libringbuffer/rseq-ppc.h

   1 /*
   2  * rseq-ppc.h
   3  *
   4  * (C) Copyright 2016 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   5  * (C) Copyright 2016 - Boqun Feng <boqun.feng@gmail.com>
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  20  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  23  * SOFTWARE.
  24  */
  25
  26 #define has_fast_acquire_release()      0
  27
  28 #ifdef __PPC64__
  29 #define has_single_copy_load_64()       1
  30 #else
  31 #define has_single_copy_load_64()       0
  32 #endif
  33
  34 /*
  35  * The __rseq_table section can be used by debuggers to better handle
  36  * single-stepping through the restartable critical sections.
  37  */
  38
  39 #ifdef __PPC64__
  40
  41 #define RSEQ_FINISH_ASM(_target_final, _to_write_final, _start_value, \
  42                 _failure, _spec_store, _spec_input, \
  43                 _final_store, _final_input, _extra_clobber, \
  44                 _setup, _teardown, _scratch) \
  45         __asm__ __volatile__ goto ( \
  46                 ".pushsection __rseq_table, \"aw\"\n\t" \
  47                 ".balign 32\n\t" \
  48                 "3:\n\t" \
  49                 ".quad 1f, 2f, 4f, 0x0\n\t" \
  50                 ".popsection\n\t" \
  51                 "1:\n\t" \
  52                 _setup \
  53                 RSEQ_INJECT_ASM(1) \
  54                 "lis %%r17, (3b)@highest\n\t" \
  55                 "ori %%r17, %%r17, (3b)@higher\n\t" \
  56                 "rldicr %%r17, %%r17, 32, 31\n\t" \
  57                 "oris %%r17, %%r17, (3b)@h\n\t" \
  58                 "ori %%r17, %%r17, (3b)@l\n\t" \
  59                 "std %%r17, 0(%[rseq_cs])\n\t" \
  60                 RSEQ_INJECT_ASM(2) \
  61                 "lwz %%r17, %[current_event_counter]\n\t" \
  62                 "cmpw cr7, %[start_event_counter], %%r17\n\t" \
  63                 "bne- cr7, 4f\n\t" \
  64                 RSEQ_INJECT_ASM(3) \
  65                 _spec_store \
  66                 _final_store \
  67                 "2:\n\t" \
  68                 RSEQ_INJECT_ASM(5) \
  69                 _teardown \
  70                 "b 5f\n\t" \
  71                 "4:\n\t" \
  72                 _teardown \
  73                 "b %l[failure]\n\t" \
  74                 "5:\n\t" \
  75                 : /* gcc asm goto does not allow outputs */ \
  76                 : [start_event_counter]"r"((_start_value).event_counter), \
  77                   [current_event_counter]"m"((_start_value).rseqp->u.e.event_counter), \
  78                   [rseq_cs]"b"(&(_start_value).rseqp->rseq_cs) \
  79                   _spec_input \
  80                   _final_input \
  81                   RSEQ_INJECT_INPUT \
  82                 : "r17", "memory", "cc" \
  83                   _extra_clobber \
  84                   RSEQ_INJECT_CLOBBER \
  85                 : _failure \
  86         )
  87
  88 #define RSEQ_FINISH_FINAL_STORE_ASM() \
  89                 "std %[to_write_final], 0(%[target_final])\n\t"
  90
  91 #define RSEQ_FINISH_FINAL_STORE_RELEASE_ASM() \
  92                 "lwsync\n\t" \
  93                 RSEQ_FINISH_FINAL_STORE_ASM()
  94
  95 #define RSEQ_FINISH_FINAL_STORE_INPUT(_target_final, _to_write_final) \
  96                 , [to_write_final]"r"(_to_write_final), \
  97                 [target_final]"b"(_target_final)
  98
  99 #define RSEQ_FINISH_SPECULATIVE_STORE_ASM() \
 100                 "std %[to_write_spec], 0(%[target_spec])\n\t" \
 101                 RSEQ_INJECT_ASM(4)
 102
 103 #define RSEQ_FINISH_SPECULATIVE_STORE_INPUT(_target_spec, _to_write_spec) \
 104                 , [to_write_spec]"r"(_to_write_spec), \
 105                 [target_spec]"b"(_target_spec)
 106
 107 /* TODO: implement a faster memcpy. */
 108 #define RSEQ_FINISH_MEMCPY_STORE_ASM() \
 109                 "cmpdi %%r19, 0\n\t" \
 110                 "beq 333f\n\t" \
 111                 "addi %%r20, %%r20, -1\n\t" \
 112                 "addi %%r21, %%r21, -1\n\t" \
 113                 "222:\n\t" \
 114                 "lbzu %%r18, 1(%%r20)\n\t" \
 115                 "stbu %%r18, 1(%%r21)\n\t" \
 116                 "addi %%r19, %%r19, -1\n\t" \
 117                 "cmpdi %%r19, 0\n\t" \
 118                 "bne 222b\n\t" \
 119                 "333:\n\t" \
 120                 RSEQ_INJECT_ASM(4)
 121
 122 #define RSEQ_FINISH_MEMCPY_STORE_INPUT(_target_memcpy, _to_write_memcpy, _len_memcpy) \
 123                 , [to_write_memcpy]"r"(_to_write_memcpy), \
 124                 [target_memcpy]"r"(_target_memcpy), \
 125                 [len_memcpy]"r"(_len_memcpy)
 126
 127 #define RSEQ_FINISH_MEMCPY_CLOBBER() \
 128                 , "r18", "r19", "r20", "r21"
 129
 130 #define RSEQ_FINISH_MEMCPY_SCRATCH()
 131
 132 /*
 133  * We use extra registers to hold the input registers, and we don't need to
 134  * save and restore the input registers.
 135  */
 136 #define RSEQ_FINISH_MEMCPY_SETUP() \
 137                 "mr %%r19, %[len_memcpy]\n\t" \
 138                 "mr %%r20, %[to_write_memcpy]\n\t" \
 139                 "mr %%r21, %[target_memcpy]\n\t" \
 140
 141 #define RSEQ_FINISH_MEMCPY_TEARDOWN()
 142
 143 #else   /* #ifdef __PPC64__ */
 144
 145 #define RSEQ_FINISH_ASM(_target_final, _to_write_final, _start_value, \
 146                 _failure, _spec_store, _spec_input, \
 147                 _final_store, _final_input, _extra_clobber, \
 148                 _setup, _teardown, _scratch) \
 149         __asm__ __volatile__ goto ( \
 150                 ".pushsection __rseq_table, \"aw\"\n\t" \
 151                 ".balign 32\n\t" \
 152                 "3:\n\t" \
 153                 /* 32-bit only supported on BE */ \
 154                 ".long 0x0, 1f, 0x0, 2f, 0x0, 4f, 0x0, 0x0\n\t" \
 155                 ".popsection\n\t" \
 156                 "1:\n\t" \
 157                 _setup \
 158                 RSEQ_INJECT_ASM(1) \
 159                 "lis %%r17, (3b)@ha\n\t" \
 160                 "addi %%r17, %%r17, (3b)@l\n\t" \
 161                 "stw %%r17, 0(%[rseq_cs])\n\t" \
 162                 RSEQ_INJECT_ASM(2) \
 163                 "lwz %%r17, %[current_event_counter]\n\t" \
 164                 "cmpw cr7, %[start_event_counter], %%r17\n\t" \
 165                 "bne- cr7, 4f\n\t" \
 166                 RSEQ_INJECT_ASM(3) \
 167                 _spec_store \
 168                 _final_store \
 169                 "2:\n\t" \
 170                 RSEQ_INJECT_ASM(5) \
 171                 _teardown \
 172                 "b 5f\n\t" \
 173                 "4:\n\t" \
 174                 _teardown \
 175                 "b %l[failure]\n\t" \
 176                 "5:\n\t" \
 177                 : /* gcc asm goto does not allow outputs */ \
 178                 : [start_event_counter]"r"((_start_value).event_counter), \
 179                   [current_event_counter]"m"((_start_value).rseqp->u.e.event_counter), \
 180                   [rseq_cs]"b"(&(_start_value).rseqp->rseq_cs) \
 181                   _spec_input \
 182                   _final_input \
 183                   RSEQ_INJECT_INPUT \
 184                 : "r17", "memory", "cc" \
 185                   _extra_clobber \
 186                   RSEQ_INJECT_CLOBBER \
 187                 : _failure \
 188         )
 189
 190 #define RSEQ_FINISH_FINAL_STORE_ASM() \
 191                 "stw %[to_write_final], 0(%[target_final])\n\t"
 192
 193 #define RSEQ_FINISH_FINAL_STORE_RELEASE_ASM() \
 194                 "lwsync\n\t" \
 195                 RSEQ_FINISH_FINAL_STORE_ASM()
 196
 197 #define RSEQ_FINISH_FINAL_STORE_INPUT(_target_final, _to_write_final) \
 198                 , [to_write_final]"r"(_to_write_final), \
 199                 [target_final]"b"(_target_final)
 200
 201 #define RSEQ_FINISH_SPECULATIVE_STORE_ASM() \
 202                 "stw %[to_write_spec], 0(%[target_spec])\n\t" \
 203                 RSEQ_INJECT_ASM(4)
 204
 205 #define RSEQ_FINISH_SPECULATIVE_STORE_INPUT(_target_spec, _to_write_spec) \
 206                 , [to_write_spec]"r"(_to_write_spec), \
 207                 [target_spec]"b"(_target_spec)
 208
 209 /* TODO: implement a faster memcpy. */
 210 #define RSEQ_FINISH_MEMCPY_STORE_ASM() \
 211                 "cmpwi %%r19, 0\n\t" \
 212                 "beq 333f\n\t" \
 213                 "addi %%r20, %%r20, -1\n\t" \
 214                 "addi %%r21, %%r21, -1\n\t" \
 215                 "222:\n\t" \
 216                 "lbzu %%r18, 1(%%r20)\n\t" \
 217                 "stbu %%r18, 1(%%r21)\n\t" \
 218                 "addi %%r19, %%r19, -1\n\t" \
 219                 "cmpwi %%r19, 0\n\t" \
 220                 "bne 222b\n\t" \
 221                 "333:\n\t" \
 222                 RSEQ_INJECT_ASM(4)
 223
 224 #define RSEQ_FINISH_MEMCPY_STORE_INPUT(_target_memcpy, _to_write_memcpy, _len_memcpy) \
 225                 , [to_write_memcpy]"r"(_to_write_memcpy), \
 226                 [target_memcpy]"r"(_target_memcpy), \
 227                 [len_memcpy]"r"(_len_memcpy)
 228
 229 #define RSEQ_FINISH_MEMCPY_CLOBBER() \
 230                 , "r18", "r19", "r20", "r21"
 231
 232 #define RSEQ_FINISH_MEMCPY_SCRATCH()
 233
 234 /*
 235  * We use extra registers to hold the input registers, and we don't need to
 236  * save and restore the input registers.
 237  */
 238 #define RSEQ_FINISH_MEMCPY_SETUP() \
 239                 "mr %%r19, %[len_memcpy]\n\t" \
 240                 "mr %%r20, %[to_write_memcpy]\n\t" \
 241                 "mr %%r21, %[target_memcpy]\n\t" \
 242
 243 #define RSEQ_FINISH_MEMCPY_TEARDOWN()
 244
 245 #endif  /* #else #ifdef __PPC64__ */