arch/x86/lib/usercopy_64.c

   1 /*
   2  * User address space access functions.
   3  *
   4  * Copyright 1997 Andi Kleen <ak@muc.de>
   5  * Copyright 1997 Linus Torvalds
   6  * Copyright 2002 Andi Kleen <ak@suse.de>
   7  */
   8 #include <linux/export.h>
   9 #include <linux/uaccess.h>
  10 #include <linux/highmem.h>
  11
  12 /*
  13  * Zero Userspace
  14  */
  15
  16 unsigned long __clear_user(void __user *addr, unsigned long size)
  17 {
  18         long __d0;
  19         might_fault();
  20         /* no memory constraint because it doesn't change any memory gcc knows
  21            about */
  22         stac();
  23         asm volatile(
  24                 "       testq  %[size8],%[size8]\n"
  25                 "       jz     4f\n"
  26                 "0:     movq $0,(%[dst])\n"
  27                 "       addq   $8,%[dst]\n"
  28                 "       decl %%ecx ; jnz   0b\n"
  29                 "4:     movq  %[size1],%%rcx\n"
  30                 "       testl %%ecx,%%ecx\n"
  31                 "       jz     2f\n"
  32                 "1:     movb   $0,(%[dst])\n"
  33                 "       incq   %[dst]\n"
  34                 "       decl %%ecx ; jnz  1b\n"
  35                 "2:\n"
  36                 ".section .fixup,\"ax\"\n"
  37                 "3:     lea 0(%[size1],%[size8],8),%[size8]\n"
  38                 "       jmp 2b\n"
  39                 ".previous\n"
  40                 _ASM_EXTABLE_UA(0b, 3b)
  41                 _ASM_EXTABLE_UA(1b, 2b)
  42                 : [size8] "=&c"(size), [dst] "=&D" (__d0)
  43                 : [size1] "r"(size & 7), "[size8]" (size / 8), "[dst]"(addr));
  44         clac();
  45         return size;
  46 }
  47 EXPORT_SYMBOL(__clear_user);
  48
  49 unsigned long clear_user(void __user *to, unsigned long n)
  50 {
  51         if (access_ok(VERIFY_WRITE, to, n))
  52                 return __clear_user(to, n);
  53         return n;
  54 }
  55 EXPORT_SYMBOL(clear_user);
  56
  57 /*
  58  * Try to copy last bytes and clear the rest if needed.
  59  * Since protection fault in copy_from/to_user is not a normal situation,
  60  * it is not necessary to optimize tail handling.
  61  */
  62 __visible unsigned long
  63 copy_user_handle_tail(char *to, char *from, unsigned len)
  64 {
  65         for (; len; --len, to++) {
  66                 char c;
  67
  68                 if (__get_user_nocheck(c, from++, sizeof(char)))
  69                         break;
  70                 if (__put_user_nocheck(c, to, sizeof(char)))
  71                         break;
  72         }
  73         clac();
  74         return len;
  75 }
  76
  77 /*
  78  * Similar to copy_user_handle_tail, probe for the write fault point,
  79  * but reuse __memcpy_mcsafe in case a new read error is encountered.
  80  * clac() is handled in _copy_to_iter_mcsafe().
  81  */
  82 __visible unsigned long
  83 mcsafe_handle_tail(char *to, char *from, unsigned len)
  84 {
  85         for (; len; --len, to++, from++) {
  86                 /*
  87                  * Call the assembly routine back directly since
  88                  * memcpy_mcsafe() may silently fallback to memcpy.
  89                  */
  90                 unsigned long rem = __memcpy_mcsafe(to, from, 1);
  91
  92                 if (rem)
  93                         break;
  94         }
  95         return len;
  96 }
  97
  98 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
  99 /**
 100  * clean_cache_range - write back a cache range with CLWB
 101  * @vaddr:      virtual start address
 102  * @size:       number of bytes to write back
 103  *
 104  * Write back a cache range using the CLWB (cache line write back)
 105  * instruction. Note that @size is internally rounded up to be cache
 106  * line size aligned.
 107  */
 108 static void clean_cache_range(void *addr, size_t size)
 109 {
 110         u16 x86_clflush_size = boot_cpu_data.x86_clflush_size;
 111         unsigned long clflush_mask = x86_clflush_size - 1;
 112         void *vend = addr + size;
 113         void *p;
 114
 115         for (p = (void *)((unsigned long)addr & ~clflush_mask);
 116              p < vend; p += x86_clflush_size)
 117                 clwb(p);
 118 }
 119
 120 void arch_wb_cache_pmem(void *addr, size_t size)
 121 {
 122         clean_cache_range(addr, size);
 123 }
 124 EXPORT_SYMBOL_GPL(arch_wb_cache_pmem);
 125
 126 long __copy_user_flushcache(void *dst, const void __user *src, unsigned size)
 127 {
 128         unsigned long flushed, dest = (unsigned long) dst;
 129         long rc = __copy_user_nocache(dst, src, size, 0);
 130
 131         /*
 132          * __copy_user_nocache() uses non-temporal stores for the bulk
 133          * of the transfer, but we need to manually flush if the
 134          * transfer is unaligned. A cached memory copy is used when
 135          * destination or size is not naturally aligned. That is:
 136          *   - Require 8-byte alignment when size is 8 bytes or larger.
 137          *   - Require 4-byte alignment when size is 4 bytes.
 138          */
 139         if (size < 8) {
 140                 if (!IS_ALIGNED(dest, 4) || size != 4)
 141                         clean_cache_range(dst, 1);
 142         } else {
 143                 if (!IS_ALIGNED(dest, 8)) {
 144                         dest = ALIGN(dest, boot_cpu_data.x86_clflush_size);
 145                         clean_cache_range(dst, 1);
 146                 }
 147
 148                 flushed = dest - (unsigned long) dst;
 149                 if (size > flushed && !IS_ALIGNED(size - flushed, 8))
 150                         clean_cache_range(dst + size - 1, 1);
 151         }
 152
 153         return rc;
 154 }
 155
 156 void __memcpy_flushcache(void *_dst, const void *_src, size_t size)
 157 {
 158         unsigned long dest = (unsigned long) _dst;
 159         unsigned long source = (unsigned long) _src;
 160
 161         /* cache copy and flush to align dest */
 162         if (!IS_ALIGNED(dest, 8)) {
 163                 unsigned len = min_t(unsigned, size, ALIGN(dest, 8) - dest);
 164
 165                 memcpy((void *) dest, (void *) source, len);
 166                 clean_cache_range((void *) dest, len);
 167                 dest += len;
 168                 source += len;
 169                 size -= len;
 170                 if (!size)
 171                         return;
 172         }
 173
 174         /* 4x8 movnti loop */
 175         while (size >= 32) {
 176                 asm("movq    (%0), %%r8\n"
 177                     "movq   8(%0), %%r9\n"
 178                     "movq  16(%0), %%r10\n"
 179                     "movq  24(%0), %%r11\n"
 180                     "movnti  %%r8,   (%1)\n"
 181                     "movnti  %%r9,  8(%1)\n"
 182                     "movnti %%r10, 16(%1)\n"
 183                     "movnti %%r11, 24(%1)\n"
 184                     :: "r" (source), "r" (dest)
 185                     : "memory", "r8", "r9", "r10", "r11");
 186                 dest += 32;
 187                 source += 32;
 188                 size -= 32;
 189         }
 190
 191         /* 1x8 movnti loop */
 192         while (size >= 8) {
 193                 asm("movq    (%0), %%r8\n"
 194                     "movnti  %%r8,   (%1)\n"
 195                     :: "r" (source), "r" (dest)
 196                     : "memory", "r8");
 197                 dest += 8;
 198                 source += 8;
 199                 size -= 8;
 200         }
 201
 202         /* 1x4 movnti loop */
 203         while (size >= 4) {
 204                 asm("movl    (%0), %%r8d\n"
 205                     "movnti  %%r8d,   (%1)\n"
 206                     :: "r" (source), "r" (dest)
 207                     : "memory", "r8");
 208                 dest += 4;
 209                 source += 4;
 210                 size -= 4;
 211         }
 212
 213         /* cache copy for remaining bytes */
 214         if (size) {
 215                 memcpy((void *) dest, (void *) source, size);
 216                 clean_cache_range((void *) dest, size);
 217         }
 218 }
 219 EXPORT_SYMBOL_GPL(__memcpy_flushcache);
 220
 221 void memcpy_page_flushcache(char *to, struct page *page, size_t offset,
 222                 size_t len)
 223 {
 224         char *from = kmap_atomic(page);
 225
 226         memcpy_flushcache(to, from + offset, len);
 227         kunmap_atomic(from);
 228 }
 229 #endif