2 * User access functions based on page table walks for enhanced
3 * system layout without hardware support.
5 * Copyright IBM Corp. 2006, 2012
6 * Author(s): Gerald Schaefer (gerald.schaefer@de.ibm.com)
9 #include <linux/errno.h>
10 #include <linux/hardirq.h>
12 #include <linux/hugetlb.h>
13 #include <asm/uaccess.h>
14 #include <asm/futex.h>
25 static size_t strnlen_kernel(size_t count, const char __user *src)
27 register unsigned long reg0 asm("0") = 0UL;
28 unsigned long tmp1, tmp2;
36 " la %0,1(%3)\n" /* strnlen_kernel results includes \0 */
40 : "+a" (count), "+a" (src), "=a" (tmp1), "=a" (tmp2)
41 : "d" (reg0) : "cc", "memory");
45 static size_t copy_in_kernel(size_t count, void __user *to,
46 const void __user *from)
55 "1: mvc 0(1,%1),0(%2)\n"
61 "2: mvc 0(256,%1),0(%2)\n"
66 "4: ex %0,1b-0b(%3)\n"
69 EX_TABLE(1b,6b) EX_TABLE(2b,0b) EX_TABLE(4b,0b)
70 : "+a" (count), "+a" (to), "+a" (from), "=a" (tmp1)
76 * Returns kernel address for user virtual address. If the returned address is
77 * >= -4095 (IS_ERR_VALUE(x) returns true), a fault has occured and the address
78 * contains the (negative) exception code.
81 static unsigned long follow_table(struct mm_struct *mm,
82 unsigned long address, int write)
84 unsigned long *table = (unsigned long *)__pa(mm->pgd);
86 switch (mm->context.asce_bits & _ASCE_TYPE_MASK) {
87 case _ASCE_TYPE_REGION1:
88 table = table + ((address >> 53) & 0x7ff);
89 if (unlikely(*table & _REGION_ENTRY_INVALID))
91 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
93 case _ASCE_TYPE_REGION2:
94 table = table + ((address >> 42) & 0x7ff);
95 if (unlikely(*table & _REGION_ENTRY_INVALID))
97 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
99 case _ASCE_TYPE_REGION3:
100 table = table + ((address >> 31) & 0x7ff);
101 if (unlikely(*table & _REGION_ENTRY_INVALID))
103 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
105 case _ASCE_TYPE_SEGMENT:
106 table = table + ((address >> 20) & 0x7ff);
107 if (unlikely(*table & _SEGMENT_ENTRY_INVALID))
109 if (unlikely(*table & _SEGMENT_ENTRY_LARGE)) {
110 if (write && (*table & _SEGMENT_ENTRY_PROTECT))
112 return (*table & _SEGMENT_ENTRY_ORIGIN_LARGE) +
113 (address & ~_SEGMENT_ENTRY_ORIGIN_LARGE);
115 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
117 table = table + ((address >> 12) & 0xff);
118 if (unlikely(*table & _PAGE_INVALID))
120 if (write && (*table & _PAGE_PROTECT))
122 return (*table & PAGE_MASK) + (address & ~PAGE_MASK);
125 #else /* CONFIG_64BIT */
127 static unsigned long follow_table(struct mm_struct *mm,
128 unsigned long address, int write)
130 unsigned long *table = (unsigned long *)__pa(mm->pgd);
132 table = table + ((address >> 20) & 0x7ff);
133 if (unlikely(*table & _SEGMENT_ENTRY_INVALID))
135 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
136 table = table + ((address >> 12) & 0xff);
137 if (unlikely(*table & _PAGE_INVALID))
139 if (write && (*table & _PAGE_PROTECT))
141 return (*table & PAGE_MASK) + (address & ~PAGE_MASK);
144 #endif /* CONFIG_64BIT */
146 static __always_inline size_t __user_copy_pt(unsigned long uaddr, void *kptr,
147 size_t n, int write_user)
149 struct mm_struct *mm = current->mm;
150 unsigned long offset, done, size, kaddr;
155 spin_lock(&mm->page_table_lock);
157 kaddr = follow_table(mm, uaddr, write_user);
158 if (IS_ERR_VALUE(kaddr))
161 offset = uaddr & ~PAGE_MASK;
162 size = min(n - done, PAGE_SIZE - offset);
167 from = (void *) kaddr;
170 memcpy(to, from, size);
174 spin_unlock(&mm->page_table_lock);
177 spin_unlock(&mm->page_table_lock);
178 if (__handle_fault(uaddr, -kaddr, write_user))
184 * Do DAT for user address by page table walk, return kernel address.
185 * This function needs to be called with current->mm->page_table_lock held.
187 static __always_inline unsigned long __dat_user_addr(unsigned long uaddr,
190 struct mm_struct *mm = current->mm;
195 kaddr = follow_table(mm, uaddr, write);
196 if (IS_ERR_VALUE(kaddr))
201 spin_unlock(&mm->page_table_lock);
202 rc = __handle_fault(uaddr, -kaddr, write);
203 spin_lock(&mm->page_table_lock);
209 size_t copy_from_user_pt(size_t n, const void __user *from, void *to)
213 if (segment_eq(get_fs(), KERNEL_DS))
214 return copy_in_kernel(n, (void __user *) to, from);
215 rc = __user_copy_pt((unsigned long) from, to, n, 0);
217 memset(to + n - rc, 0, rc);
221 size_t copy_to_user_pt(size_t n, void __user *to, const void *from)
223 if (segment_eq(get_fs(), KERNEL_DS))
224 return copy_in_kernel(n, to, (void __user *) from);
225 return __user_copy_pt((unsigned long) to, (void *) from, n, 1);
228 static size_t clear_user_pt(size_t n, void __user *to)
230 void *zpage = (void *) empty_zero_page;
231 long done, size, ret;
235 if (n - done > PAGE_SIZE)
239 if (segment_eq(get_fs(), KERNEL_DS))
240 ret = copy_in_kernel(n, to, (void __user *) zpage);
242 ret = __user_copy_pt((unsigned long) to, zpage, size, 1);
246 return ret + n - done;
251 static size_t strnlen_user_pt(size_t count, const char __user *src)
253 unsigned long uaddr = (unsigned long) src;
254 struct mm_struct *mm = current->mm;
255 unsigned long offset, done, len, kaddr;
258 if (unlikely(!count))
260 if (segment_eq(get_fs(), KERNEL_DS))
261 return strnlen_kernel(count, src);
264 spin_lock(&mm->page_table_lock);
266 kaddr = follow_table(mm, uaddr, 0);
267 if (IS_ERR_VALUE(kaddr))
270 offset = uaddr & ~PAGE_MASK;
271 len = min(count - done, PAGE_SIZE - offset);
272 len_str = strnlen((char *) kaddr, len);
275 } while ((len_str == len) && (done < count));
276 spin_unlock(&mm->page_table_lock);
279 spin_unlock(&mm->page_table_lock);
280 if (__handle_fault(uaddr, -kaddr, 0))
285 static size_t strncpy_from_user_pt(size_t count, const char __user *src,
288 size_t done, len, offset, len_str;
290 if (unlikely(!count))
294 offset = (size_t)src & ~PAGE_MASK;
295 len = min(count - done, PAGE_SIZE - offset);
296 if (segment_eq(get_fs(), KERNEL_DS)) {
297 if (copy_in_kernel(len, (void __user *) dst, src))
300 if (__user_copy_pt((unsigned long) src, dst, len, 0))
303 len_str = strnlen(dst, len);
307 } while ((len_str == len) && (done < count));
311 static size_t copy_in_user_pt(size_t n, void __user *to,
312 const void __user *from)
314 struct mm_struct *mm = current->mm;
315 unsigned long offset_max, uaddr, done, size, error_code;
316 unsigned long uaddr_from = (unsigned long) from;
317 unsigned long uaddr_to = (unsigned long) to;
318 unsigned long kaddr_to, kaddr_from;
321 if (segment_eq(get_fs(), KERNEL_DS))
322 return copy_in_kernel(n, to, from);
325 spin_lock(&mm->page_table_lock);
329 kaddr_from = follow_table(mm, uaddr_from, 0);
330 error_code = kaddr_from;
331 if (IS_ERR_VALUE(error_code))
336 kaddr_to = follow_table(mm, uaddr_to, 1);
337 error_code = (unsigned long) kaddr_to;
338 if (IS_ERR_VALUE(error_code))
341 offset_max = max(uaddr_from & ~PAGE_MASK,
342 uaddr_to & ~PAGE_MASK);
343 size = min(n - done, PAGE_SIZE - offset_max);
345 memcpy((void *) kaddr_to, (void *) kaddr_from, size);
350 spin_unlock(&mm->page_table_lock);
353 spin_unlock(&mm->page_table_lock);
354 if (__handle_fault(uaddr, -error_code, write_user))
359 #define __futex_atomic_op(insn, ret, oldval, newval, uaddr, oparg) \
360 asm volatile("0: l %1,0(%6)\n" \
362 "2: cs %1,%2,0(%6)\n" \
366 EX_TABLE(0b,4b) EX_TABLE(2b,4b) EX_TABLE(3b,4b) \
367 : "=d" (ret), "=&d" (oldval), "=&d" (newval), \
369 : "0" (-EFAULT), "d" (oparg), "a" (uaddr), \
370 "m" (*uaddr) : "cc" );
372 static int __futex_atomic_op_pt(int op, u32 __user *uaddr, int oparg, int *old)
374 int oldval = 0, newval, ret;
378 __futex_atomic_op("lr %2,%5\n",
379 ret, oldval, newval, uaddr, oparg);
382 __futex_atomic_op("lr %2,%1\nar %2,%5\n",
383 ret, oldval, newval, uaddr, oparg);
386 __futex_atomic_op("lr %2,%1\nor %2,%5\n",
387 ret, oldval, newval, uaddr, oparg);
390 __futex_atomic_op("lr %2,%1\nnr %2,%5\n",
391 ret, oldval, newval, uaddr, oparg);
394 __futex_atomic_op("lr %2,%1\nxr %2,%5\n",
395 ret, oldval, newval, uaddr, oparg);
405 int futex_atomic_op_pt(int op, u32 __user *uaddr, int oparg, int *old)
409 if (segment_eq(get_fs(), KERNEL_DS))
410 return __futex_atomic_op_pt(op, uaddr, oparg, old);
411 spin_lock(¤t->mm->page_table_lock);
412 uaddr = (u32 __force __user *)
413 __dat_user_addr((__force unsigned long) uaddr, 1);
415 spin_unlock(¤t->mm->page_table_lock);
418 get_page(virt_to_page(uaddr));
419 spin_unlock(¤t->mm->page_table_lock);
420 ret = __futex_atomic_op_pt(op, uaddr, oparg, old);
421 put_page(virt_to_page(uaddr));
425 static int __futex_atomic_cmpxchg_pt(u32 *uval, u32 __user *uaddr,
426 u32 oldval, u32 newval)
430 asm volatile("0: cs %1,%4,0(%5)\n"
433 EX_TABLE(0b,2b) EX_TABLE(1b,2b)
434 : "=d" (ret), "+d" (oldval), "=m" (*uaddr)
435 : "0" (-EFAULT), "d" (newval), "a" (uaddr), "m" (*uaddr)
441 int futex_atomic_cmpxchg_pt(u32 *uval, u32 __user *uaddr,
442 u32 oldval, u32 newval)
446 if (segment_eq(get_fs(), KERNEL_DS))
447 return __futex_atomic_cmpxchg_pt(uval, uaddr, oldval, newval);
448 spin_lock(¤t->mm->page_table_lock);
449 uaddr = (u32 __force __user *)
450 __dat_user_addr((__force unsigned long) uaddr, 1);
452 spin_unlock(¤t->mm->page_table_lock);
455 get_page(virt_to_page(uaddr));
456 spin_unlock(¤t->mm->page_table_lock);
457 ret = __futex_atomic_cmpxchg_pt(uval, uaddr, oldval, newval);
458 put_page(virt_to_page(uaddr));
462 struct uaccess_ops uaccess_pt = {
463 .copy_from_user = copy_from_user_pt,
464 .copy_to_user = copy_to_user_pt,
465 .copy_in_user = copy_in_user_pt,
466 .clear_user = clear_user_pt,
467 .strnlen_user = strnlen_user_pt,
468 .strncpy_from_user = strncpy_from_user_pt,
469 .futex_atomic_op = futex_atomic_op_pt,
470 .futex_atomic_cmpxchg = futex_atomic_cmpxchg_pt,