Thu, 14 Apr 2011 13:45:41 -0700
Merge
duke@435 | 1 | /* |
twisti@2552 | 2 | * Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved. |
duke@435 | 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
duke@435 | 4 | * |
duke@435 | 5 | * This code is free software; you can redistribute it and/or modify it |
duke@435 | 6 | * under the terms of the GNU General Public License version 2 only, as |
duke@435 | 7 | * published by the Free Software Foundation. |
duke@435 | 8 | * |
duke@435 | 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
duke@435 | 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
duke@435 | 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
duke@435 | 12 | * version 2 for more details (a copy is included in the LICENSE file that |
duke@435 | 13 | * accompanied this code). |
duke@435 | 14 | * |
duke@435 | 15 | * You should have received a copy of the GNU General Public License version |
duke@435 | 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
duke@435 | 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
duke@435 | 18 | * |
trims@1907 | 19 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
trims@1907 | 20 | * or visit www.oracle.com if you need additional information or have any |
trims@1907 | 21 | * questions. |
duke@435 | 22 | * |
duke@435 | 23 | */ |
duke@435 | 24 | |
stefank@2314 | 25 | #include "precompiled.hpp" |
stefank@2314 | 26 | #include "asm/assembler.hpp" |
stefank@2314 | 27 | #include "assembler_x86.inline.hpp" |
stefank@2314 | 28 | #include "code/debugInfoRec.hpp" |
stefank@2314 | 29 | #include "code/icBuffer.hpp" |
stefank@2314 | 30 | #include "code/vtableStubs.hpp" |
stefank@2314 | 31 | #include "interpreter/interpreter.hpp" |
stefank@2314 | 32 | #include "oops/compiledICHolderOop.hpp" |
stefank@2314 | 33 | #include "prims/jvmtiRedefineClassesTrace.hpp" |
stefank@2314 | 34 | #include "runtime/sharedRuntime.hpp" |
stefank@2314 | 35 | #include "runtime/vframeArray.hpp" |
stefank@2314 | 36 | #include "vmreg_x86.inline.hpp" |
stefank@2314 | 37 | #ifdef COMPILER1 |
stefank@2314 | 38 | #include "c1/c1_Runtime1.hpp" |
stefank@2314 | 39 | #endif |
stefank@2314 | 40 | #ifdef COMPILER2 |
stefank@2314 | 41 | #include "opto/runtime.hpp" |
stefank@2314 | 42 | #endif |
duke@435 | 43 | |
duke@435 | 44 | DeoptimizationBlob *SharedRuntime::_deopt_blob; |
duke@435 | 45 | #ifdef COMPILER2 |
duke@435 | 46 | UncommonTrapBlob *SharedRuntime::_uncommon_trap_blob; |
duke@435 | 47 | ExceptionBlob *OptoRuntime::_exception_blob; |
duke@435 | 48 | #endif // COMPILER2 |
duke@435 | 49 | |
duke@435 | 50 | SafepointBlob *SharedRuntime::_polling_page_safepoint_handler_blob; |
duke@435 | 51 | SafepointBlob *SharedRuntime::_polling_page_return_handler_blob; |
duke@435 | 52 | RuntimeStub* SharedRuntime::_wrong_method_blob; |
duke@435 | 53 | RuntimeStub* SharedRuntime::_ic_miss_blob; |
duke@435 | 54 | RuntimeStub* SharedRuntime::_resolve_opt_virtual_call_blob; |
duke@435 | 55 | RuntimeStub* SharedRuntime::_resolve_virtual_call_blob; |
duke@435 | 56 | RuntimeStub* SharedRuntime::_resolve_static_call_blob; |
duke@435 | 57 | |
xlu@959 | 58 | const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size; |
xlu@959 | 59 | |
duke@435 | 60 | #define __ masm-> |
duke@435 | 61 | |
duke@435 | 62 | class SimpleRuntimeFrame { |
duke@435 | 63 | |
duke@435 | 64 | public: |
duke@435 | 65 | |
duke@435 | 66 | // Most of the runtime stubs have this simple frame layout. |
duke@435 | 67 | // This class exists to make the layout shared in one place. |
duke@435 | 68 | // Offsets are for compiler stack slots, which are jints. |
duke@435 | 69 | enum layout { |
duke@435 | 70 | // The frame sender code expects that rbp will be in the "natural" place and |
duke@435 | 71 | // will override any oopMap setting for it. We must therefore force the layout |
duke@435 | 72 | // so that it agrees with the frame sender code. |
duke@435 | 73 | rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt, |
duke@435 | 74 | rbp_off2, |
duke@435 | 75 | return_off, return_off2, |
duke@435 | 76 | framesize |
duke@435 | 77 | }; |
duke@435 | 78 | }; |
duke@435 | 79 | |
duke@435 | 80 | class RegisterSaver { |
duke@435 | 81 | // Capture info about frame layout. Layout offsets are in jint |
duke@435 | 82 | // units because compiler frame slots are jints. |
duke@435 | 83 | #define DEF_XMM_OFFS(regnum) xmm ## regnum ## _off = xmm_off + (regnum)*16/BytesPerInt, xmm ## regnum ## H_off |
duke@435 | 84 | enum layout { |
duke@435 | 85 | fpu_state_off = frame::arg_reg_save_area_bytes/BytesPerInt, // fxsave save area |
duke@435 | 86 | xmm_off = fpu_state_off + 160/BytesPerInt, // offset in fxsave save area |
duke@435 | 87 | DEF_XMM_OFFS(0), |
duke@435 | 88 | DEF_XMM_OFFS(1), |
duke@435 | 89 | DEF_XMM_OFFS(2), |
duke@435 | 90 | DEF_XMM_OFFS(3), |
duke@435 | 91 | DEF_XMM_OFFS(4), |
duke@435 | 92 | DEF_XMM_OFFS(5), |
duke@435 | 93 | DEF_XMM_OFFS(6), |
duke@435 | 94 | DEF_XMM_OFFS(7), |
duke@435 | 95 | DEF_XMM_OFFS(8), |
duke@435 | 96 | DEF_XMM_OFFS(9), |
duke@435 | 97 | DEF_XMM_OFFS(10), |
duke@435 | 98 | DEF_XMM_OFFS(11), |
duke@435 | 99 | DEF_XMM_OFFS(12), |
duke@435 | 100 | DEF_XMM_OFFS(13), |
duke@435 | 101 | DEF_XMM_OFFS(14), |
duke@435 | 102 | DEF_XMM_OFFS(15), |
duke@435 | 103 | fpu_state_end = fpu_state_off + ((FPUStateSizeInWords-1)*wordSize / BytesPerInt), |
duke@435 | 104 | fpu_stateH_end, |
duke@435 | 105 | r15_off, r15H_off, |
duke@435 | 106 | r14_off, r14H_off, |
duke@435 | 107 | r13_off, r13H_off, |
duke@435 | 108 | r12_off, r12H_off, |
duke@435 | 109 | r11_off, r11H_off, |
duke@435 | 110 | r10_off, r10H_off, |
duke@435 | 111 | r9_off, r9H_off, |
duke@435 | 112 | r8_off, r8H_off, |
duke@435 | 113 | rdi_off, rdiH_off, |
duke@435 | 114 | rsi_off, rsiH_off, |
duke@435 | 115 | ignore_off, ignoreH_off, // extra copy of rbp |
duke@435 | 116 | rsp_off, rspH_off, |
duke@435 | 117 | rbx_off, rbxH_off, |
duke@435 | 118 | rdx_off, rdxH_off, |
duke@435 | 119 | rcx_off, rcxH_off, |
duke@435 | 120 | rax_off, raxH_off, |
duke@435 | 121 | // 16-byte stack alignment fill word: see MacroAssembler::push/pop_IU_state |
duke@435 | 122 | align_off, alignH_off, |
duke@435 | 123 | flags_off, flagsH_off, |
duke@435 | 124 | // The frame sender code expects that rbp will be in the "natural" place and |
duke@435 | 125 | // will override any oopMap setting for it. We must therefore force the layout |
duke@435 | 126 | // so that it agrees with the frame sender code. |
duke@435 | 127 | rbp_off, rbpH_off, // copy of rbp we will restore |
duke@435 | 128 | return_off, returnH_off, // slot for return address |
duke@435 | 129 | reg_save_size // size in compiler stack slots |
duke@435 | 130 | }; |
duke@435 | 131 | |
duke@435 | 132 | public: |
duke@435 | 133 | static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words); |
duke@435 | 134 | static void restore_live_registers(MacroAssembler* masm); |
duke@435 | 135 | |
duke@435 | 136 | // Offsets into the register save area |
duke@435 | 137 | // Used by deoptimization when it is managing result register |
duke@435 | 138 | // values on its own |
duke@435 | 139 | |
duke@435 | 140 | static int rax_offset_in_bytes(void) { return BytesPerInt * rax_off; } |
never@739 | 141 | static int rdx_offset_in_bytes(void) { return BytesPerInt * rdx_off; } |
duke@435 | 142 | static int rbx_offset_in_bytes(void) { return BytesPerInt * rbx_off; } |
duke@435 | 143 | static int xmm0_offset_in_bytes(void) { return BytesPerInt * xmm0_off; } |
duke@435 | 144 | static int return_offset_in_bytes(void) { return BytesPerInt * return_off; } |
duke@435 | 145 | |
duke@435 | 146 | // During deoptimization only the result registers need to be restored, |
duke@435 | 147 | // all the other values have already been extracted. |
duke@435 | 148 | static void restore_result_registers(MacroAssembler* masm); |
duke@435 | 149 | }; |
duke@435 | 150 | |
duke@435 | 151 | OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words) { |
duke@435 | 152 | |
duke@435 | 153 | // Always make the frame size 16-byte aligned |
duke@435 | 154 | int frame_size_in_bytes = round_to(additional_frame_words*wordSize + |
duke@435 | 155 | reg_save_size*BytesPerInt, 16); |
duke@435 | 156 | // OopMap frame size is in compiler stack slots (jint's) not bytes or words |
duke@435 | 157 | int frame_size_in_slots = frame_size_in_bytes / BytesPerInt; |
duke@435 | 158 | // The caller will allocate additional_frame_words |
duke@435 | 159 | int additional_frame_slots = additional_frame_words*wordSize / BytesPerInt; |
duke@435 | 160 | // CodeBlob frame size is in words. |
duke@435 | 161 | int frame_size_in_words = frame_size_in_bytes / wordSize; |
duke@435 | 162 | *total_frame_words = frame_size_in_words; |
duke@435 | 163 | |
duke@435 | 164 | // Save registers, fpu state, and flags. |
duke@435 | 165 | // We assume caller has already pushed the return address onto the |
duke@435 | 166 | // stack, so rsp is 8-byte aligned here. |
duke@435 | 167 | // We push rpb twice in this sequence because we want the real rbp |
duke@435 | 168 | // to be under the return like a normal enter. |
duke@435 | 169 | |
duke@435 | 170 | __ enter(); // rsp becomes 16-byte aligned here |
duke@435 | 171 | __ push_CPU_state(); // Push a multiple of 16 bytes |
duke@435 | 172 | if (frame::arg_reg_save_area_bytes != 0) { |
duke@435 | 173 | // Allocate argument register save area |
never@739 | 174 | __ subptr(rsp, frame::arg_reg_save_area_bytes); |
duke@435 | 175 | } |
duke@435 | 176 | |
duke@435 | 177 | // Set an oopmap for the call site. This oopmap will map all |
duke@435 | 178 | // oop-registers and debug-info registers as callee-saved. This |
duke@435 | 179 | // will allow deoptimization at this safepoint to find all possible |
duke@435 | 180 | // debug-info recordings, as well as let GC find all oops. |
duke@435 | 181 | |
duke@435 | 182 | OopMapSet *oop_maps = new OopMapSet(); |
duke@435 | 183 | OopMap* map = new OopMap(frame_size_in_slots, 0); |
duke@435 | 184 | map->set_callee_saved(VMRegImpl::stack2reg( rax_off + additional_frame_slots), rax->as_VMReg()); |
duke@435 | 185 | map->set_callee_saved(VMRegImpl::stack2reg( rcx_off + additional_frame_slots), rcx->as_VMReg()); |
duke@435 | 186 | map->set_callee_saved(VMRegImpl::stack2reg( rdx_off + additional_frame_slots), rdx->as_VMReg()); |
duke@435 | 187 | map->set_callee_saved(VMRegImpl::stack2reg( rbx_off + additional_frame_slots), rbx->as_VMReg()); |
duke@435 | 188 | // rbp location is known implicitly by the frame sender code, needs no oopmap |
duke@435 | 189 | // and the location where rbp was saved by is ignored |
duke@435 | 190 | map->set_callee_saved(VMRegImpl::stack2reg( rsi_off + additional_frame_slots), rsi->as_VMReg()); |
duke@435 | 191 | map->set_callee_saved(VMRegImpl::stack2reg( rdi_off + additional_frame_slots), rdi->as_VMReg()); |
duke@435 | 192 | map->set_callee_saved(VMRegImpl::stack2reg( r8_off + additional_frame_slots), r8->as_VMReg()); |
duke@435 | 193 | map->set_callee_saved(VMRegImpl::stack2reg( r9_off + additional_frame_slots), r9->as_VMReg()); |
duke@435 | 194 | map->set_callee_saved(VMRegImpl::stack2reg( r10_off + additional_frame_slots), r10->as_VMReg()); |
duke@435 | 195 | map->set_callee_saved(VMRegImpl::stack2reg( r11_off + additional_frame_slots), r11->as_VMReg()); |
duke@435 | 196 | map->set_callee_saved(VMRegImpl::stack2reg( r12_off + additional_frame_slots), r12->as_VMReg()); |
duke@435 | 197 | map->set_callee_saved(VMRegImpl::stack2reg( r13_off + additional_frame_slots), r13->as_VMReg()); |
duke@435 | 198 | map->set_callee_saved(VMRegImpl::stack2reg( r14_off + additional_frame_slots), r14->as_VMReg()); |
duke@435 | 199 | map->set_callee_saved(VMRegImpl::stack2reg( r15_off + additional_frame_slots), r15->as_VMReg()); |
duke@435 | 200 | map->set_callee_saved(VMRegImpl::stack2reg(xmm0_off + additional_frame_slots), xmm0->as_VMReg()); |
duke@435 | 201 | map->set_callee_saved(VMRegImpl::stack2reg(xmm1_off + additional_frame_slots), xmm1->as_VMReg()); |
duke@435 | 202 | map->set_callee_saved(VMRegImpl::stack2reg(xmm2_off + additional_frame_slots), xmm2->as_VMReg()); |
duke@435 | 203 | map->set_callee_saved(VMRegImpl::stack2reg(xmm3_off + additional_frame_slots), xmm3->as_VMReg()); |
duke@435 | 204 | map->set_callee_saved(VMRegImpl::stack2reg(xmm4_off + additional_frame_slots), xmm4->as_VMReg()); |
duke@435 | 205 | map->set_callee_saved(VMRegImpl::stack2reg(xmm5_off + additional_frame_slots), xmm5->as_VMReg()); |
duke@435 | 206 | map->set_callee_saved(VMRegImpl::stack2reg(xmm6_off + additional_frame_slots), xmm6->as_VMReg()); |
duke@435 | 207 | map->set_callee_saved(VMRegImpl::stack2reg(xmm7_off + additional_frame_slots), xmm7->as_VMReg()); |
duke@435 | 208 | map->set_callee_saved(VMRegImpl::stack2reg(xmm8_off + additional_frame_slots), xmm8->as_VMReg()); |
duke@435 | 209 | map->set_callee_saved(VMRegImpl::stack2reg(xmm9_off + additional_frame_slots), xmm9->as_VMReg()); |
duke@435 | 210 | map->set_callee_saved(VMRegImpl::stack2reg(xmm10_off + additional_frame_slots), xmm10->as_VMReg()); |
duke@435 | 211 | map->set_callee_saved(VMRegImpl::stack2reg(xmm11_off + additional_frame_slots), xmm11->as_VMReg()); |
duke@435 | 212 | map->set_callee_saved(VMRegImpl::stack2reg(xmm12_off + additional_frame_slots), xmm12->as_VMReg()); |
duke@435 | 213 | map->set_callee_saved(VMRegImpl::stack2reg(xmm13_off + additional_frame_slots), xmm13->as_VMReg()); |
duke@435 | 214 | map->set_callee_saved(VMRegImpl::stack2reg(xmm14_off + additional_frame_slots), xmm14->as_VMReg()); |
duke@435 | 215 | map->set_callee_saved(VMRegImpl::stack2reg(xmm15_off + additional_frame_slots), xmm15->as_VMReg()); |
duke@435 | 216 | |
duke@435 | 217 | // %%% These should all be a waste but we'll keep things as they were for now |
duke@435 | 218 | if (true) { |
duke@435 | 219 | map->set_callee_saved(VMRegImpl::stack2reg( raxH_off + additional_frame_slots), |
duke@435 | 220 | rax->as_VMReg()->next()); |
duke@435 | 221 | map->set_callee_saved(VMRegImpl::stack2reg( rcxH_off + additional_frame_slots), |
duke@435 | 222 | rcx->as_VMReg()->next()); |
duke@435 | 223 | map->set_callee_saved(VMRegImpl::stack2reg( rdxH_off + additional_frame_slots), |
duke@435 | 224 | rdx->as_VMReg()->next()); |
duke@435 | 225 | map->set_callee_saved(VMRegImpl::stack2reg( rbxH_off + additional_frame_slots), |
duke@435 | 226 | rbx->as_VMReg()->next()); |
duke@435 | 227 | // rbp location is known implicitly by the frame sender code, needs no oopmap |
duke@435 | 228 | map->set_callee_saved(VMRegImpl::stack2reg( rsiH_off + additional_frame_slots), |
duke@435 | 229 | rsi->as_VMReg()->next()); |
duke@435 | 230 | map->set_callee_saved(VMRegImpl::stack2reg( rdiH_off + additional_frame_slots), |
duke@435 | 231 | rdi->as_VMReg()->next()); |
duke@435 | 232 | map->set_callee_saved(VMRegImpl::stack2reg( r8H_off + additional_frame_slots), |
duke@435 | 233 | r8->as_VMReg()->next()); |
duke@435 | 234 | map->set_callee_saved(VMRegImpl::stack2reg( r9H_off + additional_frame_slots), |
duke@435 | 235 | r9->as_VMReg()->next()); |
duke@435 | 236 | map->set_callee_saved(VMRegImpl::stack2reg( r10H_off + additional_frame_slots), |
duke@435 | 237 | r10->as_VMReg()->next()); |
duke@435 | 238 | map->set_callee_saved(VMRegImpl::stack2reg( r11H_off + additional_frame_slots), |
duke@435 | 239 | r11->as_VMReg()->next()); |
duke@435 | 240 | map->set_callee_saved(VMRegImpl::stack2reg( r12H_off + additional_frame_slots), |
duke@435 | 241 | r12->as_VMReg()->next()); |
duke@435 | 242 | map->set_callee_saved(VMRegImpl::stack2reg( r13H_off + additional_frame_slots), |
duke@435 | 243 | r13->as_VMReg()->next()); |
duke@435 | 244 | map->set_callee_saved(VMRegImpl::stack2reg( r14H_off + additional_frame_slots), |
duke@435 | 245 | r14->as_VMReg()->next()); |
duke@435 | 246 | map->set_callee_saved(VMRegImpl::stack2reg( r15H_off + additional_frame_slots), |
duke@435 | 247 | r15->as_VMReg()->next()); |
duke@435 | 248 | map->set_callee_saved(VMRegImpl::stack2reg(xmm0H_off + additional_frame_slots), |
duke@435 | 249 | xmm0->as_VMReg()->next()); |
duke@435 | 250 | map->set_callee_saved(VMRegImpl::stack2reg(xmm1H_off + additional_frame_slots), |
duke@435 | 251 | xmm1->as_VMReg()->next()); |
duke@435 | 252 | map->set_callee_saved(VMRegImpl::stack2reg(xmm2H_off + additional_frame_slots), |
duke@435 | 253 | xmm2->as_VMReg()->next()); |
duke@435 | 254 | map->set_callee_saved(VMRegImpl::stack2reg(xmm3H_off + additional_frame_slots), |
duke@435 | 255 | xmm3->as_VMReg()->next()); |
duke@435 | 256 | map->set_callee_saved(VMRegImpl::stack2reg(xmm4H_off + additional_frame_slots), |
duke@435 | 257 | xmm4->as_VMReg()->next()); |
duke@435 | 258 | map->set_callee_saved(VMRegImpl::stack2reg(xmm5H_off + additional_frame_slots), |
duke@435 | 259 | xmm5->as_VMReg()->next()); |
duke@435 | 260 | map->set_callee_saved(VMRegImpl::stack2reg(xmm6H_off + additional_frame_slots), |
duke@435 | 261 | xmm6->as_VMReg()->next()); |
duke@435 | 262 | map->set_callee_saved(VMRegImpl::stack2reg(xmm7H_off + additional_frame_slots), |
duke@435 | 263 | xmm7->as_VMReg()->next()); |
duke@435 | 264 | map->set_callee_saved(VMRegImpl::stack2reg(xmm8H_off + additional_frame_slots), |
duke@435 | 265 | xmm8->as_VMReg()->next()); |
duke@435 | 266 | map->set_callee_saved(VMRegImpl::stack2reg(xmm9H_off + additional_frame_slots), |
duke@435 | 267 | xmm9->as_VMReg()->next()); |
duke@435 | 268 | map->set_callee_saved(VMRegImpl::stack2reg(xmm10H_off + additional_frame_slots), |
duke@435 | 269 | xmm10->as_VMReg()->next()); |
duke@435 | 270 | map->set_callee_saved(VMRegImpl::stack2reg(xmm11H_off + additional_frame_slots), |
duke@435 | 271 | xmm11->as_VMReg()->next()); |
duke@435 | 272 | map->set_callee_saved(VMRegImpl::stack2reg(xmm12H_off + additional_frame_slots), |
duke@435 | 273 | xmm12->as_VMReg()->next()); |
duke@435 | 274 | map->set_callee_saved(VMRegImpl::stack2reg(xmm13H_off + additional_frame_slots), |
duke@435 | 275 | xmm13->as_VMReg()->next()); |
duke@435 | 276 | map->set_callee_saved(VMRegImpl::stack2reg(xmm14H_off + additional_frame_slots), |
duke@435 | 277 | xmm14->as_VMReg()->next()); |
duke@435 | 278 | map->set_callee_saved(VMRegImpl::stack2reg(xmm15H_off + additional_frame_slots), |
duke@435 | 279 | xmm15->as_VMReg()->next()); |
duke@435 | 280 | } |
duke@435 | 281 | |
duke@435 | 282 | return map; |
duke@435 | 283 | } |
duke@435 | 284 | |
duke@435 | 285 | void RegisterSaver::restore_live_registers(MacroAssembler* masm) { |
duke@435 | 286 | if (frame::arg_reg_save_area_bytes != 0) { |
duke@435 | 287 | // Pop arg register save area |
never@739 | 288 | __ addptr(rsp, frame::arg_reg_save_area_bytes); |
duke@435 | 289 | } |
duke@435 | 290 | // Recover CPU state |
duke@435 | 291 | __ pop_CPU_state(); |
duke@435 | 292 | // Get the rbp described implicitly by the calling convention (no oopMap) |
never@739 | 293 | __ pop(rbp); |
duke@435 | 294 | } |
duke@435 | 295 | |
duke@435 | 296 | void RegisterSaver::restore_result_registers(MacroAssembler* masm) { |
duke@435 | 297 | |
duke@435 | 298 | // Just restore result register. Only used by deoptimization. By |
duke@435 | 299 | // now any callee save register that needs to be restored to a c2 |
duke@435 | 300 | // caller of the deoptee has been extracted into the vframeArray |
duke@435 | 301 | // and will be stuffed into the c2i adapter we create for later |
duke@435 | 302 | // restoration so only result registers need to be restored here. |
duke@435 | 303 | |
duke@435 | 304 | // Restore fp result register |
duke@435 | 305 | __ movdbl(xmm0, Address(rsp, xmm0_offset_in_bytes())); |
duke@435 | 306 | // Restore integer result register |
never@739 | 307 | __ movptr(rax, Address(rsp, rax_offset_in_bytes())); |
never@739 | 308 | __ movptr(rdx, Address(rsp, rdx_offset_in_bytes())); |
never@739 | 309 | |
duke@435 | 310 | // Pop all of the register save are off the stack except the return address |
never@739 | 311 | __ addptr(rsp, return_offset_in_bytes()); |
duke@435 | 312 | } |
duke@435 | 313 | |
duke@435 | 314 | // The java_calling_convention describes stack locations as ideal slots on |
duke@435 | 315 | // a frame with no abi restrictions. Since we must observe abi restrictions |
duke@435 | 316 | // (like the placement of the register window) the slots must be biased by |
duke@435 | 317 | // the following value. |
duke@435 | 318 | static int reg2offset_in(VMReg r) { |
duke@435 | 319 | // Account for saved rbp and return address |
duke@435 | 320 | // This should really be in_preserve_stack_slots |
duke@435 | 321 | return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size; |
duke@435 | 322 | } |
duke@435 | 323 | |
duke@435 | 324 | static int reg2offset_out(VMReg r) { |
duke@435 | 325 | return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size; |
duke@435 | 326 | } |
duke@435 | 327 | |
duke@435 | 328 | // --------------------------------------------------------------------------- |
duke@435 | 329 | // Read the array of BasicTypes from a signature, and compute where the |
duke@435 | 330 | // arguments should go. Values in the VMRegPair regs array refer to 4-byte |
duke@435 | 331 | // quantities. Values less than VMRegImpl::stack0 are registers, those above |
duke@435 | 332 | // refer to 4-byte stack slots. All stack slots are based off of the stack pointer |
duke@435 | 333 | // as framesizes are fixed. |
duke@435 | 334 | // VMRegImpl::stack0 refers to the first slot 0(sp). |
duke@435 | 335 | // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher. Register |
duke@435 | 336 | // up to RegisterImpl::number_of_registers) are the 64-bit |
duke@435 | 337 | // integer registers. |
duke@435 | 338 | |
duke@435 | 339 | // Note: the INPUTS in sig_bt are in units of Java argument words, which are |
duke@435 | 340 | // either 32-bit or 64-bit depending on the build. The OUTPUTS are in 32-bit |
duke@435 | 341 | // units regardless of build. Of course for i486 there is no 64 bit build |
duke@435 | 342 | |
duke@435 | 343 | // The Java calling convention is a "shifted" version of the C ABI. |
duke@435 | 344 | // By skipping the first C ABI register we can call non-static jni methods |
duke@435 | 345 | // with small numbers of arguments without having to shuffle the arguments |
duke@435 | 346 | // at all. Since we control the java ABI we ought to at least get some |
duke@435 | 347 | // advantage out of it. |
duke@435 | 348 | |
duke@435 | 349 | int SharedRuntime::java_calling_convention(const BasicType *sig_bt, |
duke@435 | 350 | VMRegPair *regs, |
duke@435 | 351 | int total_args_passed, |
duke@435 | 352 | int is_outgoing) { |
duke@435 | 353 | |
duke@435 | 354 | // Create the mapping between argument positions and |
duke@435 | 355 | // registers. |
duke@435 | 356 | static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = { |
duke@435 | 357 | j_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4, j_rarg5 |
duke@435 | 358 | }; |
duke@435 | 359 | static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_j] = { |
duke@435 | 360 | j_farg0, j_farg1, j_farg2, j_farg3, |
duke@435 | 361 | j_farg4, j_farg5, j_farg6, j_farg7 |
duke@435 | 362 | }; |
duke@435 | 363 | |
duke@435 | 364 | |
duke@435 | 365 | uint int_args = 0; |
duke@435 | 366 | uint fp_args = 0; |
duke@435 | 367 | uint stk_args = 0; // inc by 2 each time |
duke@435 | 368 | |
duke@435 | 369 | for (int i = 0; i < total_args_passed; i++) { |
duke@435 | 370 | switch (sig_bt[i]) { |
duke@435 | 371 | case T_BOOLEAN: |
duke@435 | 372 | case T_CHAR: |
duke@435 | 373 | case T_BYTE: |
duke@435 | 374 | case T_SHORT: |
duke@435 | 375 | case T_INT: |
duke@435 | 376 | if (int_args < Argument::n_int_register_parameters_j) { |
duke@435 | 377 | regs[i].set1(INT_ArgReg[int_args++]->as_VMReg()); |
duke@435 | 378 | } else { |
duke@435 | 379 | regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
duke@435 | 380 | stk_args += 2; |
duke@435 | 381 | } |
duke@435 | 382 | break; |
duke@435 | 383 | case T_VOID: |
duke@435 | 384 | // halves of T_LONG or T_DOUBLE |
duke@435 | 385 | assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half"); |
duke@435 | 386 | regs[i].set_bad(); |
duke@435 | 387 | break; |
duke@435 | 388 | case T_LONG: |
duke@435 | 389 | assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
duke@435 | 390 | // fall through |
duke@435 | 391 | case T_OBJECT: |
duke@435 | 392 | case T_ARRAY: |
duke@435 | 393 | case T_ADDRESS: |
duke@435 | 394 | if (int_args < Argument::n_int_register_parameters_j) { |
duke@435 | 395 | regs[i].set2(INT_ArgReg[int_args++]->as_VMReg()); |
duke@435 | 396 | } else { |
duke@435 | 397 | regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
duke@435 | 398 | stk_args += 2; |
duke@435 | 399 | } |
duke@435 | 400 | break; |
duke@435 | 401 | case T_FLOAT: |
duke@435 | 402 | if (fp_args < Argument::n_float_register_parameters_j) { |
duke@435 | 403 | regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg()); |
duke@435 | 404 | } else { |
duke@435 | 405 | regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
duke@435 | 406 | stk_args += 2; |
duke@435 | 407 | } |
duke@435 | 408 | break; |
duke@435 | 409 | case T_DOUBLE: |
duke@435 | 410 | assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
duke@435 | 411 | if (fp_args < Argument::n_float_register_parameters_j) { |
duke@435 | 412 | regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg()); |
duke@435 | 413 | } else { |
duke@435 | 414 | regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
duke@435 | 415 | stk_args += 2; |
duke@435 | 416 | } |
duke@435 | 417 | break; |
duke@435 | 418 | default: |
duke@435 | 419 | ShouldNotReachHere(); |
duke@435 | 420 | break; |
duke@435 | 421 | } |
duke@435 | 422 | } |
duke@435 | 423 | |
duke@435 | 424 | return round_to(stk_args, 2); |
duke@435 | 425 | } |
duke@435 | 426 | |
duke@435 | 427 | // Patch the callers callsite with entry to compiled code if it exists. |
duke@435 | 428 | static void patch_callers_callsite(MacroAssembler *masm) { |
duke@435 | 429 | Label L; |
duke@435 | 430 | __ verify_oop(rbx); |
never@739 | 431 | __ cmpptr(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int32_t)NULL_WORD); |
duke@435 | 432 | __ jcc(Assembler::equal, L); |
duke@435 | 433 | |
duke@435 | 434 | // Save the current stack pointer |
never@739 | 435 | __ mov(r13, rsp); |
duke@435 | 436 | // Schedule the branch target address early. |
duke@435 | 437 | // Call into the VM to patch the caller, then jump to compiled callee |
duke@435 | 438 | // rax isn't live so capture return address while we easily can |
never@739 | 439 | __ movptr(rax, Address(rsp, 0)); |
duke@435 | 440 | |
duke@435 | 441 | // align stack so push_CPU_state doesn't fault |
never@739 | 442 | __ andptr(rsp, -(StackAlignmentInBytes)); |
duke@435 | 443 | __ push_CPU_state(); |
duke@435 | 444 | |
duke@435 | 445 | |
duke@435 | 446 | __ verify_oop(rbx); |
duke@435 | 447 | // VM needs caller's callsite |
duke@435 | 448 | // VM needs target method |
duke@435 | 449 | // This needs to be a long call since we will relocate this adapter to |
duke@435 | 450 | // the codeBuffer and it may not reach |
duke@435 | 451 | |
duke@435 | 452 | // Allocate argument register save area |
duke@435 | 453 | if (frame::arg_reg_save_area_bytes != 0) { |
never@739 | 454 | __ subptr(rsp, frame::arg_reg_save_area_bytes); |
duke@435 | 455 | } |
never@739 | 456 | __ mov(c_rarg0, rbx); |
never@739 | 457 | __ mov(c_rarg1, rax); |
duke@435 | 458 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite))); |
duke@435 | 459 | |
duke@435 | 460 | // De-allocate argument register save area |
duke@435 | 461 | if (frame::arg_reg_save_area_bytes != 0) { |
never@739 | 462 | __ addptr(rsp, frame::arg_reg_save_area_bytes); |
duke@435 | 463 | } |
duke@435 | 464 | |
duke@435 | 465 | __ pop_CPU_state(); |
duke@435 | 466 | // restore sp |
never@739 | 467 | __ mov(rsp, r13); |
duke@435 | 468 | __ bind(L); |
duke@435 | 469 | } |
duke@435 | 470 | |
duke@435 | 471 | |
duke@435 | 472 | static void gen_c2i_adapter(MacroAssembler *masm, |
duke@435 | 473 | int total_args_passed, |
duke@435 | 474 | int comp_args_on_stack, |
duke@435 | 475 | const BasicType *sig_bt, |
duke@435 | 476 | const VMRegPair *regs, |
duke@435 | 477 | Label& skip_fixup) { |
duke@435 | 478 | // Before we get into the guts of the C2I adapter, see if we should be here |
duke@435 | 479 | // at all. We've come from compiled code and are attempting to jump to the |
duke@435 | 480 | // interpreter, which means the caller made a static call to get here |
duke@435 | 481 | // (vcalls always get a compiled target if there is one). Check for a |
duke@435 | 482 | // compiled target. If there is one, we need to patch the caller's call. |
duke@435 | 483 | patch_callers_callsite(masm); |
duke@435 | 484 | |
duke@435 | 485 | __ bind(skip_fixup); |
duke@435 | 486 | |
duke@435 | 487 | // Since all args are passed on the stack, total_args_passed * |
duke@435 | 488 | // Interpreter::stackElementSize is the space we need. Plus 1 because |
duke@435 | 489 | // we also account for the return address location since |
duke@435 | 490 | // we store it first rather than hold it in rax across all the shuffling |
duke@435 | 491 | |
twisti@1861 | 492 | int extraspace = (total_args_passed * Interpreter::stackElementSize) + wordSize; |
duke@435 | 493 | |
duke@435 | 494 | // stack is aligned, keep it that way |
duke@435 | 495 | extraspace = round_to(extraspace, 2*wordSize); |
duke@435 | 496 | |
duke@435 | 497 | // Get return address |
never@739 | 498 | __ pop(rax); |
duke@435 | 499 | |
duke@435 | 500 | // set senderSP value |
never@739 | 501 | __ mov(r13, rsp); |
never@739 | 502 | |
never@739 | 503 | __ subptr(rsp, extraspace); |
duke@435 | 504 | |
duke@435 | 505 | // Store the return address in the expected location |
never@739 | 506 | __ movptr(Address(rsp, 0), rax); |
duke@435 | 507 | |
duke@435 | 508 | // Now write the args into the outgoing interpreter space |
duke@435 | 509 | for (int i = 0; i < total_args_passed; i++) { |
duke@435 | 510 | if (sig_bt[i] == T_VOID) { |
duke@435 | 511 | assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); |
duke@435 | 512 | continue; |
duke@435 | 513 | } |
duke@435 | 514 | |
duke@435 | 515 | // offset to start parameters |
twisti@1861 | 516 | int st_off = (total_args_passed - i) * Interpreter::stackElementSize; |
twisti@1861 | 517 | int next_off = st_off - Interpreter::stackElementSize; |
duke@435 | 518 | |
duke@435 | 519 | // Say 4 args: |
duke@435 | 520 | // i st_off |
duke@435 | 521 | // 0 32 T_LONG |
duke@435 | 522 | // 1 24 T_VOID |
duke@435 | 523 | // 2 16 T_OBJECT |
duke@435 | 524 | // 3 8 T_BOOL |
duke@435 | 525 | // - 0 return address |
duke@435 | 526 | // |
duke@435 | 527 | // However to make thing extra confusing. Because we can fit a long/double in |
duke@435 | 528 | // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter |
duke@435 | 529 | // leaves one slot empty and only stores to a single slot. In this case the |
duke@435 | 530 | // slot that is occupied is the T_VOID slot. See I said it was confusing. |
duke@435 | 531 | |
duke@435 | 532 | VMReg r_1 = regs[i].first(); |
duke@435 | 533 | VMReg r_2 = regs[i].second(); |
duke@435 | 534 | if (!r_1->is_valid()) { |
duke@435 | 535 | assert(!r_2->is_valid(), ""); |
duke@435 | 536 | continue; |
duke@435 | 537 | } |
duke@435 | 538 | if (r_1->is_stack()) { |
duke@435 | 539 | // memory to memory use rax |
duke@435 | 540 | int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace; |
duke@435 | 541 | if (!r_2->is_valid()) { |
duke@435 | 542 | // sign extend?? |
duke@435 | 543 | __ movl(rax, Address(rsp, ld_off)); |
never@739 | 544 | __ movptr(Address(rsp, st_off), rax); |
duke@435 | 545 | |
duke@435 | 546 | } else { |
duke@435 | 547 | |
duke@435 | 548 | __ movq(rax, Address(rsp, ld_off)); |
duke@435 | 549 | |
duke@435 | 550 | // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG |
duke@435 | 551 | // T_DOUBLE and T_LONG use two slots in the interpreter |
duke@435 | 552 | if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) { |
duke@435 | 553 | // ld_off == LSW, ld_off+wordSize == MSW |
duke@435 | 554 | // st_off == MSW, next_off == LSW |
duke@435 | 555 | __ movq(Address(rsp, next_off), rax); |
duke@435 | 556 | #ifdef ASSERT |
duke@435 | 557 | // Overwrite the unused slot with known junk |
duke@435 | 558 | __ mov64(rax, CONST64(0xdeadffffdeadaaaa)); |
never@739 | 559 | __ movptr(Address(rsp, st_off), rax); |
duke@435 | 560 | #endif /* ASSERT */ |
duke@435 | 561 | } else { |
duke@435 | 562 | __ movq(Address(rsp, st_off), rax); |
duke@435 | 563 | } |
duke@435 | 564 | } |
duke@435 | 565 | } else if (r_1->is_Register()) { |
duke@435 | 566 | Register r = r_1->as_Register(); |
duke@435 | 567 | if (!r_2->is_valid()) { |
duke@435 | 568 | // must be only an int (or less ) so move only 32bits to slot |
duke@435 | 569 | // why not sign extend?? |
duke@435 | 570 | __ movl(Address(rsp, st_off), r); |
duke@435 | 571 | } else { |
duke@435 | 572 | // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG |
duke@435 | 573 | // T_DOUBLE and T_LONG use two slots in the interpreter |
duke@435 | 574 | if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) { |
duke@435 | 575 | // long/double in gpr |
duke@435 | 576 | #ifdef ASSERT |
duke@435 | 577 | // Overwrite the unused slot with known junk |
duke@435 | 578 | __ mov64(rax, CONST64(0xdeadffffdeadaaab)); |
never@739 | 579 | __ movptr(Address(rsp, st_off), rax); |
duke@435 | 580 | #endif /* ASSERT */ |
duke@435 | 581 | __ movq(Address(rsp, next_off), r); |
duke@435 | 582 | } else { |
never@739 | 583 | __ movptr(Address(rsp, st_off), r); |
duke@435 | 584 | } |
duke@435 | 585 | } |
duke@435 | 586 | } else { |
duke@435 | 587 | assert(r_1->is_XMMRegister(), ""); |
duke@435 | 588 | if (!r_2->is_valid()) { |
duke@435 | 589 | // only a float use just part of the slot |
duke@435 | 590 | __ movflt(Address(rsp, st_off), r_1->as_XMMRegister()); |
duke@435 | 591 | } else { |
duke@435 | 592 | #ifdef ASSERT |
duke@435 | 593 | // Overwrite the unused slot with known junk |
duke@435 | 594 | __ mov64(rax, CONST64(0xdeadffffdeadaaac)); |
never@739 | 595 | __ movptr(Address(rsp, st_off), rax); |
duke@435 | 596 | #endif /* ASSERT */ |
duke@435 | 597 | __ movdbl(Address(rsp, next_off), r_1->as_XMMRegister()); |
duke@435 | 598 | } |
duke@435 | 599 | } |
duke@435 | 600 | } |
duke@435 | 601 | |
duke@435 | 602 | // Schedule the branch target address early. |
never@739 | 603 | __ movptr(rcx, Address(rbx, in_bytes(methodOopDesc::interpreter_entry_offset()))); |
duke@435 | 604 | __ jmp(rcx); |
duke@435 | 605 | } |
duke@435 | 606 | |
duke@435 | 607 | static void gen_i2c_adapter(MacroAssembler *masm, |
duke@435 | 608 | int total_args_passed, |
duke@435 | 609 | int comp_args_on_stack, |
duke@435 | 610 | const BasicType *sig_bt, |
duke@435 | 611 | const VMRegPair *regs) { |
duke@435 | 612 | |
duke@435 | 613 | // Note: r13 contains the senderSP on entry. We must preserve it since |
duke@435 | 614 | // we may do a i2c -> c2i transition if we lose a race where compiled |
duke@435 | 615 | // code goes non-entrant while we get args ready. |
duke@435 | 616 | // In addition we use r13 to locate all the interpreter args as |
duke@435 | 617 | // we must align the stack to 16 bytes on an i2c entry else we |
duke@435 | 618 | // lose alignment we expect in all compiled code and register |
duke@435 | 619 | // save code can segv when fxsave instructions find improperly |
duke@435 | 620 | // aligned stack pointer. |
duke@435 | 621 | |
twisti@2552 | 622 | // Pick up the return address |
never@739 | 623 | __ movptr(rax, Address(rsp, 0)); |
duke@435 | 624 | |
twisti@1570 | 625 | // Must preserve original SP for loading incoming arguments because |
twisti@1570 | 626 | // we need to align the outgoing SP for compiled code. |
twisti@1570 | 627 | __ movptr(r11, rsp); |
twisti@1570 | 628 | |
duke@435 | 629 | // Cut-out for having no stack args. Since up to 2 int/oop args are passed |
duke@435 | 630 | // in registers, we will occasionally have no stack args. |
duke@435 | 631 | int comp_words_on_stack = 0; |
duke@435 | 632 | if (comp_args_on_stack) { |
duke@435 | 633 | // Sig words on the stack are greater-than VMRegImpl::stack0. Those in |
duke@435 | 634 | // registers are below. By subtracting stack0, we either get a negative |
duke@435 | 635 | // number (all values in registers) or the maximum stack slot accessed. |
duke@435 | 636 | |
duke@435 | 637 | // Convert 4-byte c2 stack slots to words. |
duke@435 | 638 | comp_words_on_stack = round_to(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord; |
duke@435 | 639 | // Round up to miminum stack alignment, in wordSize |
duke@435 | 640 | comp_words_on_stack = round_to(comp_words_on_stack, 2); |
never@739 | 641 | __ subptr(rsp, comp_words_on_stack * wordSize); |
duke@435 | 642 | } |
duke@435 | 643 | |
duke@435 | 644 | |
duke@435 | 645 | // Ensure compiled code always sees stack at proper alignment |
never@739 | 646 | __ andptr(rsp, -16); |
duke@435 | 647 | |
duke@435 | 648 | // push the return address and misalign the stack that youngest frame always sees |
duke@435 | 649 | // as far as the placement of the call instruction |
never@739 | 650 | __ push(rax); |
duke@435 | 651 | |
twisti@1570 | 652 | // Put saved SP in another register |
twisti@1570 | 653 | const Register saved_sp = rax; |
twisti@1570 | 654 | __ movptr(saved_sp, r11); |
twisti@1570 | 655 | |
duke@435 | 656 | // Will jump to the compiled code just as if compiled code was doing it. |
duke@435 | 657 | // Pre-load the register-jump target early, to schedule it better. |
never@739 | 658 | __ movptr(r11, Address(rbx, in_bytes(methodOopDesc::from_compiled_offset()))); |
duke@435 | 659 | |
duke@435 | 660 | // Now generate the shuffle code. Pick up all register args and move the |
duke@435 | 661 | // rest through the floating point stack top. |
duke@435 | 662 | for (int i = 0; i < total_args_passed; i++) { |
duke@435 | 663 | if (sig_bt[i] == T_VOID) { |
duke@435 | 664 | // Longs and doubles are passed in native word order, but misaligned |
duke@435 | 665 | // in the 32-bit build. |
duke@435 | 666 | assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); |
duke@435 | 667 | continue; |
duke@435 | 668 | } |
duke@435 | 669 | |
duke@435 | 670 | // Pick up 0, 1 or 2 words from SP+offset. |
duke@435 | 671 | |
duke@435 | 672 | assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(), |
duke@435 | 673 | "scrambled load targets?"); |
duke@435 | 674 | // Load in argument order going down. |
twisti@1861 | 675 | int ld_off = (total_args_passed - i)*Interpreter::stackElementSize; |
duke@435 | 676 | // Point to interpreter value (vs. tag) |
twisti@1861 | 677 | int next_off = ld_off - Interpreter::stackElementSize; |
duke@435 | 678 | // |
duke@435 | 679 | // |
duke@435 | 680 | // |
duke@435 | 681 | VMReg r_1 = regs[i].first(); |
duke@435 | 682 | VMReg r_2 = regs[i].second(); |
duke@435 | 683 | if (!r_1->is_valid()) { |
duke@435 | 684 | assert(!r_2->is_valid(), ""); |
duke@435 | 685 | continue; |
duke@435 | 686 | } |
duke@435 | 687 | if (r_1->is_stack()) { |
duke@435 | 688 | // Convert stack slot to an SP offset (+ wordSize to account for return address ) |
duke@435 | 689 | int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize; |
twisti@1570 | 690 | |
twisti@1570 | 691 | // We can use r13 as a temp here because compiled code doesn't need r13 as an input |
twisti@1570 | 692 | // and if we end up going thru a c2i because of a miss a reasonable value of r13 |
twisti@1570 | 693 | // will be generated. |
duke@435 | 694 | if (!r_2->is_valid()) { |
duke@435 | 695 | // sign extend??? |
twisti@1570 | 696 | __ movl(r13, Address(saved_sp, ld_off)); |
twisti@1570 | 697 | __ movptr(Address(rsp, st_off), r13); |
duke@435 | 698 | } else { |
duke@435 | 699 | // |
duke@435 | 700 | // We are using two optoregs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE |
duke@435 | 701 | // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case |
duke@435 | 702 | // So we must adjust where to pick up the data to match the interpreter. |
duke@435 | 703 | // |
duke@435 | 704 | // Interpreter local[n] == MSW, local[n+1] == LSW however locals |
duke@435 | 705 | // are accessed as negative so LSW is at LOW address |
duke@435 | 706 | |
duke@435 | 707 | // ld_off is MSW so get LSW |
duke@435 | 708 | const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)? |
duke@435 | 709 | next_off : ld_off; |
twisti@1570 | 710 | __ movq(r13, Address(saved_sp, offset)); |
duke@435 | 711 | // st_off is LSW (i.e. reg.first()) |
twisti@1570 | 712 | __ movq(Address(rsp, st_off), r13); |
duke@435 | 713 | } |
duke@435 | 714 | } else if (r_1->is_Register()) { // Register argument |
duke@435 | 715 | Register r = r_1->as_Register(); |
duke@435 | 716 | assert(r != rax, "must be different"); |
duke@435 | 717 | if (r_2->is_valid()) { |
duke@435 | 718 | // |
duke@435 | 719 | // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE |
duke@435 | 720 | // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case |
duke@435 | 721 | // So we must adjust where to pick up the data to match the interpreter. |
duke@435 | 722 | |
duke@435 | 723 | const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)? |
duke@435 | 724 | next_off : ld_off; |
duke@435 | 725 | |
duke@435 | 726 | // this can be a misaligned move |
twisti@1570 | 727 | __ movq(r, Address(saved_sp, offset)); |
duke@435 | 728 | } else { |
duke@435 | 729 | // sign extend and use a full word? |
twisti@1570 | 730 | __ movl(r, Address(saved_sp, ld_off)); |
duke@435 | 731 | } |
duke@435 | 732 | } else { |
duke@435 | 733 | if (!r_2->is_valid()) { |
twisti@1570 | 734 | __ movflt(r_1->as_XMMRegister(), Address(saved_sp, ld_off)); |
duke@435 | 735 | } else { |
twisti@1570 | 736 | __ movdbl(r_1->as_XMMRegister(), Address(saved_sp, next_off)); |
duke@435 | 737 | } |
duke@435 | 738 | } |
duke@435 | 739 | } |
duke@435 | 740 | |
duke@435 | 741 | // 6243940 We might end up in handle_wrong_method if |
duke@435 | 742 | // the callee is deoptimized as we race thru here. If that |
duke@435 | 743 | // happens we don't want to take a safepoint because the |
duke@435 | 744 | // caller frame will look interpreted and arguments are now |
duke@435 | 745 | // "compiled" so it is much better to make this transition |
duke@435 | 746 | // invisible to the stack walking code. Unfortunately if |
duke@435 | 747 | // we try and find the callee by normal means a safepoint |
duke@435 | 748 | // is possible. So we stash the desired callee in the thread |
duke@435 | 749 | // and the vm will find there should this case occur. |
duke@435 | 750 | |
never@739 | 751 | __ movptr(Address(r15_thread, JavaThread::callee_target_offset()), rbx); |
duke@435 | 752 | |
duke@435 | 753 | // put methodOop where a c2i would expect should we end up there |
duke@435 | 754 | // only needed becaus eof c2 resolve stubs return methodOop as a result in |
duke@435 | 755 | // rax |
never@739 | 756 | __ mov(rax, rbx); |
duke@435 | 757 | __ jmp(r11); |
duke@435 | 758 | } |
duke@435 | 759 | |
duke@435 | 760 | // --------------------------------------------------------------- |
duke@435 | 761 | AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm, |
duke@435 | 762 | int total_args_passed, |
duke@435 | 763 | int comp_args_on_stack, |
duke@435 | 764 | const BasicType *sig_bt, |
never@1622 | 765 | const VMRegPair *regs, |
never@1622 | 766 | AdapterFingerPrint* fingerprint) { |
duke@435 | 767 | address i2c_entry = __ pc(); |
duke@435 | 768 | |
duke@435 | 769 | gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs); |
duke@435 | 770 | |
duke@435 | 771 | // ------------------------------------------------------------------------- |
duke@435 | 772 | // Generate a C2I adapter. On entry we know rbx holds the methodOop during calls |
duke@435 | 773 | // to the interpreter. The args start out packed in the compiled layout. They |
duke@435 | 774 | // need to be unpacked into the interpreter layout. This will almost always |
duke@435 | 775 | // require some stack space. We grow the current (compiled) stack, then repack |
duke@435 | 776 | // the args. We finally end in a jump to the generic interpreter entry point. |
duke@435 | 777 | // On exit from the interpreter, the interpreter will restore our SP (lest the |
duke@435 | 778 | // compiled code, which relys solely on SP and not RBP, get sick). |
duke@435 | 779 | |
duke@435 | 780 | address c2i_unverified_entry = __ pc(); |
duke@435 | 781 | Label skip_fixup; |
duke@435 | 782 | Label ok; |
duke@435 | 783 | |
duke@435 | 784 | Register holder = rax; |
duke@435 | 785 | Register receiver = j_rarg0; |
duke@435 | 786 | Register temp = rbx; |
duke@435 | 787 | |
duke@435 | 788 | { |
duke@435 | 789 | __ verify_oop(holder); |
coleenp@548 | 790 | __ load_klass(temp, receiver); |
duke@435 | 791 | __ verify_oop(temp); |
duke@435 | 792 | |
never@739 | 793 | __ cmpptr(temp, Address(holder, compiledICHolderOopDesc::holder_klass_offset())); |
never@739 | 794 | __ movptr(rbx, Address(holder, compiledICHolderOopDesc::holder_method_offset())); |
duke@435 | 795 | __ jcc(Assembler::equal, ok); |
duke@435 | 796 | __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); |
duke@435 | 797 | |
duke@435 | 798 | __ bind(ok); |
duke@435 | 799 | // Method might have been compiled since the call site was patched to |
duke@435 | 800 | // interpreted if that is the case treat it as a miss so we can get |
duke@435 | 801 | // the call site corrected. |
never@739 | 802 | __ cmpptr(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int32_t)NULL_WORD); |
duke@435 | 803 | __ jcc(Assembler::equal, skip_fixup); |
duke@435 | 804 | __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); |
duke@435 | 805 | } |
duke@435 | 806 | |
duke@435 | 807 | address c2i_entry = __ pc(); |
duke@435 | 808 | |
duke@435 | 809 | gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup); |
duke@435 | 810 | |
duke@435 | 811 | __ flush(); |
never@1622 | 812 | return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry); |
duke@435 | 813 | } |
duke@435 | 814 | |
duke@435 | 815 | int SharedRuntime::c_calling_convention(const BasicType *sig_bt, |
duke@435 | 816 | VMRegPair *regs, |
duke@435 | 817 | int total_args_passed) { |
duke@435 | 818 | // We return the amount of VMRegImpl stack slots we need to reserve for all |
duke@435 | 819 | // the arguments NOT counting out_preserve_stack_slots. |
duke@435 | 820 | |
duke@435 | 821 | // NOTE: These arrays will have to change when c1 is ported |
duke@435 | 822 | #ifdef _WIN64 |
duke@435 | 823 | static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = { |
duke@435 | 824 | c_rarg0, c_rarg1, c_rarg2, c_rarg3 |
duke@435 | 825 | }; |
duke@435 | 826 | static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = { |
duke@435 | 827 | c_farg0, c_farg1, c_farg2, c_farg3 |
duke@435 | 828 | }; |
duke@435 | 829 | #else |
duke@435 | 830 | static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = { |
duke@435 | 831 | c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, c_rarg5 |
duke@435 | 832 | }; |
duke@435 | 833 | static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = { |
duke@435 | 834 | c_farg0, c_farg1, c_farg2, c_farg3, |
duke@435 | 835 | c_farg4, c_farg5, c_farg6, c_farg7 |
duke@435 | 836 | }; |
duke@435 | 837 | #endif // _WIN64 |
duke@435 | 838 | |
duke@435 | 839 | |
duke@435 | 840 | uint int_args = 0; |
duke@435 | 841 | uint fp_args = 0; |
duke@435 | 842 | uint stk_args = 0; // inc by 2 each time |
duke@435 | 843 | |
duke@435 | 844 | for (int i = 0; i < total_args_passed; i++) { |
duke@435 | 845 | switch (sig_bt[i]) { |
duke@435 | 846 | case T_BOOLEAN: |
duke@435 | 847 | case T_CHAR: |
duke@435 | 848 | case T_BYTE: |
duke@435 | 849 | case T_SHORT: |
duke@435 | 850 | case T_INT: |
duke@435 | 851 | if (int_args < Argument::n_int_register_parameters_c) { |
duke@435 | 852 | regs[i].set1(INT_ArgReg[int_args++]->as_VMReg()); |
duke@435 | 853 | #ifdef _WIN64 |
duke@435 | 854 | fp_args++; |
duke@435 | 855 | // Allocate slots for callee to stuff register args the stack. |
duke@435 | 856 | stk_args += 2; |
duke@435 | 857 | #endif |
duke@435 | 858 | } else { |
duke@435 | 859 | regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
duke@435 | 860 | stk_args += 2; |
duke@435 | 861 | } |
duke@435 | 862 | break; |
duke@435 | 863 | case T_LONG: |
duke@435 | 864 | assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
duke@435 | 865 | // fall through |
duke@435 | 866 | case T_OBJECT: |
duke@435 | 867 | case T_ARRAY: |
duke@435 | 868 | case T_ADDRESS: |
duke@435 | 869 | if (int_args < Argument::n_int_register_parameters_c) { |
duke@435 | 870 | regs[i].set2(INT_ArgReg[int_args++]->as_VMReg()); |
duke@435 | 871 | #ifdef _WIN64 |
duke@435 | 872 | fp_args++; |
duke@435 | 873 | stk_args += 2; |
duke@435 | 874 | #endif |
duke@435 | 875 | } else { |
duke@435 | 876 | regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
duke@435 | 877 | stk_args += 2; |
duke@435 | 878 | } |
duke@435 | 879 | break; |
duke@435 | 880 | case T_FLOAT: |
duke@435 | 881 | if (fp_args < Argument::n_float_register_parameters_c) { |
duke@435 | 882 | regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg()); |
duke@435 | 883 | #ifdef _WIN64 |
duke@435 | 884 | int_args++; |
duke@435 | 885 | // Allocate slots for callee to stuff register args the stack. |
duke@435 | 886 | stk_args += 2; |
duke@435 | 887 | #endif |
duke@435 | 888 | } else { |
duke@435 | 889 | regs[i].set1(VMRegImpl::stack2reg(stk_args)); |
duke@435 | 890 | stk_args += 2; |
duke@435 | 891 | } |
duke@435 | 892 | break; |
duke@435 | 893 | case T_DOUBLE: |
duke@435 | 894 | assert(sig_bt[i + 1] == T_VOID, "expecting half"); |
duke@435 | 895 | if (fp_args < Argument::n_float_register_parameters_c) { |
duke@435 | 896 | regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg()); |
duke@435 | 897 | #ifdef _WIN64 |
duke@435 | 898 | int_args++; |
duke@435 | 899 | // Allocate slots for callee to stuff register args the stack. |
duke@435 | 900 | stk_args += 2; |
duke@435 | 901 | #endif |
duke@435 | 902 | } else { |
duke@435 | 903 | regs[i].set2(VMRegImpl::stack2reg(stk_args)); |
duke@435 | 904 | stk_args += 2; |
duke@435 | 905 | } |
duke@435 | 906 | break; |
duke@435 | 907 | case T_VOID: // Halves of longs and doubles |
duke@435 | 908 | assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half"); |
duke@435 | 909 | regs[i].set_bad(); |
duke@435 | 910 | break; |
duke@435 | 911 | default: |
duke@435 | 912 | ShouldNotReachHere(); |
duke@435 | 913 | break; |
duke@435 | 914 | } |
duke@435 | 915 | } |
duke@435 | 916 | #ifdef _WIN64 |
duke@435 | 917 | // windows abi requires that we always allocate enough stack space |
duke@435 | 918 | // for 4 64bit registers to be stored down. |
duke@435 | 919 | if (stk_args < 8) { |
duke@435 | 920 | stk_args = 8; |
duke@435 | 921 | } |
duke@435 | 922 | #endif // _WIN64 |
duke@435 | 923 | |
duke@435 | 924 | return stk_args; |
duke@435 | 925 | } |
duke@435 | 926 | |
duke@435 | 927 | // On 64 bit we will store integer like items to the stack as |
duke@435 | 928 | // 64 bits items (sparc abi) even though java would only store |
duke@435 | 929 | // 32bits for a parameter. On 32bit it will simply be 32 bits |
duke@435 | 930 | // So this routine will do 32->32 on 32bit and 32->64 on 64bit |
duke@435 | 931 | static void move32_64(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
duke@435 | 932 | if (src.first()->is_stack()) { |
duke@435 | 933 | if (dst.first()->is_stack()) { |
duke@435 | 934 | // stack to stack |
duke@435 | 935 | __ movslq(rax, Address(rbp, reg2offset_in(src.first()))); |
duke@435 | 936 | __ movq(Address(rsp, reg2offset_out(dst.first())), rax); |
duke@435 | 937 | } else { |
duke@435 | 938 | // stack to reg |
duke@435 | 939 | __ movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()))); |
duke@435 | 940 | } |
duke@435 | 941 | } else if (dst.first()->is_stack()) { |
duke@435 | 942 | // reg to stack |
duke@435 | 943 | // Do we really have to sign extend??? |
duke@435 | 944 | // __ movslq(src.first()->as_Register(), src.first()->as_Register()); |
duke@435 | 945 | __ movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register()); |
duke@435 | 946 | } else { |
duke@435 | 947 | // Do we really have to sign extend??? |
duke@435 | 948 | // __ movslq(dst.first()->as_Register(), src.first()->as_Register()); |
duke@435 | 949 | if (dst.first() != src.first()) { |
duke@435 | 950 | __ movq(dst.first()->as_Register(), src.first()->as_Register()); |
duke@435 | 951 | } |
duke@435 | 952 | } |
duke@435 | 953 | } |
duke@435 | 954 | |
duke@435 | 955 | |
duke@435 | 956 | // An oop arg. Must pass a handle not the oop itself |
duke@435 | 957 | static void object_move(MacroAssembler* masm, |
duke@435 | 958 | OopMap* map, |
duke@435 | 959 | int oop_handle_offset, |
duke@435 | 960 | int framesize_in_slots, |
duke@435 | 961 | VMRegPair src, |
duke@435 | 962 | VMRegPair dst, |
duke@435 | 963 | bool is_receiver, |
duke@435 | 964 | int* receiver_offset) { |
duke@435 | 965 | |
duke@435 | 966 | // must pass a handle. First figure out the location we use as a handle |
duke@435 | 967 | |
duke@435 | 968 | Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register(); |
duke@435 | 969 | |
duke@435 | 970 | // See if oop is NULL if it is we need no handle |
duke@435 | 971 | |
duke@435 | 972 | if (src.first()->is_stack()) { |
duke@435 | 973 | |
duke@435 | 974 | // Oop is already on the stack as an argument |
duke@435 | 975 | int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots(); |
duke@435 | 976 | map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots)); |
duke@435 | 977 | if (is_receiver) { |
duke@435 | 978 | *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size; |
duke@435 | 979 | } |
duke@435 | 980 | |
never@739 | 981 | __ cmpptr(Address(rbp, reg2offset_in(src.first())), (int32_t)NULL_WORD); |
never@739 | 982 | __ lea(rHandle, Address(rbp, reg2offset_in(src.first()))); |
duke@435 | 983 | // conditionally move a NULL |
never@739 | 984 | __ cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first()))); |
duke@435 | 985 | } else { |
duke@435 | 986 | |
duke@435 | 987 | // Oop is in an a register we must store it to the space we reserve |
duke@435 | 988 | // on the stack for oop_handles and pass a handle if oop is non-NULL |
duke@435 | 989 | |
duke@435 | 990 | const Register rOop = src.first()->as_Register(); |
duke@435 | 991 | int oop_slot; |
duke@435 | 992 | if (rOop == j_rarg0) |
duke@435 | 993 | oop_slot = 0; |
duke@435 | 994 | else if (rOop == j_rarg1) |
duke@435 | 995 | oop_slot = 1; |
duke@435 | 996 | else if (rOop == j_rarg2) |
duke@435 | 997 | oop_slot = 2; |
duke@435 | 998 | else if (rOop == j_rarg3) |
duke@435 | 999 | oop_slot = 3; |
duke@435 | 1000 | else if (rOop == j_rarg4) |
duke@435 | 1001 | oop_slot = 4; |
duke@435 | 1002 | else { |
duke@435 | 1003 | assert(rOop == j_rarg5, "wrong register"); |
duke@435 | 1004 | oop_slot = 5; |
duke@435 | 1005 | } |
duke@435 | 1006 | |
duke@435 | 1007 | oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset; |
duke@435 | 1008 | int offset = oop_slot*VMRegImpl::stack_slot_size; |
duke@435 | 1009 | |
duke@435 | 1010 | map->set_oop(VMRegImpl::stack2reg(oop_slot)); |
duke@435 | 1011 | // Store oop in handle area, may be NULL |
never@739 | 1012 | __ movptr(Address(rsp, offset), rOop); |
duke@435 | 1013 | if (is_receiver) { |
duke@435 | 1014 | *receiver_offset = offset; |
duke@435 | 1015 | } |
duke@435 | 1016 | |
never@739 | 1017 | __ cmpptr(rOop, (int32_t)NULL_WORD); |
never@739 | 1018 | __ lea(rHandle, Address(rsp, offset)); |
duke@435 | 1019 | // conditionally move a NULL from the handle area where it was just stored |
never@739 | 1020 | __ cmovptr(Assembler::equal, rHandle, Address(rsp, offset)); |
duke@435 | 1021 | } |
duke@435 | 1022 | |
duke@435 | 1023 | // If arg is on the stack then place it otherwise it is already in correct reg. |
duke@435 | 1024 | if (dst.first()->is_stack()) { |
never@739 | 1025 | __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle); |
duke@435 | 1026 | } |
duke@435 | 1027 | } |
duke@435 | 1028 | |
duke@435 | 1029 | // A float arg may have to do float reg int reg conversion |
duke@435 | 1030 | static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
duke@435 | 1031 | assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move"); |
duke@435 | 1032 | |
duke@435 | 1033 | // The calling conventions assures us that each VMregpair is either |
duke@435 | 1034 | // all really one physical register or adjacent stack slots. |
duke@435 | 1035 | // This greatly simplifies the cases here compared to sparc. |
duke@435 | 1036 | |
duke@435 | 1037 | if (src.first()->is_stack()) { |
duke@435 | 1038 | if (dst.first()->is_stack()) { |
duke@435 | 1039 | __ movl(rax, Address(rbp, reg2offset_in(src.first()))); |
never@739 | 1040 | __ movptr(Address(rsp, reg2offset_out(dst.first())), rax); |
duke@435 | 1041 | } else { |
duke@435 | 1042 | // stack to reg |
duke@435 | 1043 | assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters"); |
duke@435 | 1044 | __ movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()))); |
duke@435 | 1045 | } |
duke@435 | 1046 | } else if (dst.first()->is_stack()) { |
duke@435 | 1047 | // reg to stack |
duke@435 | 1048 | assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters"); |
duke@435 | 1049 | __ movflt(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister()); |
duke@435 | 1050 | } else { |
duke@435 | 1051 | // reg to reg |
duke@435 | 1052 | // In theory these overlap but the ordering is such that this is likely a nop |
duke@435 | 1053 | if ( src.first() != dst.first()) { |
duke@435 | 1054 | __ movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister()); |
duke@435 | 1055 | } |
duke@435 | 1056 | } |
duke@435 | 1057 | } |
duke@435 | 1058 | |
duke@435 | 1059 | // A long move |
duke@435 | 1060 | static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
duke@435 | 1061 | |
duke@435 | 1062 | // The calling conventions assures us that each VMregpair is either |
duke@435 | 1063 | // all really one physical register or adjacent stack slots. |
duke@435 | 1064 | // This greatly simplifies the cases here compared to sparc. |
duke@435 | 1065 | |
duke@435 | 1066 | if (src.is_single_phys_reg() ) { |
duke@435 | 1067 | if (dst.is_single_phys_reg()) { |
duke@435 | 1068 | if (dst.first() != src.first()) { |
never@739 | 1069 | __ mov(dst.first()->as_Register(), src.first()->as_Register()); |
duke@435 | 1070 | } |
duke@435 | 1071 | } else { |
duke@435 | 1072 | assert(dst.is_single_reg(), "not a stack pair"); |
duke@435 | 1073 | __ movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register()); |
duke@435 | 1074 | } |
duke@435 | 1075 | } else if (dst.is_single_phys_reg()) { |
duke@435 | 1076 | assert(src.is_single_reg(), "not a stack pair"); |
duke@435 | 1077 | __ movq(dst.first()->as_Register(), Address(rbp, reg2offset_out(src.first()))); |
duke@435 | 1078 | } else { |
duke@435 | 1079 | assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs"); |
duke@435 | 1080 | __ movq(rax, Address(rbp, reg2offset_in(src.first()))); |
duke@435 | 1081 | __ movq(Address(rsp, reg2offset_out(dst.first())), rax); |
duke@435 | 1082 | } |
duke@435 | 1083 | } |
duke@435 | 1084 | |
duke@435 | 1085 | // A double move |
duke@435 | 1086 | static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
duke@435 | 1087 | |
duke@435 | 1088 | // The calling conventions assures us that each VMregpair is either |
duke@435 | 1089 | // all really one physical register or adjacent stack slots. |
duke@435 | 1090 | // This greatly simplifies the cases here compared to sparc. |
duke@435 | 1091 | |
duke@435 | 1092 | if (src.is_single_phys_reg() ) { |
duke@435 | 1093 | if (dst.is_single_phys_reg()) { |
duke@435 | 1094 | // In theory these overlap but the ordering is such that this is likely a nop |
duke@435 | 1095 | if ( src.first() != dst.first()) { |
duke@435 | 1096 | __ movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister()); |
duke@435 | 1097 | } |
duke@435 | 1098 | } else { |
duke@435 | 1099 | assert(dst.is_single_reg(), "not a stack pair"); |
duke@435 | 1100 | __ movdbl(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister()); |
duke@435 | 1101 | } |
duke@435 | 1102 | } else if (dst.is_single_phys_reg()) { |
duke@435 | 1103 | assert(src.is_single_reg(), "not a stack pair"); |
duke@435 | 1104 | __ movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_out(src.first()))); |
duke@435 | 1105 | } else { |
duke@435 | 1106 | assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs"); |
duke@435 | 1107 | __ movq(rax, Address(rbp, reg2offset_in(src.first()))); |
duke@435 | 1108 | __ movq(Address(rsp, reg2offset_out(dst.first())), rax); |
duke@435 | 1109 | } |
duke@435 | 1110 | } |
duke@435 | 1111 | |
duke@435 | 1112 | |
duke@435 | 1113 | void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { |
duke@435 | 1114 | // We always ignore the frame_slots arg and just use the space just below frame pointer |
duke@435 | 1115 | // which by this time is free to use |
duke@435 | 1116 | switch (ret_type) { |
duke@435 | 1117 | case T_FLOAT: |
duke@435 | 1118 | __ movflt(Address(rbp, -wordSize), xmm0); |
duke@435 | 1119 | break; |
duke@435 | 1120 | case T_DOUBLE: |
duke@435 | 1121 | __ movdbl(Address(rbp, -wordSize), xmm0); |
duke@435 | 1122 | break; |
duke@435 | 1123 | case T_VOID: break; |
duke@435 | 1124 | default: { |
never@739 | 1125 | __ movptr(Address(rbp, -wordSize), rax); |
duke@435 | 1126 | } |
duke@435 | 1127 | } |
duke@435 | 1128 | } |
duke@435 | 1129 | |
duke@435 | 1130 | void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { |
duke@435 | 1131 | // We always ignore the frame_slots arg and just use the space just below frame pointer |
duke@435 | 1132 | // which by this time is free to use |
duke@435 | 1133 | switch (ret_type) { |
duke@435 | 1134 | case T_FLOAT: |
duke@435 | 1135 | __ movflt(xmm0, Address(rbp, -wordSize)); |
duke@435 | 1136 | break; |
duke@435 | 1137 | case T_DOUBLE: |
duke@435 | 1138 | __ movdbl(xmm0, Address(rbp, -wordSize)); |
duke@435 | 1139 | break; |
duke@435 | 1140 | case T_VOID: break; |
duke@435 | 1141 | default: { |
never@739 | 1142 | __ movptr(rax, Address(rbp, -wordSize)); |
duke@435 | 1143 | } |
duke@435 | 1144 | } |
duke@435 | 1145 | } |
duke@435 | 1146 | |
duke@435 | 1147 | static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) { |
duke@435 | 1148 | for ( int i = first_arg ; i < arg_count ; i++ ) { |
duke@435 | 1149 | if (args[i].first()->is_Register()) { |
never@739 | 1150 | __ push(args[i].first()->as_Register()); |
duke@435 | 1151 | } else if (args[i].first()->is_XMMRegister()) { |
never@739 | 1152 | __ subptr(rsp, 2*wordSize); |
duke@435 | 1153 | __ movdbl(Address(rsp, 0), args[i].first()->as_XMMRegister()); |
duke@435 | 1154 | } |
duke@435 | 1155 | } |
duke@435 | 1156 | } |
duke@435 | 1157 | |
duke@435 | 1158 | static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) { |
duke@435 | 1159 | for ( int i = arg_count - 1 ; i >= first_arg ; i-- ) { |
duke@435 | 1160 | if (args[i].first()->is_Register()) { |
never@739 | 1161 | __ pop(args[i].first()->as_Register()); |
duke@435 | 1162 | } else if (args[i].first()->is_XMMRegister()) { |
duke@435 | 1163 | __ movdbl(args[i].first()->as_XMMRegister(), Address(rsp, 0)); |
never@739 | 1164 | __ addptr(rsp, 2*wordSize); |
duke@435 | 1165 | } |
duke@435 | 1166 | } |
duke@435 | 1167 | } |
duke@435 | 1168 | |
duke@435 | 1169 | // --------------------------------------------------------------------------- |
duke@435 | 1170 | // Generate a native wrapper for a given method. The method takes arguments |
duke@435 | 1171 | // in the Java compiled code convention, marshals them to the native |
duke@435 | 1172 | // convention (handlizes oops, etc), transitions to native, makes the call, |
duke@435 | 1173 | // returns to java state (possibly blocking), unhandlizes any result and |
duke@435 | 1174 | // returns. |
duke@435 | 1175 | nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm, |
duke@435 | 1176 | methodHandle method, |
twisti@2687 | 1177 | int compile_id, |
duke@435 | 1178 | int total_in_args, |
duke@435 | 1179 | int comp_args_on_stack, |
duke@435 | 1180 | BasicType *in_sig_bt, |
duke@435 | 1181 | VMRegPair *in_regs, |
duke@435 | 1182 | BasicType ret_type) { |
duke@435 | 1183 | // Native nmethod wrappers never take possesion of the oop arguments. |
duke@435 | 1184 | // So the caller will gc the arguments. The only thing we need an |
duke@435 | 1185 | // oopMap for is if the call is static |
duke@435 | 1186 | // |
duke@435 | 1187 | // An OopMap for lock (and class if static) |
duke@435 | 1188 | OopMapSet *oop_maps = new OopMapSet(); |
duke@435 | 1189 | intptr_t start = (intptr_t)__ pc(); |
duke@435 | 1190 | |
duke@435 | 1191 | // We have received a description of where all the java arg are located |
duke@435 | 1192 | // on entry to the wrapper. We need to convert these args to where |
duke@435 | 1193 | // the jni function will expect them. To figure out where they go |
duke@435 | 1194 | // we convert the java signature to a C signature by inserting |
duke@435 | 1195 | // the hidden arguments as arg[0] and possibly arg[1] (static method) |
duke@435 | 1196 | |
duke@435 | 1197 | int total_c_args = total_in_args + 1; |
duke@435 | 1198 | if (method->is_static()) { |
duke@435 | 1199 | total_c_args++; |
duke@435 | 1200 | } |
duke@435 | 1201 | |
duke@435 | 1202 | BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args); |
duke@435 | 1203 | VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args); |
duke@435 | 1204 | |
duke@435 | 1205 | int argc = 0; |
duke@435 | 1206 | out_sig_bt[argc++] = T_ADDRESS; |
duke@435 | 1207 | if (method->is_static()) { |
duke@435 | 1208 | out_sig_bt[argc++] = T_OBJECT; |
duke@435 | 1209 | } |
duke@435 | 1210 | |
duke@435 | 1211 | for (int i = 0; i < total_in_args ; i++ ) { |
duke@435 | 1212 | out_sig_bt[argc++] = in_sig_bt[i]; |
duke@435 | 1213 | } |
duke@435 | 1214 | |
duke@435 | 1215 | // Now figure out where the args must be stored and how much stack space |
duke@435 | 1216 | // they require. |
duke@435 | 1217 | // |
duke@435 | 1218 | int out_arg_slots; |
duke@435 | 1219 | out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args); |
duke@435 | 1220 | |
duke@435 | 1221 | // Compute framesize for the wrapper. We need to handlize all oops in |
duke@435 | 1222 | // incoming registers |
duke@435 | 1223 | |
duke@435 | 1224 | // Calculate the total number of stack slots we will need. |
duke@435 | 1225 | |
duke@435 | 1226 | // First count the abi requirement plus all of the outgoing args |
duke@435 | 1227 | int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; |
duke@435 | 1228 | |
duke@435 | 1229 | // Now the space for the inbound oop handle area |
duke@435 | 1230 | |
duke@435 | 1231 | int oop_handle_offset = stack_slots; |
duke@435 | 1232 | stack_slots += 6*VMRegImpl::slots_per_word; |
duke@435 | 1233 | |
duke@435 | 1234 | // Now any space we need for handlizing a klass if static method |
duke@435 | 1235 | |
duke@435 | 1236 | int oop_temp_slot_offset = 0; |
duke@435 | 1237 | int klass_slot_offset = 0; |
duke@435 | 1238 | int klass_offset = -1; |
duke@435 | 1239 | int lock_slot_offset = 0; |
duke@435 | 1240 | bool is_static = false; |
duke@435 | 1241 | |
duke@435 | 1242 | if (method->is_static()) { |
duke@435 | 1243 | klass_slot_offset = stack_slots; |
duke@435 | 1244 | stack_slots += VMRegImpl::slots_per_word; |
duke@435 | 1245 | klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size; |
duke@435 | 1246 | is_static = true; |
duke@435 | 1247 | } |
duke@435 | 1248 | |
duke@435 | 1249 | // Plus a lock if needed |
duke@435 | 1250 | |
duke@435 | 1251 | if (method->is_synchronized()) { |
duke@435 | 1252 | lock_slot_offset = stack_slots; |
duke@435 | 1253 | stack_slots += VMRegImpl::slots_per_word; |
duke@435 | 1254 | } |
duke@435 | 1255 | |
duke@435 | 1256 | // Now a place (+2) to save return values or temp during shuffling |
duke@435 | 1257 | // + 4 for return address (which we own) and saved rbp |
duke@435 | 1258 | stack_slots += 6; |
duke@435 | 1259 | |
duke@435 | 1260 | // Ok The space we have allocated will look like: |
duke@435 | 1261 | // |
duke@435 | 1262 | // |
duke@435 | 1263 | // FP-> | | |
duke@435 | 1264 | // |---------------------| |
duke@435 | 1265 | // | 2 slots for moves | |
duke@435 | 1266 | // |---------------------| |
duke@435 | 1267 | // | lock box (if sync) | |
duke@435 | 1268 | // |---------------------| <- lock_slot_offset |
duke@435 | 1269 | // | klass (if static) | |
duke@435 | 1270 | // |---------------------| <- klass_slot_offset |
duke@435 | 1271 | // | oopHandle area | |
duke@435 | 1272 | // |---------------------| <- oop_handle_offset (6 java arg registers) |
duke@435 | 1273 | // | outbound memory | |
duke@435 | 1274 | // | based arguments | |
duke@435 | 1275 | // | | |
duke@435 | 1276 | // |---------------------| |
duke@435 | 1277 | // | | |
duke@435 | 1278 | // SP-> | out_preserved_slots | |
duke@435 | 1279 | // |
duke@435 | 1280 | // |
duke@435 | 1281 | |
duke@435 | 1282 | |
duke@435 | 1283 | // Now compute actual number of stack words we need rounding to make |
duke@435 | 1284 | // stack properly aligned. |
xlu@959 | 1285 | stack_slots = round_to(stack_slots, StackAlignmentInSlots); |
duke@435 | 1286 | |
duke@435 | 1287 | int stack_size = stack_slots * VMRegImpl::stack_slot_size; |
duke@435 | 1288 | |
duke@435 | 1289 | |
duke@435 | 1290 | // First thing make an ic check to see if we should even be here |
duke@435 | 1291 | |
duke@435 | 1292 | // We are free to use all registers as temps without saving them and |
duke@435 | 1293 | // restoring them except rbp. rbp is the only callee save register |
duke@435 | 1294 | // as far as the interpreter and the compiler(s) are concerned. |
duke@435 | 1295 | |
duke@435 | 1296 | |
duke@435 | 1297 | const Register ic_reg = rax; |
duke@435 | 1298 | const Register receiver = j_rarg0; |
duke@435 | 1299 | |
duke@435 | 1300 | Label ok; |
duke@435 | 1301 | Label exception_pending; |
duke@435 | 1302 | |
never@1283 | 1303 | assert_different_registers(ic_reg, receiver, rscratch1); |
duke@435 | 1304 | __ verify_oop(receiver); |
never@1283 | 1305 | __ load_klass(rscratch1, receiver); |
never@1283 | 1306 | __ cmpq(ic_reg, rscratch1); |
duke@435 | 1307 | __ jcc(Assembler::equal, ok); |
duke@435 | 1308 | |
duke@435 | 1309 | __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); |
duke@435 | 1310 | |
coleenp@548 | 1311 | __ bind(ok); |
coleenp@548 | 1312 | |
duke@435 | 1313 | // Verified entry point must be aligned |
duke@435 | 1314 | __ align(8); |
duke@435 | 1315 | |
duke@435 | 1316 | int vep_offset = ((intptr_t)__ pc()) - start; |
duke@435 | 1317 | |
duke@435 | 1318 | // The instruction at the verified entry point must be 5 bytes or longer |
duke@435 | 1319 | // because it can be patched on the fly by make_non_entrant. The stack bang |
duke@435 | 1320 | // instruction fits that requirement. |
duke@435 | 1321 | |
duke@435 | 1322 | // Generate stack overflow check |
duke@435 | 1323 | |
duke@435 | 1324 | if (UseStackBanging) { |
duke@435 | 1325 | __ bang_stack_with_offset(StackShadowPages*os::vm_page_size()); |
duke@435 | 1326 | } else { |
duke@435 | 1327 | // need a 5 byte instruction to allow MT safe patching to non-entrant |
duke@435 | 1328 | __ fat_nop(); |
duke@435 | 1329 | } |
duke@435 | 1330 | |
duke@435 | 1331 | // Generate a new frame for the wrapper. |
duke@435 | 1332 | __ enter(); |
duke@435 | 1333 | // -2 because return address is already present and so is saved rbp |
never@739 | 1334 | __ subptr(rsp, stack_size - 2*wordSize); |
duke@435 | 1335 | |
duke@435 | 1336 | // Frame is now completed as far as size and linkage. |
duke@435 | 1337 | |
duke@435 | 1338 | int frame_complete = ((intptr_t)__ pc()) - start; |
duke@435 | 1339 | |
duke@435 | 1340 | #ifdef ASSERT |
duke@435 | 1341 | { |
duke@435 | 1342 | Label L; |
never@739 | 1343 | __ mov(rax, rsp); |
twisti@1040 | 1344 | __ andptr(rax, -16); // must be 16 byte boundary (see amd64 ABI) |
never@739 | 1345 | __ cmpptr(rax, rsp); |
duke@435 | 1346 | __ jcc(Assembler::equal, L); |
duke@435 | 1347 | __ stop("improperly aligned stack"); |
duke@435 | 1348 | __ bind(L); |
duke@435 | 1349 | } |
duke@435 | 1350 | #endif /* ASSERT */ |
duke@435 | 1351 | |
duke@435 | 1352 | |
duke@435 | 1353 | // We use r14 as the oop handle for the receiver/klass |
duke@435 | 1354 | // It is callee save so it survives the call to native |
duke@435 | 1355 | |
duke@435 | 1356 | const Register oop_handle_reg = r14; |
duke@435 | 1357 | |
duke@435 | 1358 | |
duke@435 | 1359 | |
duke@435 | 1360 | // |
duke@435 | 1361 | // We immediately shuffle the arguments so that any vm call we have to |
duke@435 | 1362 | // make from here on out (sync slow path, jvmti, etc.) we will have |
duke@435 | 1363 | // captured the oops from our caller and have a valid oopMap for |
duke@435 | 1364 | // them. |
duke@435 | 1365 | |
duke@435 | 1366 | // ----------------- |
duke@435 | 1367 | // The Grand Shuffle |
duke@435 | 1368 | |
duke@435 | 1369 | // The Java calling convention is either equal (linux) or denser (win64) than the |
duke@435 | 1370 | // c calling convention. However the because of the jni_env argument the c calling |
duke@435 | 1371 | // convention always has at least one more (and two for static) arguments than Java. |
duke@435 | 1372 | // Therefore if we move the args from java -> c backwards then we will never have |
duke@435 | 1373 | // a register->register conflict and we don't have to build a dependency graph |
duke@435 | 1374 | // and figure out how to break any cycles. |
duke@435 | 1375 | // |
duke@435 | 1376 | |
duke@435 | 1377 | // Record esp-based slot for receiver on stack for non-static methods |
duke@435 | 1378 | int receiver_offset = -1; |
duke@435 | 1379 | |
duke@435 | 1380 | // This is a trick. We double the stack slots so we can claim |
duke@435 | 1381 | // the oops in the caller's frame. Since we are sure to have |
duke@435 | 1382 | // more args than the caller doubling is enough to make |
duke@435 | 1383 | // sure we can capture all the incoming oop args from the |
duke@435 | 1384 | // caller. |
duke@435 | 1385 | // |
duke@435 | 1386 | OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/); |
duke@435 | 1387 | |
duke@435 | 1388 | // Mark location of rbp (someday) |
duke@435 | 1389 | // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, vmreg(rbp)); |
duke@435 | 1390 | |
duke@435 | 1391 | // Use eax, ebx as temporaries during any memory-memory moves we have to do |
duke@435 | 1392 | // All inbound args are referenced based on rbp and all outbound args via rsp. |
duke@435 | 1393 | |
duke@435 | 1394 | |
duke@435 | 1395 | #ifdef ASSERT |
duke@435 | 1396 | bool reg_destroyed[RegisterImpl::number_of_registers]; |
duke@435 | 1397 | bool freg_destroyed[XMMRegisterImpl::number_of_registers]; |
duke@435 | 1398 | for ( int r = 0 ; r < RegisterImpl::number_of_registers ; r++ ) { |
duke@435 | 1399 | reg_destroyed[r] = false; |
duke@435 | 1400 | } |
duke@435 | 1401 | for ( int f = 0 ; f < XMMRegisterImpl::number_of_registers ; f++ ) { |
duke@435 | 1402 | freg_destroyed[f] = false; |
duke@435 | 1403 | } |
duke@435 | 1404 | |
duke@435 | 1405 | #endif /* ASSERT */ |
duke@435 | 1406 | |
duke@435 | 1407 | |
duke@435 | 1408 | int c_arg = total_c_args - 1; |
duke@435 | 1409 | for ( int i = total_in_args - 1; i >= 0 ; i--, c_arg-- ) { |
duke@435 | 1410 | #ifdef ASSERT |
duke@435 | 1411 | if (in_regs[i].first()->is_Register()) { |
duke@435 | 1412 | assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!"); |
duke@435 | 1413 | } else if (in_regs[i].first()->is_XMMRegister()) { |
duke@435 | 1414 | assert(!freg_destroyed[in_regs[i].first()->as_XMMRegister()->encoding()], "destroyed reg!"); |
duke@435 | 1415 | } |
duke@435 | 1416 | if (out_regs[c_arg].first()->is_Register()) { |
duke@435 | 1417 | reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true; |
duke@435 | 1418 | } else if (out_regs[c_arg].first()->is_XMMRegister()) { |
duke@435 | 1419 | freg_destroyed[out_regs[c_arg].first()->as_XMMRegister()->encoding()] = true; |
duke@435 | 1420 | } |
duke@435 | 1421 | #endif /* ASSERT */ |
duke@435 | 1422 | switch (in_sig_bt[i]) { |
duke@435 | 1423 | case T_ARRAY: |
duke@435 | 1424 | case T_OBJECT: |
duke@435 | 1425 | object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg], |
duke@435 | 1426 | ((i == 0) && (!is_static)), |
duke@435 | 1427 | &receiver_offset); |
duke@435 | 1428 | break; |
duke@435 | 1429 | case T_VOID: |
duke@435 | 1430 | break; |
duke@435 | 1431 | |
duke@435 | 1432 | case T_FLOAT: |
duke@435 | 1433 | float_move(masm, in_regs[i], out_regs[c_arg]); |
duke@435 | 1434 | break; |
duke@435 | 1435 | |
duke@435 | 1436 | case T_DOUBLE: |
duke@435 | 1437 | assert( i + 1 < total_in_args && |
duke@435 | 1438 | in_sig_bt[i + 1] == T_VOID && |
duke@435 | 1439 | out_sig_bt[c_arg+1] == T_VOID, "bad arg list"); |
duke@435 | 1440 | double_move(masm, in_regs[i], out_regs[c_arg]); |
duke@435 | 1441 | break; |
duke@435 | 1442 | |
duke@435 | 1443 | case T_LONG : |
duke@435 | 1444 | long_move(masm, in_regs[i], out_regs[c_arg]); |
duke@435 | 1445 | break; |
duke@435 | 1446 | |
duke@435 | 1447 | case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); |
duke@435 | 1448 | |
duke@435 | 1449 | default: |
duke@435 | 1450 | move32_64(masm, in_regs[i], out_regs[c_arg]); |
duke@435 | 1451 | } |
duke@435 | 1452 | } |
duke@435 | 1453 | |
duke@435 | 1454 | // point c_arg at the first arg that is already loaded in case we |
duke@435 | 1455 | // need to spill before we call out |
duke@435 | 1456 | c_arg++; |
duke@435 | 1457 | |
duke@435 | 1458 | // Pre-load a static method's oop into r14. Used both by locking code and |
duke@435 | 1459 | // the normal JNI call code. |
duke@435 | 1460 | if (method->is_static()) { |
duke@435 | 1461 | |
duke@435 | 1462 | // load oop into a register |
duke@435 | 1463 | __ movoop(oop_handle_reg, JNIHandles::make_local(Klass::cast(method->method_holder())->java_mirror())); |
duke@435 | 1464 | |
duke@435 | 1465 | // Now handlize the static class mirror it's known not-null. |
never@739 | 1466 | __ movptr(Address(rsp, klass_offset), oop_handle_reg); |
duke@435 | 1467 | map->set_oop(VMRegImpl::stack2reg(klass_slot_offset)); |
duke@435 | 1468 | |
duke@435 | 1469 | // Now get the handle |
never@739 | 1470 | __ lea(oop_handle_reg, Address(rsp, klass_offset)); |
duke@435 | 1471 | // store the klass handle as second argument |
never@739 | 1472 | __ movptr(c_rarg1, oop_handle_reg); |
duke@435 | 1473 | // and protect the arg if we must spill |
duke@435 | 1474 | c_arg--; |
duke@435 | 1475 | } |
duke@435 | 1476 | |
duke@435 | 1477 | // Change state to native (we save the return address in the thread, since it might not |
duke@435 | 1478 | // be pushed on the stack when we do a a stack traversal). It is enough that the pc() |
duke@435 | 1479 | // points into the right code segment. It does not have to be the correct return pc. |
duke@435 | 1480 | // We use the same pc/oopMap repeatedly when we call out |
duke@435 | 1481 | |
duke@435 | 1482 | intptr_t the_pc = (intptr_t) __ pc(); |
duke@435 | 1483 | oop_maps->add_gc_map(the_pc - start, map); |
duke@435 | 1484 | |
duke@435 | 1485 | __ set_last_Java_frame(rsp, noreg, (address)the_pc); |
duke@435 | 1486 | |
duke@435 | 1487 | |
duke@435 | 1488 | // We have all of the arguments setup at this point. We must not touch any register |
duke@435 | 1489 | // argument registers at this point (what if we save/restore them there are no oop? |
duke@435 | 1490 | |
duke@435 | 1491 | { |
duke@435 | 1492 | SkipIfEqual skip(masm, &DTraceMethodProbes, false); |
duke@435 | 1493 | // protect the args we've loaded |
duke@435 | 1494 | save_args(masm, total_c_args, c_arg, out_regs); |
duke@435 | 1495 | __ movoop(c_rarg1, JNIHandles::make_local(method())); |
duke@435 | 1496 | __ call_VM_leaf( |
duke@435 | 1497 | CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), |
duke@435 | 1498 | r15_thread, c_rarg1); |
duke@435 | 1499 | restore_args(masm, total_c_args, c_arg, out_regs); |
duke@435 | 1500 | } |
duke@435 | 1501 | |
dcubed@1045 | 1502 | // RedefineClasses() tracing support for obsolete method entry |
dcubed@1045 | 1503 | if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { |
dcubed@1045 | 1504 | // protect the args we've loaded |
dcubed@1045 | 1505 | save_args(masm, total_c_args, c_arg, out_regs); |
dcubed@1045 | 1506 | __ movoop(c_rarg1, JNIHandles::make_local(method())); |
dcubed@1045 | 1507 | __ call_VM_leaf( |
dcubed@1045 | 1508 | CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), |
dcubed@1045 | 1509 | r15_thread, c_rarg1); |
dcubed@1045 | 1510 | restore_args(masm, total_c_args, c_arg, out_regs); |
dcubed@1045 | 1511 | } |
dcubed@1045 | 1512 | |
duke@435 | 1513 | // Lock a synchronized method |
duke@435 | 1514 | |
duke@435 | 1515 | // Register definitions used by locking and unlocking |
duke@435 | 1516 | |
duke@435 | 1517 | const Register swap_reg = rax; // Must use rax for cmpxchg instruction |
duke@435 | 1518 | const Register obj_reg = rbx; // Will contain the oop |
duke@435 | 1519 | const Register lock_reg = r13; // Address of compiler lock object (BasicLock) |
duke@435 | 1520 | const Register old_hdr = r13; // value of old header at unlock time |
duke@435 | 1521 | |
duke@435 | 1522 | Label slow_path_lock; |
duke@435 | 1523 | Label lock_done; |
duke@435 | 1524 | |
duke@435 | 1525 | if (method->is_synchronized()) { |
duke@435 | 1526 | |
duke@435 | 1527 | |
duke@435 | 1528 | const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes(); |
duke@435 | 1529 | |
duke@435 | 1530 | // Get the handle (the 2nd argument) |
never@739 | 1531 | __ mov(oop_handle_reg, c_rarg1); |
duke@435 | 1532 | |
duke@435 | 1533 | // Get address of the box |
duke@435 | 1534 | |
never@739 | 1535 | __ lea(lock_reg, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size)); |
duke@435 | 1536 | |
duke@435 | 1537 | // Load the oop from the handle |
never@739 | 1538 | __ movptr(obj_reg, Address(oop_handle_reg, 0)); |
duke@435 | 1539 | |
duke@435 | 1540 | if (UseBiasedLocking) { |
duke@435 | 1541 | __ biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, lock_done, &slow_path_lock); |
duke@435 | 1542 | } |
duke@435 | 1543 | |
duke@435 | 1544 | // Load immediate 1 into swap_reg %rax |
duke@435 | 1545 | __ movl(swap_reg, 1); |
duke@435 | 1546 | |
duke@435 | 1547 | // Load (object->mark() | 1) into swap_reg %rax |
never@739 | 1548 | __ orptr(swap_reg, Address(obj_reg, 0)); |
duke@435 | 1549 | |
duke@435 | 1550 | // Save (object->mark() | 1) into BasicLock's displaced header |
never@739 | 1551 | __ movptr(Address(lock_reg, mark_word_offset), swap_reg); |
duke@435 | 1552 | |
duke@435 | 1553 | if (os::is_MP()) { |
duke@435 | 1554 | __ lock(); |
duke@435 | 1555 | } |
duke@435 | 1556 | |
duke@435 | 1557 | // src -> dest iff dest == rax else rax <- dest |
never@739 | 1558 | __ cmpxchgptr(lock_reg, Address(obj_reg, 0)); |
duke@435 | 1559 | __ jcc(Assembler::equal, lock_done); |
duke@435 | 1560 | |
duke@435 | 1561 | // Hmm should this move to the slow path code area??? |
duke@435 | 1562 | |
duke@435 | 1563 | // Test if the oopMark is an obvious stack pointer, i.e., |
duke@435 | 1564 | // 1) (mark & 3) == 0, and |
duke@435 | 1565 | // 2) rsp <= mark < mark + os::pagesize() |
duke@435 | 1566 | // These 3 tests can be done by evaluating the following |
duke@435 | 1567 | // expression: ((mark - rsp) & (3 - os::vm_page_size())), |
duke@435 | 1568 | // assuming both stack pointer and pagesize have their |
duke@435 | 1569 | // least significant 2 bits clear. |
duke@435 | 1570 | // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg |
duke@435 | 1571 | |
never@739 | 1572 | __ subptr(swap_reg, rsp); |
never@739 | 1573 | __ andptr(swap_reg, 3 - os::vm_page_size()); |
duke@435 | 1574 | |
duke@435 | 1575 | // Save the test result, for recursive case, the result is zero |
never@739 | 1576 | __ movptr(Address(lock_reg, mark_word_offset), swap_reg); |
duke@435 | 1577 | __ jcc(Assembler::notEqual, slow_path_lock); |
duke@435 | 1578 | |
duke@435 | 1579 | // Slow path will re-enter here |
duke@435 | 1580 | |
duke@435 | 1581 | __ bind(lock_done); |
duke@435 | 1582 | } |
duke@435 | 1583 | |
duke@435 | 1584 | |
duke@435 | 1585 | // Finally just about ready to make the JNI call |
duke@435 | 1586 | |
duke@435 | 1587 | |
duke@435 | 1588 | // get JNIEnv* which is first argument to native |
duke@435 | 1589 | |
never@739 | 1590 | __ lea(c_rarg0, Address(r15_thread, in_bytes(JavaThread::jni_environment_offset()))); |
duke@435 | 1591 | |
duke@435 | 1592 | // Now set thread in native |
never@739 | 1593 | __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native); |
duke@435 | 1594 | |
duke@435 | 1595 | __ call(RuntimeAddress(method->native_function())); |
duke@435 | 1596 | |
duke@435 | 1597 | // Either restore the MXCSR register after returning from the JNI Call |
duke@435 | 1598 | // or verify that it wasn't changed. |
duke@435 | 1599 | if (RestoreMXCSROnJNICalls) { |
never@739 | 1600 | __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std())); |
duke@435 | 1601 | |
duke@435 | 1602 | } |
duke@435 | 1603 | else if (CheckJNICalls ) { |
never@739 | 1604 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry()))); |
duke@435 | 1605 | } |
duke@435 | 1606 | |
duke@435 | 1607 | |
duke@435 | 1608 | // Unpack native results. |
duke@435 | 1609 | switch (ret_type) { |
duke@435 | 1610 | case T_BOOLEAN: __ c2bool(rax); break; |
duke@435 | 1611 | case T_CHAR : __ movzwl(rax, rax); break; |
duke@435 | 1612 | case T_BYTE : __ sign_extend_byte (rax); break; |
duke@435 | 1613 | case T_SHORT : __ sign_extend_short(rax); break; |
duke@435 | 1614 | case T_INT : /* nothing to do */ break; |
duke@435 | 1615 | case T_DOUBLE : |
duke@435 | 1616 | case T_FLOAT : |
duke@435 | 1617 | // Result is in xmm0 we'll save as needed |
duke@435 | 1618 | break; |
duke@435 | 1619 | case T_ARRAY: // Really a handle |
duke@435 | 1620 | case T_OBJECT: // Really a handle |
duke@435 | 1621 | break; // can't de-handlize until after safepoint check |
duke@435 | 1622 | case T_VOID: break; |
duke@435 | 1623 | case T_LONG: break; |
duke@435 | 1624 | default : ShouldNotReachHere(); |
duke@435 | 1625 | } |
duke@435 | 1626 | |
duke@435 | 1627 | // Switch thread to "native transition" state before reading the synchronization state. |
duke@435 | 1628 | // This additional state is necessary because reading and testing the synchronization |
duke@435 | 1629 | // state is not atomic w.r.t. GC, as this scenario demonstrates: |
duke@435 | 1630 | // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted. |
duke@435 | 1631 | // VM thread changes sync state to synchronizing and suspends threads for GC. |
duke@435 | 1632 | // Thread A is resumed to finish this native method, but doesn't block here since it |
duke@435 | 1633 | // didn't see any synchronization is progress, and escapes. |
never@739 | 1634 | __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native_trans); |
duke@435 | 1635 | |
duke@435 | 1636 | if(os::is_MP()) { |
duke@435 | 1637 | if (UseMembar) { |
duke@435 | 1638 | // Force this write out before the read below |
duke@435 | 1639 | __ membar(Assembler::Membar_mask_bits( |
duke@435 | 1640 | Assembler::LoadLoad | Assembler::LoadStore | |
duke@435 | 1641 | Assembler::StoreLoad | Assembler::StoreStore)); |
duke@435 | 1642 | } else { |
duke@435 | 1643 | // Write serialization page so VM thread can do a pseudo remote membar. |
duke@435 | 1644 | // We use the current thread pointer to calculate a thread specific |
duke@435 | 1645 | // offset to write to within the page. This minimizes bus traffic |
duke@435 | 1646 | // due to cache line collision. |
duke@435 | 1647 | __ serialize_memory(r15_thread, rcx); |
duke@435 | 1648 | } |
duke@435 | 1649 | } |
duke@435 | 1650 | |
duke@435 | 1651 | |
duke@435 | 1652 | // check for safepoint operation in progress and/or pending suspend requests |
duke@435 | 1653 | { |
duke@435 | 1654 | Label Continue; |
duke@435 | 1655 | |
duke@435 | 1656 | __ cmp32(ExternalAddress((address)SafepointSynchronize::address_of_state()), |
duke@435 | 1657 | SafepointSynchronize::_not_synchronized); |
duke@435 | 1658 | |
duke@435 | 1659 | Label L; |
duke@435 | 1660 | __ jcc(Assembler::notEqual, L); |
duke@435 | 1661 | __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0); |
duke@435 | 1662 | __ jcc(Assembler::equal, Continue); |
duke@435 | 1663 | __ bind(L); |
duke@435 | 1664 | |
duke@435 | 1665 | // Don't use call_VM as it will see a possible pending exception and forward it |
duke@435 | 1666 | // and never return here preventing us from clearing _last_native_pc down below. |
duke@435 | 1667 | // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are |
duke@435 | 1668 | // preserved and correspond to the bcp/locals pointers. So we do a runtime call |
duke@435 | 1669 | // by hand. |
duke@435 | 1670 | // |
duke@435 | 1671 | save_native_result(masm, ret_type, stack_slots); |
never@739 | 1672 | __ mov(c_rarg0, r15_thread); |
never@739 | 1673 | __ mov(r12, rsp); // remember sp |
never@739 | 1674 | __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows |
never@739 | 1675 | __ andptr(rsp, -16); // align stack as required by ABI |
duke@435 | 1676 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans))); |
never@739 | 1677 | __ mov(rsp, r12); // restore sp |
coleenp@548 | 1678 | __ reinit_heapbase(); |
duke@435 | 1679 | // Restore any method result value |
duke@435 | 1680 | restore_native_result(masm, ret_type, stack_slots); |
duke@435 | 1681 | __ bind(Continue); |
duke@435 | 1682 | } |
duke@435 | 1683 | |
duke@435 | 1684 | // change thread state |
duke@435 | 1685 | __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java); |
duke@435 | 1686 | |
duke@435 | 1687 | Label reguard; |
duke@435 | 1688 | Label reguard_done; |
duke@435 | 1689 | __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled); |
duke@435 | 1690 | __ jcc(Assembler::equal, reguard); |
duke@435 | 1691 | __ bind(reguard_done); |
duke@435 | 1692 | |
duke@435 | 1693 | // native result if any is live |
duke@435 | 1694 | |
duke@435 | 1695 | // Unlock |
duke@435 | 1696 | Label unlock_done; |
duke@435 | 1697 | Label slow_path_unlock; |
duke@435 | 1698 | if (method->is_synchronized()) { |
duke@435 | 1699 | |
duke@435 | 1700 | // Get locked oop from the handle we passed to jni |
never@739 | 1701 | __ movptr(obj_reg, Address(oop_handle_reg, 0)); |
duke@435 | 1702 | |
duke@435 | 1703 | Label done; |
duke@435 | 1704 | |
duke@435 | 1705 | if (UseBiasedLocking) { |
duke@435 | 1706 | __ biased_locking_exit(obj_reg, old_hdr, done); |
duke@435 | 1707 | } |
duke@435 | 1708 | |
duke@435 | 1709 | // Simple recursive lock? |
duke@435 | 1710 | |
never@739 | 1711 | __ cmpptr(Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size), (int32_t)NULL_WORD); |
duke@435 | 1712 | __ jcc(Assembler::equal, done); |
duke@435 | 1713 | |
duke@435 | 1714 | // Must save rax if if it is live now because cmpxchg must use it |
duke@435 | 1715 | if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { |
duke@435 | 1716 | save_native_result(masm, ret_type, stack_slots); |
duke@435 | 1717 | } |
duke@435 | 1718 | |
duke@435 | 1719 | |
duke@435 | 1720 | // get address of the stack lock |
never@739 | 1721 | __ lea(rax, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size)); |
duke@435 | 1722 | // get old displaced header |
never@739 | 1723 | __ movptr(old_hdr, Address(rax, 0)); |
duke@435 | 1724 | |
duke@435 | 1725 | // Atomic swap old header if oop still contains the stack lock |
duke@435 | 1726 | if (os::is_MP()) { |
duke@435 | 1727 | __ lock(); |
duke@435 | 1728 | } |
never@739 | 1729 | __ cmpxchgptr(old_hdr, Address(obj_reg, 0)); |
duke@435 | 1730 | __ jcc(Assembler::notEqual, slow_path_unlock); |
duke@435 | 1731 | |
duke@435 | 1732 | // slow path re-enters here |
duke@435 | 1733 | __ bind(unlock_done); |
duke@435 | 1734 | if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { |
duke@435 | 1735 | restore_native_result(masm, ret_type, stack_slots); |
duke@435 | 1736 | } |
duke@435 | 1737 | |
duke@435 | 1738 | __ bind(done); |
duke@435 | 1739 | |
duke@435 | 1740 | } |
duke@435 | 1741 | { |
duke@435 | 1742 | SkipIfEqual skip(masm, &DTraceMethodProbes, false); |
duke@435 | 1743 | save_native_result(masm, ret_type, stack_slots); |
duke@435 | 1744 | __ movoop(c_rarg1, JNIHandles::make_local(method())); |
duke@435 | 1745 | __ call_VM_leaf( |
duke@435 | 1746 | CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), |
duke@435 | 1747 | r15_thread, c_rarg1); |
duke@435 | 1748 | restore_native_result(masm, ret_type, stack_slots); |
duke@435 | 1749 | } |
duke@435 | 1750 | |
duke@435 | 1751 | __ reset_last_Java_frame(false, true); |
duke@435 | 1752 | |
duke@435 | 1753 | // Unpack oop result |
duke@435 | 1754 | if (ret_type == T_OBJECT || ret_type == T_ARRAY) { |
duke@435 | 1755 | Label L; |
never@739 | 1756 | __ testptr(rax, rax); |
duke@435 | 1757 | __ jcc(Assembler::zero, L); |
never@739 | 1758 | __ movptr(rax, Address(rax, 0)); |
duke@435 | 1759 | __ bind(L); |
duke@435 | 1760 | __ verify_oop(rax); |
duke@435 | 1761 | } |
duke@435 | 1762 | |
duke@435 | 1763 | // reset handle block |
never@739 | 1764 | __ movptr(rcx, Address(r15_thread, JavaThread::active_handles_offset())); |
never@739 | 1765 | __ movptr(Address(rcx, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD); |
duke@435 | 1766 | |
duke@435 | 1767 | // pop our frame |
duke@435 | 1768 | |
duke@435 | 1769 | __ leave(); |
duke@435 | 1770 | |
duke@435 | 1771 | // Any exception pending? |
never@739 | 1772 | __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); |
duke@435 | 1773 | __ jcc(Assembler::notEqual, exception_pending); |
duke@435 | 1774 | |
duke@435 | 1775 | // Return |
duke@435 | 1776 | |
duke@435 | 1777 | __ ret(0); |
duke@435 | 1778 | |
duke@435 | 1779 | // Unexpected paths are out of line and go here |
duke@435 | 1780 | |
duke@435 | 1781 | // forward the exception |
duke@435 | 1782 | __ bind(exception_pending); |
duke@435 | 1783 | |
duke@435 | 1784 | // and forward the exception |
duke@435 | 1785 | __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
duke@435 | 1786 | |
duke@435 | 1787 | |
duke@435 | 1788 | // Slow path locking & unlocking |
duke@435 | 1789 | if (method->is_synchronized()) { |
duke@435 | 1790 | |
duke@435 | 1791 | // BEGIN Slow path lock |
duke@435 | 1792 | __ bind(slow_path_lock); |
duke@435 | 1793 | |
duke@435 | 1794 | // has last_Java_frame setup. No exceptions so do vanilla call not call_VM |
duke@435 | 1795 | // args are (oop obj, BasicLock* lock, JavaThread* thread) |
duke@435 | 1796 | |
duke@435 | 1797 | // protect the args we've loaded |
duke@435 | 1798 | save_args(masm, total_c_args, c_arg, out_regs); |
duke@435 | 1799 | |
never@739 | 1800 | __ mov(c_rarg0, obj_reg); |
never@739 | 1801 | __ mov(c_rarg1, lock_reg); |
never@739 | 1802 | __ mov(c_rarg2, r15_thread); |
duke@435 | 1803 | |
duke@435 | 1804 | // Not a leaf but we have last_Java_frame setup as we want |
duke@435 | 1805 | __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), 3); |
duke@435 | 1806 | restore_args(masm, total_c_args, c_arg, out_regs); |
duke@435 | 1807 | |
duke@435 | 1808 | #ifdef ASSERT |
duke@435 | 1809 | { Label L; |
never@739 | 1810 | __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); |
duke@435 | 1811 | __ jcc(Assembler::equal, L); |
duke@435 | 1812 | __ stop("no pending exception allowed on exit from monitorenter"); |
duke@435 | 1813 | __ bind(L); |
duke@435 | 1814 | } |
duke@435 | 1815 | #endif |
duke@435 | 1816 | __ jmp(lock_done); |
duke@435 | 1817 | |
duke@435 | 1818 | // END Slow path lock |
duke@435 | 1819 | |
duke@435 | 1820 | // BEGIN Slow path unlock |
duke@435 | 1821 | __ bind(slow_path_unlock); |
duke@435 | 1822 | |
duke@435 | 1823 | // If we haven't already saved the native result we must save it now as xmm registers |
duke@435 | 1824 | // are still exposed. |
duke@435 | 1825 | |
duke@435 | 1826 | if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { |
duke@435 | 1827 | save_native_result(masm, ret_type, stack_slots); |
duke@435 | 1828 | } |
duke@435 | 1829 | |
never@739 | 1830 | __ lea(c_rarg1, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size)); |
never@739 | 1831 | |
never@739 | 1832 | __ mov(c_rarg0, obj_reg); |
never@739 | 1833 | __ mov(r12, rsp); // remember sp |
never@739 | 1834 | __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows |
never@739 | 1835 | __ andptr(rsp, -16); // align stack as required by ABI |
duke@435 | 1836 | |
duke@435 | 1837 | // Save pending exception around call to VM (which contains an EXCEPTION_MARK) |
duke@435 | 1838 | // NOTE that obj_reg == rbx currently |
never@739 | 1839 | __ movptr(rbx, Address(r15_thread, in_bytes(Thread::pending_exception_offset()))); |
never@739 | 1840 | __ movptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); |
duke@435 | 1841 | |
duke@435 | 1842 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C))); |
never@739 | 1843 | __ mov(rsp, r12); // restore sp |
coleenp@548 | 1844 | __ reinit_heapbase(); |
duke@435 | 1845 | #ifdef ASSERT |
duke@435 | 1846 | { |
duke@435 | 1847 | Label L; |
never@739 | 1848 | __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD); |
duke@435 | 1849 | __ jcc(Assembler::equal, L); |
duke@435 | 1850 | __ stop("no pending exception allowed on exit complete_monitor_unlocking_C"); |
duke@435 | 1851 | __ bind(L); |
duke@435 | 1852 | } |
duke@435 | 1853 | #endif /* ASSERT */ |
duke@435 | 1854 | |
never@739 | 1855 | __ movptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), rbx); |
duke@435 | 1856 | |
duke@435 | 1857 | if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { |
duke@435 | 1858 | restore_native_result(masm, ret_type, stack_slots); |
duke@435 | 1859 | } |
duke@435 | 1860 | __ jmp(unlock_done); |
duke@435 | 1861 | |
duke@435 | 1862 | // END Slow path unlock |
duke@435 | 1863 | |
duke@435 | 1864 | } // synchronized |
duke@435 | 1865 | |
duke@435 | 1866 | // SLOW PATH Reguard the stack if needed |
duke@435 | 1867 | |
duke@435 | 1868 | __ bind(reguard); |
duke@435 | 1869 | save_native_result(masm, ret_type, stack_slots); |
never@739 | 1870 | __ mov(r12, rsp); // remember sp |
never@739 | 1871 | __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows |
never@739 | 1872 | __ andptr(rsp, -16); // align stack as required by ABI |
duke@435 | 1873 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); |
never@739 | 1874 | __ mov(rsp, r12); // restore sp |
coleenp@548 | 1875 | __ reinit_heapbase(); |
duke@435 | 1876 | restore_native_result(masm, ret_type, stack_slots); |
duke@435 | 1877 | // and continue |
duke@435 | 1878 | __ jmp(reguard_done); |
duke@435 | 1879 | |
duke@435 | 1880 | |
duke@435 | 1881 | |
duke@435 | 1882 | __ flush(); |
duke@435 | 1883 | |
duke@435 | 1884 | nmethod *nm = nmethod::new_native_nmethod(method, |
twisti@2687 | 1885 | compile_id, |
duke@435 | 1886 | masm->code(), |
duke@435 | 1887 | vep_offset, |
duke@435 | 1888 | frame_complete, |
duke@435 | 1889 | stack_slots / VMRegImpl::slots_per_word, |
duke@435 | 1890 | (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)), |
duke@435 | 1891 | in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size), |
duke@435 | 1892 | oop_maps); |
duke@435 | 1893 | return nm; |
duke@435 | 1894 | |
duke@435 | 1895 | } |
duke@435 | 1896 | |
kamg@551 | 1897 | #ifdef HAVE_DTRACE_H |
kamg@551 | 1898 | // --------------------------------------------------------------------------- |
kamg@551 | 1899 | // Generate a dtrace nmethod for a given signature. The method takes arguments |
kamg@551 | 1900 | // in the Java compiled code convention, marshals them to the native |
kamg@551 | 1901 | // abi and then leaves nops at the position you would expect to call a native |
kamg@551 | 1902 | // function. When the probe is enabled the nops are replaced with a trap |
kamg@551 | 1903 | // instruction that dtrace inserts and the trace will cause a notification |
kamg@551 | 1904 | // to dtrace. |
kamg@551 | 1905 | // |
kamg@551 | 1906 | // The probes are only able to take primitive types and java/lang/String as |
kamg@551 | 1907 | // arguments. No other java types are allowed. Strings are converted to utf8 |
kamg@551 | 1908 | // strings so that from dtrace point of view java strings are converted to C |
kamg@551 | 1909 | // strings. There is an arbitrary fixed limit on the total space that a method |
kamg@551 | 1910 | // can use for converting the strings. (256 chars per string in the signature). |
kamg@551 | 1911 | // So any java string larger then this is truncated. |
kamg@551 | 1912 | |
kamg@551 | 1913 | static int fp_offset[ConcreteRegisterImpl::number_of_registers] = { 0 }; |
kamg@551 | 1914 | static bool offsets_initialized = false; |
kamg@551 | 1915 | |
kamg@551 | 1916 | |
kamg@551 | 1917 | nmethod *SharedRuntime::generate_dtrace_nmethod(MacroAssembler *masm, |
kamg@551 | 1918 | methodHandle method) { |
kamg@551 | 1919 | |
kamg@551 | 1920 | |
kamg@551 | 1921 | // generate_dtrace_nmethod is guarded by a mutex so we are sure to |
kamg@551 | 1922 | // be single threaded in this method. |
kamg@551 | 1923 | assert(AdapterHandlerLibrary_lock->owned_by_self(), "must be"); |
kamg@551 | 1924 | |
kamg@551 | 1925 | if (!offsets_initialized) { |
kamg@551 | 1926 | fp_offset[c_rarg0->as_VMReg()->value()] = -1 * wordSize; |
kamg@551 | 1927 | fp_offset[c_rarg1->as_VMReg()->value()] = -2 * wordSize; |
kamg@551 | 1928 | fp_offset[c_rarg2->as_VMReg()->value()] = -3 * wordSize; |
kamg@551 | 1929 | fp_offset[c_rarg3->as_VMReg()->value()] = -4 * wordSize; |
kamg@551 | 1930 | fp_offset[c_rarg4->as_VMReg()->value()] = -5 * wordSize; |
kamg@551 | 1931 | fp_offset[c_rarg5->as_VMReg()->value()] = -6 * wordSize; |
kamg@551 | 1932 | |
kamg@551 | 1933 | fp_offset[c_farg0->as_VMReg()->value()] = -7 * wordSize; |
kamg@551 | 1934 | fp_offset[c_farg1->as_VMReg()->value()] = -8 * wordSize; |
kamg@551 | 1935 | fp_offset[c_farg2->as_VMReg()->value()] = -9 * wordSize; |
kamg@551 | 1936 | fp_offset[c_farg3->as_VMReg()->value()] = -10 * wordSize; |
kamg@551 | 1937 | fp_offset[c_farg4->as_VMReg()->value()] = -11 * wordSize; |
kamg@551 | 1938 | fp_offset[c_farg5->as_VMReg()->value()] = -12 * wordSize; |
kamg@551 | 1939 | fp_offset[c_farg6->as_VMReg()->value()] = -13 * wordSize; |
kamg@551 | 1940 | fp_offset[c_farg7->as_VMReg()->value()] = -14 * wordSize; |
kamg@551 | 1941 | |
kamg@551 | 1942 | offsets_initialized = true; |
kamg@551 | 1943 | } |
kamg@551 | 1944 | // Fill in the signature array, for the calling-convention call. |
kamg@551 | 1945 | int total_args_passed = method->size_of_parameters(); |
kamg@551 | 1946 | |
kamg@551 | 1947 | BasicType* in_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed); |
kamg@551 | 1948 | VMRegPair *in_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed); |
kamg@551 | 1949 | |
kamg@551 | 1950 | // The signature we are going to use for the trap that dtrace will see |
kamg@551 | 1951 | // java/lang/String is converted. We drop "this" and any other object |
kamg@551 | 1952 | // is converted to NULL. (A one-slot java/lang/Long object reference |
kamg@551 | 1953 | // is converted to a two-slot long, which is why we double the allocation). |
kamg@551 | 1954 | BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed * 2); |
kamg@551 | 1955 | VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed * 2); |
kamg@551 | 1956 | |
kamg@551 | 1957 | int i=0; |
kamg@551 | 1958 | int total_strings = 0; |
kamg@551 | 1959 | int first_arg_to_pass = 0; |
kamg@551 | 1960 | int total_c_args = 0; |
kamg@551 | 1961 | |
kamg@551 | 1962 | // Skip the receiver as dtrace doesn't want to see it |
kamg@551 | 1963 | if( !method->is_static() ) { |
kamg@551 | 1964 | in_sig_bt[i++] = T_OBJECT; |
kamg@551 | 1965 | first_arg_to_pass = 1; |
kamg@551 | 1966 | } |
kamg@551 | 1967 | |
kamg@551 | 1968 | // We need to convert the java args to where a native (non-jni) function |
kamg@551 | 1969 | // would expect them. To figure out where they go we convert the java |
kamg@551 | 1970 | // signature to a C signature. |
kamg@551 | 1971 | |
kamg@551 | 1972 | SignatureStream ss(method->signature()); |
kamg@551 | 1973 | for ( ; !ss.at_return_type(); ss.next()) { |
kamg@551 | 1974 | BasicType bt = ss.type(); |
kamg@551 | 1975 | in_sig_bt[i++] = bt; // Collect remaining bits of signature |
kamg@551 | 1976 | out_sig_bt[total_c_args++] = bt; |
kamg@551 | 1977 | if( bt == T_OBJECT) { |
coleenp@2497 | 1978 | Symbol* s = ss.as_symbol_or_null(); // symbol is created |
kamg@551 | 1979 | if (s == vmSymbols::java_lang_String()) { |
kamg@551 | 1980 | total_strings++; |
kamg@551 | 1981 | out_sig_bt[total_c_args-1] = T_ADDRESS; |
kamg@551 | 1982 | } else if (s == vmSymbols::java_lang_Boolean() || |
kamg@551 | 1983 | s == vmSymbols::java_lang_Character() || |
kamg@551 | 1984 | s == vmSymbols::java_lang_Byte() || |
kamg@551 | 1985 | s == vmSymbols::java_lang_Short() || |
kamg@551 | 1986 | s == vmSymbols::java_lang_Integer() || |
kamg@551 | 1987 | s == vmSymbols::java_lang_Float()) { |
kamg@551 | 1988 | out_sig_bt[total_c_args-1] = T_INT; |
kamg@551 | 1989 | } else if (s == vmSymbols::java_lang_Long() || |
kamg@551 | 1990 | s == vmSymbols::java_lang_Double()) { |
kamg@551 | 1991 | out_sig_bt[total_c_args-1] = T_LONG; |
kamg@551 | 1992 | out_sig_bt[total_c_args++] = T_VOID; |
kamg@551 | 1993 | } |
kamg@551 | 1994 | } else if ( bt == T_LONG || bt == T_DOUBLE ) { |
kamg@551 | 1995 | in_sig_bt[i++] = T_VOID; // Longs & doubles take 2 Java slots |
kamg@551 | 1996 | // We convert double to long |
kamg@551 | 1997 | out_sig_bt[total_c_args-1] = T_LONG; |
kamg@551 | 1998 | out_sig_bt[total_c_args++] = T_VOID; |
kamg@551 | 1999 | } else if ( bt == T_FLOAT) { |
kamg@551 | 2000 | // We convert float to int |
kamg@551 | 2001 | out_sig_bt[total_c_args-1] = T_INT; |
kamg@551 | 2002 | } |
kamg@551 | 2003 | } |
kamg@551 | 2004 | |
kamg@551 | 2005 | assert(i==total_args_passed, "validly parsed signature"); |
kamg@551 | 2006 | |
kamg@551 | 2007 | // Now get the compiled-Java layout as input arguments |
kamg@551 | 2008 | int comp_args_on_stack; |
kamg@551 | 2009 | comp_args_on_stack = SharedRuntime::java_calling_convention( |
kamg@551 | 2010 | in_sig_bt, in_regs, total_args_passed, false); |
kamg@551 | 2011 | |
kamg@551 | 2012 | // Now figure out where the args must be stored and how much stack space |
kamg@551 | 2013 | // they require (neglecting out_preserve_stack_slots but space for storing |
kamg@551 | 2014 | // the 1st six register arguments). It's weird see int_stk_helper. |
kamg@551 | 2015 | |
kamg@551 | 2016 | int out_arg_slots; |
kamg@551 | 2017 | out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args); |
kamg@551 | 2018 | |
kamg@551 | 2019 | // Calculate the total number of stack slots we will need. |
kamg@551 | 2020 | |
kamg@551 | 2021 | // First count the abi requirement plus all of the outgoing args |
kamg@551 | 2022 | int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; |
kamg@551 | 2023 | |
kamg@551 | 2024 | // Now space for the string(s) we must convert |
kamg@551 | 2025 | int* string_locs = NEW_RESOURCE_ARRAY(int, total_strings + 1); |
kamg@551 | 2026 | for (i = 0; i < total_strings ; i++) { |
kamg@551 | 2027 | string_locs[i] = stack_slots; |
kamg@551 | 2028 | stack_slots += max_dtrace_string_size / VMRegImpl::stack_slot_size; |
kamg@551 | 2029 | } |
kamg@551 | 2030 | |
kamg@551 | 2031 | // Plus the temps we might need to juggle register args |
kamg@551 | 2032 | // regs take two slots each |
kamg@551 | 2033 | stack_slots += (Argument::n_int_register_parameters_c + |
kamg@551 | 2034 | Argument::n_float_register_parameters_c) * 2; |
kamg@551 | 2035 | |
kamg@551 | 2036 | |
kamg@551 | 2037 | // + 4 for return address (which we own) and saved rbp, |
kamg@551 | 2038 | |
kamg@551 | 2039 | stack_slots += 4; |
kamg@551 | 2040 | |
kamg@551 | 2041 | // Ok The space we have allocated will look like: |
kamg@551 | 2042 | // |
kamg@551 | 2043 | // |
kamg@551 | 2044 | // FP-> | | |
kamg@551 | 2045 | // |---------------------| |
kamg@551 | 2046 | // | string[n] | |
kamg@551 | 2047 | // |---------------------| <- string_locs[n] |
kamg@551 | 2048 | // | string[n-1] | |
kamg@551 | 2049 | // |---------------------| <- string_locs[n-1] |
kamg@551 | 2050 | // | ... | |
kamg@551 | 2051 | // | ... | |
kamg@551 | 2052 | // |---------------------| <- string_locs[1] |
kamg@551 | 2053 | // | string[0] | |
kamg@551 | 2054 | // |---------------------| <- string_locs[0] |
kamg@551 | 2055 | // | outbound memory | |
kamg@551 | 2056 | // | based arguments | |
kamg@551 | 2057 | // | | |
kamg@551 | 2058 | // |---------------------| |
kamg@551 | 2059 | // | | |
kamg@551 | 2060 | // SP-> | out_preserved_slots | |
kamg@551 | 2061 | // |
kamg@551 | 2062 | // |
kamg@551 | 2063 | |
kamg@551 | 2064 | // Now compute actual number of stack words we need rounding to make |
kamg@551 | 2065 | // stack properly aligned. |
kamg@551 | 2066 | stack_slots = round_to(stack_slots, 4 * VMRegImpl::slots_per_word); |
kamg@551 | 2067 | |
kamg@551 | 2068 | int stack_size = stack_slots * VMRegImpl::stack_slot_size; |
kamg@551 | 2069 | |
kamg@551 | 2070 | intptr_t start = (intptr_t)__ pc(); |
kamg@551 | 2071 | |
kamg@551 | 2072 | // First thing make an ic check to see if we should even be here |
kamg@551 | 2073 | |
kamg@551 | 2074 | // We are free to use all registers as temps without saving them and |
kamg@551 | 2075 | // restoring them except rbp. rbp, is the only callee save register |
kamg@551 | 2076 | // as far as the interpreter and the compiler(s) are concerned. |
kamg@551 | 2077 | |
kamg@551 | 2078 | const Register ic_reg = rax; |
kamg@551 | 2079 | const Register receiver = rcx; |
kamg@551 | 2080 | Label hit; |
kamg@551 | 2081 | Label exception_pending; |
kamg@551 | 2082 | |
kamg@551 | 2083 | |
kamg@551 | 2084 | __ verify_oop(receiver); |
kamg@551 | 2085 | __ cmpl(ic_reg, Address(receiver, oopDesc::klass_offset_in_bytes())); |
kamg@551 | 2086 | __ jcc(Assembler::equal, hit); |
kamg@551 | 2087 | |
kamg@551 | 2088 | __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); |
kamg@551 | 2089 | |
kamg@551 | 2090 | // verified entry must be aligned for code patching. |
kamg@551 | 2091 | // and the first 5 bytes must be in the same cache line |
kamg@551 | 2092 | // if we align at 8 then we will be sure 5 bytes are in the same line |
kamg@551 | 2093 | __ align(8); |
kamg@551 | 2094 | |
kamg@551 | 2095 | __ bind(hit); |
kamg@551 | 2096 | |
kamg@551 | 2097 | int vep_offset = ((intptr_t)__ pc()) - start; |
kamg@551 | 2098 | |
kamg@551 | 2099 | |
kamg@551 | 2100 | // The instruction at the verified entry point must be 5 bytes or longer |
kamg@551 | 2101 | // because it can be patched on the fly by make_non_entrant. The stack bang |
kamg@551 | 2102 | // instruction fits that requirement. |
kamg@551 | 2103 | |
kamg@551 | 2104 | // Generate stack overflow check |
kamg@551 | 2105 | |
kamg@551 | 2106 | if (UseStackBanging) { |
kamg@551 | 2107 | if (stack_size <= StackShadowPages*os::vm_page_size()) { |
kamg@551 | 2108 | __ bang_stack_with_offset(StackShadowPages*os::vm_page_size()); |
kamg@551 | 2109 | } else { |
kamg@551 | 2110 | __ movl(rax, stack_size); |
kamg@551 | 2111 | __ bang_stack_size(rax, rbx); |
kamg@551 | 2112 | } |
kamg@551 | 2113 | } else { |
kamg@551 | 2114 | // need a 5 byte instruction to allow MT safe patching to non-entrant |
kamg@551 | 2115 | __ fat_nop(); |
kamg@551 | 2116 | } |
kamg@551 | 2117 | |
kamg@551 | 2118 | assert(((uintptr_t)__ pc() - start - vep_offset) >= 5, |
kamg@551 | 2119 | "valid size for make_non_entrant"); |
kamg@551 | 2120 | |
kamg@551 | 2121 | // Generate a new frame for the wrapper. |
kamg@551 | 2122 | __ enter(); |
kamg@551 | 2123 | |
kamg@551 | 2124 | // -4 because return address is already present and so is saved rbp, |
kamg@551 | 2125 | if (stack_size - 2*wordSize != 0) { |
kamg@551 | 2126 | __ subq(rsp, stack_size - 2*wordSize); |
kamg@551 | 2127 | } |
kamg@551 | 2128 | |
kamg@551 | 2129 | // Frame is now completed as far a size and linkage. |
kamg@551 | 2130 | |
kamg@551 | 2131 | int frame_complete = ((intptr_t)__ pc()) - start; |
kamg@551 | 2132 | |
kamg@551 | 2133 | int c_arg, j_arg; |
kamg@551 | 2134 | |
kamg@551 | 2135 | // State of input register args |
kamg@551 | 2136 | |
kamg@551 | 2137 | bool live[ConcreteRegisterImpl::number_of_registers]; |
kamg@551 | 2138 | |
kamg@551 | 2139 | live[j_rarg0->as_VMReg()->value()] = false; |
kamg@551 | 2140 | live[j_rarg1->as_VMReg()->value()] = false; |
kamg@551 | 2141 | live[j_rarg2->as_VMReg()->value()] = false; |
kamg@551 | 2142 | live[j_rarg3->as_VMReg()->value()] = false; |
kamg@551 | 2143 | live[j_rarg4->as_VMReg()->value()] = false; |
kamg@551 | 2144 | live[j_rarg5->as_VMReg()->value()] = false; |
kamg@551 | 2145 | |
kamg@551 | 2146 | live[j_farg0->as_VMReg()->value()] = false; |
kamg@551 | 2147 | live[j_farg1->as_VMReg()->value()] = false; |
kamg@551 | 2148 | live[j_farg2->as_VMReg()->value()] = false; |
kamg@551 | 2149 | live[j_farg3->as_VMReg()->value()] = false; |
kamg@551 | 2150 | live[j_farg4->as_VMReg()->value()] = false; |
kamg@551 | 2151 | live[j_farg5->as_VMReg()->value()] = false; |
kamg@551 | 2152 | live[j_farg6->as_VMReg()->value()] = false; |
kamg@551 | 2153 | live[j_farg7->as_VMReg()->value()] = false; |
kamg@551 | 2154 | |
kamg@551 | 2155 | |
kamg@551 | 2156 | bool rax_is_zero = false; |
kamg@551 | 2157 | |
kamg@551 | 2158 | // All args (except strings) destined for the stack are moved first |
kamg@551 | 2159 | for (j_arg = first_arg_to_pass, c_arg = 0 ; |
kamg@551 | 2160 | j_arg < total_args_passed ; j_arg++, c_arg++ ) { |
kamg@551 | 2161 | VMRegPair src = in_regs[j_arg]; |
kamg@551 | 2162 | VMRegPair dst = out_regs[c_arg]; |
kamg@551 | 2163 | |
kamg@551 | 2164 | // Get the real reg value or a dummy (rsp) |
kamg@551 | 2165 | |
kamg@551 | 2166 | int src_reg = src.first()->is_reg() ? |
kamg@551 | 2167 | src.first()->value() : |
kamg@551 | 2168 | rsp->as_VMReg()->value(); |
kamg@551 | 2169 | |
kamg@551 | 2170 | bool useless = in_sig_bt[j_arg] == T_ARRAY || |
kamg@551 | 2171 | (in_sig_bt[j_arg] == T_OBJECT && |
kamg@551 | 2172 | out_sig_bt[c_arg] != T_INT && |
kamg@551 | 2173 | out_sig_bt[c_arg] != T_ADDRESS && |
kamg@551 | 2174 | out_sig_bt[c_arg] != T_LONG); |
kamg@551 | 2175 | |
kamg@551 | 2176 | live[src_reg] = !useless; |
kamg@551 | 2177 | |
kamg@551 | 2178 | if (dst.first()->is_stack()) { |
kamg@551 | 2179 | |
kamg@551 | 2180 | // Even though a string arg in a register is still live after this loop |
kamg@551 | 2181 | // after the string conversion loop (next) it will be dead so we take |
kamg@551 | 2182 | // advantage of that now for simpler code to manage live. |
kamg@551 | 2183 | |
kamg@551 | 2184 | live[src_reg] = false; |
kamg@551 | 2185 | switch (in_sig_bt[j_arg]) { |
kamg@551 | 2186 | |
kamg@551 | 2187 | case T_ARRAY: |
kamg@551 | 2188 | case T_OBJECT: |
kamg@551 | 2189 | { |
kamg@551 | 2190 | Address stack_dst(rsp, reg2offset_out(dst.first())); |
kamg@551 | 2191 | |
kamg@551 | 2192 | if (out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) { |
kamg@551 | 2193 | // need to unbox a one-word value |
kamg@551 | 2194 | Register in_reg = rax; |
kamg@551 | 2195 | if ( src.first()->is_reg() ) { |
kamg@551 | 2196 | in_reg = src.first()->as_Register(); |
kamg@551 | 2197 | } else { |
kamg@551 | 2198 | __ movq(rax, Address(rbp, reg2offset_in(src.first()))); |
kamg@551 | 2199 | rax_is_zero = false; |
kamg@551 | 2200 | } |
kamg@551 | 2201 | Label skipUnbox; |
kamg@551 | 2202 | __ movptr(Address(rsp, reg2offset_out(dst.first())), |
kamg@551 | 2203 | (int32_t)NULL_WORD); |
kamg@551 | 2204 | __ testq(in_reg, in_reg); |
kamg@551 | 2205 | __ jcc(Assembler::zero, skipUnbox); |
kamg@551 | 2206 | |
kvn@600 | 2207 | BasicType bt = out_sig_bt[c_arg]; |
kvn@600 | 2208 | int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt); |
kamg@551 | 2209 | Address src1(in_reg, box_offset); |
kvn@600 | 2210 | if ( bt == T_LONG ) { |
kamg@551 | 2211 | __ movq(in_reg, src1); |
kamg@551 | 2212 | __ movq(stack_dst, in_reg); |
kamg@551 | 2213 | assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); |
kamg@551 | 2214 | ++c_arg; // skip over T_VOID to keep the loop indices in sync |
kamg@551 | 2215 | } else { |
kamg@551 | 2216 | __ movl(in_reg, src1); |
kamg@551 | 2217 | __ movl(stack_dst, in_reg); |
kamg@551 | 2218 | } |
kamg@551 | 2219 | |
kamg@551 | 2220 | __ bind(skipUnbox); |
kamg@551 | 2221 | } else if (out_sig_bt[c_arg] != T_ADDRESS) { |
kamg@551 | 2222 | // Convert the arg to NULL |
kamg@551 | 2223 | if (!rax_is_zero) { |
kamg@551 | 2224 | __ xorq(rax, rax); |
kamg@551 | 2225 | rax_is_zero = true; |
kamg@551 | 2226 | } |
kamg@551 | 2227 | __ movq(stack_dst, rax); |
kamg@551 | 2228 | } |
kamg@551 | 2229 | } |
kamg@551 | 2230 | break; |
kamg@551 | 2231 | |
kamg@551 | 2232 | case T_VOID: |
kamg@551 | 2233 | break; |
kamg@551 | 2234 | |
kamg@551 | 2235 | case T_FLOAT: |
kamg@551 | 2236 | // This does the right thing since we know it is destined for the |
kamg@551 | 2237 | // stack |
kamg@551 | 2238 | float_move(masm, src, dst); |
kamg@551 | 2239 | break; |
kamg@551 | 2240 | |
kamg@551 | 2241 | case T_DOUBLE: |
kamg@551 | 2242 | // This does the right thing since we know it is destined for the |
kamg@551 | 2243 | // stack |
kamg@551 | 2244 | double_move(masm, src, dst); |
kamg@551 | 2245 | break; |
kamg@551 | 2246 | |
kamg@551 | 2247 | case T_LONG : |
kamg@551 | 2248 | long_move(masm, src, dst); |
kamg@551 | 2249 | break; |
kamg@551 | 2250 | |
kamg@551 | 2251 | case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); |
kamg@551 | 2252 | |
kamg@551 | 2253 | default: |
kamg@551 | 2254 | move32_64(masm, src, dst); |
kamg@551 | 2255 | } |
kamg@551 | 2256 | } |
kamg@551 | 2257 | |
kamg@551 | 2258 | } |
kamg@551 | 2259 | |
kamg@551 | 2260 | // If we have any strings we must store any register based arg to the stack |
kamg@551 | 2261 | // This includes any still live xmm registers too. |
kamg@551 | 2262 | |
kamg@551 | 2263 | int sid = 0; |
kamg@551 | 2264 | |
kamg@551 | 2265 | if (total_strings > 0 ) { |
kamg@551 | 2266 | for (j_arg = first_arg_to_pass, c_arg = 0 ; |
kamg@551 | 2267 | j_arg < total_args_passed ; j_arg++, c_arg++ ) { |
kamg@551 | 2268 | VMRegPair src = in_regs[j_arg]; |
kamg@551 | 2269 | VMRegPair dst = out_regs[c_arg]; |
kamg@551 | 2270 | |
kamg@551 | 2271 | if (src.first()->is_reg()) { |
kamg@551 | 2272 | Address src_tmp(rbp, fp_offset[src.first()->value()]); |
kamg@551 | 2273 | |
kamg@551 | 2274 | // string oops were left untouched by the previous loop even if the |
kamg@551 | 2275 | // eventual (converted) arg is destined for the stack so park them |
kamg@551 | 2276 | // away now (except for first) |
kamg@551 | 2277 | |
kamg@551 | 2278 | if (out_sig_bt[c_arg] == T_ADDRESS) { |
kamg@551 | 2279 | Address utf8_addr = Address( |
kamg@551 | 2280 | rsp, string_locs[sid++] * VMRegImpl::stack_slot_size); |
kamg@551 | 2281 | if (sid != 1) { |
kamg@551 | 2282 | // The first string arg won't be killed until after the utf8 |
kamg@551 | 2283 | // conversion |
kamg@551 | 2284 | __ movq(utf8_addr, src.first()->as_Register()); |
kamg@551 | 2285 | } |
kamg@551 | 2286 | } else if (dst.first()->is_reg()) { |
kamg@551 | 2287 | if (in_sig_bt[j_arg] == T_FLOAT || in_sig_bt[j_arg] == T_DOUBLE) { |
kamg@551 | 2288 | |
kamg@551 | 2289 | // Convert the xmm register to an int and store it in the reserved |
kamg@551 | 2290 | // location for the eventual c register arg |
kamg@551 | 2291 | XMMRegister f = src.first()->as_XMMRegister(); |
kamg@551 | 2292 | if (in_sig_bt[j_arg] == T_FLOAT) { |
kamg@551 | 2293 | __ movflt(src_tmp, f); |
kamg@551 | 2294 | } else { |
kamg@551 | 2295 | __ movdbl(src_tmp, f); |
kamg@551 | 2296 | } |
kamg@551 | 2297 | } else { |
kamg@551 | 2298 | // If the arg is an oop type we don't support don't bother to store |
kamg@551 | 2299 | // it remember string was handled above. |
kamg@551 | 2300 | bool useless = in_sig_bt[j_arg] == T_ARRAY || |
kamg@551 | 2301 | (in_sig_bt[j_arg] == T_OBJECT && |
kamg@551 | 2302 | out_sig_bt[c_arg] != T_INT && |
kamg@551 | 2303 | out_sig_bt[c_arg] != T_LONG); |
kamg@551 | 2304 | |
kamg@551 | 2305 | if (!useless) { |
kamg@551 | 2306 | __ movq(src_tmp, src.first()->as_Register()); |
kamg@551 | 2307 | } |
kamg@551 | 2308 | } |
kamg@551 | 2309 | } |
kamg@551 | 2310 | } |
kamg@551 | 2311 | if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) { |
kamg@551 | 2312 | assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); |
kamg@551 | 2313 | ++c_arg; // skip over T_VOID to keep the loop indices in sync |
kamg@551 | 2314 | } |
kamg@551 | 2315 | } |
kamg@551 | 2316 | |
kamg@551 | 2317 | // Now that the volatile registers are safe, convert all the strings |
kamg@551 | 2318 | sid = 0; |
kamg@551 | 2319 | |
kamg@551 | 2320 | for (j_arg = first_arg_to_pass, c_arg = 0 ; |
kamg@551 | 2321 | j_arg < total_args_passed ; j_arg++, c_arg++ ) { |
kamg@551 | 2322 | if (out_sig_bt[c_arg] == T_ADDRESS) { |
kamg@551 | 2323 | // It's a string |
kamg@551 | 2324 | Address utf8_addr = Address( |
kamg@551 | 2325 | rsp, string_locs[sid++] * VMRegImpl::stack_slot_size); |
kamg@551 | 2326 | // The first string we find might still be in the original java arg |
kamg@551 | 2327 | // register |
kamg@551 | 2328 | |
kamg@551 | 2329 | VMReg src = in_regs[j_arg].first(); |
kamg@551 | 2330 | |
kamg@551 | 2331 | // We will need to eventually save the final argument to the trap |
kamg@551 | 2332 | // in the von-volatile location dedicated to src. This is the offset |
kamg@551 | 2333 | // from fp we will use. |
kamg@551 | 2334 | int src_off = src->is_reg() ? |
kamg@551 | 2335 | fp_offset[src->value()] : reg2offset_in(src); |
kamg@551 | 2336 | |
kamg@551 | 2337 | // This is where the argument will eventually reside |
kamg@551 | 2338 | VMRegPair dst = out_regs[c_arg]; |
kamg@551 | 2339 | |
kamg@551 | 2340 | if (src->is_reg()) { |
kamg@551 | 2341 | if (sid == 1) { |
kamg@551 | 2342 | __ movq(c_rarg0, src->as_Register()); |
kamg@551 | 2343 | } else { |
kamg@551 | 2344 | __ movq(c_rarg0, utf8_addr); |
kamg@551 | 2345 | } |
kamg@551 | 2346 | } else { |
kamg@551 | 2347 | // arg is still in the original location |
kamg@551 | 2348 | __ movq(c_rarg0, Address(rbp, reg2offset_in(src))); |
kamg@551 | 2349 | } |
kamg@551 | 2350 | Label done, convert; |
kamg@551 | 2351 | |
kamg@551 | 2352 | // see if the oop is NULL |
kamg@551 | 2353 | __ testq(c_rarg0, c_rarg0); |
kamg@551 | 2354 | __ jcc(Assembler::notEqual, convert); |
kamg@551 | 2355 | |
kamg@551 | 2356 | if (dst.first()->is_reg()) { |
kamg@551 | 2357 | // Save the ptr to utf string in the origina src loc or the tmp |
kamg@551 | 2358 | // dedicated to it |
kamg@551 | 2359 | __ movq(Address(rbp, src_off), c_rarg0); |
kamg@551 | 2360 | } else { |
kamg@551 | 2361 | __ movq(Address(rsp, reg2offset_out(dst.first())), c_rarg0); |
kamg@551 | 2362 | } |
kamg@551 | 2363 | __ jmp(done); |
kamg@551 | 2364 | |
kamg@551 | 2365 | __ bind(convert); |
kamg@551 | 2366 | |
kamg@551 | 2367 | __ lea(c_rarg1, utf8_addr); |
kamg@551 | 2368 | if (dst.first()->is_reg()) { |
kamg@551 | 2369 | __ movq(Address(rbp, src_off), c_rarg1); |
kamg@551 | 2370 | } else { |
kamg@551 | 2371 | __ movq(Address(rsp, reg2offset_out(dst.first())), c_rarg1); |
kamg@551 | 2372 | } |
kamg@551 | 2373 | // And do the conversion |
kamg@551 | 2374 | __ call(RuntimeAddress( |
kamg@551 | 2375 | CAST_FROM_FN_PTR(address, SharedRuntime::get_utf))); |
kamg@551 | 2376 | |
kamg@551 | 2377 | __ bind(done); |
kamg@551 | 2378 | } |
kamg@551 | 2379 | if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) { |
kamg@551 | 2380 | assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); |
kamg@551 | 2381 | ++c_arg; // skip over T_VOID to keep the loop indices in sync |
kamg@551 | 2382 | } |
kamg@551 | 2383 | } |
kamg@551 | 2384 | // The get_utf call killed all the c_arg registers |
kamg@551 | 2385 | live[c_rarg0->as_VMReg()->value()] = false; |
kamg@551 | 2386 | live[c_rarg1->as_VMReg()->value()] = false; |
kamg@551 | 2387 | live[c_rarg2->as_VMReg()->value()] = false; |
kamg@551 | 2388 | live[c_rarg3->as_VMReg()->value()] = false; |
kamg@551 | 2389 | live[c_rarg4->as_VMReg()->value()] = false; |
kamg@551 | 2390 | live[c_rarg5->as_VMReg()->value()] = false; |
kamg@551 | 2391 | |
kamg@551 | 2392 | live[c_farg0->as_VMReg()->value()] = false; |
kamg@551 | 2393 | live[c_farg1->as_VMReg()->value()] = false; |
kamg@551 | 2394 | live[c_farg2->as_VMReg()->value()] = false; |
kamg@551 | 2395 | live[c_farg3->as_VMReg()->value()] = false; |
kamg@551 | 2396 | live[c_farg4->as_VMReg()->value()] = false; |
kamg@551 | 2397 | live[c_farg5->as_VMReg()->value()] = false; |
kamg@551 | 2398 | live[c_farg6->as_VMReg()->value()] = false; |
kamg@551 | 2399 | live[c_farg7->as_VMReg()->value()] = false; |
kamg@551 | 2400 | } |
kamg@551 | 2401 | |
kamg@551 | 2402 | // Now we can finally move the register args to their desired locations |
kamg@551 | 2403 | |
kamg@551 | 2404 | rax_is_zero = false; |
kamg@551 | 2405 | |
kamg@551 | 2406 | for (j_arg = first_arg_to_pass, c_arg = 0 ; |
kamg@551 | 2407 | j_arg < total_args_passed ; j_arg++, c_arg++ ) { |
kamg@551 | 2408 | |
kamg@551 | 2409 | VMRegPair src = in_regs[j_arg]; |
kamg@551 | 2410 | VMRegPair dst = out_regs[c_arg]; |
kamg@551 | 2411 | |
kamg@551 | 2412 | // Only need to look for args destined for the interger registers (since we |
kamg@551 | 2413 | // convert float/double args to look like int/long outbound) |
kamg@551 | 2414 | if (dst.first()->is_reg()) { |
kamg@551 | 2415 | Register r = dst.first()->as_Register(); |
kamg@551 | 2416 | |
kamg@551 | 2417 | // Check if the java arg is unsupported and thereofre useless |
kamg@551 | 2418 | bool useless = in_sig_bt[j_arg] == T_ARRAY || |
kamg@551 | 2419 | (in_sig_bt[j_arg] == T_OBJECT && |
kamg@551 | 2420 | out_sig_bt[c_arg] != T_INT && |
kamg@551 | 2421 | out_sig_bt[c_arg] != T_ADDRESS && |
kamg@551 | 2422 | out_sig_bt[c_arg] != T_LONG); |
kamg@551 | 2423 | |
kamg@551 | 2424 | |
kamg@551 | 2425 | // If we're going to kill an existing arg save it first |
kamg@551 | 2426 | if (live[dst.first()->value()]) { |
kamg@551 | 2427 | // you can't kill yourself |
kamg@551 | 2428 | if (src.first() != dst.first()) { |
kamg@551 | 2429 | __ movq(Address(rbp, fp_offset[dst.first()->value()]), r); |
kamg@551 | 2430 | } |
kamg@551 | 2431 | } |
kamg@551 | 2432 | if (src.first()->is_reg()) { |
kamg@551 | 2433 | if (live[src.first()->value()] ) { |
kamg@551 | 2434 | if (in_sig_bt[j_arg] == T_FLOAT) { |
kamg@551 | 2435 | __ movdl(r, src.first()->as_XMMRegister()); |
kamg@551 | 2436 | } else if (in_sig_bt[j_arg] == T_DOUBLE) { |
kamg@551 | 2437 | __ movdq(r, src.first()->as_XMMRegister()); |
kamg@551 | 2438 | } else if (r != src.first()->as_Register()) { |
kamg@551 | 2439 | if (!useless) { |
kamg@551 | 2440 | __ movq(r, src.first()->as_Register()); |
kamg@551 | 2441 | } |
kamg@551 | 2442 | } |
kamg@551 | 2443 | } else { |
kamg@551 | 2444 | // If the arg is an oop type we don't support don't bother to store |
kamg@551 | 2445 | // it |
kamg@551 | 2446 | if (!useless) { |
kamg@551 | 2447 | if (in_sig_bt[j_arg] == T_DOUBLE || |
kamg@551 | 2448 | in_sig_bt[j_arg] == T_LONG || |
kamg@551 | 2449 | in_sig_bt[j_arg] == T_OBJECT ) { |
kamg@551 | 2450 | __ movq(r, Address(rbp, fp_offset[src.first()->value()])); |
kamg@551 | 2451 | } else { |
kamg@551 | 2452 | __ movl(r, Address(rbp, fp_offset[src.first()->value()])); |
kamg@551 | 2453 | } |
kamg@551 | 2454 | } |
kamg@551 | 2455 | } |
kamg@551 | 2456 | live[src.first()->value()] = false; |
kamg@551 | 2457 | } else if (!useless) { |
kamg@551 | 2458 | // full sized move even for int should be ok |
kamg@551 | 2459 | __ movq(r, Address(rbp, reg2offset_in(src.first()))); |
kamg@551 | 2460 | } |
kamg@551 | 2461 | |
kamg@551 | 2462 | // At this point r has the original java arg in the final location |
kamg@551 | 2463 | // (assuming it wasn't useless). If the java arg was an oop |
kamg@551 | 2464 | // we have a bit more to do |
kamg@551 | 2465 | |
kamg@551 | 2466 | if (in_sig_bt[j_arg] == T_ARRAY || in_sig_bt[j_arg] == T_OBJECT ) { |
kamg@551 | 2467 | if (out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) { |
kamg@551 | 2468 | // need to unbox a one-word value |
kamg@551 | 2469 | Label skip; |
kamg@551 | 2470 | __ testq(r, r); |
kamg@551 | 2471 | __ jcc(Assembler::equal, skip); |
kvn@600 | 2472 | BasicType bt = out_sig_bt[c_arg]; |
kvn@600 | 2473 | int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt); |
kamg@551 | 2474 | Address src1(r, box_offset); |
kvn@600 | 2475 | if ( bt == T_LONG ) { |
kamg@551 | 2476 | __ movq(r, src1); |
kamg@551 | 2477 | } else { |
kamg@551 | 2478 | __ movl(r, src1); |
kamg@551 | 2479 | } |
kamg@551 | 2480 | __ bind(skip); |
kamg@551 | 2481 | |
kamg@551 | 2482 | } else if (out_sig_bt[c_arg] != T_ADDRESS) { |
kamg@551 | 2483 | // Convert the arg to NULL |
kamg@551 | 2484 | __ xorq(r, r); |
kamg@551 | 2485 | } |
kamg@551 | 2486 | } |
kamg@551 | 2487 | |
kamg@551 | 2488 | // dst can longer be holding an input value |
kamg@551 | 2489 | live[dst.first()->value()] = false; |
kamg@551 | 2490 | } |
kamg@551 | 2491 | if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) { |
kamg@551 | 2492 | assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); |
kamg@551 | 2493 | ++c_arg; // skip over T_VOID to keep the loop indices in sync |
kamg@551 | 2494 | } |
kamg@551 | 2495 | } |
kamg@551 | 2496 | |
kamg@551 | 2497 | |
kamg@551 | 2498 | // Ok now we are done. Need to place the nop that dtrace wants in order to |
kamg@551 | 2499 | // patch in the trap |
kamg@551 | 2500 | int patch_offset = ((intptr_t)__ pc()) - start; |
kamg@551 | 2501 | |
kamg@551 | 2502 | __ nop(); |
kamg@551 | 2503 | |
kamg@551 | 2504 | |
kamg@551 | 2505 | // Return |
kamg@551 | 2506 | |
kamg@551 | 2507 | __ leave(); |
kamg@551 | 2508 | __ ret(0); |
kamg@551 | 2509 | |
kamg@551 | 2510 | __ flush(); |
kamg@551 | 2511 | |
kamg@551 | 2512 | nmethod *nm = nmethod::new_dtrace_nmethod( |
kamg@551 | 2513 | method, masm->code(), vep_offset, patch_offset, frame_complete, |
kamg@551 | 2514 | stack_slots / VMRegImpl::slots_per_word); |
kamg@551 | 2515 | return nm; |
kamg@551 | 2516 | |
kamg@551 | 2517 | } |
kamg@551 | 2518 | |
kamg@551 | 2519 | #endif // HAVE_DTRACE_H |
kamg@551 | 2520 | |
duke@435 | 2521 | // this function returns the adjust size (in number of words) to a c2i adapter |
duke@435 | 2522 | // activation for use during deoptimization |
duke@435 | 2523 | int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals ) { |
twisti@1861 | 2524 | return (callee_locals - callee_parameters) * Interpreter::stackElementWords; |
duke@435 | 2525 | } |
duke@435 | 2526 | |
duke@435 | 2527 | |
duke@435 | 2528 | uint SharedRuntime::out_preserve_stack_slots() { |
duke@435 | 2529 | return 0; |
duke@435 | 2530 | } |
duke@435 | 2531 | |
duke@435 | 2532 | |
duke@435 | 2533 | //------------------------------generate_deopt_blob---------------------------- |
duke@435 | 2534 | void SharedRuntime::generate_deopt_blob() { |
duke@435 | 2535 | // Allocate space for the code |
duke@435 | 2536 | ResourceMark rm; |
duke@435 | 2537 | // Setup code generation tools |
duke@435 | 2538 | CodeBuffer buffer("deopt_blob", 2048, 1024); |
duke@435 | 2539 | MacroAssembler* masm = new MacroAssembler(&buffer); |
duke@435 | 2540 | int frame_size_in_words; |
duke@435 | 2541 | OopMap* map = NULL; |
duke@435 | 2542 | OopMapSet *oop_maps = new OopMapSet(); |
duke@435 | 2543 | |
duke@435 | 2544 | // ------------- |
duke@435 | 2545 | // This code enters when returning to a de-optimized nmethod. A return |
duke@435 | 2546 | // address has been pushed on the the stack, and return values are in |
duke@435 | 2547 | // registers. |
duke@435 | 2548 | // If we are doing a normal deopt then we were called from the patched |
duke@435 | 2549 | // nmethod from the point we returned to the nmethod. So the return |
duke@435 | 2550 | // address on the stack is wrong by NativeCall::instruction_size |
duke@435 | 2551 | // We will adjust the value so it looks like we have the original return |
duke@435 | 2552 | // address on the stack (like when we eagerly deoptimized). |
duke@435 | 2553 | // In the case of an exception pending when deoptimizing, we enter |
duke@435 | 2554 | // with a return address on the stack that points after the call we patched |
duke@435 | 2555 | // into the exception handler. We have the following register state from, |
duke@435 | 2556 | // e.g., the forward exception stub (see stubGenerator_x86_64.cpp). |
duke@435 | 2557 | // rax: exception oop |
duke@435 | 2558 | // rbx: exception handler |
duke@435 | 2559 | // rdx: throwing pc |
duke@435 | 2560 | // So in this case we simply jam rdx into the useless return address and |
duke@435 | 2561 | // the stack looks just like we want. |
duke@435 | 2562 | // |
duke@435 | 2563 | // At this point we need to de-opt. We save the argument return |
duke@435 | 2564 | // registers. We call the first C routine, fetch_unroll_info(). This |
duke@435 | 2565 | // routine captures the return values and returns a structure which |
duke@435 | 2566 | // describes the current frame size and the sizes of all replacement frames. |
duke@435 | 2567 | // The current frame is compiled code and may contain many inlined |
duke@435 | 2568 | // functions, each with their own JVM state. We pop the current frame, then |
duke@435 | 2569 | // push all the new frames. Then we call the C routine unpack_frames() to |
duke@435 | 2570 | // populate these frames. Finally unpack_frames() returns us the new target |
duke@435 | 2571 | // address. Notice that callee-save registers are BLOWN here; they have |
duke@435 | 2572 | // already been captured in the vframeArray at the time the return PC was |
duke@435 | 2573 | // patched. |
duke@435 | 2574 | address start = __ pc(); |
duke@435 | 2575 | Label cont; |
duke@435 | 2576 | |
duke@435 | 2577 | // Prolog for non exception case! |
duke@435 | 2578 | |
duke@435 | 2579 | // Save everything in sight. |
duke@435 | 2580 | map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); |
duke@435 | 2581 | |
duke@435 | 2582 | // Normal deoptimization. Save exec mode for unpack_frames. |
coleenp@548 | 2583 | __ movl(r14, Deoptimization::Unpack_deopt); // callee-saved |
duke@435 | 2584 | __ jmp(cont); |
never@739 | 2585 | |
never@739 | 2586 | int reexecute_offset = __ pc() - start; |
never@739 | 2587 | |
never@739 | 2588 | // Reexecute case |
never@739 | 2589 | // return address is the pc describes what bci to do re-execute at |
never@739 | 2590 | |
never@739 | 2591 | // No need to update map as each call to save_live_registers will produce identical oopmap |
never@739 | 2592 | (void) RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); |
never@739 | 2593 | |
never@739 | 2594 | __ movl(r14, Deoptimization::Unpack_reexecute); // callee-saved |
never@739 | 2595 | __ jmp(cont); |
never@739 | 2596 | |
duke@435 | 2597 | int exception_offset = __ pc() - start; |
duke@435 | 2598 | |
duke@435 | 2599 | // Prolog for exception case |
duke@435 | 2600 | |
never@739 | 2601 | // all registers are dead at this entry point, except for rax, and |
never@739 | 2602 | // rdx which contain the exception oop and exception pc |
never@739 | 2603 | // respectively. Set them in TLS and fall thru to the |
never@739 | 2604 | // unpack_with_exception_in_tls entry point. |
never@739 | 2605 | |
never@739 | 2606 | __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), rdx); |
never@739 | 2607 | __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), rax); |
never@739 | 2608 | |
never@739 | 2609 | int exception_in_tls_offset = __ pc() - start; |
never@739 | 2610 | |
never@739 | 2611 | // new implementation because exception oop is now passed in JavaThread |
never@739 | 2612 | |
never@739 | 2613 | // Prolog for exception case |
never@739 | 2614 | // All registers must be preserved because they might be used by LinearScan |
never@739 | 2615 | // Exceptiop oop and throwing PC are passed in JavaThread |
never@739 | 2616 | // tos: stack at point of call to method that threw the exception (i.e. only |
never@739 | 2617 | // args are on the stack, no return address) |
never@739 | 2618 | |
never@739 | 2619 | // make room on stack for the return address |
never@739 | 2620 | // It will be patched later with the throwing pc. The correct value is not |
never@739 | 2621 | // available now because loading it from memory would destroy registers. |
never@739 | 2622 | __ push(0); |
duke@435 | 2623 | |
duke@435 | 2624 | // Save everything in sight. |
duke@435 | 2625 | map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); |
duke@435 | 2626 | |
never@739 | 2627 | // Now it is safe to overwrite any register |
never@739 | 2628 | |
duke@435 | 2629 | // Deopt during an exception. Save exec mode for unpack_frames. |
coleenp@548 | 2630 | __ movl(r14, Deoptimization::Unpack_exception); // callee-saved |
duke@435 | 2631 | |
never@739 | 2632 | // load throwing pc from JavaThread and patch it as the return address |
never@739 | 2633 | // of the current frame. Then clear the field in JavaThread |
never@739 | 2634 | |
never@739 | 2635 | __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset())); |
never@739 | 2636 | __ movptr(Address(rbp, wordSize), rdx); |
never@739 | 2637 | __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD); |
never@739 | 2638 | |
never@739 | 2639 | #ifdef ASSERT |
never@739 | 2640 | // verify that there is really an exception oop in JavaThread |
never@739 | 2641 | __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset())); |
never@739 | 2642 | __ verify_oop(rax); |
never@739 | 2643 | |
never@739 | 2644 | // verify that there is no pending exception |
never@739 | 2645 | Label no_pending_exception; |
never@739 | 2646 | __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset())); |
never@739 | 2647 | __ testptr(rax, rax); |
never@739 | 2648 | __ jcc(Assembler::zero, no_pending_exception); |
never@739 | 2649 | __ stop("must not have pending exception here"); |
never@739 | 2650 | __ bind(no_pending_exception); |
never@739 | 2651 | #endif |
never@739 | 2652 | |
duke@435 | 2653 | __ bind(cont); |
duke@435 | 2654 | |
duke@435 | 2655 | // Call C code. Need thread and this frame, but NOT official VM entry |
duke@435 | 2656 | // crud. We cannot block on this call, no GC can happen. |
duke@435 | 2657 | // |
duke@435 | 2658 | // UnrollBlock* fetch_unroll_info(JavaThread* thread) |
duke@435 | 2659 | |
duke@435 | 2660 | // fetch_unroll_info needs to call last_java_frame(). |
duke@435 | 2661 | |
duke@435 | 2662 | __ set_last_Java_frame(noreg, noreg, NULL); |
duke@435 | 2663 | #ifdef ASSERT |
duke@435 | 2664 | { Label L; |
never@739 | 2665 | __ cmpptr(Address(r15_thread, |
duke@435 | 2666 | JavaThread::last_Java_fp_offset()), |
never@739 | 2667 | (int32_t)0); |
duke@435 | 2668 | __ jcc(Assembler::equal, L); |
duke@435 | 2669 | __ stop("SharedRuntime::generate_deopt_blob: last_Java_fp not cleared"); |
duke@435 | 2670 | __ bind(L); |
duke@435 | 2671 | } |
duke@435 | 2672 | #endif // ASSERT |
never@739 | 2673 | __ mov(c_rarg0, r15_thread); |
duke@435 | 2674 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info))); |
duke@435 | 2675 | |
duke@435 | 2676 | // Need to have an oopmap that tells fetch_unroll_info where to |
duke@435 | 2677 | // find any register it might need. |
duke@435 | 2678 | oop_maps->add_gc_map(__ pc() - start, map); |
duke@435 | 2679 | |
duke@435 | 2680 | __ reset_last_Java_frame(false, false); |
duke@435 | 2681 | |
duke@435 | 2682 | // Load UnrollBlock* into rdi |
never@739 | 2683 | __ mov(rdi, rax); |
never@739 | 2684 | |
never@739 | 2685 | Label noException; |
never@1117 | 2686 | __ cmpl(r14, Deoptimization::Unpack_exception); // Was exception pending? |
never@739 | 2687 | __ jcc(Assembler::notEqual, noException); |
never@739 | 2688 | __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset())); |
never@739 | 2689 | // QQQ this is useless it was NULL above |
never@739 | 2690 | __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset())); |
never@739 | 2691 | __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD); |
never@739 | 2692 | __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD); |
never@739 | 2693 | |
never@739 | 2694 | __ verify_oop(rax); |
never@739 | 2695 | |
never@739 | 2696 | // Overwrite the result registers with the exception results. |
never@739 | 2697 | __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax); |
never@739 | 2698 | // I think this is useless |
never@739 | 2699 | __ movptr(Address(rsp, RegisterSaver::rdx_offset_in_bytes()), rdx); |
never@739 | 2700 | |
never@739 | 2701 | __ bind(noException); |
duke@435 | 2702 | |
duke@435 | 2703 | // Only register save data is on the stack. |
duke@435 | 2704 | // Now restore the result registers. Everything else is either dead |
duke@435 | 2705 | // or captured in the vframeArray. |
duke@435 | 2706 | RegisterSaver::restore_result_registers(masm); |
duke@435 | 2707 | |
duke@435 | 2708 | // All of the register save area has been popped of the stack. Only the |
duke@435 | 2709 | // return address remains. |
duke@435 | 2710 | |
duke@435 | 2711 | // Pop all the frames we must move/replace. |
duke@435 | 2712 | // |
duke@435 | 2713 | // Frame picture (youngest to oldest) |
duke@435 | 2714 | // 1: self-frame (no frame link) |
duke@435 | 2715 | // 2: deopting frame (no frame link) |
duke@435 | 2716 | // 3: caller of deopting frame (could be compiled/interpreted). |
duke@435 | 2717 | // |
duke@435 | 2718 | // Note: by leaving the return address of self-frame on the stack |
duke@435 | 2719 | // and using the size of frame 2 to adjust the stack |
duke@435 | 2720 | // when we are done the return to frame 3 will still be on the stack. |
duke@435 | 2721 | |
duke@435 | 2722 | // Pop deoptimized frame |
duke@435 | 2723 | __ movl(rcx, Address(rdi, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes())); |
never@739 | 2724 | __ addptr(rsp, rcx); |
duke@435 | 2725 | |
duke@435 | 2726 | // rsp should be pointing at the return address to the caller (3) |
duke@435 | 2727 | |
duke@435 | 2728 | // Stack bang to make sure there's enough room for these interpreter frames. |
duke@435 | 2729 | if (UseStackBanging) { |
duke@435 | 2730 | __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes())); |
duke@435 | 2731 | __ bang_stack_size(rbx, rcx); |
duke@435 | 2732 | } |
duke@435 | 2733 | |
duke@435 | 2734 | // Load address of array of frame pcs into rcx |
never@739 | 2735 | __ movptr(rcx, Address(rdi, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes())); |
duke@435 | 2736 | |
duke@435 | 2737 | // Trash the old pc |
never@739 | 2738 | __ addptr(rsp, wordSize); |
duke@435 | 2739 | |
duke@435 | 2740 | // Load address of array of frame sizes into rsi |
never@739 | 2741 | __ movptr(rsi, Address(rdi, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes())); |
duke@435 | 2742 | |
duke@435 | 2743 | // Load counter into rdx |
duke@435 | 2744 | __ movl(rdx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes())); |
duke@435 | 2745 | |
duke@435 | 2746 | // Pick up the initial fp we should save |
never@739 | 2747 | __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_fp_offset_in_bytes())); |
duke@435 | 2748 | |
duke@435 | 2749 | // Now adjust the caller's stack to make up for the extra locals |
duke@435 | 2750 | // but record the original sp so that we can save it in the skeletal interpreter |
duke@435 | 2751 | // frame and the stack walking of interpreter_sender will get the unextended sp |
duke@435 | 2752 | // value and not the "real" sp value. |
duke@435 | 2753 | |
duke@435 | 2754 | const Register sender_sp = r8; |
duke@435 | 2755 | |
never@739 | 2756 | __ mov(sender_sp, rsp); |
duke@435 | 2757 | __ movl(rbx, Address(rdi, |
duke@435 | 2758 | Deoptimization::UnrollBlock:: |
duke@435 | 2759 | caller_adjustment_offset_in_bytes())); |
never@739 | 2760 | __ subptr(rsp, rbx); |
duke@435 | 2761 | |
duke@435 | 2762 | // Push interpreter frames in a loop |
duke@435 | 2763 | Label loop; |
duke@435 | 2764 | __ bind(loop); |
never@739 | 2765 | __ movptr(rbx, Address(rsi, 0)); // Load frame size |
never@739 | 2766 | #ifdef CC_INTERP |
never@739 | 2767 | __ subptr(rbx, 4*wordSize); // we'll push pc and ebp by hand and |
never@739 | 2768 | #ifdef ASSERT |
never@739 | 2769 | __ push(0xDEADDEAD); // Make a recognizable pattern |
never@739 | 2770 | __ push(0xDEADDEAD); |
never@739 | 2771 | #else /* ASSERT */ |
never@739 | 2772 | __ subptr(rsp, 2*wordSize); // skip the "static long no_param" |
never@739 | 2773 | #endif /* ASSERT */ |
never@739 | 2774 | #else |
never@739 | 2775 | __ subptr(rbx, 2*wordSize); // We'll push pc and ebp by hand |
never@739 | 2776 | #endif // CC_INTERP |
never@739 | 2777 | __ pushptr(Address(rcx, 0)); // Save return address |
duke@435 | 2778 | __ enter(); // Save old & set new ebp |
never@739 | 2779 | __ subptr(rsp, rbx); // Prolog |
never@739 | 2780 | #ifdef CC_INTERP |
never@739 | 2781 | __ movptr(Address(rbp, |
never@739 | 2782 | -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))), |
never@739 | 2783 | sender_sp); // Make it walkable |
never@739 | 2784 | #else /* CC_INTERP */ |
duke@435 | 2785 | // This value is corrected by layout_activation_impl |
never@739 | 2786 | __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD ); |
never@739 | 2787 | __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), sender_sp); // Make it walkable |
never@739 | 2788 | #endif /* CC_INTERP */ |
never@739 | 2789 | __ mov(sender_sp, rsp); // Pass sender_sp to next frame |
never@739 | 2790 | __ addptr(rsi, wordSize); // Bump array pointer (sizes) |
never@739 | 2791 | __ addptr(rcx, wordSize); // Bump array pointer (pcs) |
duke@435 | 2792 | __ decrementl(rdx); // Decrement counter |
duke@435 | 2793 | __ jcc(Assembler::notZero, loop); |
never@739 | 2794 | __ pushptr(Address(rcx, 0)); // Save final return address |
duke@435 | 2795 | |
duke@435 | 2796 | // Re-push self-frame |
duke@435 | 2797 | __ enter(); // Save old & set new ebp |
duke@435 | 2798 | |
duke@435 | 2799 | // Allocate a full sized register save area. |
duke@435 | 2800 | // Return address and rbp are in place, so we allocate two less words. |
never@739 | 2801 | __ subptr(rsp, (frame_size_in_words - 2) * wordSize); |
duke@435 | 2802 | |
duke@435 | 2803 | // Restore frame locals after moving the frame |
duke@435 | 2804 | __ movdbl(Address(rsp, RegisterSaver::xmm0_offset_in_bytes()), xmm0); |
never@739 | 2805 | __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax); |
duke@435 | 2806 | |
duke@435 | 2807 | // Call C code. Need thread but NOT official VM entry |
duke@435 | 2808 | // crud. We cannot block on this call, no GC can happen. Call should |
duke@435 | 2809 | // restore return values to their stack-slots with the new SP. |
duke@435 | 2810 | // |
duke@435 | 2811 | // void Deoptimization::unpack_frames(JavaThread* thread, int exec_mode) |
duke@435 | 2812 | |
duke@435 | 2813 | // Use rbp because the frames look interpreted now |
duke@435 | 2814 | __ set_last_Java_frame(noreg, rbp, NULL); |
duke@435 | 2815 | |
never@739 | 2816 | __ mov(c_rarg0, r15_thread); |
coleenp@548 | 2817 | __ movl(c_rarg1, r14); // second arg: exec_mode |
duke@435 | 2818 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames))); |
duke@435 | 2819 | |
duke@435 | 2820 | // Set an oopmap for the call site |
duke@435 | 2821 | oop_maps->add_gc_map(__ pc() - start, |
duke@435 | 2822 | new OopMap( frame_size_in_words, 0 )); |
duke@435 | 2823 | |
duke@435 | 2824 | __ reset_last_Java_frame(true, false); |
duke@435 | 2825 | |
duke@435 | 2826 | // Collect return values |
duke@435 | 2827 | __ movdbl(xmm0, Address(rsp, RegisterSaver::xmm0_offset_in_bytes())); |
never@739 | 2828 | __ movptr(rax, Address(rsp, RegisterSaver::rax_offset_in_bytes())); |
never@739 | 2829 | // I think this is useless (throwing pc?) |
never@739 | 2830 | __ movptr(rdx, Address(rsp, RegisterSaver::rdx_offset_in_bytes())); |
duke@435 | 2831 | |
duke@435 | 2832 | // Pop self-frame. |
duke@435 | 2833 | __ leave(); // Epilog |
duke@435 | 2834 | |
duke@435 | 2835 | // Jump to interpreter |
duke@435 | 2836 | __ ret(0); |
duke@435 | 2837 | |
duke@435 | 2838 | // Make sure all code is generated |
duke@435 | 2839 | masm->flush(); |
duke@435 | 2840 | |
never@739 | 2841 | _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words); |
never@739 | 2842 | _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset); |
duke@435 | 2843 | } |
duke@435 | 2844 | |
duke@435 | 2845 | #ifdef COMPILER2 |
duke@435 | 2846 | //------------------------------generate_uncommon_trap_blob-------------------- |
duke@435 | 2847 | void SharedRuntime::generate_uncommon_trap_blob() { |
duke@435 | 2848 | // Allocate space for the code |
duke@435 | 2849 | ResourceMark rm; |
duke@435 | 2850 | // Setup code generation tools |
duke@435 | 2851 | CodeBuffer buffer("uncommon_trap_blob", 2048, 1024); |
duke@435 | 2852 | MacroAssembler* masm = new MacroAssembler(&buffer); |
duke@435 | 2853 | |
duke@435 | 2854 | assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned"); |
duke@435 | 2855 | |
duke@435 | 2856 | address start = __ pc(); |
duke@435 | 2857 | |
duke@435 | 2858 | // Push self-frame. We get here with a return address on the |
duke@435 | 2859 | // stack, so rsp is 8-byte aligned until we allocate our frame. |
never@739 | 2860 | __ subptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog! |
duke@435 | 2861 | |
duke@435 | 2862 | // No callee saved registers. rbp is assumed implicitly saved |
never@739 | 2863 | __ movptr(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp); |
duke@435 | 2864 | |
duke@435 | 2865 | // compiler left unloaded_class_index in j_rarg0 move to where the |
duke@435 | 2866 | // runtime expects it. |
duke@435 | 2867 | __ movl(c_rarg1, j_rarg0); |
duke@435 | 2868 | |
duke@435 | 2869 | __ set_last_Java_frame(noreg, noreg, NULL); |
duke@435 | 2870 | |
duke@435 | 2871 | // Call C code. Need thread but NOT official VM entry |
duke@435 | 2872 | // crud. We cannot block on this call, no GC can happen. Call should |
duke@435 | 2873 | // capture callee-saved registers as well as return values. |
duke@435 | 2874 | // Thread is in rdi already. |
duke@435 | 2875 | // |
duke@435 | 2876 | // UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index); |
duke@435 | 2877 | |
never@739 | 2878 | __ mov(c_rarg0, r15_thread); |
duke@435 | 2879 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap))); |
duke@435 | 2880 | |
duke@435 | 2881 | // Set an oopmap for the call site |
duke@435 | 2882 | OopMapSet* oop_maps = new OopMapSet(); |
duke@435 | 2883 | OopMap* map = new OopMap(SimpleRuntimeFrame::framesize, 0); |
duke@435 | 2884 | |
duke@435 | 2885 | // location of rbp is known implicitly by the frame sender code |
duke@435 | 2886 | |
duke@435 | 2887 | oop_maps->add_gc_map(__ pc() - start, map); |
duke@435 | 2888 | |
duke@435 | 2889 | __ reset_last_Java_frame(false, false); |
duke@435 | 2890 | |
duke@435 | 2891 | // Load UnrollBlock* into rdi |
never@739 | 2892 | __ mov(rdi, rax); |
duke@435 | 2893 | |
duke@435 | 2894 | // Pop all the frames we must move/replace. |
duke@435 | 2895 | // |
duke@435 | 2896 | // Frame picture (youngest to oldest) |
duke@435 | 2897 | // 1: self-frame (no frame link) |
duke@435 | 2898 | // 2: deopting frame (no frame link) |
duke@435 | 2899 | // 3: caller of deopting frame (could be compiled/interpreted). |
duke@435 | 2900 | |
duke@435 | 2901 | // Pop self-frame. We have no frame, and must rely only on rax and rsp. |
never@739 | 2902 | __ addptr(rsp, (SimpleRuntimeFrame::framesize - 2) << LogBytesPerInt); // Epilog! |
duke@435 | 2903 | |
duke@435 | 2904 | // Pop deoptimized frame (int) |
duke@435 | 2905 | __ movl(rcx, Address(rdi, |
duke@435 | 2906 | Deoptimization::UnrollBlock:: |
duke@435 | 2907 | size_of_deoptimized_frame_offset_in_bytes())); |
never@739 | 2908 | __ addptr(rsp, rcx); |
duke@435 | 2909 | |
duke@435 | 2910 | // rsp should be pointing at the return address to the caller (3) |
duke@435 | 2911 | |
duke@435 | 2912 | // Stack bang to make sure there's enough room for these interpreter frames. |
duke@435 | 2913 | if (UseStackBanging) { |
duke@435 | 2914 | __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes())); |
duke@435 | 2915 | __ bang_stack_size(rbx, rcx); |
duke@435 | 2916 | } |
duke@435 | 2917 | |
duke@435 | 2918 | // Load address of array of frame pcs into rcx (address*) |
never@739 | 2919 | __ movptr(rcx, |
never@739 | 2920 | Address(rdi, |
never@739 | 2921 | Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes())); |
duke@435 | 2922 | |
duke@435 | 2923 | // Trash the return pc |
never@739 | 2924 | __ addptr(rsp, wordSize); |
duke@435 | 2925 | |
duke@435 | 2926 | // Load address of array of frame sizes into rsi (intptr_t*) |
never@739 | 2927 | __ movptr(rsi, Address(rdi, |
never@739 | 2928 | Deoptimization::UnrollBlock:: |
never@739 | 2929 | frame_sizes_offset_in_bytes())); |
duke@435 | 2930 | |
duke@435 | 2931 | // Counter |
duke@435 | 2932 | __ movl(rdx, Address(rdi, |
duke@435 | 2933 | Deoptimization::UnrollBlock:: |
duke@435 | 2934 | number_of_frames_offset_in_bytes())); // (int) |
duke@435 | 2935 | |
duke@435 | 2936 | // Pick up the initial fp we should save |
never@739 | 2937 | __ movptr(rbp, |
never@739 | 2938 | Address(rdi, |
never@739 | 2939 | Deoptimization::UnrollBlock::initial_fp_offset_in_bytes())); |
duke@435 | 2940 | |
duke@435 | 2941 | // Now adjust the caller's stack to make up for the extra locals but |
duke@435 | 2942 | // record the original sp so that we can save it in the skeletal |
duke@435 | 2943 | // interpreter frame and the stack walking of interpreter_sender |
duke@435 | 2944 | // will get the unextended sp value and not the "real" sp value. |
duke@435 | 2945 | |
duke@435 | 2946 | const Register sender_sp = r8; |
duke@435 | 2947 | |
never@739 | 2948 | __ mov(sender_sp, rsp); |
duke@435 | 2949 | __ movl(rbx, Address(rdi, |
duke@435 | 2950 | Deoptimization::UnrollBlock:: |
duke@435 | 2951 | caller_adjustment_offset_in_bytes())); // (int) |
never@739 | 2952 | __ subptr(rsp, rbx); |
duke@435 | 2953 | |
duke@435 | 2954 | // Push interpreter frames in a loop |
duke@435 | 2955 | Label loop; |
duke@435 | 2956 | __ bind(loop); |
never@739 | 2957 | __ movptr(rbx, Address(rsi, 0)); // Load frame size |
never@739 | 2958 | __ subptr(rbx, 2 * wordSize); // We'll push pc and rbp by hand |
never@739 | 2959 | __ pushptr(Address(rcx, 0)); // Save return address |
never@739 | 2960 | __ enter(); // Save old & set new rbp |
never@739 | 2961 | __ subptr(rsp, rbx); // Prolog |
coleenp@955 | 2962 | #ifdef CC_INTERP |
coleenp@955 | 2963 | __ movptr(Address(rbp, |
coleenp@955 | 2964 | -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))), |
coleenp@955 | 2965 | sender_sp); // Make it walkable |
coleenp@955 | 2966 | #else // CC_INTERP |
never@739 | 2967 | __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), |
never@739 | 2968 | sender_sp); // Make it walkable |
duke@435 | 2969 | // This value is corrected by layout_activation_impl |
never@739 | 2970 | __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD ); |
coleenp@955 | 2971 | #endif // CC_INTERP |
never@739 | 2972 | __ mov(sender_sp, rsp); // Pass sender_sp to next frame |
never@739 | 2973 | __ addptr(rsi, wordSize); // Bump array pointer (sizes) |
never@739 | 2974 | __ addptr(rcx, wordSize); // Bump array pointer (pcs) |
never@739 | 2975 | __ decrementl(rdx); // Decrement counter |
duke@435 | 2976 | __ jcc(Assembler::notZero, loop); |
never@739 | 2977 | __ pushptr(Address(rcx, 0)); // Save final return address |
duke@435 | 2978 | |
duke@435 | 2979 | // Re-push self-frame |
duke@435 | 2980 | __ enter(); // Save old & set new rbp |
never@739 | 2981 | __ subptr(rsp, (SimpleRuntimeFrame::framesize - 4) << LogBytesPerInt); |
duke@435 | 2982 | // Prolog |
duke@435 | 2983 | |
duke@435 | 2984 | // Use rbp because the frames look interpreted now |
duke@435 | 2985 | __ set_last_Java_frame(noreg, rbp, NULL); |
duke@435 | 2986 | |
duke@435 | 2987 | // Call C code. Need thread but NOT official VM entry |
duke@435 | 2988 | // crud. We cannot block on this call, no GC can happen. Call should |
duke@435 | 2989 | // restore return values to their stack-slots with the new SP. |
duke@435 | 2990 | // Thread is in rdi already. |
duke@435 | 2991 | // |
duke@435 | 2992 | // BasicType unpack_frames(JavaThread* thread, int exec_mode); |
duke@435 | 2993 | |
never@739 | 2994 | __ mov(c_rarg0, r15_thread); |
duke@435 | 2995 | __ movl(c_rarg1, Deoptimization::Unpack_uncommon_trap); |
duke@435 | 2996 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames))); |
duke@435 | 2997 | |
duke@435 | 2998 | // Set an oopmap for the call site |
duke@435 | 2999 | oop_maps->add_gc_map(__ pc() - start, new OopMap(SimpleRuntimeFrame::framesize, 0)); |
duke@435 | 3000 | |
duke@435 | 3001 | __ reset_last_Java_frame(true, false); |
duke@435 | 3002 | |
duke@435 | 3003 | // Pop self-frame. |
duke@435 | 3004 | __ leave(); // Epilog |
duke@435 | 3005 | |
duke@435 | 3006 | // Jump to interpreter |
duke@435 | 3007 | __ ret(0); |
duke@435 | 3008 | |
duke@435 | 3009 | // Make sure all code is generated |
duke@435 | 3010 | masm->flush(); |
duke@435 | 3011 | |
duke@435 | 3012 | _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps, |
duke@435 | 3013 | SimpleRuntimeFrame::framesize >> 1); |
duke@435 | 3014 | } |
duke@435 | 3015 | #endif // COMPILER2 |
duke@435 | 3016 | |
duke@435 | 3017 | |
duke@435 | 3018 | //------------------------------generate_handler_blob------ |
duke@435 | 3019 | // |
duke@435 | 3020 | // Generate a special Compile2Runtime blob that saves all registers, |
duke@435 | 3021 | // and setup oopmap. |
duke@435 | 3022 | // |
duke@435 | 3023 | static SafepointBlob* generate_handler_blob(address call_ptr, bool cause_return) { |
duke@435 | 3024 | assert(StubRoutines::forward_exception_entry() != NULL, |
duke@435 | 3025 | "must be generated before"); |
duke@435 | 3026 | |
duke@435 | 3027 | ResourceMark rm; |
duke@435 | 3028 | OopMapSet *oop_maps = new OopMapSet(); |
duke@435 | 3029 | OopMap* map; |
duke@435 | 3030 | |
duke@435 | 3031 | // Allocate space for the code. Setup code generation tools. |
duke@435 | 3032 | CodeBuffer buffer("handler_blob", 2048, 1024); |
duke@435 | 3033 | MacroAssembler* masm = new MacroAssembler(&buffer); |
duke@435 | 3034 | |
duke@435 | 3035 | address start = __ pc(); |
duke@435 | 3036 | address call_pc = NULL; |
duke@435 | 3037 | int frame_size_in_words; |
duke@435 | 3038 | |
duke@435 | 3039 | // Make room for return address (or push it again) |
duke@435 | 3040 | if (!cause_return) { |
never@739 | 3041 | __ push(rbx); |
duke@435 | 3042 | } |
duke@435 | 3043 | |
duke@435 | 3044 | // Save registers, fpu state, and flags |
duke@435 | 3045 | map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); |
duke@435 | 3046 | |
duke@435 | 3047 | // The following is basically a call_VM. However, we need the precise |
duke@435 | 3048 | // address of the call in order to generate an oopmap. Hence, we do all the |
duke@435 | 3049 | // work outselves. |
duke@435 | 3050 | |
duke@435 | 3051 | __ set_last_Java_frame(noreg, noreg, NULL); |
duke@435 | 3052 | |
duke@435 | 3053 | // The return address must always be correct so that frame constructor never |
duke@435 | 3054 | // sees an invalid pc. |
duke@435 | 3055 | |
duke@435 | 3056 | if (!cause_return) { |
duke@435 | 3057 | // overwrite the dummy value we pushed on entry |
never@739 | 3058 | __ movptr(c_rarg0, Address(r15_thread, JavaThread::saved_exception_pc_offset())); |
never@739 | 3059 | __ movptr(Address(rbp, wordSize), c_rarg0); |
duke@435 | 3060 | } |
duke@435 | 3061 | |
duke@435 | 3062 | // Do the call |
never@739 | 3063 | __ mov(c_rarg0, r15_thread); |
duke@435 | 3064 | __ call(RuntimeAddress(call_ptr)); |
duke@435 | 3065 | |
duke@435 | 3066 | // Set an oopmap for the call site. This oopmap will map all |
duke@435 | 3067 | // oop-registers and debug-info registers as callee-saved. This |
duke@435 | 3068 | // will allow deoptimization at this safepoint to find all possible |
duke@435 | 3069 | // debug-info recordings, as well as let GC find all oops. |
duke@435 | 3070 | |
duke@435 | 3071 | oop_maps->add_gc_map( __ pc() - start, map); |
duke@435 | 3072 | |
duke@435 | 3073 | Label noException; |
duke@435 | 3074 | |
duke@435 | 3075 | __ reset_last_Java_frame(false, false); |
duke@435 | 3076 | |
never@739 | 3077 | __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
duke@435 | 3078 | __ jcc(Assembler::equal, noException); |
duke@435 | 3079 | |
duke@435 | 3080 | // Exception pending |
duke@435 | 3081 | |
duke@435 | 3082 | RegisterSaver::restore_live_registers(masm); |
duke@435 | 3083 | |
duke@435 | 3084 | __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
duke@435 | 3085 | |
duke@435 | 3086 | // No exception case |
duke@435 | 3087 | __ bind(noException); |
duke@435 | 3088 | |
duke@435 | 3089 | // Normal exit, restore registers and exit. |
duke@435 | 3090 | RegisterSaver::restore_live_registers(masm); |
duke@435 | 3091 | |
duke@435 | 3092 | __ ret(0); |
duke@435 | 3093 | |
duke@435 | 3094 | // Make sure all code is generated |
duke@435 | 3095 | masm->flush(); |
duke@435 | 3096 | |
duke@435 | 3097 | // Fill-out other meta info |
duke@435 | 3098 | return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words); |
duke@435 | 3099 | } |
duke@435 | 3100 | |
duke@435 | 3101 | // |
duke@435 | 3102 | // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss |
duke@435 | 3103 | // |
duke@435 | 3104 | // Generate a stub that calls into vm to find out the proper destination |
duke@435 | 3105 | // of a java call. All the argument registers are live at this point |
duke@435 | 3106 | // but since this is generic code we don't know what they are and the caller |
duke@435 | 3107 | // must do any gc of the args. |
duke@435 | 3108 | // |
duke@435 | 3109 | static RuntimeStub* generate_resolve_blob(address destination, const char* name) { |
duke@435 | 3110 | assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); |
duke@435 | 3111 | |
duke@435 | 3112 | // allocate space for the code |
duke@435 | 3113 | ResourceMark rm; |
duke@435 | 3114 | |
duke@435 | 3115 | CodeBuffer buffer(name, 1000, 512); |
duke@435 | 3116 | MacroAssembler* masm = new MacroAssembler(&buffer); |
duke@435 | 3117 | |
duke@435 | 3118 | int frame_size_in_words; |
duke@435 | 3119 | |
duke@435 | 3120 | OopMapSet *oop_maps = new OopMapSet(); |
duke@435 | 3121 | OopMap* map = NULL; |
duke@435 | 3122 | |
duke@435 | 3123 | int start = __ offset(); |
duke@435 | 3124 | |
duke@435 | 3125 | map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); |
duke@435 | 3126 | |
duke@435 | 3127 | int frame_complete = __ offset(); |
duke@435 | 3128 | |
duke@435 | 3129 | __ set_last_Java_frame(noreg, noreg, NULL); |
duke@435 | 3130 | |
never@739 | 3131 | __ mov(c_rarg0, r15_thread); |
duke@435 | 3132 | |
duke@435 | 3133 | __ call(RuntimeAddress(destination)); |
duke@435 | 3134 | |
duke@435 | 3135 | |
duke@435 | 3136 | // Set an oopmap for the call site. |
duke@435 | 3137 | // We need this not only for callee-saved registers, but also for volatile |
duke@435 | 3138 | // registers that the compiler might be keeping live across a safepoint. |
duke@435 | 3139 | |
duke@435 | 3140 | oop_maps->add_gc_map( __ offset() - start, map); |
duke@435 | 3141 | |
duke@435 | 3142 | // rax contains the address we are going to jump to assuming no exception got installed |
duke@435 | 3143 | |
duke@435 | 3144 | // clear last_Java_sp |
duke@435 | 3145 | __ reset_last_Java_frame(false, false); |
duke@435 | 3146 | // check for pending exceptions |
duke@435 | 3147 | Label pending; |
never@739 | 3148 | __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
duke@435 | 3149 | __ jcc(Assembler::notEqual, pending); |
duke@435 | 3150 | |
duke@435 | 3151 | // get the returned methodOop |
never@739 | 3152 | __ movptr(rbx, Address(r15_thread, JavaThread::vm_result_offset())); |
never@739 | 3153 | __ movptr(Address(rsp, RegisterSaver::rbx_offset_in_bytes()), rbx); |
never@739 | 3154 | |
never@739 | 3155 | __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax); |
duke@435 | 3156 | |
duke@435 | 3157 | RegisterSaver::restore_live_registers(masm); |
duke@435 | 3158 | |
duke@435 | 3159 | // We are back the the original state on entry and ready to go. |
duke@435 | 3160 | |
duke@435 | 3161 | __ jmp(rax); |
duke@435 | 3162 | |
duke@435 | 3163 | // Pending exception after the safepoint |
duke@435 | 3164 | |
duke@435 | 3165 | __ bind(pending); |
duke@435 | 3166 | |
duke@435 | 3167 | RegisterSaver::restore_live_registers(masm); |
duke@435 | 3168 | |
duke@435 | 3169 | // exception pending => remove activation and forward to exception handler |
duke@435 | 3170 | |
duke@435 | 3171 | __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int)NULL_WORD); |
duke@435 | 3172 | |
never@739 | 3173 | __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset())); |
duke@435 | 3174 | __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
duke@435 | 3175 | |
duke@435 | 3176 | // ------------- |
duke@435 | 3177 | // make sure all code is generated |
duke@435 | 3178 | masm->flush(); |
duke@435 | 3179 | |
duke@435 | 3180 | // return the blob |
duke@435 | 3181 | // frame_size_words or bytes?? |
duke@435 | 3182 | return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true); |
duke@435 | 3183 | } |
duke@435 | 3184 | |
duke@435 | 3185 | |
duke@435 | 3186 | void SharedRuntime::generate_stubs() { |
duke@435 | 3187 | |
duke@435 | 3188 | _wrong_method_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method), |
duke@435 | 3189 | "wrong_method_stub"); |
duke@435 | 3190 | _ic_miss_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method_ic_miss), |
duke@435 | 3191 | "ic_miss_stub"); |
duke@435 | 3192 | _resolve_opt_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_opt_virtual_call_C), |
duke@435 | 3193 | "resolve_opt_virtual_call"); |
duke@435 | 3194 | |
duke@435 | 3195 | _resolve_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_virtual_call_C), |
duke@435 | 3196 | "resolve_virtual_call"); |
duke@435 | 3197 | |
duke@435 | 3198 | _resolve_static_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_static_call_C), |
duke@435 | 3199 | "resolve_static_call"); |
duke@435 | 3200 | _polling_page_safepoint_handler_blob = |
duke@435 | 3201 | generate_handler_blob(CAST_FROM_FN_PTR(address, |
duke@435 | 3202 | SafepointSynchronize::handle_polling_page_exception), false); |
duke@435 | 3203 | |
duke@435 | 3204 | _polling_page_return_handler_blob = |
duke@435 | 3205 | generate_handler_blob(CAST_FROM_FN_PTR(address, |
duke@435 | 3206 | SafepointSynchronize::handle_polling_page_exception), true); |
duke@435 | 3207 | |
duke@435 | 3208 | generate_deopt_blob(); |
duke@435 | 3209 | |
duke@435 | 3210 | #ifdef COMPILER2 |
duke@435 | 3211 | generate_uncommon_trap_blob(); |
duke@435 | 3212 | #endif // COMPILER2 |
duke@435 | 3213 | } |
duke@435 | 3214 | |
duke@435 | 3215 | |
duke@435 | 3216 | #ifdef COMPILER2 |
duke@435 | 3217 | // This is here instead of runtime_x86_64.cpp because it uses SimpleRuntimeFrame |
duke@435 | 3218 | // |
duke@435 | 3219 | //------------------------------generate_exception_blob--------------------------- |
duke@435 | 3220 | // creates exception blob at the end |
duke@435 | 3221 | // Using exception blob, this code is jumped from a compiled method. |
duke@435 | 3222 | // (see emit_exception_handler in x86_64.ad file) |
duke@435 | 3223 | // |
duke@435 | 3224 | // Given an exception pc at a call we call into the runtime for the |
duke@435 | 3225 | // handler in this method. This handler might merely restore state |
duke@435 | 3226 | // (i.e. callee save registers) unwind the frame and jump to the |
duke@435 | 3227 | // exception handler for the nmethod if there is no Java level handler |
duke@435 | 3228 | // for the nmethod. |
duke@435 | 3229 | // |
duke@435 | 3230 | // This code is entered with a jmp. |
duke@435 | 3231 | // |
duke@435 | 3232 | // Arguments: |
duke@435 | 3233 | // rax: exception oop |
duke@435 | 3234 | // rdx: exception pc |
duke@435 | 3235 | // |
duke@435 | 3236 | // Results: |
duke@435 | 3237 | // rax: exception oop |
duke@435 | 3238 | // rdx: exception pc in caller or ??? |
duke@435 | 3239 | // destination: exception handler of caller |
duke@435 | 3240 | // |
duke@435 | 3241 | // Note: the exception pc MUST be at a call (precise debug information) |
duke@435 | 3242 | // Registers rax, rdx, rcx, rsi, rdi, r8-r11 are not callee saved. |
duke@435 | 3243 | // |
duke@435 | 3244 | |
duke@435 | 3245 | void OptoRuntime::generate_exception_blob() { |
duke@435 | 3246 | assert(!OptoRuntime::is_callee_saved_register(RDX_num), ""); |
duke@435 | 3247 | assert(!OptoRuntime::is_callee_saved_register(RAX_num), ""); |
duke@435 | 3248 | assert(!OptoRuntime::is_callee_saved_register(RCX_num), ""); |
duke@435 | 3249 | |
duke@435 | 3250 | assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned"); |
duke@435 | 3251 | |
duke@435 | 3252 | // Allocate space for the code |
duke@435 | 3253 | ResourceMark rm; |
duke@435 | 3254 | // Setup code generation tools |
duke@435 | 3255 | CodeBuffer buffer("exception_blob", 2048, 1024); |
duke@435 | 3256 | MacroAssembler* masm = new MacroAssembler(&buffer); |
duke@435 | 3257 | |
duke@435 | 3258 | |
duke@435 | 3259 | address start = __ pc(); |
duke@435 | 3260 | |
duke@435 | 3261 | // Exception pc is 'return address' for stack walker |
never@739 | 3262 | __ push(rdx); |
never@739 | 3263 | __ subptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Prolog |
duke@435 | 3264 | |
duke@435 | 3265 | // Save callee-saved registers. See x86_64.ad. |
duke@435 | 3266 | |
duke@435 | 3267 | // rbp is an implicitly saved callee saved register (i.e. the calling |
duke@435 | 3268 | // convention will save restore it in prolog/epilog) Other than that |
duke@435 | 3269 | // there are no callee save registers now that adapter frames are gone. |
duke@435 | 3270 | |
never@739 | 3271 | __ movptr(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp); |
duke@435 | 3272 | |
duke@435 | 3273 | // Store exception in Thread object. We cannot pass any arguments to the |
duke@435 | 3274 | // handle_exception call, since we do not want to make any assumption |
duke@435 | 3275 | // about the size of the frame where the exception happened in. |
duke@435 | 3276 | // c_rarg0 is either rdi (Linux) or rcx (Windows). |
never@739 | 3277 | __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()),rax); |
never@739 | 3278 | __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), rdx); |
duke@435 | 3279 | |
duke@435 | 3280 | // This call does all the hard work. It checks if an exception handler |
duke@435 | 3281 | // exists in the method. |
duke@435 | 3282 | // If so, it returns the handler address. |
duke@435 | 3283 | // If not, it prepares for stack-unwinding, restoring the callee-save |
duke@435 | 3284 | // registers of the frame being removed. |
duke@435 | 3285 | // |
duke@435 | 3286 | // address OptoRuntime::handle_exception_C(JavaThread* thread) |
duke@435 | 3287 | |
duke@435 | 3288 | __ set_last_Java_frame(noreg, noreg, NULL); |
never@739 | 3289 | __ mov(c_rarg0, r15_thread); |
duke@435 | 3290 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C))); |
duke@435 | 3291 | |
duke@435 | 3292 | // Set an oopmap for the call site. This oopmap will only be used if we |
duke@435 | 3293 | // are unwinding the stack. Hence, all locations will be dead. |
duke@435 | 3294 | // Callee-saved registers will be the same as the frame above (i.e., |
duke@435 | 3295 | // handle_exception_stub), since they were restored when we got the |
duke@435 | 3296 | // exception. |
duke@435 | 3297 | |
duke@435 | 3298 | OopMapSet* oop_maps = new OopMapSet(); |
duke@435 | 3299 | |
duke@435 | 3300 | oop_maps->add_gc_map( __ pc()-start, new OopMap(SimpleRuntimeFrame::framesize, 0)); |
duke@435 | 3301 | |
duke@435 | 3302 | __ reset_last_Java_frame(false, false); |
duke@435 | 3303 | |
duke@435 | 3304 | // Restore callee-saved registers |
duke@435 | 3305 | |
duke@435 | 3306 | // rbp is an implicitly saved callee saved register (i.e. the calling |
duke@435 | 3307 | // convention will save restore it in prolog/epilog) Other than that |
duke@435 | 3308 | // there are no callee save registers no that adapter frames are gone. |
duke@435 | 3309 | |
never@739 | 3310 | __ movptr(rbp, Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt)); |
never@739 | 3311 | |
never@739 | 3312 | __ addptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog |
never@739 | 3313 | __ pop(rdx); // No need for exception pc anymore |
duke@435 | 3314 | |
duke@435 | 3315 | // rax: exception handler |
duke@435 | 3316 | |
twisti@1803 | 3317 | // Restore SP from BP if the exception PC is a MethodHandle call site. |
twisti@1803 | 3318 | __ cmpl(Address(r15_thread, JavaThread::is_method_handle_return_offset()), 0); |
twisti@1922 | 3319 | __ cmovptr(Assembler::notEqual, rsp, rbp_mh_SP_save); |
twisti@1570 | 3320 | |
duke@435 | 3321 | // We have a handler in rax (could be deopt blob). |
never@739 | 3322 | __ mov(r8, rax); |
duke@435 | 3323 | |
duke@435 | 3324 | // Get the exception oop |
never@739 | 3325 | __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset())); |
duke@435 | 3326 | // Get the exception pc in case we are deoptimized |
never@739 | 3327 | __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset())); |
duke@435 | 3328 | #ifdef ASSERT |
duke@435 | 3329 | __ movptr(Address(r15_thread, JavaThread::exception_handler_pc_offset()), (int)NULL_WORD); |
duke@435 | 3330 | __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int)NULL_WORD); |
duke@435 | 3331 | #endif |
duke@435 | 3332 | // Clear the exception oop so GC no longer processes it as a root. |
duke@435 | 3333 | __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), (int)NULL_WORD); |
duke@435 | 3334 | |
duke@435 | 3335 | // rax: exception oop |
duke@435 | 3336 | // r8: exception handler |
duke@435 | 3337 | // rdx: exception pc |
duke@435 | 3338 | // Jump to handler |
duke@435 | 3339 | |
duke@435 | 3340 | __ jmp(r8); |
duke@435 | 3341 | |
duke@435 | 3342 | // Make sure all code is generated |
duke@435 | 3343 | masm->flush(); |
duke@435 | 3344 | |
duke@435 | 3345 | // Set exception blob |
duke@435 | 3346 | _exception_blob = ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1); |
duke@435 | 3347 | } |
duke@435 | 3348 | #endif // COMPILER2 |