libringbuffer/rseq-x86.h

   1 /*
   2  * rseq-x86.h
   3  *
   4  * (C) Copyright 2016 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  19  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  */
  24
  25 #ifdef __x86_64__
  26
  27 #define has_fast_acquire_release()      1
  28 #define has_single_copy_load_64()       1
  29
  30 /*
  31  * The __rseq_table section can be used by debuggers to better handle
  32  * single-stepping through the restartable critical sections.
  33  */
  34 #define RSEQ_FINISH_ASM(_target_final, _to_write_final, _start_value, \
  35                 _failure, _spec_store, _spec_input, \
  36                 _final_store, _final_input, _extra_clobber, \
  37                 _setup, _teardown, _scratch) \
  38 do { \
  39         _scratch \
  40         __asm__ __volatile__ goto ( \
  41                 ".pushsection __rseq_table, \"aw\"\n\t" \
  42                 ".balign 32\n\t" \
  43                 "3:\n\t" \
  44                 ".quad 1f, 2f, 4f, 0x0\n\t" \
  45                 ".popsection\n\t" \
  46                 "1:\n\t" \
  47                 _setup \
  48                 RSEQ_INJECT_ASM(1) \
  49                 "leaq 3b(%%rip), %%rax\n\t" \
  50                 "movq %%rax, %[rseq_cs]\n\t" \
  51                 RSEQ_INJECT_ASM(2) \
  52                 "cmpl %[start_event_counter], %[current_event_counter]\n\t" \
  53                 "jnz 4f\n\t" \
  54                 RSEQ_INJECT_ASM(3) \
  55                 _spec_store \
  56                 _final_store \
  57                 "2:\n\t" \
  58                 RSEQ_INJECT_ASM(5) \
  59                 _teardown \
  60                 ".pushsection __rseq_failure, \"a\"\n\t" \
  61                 "4:\n\t" \
  62                 _teardown \
  63                 "jmp %l[failure]\n\t" \
  64                 ".popsection\n\t" \
  65                 : /* gcc asm goto does not allow outputs */ \
  66                 : [start_event_counter]"r"((_start_value).event_counter), \
  67                   [current_event_counter]"m"((_start_value).rseqp->u.e.event_counter), \
  68                   [rseq_cs]"m"((_start_value).rseqp->rseq_cs) \
  69                   _spec_input \
  70                   _final_input \
  71                   RSEQ_INJECT_INPUT \
  72                 : "memory", "cc", "rax" \
  73                   _extra_clobber \
  74                   RSEQ_INJECT_CLOBBER \
  75                 : _failure \
  76         ); \
  77 } while (0)
  78
  79 #define RSEQ_FINISH_FINAL_STORE_ASM() \
  80                 "movq %[to_write_final], %[target_final]\n\t"
  81
  82 /* x86-64 is TSO */
  83 #define RSEQ_FINISH_FINAL_STORE_RELEASE_ASM() \
  84                 RSEQ_FINISH_FINAL_STORE_ASM()
  85
  86 #define RSEQ_FINISH_FINAL_STORE_INPUT(_target_final, _to_write_final) \
  87                 , [to_write_final]"r"(_to_write_final), \
  88                 [target_final]"m"(*(_target_final))
  89
  90 #define RSEQ_FINISH_SPECULATIVE_STORE_ASM() \
  91                 "movq %[to_write_spec], %[target_spec]\n\t" \
  92                 RSEQ_INJECT_ASM(4)
  93
  94 #define RSEQ_FINISH_SPECULATIVE_STORE_INPUT(_target_spec, _to_write_spec) \
  95                 , [to_write_spec]"r"(_to_write_spec), \
  96                 [target_spec]"m"(*(_target_spec))
  97
  98 /* TODO: implement a faster memcpy. */
  99 #define RSEQ_FINISH_MEMCPY_STORE_ASM() \
 100                 "test %[len_memcpy], %[len_memcpy]\n\t" \
 101                 "jz 333f\n\t" \
 102                 "222:\n\t" \
 103                 "movb (%[to_write_memcpy]), %%al\n\t" \
 104                 "movb %%al, (%[target_memcpy])\n\t" \
 105                 "inc %[to_write_memcpy]\n\t" \
 106                 "inc %[target_memcpy]\n\t" \
 107                 "dec %[len_memcpy]\n\t" \
 108                 "jnz 222b\n\t" \
 109                 "333:\n\t" \
 110                 RSEQ_INJECT_ASM(4)
 111
 112 #define RSEQ_FINISH_MEMCPY_STORE_INPUT(_target_memcpy, _to_write_memcpy, _len_memcpy) \
 113                 , [to_write_memcpy]"r"(_to_write_memcpy), \
 114                 [target_memcpy]"r"(_target_memcpy), \
 115                 [len_memcpy]"r"(_len_memcpy), \
 116                 [rseq_scratch0]"m"(rseq_scratch[0]), \
 117                 [rseq_scratch1]"m"(rseq_scratch[1]), \
 118                 [rseq_scratch2]"m"(rseq_scratch[2])
 119
 120 #define RSEQ_FINISH_MEMCPY_CLOBBER()    \
 121                 , "rax"
 122
 123 #define RSEQ_FINISH_MEMCPY_SCRATCH() \
 124                 uint64_t rseq_scratch[3];
 125
 126 /*
 127  * We need to save and restore those input registers so they can be
 128  * modified within the assembly.
 129  */
 130 #define RSEQ_FINISH_MEMCPY_SETUP() \
 131                 "movq %[to_write_memcpy], %[rseq_scratch0]\n\t" \
 132                 "movq %[target_memcpy], %[rseq_scratch1]\n\t" \
 133                 "movq %[len_memcpy], %[rseq_scratch2]\n\t"
 134
 135 #define RSEQ_FINISH_MEMCPY_TEARDOWN() \
 136                 "movq %[rseq_scratch2], %[len_memcpy]\n\t" \
 137                 "movq %[rseq_scratch1], %[target_memcpy]\n\t" \
 138                 "movq %[rseq_scratch0], %[to_write_memcpy]\n\t"
 139
 140 #elif __i386__
 141
 142 #define has_fast_acquire_release()      0
 143 #define has_single_copy_load_64()       0
 144
 145 /*
 146  * Use eax as scratch register and take memory operands as input to
 147  * lessen register pressure. Especially needed when compiling
 148  * do_rseq_memcpy() in O0.
 149  */
 150 #define RSEQ_FINISH_ASM(_target_final, _to_write_final, _start_value, \
 151                 _failure, _spec_store, _spec_input, \
 152                 _final_store, _final_input, _extra_clobber, \
 153                 _setup, _teardown, _scratch) \
 154 do { \
 155         _scratch \
 156         __asm__ __volatile__ goto ( \
 157                 ".pushsection __rseq_table, \"aw\"\n\t" \
 158                 ".balign 32\n\t" \
 159                 "3:\n\t" \
 160                 ".long 1f, 0x0, 2f, 0x0, 4f, 0x0, 0x0, 0x0\n\t" \
 161                 ".popsection\n\t" \
 162                 "1:\n\t" \
 163                 _setup \
 164                 RSEQ_INJECT_ASM(1) \
 165                 "movl $3b, %[rseq_cs]\n\t" \
 166                 RSEQ_INJECT_ASM(2) \
 167                 "movl %[start_event_counter], %%eax\n\t" \
 168                 "cmpl %%eax, %[current_event_counter]\n\t" \
 169                 "jnz 4f\n\t" \
 170                 RSEQ_INJECT_ASM(3) \
 171                 _spec_store \
 172                 _final_store \
 173                 "2:\n\t" \
 174                 RSEQ_INJECT_ASM(5) \
 175                 _teardown \
 176                 ".pushsection __rseq_failure, \"a\"\n\t" \
 177                 "4:\n\t" \
 178                 _teardown \
 179                 "jmp %l[failure]\n\t" \
 180                 ".popsection\n\t" \
 181                 : /* gcc asm goto does not allow outputs */ \
 182                 : [start_event_counter]"m"((_start_value).event_counter), \
 183                   [current_event_counter]"m"((_start_value).rseqp->u.e.event_counter), \
 184                   [rseq_cs]"m"((_start_value).rseqp->rseq_cs) \
 185                   _spec_input \
 186                   _final_input \
 187                   RSEQ_INJECT_INPUT \
 188                 : "memory", "cc", "eax" \
 189                   _extra_clobber \
 190                   RSEQ_INJECT_CLOBBER \
 191                 : _failure \
 192         ); \
 193 } while (0)
 194
 195 #define RSEQ_FINISH_FINAL_STORE_ASM() \
 196                 "movl %[to_write_final], %%eax\n\t" \
 197                 "movl %%eax, %[target_final]\n\t"
 198
 199 #define RSEQ_FINISH_FINAL_STORE_RELEASE_ASM() \
 200                 "lock; addl $0,0(%%esp)\n\t" \
 201                 RSEQ_FINISH_FINAL_STORE_ASM()
 202
 203 #define RSEQ_FINISH_FINAL_STORE_INPUT(_target_final, _to_write_final) \
 204                 , [to_write_final]"m"(_to_write_final), \
 205                 [target_final]"m"(*(_target_final))
 206
 207 #define RSEQ_FINISH_SPECULATIVE_STORE_ASM() \
 208                 "movl %[to_write_spec], %%eax\n\t" \
 209                 "movl %%eax, %[target_spec]\n\t" \
 210                 RSEQ_INJECT_ASM(4)
 211
 212 #define RSEQ_FINISH_SPECULATIVE_STORE_INPUT(_target_spec, _to_write_spec) \
 213                 , [to_write_spec]"m"(_to_write_spec), \
 214                 [target_spec]"m"(*(_target_spec))
 215
 216 /* TODO: implement a faster memcpy. */
 217 #define RSEQ_FINISH_MEMCPY_STORE_ASM() \
 218                 "movl %[len_memcpy], %%eax\n\t" \
 219                 "test %%eax, %%eax\n\t" \
 220                 "jz 333f\n\t" \
 221                 "222:\n\t" \
 222                 "movb (%[to_write_memcpy]), %%al\n\t" \
 223                 "movb %%al, (%[target_memcpy])\n\t" \
 224                 "inc %[to_write_memcpy]\n\t" \
 225                 "inc %[target_memcpy]\n\t" \
 226                 "decl %[rseq_scratch2]\n\t" \
 227                 "jnz 222b\n\t" \
 228                 "333:\n\t" \
 229                 RSEQ_INJECT_ASM(4)
 230
 231 #define RSEQ_FINISH_MEMCPY_STORE_INPUT(_target_memcpy, _to_write_memcpy, _len_memcpy) \
 232                 , [to_write_memcpy]"r"(_to_write_memcpy), \
 233                 [target_memcpy]"r"(_target_memcpy), \
 234                 [len_memcpy]"m"(_len_memcpy), \
 235                 [rseq_scratch0]"m"(rseq_scratch[0]), \
 236                 [rseq_scratch1]"m"(rseq_scratch[1]), \
 237                 [rseq_scratch2]"m"(rseq_scratch[2])
 238
 239 #define RSEQ_FINISH_MEMCPY_CLOBBER()
 240
 241 #define RSEQ_FINISH_MEMCPY_SCRATCH() \
 242                 uint32_t rseq_scratch[3];
 243
 244 /*
 245  * We need to save and restore those input registers so they can be
 246  * modified within the assembly.
 247  */
 248 #define RSEQ_FINISH_MEMCPY_SETUP() \
 249                 "movl %[to_write_memcpy], %[rseq_scratch0]\n\t" \
 250                 "movl %[target_memcpy], %[rseq_scratch1]\n\t" \
 251                 "movl %[len_memcpy], %%eax\n\t" \
 252                 "movl %%eax, %[rseq_scratch2]\n\t"
 253
 254 #define RSEQ_FINISH_MEMCPY_TEARDOWN() \
 255                 "movl %[rseq_scratch1], %[target_memcpy]\n\t" \
 256                 "movl %[rseq_scratch0], %[to_write_memcpy]\n\t"
 257
 258 #endif