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                 "li %%r17, 0\n\t" \
  70                 "std %%r17, 0(%[rseq_cs])\n\t" \
  71                 _teardown \
  72                 "b 5f\n\t" \
  73                 "4:\n\t" \
  74                 "li %%r17, 0\n\t" \
  75                 "std %%r17, 0(%[rseq_cs])\n\t" \
  76                 _teardown \
  77                 "b %l[failure]\n\t" \
  78                 "5:\n\t" \
  79                 : /* gcc asm goto does not allow outputs */ \
  80                 : [start_event_counter]"r"((_start_value).event_counter), \
  81                   [current_event_counter]"m"((_start_value).rseqp->u.e.event_counter), \
  82                   [rseq_cs]"b"(&(_start_value).rseqp->rseq_cs) \
  83                   _spec_input \
  84                   _final_input \
  85                   RSEQ_INJECT_INPUT \
  86                 : "r17", "memory", "cc" \
  87                   _extra_clobber \
  88                   RSEQ_INJECT_CLOBBER \
  89                 : _failure \
  90         )
  91
  92 #define RSEQ_FINISH_FINAL_STORE_ASM() \
  93                 "std %[to_write_final], 0(%[target_final])\n\t"
  94
  95 #define RSEQ_FINISH_FINAL_STORE_RELEASE_ASM() \
  96                 "lwsync\n\t" \
  97                 RSEQ_FINISH_FINAL_STORE_ASM()
  98
  99 #define RSEQ_FINISH_FINAL_STORE_INPUT(_target_final, _to_write_final) \
 100                 , [to_write_final]"r"(_to_write_final), \
 101                 [target_final]"b"(_target_final)
 102
 103 #define RSEQ_FINISH_SPECULATIVE_STORE_ASM() \
 104                 "std %[to_write_spec], 0(%[target_spec])\n\t" \
 105                 RSEQ_INJECT_ASM(4)
 106
 107 #define RSEQ_FINISH_SPECULATIVE_STORE_INPUT(_target_spec, _to_write_spec) \
 108                 , [to_write_spec]"r"(_to_write_spec), \
 109                 [target_spec]"b"(_target_spec)
 110
 111 /* TODO: implement a faster memcpy. */
 112 #define RSEQ_FINISH_MEMCPY_STORE_ASM() \
 113                 "cmpdi %%r19, 0\n\t" \
 114                 "beq 333f\n\t" \
 115                 "addi %%r20, %%r20, -1\n\t" \
 116                 "addi %%r21, %%r21, -1\n\t" \
 117                 "222:\n\t" \
 118                 "lbzu %%r18, 1(%%r20)\n\t" \
 119                 "stbu %%r18, 1(%%r21)\n\t" \
 120                 "addi %%r19, %%r19, -1\n\t" \
 121                 "cmpdi %%r19, 0\n\t" \
 122                 "bne 222b\n\t" \
 123                 "333:\n\t" \
 124                 RSEQ_INJECT_ASM(4)
 125
 126 #define RSEQ_FINISH_MEMCPY_STORE_INPUT(_target_memcpy, _to_write_memcpy, _len_memcpy) \
 127                 , [to_write_memcpy]"r"(_to_write_memcpy), \
 128                 [target_memcpy]"r"(_target_memcpy), \
 129                 [len_memcpy]"r"(_len_memcpy)
 130
 131 #define RSEQ_FINISH_MEMCPY_CLOBBER() \
 132                 , "r18", "r19", "r20", "r21"
 133
 134 #define RSEQ_FINISH_MEMCPY_SCRATCH()
 135
 136 /*
 137  * We use extra registers to hold the input registers, and we don't need to
 138  * save and restore the input registers.
 139  */
 140 #define RSEQ_FINISH_MEMCPY_SETUP() \
 141                 "mr %%r19, %[len_memcpy]\n\t" \
 142                 "mr %%r20, %[to_write_memcpy]\n\t" \
 143                 "mr %%r21, %[target_memcpy]\n\t" \
 144
 145 #define RSEQ_FINISH_MEMCPY_TEARDOWN()
 146
 147 #else   /* #ifdef __PPC64__ */
 148
 149 #define RSEQ_FINISH_ASM(_target_final, _to_write_final, _start_value, \
 150                 _failure, _spec_store, _spec_input, \
 151                 _final_store, _final_input, _extra_clobber, \
 152                 _setup, _teardown, _scratch) \
 153         __asm__ __volatile__ goto ( \
 154                 ".pushsection __rseq_table, \"aw\"\n\t" \
 155                 ".balign 32\n\t" \
 156                 "3:\n\t" \
 157                 /* 32-bit only supported on BE */ \
 158                 ".long 0x0, 1f, 0x0, 2f, 0x0, 4f, 0x0, 0x0\n\t" \
 159                 ".popsection\n\t" \
 160                 "1:\n\t" \
 161                 _setup \
 162                 RSEQ_INJECT_ASM(1) \
 163                 "lis %%r17, (3b)@ha\n\t" \
 164                 "addi %%r17, %%r17, (3b)@l\n\t" \
 165                 "stw %%r17, 0(%[rseq_cs])\n\t" \
 166                 RSEQ_INJECT_ASM(2) \
 167                 "lwz %%r17, %[current_event_counter]\n\t" \
 168                 "cmpw cr7, %[start_event_counter], %%r17\n\t" \
 169                 "bne- cr7, 4f\n\t" \
 170                 RSEQ_INJECT_ASM(3) \
 171                 _spec_store \
 172                 _final_store \
 173                 "2:\n\t" \
 174                 RSEQ_INJECT_ASM(5) \
 175                 "li %%r17, 0\n\t" \
 176                 "stw %%r17, 0(%[rseq_cs])\n\t" \
 177                 _teardown \
 178                 "b 5f\n\t" \
 179                 "4:\n\t" \
 180                 "li %%r17, 0\n\t" \
 181                 "std %%r17, 0(%[rseq_cs])\n\t" \
 182                 _teardown \
 183                 "b %l[failure]\n\t" \
 184                 "5:\n\t" \
 185                 : /* gcc asm goto does not allow outputs */ \
 186                 : [start_event_counter]"r"((_start_value).event_counter), \
 187                   [current_event_counter]"m"((_start_value).rseqp->u.e.event_counter), \
 188                   [rseq_cs]"b"(&(_start_value).rseqp->rseq_cs) \
 189                   _spec_input \
 190                   _final_input \
 191                   RSEQ_INJECT_INPUT \
 192                 : "r17", "memory", "cc" \
 193                   _extra_clobber \
 194                   RSEQ_INJECT_CLOBBER \
 195                 : _failure \
 196         )
 197
 198 #define RSEQ_FINISH_FINAL_STORE_ASM() \
 199                 "stw %[to_write_final], 0(%[target_final])\n\t"
 200
 201 #define RSEQ_FINISH_FINAL_STORE_RELEASE_ASM() \
 202                 "lwsync\n\t" \
 203                 RSEQ_FINISH_FINAL_STORE_ASM()
 204
 205 #define RSEQ_FINISH_FINAL_STORE_INPUT(_target_final, _to_write_final) \
 206                 , [to_write_final]"r"(_to_write_final), \
 207                 [target_final]"b"(_target_final)
 208
 209 #define RSEQ_FINISH_SPECULATIVE_STORE_ASM() \
 210                 "stw %[to_write_spec], 0(%[target_spec])\n\t" \
 211                 RSEQ_INJECT_ASM(4)
 212
 213 #define RSEQ_FINISH_SPECULATIVE_STORE_INPUT(_target_spec, _to_write_spec) \
 214                 , [to_write_spec]"r"(_to_write_spec), \
 215                 [target_spec]"b"(_target_spec)
 216
 217 /* TODO: implement a faster memcpy. */
 218 #define RSEQ_FINISH_MEMCPY_STORE_ASM() \
 219                 "cmpwi %%r19, 0\n\t" \
 220                 "beq 333f\n\t" \
 221                 "addi %%r20, %%r20, -1\n\t" \
 222                 "addi %%r21, %%r21, -1\n\t" \
 223                 "222:\n\t" \
 224                 "lbzu %%r18, 1(%%r20)\n\t" \
 225                 "stbu %%r18, 1(%%r21)\n\t" \
 226                 "addi %%r19, %%r19, -1\n\t" \
 227                 "cmpwi %%r19, 0\n\t" \
 228                 "bne 222b\n\t" \
 229                 "333:\n\t" \
 230                 RSEQ_INJECT_ASM(4)
 231
 232 #define RSEQ_FINISH_MEMCPY_STORE_INPUT(_target_memcpy, _to_write_memcpy, _len_memcpy) \
 233                 , [to_write_memcpy]"r"(_to_write_memcpy), \
 234                 [target_memcpy]"r"(_target_memcpy), \
 235                 [len_memcpy]"r"(_len_memcpy)
 236
 237 #define RSEQ_FINISH_MEMCPY_CLOBBER() \
 238                 , "r18", "r19", "r20", "r21"
 239
 240 #define RSEQ_FINISH_MEMCPY_SCRATCH()
 241
 242 /*
 243  * We use extra registers to hold the input registers, and we don't need to
 244  * save and restore the input registers.
 245  */
 246 #define RSEQ_FINISH_MEMCPY_SETUP() \
 247                 "mr %%r19, %[len_memcpy]\n\t" \
 248                 "mr %%r20, %[to_write_memcpy]\n\t" \
 249                 "mr %%r21, %[target_memcpy]\n\t" \
 250
 251 #define RSEQ_FINISH_MEMCPY_TEARDOWN()
 252
 253 #endif  /* #else #ifdef __PPC64__ */