(pgd_bad(*pgd_from))) {
			uaddr = uaddr_from;
			write_user = 0;
			goto fault;
		}
		pgd_to = pgd_offset(mm, uaddr_to);
		if (pgd_none(*pgd_to) || unlikely(pgd_bad(*pgd_to))) {
			uaddr = uaddr_to;
			write_user = 1;
			goto fault;
		}

		pmd_from = pmd_offset(pgd_from, uaddr_from);
		if (pmd_none(*pmd_from) || unlikely(pmd_bad(*pmd_from))) {
			uaddr = uaddr_from;
			write_user = 0;
			goto fault;
		}
		pmd_to = pmd_offset(pgd_to, uaddr_to);
		if (pmd_none(*pmd_to) || unlikely(pmd_bad(*pmd_to))) {
			uaddr = uaddr_to;
			write_user = 1;
			goto fault;
		}

		pte_from = pte_offset_map(pmd_from, uaddr_from);
		if (!pte_from || !pte_present(*pte_from)) {
			uaddr = uaddr_from;
			write_user = 0;
			goto fault;
		}
		pte_to = pte_offset_map(pmd_to, uaddr_to);
		if (!pte_to || !pte_present(*pte_to) || !pte_write(*pte_to)) {
			uaddr = uaddr_to;
			write_user = 1;
			goto fault;
		}

		pfn_from = pte_pfn(*pte_from);
		if (!pfn_valid(pfn_from))
			goto out;
		pfn_to = pte_pfn(*pte_to);
		if (!pfn_valid(pfn_to))
			goto out;

		offset_from = uaddr_from & (PAGE_SIZE-1);
		offset_to = uaddr_from & (PAGE_SIZE-1);
		offset_max = max(offset_from, offset_to);
		size = min(n - done, PAGE_SIZE - offset_max);

		memcpy((void *)(pfn_to << PAGE_SHIFT) + offset_to,
		       (void *)(pfn_from << PAGE_SHIFT) + offset_from, size);
		done += size;
		uaddr_from += size;
		uaddr_to += size;
	} while (done < n);
out:
	spin_unlock(&mm->page_table_lock);
	return n - done;
fault:
	spin_unlock(&mm->page_table_lock);
	if (__handle_fault(mm, uaddr, write_user))
		return n - done;
	goto retry;
}

#define __futex_atomic_op(insn, ret, oldval, newval, uaddr, oparg)	\
	asm volatile("0: l   %1,0(%6)\n"				\
		     "1: " insn						\
		     "2: cs  %1,%2,0(%6)\n"				\
		     "3: jl  1b\n"					\
		     "   lhi %0,0\n"					\
		     "4:\n"						\
		     EX_TABLE(0b,4b) EX_TABLE(2b,4b) EX_TABLE(3b,4b)	\
		     : "=d" (ret), "=&d" (oldval), "=&d" (newval),	\
		       "=m" (*uaddr)					\
		     : "0" (-EFAULT), "d" (oparg), "a" (uaddr),		\
		       "m" (*uaddr) : "cc" );

int futex_atomic_op_pt(int op, int __user *uaddr, int oparg, int *old)
{
	int oldval = 0, newval, ret;

	spin_lock(&current->mm->page_table_lock);
	uaddr = (int __user *) __dat_user_addr((unsigned long) uaddr);
	if (!uaddr) {
		spin_unlock(&current->mm->page_table_lock);
		return -EFAULT;
	}
	get_page(virt_to_page(uaddr));
	spin_unlock(&current->mm->page_table_lock);
	switch (op) {
	case FUTEX_OP_SET:
		__futex_atomic_op("lr %2,%5\n",
				  ret, oldval, newval, uaddr, oparg);
		break;
	case FUTEX_OP_ADD:
		__futex_atomic_op("lr %2,%1\nar %2,%5\n",
				  ret, oldval, newval, uaddr, oparg);
		break;
	case FUTEX_OP_OR:
		__futex_atomic_op("lr %2,%1\nor %2,%5\n",
				  ret, oldval, newval, uaddr, oparg);
		break;
	case FUTEX_OP_ANDN:
		__futex_atomic_op("lr %2,%1\nnr %2,%5\n",
				  ret, oldval, newval, uaddr, oparg);
		break;
	case FUTEX_OP_XOR:
		__futex_atomic_op("lr %2,%1\nxr %2,%5\n",
				  ret, oldval, newval, uaddr, oparg);
		break;
	default:
		ret = -ENOSYS;
	}
	put_page(virt_to_page(uaddr));
	*old = oldval;
	return ret;
}

int futex_atomic_cmpxchg_pt(int __user *uaddr, int oldval, int newval)
{
	int ret;

	spin_lock(&current->mm->page_table_lock);
	uaddr = (int __user *) __dat_user_addr((unsigned long) uaddr);
	if (!uaddr) {
		spin_unlock(&current->mm->page_table_lock);
		return -EFAULT;
	}
	get_page(virt_to_page(uaddr));
	spin_unlock(&current->mm->page_table_lock);
	asm volatile("   cs   %1,%4,0(%5)\n"
		     "0: lr   %0,%1\n"
		     "1:\n"
		     EX_TABLE(0b,1b)
		     : "=d" (ret), "+d" (oldval), "=m" (*uaddr)
		     : "0" (-EFAULT), "d" (newval), "a" (uaddr), "m" (*uaddr)
		     : "cc", "memory" );
	put_page(virt_to_page(uaddr));
	return ret;
}

struct uaccess_ops uaccess_pt = {
	.copy_from_user		= copy_from_user_pt,
	.copy_from_user_small	= copy_from_user_pt,
	.copy_to_user		= copy_to_user_pt,
	.copy_to_user_small	= copy_to_user_pt,
	.copy_in_user		= copy_in_user_pt,
	.clear_user		= clear_user_pt,
	.strnlen_user		= strnlen_user_pt,
	.strncpy_from_user	= strncpy_from_user_pt,
	.futex_atomic_op	= futex_atomic_op_pt,
	.futex_atomic_cmpxchg	= futex_atomic_cmpxchg_pt,
};
Commit	Line	Data
59f35d53 GS	1	/*
	2	* arch/s390/lib/uaccess_pt.c
	3	*
c1821c2e GS	4	* User access functions based on page table walks for enhanced
c1821c2e GS	5	* system layout without hardware support.
59f35d53 GS	6	*
	7	* Copyright IBM Corp. 2006
	8	* Author(s): Gerald Schaefer (gerald.schaefer@de.ibm.com)
	9	*/
	10
	11	#include <linux/errno.h>
d8ad075e	12	#include <linux/hardirq.h>
59f35d53	13	#include <linux/mm.h>
22155914	14	#include <asm/uaccess.h>
59f35d53	15	#include <asm/futex.h>
2b67fc46	16	#include "uaccess.h"
59f35d53	17
4d284cac HC	18	static int __handle_fault(struct mm_struct *mm, unsigned long address,
4d284cac HC	19	int write_access)
59f35d53 GS	20	{
	21	struct vm_area_struct *vma;
	22	int ret = -EFAULT;
83c54070	23	int fault;
59f35d53	24
d8ad075e HC	25	if (in_atomic())
d8ad075e HC	26	return ret;
59f35d53 GS	27	down_read(&mm->mmap_sem);
	28	vma = find_vma(mm, address);
	29	if (unlikely(!vma))
	30	goto out;
	31	if (unlikely(vma->vm_start > address)) {
	32	if (!(vma->vm_flags & VM_GROWSDOWN))
	33	goto out;
	34	if (expand_stack(vma, address))
	35	goto out;
	36	}
	37
	38	if (!write_access) {
	39	/* page not present, check vm flags */
	40	if (!(vma->vm_flags & (VM_READ \| VM_EXEC \| VM_WRITE)))
	41	goto out;
	42	} else {
	43	if (!(vma->vm_flags & VM_WRITE))
	44	goto out;
	45	}
	46
	47	survive:
83c54070 NP	48	fault = handle_mm_fault(mm, vma, address, write_access);
	49	if (unlikely(fault & VM_FAULT_ERROR)) {
	50	if (fault & VM_FAULT_OOM)
	51	goto out_of_memory;
	52	else if (fault & VM_FAULT_SIGBUS)
	53	goto out_sigbus;
59f35d53 GS	54	BUG();
59f35d53 GS	55	}
83c54070 NP	56	if (fault & VM_FAULT_MAJOR)
	57	current->maj_flt++;
	58	else
	59	current->min_flt++;
59f35d53 GS	60	ret = 0;
	61	out:
	62	up_read(&mm->mmap_sem);
	63	return ret;
	64
	65	out_of_memory:
	66	up_read(&mm->mmap_sem);
b460cbc5	67	if (is_global_init(current)) {
59f35d53	68	yield();
22155914	69	down_read(&mm->mmap_sem);
59f35d53 GS	70	goto survive;
	71	}
	72	printk("VM: killing process %s\n", current->comm);
	73	return ret;
	74
	75	out_sigbus:
	76	up_read(&mm->mmap_sem);
	77	current->thread.prot_addr = address;
	78	current->thread.trap_no = 0x11;
	79	force_sig(SIGBUS, current);
	80	return ret;
	81	}
	82
4d284cac HC	83	static size_t __user_copy_pt(unsigned long uaddr, void *kptr,
4d284cac HC	84	size_t n, int write_user)
59f35d53 GS	85	{
	86	struct mm_struct *mm = current->mm;
	87	unsigned long offset, pfn, done, size;
	88	pgd_t *pgd;
	89	pmd_t *pmd;
	90	pte_t *pte;
	91	void from, to;
	92
	93	done = 0;
	94	retry:
	95	spin_lock(&mm->page_table_lock);
	96	do {
	97	pgd = pgd_offset(mm, uaddr);
	98	if (pgd_none(pgd) \|\| unlikely(pgd_bad(pgd)))
	99	goto fault;
	100
	101	pmd = pmd_offset(pgd, uaddr);
	102	if (pmd_none(pmd) \|\| unlikely(pmd_bad(pmd)))
	103	goto fault;
	104
	105	pte = pte_offset_map(pmd, uaddr);
	106	if (!pte \|\| !pte_present(*pte) \|\|
	107	(write_user && !pte_write(*pte)))
	108	goto fault;
	109
	110	pfn = pte_pfn(*pte);
	111	if (!pfn_valid(pfn))
	112	goto out;
	113
	114	offset = uaddr & (PAGE_SIZE - 1);
	115	size = min(n - done, PAGE_SIZE - offset);
	116	if (write_user) {
	117	to = (void *)((pfn << PAGE_SHIFT) + offset);
	118	from = kptr + done;
	119	} else {
	120	from = (void *)((pfn << PAGE_SHIFT) + offset);
	121	to = kptr + done;
	122	}
	123	memcpy(to, from, size);
	124	done += size;
	125	uaddr += size;
	126	} while (done < n);
	127	out:
	128	spin_unlock(&mm->page_table_lock);
	129	return n - done;
	130	fault:
	131	spin_unlock(&mm->page_table_lock);
	132	if (__handle_fault(mm, uaddr, write_user))
	133	return n - done;
	134	goto retry;
	135	}
	136
c1821c2e GS	137	/*
	138	* Do DAT for user address by page table walk, return kernel address.
	139	* This function needs to be called with current->mm->page_table_lock held.
	140	*/
4d284cac	141	static unsigned long __dat_user_addr(unsigned long uaddr)
c1821c2e GS	142	{
	143	struct mm_struct *mm = current->mm;
	144	unsigned long pfn, ret;
	145	pgd_t *pgd;
	146	pmd_t *pmd;
	147	pte_t *pte;
	148	int rc;
	149
	150	ret = 0;
	151	retry:
	152	pgd = pgd_offset(mm, uaddr);
	153	if (pgd_none(pgd) \|\| unlikely(pgd_bad(pgd)))
	154	goto fault;
	155
	156	pmd = pmd_offset(pgd, uaddr);
	157	if (pmd_none(pmd) \|\| unlikely(pmd_bad(pmd)))
	158	goto fault;
	159
	160	pte = pte_offset_map(pmd, uaddr);
	161	if (!pte \|\| !pte_present(*pte))
	162	goto fault;
	163
	164	pfn = pte_pfn(*pte);
	165	if (!pfn_valid(pfn))
	166	goto out;
	167
	168	ret = (pfn << PAGE_SHIFT) + (uaddr & (PAGE_SIZE - 1));
	169	out:
	170	return ret;
	171	fault:
	172	spin_unlock(&mm->page_table_lock);
	173	rc = __handle_fault(mm, uaddr, 0);
	174	spin_lock(&mm->page_table_lock);
	175	if (rc)
	176	goto out;
	177	goto retry;
	178	}
	179
59f35d53 GS	180	size_t copy_from_user_pt(size_t n, const void __user from, void to)
	181	{
	182	size_t rc;
	183
	184	if (segment_eq(get_fs(), KERNEL_DS)) {
	185	memcpy(to, (void __kernel __force *) from, n);
	186	return 0;
	187	}
	188	rc = __user_copy_pt((unsigned long) from, to, n, 0);
	189	if (unlikely(rc))
	190	memset(to + n - rc, 0, rc);
	191	return rc;
	192	}
	193
	194	size_t copy_to_user_pt(size_t n, void __user to, const void from)
	195	{
	196	if (segment_eq(get_fs(), KERNEL_DS)) {
	197	memcpy((void __kernel __force *) to, from, n);
	198	return 0;
	199	}
	200	return __user_copy_pt((unsigned long) to, (void *) from, n, 1);
	201	}
c1821c2e GS	202
	203	static size_t clear_user_pt(size_t n, void __user *to)
	204	{
	205	long done, size, ret;
	206
	207	if (segment_eq(get_fs(), KERNEL_DS)) {
	208	memset((void __kernel __force *) to, 0, n);
	209	return 0;
	210	}
	211	done = 0;
	212	do {
	213	if (n - done > PAGE_SIZE)
	214	size = PAGE_SIZE;
	215	else
	216	size = n - done;
	217	ret = __user_copy_pt((unsigned long) to + done,
	218	&empty_zero_page, size, 1);
	219	done += size;
	220	if (ret)
	221	return ret + n - done;
	222	} while (done < n);
	223	return 0;
	224	}
	225
	226	static size_t strnlen_user_pt(size_t count, const char __user *src)
	227	{
	228	char *addr;
	229	unsigned long uaddr = (unsigned long) src;
	230	struct mm_struct *mm = current->mm;
	231	unsigned long offset, pfn, done, len;
	232	pgd_t *pgd;
	233	pmd_t *pmd;
	234	pte_t *pte;
	235	size_t len_str;
	236
	237	if (segment_eq(get_fs(), KERNEL_DS))
	238	return strnlen((const char __kernel __force *) src, count) + 1;
	239	done = 0;
	240	retry:
	241	spin_lock(&mm->page_table_lock);
	242	do {
	243	pgd = pgd_offset(mm, uaddr);
	244	if (pgd_none(pgd) \|\| unlikely(pgd_bad(pgd)))
	245	goto fault;
	246
	247	pmd = pmd_offset(pgd, uaddr);
	248	if (pmd_none(pmd) \|\| unlikely(pmd_bad(pmd)))
	249	goto fault;
	250
	251	pte = pte_offset_map(pmd, uaddr);
	252	if (!pte \|\| !pte_present(*pte))
	253	goto fault;
	254
	255	pfn = pte_pfn(*pte);
	256	if (!pfn_valid(pfn)) {
	257	done = -1;
	258	goto out;
	259	}
	260
	261	offset = uaddr & (PAGE_SIZE-1);
	262	addr = (char *)(pfn << PAGE_SHIFT) + offset;
	263	len = min(count - done, PAGE_SIZE - offset);
	264	len_str = strnlen(addr, len);
	265	done += len_str;
266	uaddr += len_str;
267	} while ((len_str == len) && (done < count));
268	out:
269	spin_unlock(&mm->page_table_lock);
270	return done + 1;
271	fault:
272	spin_unlock(&mm->page_table_lock);
273	if (__handle_fault(mm, uaddr, 0)) {
274	return 0;
275	}
276	goto retry;
277	}
278
279	static size_t strncpy_from_user_pt(size_t count, const char __user *src,
280	char *dst)
281	{
282	size_t n = strnlen_user_pt(count, src);
283
284	if (!n)
285	return -EFAULT;
286	if (n > count)
287	n = count;
288	if (segment_eq(get_fs(), KERNEL_DS)) {
289	memcpy(dst, (const char __kernel __force *) src, n);
290	if (dst[n-1] == '\0')
291	return n-1;
292	else
293	return n;
294	}
295	if (__user_copy_pt((unsigned long) src, dst, n, 0))
296	return -EFAULT;
297	if (dst[n-1] == '\0')
298	return n-1;
299	else
300	return n;
301	}
302
303	static size_t copy_in_user_pt(size_t n, void __user *to,
304	const void __user *from)
305	{
306	struct mm_struct *mm = current->mm;
307	unsigned long offset_from, offset_to, offset_max, pfn_from, pfn_to,
308	uaddr, done, size;
309	unsigned long uaddr_from = (unsigned long) from;
310	unsigned long uaddr_to = (unsigned long) to;
311	pgd_t pgd_from, pgd_to;
312	pmd_t pmd_from, pmd_to;
313	pte_t pte_from, pte_to;
314	int write_user;
315
316	done = 0;
317	retry:
318	spin_lock(&mm->page_table_lock);
319	do {
320	pgd_from = pgd_offset(mm, uaddr_from);
321	if (pgd_none(pgd_from) \|\| unlikely(pgd_bad(pgd_from))) {
322	uaddr = uaddr_from;
323	write_user = 0;
324	goto fault;
325	}
326	pgd_to = pgd_offset(mm, uaddr_to);
327	if (pgd_none(pgd_to) \|\| unlikely(pgd_bad(pgd_to))) {
328	uaddr = uaddr_to;
329	write_user = 1;
330	goto fault;
331	}
332
333	pmd_from = pmd_offset(pgd_from, uaddr_from);
334	if (pmd_none(pmd_from) \|\| unlikely(pmd_bad(pmd_from))) {
335	uaddr = uaddr_from;
336	write_user = 0;
337	goto fault;
338	}
339	pmd_to = pmd_offset(pgd_to, uaddr_to);
340	if (pmd_none(pmd_to) \|\| unlikely(pmd_bad(pmd_to))) {
341	uaddr = uaddr_to;
342	write_user = 1;
343	goto fault;
344	}
345
346	pte_from = pte_offset_map(pmd_from, uaddr_from);
347	if (!pte_from \|\| !pte_present(*pte_from)) {
348	uaddr = uaddr_from;
349	write_user = 0;
350	goto fault;
351	}
352	pte_to = pte_offset_map(pmd_to, uaddr_to);
353	if (!pte_to \|\| !pte_present(pte_to) \|\| !pte_write(pte_to)) {
354	uaddr = uaddr_to;
355	write_user = 1;
356	goto fault;
357	}
358
359	pfn_from = pte_pfn(*pte_from);
360	if (!pfn_valid(pfn_from))
361	goto out;
362	pfn_to = pte_pfn(*pte_to);
363	if (!pfn_valid(pfn_to))
364	goto out;
365
366	offset_from = uaddr_from & (PAGE_SIZE-1);
367	offset_to = uaddr_from & (PAGE_SIZE-1);
368	offset_max = max(offset_from, offset_to);
369	size = min(n - done, PAGE_SIZE - offset_max);
370
371	memcpy((void *)(pfn_to << PAGE_SHIFT) + offset_to,
372	(void *)(pfn_from << PAGE_SHIFT) + offset_from, size);
373	done += size;
374	uaddr_from += size;
375	uaddr_to += size;
376	} while (done < n);
377	out:
378	spin_unlock(&mm->page_table_lock);
379	return n - done;
380	fault:
381	spin_unlock(&mm->page_table_lock);
382	if (__handle_fault(mm, uaddr, write_user))
383	return n - done;
384	goto retry;
385	}
386
387	#define __futex_atomic_op(insn, ret, oldval, newval, uaddr, oparg) \
388	asm volatile("0: l %1,0(%6)\n" \
389	"1: " insn \
390	"2: cs %1,%2,0(%6)\n" \
391	"3: jl 1b\n" \
392	" lhi %0,0\n" \
393	"4:\n" \
394	EX_TABLE(0b,4b) EX_TABLE(2b,4b) EX_TABLE(3b,4b) \
395	: "=d" (ret), "=&d" (oldval), "=&d" (newval), \
396	"=m" (*uaddr) \
397	: "0" (-EFAULT), "d" (oparg), "a" (uaddr), \
398	"m" (*uaddr) : "cc" );
399
400	int futex_atomic_op_pt(int op, int __user uaddr, int oparg, int old)
401	{
402	int oldval = 0, newval, ret;
403
404	spin_lock(&current->mm->page_table_lock);
405	uaddr = (int __user *) __dat_user_addr((unsigned long) uaddr);
406	if (!uaddr) {
407	spin_unlock(&current->mm->page_table_lock);
408	return -EFAULT;
409	}
410	get_page(virt_to_page(uaddr));
411	spin_unlock(&current->mm->page_table_lock);
412	switch (op) {
413	case FUTEX_OP_SET:
414	__futex_atomic_op("lr %2,%5\n",
415	ret, oldval, newval, uaddr, oparg);
416	break;
417	case FUTEX_OP_ADD:
418	__futex_atomic_op("lr %2,%1\nar %2,%5\n",
419	ret, oldval, newval, uaddr, oparg);
420	break;
421	case FUTEX_OP_OR:
422	__futex_atomic_op("lr %2,%1\nor %2,%5\n",
423	ret, oldval, newval, uaddr, oparg);
424	break;
425	case FUTEX_OP_ANDN:
426	__futex_atomic_op("lr %2,%1\nnr %2,%5\n",
427	ret, oldval, newval, uaddr, oparg);
428	break;
429	case FUTEX_OP_XOR:
430	__futex_atomic_op("lr %2,%1\nxr %2,%5\n",
431	ret, oldval, newval, uaddr, oparg);
432	break;
433	default:
434	ret = -ENOSYS;
435	}
436	put_page(virt_to_page(uaddr));
437	*old = oldval;
438	return ret;
439	}
440
441	int futex_atomic_cmpxchg_pt(int __user *uaddr, int oldval, int newval)
442	{
443	int ret;
444
445	spin_lock(&current->mm->page_table_lock);
446	uaddr = (int __user *) __dat_user_addr((unsigned long) uaddr);
447	if (!uaddr) {
448	spin_unlock(&current->mm->page_table_lock);
449	return -EFAULT;
450	}
451	get_page(virt_to_page(uaddr));
452	spin_unlock(&current->mm->page_table_lock);
453	asm volatile(" cs %1,%4,0(%5)\n"
454	"0: lr %0,%1\n"
455	"1:\n"
456	EX_TABLE(0b,1b)
457	: "=d" (ret), "+d" (oldval), "=m" (*uaddr)
458	: "0" (-EFAULT), "d" (newval), "a" (uaddr), "m" (*uaddr)
459	: "cc", "memory" );
460	put_page(virt_to_page(uaddr));
461	return ret;
462	}
463
464	struct uaccess_ops uaccess_pt = {
465	.copy_from_user = copy_from_user_pt,
466	.copy_from_user_small = copy_from_user_pt,
467	.copy_to_user = copy_to_user_pt,
468	.copy_to_user_small = copy_to_user_pt,
469	.copy_in_user = copy_in_user_pt,
470	.clear_user = clear_user_pt,
471	.strnlen_user = strnlen_user_pt,
472	.strncpy_from_user = strncpy_from_user_pt,
473	.futex_atomic_op = futex_atomic_op_pt,
474	.futex_atomic_cmpxchg = futex_atomic_cmpxchg_pt,
475	};