Fri, 03 Jun 2011 22:31:43 -0700
7045514: SPARC assembly code for JSR 292 ricochet frames
Reviewed-by: kvn, jrose
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 | #define __ masm-> |
duke@435 | 45 | |
xlu@959 | 46 | const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size; |
xlu@959 | 47 | |
duke@435 | 48 | class RegisterSaver { |
duke@435 | 49 | enum { FPU_regs_live = 8 /*for the FPU stack*/+8/*eight more for XMM registers*/ }; |
duke@435 | 50 | // Capture info about frame layout |
duke@435 | 51 | enum layout { |
duke@435 | 52 | fpu_state_off = 0, |
duke@435 | 53 | fpu_state_end = fpu_state_off+FPUStateSizeInWords-1, |
duke@435 | 54 | st0_off, st0H_off, |
duke@435 | 55 | st1_off, st1H_off, |
duke@435 | 56 | st2_off, st2H_off, |
duke@435 | 57 | st3_off, st3H_off, |
duke@435 | 58 | st4_off, st4H_off, |
duke@435 | 59 | st5_off, st5H_off, |
duke@435 | 60 | st6_off, st6H_off, |
duke@435 | 61 | st7_off, st7H_off, |
duke@435 | 62 | |
duke@435 | 63 | xmm0_off, xmm0H_off, |
duke@435 | 64 | xmm1_off, xmm1H_off, |
duke@435 | 65 | xmm2_off, xmm2H_off, |
duke@435 | 66 | xmm3_off, xmm3H_off, |
duke@435 | 67 | xmm4_off, xmm4H_off, |
duke@435 | 68 | xmm5_off, xmm5H_off, |
duke@435 | 69 | xmm6_off, xmm6H_off, |
duke@435 | 70 | xmm7_off, xmm7H_off, |
duke@435 | 71 | flags_off, |
duke@435 | 72 | rdi_off, |
duke@435 | 73 | rsi_off, |
duke@435 | 74 | ignore_off, // extra copy of rbp, |
duke@435 | 75 | rsp_off, |
duke@435 | 76 | rbx_off, |
duke@435 | 77 | rdx_off, |
duke@435 | 78 | rcx_off, |
duke@435 | 79 | rax_off, |
duke@435 | 80 | // The frame sender code expects that rbp will be in the "natural" place and |
duke@435 | 81 | // will override any oopMap setting for it. We must therefore force the layout |
duke@435 | 82 | // so that it agrees with the frame sender code. |
duke@435 | 83 | rbp_off, |
duke@435 | 84 | return_off, // slot for return address |
duke@435 | 85 | reg_save_size }; |
duke@435 | 86 | |
duke@435 | 87 | |
duke@435 | 88 | public: |
duke@435 | 89 | |
duke@435 | 90 | static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, |
duke@435 | 91 | int* total_frame_words, bool verify_fpu = true); |
duke@435 | 92 | static void restore_live_registers(MacroAssembler* masm); |
duke@435 | 93 | |
duke@435 | 94 | static int rax_offset() { return rax_off; } |
duke@435 | 95 | static int rbx_offset() { return rbx_off; } |
duke@435 | 96 | |
duke@435 | 97 | // Offsets into the register save area |
duke@435 | 98 | // Used by deoptimization when it is managing result register |
duke@435 | 99 | // values on its own |
duke@435 | 100 | |
duke@435 | 101 | static int raxOffset(void) { return rax_off; } |
duke@435 | 102 | static int rdxOffset(void) { return rdx_off; } |
duke@435 | 103 | static int rbxOffset(void) { return rbx_off; } |
duke@435 | 104 | static int xmm0Offset(void) { return xmm0_off; } |
duke@435 | 105 | // This really returns a slot in the fp save area, which one is not important |
duke@435 | 106 | static int fpResultOffset(void) { return st0_off; } |
duke@435 | 107 | |
duke@435 | 108 | // During deoptimization only the result register need to be restored |
duke@435 | 109 | // all the other values have already been extracted. |
duke@435 | 110 | |
duke@435 | 111 | static void restore_result_registers(MacroAssembler* masm); |
duke@435 | 112 | |
duke@435 | 113 | }; |
duke@435 | 114 | |
duke@435 | 115 | OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, |
duke@435 | 116 | int* total_frame_words, bool verify_fpu) { |
duke@435 | 117 | |
duke@435 | 118 | int frame_size_in_bytes = (reg_save_size + additional_frame_words) * wordSize; |
duke@435 | 119 | int frame_words = frame_size_in_bytes / wordSize; |
duke@435 | 120 | *total_frame_words = frame_words; |
duke@435 | 121 | |
duke@435 | 122 | assert(FPUStateSizeInWords == 27, "update stack layout"); |
duke@435 | 123 | |
duke@435 | 124 | // save registers, fpu state, and flags |
duke@435 | 125 | // We assume caller has already has return address slot on the stack |
duke@435 | 126 | // We push epb twice in this sequence because we want the real rbp, |
never@739 | 127 | // to be under the return like a normal enter and we want to use pusha |
duke@435 | 128 | // We push by hand instead of pusing push |
duke@435 | 129 | __ enter(); |
never@739 | 130 | __ pusha(); |
never@739 | 131 | __ pushf(); |
never@739 | 132 | __ subptr(rsp,FPU_regs_live*sizeof(jdouble)); // Push FPU registers space |
duke@435 | 133 | __ push_FPU_state(); // Save FPU state & init |
duke@435 | 134 | |
duke@435 | 135 | if (verify_fpu) { |
duke@435 | 136 | // Some stubs may have non standard FPU control word settings so |
duke@435 | 137 | // only check and reset the value when it required to be the |
duke@435 | 138 | // standard value. The safepoint blob in particular can be used |
duke@435 | 139 | // in methods which are using the 24 bit control word for |
duke@435 | 140 | // optimized float math. |
duke@435 | 141 | |
duke@435 | 142 | #ifdef ASSERT |
duke@435 | 143 | // Make sure the control word has the expected value |
duke@435 | 144 | Label ok; |
duke@435 | 145 | __ cmpw(Address(rsp, 0), StubRoutines::fpu_cntrl_wrd_std()); |
duke@435 | 146 | __ jccb(Assembler::equal, ok); |
duke@435 | 147 | __ stop("corrupted control word detected"); |
duke@435 | 148 | __ bind(ok); |
duke@435 | 149 | #endif |
duke@435 | 150 | |
duke@435 | 151 | // Reset the control word to guard against exceptions being unmasked |
duke@435 | 152 | // since fstp_d can cause FPU stack underflow exceptions. Write it |
duke@435 | 153 | // into the on stack copy and then reload that to make sure that the |
duke@435 | 154 | // current and future values are correct. |
duke@435 | 155 | __ movw(Address(rsp, 0), StubRoutines::fpu_cntrl_wrd_std()); |
duke@435 | 156 | } |
duke@435 | 157 | |
duke@435 | 158 | __ frstor(Address(rsp, 0)); |
duke@435 | 159 | if (!verify_fpu) { |
duke@435 | 160 | // Set the control word so that exceptions are masked for the |
duke@435 | 161 | // following code. |
duke@435 | 162 | __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); |
duke@435 | 163 | } |
duke@435 | 164 | |
duke@435 | 165 | // Save the FPU registers in de-opt-able form |
duke@435 | 166 | |
duke@435 | 167 | __ fstp_d(Address(rsp, st0_off*wordSize)); // st(0) |
duke@435 | 168 | __ fstp_d(Address(rsp, st1_off*wordSize)); // st(1) |
duke@435 | 169 | __ fstp_d(Address(rsp, st2_off*wordSize)); // st(2) |
duke@435 | 170 | __ fstp_d(Address(rsp, st3_off*wordSize)); // st(3) |
duke@435 | 171 | __ fstp_d(Address(rsp, st4_off*wordSize)); // st(4) |
duke@435 | 172 | __ fstp_d(Address(rsp, st5_off*wordSize)); // st(5) |
duke@435 | 173 | __ fstp_d(Address(rsp, st6_off*wordSize)); // st(6) |
duke@435 | 174 | __ fstp_d(Address(rsp, st7_off*wordSize)); // st(7) |
duke@435 | 175 | |
duke@435 | 176 | if( UseSSE == 1 ) { // Save the XMM state |
duke@435 | 177 | __ movflt(Address(rsp,xmm0_off*wordSize),xmm0); |
duke@435 | 178 | __ movflt(Address(rsp,xmm1_off*wordSize),xmm1); |
duke@435 | 179 | __ movflt(Address(rsp,xmm2_off*wordSize),xmm2); |
duke@435 | 180 | __ movflt(Address(rsp,xmm3_off*wordSize),xmm3); |
duke@435 | 181 | __ movflt(Address(rsp,xmm4_off*wordSize),xmm4); |
duke@435 | 182 | __ movflt(Address(rsp,xmm5_off*wordSize),xmm5); |
duke@435 | 183 | __ movflt(Address(rsp,xmm6_off*wordSize),xmm6); |
duke@435 | 184 | __ movflt(Address(rsp,xmm7_off*wordSize),xmm7); |
duke@435 | 185 | } else if( UseSSE >= 2 ) { |
duke@435 | 186 | __ movdbl(Address(rsp,xmm0_off*wordSize),xmm0); |
duke@435 | 187 | __ movdbl(Address(rsp,xmm1_off*wordSize),xmm1); |
duke@435 | 188 | __ movdbl(Address(rsp,xmm2_off*wordSize),xmm2); |
duke@435 | 189 | __ movdbl(Address(rsp,xmm3_off*wordSize),xmm3); |
duke@435 | 190 | __ movdbl(Address(rsp,xmm4_off*wordSize),xmm4); |
duke@435 | 191 | __ movdbl(Address(rsp,xmm5_off*wordSize),xmm5); |
duke@435 | 192 | __ movdbl(Address(rsp,xmm6_off*wordSize),xmm6); |
duke@435 | 193 | __ movdbl(Address(rsp,xmm7_off*wordSize),xmm7); |
duke@435 | 194 | } |
duke@435 | 195 | |
duke@435 | 196 | // Set an oopmap for the call site. This oopmap will map all |
duke@435 | 197 | // oop-registers and debug-info registers as callee-saved. This |
duke@435 | 198 | // will allow deoptimization at this safepoint to find all possible |
duke@435 | 199 | // debug-info recordings, as well as let GC find all oops. |
duke@435 | 200 | |
duke@435 | 201 | OopMapSet *oop_maps = new OopMapSet(); |
duke@435 | 202 | OopMap* map = new OopMap( frame_words, 0 ); |
duke@435 | 203 | |
duke@435 | 204 | #define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_words) |
duke@435 | 205 | |
duke@435 | 206 | map->set_callee_saved(STACK_OFFSET( rax_off), rax->as_VMReg()); |
duke@435 | 207 | map->set_callee_saved(STACK_OFFSET( rcx_off), rcx->as_VMReg()); |
duke@435 | 208 | map->set_callee_saved(STACK_OFFSET( rdx_off), rdx->as_VMReg()); |
duke@435 | 209 | map->set_callee_saved(STACK_OFFSET( rbx_off), rbx->as_VMReg()); |
duke@435 | 210 | // rbp, location is known implicitly, no oopMap |
duke@435 | 211 | map->set_callee_saved(STACK_OFFSET( rsi_off), rsi->as_VMReg()); |
duke@435 | 212 | map->set_callee_saved(STACK_OFFSET( rdi_off), rdi->as_VMReg()); |
duke@435 | 213 | map->set_callee_saved(STACK_OFFSET(st0_off), as_FloatRegister(0)->as_VMReg()); |
duke@435 | 214 | map->set_callee_saved(STACK_OFFSET(st1_off), as_FloatRegister(1)->as_VMReg()); |
duke@435 | 215 | map->set_callee_saved(STACK_OFFSET(st2_off), as_FloatRegister(2)->as_VMReg()); |
duke@435 | 216 | map->set_callee_saved(STACK_OFFSET(st3_off), as_FloatRegister(3)->as_VMReg()); |
duke@435 | 217 | map->set_callee_saved(STACK_OFFSET(st4_off), as_FloatRegister(4)->as_VMReg()); |
duke@435 | 218 | map->set_callee_saved(STACK_OFFSET(st5_off), as_FloatRegister(5)->as_VMReg()); |
duke@435 | 219 | map->set_callee_saved(STACK_OFFSET(st6_off), as_FloatRegister(6)->as_VMReg()); |
duke@435 | 220 | map->set_callee_saved(STACK_OFFSET(st7_off), as_FloatRegister(7)->as_VMReg()); |
duke@435 | 221 | map->set_callee_saved(STACK_OFFSET(xmm0_off), xmm0->as_VMReg()); |
duke@435 | 222 | map->set_callee_saved(STACK_OFFSET(xmm1_off), xmm1->as_VMReg()); |
duke@435 | 223 | map->set_callee_saved(STACK_OFFSET(xmm2_off), xmm2->as_VMReg()); |
duke@435 | 224 | map->set_callee_saved(STACK_OFFSET(xmm3_off), xmm3->as_VMReg()); |
duke@435 | 225 | map->set_callee_saved(STACK_OFFSET(xmm4_off), xmm4->as_VMReg()); |
duke@435 | 226 | map->set_callee_saved(STACK_OFFSET(xmm5_off), xmm5->as_VMReg()); |
duke@435 | 227 | map->set_callee_saved(STACK_OFFSET(xmm6_off), xmm6->as_VMReg()); |
duke@435 | 228 | map->set_callee_saved(STACK_OFFSET(xmm7_off), xmm7->as_VMReg()); |
duke@435 | 229 | // %%% This is really a waste but we'll keep things as they were for now |
duke@435 | 230 | if (true) { |
duke@435 | 231 | #define NEXTREG(x) (x)->as_VMReg()->next() |
duke@435 | 232 | map->set_callee_saved(STACK_OFFSET(st0H_off), NEXTREG(as_FloatRegister(0))); |
duke@435 | 233 | map->set_callee_saved(STACK_OFFSET(st1H_off), NEXTREG(as_FloatRegister(1))); |
duke@435 | 234 | map->set_callee_saved(STACK_OFFSET(st2H_off), NEXTREG(as_FloatRegister(2))); |
duke@435 | 235 | map->set_callee_saved(STACK_OFFSET(st3H_off), NEXTREG(as_FloatRegister(3))); |
duke@435 | 236 | map->set_callee_saved(STACK_OFFSET(st4H_off), NEXTREG(as_FloatRegister(4))); |
duke@435 | 237 | map->set_callee_saved(STACK_OFFSET(st5H_off), NEXTREG(as_FloatRegister(5))); |
duke@435 | 238 | map->set_callee_saved(STACK_OFFSET(st6H_off), NEXTREG(as_FloatRegister(6))); |
duke@435 | 239 | map->set_callee_saved(STACK_OFFSET(st7H_off), NEXTREG(as_FloatRegister(7))); |
duke@435 | 240 | map->set_callee_saved(STACK_OFFSET(xmm0H_off), NEXTREG(xmm0)); |
duke@435 | 241 | map->set_callee_saved(STACK_OFFSET(xmm1H_off), NEXTREG(xmm1)); |
duke@435 | 242 | map->set_callee_saved(STACK_OFFSET(xmm2H_off), NEXTREG(xmm2)); |
duke@435 | 243 | map->set_callee_saved(STACK_OFFSET(xmm3H_off), NEXTREG(xmm3)); |
duke@435 | 244 | map->set_callee_saved(STACK_OFFSET(xmm4H_off), NEXTREG(xmm4)); |
duke@435 | 245 | map->set_callee_saved(STACK_OFFSET(xmm5H_off), NEXTREG(xmm5)); |
duke@435 | 246 | map->set_callee_saved(STACK_OFFSET(xmm6H_off), NEXTREG(xmm6)); |
duke@435 | 247 | map->set_callee_saved(STACK_OFFSET(xmm7H_off), NEXTREG(xmm7)); |
duke@435 | 248 | #undef NEXTREG |
duke@435 | 249 | #undef STACK_OFFSET |
duke@435 | 250 | } |
duke@435 | 251 | |
duke@435 | 252 | return map; |
duke@435 | 253 | |
duke@435 | 254 | } |
duke@435 | 255 | |
duke@435 | 256 | void RegisterSaver::restore_live_registers(MacroAssembler* masm) { |
duke@435 | 257 | |
duke@435 | 258 | // Recover XMM & FPU state |
duke@435 | 259 | if( UseSSE == 1 ) { |
duke@435 | 260 | __ movflt(xmm0,Address(rsp,xmm0_off*wordSize)); |
duke@435 | 261 | __ movflt(xmm1,Address(rsp,xmm1_off*wordSize)); |
duke@435 | 262 | __ movflt(xmm2,Address(rsp,xmm2_off*wordSize)); |
duke@435 | 263 | __ movflt(xmm3,Address(rsp,xmm3_off*wordSize)); |
duke@435 | 264 | __ movflt(xmm4,Address(rsp,xmm4_off*wordSize)); |
duke@435 | 265 | __ movflt(xmm5,Address(rsp,xmm5_off*wordSize)); |
duke@435 | 266 | __ movflt(xmm6,Address(rsp,xmm6_off*wordSize)); |
duke@435 | 267 | __ movflt(xmm7,Address(rsp,xmm7_off*wordSize)); |
duke@435 | 268 | } else if( UseSSE >= 2 ) { |
duke@435 | 269 | __ movdbl(xmm0,Address(rsp,xmm0_off*wordSize)); |
duke@435 | 270 | __ movdbl(xmm1,Address(rsp,xmm1_off*wordSize)); |
duke@435 | 271 | __ movdbl(xmm2,Address(rsp,xmm2_off*wordSize)); |
duke@435 | 272 | __ movdbl(xmm3,Address(rsp,xmm3_off*wordSize)); |
duke@435 | 273 | __ movdbl(xmm4,Address(rsp,xmm4_off*wordSize)); |
duke@435 | 274 | __ movdbl(xmm5,Address(rsp,xmm5_off*wordSize)); |
duke@435 | 275 | __ movdbl(xmm6,Address(rsp,xmm6_off*wordSize)); |
duke@435 | 276 | __ movdbl(xmm7,Address(rsp,xmm7_off*wordSize)); |
duke@435 | 277 | } |
duke@435 | 278 | __ pop_FPU_state(); |
never@739 | 279 | __ addptr(rsp, FPU_regs_live*sizeof(jdouble)); // Pop FPU registers |
never@739 | 280 | |
never@739 | 281 | __ popf(); |
never@739 | 282 | __ popa(); |
duke@435 | 283 | // Get the rbp, described implicitly by the frame sender code (no oopMap) |
never@739 | 284 | __ pop(rbp); |
duke@435 | 285 | |
duke@435 | 286 | } |
duke@435 | 287 | |
duke@435 | 288 | void RegisterSaver::restore_result_registers(MacroAssembler* masm) { |
duke@435 | 289 | |
duke@435 | 290 | // Just restore result register. Only used by deoptimization. By |
duke@435 | 291 | // now any callee save register that needs to be restore to a c2 |
duke@435 | 292 | // caller of the deoptee has been extracted into the vframeArray |
duke@435 | 293 | // and will be stuffed into the c2i adapter we create for later |
duke@435 | 294 | // restoration so only result registers need to be restored here. |
duke@435 | 295 | // |
duke@435 | 296 | |
duke@435 | 297 | __ frstor(Address(rsp, 0)); // Restore fpu state |
duke@435 | 298 | |
duke@435 | 299 | // Recover XMM & FPU state |
duke@435 | 300 | if( UseSSE == 1 ) { |
duke@435 | 301 | __ movflt(xmm0, Address(rsp, xmm0_off*wordSize)); |
duke@435 | 302 | } else if( UseSSE >= 2 ) { |
duke@435 | 303 | __ movdbl(xmm0, Address(rsp, xmm0_off*wordSize)); |
duke@435 | 304 | } |
never@739 | 305 | __ movptr(rax, Address(rsp, rax_off*wordSize)); |
never@739 | 306 | __ movptr(rdx, Address(rsp, rdx_off*wordSize)); |
duke@435 | 307 | // Pop all of the register save are off the stack except the return address |
never@739 | 308 | __ addptr(rsp, return_off * wordSize); |
duke@435 | 309 | } |
duke@435 | 310 | |
duke@435 | 311 | // The java_calling_convention describes stack locations as ideal slots on |
duke@435 | 312 | // a frame with no abi restrictions. Since we must observe abi restrictions |
duke@435 | 313 | // (like the placement of the register window) the slots must be biased by |
duke@435 | 314 | // the following value. |
duke@435 | 315 | static int reg2offset_in(VMReg r) { |
duke@435 | 316 | // Account for saved rbp, and return address |
duke@435 | 317 | // This should really be in_preserve_stack_slots |
duke@435 | 318 | return (r->reg2stack() + 2) * VMRegImpl::stack_slot_size; |
duke@435 | 319 | } |
duke@435 | 320 | |
duke@435 | 321 | static int reg2offset_out(VMReg r) { |
duke@435 | 322 | return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size; |
duke@435 | 323 | } |
duke@435 | 324 | |
duke@435 | 325 | // --------------------------------------------------------------------------- |
duke@435 | 326 | // Read the array of BasicTypes from a signature, and compute where the |
duke@435 | 327 | // arguments should go. Values in the VMRegPair regs array refer to 4-byte |
duke@435 | 328 | // quantities. Values less than SharedInfo::stack0 are registers, those above |
duke@435 | 329 | // refer to 4-byte stack slots. All stack slots are based off of the stack pointer |
duke@435 | 330 | // as framesizes are fixed. |
duke@435 | 331 | // VMRegImpl::stack0 refers to the first slot 0(sp). |
duke@435 | 332 | // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher. Register |
duke@435 | 333 | // up to RegisterImpl::number_of_registers) are the 32-bit |
duke@435 | 334 | // integer registers. |
duke@435 | 335 | |
duke@435 | 336 | // Pass first two oop/int args in registers ECX and EDX. |
duke@435 | 337 | // Pass first two float/double args in registers XMM0 and XMM1. |
duke@435 | 338 | // Doubles have precedence, so if you pass a mix of floats and doubles |
duke@435 | 339 | // the doubles will grab the registers before the floats will. |
duke@435 | 340 | |
duke@435 | 341 | // Note: the INPUTS in sig_bt are in units of Java argument words, which are |
duke@435 | 342 | // either 32-bit or 64-bit depending on the build. The OUTPUTS are in 32-bit |
duke@435 | 343 | // units regardless of build. Of course for i486 there is no 64 bit build |
duke@435 | 344 | |
duke@435 | 345 | |
duke@435 | 346 | // --------------------------------------------------------------------------- |
duke@435 | 347 | // The compiled Java calling convention. |
duke@435 | 348 | // Pass first two oop/int args in registers ECX and EDX. |
duke@435 | 349 | // Pass first two float/double args in registers XMM0 and XMM1. |
duke@435 | 350 | // Doubles have precedence, so if you pass a mix of floats and doubles |
duke@435 | 351 | // the doubles will grab the registers before the floats will. |
duke@435 | 352 | int SharedRuntime::java_calling_convention(const BasicType *sig_bt, |
duke@435 | 353 | VMRegPair *regs, |
duke@435 | 354 | int total_args_passed, |
duke@435 | 355 | int is_outgoing) { |
duke@435 | 356 | uint stack = 0; // Starting stack position for args on stack |
duke@435 | 357 | |
duke@435 | 358 | |
duke@435 | 359 | // Pass first two oop/int args in registers ECX and EDX. |
duke@435 | 360 | uint reg_arg0 = 9999; |
duke@435 | 361 | uint reg_arg1 = 9999; |
duke@435 | 362 | |
duke@435 | 363 | // Pass first two float/double args in registers XMM0 and XMM1. |
duke@435 | 364 | // Doubles have precedence, so if you pass a mix of floats and doubles |
duke@435 | 365 | // the doubles will grab the registers before the floats will. |
duke@435 | 366 | // CNC - TURNED OFF FOR non-SSE. |
duke@435 | 367 | // On Intel we have to round all doubles (and most floats) at |
duke@435 | 368 | // call sites by storing to the stack in any case. |
duke@435 | 369 | // UseSSE=0 ==> Don't Use ==> 9999+0 |
duke@435 | 370 | // UseSSE=1 ==> Floats only ==> 9999+1 |
duke@435 | 371 | // UseSSE>=2 ==> Floats or doubles ==> 9999+2 |
duke@435 | 372 | enum { fltarg_dontuse = 9999+0, fltarg_float_only = 9999+1, fltarg_flt_dbl = 9999+2 }; |
duke@435 | 373 | uint fargs = (UseSSE>=2) ? 2 : UseSSE; |
duke@435 | 374 | uint freg_arg0 = 9999+fargs; |
duke@435 | 375 | uint freg_arg1 = 9999+fargs; |
duke@435 | 376 | |
duke@435 | 377 | // Pass doubles & longs aligned on the stack. First count stack slots for doubles |
duke@435 | 378 | int i; |
duke@435 | 379 | for( i = 0; i < total_args_passed; i++) { |
duke@435 | 380 | if( sig_bt[i] == T_DOUBLE ) { |
duke@435 | 381 | // first 2 doubles go in registers |
duke@435 | 382 | if( freg_arg0 == fltarg_flt_dbl ) freg_arg0 = i; |
duke@435 | 383 | else if( freg_arg1 == fltarg_flt_dbl ) freg_arg1 = i; |
duke@435 | 384 | else // Else double is passed low on the stack to be aligned. |
duke@435 | 385 | stack += 2; |
duke@435 | 386 | } else if( sig_bt[i] == T_LONG ) { |
duke@435 | 387 | stack += 2; |
duke@435 | 388 | } |
duke@435 | 389 | } |
duke@435 | 390 | int dstack = 0; // Separate counter for placing doubles |
duke@435 | 391 | |
duke@435 | 392 | // Now pick where all else goes. |
duke@435 | 393 | for( i = 0; i < total_args_passed; i++) { |
duke@435 | 394 | // From the type and the argument number (count) compute the location |
duke@435 | 395 | switch( sig_bt[i] ) { |
duke@435 | 396 | case T_SHORT: |
duke@435 | 397 | case T_CHAR: |
duke@435 | 398 | case T_BYTE: |
duke@435 | 399 | case T_BOOLEAN: |
duke@435 | 400 | case T_INT: |
duke@435 | 401 | case T_ARRAY: |
duke@435 | 402 | case T_OBJECT: |
duke@435 | 403 | case T_ADDRESS: |
duke@435 | 404 | if( reg_arg0 == 9999 ) { |
duke@435 | 405 | reg_arg0 = i; |
duke@435 | 406 | regs[i].set1(rcx->as_VMReg()); |
duke@435 | 407 | } else if( reg_arg1 == 9999 ) { |
duke@435 | 408 | reg_arg1 = i; |
duke@435 | 409 | regs[i].set1(rdx->as_VMReg()); |
duke@435 | 410 | } else { |
duke@435 | 411 | regs[i].set1(VMRegImpl::stack2reg(stack++)); |
duke@435 | 412 | } |
duke@435 | 413 | break; |
duke@435 | 414 | case T_FLOAT: |
duke@435 | 415 | if( freg_arg0 == fltarg_flt_dbl || freg_arg0 == fltarg_float_only ) { |
duke@435 | 416 | freg_arg0 = i; |
duke@435 | 417 | regs[i].set1(xmm0->as_VMReg()); |
duke@435 | 418 | } else if( freg_arg1 == fltarg_flt_dbl || freg_arg1 == fltarg_float_only ) { |
duke@435 | 419 | freg_arg1 = i; |
duke@435 | 420 | regs[i].set1(xmm1->as_VMReg()); |
duke@435 | 421 | } else { |
duke@435 | 422 | regs[i].set1(VMRegImpl::stack2reg(stack++)); |
duke@435 | 423 | } |
duke@435 | 424 | break; |
duke@435 | 425 | case T_LONG: |
duke@435 | 426 | assert(sig_bt[i+1] == T_VOID, "missing Half" ); |
duke@435 | 427 | regs[i].set2(VMRegImpl::stack2reg(dstack)); |
duke@435 | 428 | dstack += 2; |
duke@435 | 429 | break; |
duke@435 | 430 | case T_DOUBLE: |
duke@435 | 431 | assert(sig_bt[i+1] == T_VOID, "missing Half" ); |
duke@435 | 432 | if( freg_arg0 == (uint)i ) { |
duke@435 | 433 | regs[i].set2(xmm0->as_VMReg()); |
duke@435 | 434 | } else if( freg_arg1 == (uint)i ) { |
duke@435 | 435 | regs[i].set2(xmm1->as_VMReg()); |
duke@435 | 436 | } else { |
duke@435 | 437 | regs[i].set2(VMRegImpl::stack2reg(dstack)); |
duke@435 | 438 | dstack += 2; |
duke@435 | 439 | } |
duke@435 | 440 | break; |
duke@435 | 441 | case T_VOID: regs[i].set_bad(); break; |
duke@435 | 442 | break; |
duke@435 | 443 | default: |
duke@435 | 444 | ShouldNotReachHere(); |
duke@435 | 445 | break; |
duke@435 | 446 | } |
duke@435 | 447 | } |
duke@435 | 448 | |
duke@435 | 449 | // return value can be odd number of VMRegImpl stack slots make multiple of 2 |
duke@435 | 450 | return round_to(stack, 2); |
duke@435 | 451 | } |
duke@435 | 452 | |
duke@435 | 453 | // Patch the callers callsite with entry to compiled code if it exists. |
duke@435 | 454 | static void patch_callers_callsite(MacroAssembler *masm) { |
duke@435 | 455 | Label L; |
duke@435 | 456 | __ verify_oop(rbx); |
never@739 | 457 | __ cmpptr(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int32_t)NULL_WORD); |
duke@435 | 458 | __ jcc(Assembler::equal, L); |
duke@435 | 459 | // Schedule the branch target address early. |
duke@435 | 460 | // Call into the VM to patch the caller, then jump to compiled callee |
duke@435 | 461 | // rax, isn't live so capture return address while we easily can |
never@739 | 462 | __ movptr(rax, Address(rsp, 0)); |
never@739 | 463 | __ pusha(); |
never@739 | 464 | __ pushf(); |
duke@435 | 465 | |
duke@435 | 466 | if (UseSSE == 1) { |
never@739 | 467 | __ subptr(rsp, 2*wordSize); |
duke@435 | 468 | __ movflt(Address(rsp, 0), xmm0); |
duke@435 | 469 | __ movflt(Address(rsp, wordSize), xmm1); |
duke@435 | 470 | } |
duke@435 | 471 | if (UseSSE >= 2) { |
never@739 | 472 | __ subptr(rsp, 4*wordSize); |
duke@435 | 473 | __ movdbl(Address(rsp, 0), xmm0); |
duke@435 | 474 | __ movdbl(Address(rsp, 2*wordSize), xmm1); |
duke@435 | 475 | } |
duke@435 | 476 | #ifdef COMPILER2 |
duke@435 | 477 | // C2 may leave the stack dirty if not in SSE2+ mode |
duke@435 | 478 | if (UseSSE >= 2) { |
duke@435 | 479 | __ verify_FPU(0, "c2i transition should have clean FPU stack"); |
duke@435 | 480 | } else { |
duke@435 | 481 | __ empty_FPU_stack(); |
duke@435 | 482 | } |
duke@435 | 483 | #endif /* COMPILER2 */ |
duke@435 | 484 | |
duke@435 | 485 | // VM needs caller's callsite |
never@739 | 486 | __ push(rax); |
duke@435 | 487 | // VM needs target method |
never@739 | 488 | __ push(rbx); |
duke@435 | 489 | __ verify_oop(rbx); |
duke@435 | 490 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite))); |
never@739 | 491 | __ addptr(rsp, 2*wordSize); |
duke@435 | 492 | |
duke@435 | 493 | if (UseSSE == 1) { |
duke@435 | 494 | __ movflt(xmm0, Address(rsp, 0)); |
duke@435 | 495 | __ movflt(xmm1, Address(rsp, wordSize)); |
never@739 | 496 | __ addptr(rsp, 2*wordSize); |
duke@435 | 497 | } |
duke@435 | 498 | if (UseSSE >= 2) { |
duke@435 | 499 | __ movdbl(xmm0, Address(rsp, 0)); |
duke@435 | 500 | __ movdbl(xmm1, Address(rsp, 2*wordSize)); |
never@739 | 501 | __ addptr(rsp, 4*wordSize); |
duke@435 | 502 | } |
duke@435 | 503 | |
never@739 | 504 | __ popf(); |
never@739 | 505 | __ popa(); |
duke@435 | 506 | __ bind(L); |
duke@435 | 507 | } |
duke@435 | 508 | |
duke@435 | 509 | |
duke@435 | 510 | static void move_c2i_double(MacroAssembler *masm, XMMRegister r, int st_off) { |
twisti@1861 | 511 | int next_off = st_off - Interpreter::stackElementSize; |
twisti@1861 | 512 | __ movdbl(Address(rsp, next_off), r); |
duke@435 | 513 | } |
duke@435 | 514 | |
duke@435 | 515 | static void gen_c2i_adapter(MacroAssembler *masm, |
duke@435 | 516 | int total_args_passed, |
duke@435 | 517 | int comp_args_on_stack, |
duke@435 | 518 | const BasicType *sig_bt, |
duke@435 | 519 | const VMRegPair *regs, |
duke@435 | 520 | Label& skip_fixup) { |
duke@435 | 521 | // Before we get into the guts of the C2I adapter, see if we should be here |
duke@435 | 522 | // at all. We've come from compiled code and are attempting to jump to the |
duke@435 | 523 | // interpreter, which means the caller made a static call to get here |
duke@435 | 524 | // (vcalls always get a compiled target if there is one). Check for a |
duke@435 | 525 | // compiled target. If there is one, we need to patch the caller's call. |
duke@435 | 526 | patch_callers_callsite(masm); |
duke@435 | 527 | |
duke@435 | 528 | __ bind(skip_fixup); |
duke@435 | 529 | |
duke@435 | 530 | #ifdef COMPILER2 |
duke@435 | 531 | // C2 may leave the stack dirty if not in SSE2+ mode |
duke@435 | 532 | if (UseSSE >= 2) { |
duke@435 | 533 | __ verify_FPU(0, "c2i transition should have clean FPU stack"); |
duke@435 | 534 | } else { |
duke@435 | 535 | __ empty_FPU_stack(); |
duke@435 | 536 | } |
duke@435 | 537 | #endif /* COMPILER2 */ |
duke@435 | 538 | |
duke@435 | 539 | // Since all args are passed on the stack, total_args_passed * interpreter_ |
duke@435 | 540 | // stack_element_size is the |
duke@435 | 541 | // space we need. |
twisti@1861 | 542 | int extraspace = total_args_passed * Interpreter::stackElementSize; |
duke@435 | 543 | |
duke@435 | 544 | // Get return address |
never@739 | 545 | __ pop(rax); |
duke@435 | 546 | |
duke@435 | 547 | // set senderSP value |
never@739 | 548 | __ movptr(rsi, rsp); |
never@739 | 549 | |
never@739 | 550 | __ subptr(rsp, extraspace); |
duke@435 | 551 | |
duke@435 | 552 | // Now write the args into the outgoing interpreter space |
duke@435 | 553 | for (int i = 0; i < total_args_passed; i++) { |
duke@435 | 554 | if (sig_bt[i] == T_VOID) { |
duke@435 | 555 | assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); |
duke@435 | 556 | continue; |
duke@435 | 557 | } |
duke@435 | 558 | |
duke@435 | 559 | // st_off points to lowest address on stack. |
twisti@1861 | 560 | int st_off = ((total_args_passed - 1) - i) * Interpreter::stackElementSize; |
twisti@1861 | 561 | int next_off = st_off - Interpreter::stackElementSize; |
never@739 | 562 | |
duke@435 | 563 | // Say 4 args: |
duke@435 | 564 | // i st_off |
duke@435 | 565 | // 0 12 T_LONG |
duke@435 | 566 | // 1 8 T_VOID |
duke@435 | 567 | // 2 4 T_OBJECT |
duke@435 | 568 | // 3 0 T_BOOL |
duke@435 | 569 | VMReg r_1 = regs[i].first(); |
duke@435 | 570 | VMReg r_2 = regs[i].second(); |
duke@435 | 571 | if (!r_1->is_valid()) { |
duke@435 | 572 | assert(!r_2->is_valid(), ""); |
duke@435 | 573 | continue; |
duke@435 | 574 | } |
duke@435 | 575 | |
duke@435 | 576 | if (r_1->is_stack()) { |
duke@435 | 577 | // memory to memory use fpu stack top |
duke@435 | 578 | int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace; |
duke@435 | 579 | |
duke@435 | 580 | if (!r_2->is_valid()) { |
duke@435 | 581 | __ movl(rdi, Address(rsp, ld_off)); |
never@739 | 582 | __ movptr(Address(rsp, st_off), rdi); |
duke@435 | 583 | } else { |
duke@435 | 584 | |
duke@435 | 585 | // ld_off == LSW, ld_off+VMRegImpl::stack_slot_size == MSW |
duke@435 | 586 | // st_off == MSW, st_off-wordSize == LSW |
duke@435 | 587 | |
never@739 | 588 | __ movptr(rdi, Address(rsp, ld_off)); |
never@739 | 589 | __ movptr(Address(rsp, next_off), rdi); |
never@739 | 590 | #ifndef _LP64 |
never@739 | 591 | __ movptr(rdi, Address(rsp, ld_off + wordSize)); |
never@739 | 592 | __ movptr(Address(rsp, st_off), rdi); |
never@739 | 593 | #else |
never@739 | 594 | #ifdef ASSERT |
never@739 | 595 | // Overwrite the unused slot with known junk |
never@739 | 596 | __ mov64(rax, CONST64(0xdeadffffdeadaaaa)); |
never@739 | 597 | __ movptr(Address(rsp, st_off), rax); |
never@739 | 598 | #endif /* ASSERT */ |
never@739 | 599 | #endif // _LP64 |
duke@435 | 600 | } |
duke@435 | 601 | } else if (r_1->is_Register()) { |
duke@435 | 602 | Register r = r_1->as_Register(); |
duke@435 | 603 | if (!r_2->is_valid()) { |
duke@435 | 604 | __ movl(Address(rsp, st_off), r); |
duke@435 | 605 | } else { |
duke@435 | 606 | // long/double in gpr |
never@739 | 607 | NOT_LP64(ShouldNotReachHere()); |
never@739 | 608 | // Two VMRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG |
never@739 | 609 | // T_DOUBLE and T_LONG use two slots in the interpreter |
never@739 | 610 | if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) { |
never@739 | 611 | // long/double in gpr |
never@739 | 612 | #ifdef ASSERT |
never@739 | 613 | // Overwrite the unused slot with known junk |
never@739 | 614 | LP64_ONLY(__ mov64(rax, CONST64(0xdeadffffdeadaaab))); |
never@739 | 615 | __ movptr(Address(rsp, st_off), rax); |
never@739 | 616 | #endif /* ASSERT */ |
never@739 | 617 | __ movptr(Address(rsp, next_off), r); |
never@739 | 618 | } else { |
never@739 | 619 | __ movptr(Address(rsp, st_off), r); |
never@739 | 620 | } |
duke@435 | 621 | } |
duke@435 | 622 | } else { |
duke@435 | 623 | assert(r_1->is_XMMRegister(), ""); |
duke@435 | 624 | if (!r_2->is_valid()) { |
duke@435 | 625 | __ movflt(Address(rsp, st_off), r_1->as_XMMRegister()); |
duke@435 | 626 | } else { |
duke@435 | 627 | assert(sig_bt[i] == T_DOUBLE || sig_bt[i] == T_LONG, "wrong type"); |
duke@435 | 628 | move_c2i_double(masm, r_1->as_XMMRegister(), st_off); |
duke@435 | 629 | } |
duke@435 | 630 | } |
duke@435 | 631 | } |
duke@435 | 632 | |
duke@435 | 633 | // Schedule the branch target address early. |
never@739 | 634 | __ movptr(rcx, Address(rbx, in_bytes(methodOopDesc::interpreter_entry_offset()))); |
duke@435 | 635 | // And repush original return address |
never@739 | 636 | __ push(rax); |
duke@435 | 637 | __ jmp(rcx); |
duke@435 | 638 | } |
duke@435 | 639 | |
duke@435 | 640 | |
duke@435 | 641 | static void move_i2c_double(MacroAssembler *masm, XMMRegister r, Register saved_sp, int ld_off) { |
twisti@1861 | 642 | int next_val_off = ld_off - Interpreter::stackElementSize; |
twisti@1861 | 643 | __ movdbl(r, Address(saved_sp, next_val_off)); |
duke@435 | 644 | } |
duke@435 | 645 | |
duke@435 | 646 | static void gen_i2c_adapter(MacroAssembler *masm, |
duke@435 | 647 | int total_args_passed, |
duke@435 | 648 | int comp_args_on_stack, |
duke@435 | 649 | const BasicType *sig_bt, |
duke@435 | 650 | const VMRegPair *regs) { |
duke@435 | 651 | |
duke@435 | 652 | // Note: rsi contains the senderSP on entry. We must preserve it since |
duke@435 | 653 | // we may do a i2c -> c2i transition if we lose a race where compiled |
duke@435 | 654 | // code goes non-entrant while we get args ready. |
duke@435 | 655 | |
duke@435 | 656 | // Pick up the return address |
never@739 | 657 | __ movptr(rax, Address(rsp, 0)); |
duke@435 | 658 | |
duke@435 | 659 | // Must preserve original SP for loading incoming arguments because |
duke@435 | 660 | // we need to align the outgoing SP for compiled code. |
never@739 | 661 | __ movptr(rdi, rsp); |
duke@435 | 662 | |
duke@435 | 663 | // Cut-out for having no stack args. Since up to 2 int/oop args are passed |
duke@435 | 664 | // in registers, we will occasionally have no stack args. |
duke@435 | 665 | int comp_words_on_stack = 0; |
duke@435 | 666 | if (comp_args_on_stack) { |
duke@435 | 667 | // Sig words on the stack are greater-than VMRegImpl::stack0. Those in |
duke@435 | 668 | // registers are below. By subtracting stack0, we either get a negative |
duke@435 | 669 | // number (all values in registers) or the maximum stack slot accessed. |
duke@435 | 670 | // int comp_args_on_stack = VMRegImpl::reg2stack(max_arg); |
duke@435 | 671 | // Convert 4-byte stack slots to words. |
duke@435 | 672 | comp_words_on_stack = round_to(comp_args_on_stack*4, wordSize)>>LogBytesPerWord; |
duke@435 | 673 | // Round up to miminum stack alignment, in wordSize |
duke@435 | 674 | comp_words_on_stack = round_to(comp_words_on_stack, 2); |
never@739 | 675 | __ subptr(rsp, comp_words_on_stack * wordSize); |
duke@435 | 676 | } |
duke@435 | 677 | |
duke@435 | 678 | // Align the outgoing SP |
never@739 | 679 | __ andptr(rsp, -(StackAlignmentInBytes)); |
duke@435 | 680 | |
duke@435 | 681 | // push the return address on the stack (note that pushing, rather |
duke@435 | 682 | // than storing it, yields the correct frame alignment for the callee) |
never@739 | 683 | __ push(rax); |
duke@435 | 684 | |
duke@435 | 685 | // Put saved SP in another register |
duke@435 | 686 | const Register saved_sp = rax; |
never@739 | 687 | __ movptr(saved_sp, rdi); |
duke@435 | 688 | |
duke@435 | 689 | |
duke@435 | 690 | // Will jump to the compiled code just as if compiled code was doing it. |
duke@435 | 691 | // Pre-load the register-jump target early, to schedule it better. |
never@739 | 692 | __ movptr(rdi, Address(rbx, in_bytes(methodOopDesc::from_compiled_offset()))); |
duke@435 | 693 | |
duke@435 | 694 | // Now generate the shuffle code. Pick up all register args and move the |
duke@435 | 695 | // rest through the floating point stack top. |
duke@435 | 696 | for (int i = 0; i < total_args_passed; i++) { |
duke@435 | 697 | if (sig_bt[i] == T_VOID) { |
duke@435 | 698 | // Longs and doubles are passed in native word order, but misaligned |
duke@435 | 699 | // in the 32-bit build. |
duke@435 | 700 | assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half"); |
duke@435 | 701 | continue; |
duke@435 | 702 | } |
duke@435 | 703 | |
duke@435 | 704 | // Pick up 0, 1 or 2 words from SP+offset. |
duke@435 | 705 | |
duke@435 | 706 | assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(), |
duke@435 | 707 | "scrambled load targets?"); |
duke@435 | 708 | // Load in argument order going down. |
twisti@1861 | 709 | int ld_off = (total_args_passed - i) * Interpreter::stackElementSize; |
duke@435 | 710 | // Point to interpreter value (vs. tag) |
twisti@1861 | 711 | int next_off = ld_off - Interpreter::stackElementSize; |
duke@435 | 712 | // |
duke@435 | 713 | // |
duke@435 | 714 | // |
duke@435 | 715 | VMReg r_1 = regs[i].first(); |
duke@435 | 716 | VMReg r_2 = regs[i].second(); |
duke@435 | 717 | if (!r_1->is_valid()) { |
duke@435 | 718 | assert(!r_2->is_valid(), ""); |
duke@435 | 719 | continue; |
duke@435 | 720 | } |
duke@435 | 721 | if (r_1->is_stack()) { |
duke@435 | 722 | // Convert stack slot to an SP offset (+ wordSize to account for return address ) |
duke@435 | 723 | int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize; |
duke@435 | 724 | |
duke@435 | 725 | // We can use rsi as a temp here because compiled code doesn't need rsi as an input |
duke@435 | 726 | // and if we end up going thru a c2i because of a miss a reasonable value of rsi |
duke@435 | 727 | // we be generated. |
duke@435 | 728 | if (!r_2->is_valid()) { |
duke@435 | 729 | // __ fld_s(Address(saved_sp, ld_off)); |
duke@435 | 730 | // __ fstp_s(Address(rsp, st_off)); |
duke@435 | 731 | __ movl(rsi, Address(saved_sp, ld_off)); |
never@739 | 732 | __ movptr(Address(rsp, st_off), rsi); |
duke@435 | 733 | } else { |
duke@435 | 734 | // Interpreter local[n] == MSW, local[n+1] == LSW however locals |
duke@435 | 735 | // are accessed as negative so LSW is at LOW address |
duke@435 | 736 | |
duke@435 | 737 | // ld_off is MSW so get LSW |
duke@435 | 738 | // st_off is LSW (i.e. reg.first()) |
duke@435 | 739 | // __ fld_d(Address(saved_sp, next_off)); |
duke@435 | 740 | // __ fstp_d(Address(rsp, st_off)); |
never@739 | 741 | // |
never@739 | 742 | // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE |
never@739 | 743 | // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case |
never@739 | 744 | // So we must adjust where to pick up the data to match the interpreter. |
never@739 | 745 | // |
never@739 | 746 | // Interpreter local[n] == MSW, local[n+1] == LSW however locals |
never@739 | 747 | // are accessed as negative so LSW is at LOW address |
never@739 | 748 | |
never@739 | 749 | // ld_off is MSW so get LSW |
never@739 | 750 | const int offset = (NOT_LP64(true ||) sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)? |
never@739 | 751 | next_off : ld_off; |
never@739 | 752 | __ movptr(rsi, Address(saved_sp, offset)); |
never@739 | 753 | __ movptr(Address(rsp, st_off), rsi); |
never@739 | 754 | #ifndef _LP64 |
never@739 | 755 | __ movptr(rsi, Address(saved_sp, ld_off)); |
never@739 | 756 | __ movptr(Address(rsp, st_off + wordSize), rsi); |
never@739 | 757 | #endif // _LP64 |
duke@435 | 758 | } |
duke@435 | 759 | } else if (r_1->is_Register()) { // Register argument |
duke@435 | 760 | Register r = r_1->as_Register(); |
duke@435 | 761 | assert(r != rax, "must be different"); |
duke@435 | 762 | if (r_2->is_valid()) { |
never@739 | 763 | // |
never@739 | 764 | // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE |
never@739 | 765 | // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case |
never@739 | 766 | // So we must adjust where to pick up the data to match the interpreter. |
never@739 | 767 | |
never@739 | 768 | const int offset = (NOT_LP64(true ||) sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)? |
never@739 | 769 | next_off : ld_off; |
never@739 | 770 | |
never@739 | 771 | // this can be a misaligned move |
never@739 | 772 | __ movptr(r, Address(saved_sp, offset)); |
never@739 | 773 | #ifndef _LP64 |
duke@435 | 774 | assert(r_2->as_Register() != rax, "need another temporary register"); |
duke@435 | 775 | // Remember r_1 is low address (and LSB on x86) |
duke@435 | 776 | // So r_2 gets loaded from high address regardless of the platform |
never@739 | 777 | __ movptr(r_2->as_Register(), Address(saved_sp, ld_off)); |
never@739 | 778 | #endif // _LP64 |
duke@435 | 779 | } else { |
duke@435 | 780 | __ movl(r, Address(saved_sp, ld_off)); |
duke@435 | 781 | } |
duke@435 | 782 | } else { |
duke@435 | 783 | assert(r_1->is_XMMRegister(), ""); |
duke@435 | 784 | if (!r_2->is_valid()) { |
duke@435 | 785 | __ movflt(r_1->as_XMMRegister(), Address(saved_sp, ld_off)); |
duke@435 | 786 | } else { |
duke@435 | 787 | move_i2c_double(masm, r_1->as_XMMRegister(), saved_sp, ld_off); |
duke@435 | 788 | } |
duke@435 | 789 | } |
duke@435 | 790 | } |
duke@435 | 791 | |
duke@435 | 792 | // 6243940 We might end up in handle_wrong_method if |
duke@435 | 793 | // the callee is deoptimized as we race thru here. If that |
duke@435 | 794 | // happens we don't want to take a safepoint because the |
duke@435 | 795 | // caller frame will look interpreted and arguments are now |
duke@435 | 796 | // "compiled" so it is much better to make this transition |
duke@435 | 797 | // invisible to the stack walking code. Unfortunately if |
duke@435 | 798 | // we try and find the callee by normal means a safepoint |
duke@435 | 799 | // is possible. So we stash the desired callee in the thread |
duke@435 | 800 | // and the vm will find there should this case occur. |
duke@435 | 801 | |
duke@435 | 802 | __ get_thread(rax); |
never@739 | 803 | __ movptr(Address(rax, JavaThread::callee_target_offset()), rbx); |
duke@435 | 804 | |
duke@435 | 805 | // move methodOop to rax, in case we end up in an c2i adapter. |
duke@435 | 806 | // the c2i adapters expect methodOop in rax, (c2) because c2's |
duke@435 | 807 | // resolve stubs return the result (the method) in rax,. |
duke@435 | 808 | // I'd love to fix this. |
never@739 | 809 | __ mov(rax, rbx); |
duke@435 | 810 | |
duke@435 | 811 | __ jmp(rdi); |
duke@435 | 812 | } |
duke@435 | 813 | |
duke@435 | 814 | // --------------------------------------------------------------- |
duke@435 | 815 | AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm, |
duke@435 | 816 | int total_args_passed, |
duke@435 | 817 | int comp_args_on_stack, |
duke@435 | 818 | const BasicType *sig_bt, |
never@1622 | 819 | const VMRegPair *regs, |
never@1622 | 820 | AdapterFingerPrint* fingerprint) { |
duke@435 | 821 | address i2c_entry = __ pc(); |
duke@435 | 822 | |
duke@435 | 823 | gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs); |
duke@435 | 824 | |
duke@435 | 825 | // ------------------------------------------------------------------------- |
duke@435 | 826 | // Generate a C2I adapter. On entry we know rbx, holds the methodOop during calls |
duke@435 | 827 | // to the interpreter. The args start out packed in the compiled layout. They |
duke@435 | 828 | // need to be unpacked into the interpreter layout. This will almost always |
duke@435 | 829 | // require some stack space. We grow the current (compiled) stack, then repack |
duke@435 | 830 | // the args. We finally end in a jump to the generic interpreter entry point. |
duke@435 | 831 | // On exit from the interpreter, the interpreter will restore our SP (lest the |
duke@435 | 832 | // compiled code, which relys solely on SP and not EBP, get sick). |
duke@435 | 833 | |
duke@435 | 834 | address c2i_unverified_entry = __ pc(); |
duke@435 | 835 | Label skip_fixup; |
duke@435 | 836 | |
duke@435 | 837 | Register holder = rax; |
duke@435 | 838 | Register receiver = rcx; |
duke@435 | 839 | Register temp = rbx; |
duke@435 | 840 | |
duke@435 | 841 | { |
duke@435 | 842 | |
duke@435 | 843 | Label missed; |
duke@435 | 844 | |
duke@435 | 845 | __ verify_oop(holder); |
never@739 | 846 | __ movptr(temp, Address(receiver, oopDesc::klass_offset_in_bytes())); |
duke@435 | 847 | __ verify_oop(temp); |
duke@435 | 848 | |
never@739 | 849 | __ cmpptr(temp, Address(holder, compiledICHolderOopDesc::holder_klass_offset())); |
never@739 | 850 | __ movptr(rbx, Address(holder, compiledICHolderOopDesc::holder_method_offset())); |
duke@435 | 851 | __ jcc(Assembler::notEqual, missed); |
duke@435 | 852 | // Method might have been compiled since the call site was patched to |
duke@435 | 853 | // interpreted if that is the case treat it as a miss so we can get |
duke@435 | 854 | // the call site corrected. |
never@739 | 855 | __ cmpptr(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int32_t)NULL_WORD); |
duke@435 | 856 | __ jcc(Assembler::equal, skip_fixup); |
duke@435 | 857 | |
duke@435 | 858 | __ bind(missed); |
duke@435 | 859 | __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); |
duke@435 | 860 | } |
duke@435 | 861 | |
duke@435 | 862 | address c2i_entry = __ pc(); |
duke@435 | 863 | |
duke@435 | 864 | gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup); |
duke@435 | 865 | |
duke@435 | 866 | __ flush(); |
never@1622 | 867 | return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry); |
duke@435 | 868 | } |
duke@435 | 869 | |
duke@435 | 870 | int SharedRuntime::c_calling_convention(const BasicType *sig_bt, |
duke@435 | 871 | VMRegPair *regs, |
duke@435 | 872 | int total_args_passed) { |
duke@435 | 873 | // We return the amount of VMRegImpl stack slots we need to reserve for all |
duke@435 | 874 | // the arguments NOT counting out_preserve_stack_slots. |
duke@435 | 875 | |
duke@435 | 876 | uint stack = 0; // All arguments on stack |
duke@435 | 877 | |
duke@435 | 878 | for( int i = 0; i < total_args_passed; i++) { |
duke@435 | 879 | // From the type and the argument number (count) compute the location |
duke@435 | 880 | switch( sig_bt[i] ) { |
duke@435 | 881 | case T_BOOLEAN: |
duke@435 | 882 | case T_CHAR: |
duke@435 | 883 | case T_FLOAT: |
duke@435 | 884 | case T_BYTE: |
duke@435 | 885 | case T_SHORT: |
duke@435 | 886 | case T_INT: |
duke@435 | 887 | case T_OBJECT: |
duke@435 | 888 | case T_ARRAY: |
duke@435 | 889 | case T_ADDRESS: |
duke@435 | 890 | regs[i].set1(VMRegImpl::stack2reg(stack++)); |
duke@435 | 891 | break; |
duke@435 | 892 | case T_LONG: |
duke@435 | 893 | case T_DOUBLE: // The stack numbering is reversed from Java |
duke@435 | 894 | // Since C arguments do not get reversed, the ordering for |
duke@435 | 895 | // doubles on the stack must be opposite the Java convention |
duke@435 | 896 | assert(sig_bt[i+1] == T_VOID, "missing Half" ); |
duke@435 | 897 | regs[i].set2(VMRegImpl::stack2reg(stack)); |
duke@435 | 898 | stack += 2; |
duke@435 | 899 | break; |
duke@435 | 900 | case T_VOID: regs[i].set_bad(); break; |
duke@435 | 901 | default: |
duke@435 | 902 | ShouldNotReachHere(); |
duke@435 | 903 | break; |
duke@435 | 904 | } |
duke@435 | 905 | } |
duke@435 | 906 | return stack; |
duke@435 | 907 | } |
duke@435 | 908 | |
duke@435 | 909 | // A simple move of integer like type |
duke@435 | 910 | static void simple_move32(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
duke@435 | 911 | if (src.first()->is_stack()) { |
duke@435 | 912 | if (dst.first()->is_stack()) { |
duke@435 | 913 | // stack to stack |
duke@435 | 914 | // __ ld(FP, reg2offset(src.first()) + STACK_BIAS, L5); |
duke@435 | 915 | // __ st(L5, SP, reg2offset(dst.first()) + STACK_BIAS); |
never@739 | 916 | __ movl2ptr(rax, Address(rbp, reg2offset_in(src.first()))); |
never@739 | 917 | __ movptr(Address(rsp, reg2offset_out(dst.first())), rax); |
duke@435 | 918 | } else { |
duke@435 | 919 | // stack to reg |
never@739 | 920 | __ movl2ptr(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()))); |
duke@435 | 921 | } |
duke@435 | 922 | } else if (dst.first()->is_stack()) { |
duke@435 | 923 | // reg to stack |
never@739 | 924 | // no need to sign extend on 64bit |
never@739 | 925 | __ movptr(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register()); |
duke@435 | 926 | } else { |
never@739 | 927 | if (dst.first() != src.first()) { |
never@739 | 928 | __ mov(dst.first()->as_Register(), src.first()->as_Register()); |
never@739 | 929 | } |
duke@435 | 930 | } |
duke@435 | 931 | } |
duke@435 | 932 | |
duke@435 | 933 | // An oop arg. Must pass a handle not the oop itself |
duke@435 | 934 | static void object_move(MacroAssembler* masm, |
duke@435 | 935 | OopMap* map, |
duke@435 | 936 | int oop_handle_offset, |
duke@435 | 937 | int framesize_in_slots, |
duke@435 | 938 | VMRegPair src, |
duke@435 | 939 | VMRegPair dst, |
duke@435 | 940 | bool is_receiver, |
duke@435 | 941 | int* receiver_offset) { |
duke@435 | 942 | |
duke@435 | 943 | // Because of the calling conventions we know that src can be a |
duke@435 | 944 | // register or a stack location. dst can only be a stack location. |
duke@435 | 945 | |
duke@435 | 946 | assert(dst.first()->is_stack(), "must be stack"); |
duke@435 | 947 | // must pass a handle. First figure out the location we use as a handle |
duke@435 | 948 | |
duke@435 | 949 | if (src.first()->is_stack()) { |
duke@435 | 950 | // Oop is already on the stack as an argument |
duke@435 | 951 | Register rHandle = rax; |
duke@435 | 952 | Label nil; |
never@739 | 953 | __ xorptr(rHandle, rHandle); |
never@739 | 954 | __ cmpptr(Address(rbp, reg2offset_in(src.first())), (int32_t)NULL_WORD); |
duke@435 | 955 | __ jcc(Assembler::equal, nil); |
never@739 | 956 | __ lea(rHandle, Address(rbp, reg2offset_in(src.first()))); |
duke@435 | 957 | __ bind(nil); |
never@739 | 958 | __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle); |
duke@435 | 959 | |
duke@435 | 960 | int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots(); |
duke@435 | 961 | map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots)); |
duke@435 | 962 | if (is_receiver) { |
duke@435 | 963 | *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size; |
duke@435 | 964 | } |
duke@435 | 965 | } else { |
duke@435 | 966 | // Oop is in an a register we must store it to the space we reserve |
duke@435 | 967 | // on the stack for oop_handles |
duke@435 | 968 | const Register rOop = src.first()->as_Register(); |
duke@435 | 969 | const Register rHandle = rax; |
duke@435 | 970 | int oop_slot = (rOop == rcx ? 0 : 1) * VMRegImpl::slots_per_word + oop_handle_offset; |
duke@435 | 971 | int offset = oop_slot*VMRegImpl::stack_slot_size; |
duke@435 | 972 | Label skip; |
never@739 | 973 | __ movptr(Address(rsp, offset), rOop); |
duke@435 | 974 | map->set_oop(VMRegImpl::stack2reg(oop_slot)); |
never@739 | 975 | __ xorptr(rHandle, rHandle); |
never@739 | 976 | __ cmpptr(rOop, (int32_t)NULL_WORD); |
duke@435 | 977 | __ jcc(Assembler::equal, skip); |
never@739 | 978 | __ lea(rHandle, Address(rsp, offset)); |
duke@435 | 979 | __ bind(skip); |
duke@435 | 980 | // Store the handle parameter |
never@739 | 981 | __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle); |
duke@435 | 982 | if (is_receiver) { |
duke@435 | 983 | *receiver_offset = offset; |
duke@435 | 984 | } |
duke@435 | 985 | } |
duke@435 | 986 | } |
duke@435 | 987 | |
duke@435 | 988 | // A float arg may have to do float reg int reg conversion |
duke@435 | 989 | static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
duke@435 | 990 | assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move"); |
duke@435 | 991 | |
duke@435 | 992 | // Because of the calling convention we know that src is either a stack location |
duke@435 | 993 | // or an xmm register. dst can only be a stack location. |
duke@435 | 994 | |
duke@435 | 995 | assert(dst.first()->is_stack() && ( src.first()->is_stack() || src.first()->is_XMMRegister()), "bad parameters"); |
duke@435 | 996 | |
duke@435 | 997 | if (src.first()->is_stack()) { |
duke@435 | 998 | __ movl(rax, Address(rbp, reg2offset_in(src.first()))); |
never@739 | 999 | __ movptr(Address(rsp, reg2offset_out(dst.first())), rax); |
duke@435 | 1000 | } else { |
duke@435 | 1001 | // reg to stack |
duke@435 | 1002 | __ movflt(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister()); |
duke@435 | 1003 | } |
duke@435 | 1004 | } |
duke@435 | 1005 | |
duke@435 | 1006 | // A long move |
duke@435 | 1007 | static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
duke@435 | 1008 | |
duke@435 | 1009 | // The only legal possibility for a long_move VMRegPair is: |
duke@435 | 1010 | // 1: two stack slots (possibly unaligned) |
duke@435 | 1011 | // as neither the java or C calling convention will use registers |
duke@435 | 1012 | // for longs. |
duke@435 | 1013 | |
duke@435 | 1014 | if (src.first()->is_stack() && dst.first()->is_stack()) { |
duke@435 | 1015 | assert(src.second()->is_stack() && dst.second()->is_stack(), "must be all stack"); |
never@739 | 1016 | __ movptr(rax, Address(rbp, reg2offset_in(src.first()))); |
never@739 | 1017 | NOT_LP64(__ movptr(rbx, Address(rbp, reg2offset_in(src.second())))); |
never@739 | 1018 | __ movptr(Address(rsp, reg2offset_out(dst.first())), rax); |
never@739 | 1019 | NOT_LP64(__ movptr(Address(rsp, reg2offset_out(dst.second())), rbx)); |
duke@435 | 1020 | } else { |
duke@435 | 1021 | ShouldNotReachHere(); |
duke@435 | 1022 | } |
duke@435 | 1023 | } |
duke@435 | 1024 | |
duke@435 | 1025 | // A double move |
duke@435 | 1026 | static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) { |
duke@435 | 1027 | |
duke@435 | 1028 | // The only legal possibilities for a double_move VMRegPair are: |
duke@435 | 1029 | // The painful thing here is that like long_move a VMRegPair might be |
duke@435 | 1030 | |
duke@435 | 1031 | // Because of the calling convention we know that src is either |
duke@435 | 1032 | // 1: a single physical register (xmm registers only) |
duke@435 | 1033 | // 2: two stack slots (possibly unaligned) |
duke@435 | 1034 | // dst can only be a pair of stack slots. |
duke@435 | 1035 | |
duke@435 | 1036 | assert(dst.first()->is_stack() && (src.first()->is_XMMRegister() || src.first()->is_stack()), "bad args"); |
duke@435 | 1037 | |
duke@435 | 1038 | if (src.first()->is_stack()) { |
duke@435 | 1039 | // source is all stack |
never@739 | 1040 | __ movptr(rax, Address(rbp, reg2offset_in(src.first()))); |
never@739 | 1041 | NOT_LP64(__ movptr(rbx, Address(rbp, reg2offset_in(src.second())))); |
never@739 | 1042 | __ movptr(Address(rsp, reg2offset_out(dst.first())), rax); |
never@739 | 1043 | NOT_LP64(__ movptr(Address(rsp, reg2offset_out(dst.second())), rbx)); |
duke@435 | 1044 | } else { |
duke@435 | 1045 | // reg to stack |
duke@435 | 1046 | // No worries about stack alignment |
duke@435 | 1047 | __ movdbl(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister()); |
duke@435 | 1048 | } |
duke@435 | 1049 | } |
duke@435 | 1050 | |
duke@435 | 1051 | |
duke@435 | 1052 | void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { |
duke@435 | 1053 | // We always ignore the frame_slots arg and just use the space just below frame pointer |
duke@435 | 1054 | // which by this time is free to use |
duke@435 | 1055 | switch (ret_type) { |
duke@435 | 1056 | case T_FLOAT: |
duke@435 | 1057 | __ fstp_s(Address(rbp, -wordSize)); |
duke@435 | 1058 | break; |
duke@435 | 1059 | case T_DOUBLE: |
duke@435 | 1060 | __ fstp_d(Address(rbp, -2*wordSize)); |
duke@435 | 1061 | break; |
duke@435 | 1062 | case T_VOID: break; |
duke@435 | 1063 | case T_LONG: |
never@739 | 1064 | __ movptr(Address(rbp, -wordSize), rax); |
never@739 | 1065 | NOT_LP64(__ movptr(Address(rbp, -2*wordSize), rdx)); |
duke@435 | 1066 | break; |
duke@435 | 1067 | default: { |
never@739 | 1068 | __ movptr(Address(rbp, -wordSize), rax); |
duke@435 | 1069 | } |
duke@435 | 1070 | } |
duke@435 | 1071 | } |
duke@435 | 1072 | |
duke@435 | 1073 | void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) { |
duke@435 | 1074 | // We always ignore the frame_slots arg and just use the space just below frame pointer |
duke@435 | 1075 | // which by this time is free to use |
duke@435 | 1076 | switch (ret_type) { |
duke@435 | 1077 | case T_FLOAT: |
duke@435 | 1078 | __ fld_s(Address(rbp, -wordSize)); |
duke@435 | 1079 | break; |
duke@435 | 1080 | case T_DOUBLE: |
duke@435 | 1081 | __ fld_d(Address(rbp, -2*wordSize)); |
duke@435 | 1082 | break; |
duke@435 | 1083 | case T_LONG: |
never@739 | 1084 | __ movptr(rax, Address(rbp, -wordSize)); |
never@739 | 1085 | NOT_LP64(__ movptr(rdx, Address(rbp, -2*wordSize))); |
duke@435 | 1086 | break; |
duke@435 | 1087 | case T_VOID: break; |
duke@435 | 1088 | default: { |
never@739 | 1089 | __ movptr(rax, Address(rbp, -wordSize)); |
duke@435 | 1090 | } |
duke@435 | 1091 | } |
duke@435 | 1092 | } |
duke@435 | 1093 | |
duke@435 | 1094 | // --------------------------------------------------------------------------- |
duke@435 | 1095 | // Generate a native wrapper for a given method. The method takes arguments |
duke@435 | 1096 | // in the Java compiled code convention, marshals them to the native |
duke@435 | 1097 | // convention (handlizes oops, etc), transitions to native, makes the call, |
duke@435 | 1098 | // returns to java state (possibly blocking), unhandlizes any result and |
duke@435 | 1099 | // returns. |
duke@435 | 1100 | nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm, |
duke@435 | 1101 | methodHandle method, |
twisti@2687 | 1102 | int compile_id, |
duke@435 | 1103 | int total_in_args, |
duke@435 | 1104 | int comp_args_on_stack, |
duke@435 | 1105 | BasicType *in_sig_bt, |
duke@435 | 1106 | VMRegPair *in_regs, |
duke@435 | 1107 | BasicType ret_type) { |
duke@435 | 1108 | |
duke@435 | 1109 | // An OopMap for lock (and class if static) |
duke@435 | 1110 | OopMapSet *oop_maps = new OopMapSet(); |
duke@435 | 1111 | |
duke@435 | 1112 | // We have received a description of where all the java arg are located |
duke@435 | 1113 | // on entry to the wrapper. We need to convert these args to where |
duke@435 | 1114 | // the jni function will expect them. To figure out where they go |
duke@435 | 1115 | // we convert the java signature to a C signature by inserting |
duke@435 | 1116 | // the hidden arguments as arg[0] and possibly arg[1] (static method) |
duke@435 | 1117 | |
duke@435 | 1118 | int total_c_args = total_in_args + 1; |
duke@435 | 1119 | if (method->is_static()) { |
duke@435 | 1120 | total_c_args++; |
duke@435 | 1121 | } |
duke@435 | 1122 | |
duke@435 | 1123 | BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args); |
duke@435 | 1124 | VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args); |
duke@435 | 1125 | |
duke@435 | 1126 | int argc = 0; |
duke@435 | 1127 | out_sig_bt[argc++] = T_ADDRESS; |
duke@435 | 1128 | if (method->is_static()) { |
duke@435 | 1129 | out_sig_bt[argc++] = T_OBJECT; |
duke@435 | 1130 | } |
duke@435 | 1131 | |
duke@435 | 1132 | int i; |
duke@435 | 1133 | for (i = 0; i < total_in_args ; i++ ) { |
duke@435 | 1134 | out_sig_bt[argc++] = in_sig_bt[i]; |
duke@435 | 1135 | } |
duke@435 | 1136 | |
duke@435 | 1137 | |
duke@435 | 1138 | // Now figure out where the args must be stored and how much stack space |
duke@435 | 1139 | // they require (neglecting out_preserve_stack_slots but space for storing |
duke@435 | 1140 | // the 1st six register arguments). It's weird see int_stk_helper. |
duke@435 | 1141 | // |
duke@435 | 1142 | int out_arg_slots; |
duke@435 | 1143 | out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args); |
duke@435 | 1144 | |
duke@435 | 1145 | // Compute framesize for the wrapper. We need to handlize all oops in |
duke@435 | 1146 | // registers a max of 2 on x86. |
duke@435 | 1147 | |
duke@435 | 1148 | // Calculate the total number of stack slots we will need. |
duke@435 | 1149 | |
duke@435 | 1150 | // First count the abi requirement plus all of the outgoing args |
duke@435 | 1151 | int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; |
duke@435 | 1152 | |
duke@435 | 1153 | // Now the space for the inbound oop handle area |
duke@435 | 1154 | |
duke@435 | 1155 | int oop_handle_offset = stack_slots; |
duke@435 | 1156 | stack_slots += 2*VMRegImpl::slots_per_word; |
duke@435 | 1157 | |
duke@435 | 1158 | // Now any space we need for handlizing a klass if static method |
duke@435 | 1159 | |
duke@435 | 1160 | int klass_slot_offset = 0; |
duke@435 | 1161 | int klass_offset = -1; |
duke@435 | 1162 | int lock_slot_offset = 0; |
duke@435 | 1163 | bool is_static = false; |
duke@435 | 1164 | int oop_temp_slot_offset = 0; |
duke@435 | 1165 | |
duke@435 | 1166 | if (method->is_static()) { |
duke@435 | 1167 | klass_slot_offset = stack_slots; |
duke@435 | 1168 | stack_slots += VMRegImpl::slots_per_word; |
duke@435 | 1169 | klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size; |
duke@435 | 1170 | is_static = true; |
duke@435 | 1171 | } |
duke@435 | 1172 | |
duke@435 | 1173 | // Plus a lock if needed |
duke@435 | 1174 | |
duke@435 | 1175 | if (method->is_synchronized()) { |
duke@435 | 1176 | lock_slot_offset = stack_slots; |
duke@435 | 1177 | stack_slots += VMRegImpl::slots_per_word; |
duke@435 | 1178 | } |
duke@435 | 1179 | |
duke@435 | 1180 | // Now a place (+2) to save return values or temp during shuffling |
duke@435 | 1181 | // + 2 for return address (which we own) and saved rbp, |
duke@435 | 1182 | stack_slots += 4; |
duke@435 | 1183 | |
duke@435 | 1184 | // Ok The space we have allocated will look like: |
duke@435 | 1185 | // |
duke@435 | 1186 | // |
duke@435 | 1187 | // FP-> | | |
duke@435 | 1188 | // |---------------------| |
duke@435 | 1189 | // | 2 slots for moves | |
duke@435 | 1190 | // |---------------------| |
duke@435 | 1191 | // | lock box (if sync) | |
duke@435 | 1192 | // |---------------------| <- lock_slot_offset (-lock_slot_rbp_offset) |
duke@435 | 1193 | // | klass (if static) | |
duke@435 | 1194 | // |---------------------| <- klass_slot_offset |
duke@435 | 1195 | // | oopHandle area | |
duke@435 | 1196 | // |---------------------| <- oop_handle_offset (a max of 2 registers) |
duke@435 | 1197 | // | outbound memory | |
duke@435 | 1198 | // | based arguments | |
duke@435 | 1199 | // | | |
duke@435 | 1200 | // |---------------------| |
duke@435 | 1201 | // | | |
duke@435 | 1202 | // SP-> | out_preserved_slots | |
duke@435 | 1203 | // |
duke@435 | 1204 | // |
duke@435 | 1205 | // **************************************************************************** |
duke@435 | 1206 | // WARNING - on Windows Java Natives use pascal calling convention and pop the |
duke@435 | 1207 | // arguments off of the stack after the jni call. Before the call we can use |
duke@435 | 1208 | // instructions that are SP relative. After the jni call we switch to FP |
duke@435 | 1209 | // relative instructions instead of re-adjusting the stack on windows. |
duke@435 | 1210 | // **************************************************************************** |
duke@435 | 1211 | |
duke@435 | 1212 | |
duke@435 | 1213 | // Now compute actual number of stack words we need rounding to make |
duke@435 | 1214 | // stack properly aligned. |
xlu@959 | 1215 | stack_slots = round_to(stack_slots, StackAlignmentInSlots); |
duke@435 | 1216 | |
duke@435 | 1217 | int stack_size = stack_slots * VMRegImpl::stack_slot_size; |
duke@435 | 1218 | |
duke@435 | 1219 | intptr_t start = (intptr_t)__ pc(); |
duke@435 | 1220 | |
duke@435 | 1221 | // First thing make an ic check to see if we should even be here |
duke@435 | 1222 | |
duke@435 | 1223 | // We are free to use all registers as temps without saving them and |
duke@435 | 1224 | // restoring them except rbp,. rbp, is the only callee save register |
duke@435 | 1225 | // as far as the interpreter and the compiler(s) are concerned. |
duke@435 | 1226 | |
duke@435 | 1227 | |
duke@435 | 1228 | const Register ic_reg = rax; |
duke@435 | 1229 | const Register receiver = rcx; |
duke@435 | 1230 | Label hit; |
duke@435 | 1231 | Label exception_pending; |
duke@435 | 1232 | |
duke@435 | 1233 | |
duke@435 | 1234 | __ verify_oop(receiver); |
never@739 | 1235 | __ cmpptr(ic_reg, Address(receiver, oopDesc::klass_offset_in_bytes())); |
duke@435 | 1236 | __ jcc(Assembler::equal, hit); |
duke@435 | 1237 | |
duke@435 | 1238 | __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); |
duke@435 | 1239 | |
duke@435 | 1240 | // verified entry must be aligned for code patching. |
duke@435 | 1241 | // and the first 5 bytes must be in the same cache line |
duke@435 | 1242 | // if we align at 8 then we will be sure 5 bytes are in the same line |
duke@435 | 1243 | __ align(8); |
duke@435 | 1244 | |
duke@435 | 1245 | __ bind(hit); |
duke@435 | 1246 | |
duke@435 | 1247 | int vep_offset = ((intptr_t)__ pc()) - start; |
duke@435 | 1248 | |
duke@435 | 1249 | #ifdef COMPILER1 |
duke@435 | 1250 | if (InlineObjectHash && method->intrinsic_id() == vmIntrinsics::_hashCode) { |
duke@435 | 1251 | // Object.hashCode can pull the hashCode from the header word |
duke@435 | 1252 | // instead of doing a full VM transition once it's been computed. |
duke@435 | 1253 | // Since hashCode is usually polymorphic at call sites we can't do |
duke@435 | 1254 | // this optimization at the call site without a lot of work. |
duke@435 | 1255 | Label slowCase; |
duke@435 | 1256 | Register receiver = rcx; |
duke@435 | 1257 | Register result = rax; |
never@739 | 1258 | __ movptr(result, Address(receiver, oopDesc::mark_offset_in_bytes())); |
duke@435 | 1259 | |
duke@435 | 1260 | // check if locked |
never@739 | 1261 | __ testptr(result, markOopDesc::unlocked_value); |
duke@435 | 1262 | __ jcc (Assembler::zero, slowCase); |
duke@435 | 1263 | |
duke@435 | 1264 | if (UseBiasedLocking) { |
duke@435 | 1265 | // Check if biased and fall through to runtime if so |
never@739 | 1266 | __ testptr(result, markOopDesc::biased_lock_bit_in_place); |
duke@435 | 1267 | __ jcc (Assembler::notZero, slowCase); |
duke@435 | 1268 | } |
duke@435 | 1269 | |
duke@435 | 1270 | // get hash |
never@739 | 1271 | __ andptr(result, markOopDesc::hash_mask_in_place); |
duke@435 | 1272 | // test if hashCode exists |
duke@435 | 1273 | __ jcc (Assembler::zero, slowCase); |
never@739 | 1274 | __ shrptr(result, markOopDesc::hash_shift); |
duke@435 | 1275 | __ ret(0); |
duke@435 | 1276 | __ bind (slowCase); |
duke@435 | 1277 | } |
duke@435 | 1278 | #endif // COMPILER1 |
duke@435 | 1279 | |
duke@435 | 1280 | // The instruction at the verified entry point must be 5 bytes or longer |
duke@435 | 1281 | // because it can be patched on the fly by make_non_entrant. The stack bang |
duke@435 | 1282 | // instruction fits that requirement. |
duke@435 | 1283 | |
duke@435 | 1284 | // Generate stack overflow check |
duke@435 | 1285 | |
duke@435 | 1286 | if (UseStackBanging) { |
duke@435 | 1287 | __ bang_stack_with_offset(StackShadowPages*os::vm_page_size()); |
duke@435 | 1288 | } else { |
duke@435 | 1289 | // need a 5 byte instruction to allow MT safe patching to non-entrant |
duke@435 | 1290 | __ fat_nop(); |
duke@435 | 1291 | } |
duke@435 | 1292 | |
duke@435 | 1293 | // Generate a new frame for the wrapper. |
duke@435 | 1294 | __ enter(); |
duke@435 | 1295 | // -2 because return address is already present and so is saved rbp, |
never@739 | 1296 | __ subptr(rsp, stack_size - 2*wordSize); |
duke@435 | 1297 | |
duke@435 | 1298 | // Frame is now completed as far a size and linkage. |
duke@435 | 1299 | |
duke@435 | 1300 | int frame_complete = ((intptr_t)__ pc()) - start; |
duke@435 | 1301 | |
duke@435 | 1302 | // Calculate the difference between rsp and rbp,. We need to know it |
duke@435 | 1303 | // after the native call because on windows Java Natives will pop |
duke@435 | 1304 | // the arguments and it is painful to do rsp relative addressing |
duke@435 | 1305 | // in a platform independent way. So after the call we switch to |
duke@435 | 1306 | // rbp, relative addressing. |
duke@435 | 1307 | |
duke@435 | 1308 | int fp_adjustment = stack_size - 2*wordSize; |
duke@435 | 1309 | |
duke@435 | 1310 | #ifdef COMPILER2 |
duke@435 | 1311 | // C2 may leave the stack dirty if not in SSE2+ mode |
duke@435 | 1312 | if (UseSSE >= 2) { |
duke@435 | 1313 | __ verify_FPU(0, "c2i transition should have clean FPU stack"); |
duke@435 | 1314 | } else { |
duke@435 | 1315 | __ empty_FPU_stack(); |
duke@435 | 1316 | } |
duke@435 | 1317 | #endif /* COMPILER2 */ |
duke@435 | 1318 | |
duke@435 | 1319 | // Compute the rbp, offset for any slots used after the jni call |
duke@435 | 1320 | |
duke@435 | 1321 | int lock_slot_rbp_offset = (lock_slot_offset*VMRegImpl::stack_slot_size) - fp_adjustment; |
duke@435 | 1322 | int oop_temp_slot_rbp_offset = (oop_temp_slot_offset*VMRegImpl::stack_slot_size) - fp_adjustment; |
duke@435 | 1323 | |
duke@435 | 1324 | // We use rdi as a thread pointer because it is callee save and |
duke@435 | 1325 | // if we load it once it is usable thru the entire wrapper |
duke@435 | 1326 | const Register thread = rdi; |
duke@435 | 1327 | |
duke@435 | 1328 | // We use rsi as the oop handle for the receiver/klass |
duke@435 | 1329 | // It is callee save so it survives the call to native |
duke@435 | 1330 | |
duke@435 | 1331 | const Register oop_handle_reg = rsi; |
duke@435 | 1332 | |
duke@435 | 1333 | __ get_thread(thread); |
duke@435 | 1334 | |
duke@435 | 1335 | |
duke@435 | 1336 | // |
duke@435 | 1337 | // We immediately shuffle the arguments so that any vm call we have to |
duke@435 | 1338 | // make from here on out (sync slow path, jvmti, etc.) we will have |
duke@435 | 1339 | // captured the oops from our caller and have a valid oopMap for |
duke@435 | 1340 | // them. |
duke@435 | 1341 | |
duke@435 | 1342 | // ----------------- |
duke@435 | 1343 | // The Grand Shuffle |
duke@435 | 1344 | // |
duke@435 | 1345 | // Natives require 1 or 2 extra arguments over the normal ones: the JNIEnv* |
duke@435 | 1346 | // and, if static, the class mirror instead of a receiver. This pretty much |
duke@435 | 1347 | // guarantees that register layout will not match (and x86 doesn't use reg |
duke@435 | 1348 | // parms though amd does). Since the native abi doesn't use register args |
duke@435 | 1349 | // and the java conventions does we don't have to worry about collisions. |
duke@435 | 1350 | // All of our moved are reg->stack or stack->stack. |
duke@435 | 1351 | // We ignore the extra arguments during the shuffle and handle them at the |
duke@435 | 1352 | // last moment. The shuffle is described by the two calling convention |
duke@435 | 1353 | // vectors we have in our possession. We simply walk the java vector to |
duke@435 | 1354 | // get the source locations and the c vector to get the destinations. |
duke@435 | 1355 | |
duke@435 | 1356 | int c_arg = method->is_static() ? 2 : 1 ; |
duke@435 | 1357 | |
duke@435 | 1358 | // Record rsp-based slot for receiver on stack for non-static methods |
duke@435 | 1359 | int receiver_offset = -1; |
duke@435 | 1360 | |
duke@435 | 1361 | // This is a trick. We double the stack slots so we can claim |
duke@435 | 1362 | // the oops in the caller's frame. Since we are sure to have |
duke@435 | 1363 | // more args than the caller doubling is enough to make |
duke@435 | 1364 | // sure we can capture all the incoming oop args from the |
duke@435 | 1365 | // caller. |
duke@435 | 1366 | // |
duke@435 | 1367 | OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/); |
duke@435 | 1368 | |
duke@435 | 1369 | // Mark location of rbp, |
duke@435 | 1370 | // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, rbp->as_VMReg()); |
duke@435 | 1371 | |
duke@435 | 1372 | // We know that we only have args in at most two integer registers (rcx, rdx). So rax, rbx |
duke@435 | 1373 | // Are free to temporaries if we have to do stack to steck moves. |
duke@435 | 1374 | // All inbound args are referenced based on rbp, and all outbound args via rsp. |
duke@435 | 1375 | |
duke@435 | 1376 | for (i = 0; i < total_in_args ; i++, c_arg++ ) { |
duke@435 | 1377 | switch (in_sig_bt[i]) { |
duke@435 | 1378 | case T_ARRAY: |
duke@435 | 1379 | case T_OBJECT: |
duke@435 | 1380 | object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg], |
duke@435 | 1381 | ((i == 0) && (!is_static)), |
duke@435 | 1382 | &receiver_offset); |
duke@435 | 1383 | break; |
duke@435 | 1384 | case T_VOID: |
duke@435 | 1385 | break; |
duke@435 | 1386 | |
duke@435 | 1387 | case T_FLOAT: |
duke@435 | 1388 | float_move(masm, in_regs[i], out_regs[c_arg]); |
duke@435 | 1389 | break; |
duke@435 | 1390 | |
duke@435 | 1391 | case T_DOUBLE: |
duke@435 | 1392 | assert( i + 1 < total_in_args && |
duke@435 | 1393 | in_sig_bt[i + 1] == T_VOID && |
duke@435 | 1394 | out_sig_bt[c_arg+1] == T_VOID, "bad arg list"); |
duke@435 | 1395 | double_move(masm, in_regs[i], out_regs[c_arg]); |
duke@435 | 1396 | break; |
duke@435 | 1397 | |
duke@435 | 1398 | case T_LONG : |
duke@435 | 1399 | long_move(masm, in_regs[i], out_regs[c_arg]); |
duke@435 | 1400 | break; |
duke@435 | 1401 | |
duke@435 | 1402 | case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); |
duke@435 | 1403 | |
duke@435 | 1404 | default: |
duke@435 | 1405 | simple_move32(masm, in_regs[i], out_regs[c_arg]); |
duke@435 | 1406 | } |
duke@435 | 1407 | } |
duke@435 | 1408 | |
duke@435 | 1409 | // Pre-load a static method's oop into rsi. Used both by locking code and |
duke@435 | 1410 | // the normal JNI call code. |
duke@435 | 1411 | if (method->is_static()) { |
duke@435 | 1412 | |
duke@435 | 1413 | // load opp into a register |
duke@435 | 1414 | __ movoop(oop_handle_reg, JNIHandles::make_local(Klass::cast(method->method_holder())->java_mirror())); |
duke@435 | 1415 | |
duke@435 | 1416 | // Now handlize the static class mirror it's known not-null. |
never@739 | 1417 | __ movptr(Address(rsp, klass_offset), oop_handle_reg); |
duke@435 | 1418 | map->set_oop(VMRegImpl::stack2reg(klass_slot_offset)); |
duke@435 | 1419 | |
duke@435 | 1420 | // Now get the handle |
never@739 | 1421 | __ lea(oop_handle_reg, Address(rsp, klass_offset)); |
duke@435 | 1422 | // store the klass handle as second argument |
never@739 | 1423 | __ movptr(Address(rsp, wordSize), oop_handle_reg); |
duke@435 | 1424 | } |
duke@435 | 1425 | |
duke@435 | 1426 | // Change state to native (we save the return address in the thread, since it might not |
duke@435 | 1427 | // be pushed on the stack when we do a a stack traversal). It is enough that the pc() |
duke@435 | 1428 | // points into the right code segment. It does not have to be the correct return pc. |
duke@435 | 1429 | // We use the same pc/oopMap repeatedly when we call out |
duke@435 | 1430 | |
duke@435 | 1431 | intptr_t the_pc = (intptr_t) __ pc(); |
duke@435 | 1432 | oop_maps->add_gc_map(the_pc - start, map); |
duke@435 | 1433 | |
duke@435 | 1434 | __ set_last_Java_frame(thread, rsp, noreg, (address)the_pc); |
duke@435 | 1435 | |
duke@435 | 1436 | |
duke@435 | 1437 | // We have all of the arguments setup at this point. We must not touch any register |
duke@435 | 1438 | // argument registers at this point (what if we save/restore them there are no oop? |
duke@435 | 1439 | |
duke@435 | 1440 | { |
duke@435 | 1441 | SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0); |
duke@435 | 1442 | __ movoop(rax, JNIHandles::make_local(method())); |
duke@435 | 1443 | __ call_VM_leaf( |
duke@435 | 1444 | CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), |
duke@435 | 1445 | thread, rax); |
duke@435 | 1446 | } |
duke@435 | 1447 | |
dcubed@1045 | 1448 | // RedefineClasses() tracing support for obsolete method entry |
dcubed@1045 | 1449 | if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { |
dcubed@1045 | 1450 | __ movoop(rax, JNIHandles::make_local(method())); |
dcubed@1045 | 1451 | __ call_VM_leaf( |
dcubed@1045 | 1452 | CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), |
dcubed@1045 | 1453 | thread, rax); |
dcubed@1045 | 1454 | } |
dcubed@1045 | 1455 | |
duke@435 | 1456 | // These are register definitions we need for locking/unlocking |
duke@435 | 1457 | const Register swap_reg = rax; // Must use rax, for cmpxchg instruction |
duke@435 | 1458 | const Register obj_reg = rcx; // Will contain the oop |
duke@435 | 1459 | const Register lock_reg = rdx; // Address of compiler lock object (BasicLock) |
duke@435 | 1460 | |
duke@435 | 1461 | Label slow_path_lock; |
duke@435 | 1462 | Label lock_done; |
duke@435 | 1463 | |
duke@435 | 1464 | // Lock a synchronized method |
duke@435 | 1465 | if (method->is_synchronized()) { |
duke@435 | 1466 | |
duke@435 | 1467 | |
duke@435 | 1468 | const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes(); |
duke@435 | 1469 | |
duke@435 | 1470 | // Get the handle (the 2nd argument) |
never@739 | 1471 | __ movptr(oop_handle_reg, Address(rsp, wordSize)); |
duke@435 | 1472 | |
duke@435 | 1473 | // Get address of the box |
duke@435 | 1474 | |
never@739 | 1475 | __ lea(lock_reg, Address(rbp, lock_slot_rbp_offset)); |
duke@435 | 1476 | |
duke@435 | 1477 | // Load the oop from the handle |
never@739 | 1478 | __ movptr(obj_reg, Address(oop_handle_reg, 0)); |
duke@435 | 1479 | |
duke@435 | 1480 | if (UseBiasedLocking) { |
duke@435 | 1481 | // Note that oop_handle_reg is trashed during this call |
duke@435 | 1482 | __ biased_locking_enter(lock_reg, obj_reg, swap_reg, oop_handle_reg, false, lock_done, &slow_path_lock); |
duke@435 | 1483 | } |
duke@435 | 1484 | |
duke@435 | 1485 | // Load immediate 1 into swap_reg %rax, |
never@739 | 1486 | __ movptr(swap_reg, 1); |
duke@435 | 1487 | |
duke@435 | 1488 | // Load (object->mark() | 1) into swap_reg %rax, |
never@739 | 1489 | __ orptr(swap_reg, Address(obj_reg, 0)); |
duke@435 | 1490 | |
duke@435 | 1491 | // Save (object->mark() | 1) into BasicLock's displaced header |
never@739 | 1492 | __ movptr(Address(lock_reg, mark_word_offset), swap_reg); |
duke@435 | 1493 | |
duke@435 | 1494 | if (os::is_MP()) { |
duke@435 | 1495 | __ lock(); |
duke@435 | 1496 | } |
duke@435 | 1497 | |
duke@435 | 1498 | // src -> dest iff dest == rax, else rax, <- dest |
duke@435 | 1499 | // *obj_reg = lock_reg iff *obj_reg == rax, else rax, = *(obj_reg) |
never@739 | 1500 | __ cmpxchgptr(lock_reg, Address(obj_reg, 0)); |
duke@435 | 1501 | __ jcc(Assembler::equal, lock_done); |
duke@435 | 1502 | |
duke@435 | 1503 | // Test if the oopMark is an obvious stack pointer, i.e., |
duke@435 | 1504 | // 1) (mark & 3) == 0, and |
duke@435 | 1505 | // 2) rsp <= mark < mark + os::pagesize() |
duke@435 | 1506 | // These 3 tests can be done by evaluating the following |
duke@435 | 1507 | // expression: ((mark - rsp) & (3 - os::vm_page_size())), |
duke@435 | 1508 | // assuming both stack pointer and pagesize have their |
duke@435 | 1509 | // least significant 2 bits clear. |
duke@435 | 1510 | // NOTE: the oopMark is in swap_reg %rax, as the result of cmpxchg |
duke@435 | 1511 | |
never@739 | 1512 | __ subptr(swap_reg, rsp); |
never@739 | 1513 | __ andptr(swap_reg, 3 - os::vm_page_size()); |
duke@435 | 1514 | |
duke@435 | 1515 | // Save the test result, for recursive case, the result is zero |
never@739 | 1516 | __ movptr(Address(lock_reg, mark_word_offset), swap_reg); |
duke@435 | 1517 | __ jcc(Assembler::notEqual, slow_path_lock); |
duke@435 | 1518 | // Slow path will re-enter here |
duke@435 | 1519 | __ bind(lock_done); |
duke@435 | 1520 | |
duke@435 | 1521 | if (UseBiasedLocking) { |
duke@435 | 1522 | // Re-fetch oop_handle_reg as we trashed it above |
never@739 | 1523 | __ movptr(oop_handle_reg, Address(rsp, wordSize)); |
duke@435 | 1524 | } |
duke@435 | 1525 | } |
duke@435 | 1526 | |
duke@435 | 1527 | |
duke@435 | 1528 | // Finally just about ready to make the JNI call |
duke@435 | 1529 | |
duke@435 | 1530 | |
duke@435 | 1531 | // get JNIEnv* which is first argument to native |
duke@435 | 1532 | |
never@739 | 1533 | __ lea(rdx, Address(thread, in_bytes(JavaThread::jni_environment_offset()))); |
never@739 | 1534 | __ movptr(Address(rsp, 0), rdx); |
duke@435 | 1535 | |
duke@435 | 1536 | // Now set thread in native |
duke@435 | 1537 | __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native); |
duke@435 | 1538 | |
duke@435 | 1539 | __ call(RuntimeAddress(method->native_function())); |
duke@435 | 1540 | |
duke@435 | 1541 | // WARNING - on Windows Java Natives use pascal calling convention and pop the |
duke@435 | 1542 | // arguments off of the stack. We could just re-adjust the stack pointer here |
duke@435 | 1543 | // and continue to do SP relative addressing but we instead switch to FP |
duke@435 | 1544 | // relative addressing. |
duke@435 | 1545 | |
duke@435 | 1546 | // Unpack native results. |
duke@435 | 1547 | switch (ret_type) { |
duke@435 | 1548 | case T_BOOLEAN: __ c2bool(rax); break; |
never@739 | 1549 | case T_CHAR : __ andptr(rax, 0xFFFF); break; |
duke@435 | 1550 | case T_BYTE : __ sign_extend_byte (rax); break; |
duke@435 | 1551 | case T_SHORT : __ sign_extend_short(rax); break; |
duke@435 | 1552 | case T_INT : /* nothing to do */ break; |
duke@435 | 1553 | case T_DOUBLE : |
duke@435 | 1554 | case T_FLOAT : |
duke@435 | 1555 | // Result is in st0 we'll save as needed |
duke@435 | 1556 | break; |
duke@435 | 1557 | case T_ARRAY: // Really a handle |
duke@435 | 1558 | case T_OBJECT: // Really a handle |
duke@435 | 1559 | break; // can't de-handlize until after safepoint check |
duke@435 | 1560 | case T_VOID: break; |
duke@435 | 1561 | case T_LONG: break; |
duke@435 | 1562 | default : ShouldNotReachHere(); |
duke@435 | 1563 | } |
duke@435 | 1564 | |
duke@435 | 1565 | // Switch thread to "native transition" state before reading the synchronization state. |
duke@435 | 1566 | // This additional state is necessary because reading and testing the synchronization |
duke@435 | 1567 | // state is not atomic w.r.t. GC, as this scenario demonstrates: |
duke@435 | 1568 | // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted. |
duke@435 | 1569 | // VM thread changes sync state to synchronizing and suspends threads for GC. |
duke@435 | 1570 | // Thread A is resumed to finish this native method, but doesn't block here since it |
duke@435 | 1571 | // didn't see any synchronization is progress, and escapes. |
duke@435 | 1572 | __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans); |
duke@435 | 1573 | |
duke@435 | 1574 | if(os::is_MP()) { |
duke@435 | 1575 | if (UseMembar) { |
never@739 | 1576 | // Force this write out before the read below |
never@739 | 1577 | __ membar(Assembler::Membar_mask_bits( |
never@739 | 1578 | Assembler::LoadLoad | Assembler::LoadStore | |
never@739 | 1579 | Assembler::StoreLoad | Assembler::StoreStore)); |
duke@435 | 1580 | } else { |
duke@435 | 1581 | // Write serialization page so VM thread can do a pseudo remote membar. |
duke@435 | 1582 | // We use the current thread pointer to calculate a thread specific |
duke@435 | 1583 | // offset to write to within the page. This minimizes bus traffic |
duke@435 | 1584 | // due to cache line collision. |
duke@435 | 1585 | __ serialize_memory(thread, rcx); |
duke@435 | 1586 | } |
duke@435 | 1587 | } |
duke@435 | 1588 | |
duke@435 | 1589 | if (AlwaysRestoreFPU) { |
duke@435 | 1590 | // Make sure the control word is correct. |
duke@435 | 1591 | __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); |
duke@435 | 1592 | } |
duke@435 | 1593 | |
duke@435 | 1594 | // check for safepoint operation in progress and/or pending suspend requests |
duke@435 | 1595 | { Label Continue; |
duke@435 | 1596 | |
duke@435 | 1597 | __ cmp32(ExternalAddress((address)SafepointSynchronize::address_of_state()), |
duke@435 | 1598 | SafepointSynchronize::_not_synchronized); |
duke@435 | 1599 | |
duke@435 | 1600 | Label L; |
duke@435 | 1601 | __ jcc(Assembler::notEqual, L); |
duke@435 | 1602 | __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0); |
duke@435 | 1603 | __ jcc(Assembler::equal, Continue); |
duke@435 | 1604 | __ bind(L); |
duke@435 | 1605 | |
duke@435 | 1606 | // Don't use call_VM as it will see a possible pending exception and forward it |
duke@435 | 1607 | // and never return here preventing us from clearing _last_native_pc down below. |
duke@435 | 1608 | // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are |
duke@435 | 1609 | // preserved and correspond to the bcp/locals pointers. So we do a runtime call |
duke@435 | 1610 | // by hand. |
duke@435 | 1611 | // |
duke@435 | 1612 | save_native_result(masm, ret_type, stack_slots); |
never@739 | 1613 | __ push(thread); |
duke@435 | 1614 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, |
duke@435 | 1615 | JavaThread::check_special_condition_for_native_trans))); |
duke@435 | 1616 | __ increment(rsp, wordSize); |
duke@435 | 1617 | // Restore any method result value |
duke@435 | 1618 | restore_native_result(masm, ret_type, stack_slots); |
duke@435 | 1619 | |
duke@435 | 1620 | __ bind(Continue); |
duke@435 | 1621 | } |
duke@435 | 1622 | |
duke@435 | 1623 | // change thread state |
duke@435 | 1624 | __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java); |
duke@435 | 1625 | |
duke@435 | 1626 | Label reguard; |
duke@435 | 1627 | Label reguard_done; |
duke@435 | 1628 | __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled); |
duke@435 | 1629 | __ jcc(Assembler::equal, reguard); |
duke@435 | 1630 | |
duke@435 | 1631 | // slow path reguard re-enters here |
duke@435 | 1632 | __ bind(reguard_done); |
duke@435 | 1633 | |
duke@435 | 1634 | // Handle possible exception (will unlock if necessary) |
duke@435 | 1635 | |
duke@435 | 1636 | // native result if any is live |
duke@435 | 1637 | |
duke@435 | 1638 | // Unlock |
duke@435 | 1639 | Label slow_path_unlock; |
duke@435 | 1640 | Label unlock_done; |
duke@435 | 1641 | if (method->is_synchronized()) { |
duke@435 | 1642 | |
duke@435 | 1643 | Label done; |
duke@435 | 1644 | |
duke@435 | 1645 | // Get locked oop from the handle we passed to jni |
never@739 | 1646 | __ movptr(obj_reg, Address(oop_handle_reg, 0)); |
duke@435 | 1647 | |
duke@435 | 1648 | if (UseBiasedLocking) { |
duke@435 | 1649 | __ biased_locking_exit(obj_reg, rbx, done); |
duke@435 | 1650 | } |
duke@435 | 1651 | |
duke@435 | 1652 | // Simple recursive lock? |
duke@435 | 1653 | |
never@739 | 1654 | __ cmpptr(Address(rbp, lock_slot_rbp_offset), (int32_t)NULL_WORD); |
duke@435 | 1655 | __ jcc(Assembler::equal, done); |
duke@435 | 1656 | |
duke@435 | 1657 | // Must save rax, if if it is live now because cmpxchg must use it |
duke@435 | 1658 | if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { |
duke@435 | 1659 | save_native_result(masm, ret_type, stack_slots); |
duke@435 | 1660 | } |
duke@435 | 1661 | |
duke@435 | 1662 | // get old displaced header |
never@739 | 1663 | __ movptr(rbx, Address(rbp, lock_slot_rbp_offset)); |
duke@435 | 1664 | |
duke@435 | 1665 | // get address of the stack lock |
never@739 | 1666 | __ lea(rax, Address(rbp, lock_slot_rbp_offset)); |
duke@435 | 1667 | |
duke@435 | 1668 | // Atomic swap old header if oop still contains the stack lock |
duke@435 | 1669 | if (os::is_MP()) { |
duke@435 | 1670 | __ lock(); |
duke@435 | 1671 | } |
duke@435 | 1672 | |
duke@435 | 1673 | // src -> dest iff dest == rax, else rax, <- dest |
duke@435 | 1674 | // *obj_reg = rbx, iff *obj_reg == rax, else rax, = *(obj_reg) |
never@739 | 1675 | __ cmpxchgptr(rbx, Address(obj_reg, 0)); |
duke@435 | 1676 | __ jcc(Assembler::notEqual, slow_path_unlock); |
duke@435 | 1677 | |
duke@435 | 1678 | // slow path re-enters here |
duke@435 | 1679 | __ bind(unlock_done); |
duke@435 | 1680 | if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) { |
duke@435 | 1681 | restore_native_result(masm, ret_type, stack_slots); |
duke@435 | 1682 | } |
duke@435 | 1683 | |
duke@435 | 1684 | __ bind(done); |
duke@435 | 1685 | |
duke@435 | 1686 | } |
duke@435 | 1687 | |
duke@435 | 1688 | { |
duke@435 | 1689 | SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0); |
duke@435 | 1690 | // Tell dtrace about this method exit |
duke@435 | 1691 | save_native_result(masm, ret_type, stack_slots); |
duke@435 | 1692 | __ movoop(rax, JNIHandles::make_local(method())); |
duke@435 | 1693 | __ call_VM_leaf( |
duke@435 | 1694 | CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), |
duke@435 | 1695 | thread, rax); |
duke@435 | 1696 | restore_native_result(masm, ret_type, stack_slots); |
duke@435 | 1697 | } |
duke@435 | 1698 | |
duke@435 | 1699 | // We can finally stop using that last_Java_frame we setup ages ago |
duke@435 | 1700 | |
duke@435 | 1701 | __ reset_last_Java_frame(thread, false, true); |
duke@435 | 1702 | |
duke@435 | 1703 | // Unpack oop result |
duke@435 | 1704 | if (ret_type == T_OBJECT || ret_type == T_ARRAY) { |
duke@435 | 1705 | Label L; |
never@739 | 1706 | __ cmpptr(rax, (int32_t)NULL_WORD); |
duke@435 | 1707 | __ jcc(Assembler::equal, L); |
never@739 | 1708 | __ movptr(rax, Address(rax, 0)); |
duke@435 | 1709 | __ bind(L); |
duke@435 | 1710 | __ verify_oop(rax); |
duke@435 | 1711 | } |
duke@435 | 1712 | |
duke@435 | 1713 | // reset handle block |
never@739 | 1714 | __ movptr(rcx, Address(thread, JavaThread::active_handles_offset())); |
never@739 | 1715 | |
xlu@947 | 1716 | __ movptr(Address(rcx, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD); |
duke@435 | 1717 | |
duke@435 | 1718 | // Any exception pending? |
never@739 | 1719 | __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); |
duke@435 | 1720 | __ jcc(Assembler::notEqual, exception_pending); |
duke@435 | 1721 | |
duke@435 | 1722 | |
duke@435 | 1723 | // no exception, we're almost done |
duke@435 | 1724 | |
duke@435 | 1725 | // check that only result value is on FPU stack |
duke@435 | 1726 | __ verify_FPU(ret_type == T_FLOAT || ret_type == T_DOUBLE ? 1 : 0, "native_wrapper normal exit"); |
duke@435 | 1727 | |
duke@435 | 1728 | // Fixup floating pointer results so that result looks like a return from a compiled method |
duke@435 | 1729 | if (ret_type == T_FLOAT) { |
duke@435 | 1730 | if (UseSSE >= 1) { |
duke@435 | 1731 | // Pop st0 and store as float and reload into xmm register |
duke@435 | 1732 | __ fstp_s(Address(rbp, -4)); |
duke@435 | 1733 | __ movflt(xmm0, Address(rbp, -4)); |
duke@435 | 1734 | } |
duke@435 | 1735 | } else if (ret_type == T_DOUBLE) { |
duke@435 | 1736 | if (UseSSE >= 2) { |
duke@435 | 1737 | // Pop st0 and store as double and reload into xmm register |
duke@435 | 1738 | __ fstp_d(Address(rbp, -8)); |
duke@435 | 1739 | __ movdbl(xmm0, Address(rbp, -8)); |
duke@435 | 1740 | } |
duke@435 | 1741 | } |
duke@435 | 1742 | |
duke@435 | 1743 | // Return |
duke@435 | 1744 | |
duke@435 | 1745 | __ leave(); |
duke@435 | 1746 | __ ret(0); |
duke@435 | 1747 | |
duke@435 | 1748 | // Unexpected paths are out of line and go here |
duke@435 | 1749 | |
duke@435 | 1750 | // Slow path locking & unlocking |
duke@435 | 1751 | if (method->is_synchronized()) { |
duke@435 | 1752 | |
duke@435 | 1753 | // BEGIN Slow path lock |
duke@435 | 1754 | |
duke@435 | 1755 | __ bind(slow_path_lock); |
duke@435 | 1756 | |
duke@435 | 1757 | // has last_Java_frame setup. No exceptions so do vanilla call not call_VM |
duke@435 | 1758 | // args are (oop obj, BasicLock* lock, JavaThread* thread) |
never@739 | 1759 | __ push(thread); |
never@739 | 1760 | __ push(lock_reg); |
never@739 | 1761 | __ push(obj_reg); |
duke@435 | 1762 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C))); |
never@739 | 1763 | __ addptr(rsp, 3*wordSize); |
duke@435 | 1764 | |
duke@435 | 1765 | #ifdef ASSERT |
duke@435 | 1766 | { Label L; |
never@739 | 1767 | __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD); |
duke@435 | 1768 | __ jcc(Assembler::equal, L); |
duke@435 | 1769 | __ stop("no pending exception allowed on exit from monitorenter"); |
duke@435 | 1770 | __ bind(L); |
duke@435 | 1771 | } |
duke@435 | 1772 | #endif |
duke@435 | 1773 | __ jmp(lock_done); |
duke@435 | 1774 | |
duke@435 | 1775 | // END Slow path lock |
duke@435 | 1776 | |
duke@435 | 1777 | // BEGIN Slow path unlock |
duke@435 | 1778 | __ bind(slow_path_unlock); |
duke@435 | 1779 | |
duke@435 | 1780 | // Slow path unlock |
duke@435 | 1781 | |
duke@435 | 1782 | if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { |
duke@435 | 1783 | save_native_result(masm, ret_type, stack_slots); |
duke@435 | 1784 | } |
duke@435 | 1785 | // Save pending exception around call to VM (which contains an EXCEPTION_MARK) |
duke@435 | 1786 | |
never@739 | 1787 | __ pushptr(Address(thread, in_bytes(Thread::pending_exception_offset()))); |
xlu@947 | 1788 | __ movptr(Address(thread, in_bytes(Thread::pending_exception_offset())), NULL_WORD); |
duke@435 | 1789 | |
duke@435 | 1790 | |
duke@435 | 1791 | // should be a peal |
duke@435 | 1792 | // +wordSize because of the push above |
never@739 | 1793 | __ lea(rax, Address(rbp, lock_slot_rbp_offset)); |
never@739 | 1794 | __ push(rax); |
never@739 | 1795 | |
never@739 | 1796 | __ push(obj_reg); |
duke@435 | 1797 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C))); |
never@739 | 1798 | __ addptr(rsp, 2*wordSize); |
duke@435 | 1799 | #ifdef ASSERT |
duke@435 | 1800 | { |
duke@435 | 1801 | Label L; |
never@739 | 1802 | __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD); |
duke@435 | 1803 | __ jcc(Assembler::equal, L); |
duke@435 | 1804 | __ stop("no pending exception allowed on exit complete_monitor_unlocking_C"); |
duke@435 | 1805 | __ bind(L); |
duke@435 | 1806 | } |
duke@435 | 1807 | #endif /* ASSERT */ |
duke@435 | 1808 | |
never@739 | 1809 | __ popptr(Address(thread, in_bytes(Thread::pending_exception_offset()))); |
duke@435 | 1810 | |
duke@435 | 1811 | if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) { |
duke@435 | 1812 | restore_native_result(masm, ret_type, stack_slots); |
duke@435 | 1813 | } |
duke@435 | 1814 | __ jmp(unlock_done); |
duke@435 | 1815 | // END Slow path unlock |
duke@435 | 1816 | |
duke@435 | 1817 | } |
duke@435 | 1818 | |
duke@435 | 1819 | // SLOW PATH Reguard the stack if needed |
duke@435 | 1820 | |
duke@435 | 1821 | __ bind(reguard); |
duke@435 | 1822 | save_native_result(masm, ret_type, stack_slots); |
duke@435 | 1823 | { |
duke@435 | 1824 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); |
duke@435 | 1825 | } |
duke@435 | 1826 | restore_native_result(masm, ret_type, stack_slots); |
duke@435 | 1827 | __ jmp(reguard_done); |
duke@435 | 1828 | |
duke@435 | 1829 | |
duke@435 | 1830 | // BEGIN EXCEPTION PROCESSING |
duke@435 | 1831 | |
duke@435 | 1832 | // Forward the exception |
duke@435 | 1833 | __ bind(exception_pending); |
duke@435 | 1834 | |
duke@435 | 1835 | // remove possible return value from FPU register stack |
duke@435 | 1836 | __ empty_FPU_stack(); |
duke@435 | 1837 | |
duke@435 | 1838 | // pop our frame |
duke@435 | 1839 | __ leave(); |
duke@435 | 1840 | // and forward the exception |
duke@435 | 1841 | __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
duke@435 | 1842 | |
duke@435 | 1843 | __ flush(); |
duke@435 | 1844 | |
duke@435 | 1845 | nmethod *nm = nmethod::new_native_nmethod(method, |
twisti@2687 | 1846 | compile_id, |
duke@435 | 1847 | masm->code(), |
duke@435 | 1848 | vep_offset, |
duke@435 | 1849 | frame_complete, |
duke@435 | 1850 | stack_slots / VMRegImpl::slots_per_word, |
duke@435 | 1851 | (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)), |
duke@435 | 1852 | in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size), |
duke@435 | 1853 | oop_maps); |
duke@435 | 1854 | return nm; |
duke@435 | 1855 | |
duke@435 | 1856 | } |
duke@435 | 1857 | |
kamg@551 | 1858 | #ifdef HAVE_DTRACE_H |
kamg@551 | 1859 | // --------------------------------------------------------------------------- |
kamg@551 | 1860 | // Generate a dtrace nmethod for a given signature. The method takes arguments |
kamg@551 | 1861 | // in the Java compiled code convention, marshals them to the native |
kamg@551 | 1862 | // abi and then leaves nops at the position you would expect to call a native |
kamg@551 | 1863 | // function. When the probe is enabled the nops are replaced with a trap |
kamg@551 | 1864 | // instruction that dtrace inserts and the trace will cause a notification |
kamg@551 | 1865 | // to dtrace. |
kamg@551 | 1866 | // |
kamg@551 | 1867 | // The probes are only able to take primitive types and java/lang/String as |
kamg@551 | 1868 | // arguments. No other java types are allowed. Strings are converted to utf8 |
kamg@551 | 1869 | // strings so that from dtrace point of view java strings are converted to C |
kamg@551 | 1870 | // strings. There is an arbitrary fixed limit on the total space that a method |
kamg@551 | 1871 | // can use for converting the strings. (256 chars per string in the signature). |
kamg@551 | 1872 | // So any java string larger then this is truncated. |
kamg@551 | 1873 | |
kamg@551 | 1874 | nmethod *SharedRuntime::generate_dtrace_nmethod( |
kamg@551 | 1875 | MacroAssembler *masm, methodHandle method) { |
kamg@551 | 1876 | |
kamg@551 | 1877 | // generate_dtrace_nmethod is guarded by a mutex so we are sure to |
kamg@551 | 1878 | // be single threaded in this method. |
kamg@551 | 1879 | assert(AdapterHandlerLibrary_lock->owned_by_self(), "must be"); |
kamg@551 | 1880 | |
kamg@551 | 1881 | // Fill in the signature array, for the calling-convention call. |
kamg@551 | 1882 | int total_args_passed = method->size_of_parameters(); |
kamg@551 | 1883 | |
kamg@551 | 1884 | BasicType* in_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed); |
kamg@551 | 1885 | VMRegPair *in_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed); |
kamg@551 | 1886 | |
kamg@551 | 1887 | // The signature we are going to use for the trap that dtrace will see |
kamg@551 | 1888 | // java/lang/String is converted. We drop "this" and any other object |
kamg@551 | 1889 | // is converted to NULL. (A one-slot java/lang/Long object reference |
kamg@551 | 1890 | // is converted to a two-slot long, which is why we double the allocation). |
kamg@551 | 1891 | BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed * 2); |
kamg@551 | 1892 | VMRegPair* out_regs = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed * 2); |
kamg@551 | 1893 | |
kamg@551 | 1894 | int i=0; |
kamg@551 | 1895 | int total_strings = 0; |
kamg@551 | 1896 | int first_arg_to_pass = 0; |
kamg@551 | 1897 | int total_c_args = 0; |
kamg@551 | 1898 | |
kamg@551 | 1899 | if( !method->is_static() ) { // Pass in receiver first |
kamg@551 | 1900 | in_sig_bt[i++] = T_OBJECT; |
kamg@551 | 1901 | first_arg_to_pass = 1; |
kamg@551 | 1902 | } |
kamg@551 | 1903 | |
kamg@551 | 1904 | // We need to convert the java args to where a native (non-jni) function |
kamg@551 | 1905 | // would expect them. To figure out where they go we convert the java |
kamg@551 | 1906 | // signature to a C signature. |
kamg@551 | 1907 | |
kamg@551 | 1908 | SignatureStream ss(method->signature()); |
kamg@551 | 1909 | for ( ; !ss.at_return_type(); ss.next()) { |
kamg@551 | 1910 | BasicType bt = ss.type(); |
kamg@551 | 1911 | in_sig_bt[i++] = bt; // Collect remaining bits of signature |
kamg@551 | 1912 | out_sig_bt[total_c_args++] = bt; |
kamg@551 | 1913 | if( bt == T_OBJECT) { |
coleenp@2497 | 1914 | Symbol* s = ss.as_symbol_or_null(); // symbol is created |
kamg@551 | 1915 | if (s == vmSymbols::java_lang_String()) { |
kamg@551 | 1916 | total_strings++; |
kamg@551 | 1917 | out_sig_bt[total_c_args-1] = T_ADDRESS; |
kamg@551 | 1918 | } else if (s == vmSymbols::java_lang_Boolean() || |
kamg@551 | 1919 | s == vmSymbols::java_lang_Character() || |
kamg@551 | 1920 | s == vmSymbols::java_lang_Byte() || |
kamg@551 | 1921 | s == vmSymbols::java_lang_Short() || |
kamg@551 | 1922 | s == vmSymbols::java_lang_Integer() || |
kamg@551 | 1923 | s == vmSymbols::java_lang_Float()) { |
kamg@551 | 1924 | out_sig_bt[total_c_args-1] = T_INT; |
kamg@551 | 1925 | } else if (s == vmSymbols::java_lang_Long() || |
kamg@551 | 1926 | s == vmSymbols::java_lang_Double()) { |
kamg@551 | 1927 | out_sig_bt[total_c_args-1] = T_LONG; |
kamg@551 | 1928 | out_sig_bt[total_c_args++] = T_VOID; |
kamg@551 | 1929 | } |
kamg@551 | 1930 | } else if ( bt == T_LONG || bt == T_DOUBLE ) { |
kamg@551 | 1931 | in_sig_bt[i++] = T_VOID; // Longs & doubles take 2 Java slots |
kamg@551 | 1932 | out_sig_bt[total_c_args++] = T_VOID; |
kamg@551 | 1933 | } |
kamg@551 | 1934 | } |
kamg@551 | 1935 | |
kamg@551 | 1936 | assert(i==total_args_passed, "validly parsed signature"); |
kamg@551 | 1937 | |
kamg@551 | 1938 | // Now get the compiled-Java layout as input arguments |
kamg@551 | 1939 | int comp_args_on_stack; |
kamg@551 | 1940 | comp_args_on_stack = SharedRuntime::java_calling_convention( |
kamg@551 | 1941 | in_sig_bt, in_regs, total_args_passed, false); |
kamg@551 | 1942 | |
kamg@551 | 1943 | // Now figure out where the args must be stored and how much stack space |
kamg@551 | 1944 | // they require (neglecting out_preserve_stack_slots). |
kamg@551 | 1945 | |
kamg@551 | 1946 | int out_arg_slots; |
kamg@551 | 1947 | out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args); |
kamg@551 | 1948 | |
kamg@551 | 1949 | // Calculate the total number of stack slots we will need. |
kamg@551 | 1950 | |
kamg@551 | 1951 | // First count the abi requirement plus all of the outgoing args |
kamg@551 | 1952 | int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots; |
kamg@551 | 1953 | |
kamg@551 | 1954 | // Now space for the string(s) we must convert |
kamg@551 | 1955 | |
kamg@551 | 1956 | int* string_locs = NEW_RESOURCE_ARRAY(int, total_strings + 1); |
kamg@551 | 1957 | for (i = 0; i < total_strings ; i++) { |
kamg@551 | 1958 | string_locs[i] = stack_slots; |
kamg@551 | 1959 | stack_slots += max_dtrace_string_size / VMRegImpl::stack_slot_size; |
kamg@551 | 1960 | } |
kamg@551 | 1961 | |
kamg@551 | 1962 | // + 2 for return address (which we own) and saved rbp, |
kamg@551 | 1963 | |
kamg@551 | 1964 | stack_slots += 2; |
kamg@551 | 1965 | |
kamg@551 | 1966 | // Ok The space we have allocated will look like: |
kamg@551 | 1967 | // |
kamg@551 | 1968 | // |
kamg@551 | 1969 | // FP-> | | |
kamg@551 | 1970 | // |---------------------| |
kamg@551 | 1971 | // | string[n] | |
kamg@551 | 1972 | // |---------------------| <- string_locs[n] |
kamg@551 | 1973 | // | string[n-1] | |
kamg@551 | 1974 | // |---------------------| <- string_locs[n-1] |
kamg@551 | 1975 | // | ... | |
kamg@551 | 1976 | // | ... | |
kamg@551 | 1977 | // |---------------------| <- string_locs[1] |
kamg@551 | 1978 | // | string[0] | |
kamg@551 | 1979 | // |---------------------| <- string_locs[0] |
kamg@551 | 1980 | // | outbound memory | |
kamg@551 | 1981 | // | based arguments | |
kamg@551 | 1982 | // | | |
kamg@551 | 1983 | // |---------------------| |
kamg@551 | 1984 | // | | |
kamg@551 | 1985 | // SP-> | out_preserved_slots | |
kamg@551 | 1986 | // |
kamg@551 | 1987 | // |
kamg@551 | 1988 | |
kamg@551 | 1989 | // Now compute actual number of stack words we need rounding to make |
kamg@551 | 1990 | // stack properly aligned. |
kamg@551 | 1991 | stack_slots = round_to(stack_slots, 2 * VMRegImpl::slots_per_word); |
kamg@551 | 1992 | |
kamg@551 | 1993 | int stack_size = stack_slots * VMRegImpl::stack_slot_size; |
kamg@551 | 1994 | |
kamg@551 | 1995 | intptr_t start = (intptr_t)__ pc(); |
kamg@551 | 1996 | |
kamg@551 | 1997 | // First thing make an ic check to see if we should even be here |
kamg@551 | 1998 | |
kamg@551 | 1999 | // We are free to use all registers as temps without saving them and |
kamg@551 | 2000 | // restoring them except rbp. rbp, is the only callee save register |
kamg@551 | 2001 | // as far as the interpreter and the compiler(s) are concerned. |
kamg@551 | 2002 | |
kamg@551 | 2003 | const Register ic_reg = rax; |
kamg@551 | 2004 | const Register receiver = rcx; |
kamg@551 | 2005 | Label hit; |
kamg@551 | 2006 | Label exception_pending; |
kamg@551 | 2007 | |
kamg@551 | 2008 | |
kamg@551 | 2009 | __ verify_oop(receiver); |
kamg@551 | 2010 | __ cmpl(ic_reg, Address(receiver, oopDesc::klass_offset_in_bytes())); |
kamg@551 | 2011 | __ jcc(Assembler::equal, hit); |
kamg@551 | 2012 | |
kamg@551 | 2013 | __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); |
kamg@551 | 2014 | |
kamg@551 | 2015 | // verified entry must be aligned for code patching. |
kamg@551 | 2016 | // and the first 5 bytes must be in the same cache line |
kamg@551 | 2017 | // if we align at 8 then we will be sure 5 bytes are in the same line |
kamg@551 | 2018 | __ align(8); |
kamg@551 | 2019 | |
kamg@551 | 2020 | __ bind(hit); |
kamg@551 | 2021 | |
kamg@551 | 2022 | int vep_offset = ((intptr_t)__ pc()) - start; |
kamg@551 | 2023 | |
kamg@551 | 2024 | |
kamg@551 | 2025 | // The instruction at the verified entry point must be 5 bytes or longer |
kamg@551 | 2026 | // because it can be patched on the fly by make_non_entrant. The stack bang |
kamg@551 | 2027 | // instruction fits that requirement. |
kamg@551 | 2028 | |
kamg@551 | 2029 | // Generate stack overflow check |
kamg@551 | 2030 | |
kamg@551 | 2031 | |
kamg@551 | 2032 | if (UseStackBanging) { |
kamg@551 | 2033 | if (stack_size <= StackShadowPages*os::vm_page_size()) { |
kamg@551 | 2034 | __ bang_stack_with_offset(StackShadowPages*os::vm_page_size()); |
kamg@551 | 2035 | } else { |
kamg@551 | 2036 | __ movl(rax, stack_size); |
kamg@551 | 2037 | __ bang_stack_size(rax, rbx); |
kamg@551 | 2038 | } |
kamg@551 | 2039 | } else { |
kamg@551 | 2040 | // need a 5 byte instruction to allow MT safe patching to non-entrant |
kamg@551 | 2041 | __ fat_nop(); |
kamg@551 | 2042 | } |
kamg@551 | 2043 | |
kamg@551 | 2044 | assert(((int)__ pc() - start - vep_offset) >= 5, |
kamg@551 | 2045 | "valid size for make_non_entrant"); |
kamg@551 | 2046 | |
kamg@551 | 2047 | // Generate a new frame for the wrapper. |
kamg@551 | 2048 | __ enter(); |
kamg@551 | 2049 | |
kamg@551 | 2050 | // -2 because return address is already present and so is saved rbp, |
kamg@551 | 2051 | if (stack_size - 2*wordSize != 0) { |
kamg@551 | 2052 | __ subl(rsp, stack_size - 2*wordSize); |
kamg@551 | 2053 | } |
kamg@551 | 2054 | |
kamg@551 | 2055 | // Frame is now completed as far a size and linkage. |
kamg@551 | 2056 | |
kamg@551 | 2057 | int frame_complete = ((intptr_t)__ pc()) - start; |
kamg@551 | 2058 | |
kamg@551 | 2059 | // First thing we do store all the args as if we are doing the call. |
kamg@551 | 2060 | // Since the C calling convention is stack based that ensures that |
kamg@551 | 2061 | // all the Java register args are stored before we need to convert any |
kamg@551 | 2062 | // string we might have. |
kamg@551 | 2063 | |
kamg@551 | 2064 | int sid = 0; |
kamg@551 | 2065 | int c_arg, j_arg; |
kamg@551 | 2066 | int string_reg = 0; |
kamg@551 | 2067 | |
kamg@551 | 2068 | for (j_arg = first_arg_to_pass, c_arg = 0 ; |
kamg@551 | 2069 | j_arg < total_args_passed ; j_arg++, c_arg++ ) { |
kamg@551 | 2070 | |
kamg@551 | 2071 | VMRegPair src = in_regs[j_arg]; |
kamg@551 | 2072 | VMRegPair dst = out_regs[c_arg]; |
kamg@551 | 2073 | assert(dst.first()->is_stack() || in_sig_bt[j_arg] == T_VOID, |
kamg@551 | 2074 | "stack based abi assumed"); |
kamg@551 | 2075 | |
kamg@551 | 2076 | switch (in_sig_bt[j_arg]) { |
kamg@551 | 2077 | |
kamg@551 | 2078 | case T_ARRAY: |
kamg@551 | 2079 | case T_OBJECT: |
kamg@551 | 2080 | if (out_sig_bt[c_arg] == T_ADDRESS) { |
kamg@551 | 2081 | // Any register based arg for a java string after the first |
kamg@551 | 2082 | // will be destroyed by the call to get_utf so we store |
kamg@551 | 2083 | // the original value in the location the utf string address |
kamg@551 | 2084 | // will eventually be stored. |
kamg@551 | 2085 | if (src.first()->is_reg()) { |
kamg@551 | 2086 | if (string_reg++ != 0) { |
kamg@551 | 2087 | simple_move32(masm, src, dst); |
kamg@551 | 2088 | } |
kamg@551 | 2089 | } |
kamg@551 | 2090 | } else if (out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) { |
kamg@551 | 2091 | // need to unbox a one-word value |
kamg@551 | 2092 | Register in_reg = rax; |
kamg@551 | 2093 | if ( src.first()->is_reg() ) { |
kamg@551 | 2094 | in_reg = src.first()->as_Register(); |
kamg@551 | 2095 | } else { |
kamg@551 | 2096 | simple_move32(masm, src, in_reg->as_VMReg()); |
kamg@551 | 2097 | } |
kamg@551 | 2098 | Label skipUnbox; |
kamg@551 | 2099 | __ movl(Address(rsp, reg2offset_out(dst.first())), NULL_WORD); |
kamg@551 | 2100 | if ( out_sig_bt[c_arg] == T_LONG ) { |
kamg@551 | 2101 | __ movl(Address(rsp, reg2offset_out(dst.second())), NULL_WORD); |
kamg@551 | 2102 | } |
kamg@551 | 2103 | __ testl(in_reg, in_reg); |
kamg@551 | 2104 | __ jcc(Assembler::zero, skipUnbox); |
kamg@551 | 2105 | assert(dst.first()->is_stack() && |
kamg@551 | 2106 | (!dst.second()->is_valid() || dst.second()->is_stack()), |
kamg@551 | 2107 | "value(s) must go into stack slots"); |
kvn@600 | 2108 | |
kvn@600 | 2109 | BasicType bt = out_sig_bt[c_arg]; |
kvn@600 | 2110 | int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt); |
kvn@600 | 2111 | if ( bt == T_LONG ) { |
kamg@551 | 2112 | __ movl(rbx, Address(in_reg, |
kamg@551 | 2113 | box_offset + VMRegImpl::stack_slot_size)); |
kamg@551 | 2114 | __ movl(Address(rsp, reg2offset_out(dst.second())), rbx); |
kamg@551 | 2115 | } |
kamg@551 | 2116 | __ movl(in_reg, Address(in_reg, box_offset)); |
kamg@551 | 2117 | __ movl(Address(rsp, reg2offset_out(dst.first())), in_reg); |
kamg@551 | 2118 | __ bind(skipUnbox); |
kamg@551 | 2119 | } else { |
kamg@551 | 2120 | // Convert the arg to NULL |
kamg@551 | 2121 | __ movl(Address(rsp, reg2offset_out(dst.first())), NULL_WORD); |
kamg@551 | 2122 | } |
kamg@551 | 2123 | if (out_sig_bt[c_arg] == T_LONG) { |
kamg@551 | 2124 | assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); |
kamg@551 | 2125 | ++c_arg; // Move over the T_VOID To keep the loop indices in sync |
kamg@551 | 2126 | } |
kamg@551 | 2127 | break; |
kamg@551 | 2128 | |
kamg@551 | 2129 | case T_VOID: |
kamg@551 | 2130 | break; |
kamg@551 | 2131 | |
kamg@551 | 2132 | case T_FLOAT: |
kamg@551 | 2133 | float_move(masm, src, dst); |
kamg@551 | 2134 | break; |
kamg@551 | 2135 | |
kamg@551 | 2136 | case T_DOUBLE: |
kamg@551 | 2137 | assert( j_arg + 1 < total_args_passed && |
kamg@551 | 2138 | in_sig_bt[j_arg + 1] == T_VOID, "bad arg list"); |
kamg@551 | 2139 | double_move(masm, src, dst); |
kamg@551 | 2140 | break; |
kamg@551 | 2141 | |
kamg@551 | 2142 | case T_LONG : |
kamg@551 | 2143 | long_move(masm, src, dst); |
kamg@551 | 2144 | break; |
kamg@551 | 2145 | |
kamg@551 | 2146 | case T_ADDRESS: assert(false, "found T_ADDRESS in java args"); |
kamg@551 | 2147 | |
kamg@551 | 2148 | default: |
kamg@551 | 2149 | simple_move32(masm, src, dst); |
kamg@551 | 2150 | } |
kamg@551 | 2151 | } |
kamg@551 | 2152 | |
kamg@551 | 2153 | // Now we must convert any string we have to utf8 |
kamg@551 | 2154 | // |
kamg@551 | 2155 | |
kamg@551 | 2156 | for (sid = 0, j_arg = first_arg_to_pass, c_arg = 0 ; |
kamg@551 | 2157 | sid < total_strings ; j_arg++, c_arg++ ) { |
kamg@551 | 2158 | |
kamg@551 | 2159 | if (out_sig_bt[c_arg] == T_ADDRESS) { |
kamg@551 | 2160 | |
kamg@551 | 2161 | Address utf8_addr = Address( |
kamg@551 | 2162 | rsp, string_locs[sid++] * VMRegImpl::stack_slot_size); |
kamg@551 | 2163 | __ leal(rax, utf8_addr); |
kamg@551 | 2164 | |
kamg@551 | 2165 | // The first string we find might still be in the original java arg |
kamg@551 | 2166 | // register |
kamg@551 | 2167 | VMReg orig_loc = in_regs[j_arg].first(); |
kamg@551 | 2168 | Register string_oop; |
kamg@551 | 2169 | |
kamg@551 | 2170 | // This is where the argument will eventually reside |
kamg@551 | 2171 | Address dest = Address(rsp, reg2offset_out(out_regs[c_arg].first())); |
kamg@551 | 2172 | |
kamg@551 | 2173 | if (sid == 1 && orig_loc->is_reg()) { |
kamg@551 | 2174 | string_oop = orig_loc->as_Register(); |
kamg@551 | 2175 | assert(string_oop != rax, "smashed arg"); |
kamg@551 | 2176 | } else { |
kamg@551 | 2177 | |
kamg@551 | 2178 | if (orig_loc->is_reg()) { |
kamg@551 | 2179 | // Get the copy of the jls object |
kamg@551 | 2180 | __ movl(rcx, dest); |
kamg@551 | 2181 | } else { |
kamg@551 | 2182 | // arg is still in the original location |
kamg@551 | 2183 | __ movl(rcx, Address(rbp, reg2offset_in(orig_loc))); |
kamg@551 | 2184 | } |
kamg@551 | 2185 | string_oop = rcx; |
kamg@551 | 2186 | |
kamg@551 | 2187 | } |
kamg@551 | 2188 | Label nullString; |
kamg@551 | 2189 | __ movl(dest, NULL_WORD); |
kamg@551 | 2190 | __ testl(string_oop, string_oop); |
kamg@551 | 2191 | __ jcc(Assembler::zero, nullString); |
kamg@551 | 2192 | |
kamg@551 | 2193 | // Now we can store the address of the utf string as the argument |
kamg@551 | 2194 | __ movl(dest, rax); |
kamg@551 | 2195 | |
kamg@551 | 2196 | // And do the conversion |
kamg@551 | 2197 | __ call_VM_leaf(CAST_FROM_FN_PTR( |
kamg@551 | 2198 | address, SharedRuntime::get_utf), string_oop, rax); |
kamg@551 | 2199 | __ bind(nullString); |
kamg@551 | 2200 | } |
kamg@551 | 2201 | |
kamg@551 | 2202 | if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) { |
kamg@551 | 2203 | assert(out_sig_bt[c_arg+1] == T_VOID, "must be"); |
kamg@551 | 2204 | ++c_arg; // Move over the T_VOID To keep the loop indices in sync |
kamg@551 | 2205 | } |
kamg@551 | 2206 | } |
kamg@551 | 2207 | |
kamg@551 | 2208 | |
kamg@551 | 2209 | // Ok now we are done. Need to place the nop that dtrace wants in order to |
kamg@551 | 2210 | // patch in the trap |
kamg@551 | 2211 | |
kamg@551 | 2212 | int patch_offset = ((intptr_t)__ pc()) - start; |
kamg@551 | 2213 | |
kamg@551 | 2214 | __ nop(); |
kamg@551 | 2215 | |
kamg@551 | 2216 | |
kamg@551 | 2217 | // Return |
kamg@551 | 2218 | |
kamg@551 | 2219 | __ leave(); |
kamg@551 | 2220 | __ ret(0); |
kamg@551 | 2221 | |
kamg@551 | 2222 | __ flush(); |
kamg@551 | 2223 | |
kamg@551 | 2224 | nmethod *nm = nmethod::new_dtrace_nmethod( |
kamg@551 | 2225 | method, masm->code(), vep_offset, patch_offset, frame_complete, |
kamg@551 | 2226 | stack_slots / VMRegImpl::slots_per_word); |
kamg@551 | 2227 | return nm; |
kamg@551 | 2228 | |
kamg@551 | 2229 | } |
kamg@551 | 2230 | |
kamg@551 | 2231 | #endif // HAVE_DTRACE_H |
kamg@551 | 2232 | |
duke@435 | 2233 | // this function returns the adjust size (in number of words) to a c2i adapter |
duke@435 | 2234 | // activation for use during deoptimization |
duke@435 | 2235 | int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals ) { |
twisti@1861 | 2236 | return (callee_locals - callee_parameters) * Interpreter::stackElementWords; |
duke@435 | 2237 | } |
duke@435 | 2238 | |
duke@435 | 2239 | |
duke@435 | 2240 | uint SharedRuntime::out_preserve_stack_slots() { |
duke@435 | 2241 | return 0; |
duke@435 | 2242 | } |
duke@435 | 2243 | |
duke@435 | 2244 | |
duke@435 | 2245 | //------------------------------generate_deopt_blob---------------------------- |
duke@435 | 2246 | void SharedRuntime::generate_deopt_blob() { |
duke@435 | 2247 | // allocate space for the code |
duke@435 | 2248 | ResourceMark rm; |
duke@435 | 2249 | // setup code generation tools |
duke@435 | 2250 | CodeBuffer buffer("deopt_blob", 1024, 1024); |
duke@435 | 2251 | MacroAssembler* masm = new MacroAssembler(&buffer); |
duke@435 | 2252 | int frame_size_in_words; |
duke@435 | 2253 | OopMap* map = NULL; |
duke@435 | 2254 | // Account for the extra args we place on the stack |
duke@435 | 2255 | // by the time we call fetch_unroll_info |
duke@435 | 2256 | const int additional_words = 2; // deopt kind, thread |
duke@435 | 2257 | |
duke@435 | 2258 | OopMapSet *oop_maps = new OopMapSet(); |
duke@435 | 2259 | |
duke@435 | 2260 | // ------------- |
duke@435 | 2261 | // This code enters when returning to a de-optimized nmethod. A return |
duke@435 | 2262 | // address has been pushed on the the stack, and return values are in |
duke@435 | 2263 | // registers. |
duke@435 | 2264 | // If we are doing a normal deopt then we were called from the patched |
duke@435 | 2265 | // nmethod from the point we returned to the nmethod. So the return |
duke@435 | 2266 | // address on the stack is wrong by NativeCall::instruction_size |
duke@435 | 2267 | // We will adjust the value to it looks like we have the original return |
duke@435 | 2268 | // address on the stack (like when we eagerly deoptimized). |
duke@435 | 2269 | // In the case of an exception pending with deoptimized then we enter |
duke@435 | 2270 | // with a return address on the stack that points after the call we patched |
duke@435 | 2271 | // into the exception handler. We have the following register state: |
duke@435 | 2272 | // rax,: exception |
duke@435 | 2273 | // rbx,: exception handler |
duke@435 | 2274 | // rdx: throwing pc |
duke@435 | 2275 | // So in this case we simply jam rdx into the useless return address and |
duke@435 | 2276 | // the stack looks just like we want. |
duke@435 | 2277 | // |
duke@435 | 2278 | // At this point we need to de-opt. We save the argument return |
duke@435 | 2279 | // registers. We call the first C routine, fetch_unroll_info(). This |
duke@435 | 2280 | // routine captures the return values and returns a structure which |
duke@435 | 2281 | // describes the current frame size and the sizes of all replacement frames. |
duke@435 | 2282 | // The current frame is compiled code and may contain many inlined |
duke@435 | 2283 | // functions, each with their own JVM state. We pop the current frame, then |
duke@435 | 2284 | // push all the new frames. Then we call the C routine unpack_frames() to |
duke@435 | 2285 | // populate these frames. Finally unpack_frames() returns us the new target |
duke@435 | 2286 | // address. Notice that callee-save registers are BLOWN here; they have |
duke@435 | 2287 | // already been captured in the vframeArray at the time the return PC was |
duke@435 | 2288 | // patched. |
duke@435 | 2289 | address start = __ pc(); |
duke@435 | 2290 | Label cont; |
duke@435 | 2291 | |
duke@435 | 2292 | // Prolog for non exception case! |
duke@435 | 2293 | |
duke@435 | 2294 | // Save everything in sight. |
duke@435 | 2295 | |
cfang@1361 | 2296 | map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false); |
duke@435 | 2297 | // Normal deoptimization |
never@739 | 2298 | __ push(Deoptimization::Unpack_deopt); |
duke@435 | 2299 | __ jmp(cont); |
duke@435 | 2300 | |
duke@435 | 2301 | int reexecute_offset = __ pc() - start; |
duke@435 | 2302 | |
duke@435 | 2303 | // Reexecute case |
duke@435 | 2304 | // return address is the pc describes what bci to do re-execute at |
duke@435 | 2305 | |
duke@435 | 2306 | // No need to update map as each call to save_live_registers will produce identical oopmap |
cfang@1361 | 2307 | (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false); |
duke@435 | 2308 | |
never@739 | 2309 | __ push(Deoptimization::Unpack_reexecute); |
duke@435 | 2310 | __ jmp(cont); |
duke@435 | 2311 | |
duke@435 | 2312 | int exception_offset = __ pc() - start; |
duke@435 | 2313 | |
duke@435 | 2314 | // Prolog for exception case |
duke@435 | 2315 | |
duke@435 | 2316 | // all registers are dead at this entry point, except for rax, and |
duke@435 | 2317 | // rdx which contain the exception oop and exception pc |
duke@435 | 2318 | // respectively. Set them in TLS and fall thru to the |
duke@435 | 2319 | // unpack_with_exception_in_tls entry point. |
duke@435 | 2320 | |
duke@435 | 2321 | __ get_thread(rdi); |
never@739 | 2322 | __ movptr(Address(rdi, JavaThread::exception_pc_offset()), rdx); |
never@739 | 2323 | __ movptr(Address(rdi, JavaThread::exception_oop_offset()), rax); |
duke@435 | 2324 | |
duke@435 | 2325 | int exception_in_tls_offset = __ pc() - start; |
duke@435 | 2326 | |
duke@435 | 2327 | // new implementation because exception oop is now passed in JavaThread |
duke@435 | 2328 | |
duke@435 | 2329 | // Prolog for exception case |
duke@435 | 2330 | // All registers must be preserved because they might be used by LinearScan |
duke@435 | 2331 | // Exceptiop oop and throwing PC are passed in JavaThread |
duke@435 | 2332 | // tos: stack at point of call to method that threw the exception (i.e. only |
duke@435 | 2333 | // args are on the stack, no return address) |
duke@435 | 2334 | |
duke@435 | 2335 | // make room on stack for the return address |
duke@435 | 2336 | // It will be patched later with the throwing pc. The correct value is not |
duke@435 | 2337 | // available now because loading it from memory would destroy registers. |
never@739 | 2338 | __ push(0); |
duke@435 | 2339 | |
duke@435 | 2340 | // Save everything in sight. |
duke@435 | 2341 | |
duke@435 | 2342 | // No need to update map as each call to save_live_registers will produce identical oopmap |
cfang@1361 | 2343 | (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false); |
duke@435 | 2344 | |
duke@435 | 2345 | // Now it is safe to overwrite any register |
duke@435 | 2346 | |
duke@435 | 2347 | // store the correct deoptimization type |
never@739 | 2348 | __ push(Deoptimization::Unpack_exception); |
duke@435 | 2349 | |
duke@435 | 2350 | // load throwing pc from JavaThread and patch it as the return address |
duke@435 | 2351 | // of the current frame. Then clear the field in JavaThread |
duke@435 | 2352 | __ get_thread(rdi); |
never@739 | 2353 | __ movptr(rdx, Address(rdi, JavaThread::exception_pc_offset())); |
never@739 | 2354 | __ movptr(Address(rbp, wordSize), rdx); |
xlu@947 | 2355 | __ movptr(Address(rdi, JavaThread::exception_pc_offset()), NULL_WORD); |
duke@435 | 2356 | |
duke@435 | 2357 | #ifdef ASSERT |
duke@435 | 2358 | // verify that there is really an exception oop in JavaThread |
never@739 | 2359 | __ movptr(rax, Address(rdi, JavaThread::exception_oop_offset())); |
duke@435 | 2360 | __ verify_oop(rax); |
duke@435 | 2361 | |
duke@435 | 2362 | // verify that there is no pending exception |
duke@435 | 2363 | Label no_pending_exception; |
never@739 | 2364 | __ movptr(rax, Address(rdi, Thread::pending_exception_offset())); |
never@739 | 2365 | __ testptr(rax, rax); |
duke@435 | 2366 | __ jcc(Assembler::zero, no_pending_exception); |
duke@435 | 2367 | __ stop("must not have pending exception here"); |
duke@435 | 2368 | __ bind(no_pending_exception); |
duke@435 | 2369 | #endif |
duke@435 | 2370 | |
duke@435 | 2371 | __ bind(cont); |
duke@435 | 2372 | |
duke@435 | 2373 | // Compiled code leaves the floating point stack dirty, empty it. |
duke@435 | 2374 | __ empty_FPU_stack(); |
duke@435 | 2375 | |
duke@435 | 2376 | |
duke@435 | 2377 | // Call C code. Need thread and this frame, but NOT official VM entry |
duke@435 | 2378 | // crud. We cannot block on this call, no GC can happen. |
duke@435 | 2379 | __ get_thread(rcx); |
never@739 | 2380 | __ push(rcx); |
duke@435 | 2381 | // fetch_unroll_info needs to call last_java_frame() |
duke@435 | 2382 | __ set_last_Java_frame(rcx, noreg, noreg, NULL); |
duke@435 | 2383 | |
duke@435 | 2384 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info))); |
duke@435 | 2385 | |
duke@435 | 2386 | // Need to have an oopmap that tells fetch_unroll_info where to |
duke@435 | 2387 | // find any register it might need. |
duke@435 | 2388 | |
duke@435 | 2389 | oop_maps->add_gc_map( __ pc()-start, map); |
duke@435 | 2390 | |
duke@435 | 2391 | // Discard arg to fetch_unroll_info |
never@739 | 2392 | __ pop(rcx); |
duke@435 | 2393 | |
duke@435 | 2394 | __ get_thread(rcx); |
duke@435 | 2395 | __ reset_last_Java_frame(rcx, false, false); |
duke@435 | 2396 | |
duke@435 | 2397 | // Load UnrollBlock into EDI |
never@739 | 2398 | __ mov(rdi, rax); |
duke@435 | 2399 | |
duke@435 | 2400 | // Move the unpack kind to a safe place in the UnrollBlock because |
duke@435 | 2401 | // we are very short of registers |
duke@435 | 2402 | |
duke@435 | 2403 | Address unpack_kind(rdi, Deoptimization::UnrollBlock::unpack_kind_offset_in_bytes()); |
duke@435 | 2404 | // retrieve the deopt kind from where we left it. |
never@739 | 2405 | __ pop(rax); |
duke@435 | 2406 | __ movl(unpack_kind, rax); // save the unpack_kind value |
duke@435 | 2407 | |
duke@435 | 2408 | Label noException; |
duke@435 | 2409 | __ cmpl(rax, Deoptimization::Unpack_exception); // Was exception pending? |
duke@435 | 2410 | __ jcc(Assembler::notEqual, noException); |
never@739 | 2411 | __ movptr(rax, Address(rcx, JavaThread::exception_oop_offset())); |
never@739 | 2412 | __ movptr(rdx, Address(rcx, JavaThread::exception_pc_offset())); |
xlu@947 | 2413 | __ movptr(Address(rcx, JavaThread::exception_oop_offset()), NULL_WORD); |
xlu@947 | 2414 | __ movptr(Address(rcx, JavaThread::exception_pc_offset()), NULL_WORD); |
duke@435 | 2415 | |
duke@435 | 2416 | __ verify_oop(rax); |
duke@435 | 2417 | |
duke@435 | 2418 | // Overwrite the result registers with the exception results. |
never@739 | 2419 | __ movptr(Address(rsp, RegisterSaver::raxOffset()*wordSize), rax); |
never@739 | 2420 | __ movptr(Address(rsp, RegisterSaver::rdxOffset()*wordSize), rdx); |
duke@435 | 2421 | |
duke@435 | 2422 | __ bind(noException); |
duke@435 | 2423 | |
duke@435 | 2424 | // Stack is back to only having register save data on the stack. |
duke@435 | 2425 | // Now restore the result registers. Everything else is either dead or captured |
duke@435 | 2426 | // in the vframeArray. |
duke@435 | 2427 | |
duke@435 | 2428 | RegisterSaver::restore_result_registers(masm); |
duke@435 | 2429 | |
cfang@1361 | 2430 | // Non standard control word may be leaked out through a safepoint blob, and we can |
cfang@1361 | 2431 | // deopt at a poll point with the non standard control word. However, we should make |
cfang@1361 | 2432 | // sure the control word is correct after restore_result_registers. |
cfang@1361 | 2433 | __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); |
cfang@1361 | 2434 | |
duke@435 | 2435 | // All of the register save area has been popped of the stack. Only the |
duke@435 | 2436 | // return address remains. |
duke@435 | 2437 | |
duke@435 | 2438 | // Pop all the frames we must move/replace. |
duke@435 | 2439 | // |
duke@435 | 2440 | // Frame picture (youngest to oldest) |
duke@435 | 2441 | // 1: self-frame (no frame link) |
duke@435 | 2442 | // 2: deopting frame (no frame link) |
duke@435 | 2443 | // 3: caller of deopting frame (could be compiled/interpreted). |
duke@435 | 2444 | // |
duke@435 | 2445 | // Note: by leaving the return address of self-frame on the stack |
duke@435 | 2446 | // and using the size of frame 2 to adjust the stack |
duke@435 | 2447 | // when we are done the return to frame 3 will still be on the stack. |
duke@435 | 2448 | |
duke@435 | 2449 | // Pop deoptimized frame |
never@739 | 2450 | __ addptr(rsp, Address(rdi,Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes())); |
duke@435 | 2451 | |
duke@435 | 2452 | // sp should be pointing at the return address to the caller (3) |
duke@435 | 2453 | |
duke@435 | 2454 | // Stack bang to make sure there's enough room for these interpreter frames. |
duke@435 | 2455 | if (UseStackBanging) { |
duke@435 | 2456 | __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes())); |
duke@435 | 2457 | __ bang_stack_size(rbx, rcx); |
duke@435 | 2458 | } |
duke@435 | 2459 | |
duke@435 | 2460 | // Load array of frame pcs into ECX |
never@739 | 2461 | __ movptr(rcx,Address(rdi,Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes())); |
never@739 | 2462 | |
never@739 | 2463 | __ pop(rsi); // trash the old pc |
duke@435 | 2464 | |
duke@435 | 2465 | // Load array of frame sizes into ESI |
never@739 | 2466 | __ movptr(rsi,Address(rdi,Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes())); |
duke@435 | 2467 | |
duke@435 | 2468 | Address counter(rdi, Deoptimization::UnrollBlock::counter_temp_offset_in_bytes()); |
duke@435 | 2469 | |
duke@435 | 2470 | __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes())); |
duke@435 | 2471 | __ movl(counter, rbx); |
duke@435 | 2472 | |
duke@435 | 2473 | // Pick up the initial fp we should save |
never@739 | 2474 | __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_fp_offset_in_bytes())); |
duke@435 | 2475 | |
duke@435 | 2476 | // Now adjust the caller's stack to make up for the extra locals |
duke@435 | 2477 | // but record the original sp so that we can save it in the skeletal interpreter |
duke@435 | 2478 | // frame and the stack walking of interpreter_sender will get the unextended sp |
duke@435 | 2479 | // value and not the "real" sp value. |
duke@435 | 2480 | |
duke@435 | 2481 | Address sp_temp(rdi, Deoptimization::UnrollBlock::sender_sp_temp_offset_in_bytes()); |
never@739 | 2482 | __ movptr(sp_temp, rsp); |
never@739 | 2483 | __ movl2ptr(rbx, Address(rdi, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes())); |
never@739 | 2484 | __ subptr(rsp, rbx); |
duke@435 | 2485 | |
duke@435 | 2486 | // Push interpreter frames in a loop |
duke@435 | 2487 | Label loop; |
duke@435 | 2488 | __ bind(loop); |
never@739 | 2489 | __ movptr(rbx, Address(rsi, 0)); // Load frame size |
duke@435 | 2490 | #ifdef CC_INTERP |
never@739 | 2491 | __ subptr(rbx, 4*wordSize); // we'll push pc and ebp by hand and |
duke@435 | 2492 | #ifdef ASSERT |
never@739 | 2493 | __ push(0xDEADDEAD); // Make a recognizable pattern |
never@739 | 2494 | __ push(0xDEADDEAD); |
duke@435 | 2495 | #else /* ASSERT */ |
never@739 | 2496 | __ subptr(rsp, 2*wordSize); // skip the "static long no_param" |
duke@435 | 2497 | #endif /* ASSERT */ |
duke@435 | 2498 | #else /* CC_INTERP */ |
never@739 | 2499 | __ subptr(rbx, 2*wordSize); // we'll push pc and rbp, by hand |
duke@435 | 2500 | #endif /* CC_INTERP */ |
never@739 | 2501 | __ pushptr(Address(rcx, 0)); // save return address |
duke@435 | 2502 | __ enter(); // save old & set new rbp, |
never@739 | 2503 | __ subptr(rsp, rbx); // Prolog! |
never@739 | 2504 | __ movptr(rbx, sp_temp); // sender's sp |
duke@435 | 2505 | #ifdef CC_INTERP |
never@739 | 2506 | __ movptr(Address(rbp, |
duke@435 | 2507 | -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))), |
duke@435 | 2508 | rbx); // Make it walkable |
duke@435 | 2509 | #else /* CC_INTERP */ |
duke@435 | 2510 | // This value is corrected by layout_activation_impl |
xlu@947 | 2511 | __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); |
never@739 | 2512 | __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), rbx); // Make it walkable |
duke@435 | 2513 | #endif /* CC_INTERP */ |
never@739 | 2514 | __ movptr(sp_temp, rsp); // pass to next frame |
never@739 | 2515 | __ addptr(rsi, wordSize); // Bump array pointer (sizes) |
never@739 | 2516 | __ addptr(rcx, wordSize); // Bump array pointer (pcs) |
never@739 | 2517 | __ decrementl(counter); // decrement counter |
duke@435 | 2518 | __ jcc(Assembler::notZero, loop); |
never@739 | 2519 | __ pushptr(Address(rcx, 0)); // save final return address |
duke@435 | 2520 | |
duke@435 | 2521 | // Re-push self-frame |
duke@435 | 2522 | __ enter(); // save old & set new rbp, |
duke@435 | 2523 | |
duke@435 | 2524 | // Return address and rbp, are in place |
duke@435 | 2525 | // We'll push additional args later. Just allocate a full sized |
duke@435 | 2526 | // register save area |
never@739 | 2527 | __ subptr(rsp, (frame_size_in_words-additional_words - 2) * wordSize); |
duke@435 | 2528 | |
duke@435 | 2529 | // Restore frame locals after moving the frame |
never@739 | 2530 | __ movptr(Address(rsp, RegisterSaver::raxOffset()*wordSize), rax); |
never@739 | 2531 | __ movptr(Address(rsp, RegisterSaver::rdxOffset()*wordSize), rdx); |
duke@435 | 2532 | __ fstp_d(Address(rsp, RegisterSaver::fpResultOffset()*wordSize)); // Pop float stack and store in local |
duke@435 | 2533 | if( UseSSE>=2 ) __ movdbl(Address(rsp, RegisterSaver::xmm0Offset()*wordSize), xmm0); |
duke@435 | 2534 | if( UseSSE==1 ) __ movflt(Address(rsp, RegisterSaver::xmm0Offset()*wordSize), xmm0); |
duke@435 | 2535 | |
duke@435 | 2536 | // Set up the args to unpack_frame |
duke@435 | 2537 | |
duke@435 | 2538 | __ pushl(unpack_kind); // get the unpack_kind value |
duke@435 | 2539 | __ get_thread(rcx); |
never@739 | 2540 | __ push(rcx); |
duke@435 | 2541 | |
duke@435 | 2542 | // set last_Java_sp, last_Java_fp |
duke@435 | 2543 | __ set_last_Java_frame(rcx, noreg, rbp, NULL); |
duke@435 | 2544 | |
duke@435 | 2545 | // Call C code. Need thread but NOT official VM entry |
duke@435 | 2546 | // crud. We cannot block on this call, no GC can happen. Call should |
duke@435 | 2547 | // restore return values to their stack-slots with the new SP. |
duke@435 | 2548 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames))); |
duke@435 | 2549 | // Set an oopmap for the call site |
duke@435 | 2550 | oop_maps->add_gc_map( __ pc()-start, new OopMap( frame_size_in_words, 0 )); |
duke@435 | 2551 | |
duke@435 | 2552 | // rax, contains the return result type |
never@739 | 2553 | __ push(rax); |
duke@435 | 2554 | |
duke@435 | 2555 | __ get_thread(rcx); |
duke@435 | 2556 | __ reset_last_Java_frame(rcx, false, false); |
duke@435 | 2557 | |
duke@435 | 2558 | // Collect return values |
never@739 | 2559 | __ movptr(rax,Address(rsp, (RegisterSaver::raxOffset() + additional_words + 1)*wordSize)); |
never@739 | 2560 | __ movptr(rdx,Address(rsp, (RegisterSaver::rdxOffset() + additional_words + 1)*wordSize)); |
duke@435 | 2561 | |
duke@435 | 2562 | // Clear floating point stack before returning to interpreter |
duke@435 | 2563 | __ empty_FPU_stack(); |
duke@435 | 2564 | |
duke@435 | 2565 | // Check if we should push the float or double return value. |
duke@435 | 2566 | Label results_done, yes_double_value; |
duke@435 | 2567 | __ cmpl(Address(rsp, 0), T_DOUBLE); |
duke@435 | 2568 | __ jcc (Assembler::zero, yes_double_value); |
duke@435 | 2569 | __ cmpl(Address(rsp, 0), T_FLOAT); |
duke@435 | 2570 | __ jcc (Assembler::notZero, results_done); |
duke@435 | 2571 | |
duke@435 | 2572 | // return float value as expected by interpreter |
duke@435 | 2573 | if( UseSSE>=1 ) __ movflt(xmm0, Address(rsp, (RegisterSaver::xmm0Offset() + additional_words + 1)*wordSize)); |
duke@435 | 2574 | else __ fld_d(Address(rsp, (RegisterSaver::fpResultOffset() + additional_words + 1)*wordSize)); |
duke@435 | 2575 | __ jmp(results_done); |
duke@435 | 2576 | |
duke@435 | 2577 | // return double value as expected by interpreter |
duke@435 | 2578 | __ bind(yes_double_value); |
duke@435 | 2579 | if( UseSSE>=2 ) __ movdbl(xmm0, Address(rsp, (RegisterSaver::xmm0Offset() + additional_words + 1)*wordSize)); |
duke@435 | 2580 | else __ fld_d(Address(rsp, (RegisterSaver::fpResultOffset() + additional_words + 1)*wordSize)); |
duke@435 | 2581 | |
duke@435 | 2582 | __ bind(results_done); |
duke@435 | 2583 | |
duke@435 | 2584 | // Pop self-frame. |
duke@435 | 2585 | __ leave(); // Epilog! |
duke@435 | 2586 | |
duke@435 | 2587 | // Jump to interpreter |
duke@435 | 2588 | __ ret(0); |
duke@435 | 2589 | |
duke@435 | 2590 | // ------------- |
duke@435 | 2591 | // make sure all code is generated |
duke@435 | 2592 | masm->flush(); |
duke@435 | 2593 | |
duke@435 | 2594 | _deopt_blob = DeoptimizationBlob::create( &buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words); |
duke@435 | 2595 | _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset); |
duke@435 | 2596 | } |
duke@435 | 2597 | |
duke@435 | 2598 | |
duke@435 | 2599 | #ifdef COMPILER2 |
duke@435 | 2600 | //------------------------------generate_uncommon_trap_blob-------------------- |
duke@435 | 2601 | void SharedRuntime::generate_uncommon_trap_blob() { |
duke@435 | 2602 | // allocate space for the code |
duke@435 | 2603 | ResourceMark rm; |
duke@435 | 2604 | // setup code generation tools |
duke@435 | 2605 | CodeBuffer buffer("uncommon_trap_blob", 512, 512); |
duke@435 | 2606 | MacroAssembler* masm = new MacroAssembler(&buffer); |
duke@435 | 2607 | |
duke@435 | 2608 | enum frame_layout { |
duke@435 | 2609 | arg0_off, // thread sp + 0 // Arg location for |
duke@435 | 2610 | arg1_off, // unloaded_class_index sp + 1 // calling C |
duke@435 | 2611 | // The frame sender code expects that rbp will be in the "natural" place and |
duke@435 | 2612 | // will override any oopMap setting for it. We must therefore force the layout |
duke@435 | 2613 | // so that it agrees with the frame sender code. |
duke@435 | 2614 | rbp_off, // callee saved register sp + 2 |
duke@435 | 2615 | return_off, // slot for return address sp + 3 |
duke@435 | 2616 | framesize |
duke@435 | 2617 | }; |
duke@435 | 2618 | |
duke@435 | 2619 | address start = __ pc(); |
duke@435 | 2620 | // Push self-frame. |
never@739 | 2621 | __ subptr(rsp, return_off*wordSize); // Epilog! |
duke@435 | 2622 | |
duke@435 | 2623 | // rbp, is an implicitly saved callee saved register (i.e. the calling |
duke@435 | 2624 | // convention will save restore it in prolog/epilog) Other than that |
duke@435 | 2625 | // there are no callee save registers no that adapter frames are gone. |
never@739 | 2626 | __ movptr(Address(rsp, rbp_off*wordSize), rbp); |
duke@435 | 2627 | |
duke@435 | 2628 | // Clear the floating point exception stack |
duke@435 | 2629 | __ empty_FPU_stack(); |
duke@435 | 2630 | |
duke@435 | 2631 | // set last_Java_sp |
duke@435 | 2632 | __ get_thread(rdx); |
duke@435 | 2633 | __ set_last_Java_frame(rdx, noreg, noreg, NULL); |
duke@435 | 2634 | |
duke@435 | 2635 | // Call C code. Need thread but NOT official VM entry |
duke@435 | 2636 | // crud. We cannot block on this call, no GC can happen. Call should |
duke@435 | 2637 | // capture callee-saved registers as well as return values. |
never@739 | 2638 | __ movptr(Address(rsp, arg0_off*wordSize), rdx); |
duke@435 | 2639 | // argument already in ECX |
duke@435 | 2640 | __ movl(Address(rsp, arg1_off*wordSize),rcx); |
duke@435 | 2641 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap))); |
duke@435 | 2642 | |
duke@435 | 2643 | // Set an oopmap for the call site |
duke@435 | 2644 | OopMapSet *oop_maps = new OopMapSet(); |
duke@435 | 2645 | OopMap* map = new OopMap( framesize, 0 ); |
duke@435 | 2646 | // No oopMap for rbp, it is known implicitly |
duke@435 | 2647 | |
duke@435 | 2648 | oop_maps->add_gc_map( __ pc()-start, map); |
duke@435 | 2649 | |
duke@435 | 2650 | __ get_thread(rcx); |
duke@435 | 2651 | |
duke@435 | 2652 | __ reset_last_Java_frame(rcx, false, false); |
duke@435 | 2653 | |
duke@435 | 2654 | // Load UnrollBlock into EDI |
never@739 | 2655 | __ movptr(rdi, rax); |
duke@435 | 2656 | |
duke@435 | 2657 | // Pop all the frames we must move/replace. |
duke@435 | 2658 | // |
duke@435 | 2659 | // Frame picture (youngest to oldest) |
duke@435 | 2660 | // 1: self-frame (no frame link) |
duke@435 | 2661 | // 2: deopting frame (no frame link) |
duke@435 | 2662 | // 3: caller of deopting frame (could be compiled/interpreted). |
duke@435 | 2663 | |
duke@435 | 2664 | // Pop self-frame. We have no frame, and must rely only on EAX and ESP. |
never@739 | 2665 | __ addptr(rsp,(framesize-1)*wordSize); // Epilog! |
duke@435 | 2666 | |
duke@435 | 2667 | // Pop deoptimized frame |
never@739 | 2668 | __ movl2ptr(rcx, Address(rdi,Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes())); |
never@739 | 2669 | __ addptr(rsp, rcx); |
duke@435 | 2670 | |
duke@435 | 2671 | // sp should be pointing at the return address to the caller (3) |
duke@435 | 2672 | |
duke@435 | 2673 | // Stack bang to make sure there's enough room for these interpreter frames. |
duke@435 | 2674 | if (UseStackBanging) { |
duke@435 | 2675 | __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes())); |
duke@435 | 2676 | __ bang_stack_size(rbx, rcx); |
duke@435 | 2677 | } |
duke@435 | 2678 | |
duke@435 | 2679 | |
duke@435 | 2680 | // Load array of frame pcs into ECX |
duke@435 | 2681 | __ movl(rcx,Address(rdi,Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes())); |
duke@435 | 2682 | |
never@739 | 2683 | __ pop(rsi); // trash the pc |
duke@435 | 2684 | |
duke@435 | 2685 | // Load array of frame sizes into ESI |
never@739 | 2686 | __ movptr(rsi,Address(rdi,Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes())); |
duke@435 | 2687 | |
duke@435 | 2688 | Address counter(rdi, Deoptimization::UnrollBlock::counter_temp_offset_in_bytes()); |
duke@435 | 2689 | |
duke@435 | 2690 | __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes())); |
duke@435 | 2691 | __ movl(counter, rbx); |
duke@435 | 2692 | |
duke@435 | 2693 | // Pick up the initial fp we should save |
never@739 | 2694 | __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_fp_offset_in_bytes())); |
duke@435 | 2695 | |
duke@435 | 2696 | // Now adjust the caller's stack to make up for the extra locals |
duke@435 | 2697 | // but record the original sp so that we can save it in the skeletal interpreter |
duke@435 | 2698 | // frame and the stack walking of interpreter_sender will get the unextended sp |
duke@435 | 2699 | // value and not the "real" sp value. |
duke@435 | 2700 | |
duke@435 | 2701 | Address sp_temp(rdi, Deoptimization::UnrollBlock::sender_sp_temp_offset_in_bytes()); |
never@739 | 2702 | __ movptr(sp_temp, rsp); |
never@739 | 2703 | __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes())); |
never@739 | 2704 | __ subptr(rsp, rbx); |
duke@435 | 2705 | |
duke@435 | 2706 | // Push interpreter frames in a loop |
duke@435 | 2707 | Label loop; |
duke@435 | 2708 | __ bind(loop); |
never@739 | 2709 | __ movptr(rbx, Address(rsi, 0)); // Load frame size |
duke@435 | 2710 | #ifdef CC_INTERP |
never@739 | 2711 | __ subptr(rbx, 4*wordSize); // we'll push pc and ebp by hand and |
duke@435 | 2712 | #ifdef ASSERT |
never@739 | 2713 | __ push(0xDEADDEAD); // Make a recognizable pattern |
never@739 | 2714 | __ push(0xDEADDEAD); // (parm to RecursiveInterpreter...) |
duke@435 | 2715 | #else /* ASSERT */ |
never@739 | 2716 | __ subptr(rsp, 2*wordSize); // skip the "static long no_param" |
duke@435 | 2717 | #endif /* ASSERT */ |
duke@435 | 2718 | #else /* CC_INTERP */ |
never@739 | 2719 | __ subptr(rbx, 2*wordSize); // we'll push pc and rbp, by hand |
duke@435 | 2720 | #endif /* CC_INTERP */ |
never@739 | 2721 | __ pushptr(Address(rcx, 0)); // save return address |
duke@435 | 2722 | __ enter(); // save old & set new rbp, |
never@739 | 2723 | __ subptr(rsp, rbx); // Prolog! |
never@739 | 2724 | __ movptr(rbx, sp_temp); // sender's sp |
duke@435 | 2725 | #ifdef CC_INTERP |
never@739 | 2726 | __ movptr(Address(rbp, |
duke@435 | 2727 | -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))), |
duke@435 | 2728 | rbx); // Make it walkable |
duke@435 | 2729 | #else /* CC_INTERP */ |
duke@435 | 2730 | // This value is corrected by layout_activation_impl |
xlu@947 | 2731 | __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD ); |
never@739 | 2732 | __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), rbx); // Make it walkable |
duke@435 | 2733 | #endif /* CC_INTERP */ |
never@739 | 2734 | __ movptr(sp_temp, rsp); // pass to next frame |
never@739 | 2735 | __ addptr(rsi, wordSize); // Bump array pointer (sizes) |
never@739 | 2736 | __ addptr(rcx, wordSize); // Bump array pointer (pcs) |
never@739 | 2737 | __ decrementl(counter); // decrement counter |
duke@435 | 2738 | __ jcc(Assembler::notZero, loop); |
never@739 | 2739 | __ pushptr(Address(rcx, 0)); // save final return address |
duke@435 | 2740 | |
duke@435 | 2741 | // Re-push self-frame |
duke@435 | 2742 | __ enter(); // save old & set new rbp, |
never@739 | 2743 | __ subptr(rsp, (framesize-2) * wordSize); // Prolog! |
duke@435 | 2744 | |
duke@435 | 2745 | |
duke@435 | 2746 | // set last_Java_sp, last_Java_fp |
duke@435 | 2747 | __ get_thread(rdi); |
duke@435 | 2748 | __ set_last_Java_frame(rdi, noreg, rbp, NULL); |
duke@435 | 2749 | |
duke@435 | 2750 | // Call C code. Need thread but NOT official VM entry |
duke@435 | 2751 | // crud. We cannot block on this call, no GC can happen. Call should |
duke@435 | 2752 | // restore return values to their stack-slots with the new SP. |
never@739 | 2753 | __ movptr(Address(rsp,arg0_off*wordSize),rdi); |
duke@435 | 2754 | __ movl(Address(rsp,arg1_off*wordSize), Deoptimization::Unpack_uncommon_trap); |
duke@435 | 2755 | __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames))); |
duke@435 | 2756 | // Set an oopmap for the call site |
duke@435 | 2757 | oop_maps->add_gc_map( __ pc()-start, new OopMap( framesize, 0 ) ); |
duke@435 | 2758 | |
duke@435 | 2759 | __ get_thread(rdi); |
duke@435 | 2760 | __ reset_last_Java_frame(rdi, true, false); |
duke@435 | 2761 | |
duke@435 | 2762 | // Pop self-frame. |
duke@435 | 2763 | __ leave(); // Epilog! |
duke@435 | 2764 | |
duke@435 | 2765 | // Jump to interpreter |
duke@435 | 2766 | __ ret(0); |
duke@435 | 2767 | |
duke@435 | 2768 | // ------------- |
duke@435 | 2769 | // make sure all code is generated |
duke@435 | 2770 | masm->flush(); |
duke@435 | 2771 | |
duke@435 | 2772 | _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps, framesize); |
duke@435 | 2773 | } |
duke@435 | 2774 | #endif // COMPILER2 |
duke@435 | 2775 | |
duke@435 | 2776 | //------------------------------generate_handler_blob------ |
duke@435 | 2777 | // |
duke@435 | 2778 | // Generate a special Compile2Runtime blob that saves all registers, |
duke@435 | 2779 | // setup oopmap, and calls safepoint code to stop the compiled code for |
duke@435 | 2780 | // a safepoint. |
duke@435 | 2781 | // |
never@2950 | 2782 | SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, bool cause_return) { |
duke@435 | 2783 | |
duke@435 | 2784 | // Account for thread arg in our frame |
duke@435 | 2785 | const int additional_words = 1; |
duke@435 | 2786 | int frame_size_in_words; |
duke@435 | 2787 | |
duke@435 | 2788 | assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); |
duke@435 | 2789 | |
duke@435 | 2790 | ResourceMark rm; |
duke@435 | 2791 | OopMapSet *oop_maps = new OopMapSet(); |
duke@435 | 2792 | OopMap* map; |
duke@435 | 2793 | |
duke@435 | 2794 | // allocate space for the code |
duke@435 | 2795 | // setup code generation tools |
duke@435 | 2796 | CodeBuffer buffer("handler_blob", 1024, 512); |
duke@435 | 2797 | MacroAssembler* masm = new MacroAssembler(&buffer); |
duke@435 | 2798 | |
duke@435 | 2799 | const Register java_thread = rdi; // callee-saved for VC++ |
duke@435 | 2800 | address start = __ pc(); |
duke@435 | 2801 | address call_pc = NULL; |
duke@435 | 2802 | |
duke@435 | 2803 | // If cause_return is true we are at a poll_return and there is |
duke@435 | 2804 | // the return address on the stack to the caller on the nmethod |
duke@435 | 2805 | // that is safepoint. We can leave this return on the stack and |
duke@435 | 2806 | // effectively complete the return and safepoint in the caller. |
duke@435 | 2807 | // Otherwise we push space for a return address that the safepoint |
duke@435 | 2808 | // handler will install later to make the stack walking sensible. |
duke@435 | 2809 | if( !cause_return ) |
never@739 | 2810 | __ push(rbx); // Make room for return address (or push it again) |
duke@435 | 2811 | |
duke@435 | 2812 | map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false); |
duke@435 | 2813 | |
duke@435 | 2814 | // The following is basically a call_VM. However, we need the precise |
duke@435 | 2815 | // address of the call in order to generate an oopmap. Hence, we do all the |
duke@435 | 2816 | // work ourselves. |
duke@435 | 2817 | |
duke@435 | 2818 | // Push thread argument and setup last_Java_sp |
duke@435 | 2819 | __ get_thread(java_thread); |
never@739 | 2820 | __ push(java_thread); |
duke@435 | 2821 | __ set_last_Java_frame(java_thread, noreg, noreg, NULL); |
duke@435 | 2822 | |
duke@435 | 2823 | // if this was not a poll_return then we need to correct the return address now. |
duke@435 | 2824 | if( !cause_return ) { |
never@739 | 2825 | __ movptr(rax, Address(java_thread, JavaThread::saved_exception_pc_offset())); |
never@739 | 2826 | __ movptr(Address(rbp, wordSize), rax); |
duke@435 | 2827 | } |
duke@435 | 2828 | |
duke@435 | 2829 | // do the call |
duke@435 | 2830 | __ call(RuntimeAddress(call_ptr)); |
duke@435 | 2831 | |
duke@435 | 2832 | // Set an oopmap for the call site. This oopmap will map all |
duke@435 | 2833 | // oop-registers and debug-info registers as callee-saved. This |
duke@435 | 2834 | // will allow deoptimization at this safepoint to find all possible |
duke@435 | 2835 | // debug-info recordings, as well as let GC find all oops. |
duke@435 | 2836 | |
duke@435 | 2837 | oop_maps->add_gc_map( __ pc() - start, map); |
duke@435 | 2838 | |
duke@435 | 2839 | // Discard arg |
never@739 | 2840 | __ pop(rcx); |
duke@435 | 2841 | |
duke@435 | 2842 | Label noException; |
duke@435 | 2843 | |
duke@435 | 2844 | // Clear last_Java_sp again |
duke@435 | 2845 | __ get_thread(java_thread); |
duke@435 | 2846 | __ reset_last_Java_frame(java_thread, false, false); |
duke@435 | 2847 | |
never@739 | 2848 | __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
duke@435 | 2849 | __ jcc(Assembler::equal, noException); |
duke@435 | 2850 | |
duke@435 | 2851 | // Exception pending |
duke@435 | 2852 | |
duke@435 | 2853 | RegisterSaver::restore_live_registers(masm); |
duke@435 | 2854 | |
duke@435 | 2855 | __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
duke@435 | 2856 | |
duke@435 | 2857 | __ bind(noException); |
duke@435 | 2858 | |
duke@435 | 2859 | // Normal exit, register restoring and exit |
duke@435 | 2860 | RegisterSaver::restore_live_registers(masm); |
duke@435 | 2861 | |
duke@435 | 2862 | __ ret(0); |
duke@435 | 2863 | |
duke@435 | 2864 | // make sure all code is generated |
duke@435 | 2865 | masm->flush(); |
duke@435 | 2866 | |
duke@435 | 2867 | // Fill-out other meta info |
duke@435 | 2868 | return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words); |
duke@435 | 2869 | } |
duke@435 | 2870 | |
duke@435 | 2871 | // |
duke@435 | 2872 | // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss |
duke@435 | 2873 | // |
duke@435 | 2874 | // Generate a stub that calls into vm to find out the proper destination |
duke@435 | 2875 | // of a java call. All the argument registers are live at this point |
duke@435 | 2876 | // but since this is generic code we don't know what they are and the caller |
duke@435 | 2877 | // must do any gc of the args. |
duke@435 | 2878 | // |
never@2950 | 2879 | RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) { |
duke@435 | 2880 | assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before"); |
duke@435 | 2881 | |
duke@435 | 2882 | // allocate space for the code |
duke@435 | 2883 | ResourceMark rm; |
duke@435 | 2884 | |
duke@435 | 2885 | CodeBuffer buffer(name, 1000, 512); |
duke@435 | 2886 | MacroAssembler* masm = new MacroAssembler(&buffer); |
duke@435 | 2887 | |
duke@435 | 2888 | int frame_size_words; |
duke@435 | 2889 | enum frame_layout { |
duke@435 | 2890 | thread_off, |
duke@435 | 2891 | extra_words }; |
duke@435 | 2892 | |
duke@435 | 2893 | OopMapSet *oop_maps = new OopMapSet(); |
duke@435 | 2894 | OopMap* map = NULL; |
duke@435 | 2895 | |
duke@435 | 2896 | int start = __ offset(); |
duke@435 | 2897 | |
duke@435 | 2898 | map = RegisterSaver::save_live_registers(masm, extra_words, &frame_size_words); |
duke@435 | 2899 | |
duke@435 | 2900 | int frame_complete = __ offset(); |
duke@435 | 2901 | |
duke@435 | 2902 | const Register thread = rdi; |
duke@435 | 2903 | __ get_thread(rdi); |
duke@435 | 2904 | |
never@739 | 2905 | __ push(thread); |
duke@435 | 2906 | __ set_last_Java_frame(thread, noreg, rbp, NULL); |
duke@435 | 2907 | |
duke@435 | 2908 | __ call(RuntimeAddress(destination)); |
duke@435 | 2909 | |
duke@435 | 2910 | |
duke@435 | 2911 | // Set an oopmap for the call site. |
duke@435 | 2912 | // We need this not only for callee-saved registers, but also for volatile |
duke@435 | 2913 | // registers that the compiler might be keeping live across a safepoint. |
duke@435 | 2914 | |
duke@435 | 2915 | oop_maps->add_gc_map( __ offset() - start, map); |
duke@435 | 2916 | |
duke@435 | 2917 | // rax, contains the address we are going to jump to assuming no exception got installed |
duke@435 | 2918 | |
never@739 | 2919 | __ addptr(rsp, wordSize); |
duke@435 | 2920 | |
duke@435 | 2921 | // clear last_Java_sp |
duke@435 | 2922 | __ reset_last_Java_frame(thread, true, false); |
duke@435 | 2923 | // check for pending exceptions |
duke@435 | 2924 | Label pending; |
never@739 | 2925 | __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
duke@435 | 2926 | __ jcc(Assembler::notEqual, pending); |
duke@435 | 2927 | |
duke@435 | 2928 | // get the returned methodOop |
never@739 | 2929 | __ movptr(rbx, Address(thread, JavaThread::vm_result_offset())); |
never@739 | 2930 | __ movptr(Address(rsp, RegisterSaver::rbx_offset() * wordSize), rbx); |
never@739 | 2931 | |
never@739 | 2932 | __ movptr(Address(rsp, RegisterSaver::rax_offset() * wordSize), rax); |
duke@435 | 2933 | |
duke@435 | 2934 | RegisterSaver::restore_live_registers(masm); |
duke@435 | 2935 | |
duke@435 | 2936 | // We are back the the original state on entry and ready to go. |
duke@435 | 2937 | |
duke@435 | 2938 | __ jmp(rax); |
duke@435 | 2939 | |
duke@435 | 2940 | // Pending exception after the safepoint |
duke@435 | 2941 | |
duke@435 | 2942 | __ bind(pending); |
duke@435 | 2943 | |
duke@435 | 2944 | RegisterSaver::restore_live_registers(masm); |
duke@435 | 2945 | |
duke@435 | 2946 | // exception pending => remove activation and forward to exception handler |
duke@435 | 2947 | |
duke@435 | 2948 | __ get_thread(thread); |
xlu@947 | 2949 | __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD); |
never@739 | 2950 | __ movptr(rax, Address(thread, Thread::pending_exception_offset())); |
duke@435 | 2951 | __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
duke@435 | 2952 | |
duke@435 | 2953 | // ------------- |
duke@435 | 2954 | // make sure all code is generated |
duke@435 | 2955 | masm->flush(); |
duke@435 | 2956 | |
duke@435 | 2957 | // return the blob |
duke@435 | 2958 | // frame_size_words or bytes?? |
duke@435 | 2959 | return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_words, oop_maps, true); |
duke@435 | 2960 | } |