sewardj | 112711a | 2015-04-10 12:30:09 +0000 | [diff] [blame] | 1 | |
| 2 | /*--------------------------------------------------------------------*/ |
| 3 | /*--- begin dispatch-tilegx-linux.S ---*/ |
| 4 | /*--------------------------------------------------------------------*/ |
| 5 | |
| 6 | /* |
| 7 | This file is part of Valgrind, a dynamic binary instrumentation |
| 8 | framework. |
| 9 | |
sewardj | b3a1e4b | 2015-08-21 11:32:26 +0000 | [diff] [blame] | 10 | Copyright (C) 2010-2015 Tilera Corp. |
sewardj | 112711a | 2015-04-10 12:30:09 +0000 | [diff] [blame] | 11 | |
| 12 | This program is free software; you can redistribute it and/or |
| 13 | modify it under the terms of the GNU General Public License as |
| 14 | published by the Free Software Foundation; either version 2 of the |
| 15 | License, or (at your option) any later version. |
| 16 | |
| 17 | This program is distributed in the hope that it will be useful, but |
| 18 | WITHOUT ANY WARRANTY; without even the implied warranty of |
| 19 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 20 | General Public License for more details. |
| 21 | |
| 22 | You should have received a copy of the GNU General Public License |
| 23 | along with this program; if not, write to the Free Software |
| 24 | Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA |
| 25 | 02111-1307, USA. |
| 26 | |
| 27 | The GNU General Public License is contained in the file COPYING. |
| 28 | */ |
| 29 | |
| 30 | /* Contributed by Zhi-Gang Liu <zliu at tilera dot com> */ |
| 31 | |
sewardj | 112711a | 2015-04-10 12:30:09 +0000 | [diff] [blame] | 32 | #include "pub_core_basics_asm.h" |
florian | 3f1d613 | 2015-09-30 20:30:48 +0000 | [diff] [blame] | 33 | |
| 34 | #if defined(VGP_tilegx_linux) |
sewardj | 112711a | 2015-04-10 12:30:09 +0000 | [diff] [blame] | 35 | #include "pub_core_dispatch_asm.h" |
| 36 | #include "pub_core_transtab_asm.h" |
| 37 | #include "libvex_guest_offsets.h" /* for OFFSET_tilegx_PC */ |
| 38 | |
| 39 | /*------------------------------------------------------------*/ |
| 40 | /*--- ---*/ |
| 41 | /*--- The dispatch loop. VG_(run_innerloop) is used to ---*/ |
| 42 | /*--- run all translations except no-redir ones. ---*/ |
| 43 | /*--- ---*/ |
| 44 | /*------------------------------------------------------------*/ |
| 45 | |
| 46 | /*----------------------------------------------------*/ |
| 47 | /*--- Preamble (set everything up) ---*/ |
| 48 | /*----------------------------------------------------*/ |
| 49 | |
| 50 | /* signature: |
| 51 | void VG_(disp_run_translations)(UWord* two_words, |
| 52 | void* guest_state, |
| 53 | Addr host_addr ); |
| 54 | UWord VG_(run_innerloop) ( void* guest_state, UWord do_profiling ); |
| 55 | */ |
| 56 | |
| 57 | .text |
| 58 | .globl VG_(disp_run_translations) |
| 59 | VG_(disp_run_translations): |
| 60 | |
| 61 | /* r0 holds two_words |
| 62 | r1 holds guest_state |
| 63 | r2 holds host_addr */ |
| 64 | |
| 65 | /* New stack frame */ |
| 66 | addli sp, sp, -256 |
| 67 | addi r29, sp, 8 |
| 68 | /* |
| 69 | high memory of stack |
| 70 | 216 lr |
| 71 | 208 r53 |
| 72 | 200 r52 |
| 73 | 192 r51 |
| 74 | ... |
| 75 | 48 r33 |
| 76 | 40 r32 |
| 77 | 32 r31 |
| 78 | 24 r30 |
| 79 | 16 r1 <--- |
| 80 | 8 r0 |
| 81 | 0 <-sp |
| 82 | */ |
| 83 | st_add r29, r0, 8 |
| 84 | st_add r29, r1, 8 |
| 85 | |
| 86 | /* ... and r30 - r53 */ |
| 87 | st_add r29, r30, 8 |
| 88 | st_add r29, r31, 8 |
| 89 | st_add r29, r32, 8 |
| 90 | st_add r29, r33, 8 |
| 91 | st_add r29, r34, 8 |
| 92 | st_add r29, r35, 8 |
| 93 | st_add r29, r36, 8 |
| 94 | st_add r29, r37, 8 |
| 95 | st_add r29, r38, 8 |
| 96 | st_add r29, r39, 8 |
| 97 | st_add r29, r40, 8 |
| 98 | st_add r29, r41, 8 |
| 99 | st_add r29, r42, 8 |
| 100 | st_add r29, r43, 8 |
| 101 | st_add r29, r44, 8 |
| 102 | st_add r29, r45, 8 |
| 103 | st_add r29, r46, 8 |
| 104 | st_add r29, r47, 8 |
| 105 | st_add r29, r48, 8 |
| 106 | st_add r29, r49, 8 |
| 107 | st_add r29, r50, 8 |
| 108 | st_add r29, r51, 8 |
| 109 | st_add r29, r52, 8 |
| 110 | st_add r29, r53, 8 |
| 111 | st r29, lr |
| 112 | |
| 113 | /* Load the address of guest state into guest state register r50. */ |
| 114 | move r50, r1 |
| 115 | |
| 116 | //j postamble |
| 117 | |
| 118 | /* jump to the code cache. */ |
| 119 | jr r2 |
| 120 | /*NOTREACHED*/ |
| 121 | |
| 122 | |
| 123 | /*----------------------------------------------------*/ |
| 124 | /*--- Postamble and exit. ---*/ |
| 125 | /*----------------------------------------------------*/ |
| 126 | |
| 127 | postamble: |
| 128 | /* At this point, r12 and r13 contain two |
| 129 | words to be returned to the caller. r12 |
| 130 | holds a TRC value, and r13 optionally may |
| 131 | hold another word (for CHAIN_ME exits, the |
| 132 | address of the place to patch.) */ |
| 133 | |
| 134 | /* run_innerloop_exit_REALLY: |
| 135 | r50 holds VG_TRC_* value to return |
| 136 | Return to parent stack |
| 137 | addli sp, sp, 256 */ |
| 138 | |
| 139 | addi r29, sp, 8 |
| 140 | |
| 141 | /* Restore r0 from stack; holding address of twp words */ |
| 142 | ld_add r0, r29, 16 |
| 143 | /* store r12 in two_words[0] */ |
| 144 | st_add r0, r12, 8 |
| 145 | /* store r13 in two_words[1] */ |
| 146 | st r0, r13 |
| 147 | |
| 148 | /* Restore callee-saved registers... */ |
| 149 | ld_add r30, r29, 8 |
| 150 | ld_add r31, r29, 8 |
| 151 | ld_add r32, r29, 8 |
| 152 | ld_add r33, r29, 8 |
| 153 | ld_add r34, r29, 8 |
| 154 | ld_add r35, r29, 8 |
| 155 | ld_add r36, r29, 8 |
| 156 | ld_add r37, r29, 8 |
| 157 | ld_add r38, r29, 8 |
| 158 | ld_add r39, r29, 8 |
| 159 | ld_add r40, r29, 8 |
| 160 | ld_add r41, r29, 8 |
| 161 | ld_add r42, r29, 8 |
| 162 | ld_add r43, r29, 8 |
| 163 | ld_add r44, r29, 8 |
| 164 | ld_add r45, r29, 8 |
| 165 | ld_add r46, r29, 8 |
| 166 | ld_add r47, r29, 8 |
| 167 | ld_add r48, r29, 8 |
| 168 | ld_add r49, r29, 8 |
| 169 | ld_add r50, r29, 8 |
| 170 | ld_add r51, r29, 8 |
| 171 | ld_add r52, r29, 8 |
| 172 | ld_add r53, r29, 8 |
| 173 | ld lr, r29 |
| 174 | addli sp, sp, 256 /* stack_size */ |
| 175 | jr lr |
| 176 | nop |
| 177 | |
| 178 | |
| 179 | /*----------------------------------------------------*/ |
| 180 | /*--- Continuation points ---*/ |
| 181 | /*----------------------------------------------------*/ |
| 182 | |
| 183 | /* ------ Chain me to slow entry point ------ */ |
| 184 | .global VG_(disp_cp_chain_me_to_slowEP) |
| 185 | VG_(disp_cp_chain_me_to_slowEP): |
| 186 | /* We got called. The return address indicates |
| 187 | where the patching needs to happen. Collect |
| 188 | the return address and, exit back to C land, |
| 189 | handing the caller the pair (Chain_me_S, RA) */ |
| 190 | # if (VG_TRC_CHAIN_ME_TO_SLOW_EP > 128) |
| 191 | # error ("VG_TRC_CHAIN_ME_TO_SLOW_EP is > 128"); |
| 192 | # endif |
| 193 | moveli r12, VG_TRC_CHAIN_ME_TO_SLOW_EP |
| 194 | move r13, lr |
| 195 | /* 32 = mkLoadImm_EXACTLY4 |
| 196 | 8 = jalr r9 |
| 197 | 8 = nop */ |
| 198 | addi r13, r13, -40 |
| 199 | j postamble |
| 200 | |
| 201 | /* ------ Chain me to slow entry point ------ */ |
| 202 | .global VG_(disp_cp_chain_me_to_fastEP) |
| 203 | VG_(disp_cp_chain_me_to_fastEP): |
| 204 | /* We got called. The return address indicates |
| 205 | where the patching needs to happen. Collect |
| 206 | the return address and, exit back to C land, |
| 207 | handing the caller the pair (Chain_me_S, RA) */ |
| 208 | # if (VG_TRC_CHAIN_ME_TO_FAST_EP > 128) |
| 209 | # error ("VG_TRC_CHAIN_ME_TO_FAST_EP is > 128"); |
| 210 | # endif |
| 211 | moveli r12, VG_TRC_CHAIN_ME_TO_FAST_EP |
| 212 | move r13, lr |
| 213 | /* 32 = mkLoadImm_EXACTLY4 |
| 214 | 8 = jalr r9 |
| 215 | 8 = nop */ |
| 216 | addi r13, r13, -40 |
| 217 | j postamble |
| 218 | |
| 219 | /* ------ Indirect but boring jump ------ */ |
| 220 | .global VG_(disp_cp_xindir) |
| 221 | VG_(disp_cp_xindir): |
| 222 | /* Where are we going? */ |
| 223 | addli r11, r50, OFFSET_tilegx_pc |
| 224 | ld r11, r11 |
| 225 | |
| 226 | moveli r7, hw2_last(VG_(stats__n_xindirs_32)) |
| 227 | shl16insli r7, r7, hw1(VG_(stats__n_xindirs_32)) |
| 228 | shl16insli r7, r7, hw0(VG_(stats__n_xindirs_32)) |
| 229 | ld4u r6, r7 |
| 230 | addi r6, r6, 1 |
| 231 | st4 r7, r6 |
| 232 | |
| 233 | /* try a fast lookup in the translation cache */ |
| 234 | /* r14 = VG_TT_FAST_HASH(addr) * sizeof(ULong*) |
| 235 | = (t8 >> 3 & VG_TT_FAST_MASK) << 3 */ |
| 236 | |
| 237 | move r14, r11 |
| 238 | /* Assume VG_TT_FAST_MASK < 4G */ |
| 239 | moveli r12, hw1(VG_TT_FAST_MASK) |
| 240 | shl16insli r12, r12, hw0(VG_TT_FAST_MASK) |
| 241 | shrui r14, r14, 3 |
| 242 | and r14, r14, r12 |
| 243 | shli r14, r14, 4 |
| 244 | /* Note, each tt_fast hash entry has two pointers i.e. 16 Bytes. */ |
| 245 | |
| 246 | /* r13 = (addr of VG_(tt_fast)) + r14 */ |
| 247 | moveli r13, hw2_last(VG_(tt_fast)) |
| 248 | shl16insli r13, r13, hw1(VG_(tt_fast)) |
| 249 | shl16insli r13, r13, hw0(VG_(tt_fast)) |
| 250 | |
| 251 | add r13, r13, r14 |
| 252 | |
| 253 | /* r12 = VG_(tt_fast)[hash] :: ULong* */ |
| 254 | ld_add r12, r13, 8 |
| 255 | |
| 256 | { |
| 257 | ld r25, r13 |
| 258 | sub r7, r12, r11 |
| 259 | } |
| 260 | |
| 261 | bnez r7, fast_lookup_failed |
| 262 | |
| 263 | /* Run the translation */ |
| 264 | jr r25 |
| 265 | |
| 266 | .quad 0x0 |
| 267 | |
| 268 | fast_lookup_failed: |
| 269 | /* %PC is up to date */ |
| 270 | /* back out decrement of the dispatch counter */ |
| 271 | /* hold dispatch_ctr in t0 (r8) */ |
| 272 | |
| 273 | moveli r7, hw2_last(VG_(stats__n_xindir_misses_32)) |
| 274 | shl16insli r7, r7, hw1(VG_(stats__n_xindir_misses_32)) |
| 275 | shl16insli r7, r7, hw0(VG_(stats__n_xindir_misses_32)) |
| 276 | ld4u r6, r7 |
| 277 | addi r6, r6, 1 |
| 278 | st4 r7, r6 |
| 279 | moveli r12, VG_TRC_INNER_FASTMISS |
| 280 | movei r13, 0 |
| 281 | j postamble |
| 282 | |
| 283 | /* ------ Assisted jump ------ */ |
| 284 | .global VG_(disp_cp_xassisted) |
| 285 | VG_(disp_cp_xassisted): |
| 286 | /* guest-state-pointer contains the TRC. Put the value into the |
| 287 | return register */ |
| 288 | move r12, r50 |
| 289 | movei r13, 0 |
| 290 | j postamble |
| 291 | |
| 292 | /* ------ Event check failed ------ */ |
| 293 | .global VG_(disp_cp_evcheck_fail) |
| 294 | VG_(disp_cp_evcheck_fail): |
| 295 | moveli r12, VG_TRC_INNER_COUNTERZERO |
| 296 | movei r13, 0 |
| 297 | j postamble |
| 298 | |
| 299 | .size VG_(disp_run_translations), .-VG_(disp_run_translations) |
| 300 | |
sewardj | 112711a | 2015-04-10 12:30:09 +0000 | [diff] [blame] | 301 | #endif /* defined(VGP_tilegx_linux) */ |
florian | 3f1d613 | 2015-09-30 20:30:48 +0000 | [diff] [blame] | 302 | |
| 303 | /* Let the linker know we don't need an executable stack */ |
| 304 | MARK_STACK_NO_EXEC |
| 305 | |
sewardj | 112711a | 2015-04-10 12:30:09 +0000 | [diff] [blame] | 306 | /*--------------------------------------------------------------------*/ |
| 307 | /*--- end ---*/ |
| 308 | /*--------------------------------------------------------------------*/ |
| 309 | |