annotate src/cpu/x86/vm/cppInterpreter_x86.cpp @ 1472:c18cbe5936b8

6941466: Oracle rebranding changes for Hotspot repositories Summary: Change all the Sun copyrights to Oracle copyright Reviewed-by: ohair
author trims
date Thu, 27 May 2010 19:08:38 -0700
parents e5b0439ef4ae
children f95d63e2154a
rev   line source
duke@0 1 /*
trims@1472 2 * Copyright (c) 2007, 2009, Oracle and/or its affiliates. All rights reserved.
duke@0 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@0 4 *
duke@0 5 * This code is free software; you can redistribute it and/or modify it
duke@0 6 * under the terms of the GNU General Public License version 2 only, as
duke@0 7 * published by the Free Software Foundation.
duke@0 8 *
duke@0 9 * This code is distributed in the hope that it will be useful, but WITHOUT
duke@0 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@0 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
duke@0 12 * version 2 for more details (a copy is included in the LICENSE file that
duke@0 13 * accompanied this code).
duke@0 14 *
duke@0 15 * You should have received a copy of the GNU General Public License version
duke@0 16 * 2 along with this work; if not, write to the Free Software Foundation,
duke@0 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@0 18 *
trims@1472 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1472 20 * or visit www.oracle.com if you need additional information or have any
trims@1472 21 * questions.
duke@0 22 *
duke@0 23 */
duke@0 24
duke@0 25 #include "incls/_precompiled.incl"
duke@0 26 #include "incls/_cppInterpreter_x86.cpp.incl"
duke@0 27
duke@0 28 #ifdef CC_INTERP
duke@0 29
duke@0 30 // Routine exists to make tracebacks look decent in debugger
duke@0 31 // while we are recursed in the frame manager/c++ interpreter.
duke@0 32 // We could use an address in the frame manager but having
duke@0 33 // frames look natural in the debugger is a plus.
duke@0 34 extern "C" void RecursiveInterpreterActivation(interpreterState istate )
duke@0 35 {
duke@0 36 //
duke@0 37 ShouldNotReachHere();
duke@0 38 }
duke@0 39
duke@0 40
duke@0 41 #define __ _masm->
duke@0 42 #define STATE(field_name) (Address(state, byte_offset_of(BytecodeInterpreter, field_name)))
duke@0 43
duke@0 44 Label fast_accessor_slow_entry_path; // fast accessor methods need to be able to jmp to unsynchronized
duke@0 45 // c++ interpreter entry point this holds that entry point label.
duke@0 46
never@304 47 // default registers for state and sender_sp
never@304 48 // state and sender_sp are the same on 32bit because we have no choice.
never@304 49 // state could be rsi on 64bit but it is an arg reg and not callee save
never@304 50 // so r13 is better choice.
never@304 51
never@304 52 const Register state = NOT_LP64(rsi) LP64_ONLY(r13);
never@304 53 const Register sender_sp_on_entry = NOT_LP64(rsi) LP64_ONLY(r13);
never@304 54
duke@0 55 // NEEDED for JVMTI?
duke@0 56 // address AbstractInterpreter::_remove_activation_preserving_args_entry;
duke@0 57
duke@0 58 static address unctrap_frame_manager_entry = NULL;
duke@0 59
duke@0 60 static address deopt_frame_manager_return_atos = NULL;
duke@0 61 static address deopt_frame_manager_return_btos = NULL;
duke@0 62 static address deopt_frame_manager_return_itos = NULL;
duke@0 63 static address deopt_frame_manager_return_ltos = NULL;
duke@0 64 static address deopt_frame_manager_return_ftos = NULL;
duke@0 65 static address deopt_frame_manager_return_dtos = NULL;
duke@0 66 static address deopt_frame_manager_return_vtos = NULL;
duke@0 67
duke@0 68 int AbstractInterpreter::BasicType_as_index(BasicType type) {
duke@0 69 int i = 0;
duke@0 70 switch (type) {
duke@0 71 case T_BOOLEAN: i = 0; break;
duke@0 72 case T_CHAR : i = 1; break;
duke@0 73 case T_BYTE : i = 2; break;
duke@0 74 case T_SHORT : i = 3; break;
duke@0 75 case T_INT : i = 4; break;
duke@0 76 case T_VOID : i = 5; break;
duke@0 77 case T_FLOAT : i = 8; break;
duke@0 78 case T_LONG : i = 9; break;
duke@0 79 case T_DOUBLE : i = 6; break;
duke@0 80 case T_OBJECT : // fall through
duke@0 81 case T_ARRAY : i = 7; break;
duke@0 82 default : ShouldNotReachHere();
duke@0 83 }
duke@0 84 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
duke@0 85 return i;
duke@0 86 }
duke@0 87
duke@0 88 // Is this pc anywhere within code owned by the interpreter?
duke@0 89 // This only works for pc that might possibly be exposed to frame
duke@0 90 // walkers. It clearly misses all of the actual c++ interpreter
duke@0 91 // implementation
duke@0 92 bool CppInterpreter::contains(address pc) {
duke@0 93 return (_code->contains(pc) ||
duke@0 94 pc == CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation));
duke@0 95 }
duke@0 96
duke@0 97
duke@0 98 address CppInterpreterGenerator::generate_result_handler_for(BasicType type) {
duke@0 99 address entry = __ pc();
duke@0 100 switch (type) {
duke@0 101 case T_BOOLEAN: __ c2bool(rax); break;
duke@0 102 case T_CHAR : __ andl(rax, 0xFFFF); break;
duke@0 103 case T_BYTE : __ sign_extend_byte (rax); break;
duke@0 104 case T_SHORT : __ sign_extend_short(rax); break;
duke@0 105 case T_VOID : // fall thru
duke@0 106 case T_LONG : // fall thru
duke@0 107 case T_INT : /* nothing to do */ break;
never@304 108
duke@0 109 case T_DOUBLE :
duke@0 110 case T_FLOAT :
never@304 111 {
never@304 112 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
never@304 113 __ pop(t); // remove return address first
duke@0 114 // Must return a result for interpreter or compiler. In SSE
duke@0 115 // mode, results are returned in xmm0 and the FPU stack must
duke@0 116 // be empty.
duke@0 117 if (type == T_FLOAT && UseSSE >= 1) {
never@304 118 #ifndef _LP64
duke@0 119 // Load ST0
duke@0 120 __ fld_d(Address(rsp, 0));
duke@0 121 // Store as float and empty fpu stack
duke@0 122 __ fstp_s(Address(rsp, 0));
never@304 123 #endif // !_LP64
duke@0 124 // and reload
duke@0 125 __ movflt(xmm0, Address(rsp, 0));
duke@0 126 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
duke@0 127 __ movdbl(xmm0, Address(rsp, 0));
duke@0 128 } else {
duke@0 129 // restore ST0
duke@0 130 __ fld_d(Address(rsp, 0));
duke@0 131 }
duke@0 132 // and pop the temp
never@304 133 __ addptr(rsp, 2 * wordSize);
never@304 134 __ push(t); // restore return address
duke@0 135 }
duke@0 136 break;
duke@0 137 case T_OBJECT :
duke@0 138 // retrieve result from frame
never@304 139 __ movptr(rax, STATE(_oop_temp));
duke@0 140 // and verify it
duke@0 141 __ verify_oop(rax);
duke@0 142 break;
duke@0 143 default : ShouldNotReachHere();
duke@0 144 }
duke@0 145 __ ret(0); // return from result handler
duke@0 146 return entry;
duke@0 147 }
duke@0 148
duke@0 149 // tosca based result to c++ interpreter stack based result.
duke@0 150 // Result goes to top of native stack.
duke@0 151
duke@0 152 #undef EXTEND // SHOULD NOT BE NEEDED
duke@0 153 address CppInterpreterGenerator::generate_tosca_to_stack_converter(BasicType type) {
duke@0 154 // A result is in the tosca (abi result) from either a native method call or compiled
duke@0 155 // code. Place this result on the java expression stack so C++ interpreter can use it.
duke@0 156 address entry = __ pc();
duke@0 157
duke@0 158 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
never@304 159 __ pop(t); // remove return address first
duke@0 160 switch (type) {
duke@0 161 case T_VOID:
duke@0 162 break;
duke@0 163 case T_BOOLEAN:
duke@0 164 #ifdef EXTEND
duke@0 165 __ c2bool(rax);
duke@0 166 #endif
never@304 167 __ push(rax);
duke@0 168 break;
duke@0 169 case T_CHAR :
duke@0 170 #ifdef EXTEND
duke@0 171 __ andl(rax, 0xFFFF);
duke@0 172 #endif
never@304 173 __ push(rax);
duke@0 174 break;
duke@0 175 case T_BYTE :
duke@0 176 #ifdef EXTEND
duke@0 177 __ sign_extend_byte (rax);
duke@0 178 #endif
never@304 179 __ push(rax);
duke@0 180 break;
duke@0 181 case T_SHORT :
duke@0 182 #ifdef EXTEND
duke@0 183 __ sign_extend_short(rax);
duke@0 184 #endif
never@304 185 __ push(rax);
duke@0 186 break;
duke@0 187 case T_LONG :
never@304 188 __ push(rdx); // pushes useless junk on 64bit
never@304 189 __ push(rax);
duke@0 190 break;
duke@0 191 case T_INT :
never@304 192 __ push(rax);
duke@0 193 break;
duke@0 194 case T_FLOAT :
never@304 195 // Result is in ST(0)/xmm0
never@304 196 __ subptr(rsp, wordSize);
duke@0 197 if ( UseSSE < 1) {
never@304 198 __ fstp_s(Address(rsp, 0));
duke@0 199 } else {
duke@0 200 __ movflt(Address(rsp, 0), xmm0);
duke@0 201 }
duke@0 202 break;
duke@0 203 case T_DOUBLE :
never@304 204 __ subptr(rsp, 2*wordSize);
duke@0 205 if ( UseSSE < 2 ) {
never@304 206 __ fstp_d(Address(rsp, 0));
duke@0 207 } else {
duke@0 208 __ movdbl(Address(rsp, 0), xmm0);
duke@0 209 }
duke@0 210 break;
duke@0 211 case T_OBJECT :
duke@0 212 __ verify_oop(rax); // verify it
never@304 213 __ push(rax);
duke@0 214 break;
duke@0 215 default : ShouldNotReachHere();
duke@0 216 }
duke@0 217 __ jmp(t); // return from result handler
duke@0 218 return entry;
duke@0 219 }
duke@0 220
duke@0 221 address CppInterpreterGenerator::generate_stack_to_stack_converter(BasicType type) {
duke@0 222 // A result is in the java expression stack of the interpreted method that has just
duke@0 223 // returned. Place this result on the java expression stack of the caller.
duke@0 224 //
never@304 225 // The current interpreter activation in rsi/r13 is for the method just returning its
duke@0 226 // result. So we know that the result of this method is on the top of the current
duke@0 227 // execution stack (which is pre-pushed) and will be return to the top of the caller
duke@0 228 // stack. The top of the callers stack is the bottom of the locals of the current
duke@0 229 // activation.
duke@0 230 // Because of the way activation are managed by the frame manager the value of rsp is
duke@0 231 // below both the stack top of the current activation and naturally the stack top
duke@0 232 // of the calling activation. This enable this routine to leave the return address
duke@0 233 // to the frame manager on the stack and do a vanilla return.
duke@0 234 //
never@304 235 // On entry: rsi/r13 - interpreter state of activation returning a (potential) result
never@304 236 // On Return: rsi/r13 - unchanged
duke@0 237 // rax - new stack top for caller activation (i.e. activation in _prev_link)
duke@0 238 //
duke@0 239 // Can destroy rdx, rcx.
duke@0 240 //
duke@0 241
duke@0 242 address entry = __ pc();
duke@0 243 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
duke@0 244 switch (type) {
duke@0 245 case T_VOID:
never@304 246 __ movptr(rax, STATE(_locals)); // pop parameters get new stack value
never@304 247 __ addptr(rax, wordSize); // account for prepush before we return
duke@0 248 break;
duke@0 249 case T_FLOAT :
duke@0 250 case T_BOOLEAN:
duke@0 251 case T_CHAR :
duke@0 252 case T_BYTE :
duke@0 253 case T_SHORT :
duke@0 254 case T_INT :
duke@0 255 // 1 word result
never@304 256 __ movptr(rdx, STATE(_stack));
never@304 257 __ movptr(rax, STATE(_locals)); // address for result
duke@0 258 __ movl(rdx, Address(rdx, wordSize)); // get result
never@304 259 __ movptr(Address(rax, 0), rdx); // and store it
duke@0 260 break;
duke@0 261 case T_LONG :
duke@0 262 case T_DOUBLE :
duke@0 263 // return top two words on current expression stack to caller's expression stack
duke@0 264 // The caller's expression stack is adjacent to the current frame manager's intepretState
duke@0 265 // except we allocated one extra word for this intepretState so we won't overwrite it
duke@0 266 // when we return a two word result.
duke@0 267
never@304 268 __ movptr(rax, STATE(_locals)); // address for result
never@304 269 __ movptr(rcx, STATE(_stack));
never@304 270 __ subptr(rax, wordSize); // need addition word besides locals[0]
never@304 271 __ movptr(rdx, Address(rcx, 2*wordSize)); // get result word (junk in 64bit)
never@304 272 __ movptr(Address(rax, wordSize), rdx); // and store it
never@304 273 __ movptr(rdx, Address(rcx, wordSize)); // get result word
never@304 274 __ movptr(Address(rax, 0), rdx); // and store it
duke@0 275 break;
duke@0 276 case T_OBJECT :
never@304 277 __ movptr(rdx, STATE(_stack));
never@304 278 __ movptr(rax, STATE(_locals)); // address for result
never@304 279 __ movptr(rdx, Address(rdx, wordSize)); // get result
duke@0 280 __ verify_oop(rdx); // verify it
never@304 281 __ movptr(Address(rax, 0), rdx); // and store it
duke@0 282 break;
duke@0 283 default : ShouldNotReachHere();
duke@0 284 }
duke@0 285 __ ret(0);
duke@0 286 return entry;
duke@0 287 }
duke@0 288
duke@0 289 address CppInterpreterGenerator::generate_stack_to_native_abi_converter(BasicType type) {
duke@0 290 // A result is in the java expression stack of the interpreted method that has just
duke@0 291 // returned. Place this result in the native abi that the caller expects.
duke@0 292 //
duke@0 293 // Similar to generate_stack_to_stack_converter above. Called at a similar time from the
duke@0 294 // frame manager execept in this situation the caller is native code (c1/c2/call_stub)
duke@0 295 // and so rather than return result onto caller's java expression stack we return the
duke@0 296 // result in the expected location based on the native abi.
never@304 297 // On entry: rsi/r13 - interpreter state of activation returning a (potential) result
never@304 298 // On Return: rsi/r13 - unchanged
duke@0 299 // Other registers changed [rax/rdx/ST(0) as needed for the result returned]
duke@0 300
duke@0 301 address entry = __ pc();
duke@0 302 switch (type) {
duke@0 303 case T_VOID:
duke@0 304 break;
duke@0 305 case T_BOOLEAN:
duke@0 306 case T_CHAR :
duke@0 307 case T_BYTE :
duke@0 308 case T_SHORT :
duke@0 309 case T_INT :
never@304 310 __ movptr(rdx, STATE(_stack)); // get top of stack
duke@0 311 __ movl(rax, Address(rdx, wordSize)); // get result word 1
duke@0 312 break;
duke@0 313 case T_LONG :
never@304 314 __ movptr(rdx, STATE(_stack)); // get top of stack
never@304 315 __ movptr(rax, Address(rdx, wordSize)); // get result low word
never@304 316 NOT_LP64(__ movl(rdx, Address(rdx, 2*wordSize));) // get result high word
duke@0 317 break;
duke@0 318 case T_FLOAT :
never@304 319 __ movptr(rdx, STATE(_stack)); // get top of stack
duke@0 320 if ( UseSSE >= 1) {
duke@0 321 __ movflt(xmm0, Address(rdx, wordSize));
duke@0 322 } else {
duke@0 323 __ fld_s(Address(rdx, wordSize)); // pushd float result
duke@0 324 }
duke@0 325 break;
duke@0 326 case T_DOUBLE :
never@304 327 __ movptr(rdx, STATE(_stack)); // get top of stack
duke@0 328 if ( UseSSE > 1) {
duke@0 329 __ movdbl(xmm0, Address(rdx, wordSize));
duke@0 330 } else {
duke@0 331 __ fld_d(Address(rdx, wordSize)); // push double result
duke@0 332 }
duke@0 333 break;
duke@0 334 case T_OBJECT :
never@304 335 __ movptr(rdx, STATE(_stack)); // get top of stack
never@304 336 __ movptr(rax, Address(rdx, wordSize)); // get result word 1
duke@0 337 __ verify_oop(rax); // verify it
duke@0 338 break;
duke@0 339 default : ShouldNotReachHere();
duke@0 340 }
duke@0 341 __ ret(0);
duke@0 342 return entry;
duke@0 343 }
duke@0 344
duke@0 345 address CppInterpreter::return_entry(TosState state, int length) {
duke@0 346 // make it look good in the debugger
duke@0 347 return CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation);
duke@0 348 }
duke@0 349
duke@0 350 address CppInterpreter::deopt_entry(TosState state, int length) {
duke@0 351 address ret = NULL;
duke@0 352 if (length != 0) {
duke@0 353 switch (state) {
duke@0 354 case atos: ret = deopt_frame_manager_return_atos; break;
duke@0 355 case btos: ret = deopt_frame_manager_return_btos; break;
duke@0 356 case ctos:
duke@0 357 case stos:
duke@0 358 case itos: ret = deopt_frame_manager_return_itos; break;
duke@0 359 case ltos: ret = deopt_frame_manager_return_ltos; break;
duke@0 360 case ftos: ret = deopt_frame_manager_return_ftos; break;
duke@0 361 case dtos: ret = deopt_frame_manager_return_dtos; break;
duke@0 362 case vtos: ret = deopt_frame_manager_return_vtos; break;
duke@0 363 }
duke@0 364 } else {
duke@0 365 ret = unctrap_frame_manager_entry; // re-execute the bytecode ( e.g. uncommon trap)
duke@0 366 }
duke@0 367 assert(ret != NULL, "Not initialized");
duke@0 368 return ret;
duke@0 369 }
duke@0 370
duke@0 371 // C++ Interpreter
duke@0 372 void CppInterpreterGenerator::generate_compute_interpreter_state(const Register state,
duke@0 373 const Register locals,
duke@0 374 const Register sender_sp,
duke@0 375 bool native) {
duke@0 376
duke@0 377 // On entry the "locals" argument points to locals[0] (or where it would be in case no locals in
duke@0 378 // a static method). "state" contains any previous frame manager state which we must save a link
duke@0 379 // to in the newly generated state object. On return "state" is a pointer to the newly allocated
duke@0 380 // state object. We must allocate and initialize a new interpretState object and the method
duke@0 381 // expression stack. Because the returned result (if any) of the method will be placed on the caller's
duke@0 382 // expression stack and this will overlap with locals[0] (and locals[1] if double/long) we must
duke@0 383 // be sure to leave space on the caller's stack so that this result will not overwrite values when
duke@0 384 // locals[0] and locals[1] do not exist (and in fact are return address and saved rbp). So when
duke@0 385 // we are non-native we in essence ensure that locals[0-1] exist. We play an extra trick in
duke@0 386 // non-product builds and initialize this last local with the previous interpreterState as
duke@0 387 // this makes things look real nice in the debugger.
duke@0 388
duke@0 389 // State on entry
duke@0 390 // Assumes locals == &locals[0]
duke@0 391 // Assumes state == any previous frame manager state (assuming call path from c++ interpreter)
duke@0 392 // Assumes rax = return address
duke@0 393 // rcx == senders_sp
duke@0 394 // rbx == method
duke@0 395 // Modifies rcx, rdx, rax
duke@0 396 // Returns:
duke@0 397 // state == address of new interpreterState
duke@0 398 // rsp == bottom of method's expression stack.
duke@0 399
duke@0 400 const Address const_offset (rbx, methodOopDesc::const_offset());
duke@0 401
duke@0 402
duke@0 403 // On entry sp is the sender's sp. This includes the space for the arguments
duke@0 404 // that the sender pushed. If the sender pushed no args (a static) and the
duke@0 405 // caller returns a long then we need two words on the sender's stack which
duke@0 406 // are not present (although when we return a restore full size stack the
duke@0 407 // space will be present). If we didn't allocate two words here then when
duke@0 408 // we "push" the result of the caller's stack we would overwrite the return
duke@0 409 // address and the saved rbp. Not good. So simply allocate 2 words now
duke@0 410 // just to be safe. This is the "static long no_params() method" issue.
duke@0 411 // See Lo.java for a testcase.
duke@0 412 // We don't need this for native calls because they return result in
duke@0 413 // register and the stack is expanded in the caller before we store
duke@0 414 // the results on the stack.
duke@0 415
duke@0 416 if (!native) {
duke@0 417 #ifdef PRODUCT
never@304 418 __ subptr(rsp, 2*wordSize);
duke@0 419 #else /* PRODUCT */
never@304 420 __ push((int32_t)NULL_WORD);
never@304 421 __ push(state); // make it look like a real argument
duke@0 422 #endif /* PRODUCT */
duke@0 423 }
duke@0 424
duke@0 425 // Now that we are assure of space for stack result, setup typical linkage
duke@0 426
never@304 427 __ push(rax);
duke@0 428 __ enter();
duke@0 429
never@304 430 __ mov(rax, state); // save current state
never@304 431
never@304 432 __ lea(rsp, Address(rsp, -(int)sizeof(BytecodeInterpreter)));
never@304 433 __ mov(state, rsp);
never@304 434
never@304 435 // rsi/r13 == state/locals rax == prevstate
duke@0 436
duke@0 437 // initialize the "shadow" frame so that use since C++ interpreter not directly
duke@0 438 // recursive. Simpler to recurse but we can't trim expression stack as we call
duke@0 439 // new methods.
never@304 440 __ movptr(STATE(_locals), locals); // state->_locals = locals()
never@304 441 __ movptr(STATE(_self_link), state); // point to self
never@304 442 __ movptr(STATE(_prev_link), rax); // state->_link = state on entry (NULL or previous state)
never@304 443 __ movptr(STATE(_sender_sp), sender_sp); // state->_sender_sp = sender_sp
never@304 444 #ifdef _LP64
never@304 445 __ movptr(STATE(_thread), r15_thread); // state->_bcp = codes()
never@304 446 #else
duke@0 447 __ get_thread(rax); // get vm's javathread*
never@304 448 __ movptr(STATE(_thread), rax); // state->_bcp = codes()
never@304 449 #endif // _LP64
never@304 450 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset())); // get constantMethodOop
never@304 451 __ lea(rdx, Address(rdx, constMethodOopDesc::codes_offset())); // get code base
duke@0 452 if (native) {
never@304 453 __ movptr(STATE(_bcp), (int32_t)NULL_WORD); // state->_bcp = NULL
duke@0 454 } else {
never@304 455 __ movptr(STATE(_bcp), rdx); // state->_bcp = codes()
duke@0 456 }
never@304 457 __ xorptr(rdx, rdx);
never@304 458 __ movptr(STATE(_oop_temp), rdx); // state->_oop_temp = NULL (only really needed for native)
never@304 459 __ movptr(STATE(_mdx), rdx); // state->_mdx = NULL
never@304 460 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
never@304 461 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
never@304 462 __ movptr(STATE(_constants), rdx); // state->_constants = constants()
never@304 463
never@304 464 __ movptr(STATE(_method), rbx); // state->_method = method()
never@304 465 __ movl(STATE(_msg), (int32_t) BytecodeInterpreter::method_entry); // state->_msg = initial method entry
never@304 466 __ movptr(STATE(_result._to_call._callee), (int32_t) NULL_WORD); // state->_result._to_call._callee_callee = NULL
never@304 467
never@304 468
never@304 469 __ movptr(STATE(_monitor_base), rsp); // set monitor block bottom (grows down) this would point to entry [0]
duke@0 470 // entries run from -1..x where &monitor[x] ==
duke@0 471
duke@0 472 {
duke@0 473 // Must not attempt to lock method until we enter interpreter as gc won't be able to find the
duke@0 474 // initial frame. However we allocate a free monitor so we don't have to shuffle the expression stack
duke@0 475 // immediately.
duke@0 476
duke@0 477 // synchronize method
duke@0 478 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
duke@0 479 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
duke@0 480 Label not_synced;
duke@0 481
duke@0 482 __ movl(rax, access_flags);
duke@0 483 __ testl(rax, JVM_ACC_SYNCHRONIZED);
duke@0 484 __ jcc(Assembler::zero, not_synced);
duke@0 485
duke@0 486 // Allocate initial monitor and pre initialize it
duke@0 487 // get synchronization object
duke@0 488
duke@0 489 Label done;
duke@0 490 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
duke@0 491 __ movl(rax, access_flags);
duke@0 492 __ testl(rax, JVM_ACC_STATIC);
never@304 493 __ movptr(rax, Address(locals, 0)); // get receiver (assume this is frequent case)
duke@0 494 __ jcc(Assembler::zero, done);
never@304 495 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
never@304 496 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
never@304 497 __ movptr(rax, Address(rax, mirror_offset));
duke@0 498 __ bind(done);
duke@0 499 // add space for monitor & lock
never@304 500 __ subptr(rsp, entry_size); // add space for a monitor entry
never@304 501 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
duke@0 502 __ bind(not_synced);
duke@0 503 }
duke@0 504
never@304 505 __ movptr(STATE(_stack_base), rsp); // set expression stack base ( == &monitors[-count])
duke@0 506 if (native) {
never@304 507 __ movptr(STATE(_stack), rsp); // set current expression stack tos
never@304 508 __ movptr(STATE(_stack_limit), rsp);
duke@0 509 } else {
never@304 510 __ subptr(rsp, wordSize); // pre-push stack
never@304 511 __ movptr(STATE(_stack), rsp); // set current expression stack tos
duke@0 512
duke@0 513 // compute full expression stack limit
duke@0 514
duke@0 515 const Address size_of_stack (rbx, methodOopDesc::max_stack_offset());
jrose@710 516 const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_words();
jrose@622 517 __ load_unsigned_short(rdx, size_of_stack); // get size of expression stack in words
never@304 518 __ negptr(rdx); // so we can subtract in next step
duke@0 519 // Allocate expression stack
jrose@710 520 __ lea(rsp, Address(rsp, rdx, Address::times_ptr, -extra_stack));
never@304 521 __ movptr(STATE(_stack_limit), rsp);
duke@0 522 }
duke@0 523
never@304 524 #ifdef _LP64
never@304 525 // Make sure stack is properly aligned and sized for the abi
never@304 526 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
twisti@605 527 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
never@304 528 #endif // _LP64
never@304 529
never@304 530
never@304 531
duke@0 532 }
duke@0 533
duke@0 534 // Helpers for commoning out cases in the various type of method entries.
duke@0 535 //
duke@0 536
duke@0 537 // increment invocation count & check for overflow
duke@0 538 //
duke@0 539 // Note: checking for negative value instead of overflow
duke@0 540 // so we have a 'sticky' overflow test
duke@0 541 //
duke@0 542 // rbx,: method
duke@0 543 // rcx: invocation counter
duke@0 544 //
duke@0 545 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
duke@0 546
duke@0 547 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
duke@0 548 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset());
duke@0 549
duke@0 550 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
never@304 551 __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset()));
duke@0 552 }
duke@0 553 // Update standard invocation counters
duke@0 554 __ movl(rax, backedge_counter); // load backedge counter
duke@0 555
duke@0 556 __ increment(rcx, InvocationCounter::count_increment);
duke@0 557 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
duke@0 558
duke@0 559 __ movl(invocation_counter, rcx); // save invocation count
duke@0 560 __ addl(rcx, rax); // add both counters
duke@0 561
duke@0 562 // profile_method is non-null only for interpreted method so
duke@0 563 // profile_method != NULL == !native_call
duke@0 564 // BytecodeInterpreter only calls for native so code is elided.
duke@0 565
duke@0 566 __ cmp32(rcx,
duke@0 567 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
duke@0 568 __ jcc(Assembler::aboveEqual, *overflow);
duke@0 569
duke@0 570 }
duke@0 571
duke@0 572 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
duke@0 573
duke@0 574 // C++ interpreter on entry
never@304 575 // rsi/r13 - new interpreter state pointer
duke@0 576 // rbp - interpreter frame pointer
duke@0 577 // rbx - method
duke@0 578
duke@0 579 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
duke@0 580 // rbx, - method
duke@0 581 // rcx - rcvr (assuming there is one)
duke@0 582 // top of stack return address of interpreter caller
duke@0 583 // rsp - sender_sp
duke@0 584
duke@0 585 // C++ interpreter only
never@304 586 // rsi/r13 - previous interpreter state pointer
duke@0 587
duke@0 588 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
duke@0 589
duke@0 590 // InterpreterRuntime::frequency_counter_overflow takes one argument
duke@0 591 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
duke@0 592 // The call returns the address of the verified entry point for the method or NULL
duke@0 593 // if the compilation did not complete (either went background or bailed out).
never@304 594 __ movptr(rax, (int32_t)false);
duke@0 595 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
duke@0 596
duke@0 597 // for c++ interpreter can rsi really be munged?
coleenp@520 598 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); // restore state
never@304 599 __ movptr(rbx, Address(state, byte_offset_of(BytecodeInterpreter, _method))); // restore method
never@304 600 __ movptr(rdi, Address(state, byte_offset_of(BytecodeInterpreter, _locals))); // get locals pointer
never@304 601
duke@0 602 __ jmp(*do_continue, relocInfo::none);
duke@0 603
duke@0 604 }
duke@0 605
duke@0 606 void InterpreterGenerator::generate_stack_overflow_check(void) {
duke@0 607 // see if we've got enough room on the stack for locals plus overhead.
duke@0 608 // the expression stack grows down incrementally, so the normal guard
duke@0 609 // page mechanism will work for that.
duke@0 610 //
duke@0 611 // Registers live on entry:
duke@0 612 //
duke@0 613 // Asm interpreter
duke@0 614 // rdx: number of additional locals this frame needs (what we must check)
duke@0 615 // rbx,: methodOop
duke@0 616
duke@0 617 // C++ Interpreter
never@304 618 // rsi/r13: previous interpreter frame state object
duke@0 619 // rdi: &locals[0]
duke@0 620 // rcx: # of locals
duke@0 621 // rdx: number of additional locals this frame needs (what we must check)
duke@0 622 // rbx: methodOop
duke@0 623
duke@0 624 // destroyed on exit
duke@0 625 // rax,
duke@0 626
duke@0 627 // NOTE: since the additional locals are also always pushed (wasn't obvious in
duke@0 628 // generate_method_entry) so the guard should work for them too.
duke@0 629 //
duke@0 630
duke@0 631 // monitor entry size: see picture of stack set (generate_method_entry) and frame_i486.hpp
duke@0 632 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
duke@0 633
duke@0 634 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
duke@0 635 // be sure to change this if you add/subtract anything to/from the overhead area
duke@0 636 const int overhead_size = (int)sizeof(BytecodeInterpreter);
duke@0 637
duke@0 638 const int page_size = os::vm_page_size();
duke@0 639
duke@0 640 Label after_frame_check;
duke@0 641
duke@0 642 // compute rsp as if this were going to be the last frame on
duke@0 643 // the stack before the red zone
duke@0 644
duke@0 645 Label after_frame_check_pop;
duke@0 646
duke@0 647 // save rsi == caller's bytecode ptr (c++ previous interp. state)
duke@0 648 // QQQ problem here?? rsi overload????
never@304 649 __ push(state);
never@304 650
never@304 651 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rsi);
never@304 652
never@304 653 NOT_LP64(__ get_thread(thread));
duke@0 654
duke@0 655 const Address stack_base(thread, Thread::stack_base_offset());
duke@0 656 const Address stack_size(thread, Thread::stack_size_offset());
duke@0 657
duke@0 658 // locals + overhead, in bytes
duke@0 659 const Address size_of_stack (rbx, methodOopDesc::max_stack_offset());
duke@0 660 // Always give one monitor to allow us to start interp if sync method.
duke@0 661 // Any additional monitors need a check when moving the expression stack
coleenp@520 662 const int one_monitor = frame::interpreter_frame_monitor_size() * wordSize;
jrose@710 663 const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries();
jrose@622 664 __ load_unsigned_short(rax, size_of_stack); // get size of expression stack in words
jrose@710 665 __ lea(rax, Address(noreg, rax, Interpreter::stackElementScale(), extra_stack + one_monitor));
never@304 666 __ lea(rax, Address(rax, rdx, Interpreter::stackElementScale(), overhead_size));
duke@0 667
duke@0 668 #ifdef ASSERT
duke@0 669 Label stack_base_okay, stack_size_okay;
duke@0 670 // verify that thread stack base is non-zero
never@304 671 __ cmpptr(stack_base, (int32_t)0);
duke@0 672 __ jcc(Assembler::notEqual, stack_base_okay);
duke@0 673 __ stop("stack base is zero");
duke@0 674 __ bind(stack_base_okay);
duke@0 675 // verify that thread stack size is non-zero
never@304 676 __ cmpptr(stack_size, (int32_t)0);
duke@0 677 __ jcc(Assembler::notEqual, stack_size_okay);
duke@0 678 __ stop("stack size is zero");
duke@0 679 __ bind(stack_size_okay);
duke@0 680 #endif
duke@0 681
duke@0 682 // Add stack base to locals and subtract stack size
never@304 683 __ addptr(rax, stack_base);
never@304 684 __ subptr(rax, stack_size);
duke@0 685
duke@0 686 // We should have a magic number here for the size of the c++ interpreter frame.
duke@0 687 // We can't actually tell this ahead of time. The debug version size is around 3k
duke@0 688 // product is 1k and fastdebug is 4k
duke@0 689 const int slop = 6 * K;
duke@0 690
duke@0 691 // Use the maximum number of pages we might bang.
duke@0 692 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
duke@0 693 (StackRedPages+StackYellowPages);
duke@0 694 // Only need this if we are stack banging which is temporary while
duke@0 695 // we're debugging.
never@304 696 __ addptr(rax, slop + 2*max_pages * page_size);
duke@0 697
duke@0 698 // check against the current stack bottom
never@304 699 __ cmpptr(rsp, rax);
duke@0 700 __ jcc(Assembler::above, after_frame_check_pop);
duke@0 701
never@304 702 __ pop(state); // get c++ prev state.
duke@0 703
duke@0 704 // throw exception return address becomes throwing pc
duke@0 705 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
duke@0 706
duke@0 707 // all done with frame size check
duke@0 708 __ bind(after_frame_check_pop);
never@304 709 __ pop(state);
duke@0 710
duke@0 711 __ bind(after_frame_check);
duke@0 712 }
duke@0 713
duke@0 714 // Find preallocated monitor and lock method (C++ interpreter)
duke@0 715 // rbx - methodOop
duke@0 716 //
duke@0 717 void InterpreterGenerator::lock_method(void) {
never@304 718 // assumes state == rsi/r13 == pointer to current interpreterState
never@304 719 // minimally destroys rax, rdx|c_rarg1, rdi
duke@0 720 //
duke@0 721 // synchronize method
duke@0 722 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
duke@0 723 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
duke@0 724
never@304 725 const Register monitor = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
never@304 726
duke@0 727 // find initial monitor i.e. monitors[-1]
never@304 728 __ movptr(monitor, STATE(_monitor_base)); // get monitor bottom limit
never@304 729 __ subptr(monitor, entry_size); // point to initial monitor
duke@0 730
duke@0 731 #ifdef ASSERT
duke@0 732 { Label L;
duke@0 733 __ movl(rax, access_flags);
duke@0 734 __ testl(rax, JVM_ACC_SYNCHRONIZED);
duke@0 735 __ jcc(Assembler::notZero, L);
duke@0 736 __ stop("method doesn't need synchronization");
duke@0 737 __ bind(L);
duke@0 738 }
duke@0 739 #endif // ASSERT
duke@0 740 // get synchronization object
duke@0 741 { Label done;
duke@0 742 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
duke@0 743 __ movl(rax, access_flags);
never@304 744 __ movptr(rdi, STATE(_locals)); // prepare to get receiver (assume common case)
duke@0 745 __ testl(rax, JVM_ACC_STATIC);
never@304 746 __ movptr(rax, Address(rdi, 0)); // get receiver (assume this is frequent case)
duke@0 747 __ jcc(Assembler::zero, done);
never@304 748 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
never@304 749 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
never@304 750 __ movptr(rax, Address(rax, mirror_offset));
duke@0 751 __ bind(done);
duke@0 752 }
duke@0 753 #ifdef ASSERT
duke@0 754 { Label L;
never@304 755 __ cmpptr(rax, Address(monitor, BasicObjectLock::obj_offset_in_bytes())); // correct object?
duke@0 756 __ jcc(Assembler::equal, L);
duke@0 757 __ stop("wrong synchronization lobject");
duke@0 758 __ bind(L);
duke@0 759 }
duke@0 760 #endif // ASSERT
never@304 761 // can destroy rax, rdx|c_rarg1, rcx, and (via call_VM) rdi!
never@304 762 __ lock_object(monitor);
duke@0 763 }
duke@0 764
duke@0 765 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
duke@0 766
duke@0 767 address InterpreterGenerator::generate_accessor_entry(void) {
duke@0 768
never@304 769 // rbx: methodOop
never@304 770
never@304 771 // rsi/r13: senderSP must preserved for slow path, set SP to it on fast path
duke@0 772
duke@0 773 Label xreturn_path;
duke@0 774
duke@0 775 // do fastpath for resolved accessor methods
duke@0 776 if (UseFastAccessorMethods) {
duke@0 777
duke@0 778 address entry_point = __ pc();
duke@0 779
duke@0 780 Label slow_path;
duke@0 781 // If we need a safepoint check, generate full interpreter entry.
duke@0 782 ExternalAddress state(SafepointSynchronize::address_of_state());
duke@0 783 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
duke@0 784 SafepointSynchronize::_not_synchronized);
duke@0 785
duke@0 786 __ jcc(Assembler::notEqual, slow_path);
duke@0 787 // ASM/C++ Interpreter
duke@0 788 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
duke@0 789 // Note: We can only use this code if the getfield has been resolved
duke@0 790 // and if we don't have a null-pointer exception => check for
duke@0 791 // these conditions first and use slow path if necessary.
duke@0 792 // rbx,: method
duke@0 793 // rcx: receiver
never@304 794 __ movptr(rax, Address(rsp, wordSize));
duke@0 795
duke@0 796 // check if local 0 != NULL and read field
never@304 797 __ testptr(rax, rax);
duke@0 798 __ jcc(Assembler::zero, slow_path);
duke@0 799
never@304 800 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
duke@0 801 // read first instruction word and extract bytecode @ 1 and index @ 2
never@304 802 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
duke@0 803 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
duke@0 804 // Shift codes right to get the index on the right.
duke@0 805 // The bytecode fetched looks like <index><0xb4><0x2a>
duke@0 806 __ shrl(rdx, 2*BitsPerByte);
duke@0 807 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
never@304 808 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
duke@0 809
duke@0 810 // rax,: local 0
duke@0 811 // rbx,: method
duke@0 812 // rcx: receiver - do not destroy since it is needed for slow path!
duke@0 813 // rcx: scratch
duke@0 814 // rdx: constant pool cache index
duke@0 815 // rdi: constant pool cache
never@304 816 // rsi/r13: sender sp
duke@0 817
duke@0 818 // check if getfield has been resolved and read constant pool cache entry
duke@0 819 // check the validity of the cache entry by testing whether _indices field
duke@0 820 // contains Bytecode::_getfield in b1 byte.
duke@0 821 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
duke@0 822 __ movl(rcx,
duke@0 823 Address(rdi,
duke@0 824 rdx,
never@304 825 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
duke@0 826 __ shrl(rcx, 2*BitsPerByte);
duke@0 827 __ andl(rcx, 0xFF);
duke@0 828 __ cmpl(rcx, Bytecodes::_getfield);
duke@0 829 __ jcc(Assembler::notEqual, slow_path);
duke@0 830
duke@0 831 // Note: constant pool entry is not valid before bytecode is resolved
never@304 832 __ movptr(rcx,
duke@0 833 Address(rdi,
duke@0 834 rdx,
never@304 835 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()));
duke@0 836 __ movl(rdx,
duke@0 837 Address(rdi,
duke@0 838 rdx,
never@304 839 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset()));
duke@0 840
duke@0 841 Label notByte, notShort, notChar;
duke@0 842 const Address field_address (rax, rcx, Address::times_1);
duke@0 843
duke@0 844 // Need to differentiate between igetfield, agetfield, bgetfield etc.
duke@0 845 // because they are different sizes.
duke@0 846 // Use the type from the constant pool cache
duke@0 847 __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
duke@0 848 // Make sure we don't need to mask rdx for tosBits after the above shift
duke@0 849 ConstantPoolCacheEntry::verify_tosBits();
never@304 850 #ifdef _LP64
never@304 851 Label notObj;
never@304 852 __ cmpl(rdx, atos);
never@304 853 __ jcc(Assembler::notEqual, notObj);
never@304 854 // atos
never@304 855 __ movptr(rax, field_address);
never@304 856 __ jmp(xreturn_path);
never@304 857
never@304 858 __ bind(notObj);
never@304 859 #endif // _LP64
duke@0 860 __ cmpl(rdx, btos);
duke@0 861 __ jcc(Assembler::notEqual, notByte);
duke@0 862 __ load_signed_byte(rax, field_address);
duke@0 863 __ jmp(xreturn_path);
duke@0 864
duke@0 865 __ bind(notByte);
duke@0 866 __ cmpl(rdx, stos);
duke@0 867 __ jcc(Assembler::notEqual, notShort);
jrose@622 868 __ load_signed_short(rax, field_address);
duke@0 869 __ jmp(xreturn_path);
duke@0 870
duke@0 871 __ bind(notShort);
duke@0 872 __ cmpl(rdx, ctos);
duke@0 873 __ jcc(Assembler::notEqual, notChar);
jrose@622 874 __ load_unsigned_short(rax, field_address);
duke@0 875 __ jmp(xreturn_path);
duke@0 876
duke@0 877 __ bind(notChar);
duke@0 878 #ifdef ASSERT
duke@0 879 Label okay;
never@304 880 #ifndef _LP64
duke@0 881 __ cmpl(rdx, atos);
duke@0 882 __ jcc(Assembler::equal, okay);
never@304 883 #endif // _LP64
duke@0 884 __ cmpl(rdx, itos);
duke@0 885 __ jcc(Assembler::equal, okay);
duke@0 886 __ stop("what type is this?");
duke@0 887 __ bind(okay);
duke@0 888 #endif // ASSERT
duke@0 889 // All the rest are a 32 bit wordsize
duke@0 890 __ movl(rax, field_address);
duke@0 891
duke@0 892 __ bind(xreturn_path);
duke@0 893
duke@0 894 // _ireturn/_areturn
never@304 895 __ pop(rdi); // get return address
never@304 896 __ mov(rsp, sender_sp_on_entry); // set sp to sender sp
duke@0 897 __ jmp(rdi);
duke@0 898
duke@0 899 // generate a vanilla interpreter entry as the slow path
duke@0 900 __ bind(slow_path);
duke@0 901 // We will enter c++ interpreter looking like it was
duke@0 902 // called by the call_stub this will cause it to return
duke@0 903 // a tosca result to the invoker which might have been
duke@0 904 // the c++ interpreter itself.
duke@0 905
duke@0 906 __ jmp(fast_accessor_slow_entry_path);
duke@0 907 return entry_point;
duke@0 908
duke@0 909 } else {
duke@0 910 return NULL;
duke@0 911 }
duke@0 912
duke@0 913 }
duke@0 914
duke@0 915 //
duke@0 916 // C++ Interpreter stub for calling a native method.
duke@0 917 // This sets up a somewhat different looking stack for calling the native method
duke@0 918 // than the typical interpreter frame setup but still has the pointer to
duke@0 919 // an interpreter state.
duke@0 920 //
duke@0 921
duke@0 922 address InterpreterGenerator::generate_native_entry(bool synchronized) {
duke@0 923 // determine code generation flags
duke@0 924 bool inc_counter = UseCompiler || CountCompiledCalls;
duke@0 925
duke@0 926 // rbx: methodOop
duke@0 927 // rcx: receiver (unused)
never@304 928 // rsi/r13: previous interpreter state (if called from C++ interpreter) must preserve
never@304 929 // in any case. If called via c1/c2/call_stub rsi/r13 is junk (to use) but harmless
duke@0 930 // to save/restore.
duke@0 931 address entry_point = __ pc();
duke@0 932
duke@0 933 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
duke@0 934 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset());
duke@0 935 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
duke@0 936 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
duke@0 937
never@304 938 // rsi/r13 == state/locals rdi == prevstate
duke@0 939 const Register locals = rdi;
duke@0 940
duke@0 941 // get parameter size (always needed)
jrose@622 942 __ load_unsigned_short(rcx, size_of_parameters);
duke@0 943
duke@0 944 // rbx: methodOop
duke@0 945 // rcx: size of parameters
never@304 946 __ pop(rax); // get return address
duke@0 947 // for natives the size of locals is zero
duke@0 948
duke@0 949 // compute beginning of parameters /locals
never@304 950 __ lea(locals, Address(rsp, rcx, Address::times_ptr, -wordSize));
duke@0 951
duke@0 952 // initialize fixed part of activation frame
duke@0 953
duke@0 954 // Assumes rax = return address
duke@0 955
duke@0 956 // allocate and initialize new interpreterState and method expression stack
duke@0 957 // IN(locals) -> locals
duke@0 958 // IN(state) -> previous frame manager state (NULL from stub/c1/c2)
duke@0 959 // destroys rax, rcx, rdx
duke@0 960 // OUT (state) -> new interpreterState
duke@0 961 // OUT(rsp) -> bottom of methods expression stack
duke@0 962
duke@0 963 // save sender_sp
never@304 964 __ mov(rcx, sender_sp_on_entry);
duke@0 965 // start with NULL previous state
never@304 966 __ movptr(state, (int32_t)NULL_WORD);
duke@0 967 generate_compute_interpreter_state(state, locals, rcx, true);
duke@0 968
duke@0 969 #ifdef ASSERT
duke@0 970 { Label L;
never@304 971 __ movptr(rax, STATE(_stack_base));
never@304 972 #ifdef _LP64
never@304 973 // duplicate the alignment rsp got after setting stack_base
never@304 974 __ subptr(rax, frame::arg_reg_save_area_bytes); // windows
twisti@605 975 __ andptr(rax, -16); // must be 16 byte boundary (see amd64 ABI)
never@304 976 #endif // _LP64
never@304 977 __ cmpptr(rax, rsp);
duke@0 978 __ jcc(Assembler::equal, L);
duke@0 979 __ stop("broken stack frame setup in interpreter");
duke@0 980 __ bind(L);
duke@0 981 }
duke@0 982 #endif
duke@0 983
duke@0 984 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
duke@0 985
never@304 986 const Register unlock_thread = LP64_ONLY(r15_thread) NOT_LP64(rax);
never@304 987 NOT_LP64(__ movptr(unlock_thread, STATE(_thread));) // get thread
duke@0 988 // Since at this point in the method invocation the exception handler
duke@0 989 // would try to exit the monitor of synchronized methods which hasn't
duke@0 990 // been entered yet, we set the thread local variable
duke@0 991 // _do_not_unlock_if_synchronized to true. The remove_activation will
duke@0 992 // check this flag.
duke@0 993
never@304 994 const Address do_not_unlock_if_synchronized(unlock_thread,
duke@0 995 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
duke@0 996 __ movbool(do_not_unlock_if_synchronized, true);
duke@0 997
duke@0 998 // make sure method is native & not abstract
duke@0 999 #ifdef ASSERT
duke@0 1000 __ movl(rax, access_flags);
duke@0 1001 {
duke@0 1002 Label L;
duke@0 1003 __ testl(rax, JVM_ACC_NATIVE);
duke@0 1004 __ jcc(Assembler::notZero, L);
duke@0 1005 __ stop("tried to execute non-native method as native");
duke@0 1006 __ bind(L);
duke@0 1007 }
duke@0 1008 { Label L;
duke@0 1009 __ testl(rax, JVM_ACC_ABSTRACT);
duke@0 1010 __ jcc(Assembler::zero, L);
duke@0 1011 __ stop("tried to execute abstract method in interpreter");
duke@0 1012 __ bind(L);
duke@0 1013 }
duke@0 1014 #endif
duke@0 1015
duke@0 1016
duke@0 1017 // increment invocation count & check for overflow
duke@0 1018 Label invocation_counter_overflow;
duke@0 1019 if (inc_counter) {
duke@0 1020 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
duke@0 1021 }
duke@0 1022
duke@0 1023 Label continue_after_compile;
duke@0 1024
duke@0 1025 __ bind(continue_after_compile);
duke@0 1026
duke@0 1027 bang_stack_shadow_pages(true);
duke@0 1028
duke@0 1029 // reset the _do_not_unlock_if_synchronized flag
never@304 1030 NOT_LP64(__ movl(rax, STATE(_thread));) // get thread
duke@0 1031 __ movbool(do_not_unlock_if_synchronized, false);
duke@0 1032
duke@0 1033
duke@0 1034 // check for synchronized native methods
duke@0 1035 //
duke@0 1036 // Note: This must happen *after* invocation counter check, since
duke@0 1037 // when overflow happens, the method should not be locked.
duke@0 1038 if (synchronized) {
duke@0 1039 // potentially kills rax, rcx, rdx, rdi
duke@0 1040 lock_method();
duke@0 1041 } else {
duke@0 1042 // no synchronization necessary
duke@0 1043 #ifdef ASSERT
duke@0 1044 { Label L;
duke@0 1045 __ movl(rax, access_flags);
duke@0 1046 __ testl(rax, JVM_ACC_SYNCHRONIZED);
duke@0 1047 __ jcc(Assembler::zero, L);
duke@0 1048 __ stop("method needs synchronization");
duke@0 1049 __ bind(L);
duke@0 1050 }
duke@0 1051 #endif
duke@0 1052 }
duke@0 1053
duke@0 1054 // start execution
duke@0 1055
duke@0 1056 // jvmti support
duke@0 1057 __ notify_method_entry();
duke@0 1058
duke@0 1059 // work registers
duke@0 1060 const Register method = rbx;
never@304 1061 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rdi);
never@304 1062 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); // rcx|rscratch1
duke@0 1063
duke@0 1064 // allocate space for parameters
never@304 1065 __ movptr(method, STATE(_method));
duke@0 1066 __ verify_oop(method);
jrose@622 1067 __ load_unsigned_short(t, Address(method, methodOopDesc::size_of_parameters_offset()));
duke@0 1068 __ shll(t, 2);
never@304 1069 #ifdef _LP64
never@304 1070 __ subptr(rsp, t);
never@304 1071 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
twisti@605 1072 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
never@304 1073 #else
never@304 1074 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
never@304 1075 __ subptr(rsp, t);
never@304 1076 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
never@304 1077 #endif // _LP64
duke@0 1078
duke@0 1079 // get signature handler
duke@0 1080 Label pending_exception_present;
duke@0 1081
duke@0 1082 { Label L;
never@304 1083 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
never@304 1084 __ testptr(t, t);
duke@0 1085 __ jcc(Assembler::notZero, L);
duke@0 1086 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method, false);
never@304 1087 __ movptr(method, STATE(_method));
never@304 1088 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
duke@0 1089 __ jcc(Assembler::notEqual, pending_exception_present);
duke@0 1090 __ verify_oop(method);
never@304 1091 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
duke@0 1092 __ bind(L);
duke@0 1093 }
duke@0 1094 #ifdef ASSERT
duke@0 1095 {
duke@0 1096 Label L;
never@304 1097 __ push(t);
duke@0 1098 __ get_thread(t); // get vm's javathread*
never@304 1099 __ cmpptr(t, STATE(_thread));
duke@0 1100 __ jcc(Assembler::equal, L);
duke@0 1101 __ int3();
duke@0 1102 __ bind(L);
never@304 1103 __ pop(t);
duke@0 1104 }
duke@0 1105 #endif //
duke@0 1106
never@304 1107 const Register from_ptr = InterpreterRuntime::SignatureHandlerGenerator::from();
duke@0 1108 // call signature handler
duke@0 1109 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
never@304 1110
duke@0 1111 // The generated handlers do not touch RBX (the method oop).
duke@0 1112 // However, large signatures cannot be cached and are generated
duke@0 1113 // each time here. The slow-path generator will blow RBX
duke@0 1114 // sometime, so we must reload it after the call.
never@304 1115 __ movptr(from_ptr, STATE(_locals)); // get the from pointer
duke@0 1116 __ call(t);
never@304 1117 __ movptr(method, STATE(_method));
duke@0 1118 __ verify_oop(method);
duke@0 1119
duke@0 1120 // result handler is in rax
duke@0 1121 // set result handler
never@304 1122 __ movptr(STATE(_result_handler), rax);
never@304 1123
never@304 1124
never@304 1125 // get native function entry point
never@304 1126 { Label L;
never@304 1127 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
never@304 1128 __ testptr(rax, rax);
never@304 1129 __ jcc(Assembler::notZero, L);
never@304 1130 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
never@304 1131 __ movptr(method, STATE(_method));
never@304 1132 __ verify_oop(method);
never@304 1133 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
never@304 1134 __ bind(L);
never@304 1135 }
duke@0 1136
duke@0 1137 // pass mirror handle if static call
duke@0 1138 { Label L;
duke@0 1139 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
duke@0 1140 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
duke@0 1141 __ testl(t, JVM_ACC_STATIC);
duke@0 1142 __ jcc(Assembler::zero, L);
duke@0 1143 // get mirror
never@304 1144 __ movptr(t, Address(method, methodOopDesc:: constants_offset()));
never@304 1145 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
never@304 1146 __ movptr(t, Address(t, mirror_offset));
duke@0 1147 // copy mirror into activation object
never@304 1148 __ movptr(STATE(_oop_temp), t);
duke@0 1149 // pass handle to mirror
never@304 1150 #ifdef _LP64
never@304 1151 __ lea(c_rarg1, STATE(_oop_temp));
never@304 1152 #else
never@304 1153 __ lea(t, STATE(_oop_temp));
never@304 1154 __ movptr(Address(rsp, wordSize), t);
never@304 1155 #endif // _LP64
duke@0 1156 __ bind(L);
duke@0 1157 }
duke@0 1158 #ifdef ASSERT
duke@0 1159 {
duke@0 1160 Label L;
never@304 1161 __ push(t);
duke@0 1162 __ get_thread(t); // get vm's javathread*
never@304 1163 __ cmpptr(t, STATE(_thread));
duke@0 1164 __ jcc(Assembler::equal, L);
duke@0 1165 __ int3();
duke@0 1166 __ bind(L);
never@304 1167 __ pop(t);
duke@0 1168 }
duke@0 1169 #endif //
duke@0 1170
duke@0 1171 // pass JNIEnv
never@304 1172 #ifdef _LP64
never@304 1173 __ lea(c_rarg0, Address(thread, JavaThread::jni_environment_offset()));
never@304 1174 #else
never@304 1175 __ movptr(thread, STATE(_thread)); // get thread
never@304 1176 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
never@304 1177
never@304 1178 __ movptr(Address(rsp, 0), t);
never@304 1179 #endif // _LP64
never@304 1180
duke@0 1181 #ifdef ASSERT
duke@0 1182 {
duke@0 1183 Label L;
never@304 1184 __ push(t);
duke@0 1185 __ get_thread(t); // get vm's javathread*
never@304 1186 __ cmpptr(t, STATE(_thread));
duke@0 1187 __ jcc(Assembler::equal, L);
duke@0 1188 __ int3();
duke@0 1189 __ bind(L);
never@304 1190 __ pop(t);
duke@0 1191 }
duke@0 1192 #endif //
duke@0 1193
duke@0 1194 #ifdef ASSERT
duke@0 1195 { Label L;
duke@0 1196 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
duke@0 1197 __ cmpl(t, _thread_in_Java);
duke@0 1198 __ jcc(Assembler::equal, L);
duke@0 1199 __ stop("Wrong thread state in native stub");
duke@0 1200 __ bind(L);
duke@0 1201 }
duke@0 1202 #endif
duke@0 1203
duke@0 1204 // Change state to native (we save the return address in the thread, since it might not
duke@0 1205 // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
duke@0 1206 // points into the right code segment. It does not have to be the correct return pc.
duke@0 1207
duke@0 1208 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
duke@0 1209
duke@0 1210 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
duke@0 1211
duke@0 1212 __ call(rax);
duke@0 1213
duke@0 1214 // result potentially in rdx:rax or ST0
never@304 1215 __ movptr(method, STATE(_method));
never@304 1216 NOT_LP64(__ movptr(thread, STATE(_thread));) // get thread
duke@0 1217
duke@0 1218 // The potential result is in ST(0) & rdx:rax
duke@0 1219 // With C++ interpreter we leave any possible result in ST(0) until we are in result handler and then
duke@0 1220 // we do the appropriate stuff for returning the result. rdx:rax must always be saved because just about
duke@0 1221 // anything we do here will destroy it, st(0) is only saved if we re-enter the vm where it would
duke@0 1222 // be destroyed.
duke@0 1223 // It is safe to do these pushes because state is _thread_in_native and return address will be found
duke@0 1224 // via _last_native_pc and not via _last_jave_sp
duke@0 1225
never@304 1226 // Must save the value of ST(0)/xmm0 since it could be destroyed before we get to result handler
duke@0 1227 { Label Lpush, Lskip;
duke@0 1228 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
duke@0 1229 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
duke@0 1230 __ cmpptr(STATE(_result_handler), float_handler.addr());
duke@0 1231 __ jcc(Assembler::equal, Lpush);
duke@0 1232 __ cmpptr(STATE(_result_handler), double_handler.addr());
duke@0 1233 __ jcc(Assembler::notEqual, Lskip);
duke@0 1234 __ bind(Lpush);
never@304 1235 __ subptr(rsp, 2*wordSize);
never@304 1236 if ( UseSSE < 2 ) {
never@304 1237 __ fstp_d(Address(rsp, 0));
never@304 1238 } else {
never@304 1239 __ movdbl(Address(rsp, 0), xmm0);
never@304 1240 }
duke@0 1241 __ bind(Lskip);
duke@0 1242 }
duke@0 1243
never@304 1244 // save rax:rdx for potential use by result handler.
never@304 1245 __ push(rax);
never@304 1246 #ifndef _LP64
never@304 1247 __ push(rdx);
never@304 1248 #endif // _LP64
duke@0 1249
duke@0 1250 // Either restore the MXCSR register after returning from the JNI Call
duke@0 1251 // or verify that it wasn't changed.
duke@0 1252 if (VM_Version::supports_sse()) {
duke@0 1253 if (RestoreMXCSROnJNICalls) {
duke@0 1254 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
duke@0 1255 }
duke@0 1256 else if (CheckJNICalls ) {
never@304 1257 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
duke@0 1258 }
duke@0 1259 }
duke@0 1260
never@304 1261 #ifndef _LP64
duke@0 1262 // Either restore the x87 floating pointer control word after returning
duke@0 1263 // from the JNI call or verify that it wasn't changed.
duke@0 1264 if (CheckJNICalls) {
never@304 1265 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
duke@0 1266 }
never@304 1267 #endif // _LP64
duke@0 1268
duke@0 1269
duke@0 1270 // change thread state
duke@0 1271 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
duke@0 1272 if(os::is_MP()) {
duke@0 1273 // Write serialization page so VM thread can do a pseudo remote membar.
duke@0 1274 // We use the current thread pointer to calculate a thread specific
duke@0 1275 // offset to write to within the page. This minimizes bus traffic
duke@0 1276 // due to cache line collision.
duke@0 1277 __ serialize_memory(thread, rcx);
duke@0 1278 }
duke@0 1279
duke@0 1280 // check for safepoint operation in progress and/or pending suspend requests
duke@0 1281 { Label Continue;
duke@0 1282
duke@0 1283 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
duke@0 1284 SafepointSynchronize::_not_synchronized);
duke@0 1285
duke@0 1286 // threads running native code and they are expected to self-suspend
duke@0 1287 // when leaving the _thread_in_native state. We need to check for
duke@0 1288 // pending suspend requests here.
duke@0 1289 Label L;
duke@0 1290 __ jcc(Assembler::notEqual, L);
duke@0 1291 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
duke@0 1292 __ jcc(Assembler::equal, Continue);
duke@0 1293 __ bind(L);
duke@0 1294
duke@0 1295 // Don't use call_VM as it will see a possible pending exception and forward it
duke@0 1296 // and never return here preventing us from clearing _last_native_pc down below.
duke@0 1297 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
never@304 1298 // preserved and correspond to the bcp/locals pointers.
duke@0 1299 //
never@304 1300
never@304 1301 ((MacroAssembler*)_masm)->call_VM_leaf(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
never@304 1302 thread);
duke@0 1303 __ increment(rsp, wordSize);
duke@0 1304
never@304 1305 __ movptr(method, STATE(_method));
duke@0 1306 __ verify_oop(method);
never@304 1307 __ movptr(thread, STATE(_thread)); // get thread
duke@0 1308
duke@0 1309 __ bind(Continue);
duke@0 1310 }
duke@0 1311
duke@0 1312 // change thread state
duke@0 1313 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
duke@0 1314
duke@0 1315 __ reset_last_Java_frame(thread, true, true);
duke@0 1316
duke@0 1317 // reset handle block
never@304 1318 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
never@304 1319 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
duke@0 1320
duke@0 1321 // If result was an oop then unbox and save it in the frame
duke@0 1322 { Label L;
duke@0 1323 Label no_oop, store_result;
duke@0 1324 ExternalAddress oop_handler(AbstractInterpreter::result_handler(T_OBJECT));
duke@0 1325 __ cmpptr(STATE(_result_handler), oop_handler.addr());
duke@0 1326 __ jcc(Assembler::notEqual, no_oop);
never@304 1327 #ifndef _LP64
never@304 1328 __ pop(rdx);
never@304 1329 #endif // _LP64
never@304 1330 __ pop(rax);
never@304 1331 __ testptr(rax, rax);
duke@0 1332 __ jcc(Assembler::zero, store_result);
duke@0 1333 // unbox
never@304 1334 __ movptr(rax, Address(rax, 0));
duke@0 1335 __ bind(store_result);
never@304 1336 __ movptr(STATE(_oop_temp), rax);
duke@0 1337 // keep stack depth as expected by pushing oop which will eventually be discarded
never@304 1338 __ push(rax);
never@304 1339 #ifndef _LP64
never@304 1340 __ push(rdx);
never@304 1341 #endif // _LP64
duke@0 1342 __ bind(no_oop);
duke@0 1343 }
duke@0 1344
duke@0 1345 {
duke@0 1346 Label no_reguard;
duke@0 1347 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
duke@0 1348 __ jcc(Assembler::notEqual, no_reguard);
duke@0 1349
never@304 1350 __ pusha();
duke@0 1351 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
never@304 1352 __ popa();
duke@0 1353
duke@0 1354 __ bind(no_reguard);
duke@0 1355 }
duke@0 1356
duke@0 1357
duke@0 1358 // QQQ Seems like for native methods we simply return and the caller will see the pending
duke@0 1359 // exception and do the right thing. Certainly the interpreter will, don't know about
duke@0 1360 // compiled methods.
duke@0 1361 // Seems that the answer to above is no this is wrong. The old code would see the exception
duke@0 1362 // and forward it before doing the unlocking and notifying jvmdi that method has exited.
duke@0 1363 // This seems wrong need to investigate the spec.
duke@0 1364
duke@0 1365 // handle exceptions (exception handling will handle unlocking!)
duke@0 1366 { Label L;
never@304 1367 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
duke@0 1368 __ jcc(Assembler::zero, L);
duke@0 1369 __ bind(pending_exception_present);
duke@0 1370
duke@0 1371 // There are potential results on the stack (rax/rdx, ST(0)) we ignore these and simply
duke@0 1372 // return and let caller deal with exception. This skips the unlocking here which
duke@0 1373 // seems wrong but seems to be what asm interpreter did. Can't find this in the spec.
duke@0 1374 // Note: must preverve method in rbx
duke@0 1375 //
duke@0 1376
duke@0 1377 // remove activation
duke@0 1378
never@304 1379 __ movptr(t, STATE(_sender_sp));
duke@0 1380 __ leave(); // remove frame anchor
never@304 1381 __ pop(rdi); // get return address
never@304 1382 __ movptr(state, STATE(_prev_link)); // get previous state for return
never@304 1383 __ mov(rsp, t); // set sp to sender sp
never@304 1384 __ push(rdi); // push throwing pc
duke@0 1385 // The skips unlocking!! This seems to be what asm interpreter does but seems
duke@0 1386 // very wrong. Not clear if this violates the spec.
duke@0 1387 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
duke@0 1388 __ bind(L);
duke@0 1389 }
duke@0 1390
duke@0 1391 // do unlocking if necessary
duke@0 1392 { Label L;
duke@0 1393 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
duke@0 1394 __ testl(t, JVM_ACC_SYNCHRONIZED);
duke@0 1395 __ jcc(Assembler::zero, L);
duke@0 1396 // the code below should be shared with interpreter macro assembler implementation
duke@0 1397 { Label unlock;
never@304 1398 const Register monitor = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
duke@0 1399 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
duke@0 1400 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
never@304 1401 __ movptr(monitor, STATE(_monitor_base));
never@304 1402 __ subptr(monitor, frame::interpreter_frame_monitor_size() * wordSize); // address of initial monitor
never@304 1403
never@304 1404 __ movptr(t, Address(monitor, BasicObjectLock::obj_offset_in_bytes()));
never@304 1405 __ testptr(t, t);
duke@0 1406 __ jcc(Assembler::notZero, unlock);
duke@0 1407
duke@0 1408 // Entry already unlocked, need to throw exception
duke@0 1409 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
duke@0 1410 __ should_not_reach_here();
duke@0 1411
duke@0 1412 __ bind(unlock);
never@304 1413 __ unlock_object(monitor);
duke@0 1414 // unlock can blow rbx so restore it for path that needs it below
never@304 1415 __ movptr(method, STATE(_method));
duke@0 1416 }
duke@0 1417 __ bind(L);
duke@0 1418 }
duke@0 1419
duke@0 1420 // jvmti support
duke@0 1421 // Note: This must happen _after_ handling/throwing any exceptions since
duke@0 1422 // the exception handler code notifies the runtime of method exits
duke@0 1423 // too. If this happens before, method entry/exit notifications are
duke@0 1424 // not properly paired (was bug - gri 11/22/99).
duke@0 1425 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
duke@0 1426
duke@0 1427 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
never@304 1428 #ifndef _LP64
never@304 1429 __ pop(rdx);
never@304 1430 #endif // _LP64
never@304 1431 __ pop(rax);
never@304 1432 __ movptr(t, STATE(_result_handler)); // get result handler
duke@0 1433 __ call(t); // call result handler to convert to tosca form
duke@0 1434
duke@0 1435 // remove activation
duke@0 1436
never@304 1437 __ movptr(t, STATE(_sender_sp));
duke@0 1438
duke@0 1439 __ leave(); // remove frame anchor
never@304 1440 __ pop(rdi); // get return address
never@304 1441 __ movptr(state, STATE(_prev_link)); // get previous state for return (if c++ interpreter was caller)
never@304 1442 __ mov(rsp, t); // set sp to sender sp
duke@0 1443 __ jmp(rdi);
duke@0 1444
duke@0 1445 // invocation counter overflow
duke@0 1446 if (inc_counter) {
duke@0 1447 // Handle overflow of counter and compile method
duke@0 1448 __ bind(invocation_counter_overflow);
duke@0 1449 generate_counter_overflow(&continue_after_compile);
duke@0 1450 }
duke@0 1451
duke@0 1452 return entry_point;
duke@0 1453 }
duke@0 1454
duke@0 1455 // Generate entries that will put a result type index into rcx
duke@0 1456 void CppInterpreterGenerator::generate_deopt_handling() {
duke@0 1457
duke@0 1458 Label return_from_deopt_common;
duke@0 1459
duke@0 1460 // Generate entries that will put a result type index into rcx
duke@0 1461 // deopt needs to jump to here to enter the interpreter (return a result)
duke@0 1462 deopt_frame_manager_return_atos = __ pc();
duke@0 1463
duke@0 1464 // rax is live here
duke@0 1465 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_OBJECT)); // Result stub address array index
duke@0 1466 __ jmp(return_from_deopt_common);
duke@0 1467
duke@0 1468
duke@0 1469 // deopt needs to jump to here to enter the interpreter (return a result)
duke@0 1470 deopt_frame_manager_return_btos = __ pc();
duke@0 1471
duke@0 1472 // rax is live here
duke@0 1473 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_BOOLEAN)); // Result stub address array index
duke@0 1474 __ jmp(return_from_deopt_common);
duke@0 1475
duke@0 1476 // deopt needs to jump to here to enter the interpreter (return a result)
duke@0 1477 deopt_frame_manager_return_itos = __ pc();
duke@0 1478
duke@0 1479 // rax is live here
duke@0 1480 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_INT)); // Result stub address array index
duke@0 1481 __ jmp(return_from_deopt_common);
duke@0 1482
duke@0 1483 // deopt needs to jump to here to enter the interpreter (return a result)
duke@0 1484
duke@0 1485 deopt_frame_manager_return_ltos = __ pc();
duke@0 1486 // rax,rdx are live here
duke@0 1487 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_LONG)); // Result stub address array index
duke@0 1488 __ jmp(return_from_deopt_common);
duke@0 1489
duke@0 1490 // deopt needs to jump to here to enter the interpreter (return a result)
duke@0 1491
duke@0 1492 deopt_frame_manager_return_ftos = __ pc();
duke@0 1493 // st(0) is live here
duke@0 1494 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index
duke@0 1495 __ jmp(return_from_deopt_common);
duke@0 1496
duke@0 1497 // deopt needs to jump to here to enter the interpreter (return a result)
duke@0 1498 deopt_frame_manager_return_dtos = __ pc();
duke@0 1499
duke@0 1500 // st(0) is live here
duke@0 1501 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index
duke@0 1502 __ jmp(return_from_deopt_common);
duke@0 1503
duke@0 1504 // deopt needs to jump to here to enter the interpreter (return a result)
duke@0 1505 deopt_frame_manager_return_vtos = __ pc();
duke@0 1506
duke@0 1507 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_VOID));
duke@0 1508
duke@0 1509 // Deopt return common
duke@0 1510 // an index is present in rcx that lets us move any possible result being
duke@0 1511 // return to the interpreter's stack
duke@0 1512 //
duke@0 1513 // Because we have a full sized interpreter frame on the youngest
duke@0 1514 // activation the stack is pushed too deep to share the tosca to
duke@0 1515 // stack converters directly. We shrink the stack to the desired
duke@0 1516 // amount and then push result and then re-extend the stack.
duke@0 1517 // We could have the code in size_activation layout a short
duke@0 1518 // frame for the top activation but that would look different
duke@0 1519 // than say sparc (which needs a full size activation because
duke@0 1520 // the windows are in the way. Really it could be short? QQQ
duke@0 1521 //
duke@0 1522 __ bind(return_from_deopt_common);
duke@0 1523
never@304 1524 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
duke@0 1525
duke@0 1526 // setup rsp so we can push the "result" as needed.
never@304 1527 __ movptr(rsp, STATE(_stack)); // trim stack (is prepushed)
never@304 1528 __ addptr(rsp, wordSize); // undo prepush
duke@0 1529
duke@0 1530 ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack);
never@304 1531 // Address index(noreg, rcx, Address::times_ptr);
never@304 1532 __ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_ptr)));
never@304 1533 // __ movl(rcx, Address(noreg, rcx, Address::times_ptr, int(AbstractInterpreter::_tosca_to_stack)));
duke@0 1534 __ call(rcx); // call result converter
duke@0 1535
duke@0 1536 __ movl(STATE(_msg), (int)BytecodeInterpreter::deopt_resume);
never@304 1537 __ lea(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present)
never@304 1538 __ movptr(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed,
duke@0 1539 // result if any on stack already )
never@304 1540 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth
duke@0 1541 }
duke@0 1542
duke@0 1543 // Generate the code to handle a more_monitors message from the c++ interpreter
duke@0 1544 void CppInterpreterGenerator::generate_more_monitors() {
duke@0 1545
duke@0 1546
duke@0 1547 Label entry, loop;
duke@0 1548 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
never@304 1549 // 1. compute new pointers // rsp: old expression stack top
never@304 1550 __ movptr(rdx, STATE(_stack_base)); // rdx: old expression stack bottom
never@304 1551 __ subptr(rsp, entry_size); // move expression stack top limit
never@304 1552 __ subptr(STATE(_stack), entry_size); // update interpreter stack top
never@304 1553 __ subptr(STATE(_stack_limit), entry_size); // inform interpreter
never@304 1554 __ subptr(rdx, entry_size); // move expression stack bottom
never@304 1555 __ movptr(STATE(_stack_base), rdx); // inform interpreter
never@304 1556 __ movptr(rcx, STATE(_stack)); // set start value for copy loop
duke@0 1557 __ jmp(entry);
duke@0 1558 // 2. move expression stack contents
duke@0 1559 __ bind(loop);
never@304 1560 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location
never@304 1561 __ movptr(Address(rcx, 0), rbx); // and store it at new location
never@304 1562 __ addptr(rcx, wordSize); // advance to next word
duke@0 1563 __ bind(entry);
never@304 1564 __ cmpptr(rcx, rdx); // check if bottom reached
never@304 1565 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word
duke@0 1566 // now zero the slot so we can find it.
never@304 1567 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
duke@0 1568 __ movl(STATE(_msg), (int)BytecodeInterpreter::got_monitors);
duke@0 1569 }
duke@0 1570
duke@0 1571
duke@0 1572 // Initial entry to C++ interpreter from the call_stub.
duke@0 1573 // This entry point is called the frame manager since it handles the generation
duke@0 1574 // of interpreter activation frames via requests directly from the vm (via call_stub)
duke@0 1575 // and via requests from the interpreter. The requests from the call_stub happen
duke@0 1576 // directly thru the entry point. Requests from the interpreter happen via returning
duke@0 1577 // from the interpreter and examining the message the interpreter has returned to
duke@0 1578 // the frame manager. The frame manager can take the following requests:
duke@0 1579
duke@0 1580 // NO_REQUEST - error, should never happen.
duke@0 1581 // MORE_MONITORS - need a new monitor. Shuffle the expression stack on down and
duke@0 1582 // allocate a new monitor.
duke@0 1583 // CALL_METHOD - setup a new activation to call a new method. Very similar to what
duke@0 1584 // happens during entry during the entry via the call stub.
duke@0 1585 // RETURN_FROM_METHOD - remove an activation. Return to interpreter or call stub.
duke@0 1586 //
duke@0 1587 // Arguments:
duke@0 1588 //
duke@0 1589 // rbx: methodOop
duke@0 1590 // rcx: receiver - unused (retrieved from stack as needed)
never@304 1591 // rsi/r13: previous frame manager state (NULL from the call_stub/c1/c2)
duke@0 1592 //
duke@0 1593 //
duke@0 1594 // Stack layout at entry
duke@0 1595 //
duke@0 1596 // [ return address ] <--- rsp
duke@0 1597 // [ parameter n ]
duke@0 1598 // ...
duke@0 1599 // [ parameter 1 ]
duke@0 1600 // [ expression stack ]
duke@0 1601 //
duke@0 1602 //
duke@0 1603 // We are free to blow any registers we like because the call_stub which brought us here
duke@0 1604 // initially has preserved the callee save registers already.
duke@0 1605 //
duke@0 1606 //
duke@0 1607
duke@0 1608 static address interpreter_frame_manager = NULL;
duke@0 1609
duke@0 1610 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
duke@0 1611
duke@0 1612 // rbx: methodOop
never@304 1613 // rsi/r13: sender sp
duke@0 1614
duke@0 1615 // Because we redispatch "recursive" interpreter entries thru this same entry point
duke@0 1616 // the "input" register usage is a little strange and not what you expect coming
duke@0 1617 // from the call_stub. From the call stub rsi/rdi (current/previous) interpreter
duke@0 1618 // state are NULL but on "recursive" dispatches they are what you'd expect.
duke@0 1619 // rsi: current interpreter state (C++ interpreter) must preserve (null from call_stub/c1/c2)
duke@0 1620
duke@0 1621
duke@0 1622 // A single frame manager is plenty as we don't specialize for synchronized. We could and
duke@0 1623 // the code is pretty much ready. Would need to change the test below and for good measure
duke@0 1624 // modify generate_interpreter_state to only do the (pre) sync stuff stuff for synchronized
duke@0 1625 // routines. Not clear this is worth it yet.
duke@0 1626
duke@0 1627 if (interpreter_frame_manager) return interpreter_frame_manager;
duke@0 1628
duke@0 1629 address entry_point = __ pc();
duke@0 1630
duke@0 1631 // Fast accessor methods share this entry point.
duke@0 1632 // This works because frame manager is in the same codelet
duke@0 1633 if (UseFastAccessorMethods && !synchronized) __ bind(fast_accessor_slow_entry_path);
duke@0 1634
duke@0 1635 Label dispatch_entry_2;
never@304 1636 __ movptr(rcx, sender_sp_on_entry);
never@304 1637 __ movptr(state, (int32_t)NULL_WORD); // no current activation
duke@0 1638
duke@0 1639 __ jmp(dispatch_entry_2);
duke@0 1640
duke@0 1641 const Register locals = rdi;
duke@0 1642
duke@0 1643 Label re_dispatch;
duke@0 1644
duke@0 1645 __ bind(re_dispatch);
duke@0 1646
duke@0 1647 // save sender sp (doesn't include return address
never@304 1648 __ lea(rcx, Address(rsp, wordSize));
duke@0 1649
duke@0 1650 __ bind(dispatch_entry_2);
duke@0 1651
duke@0 1652 // save sender sp
never@304 1653 __ push(rcx);
duke@0 1654
duke@0 1655 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
duke@0 1656 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset());
duke@0 1657 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
duke@0 1658
duke@0 1659 // const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
duke@0 1660 // const Address monitor_block_bot (rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
duke@0 1661 // const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
duke@0 1662
duke@0 1663 // get parameter size (always needed)
jrose@622 1664 __ load_unsigned_short(rcx, size_of_parameters);
duke@0 1665
duke@0 1666 // rbx: methodOop
duke@0 1667 // rcx: size of parameters
jrose@622 1668 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
duke@0 1669
never@304 1670 __ subptr(rdx, rcx); // rdx = no. of additional locals
duke@0 1671
duke@0 1672 // see if we've got enough room on the stack for locals plus overhead.
duke@0 1673 generate_stack_overflow_check(); // C++
duke@0 1674
duke@0 1675 // c++ interpreter does not use stack banging or any implicit exceptions
duke@0 1676 // leave for now to verify that check is proper.
duke@0 1677 bang_stack_shadow_pages(false);
duke@0 1678
duke@0 1679
duke@0 1680
duke@0 1681 // compute beginning of parameters (rdi)
never@304 1682 __ lea(locals, Address(rsp, rcx, Address::times_ptr, wordSize));
duke@0 1683
duke@0 1684 // save sender's sp
duke@0 1685 // __ movl(rcx, rsp);
duke@0 1686
duke@0 1687 // get sender's sp
never@304 1688 __ pop(rcx);
duke@0 1689
duke@0 1690 // get return address
never@304 1691 __ pop(rax);
duke@0 1692
duke@0 1693 // rdx - # of additional locals
duke@0 1694 // allocate space for locals
duke@0 1695 // explicitly initialize locals
duke@0 1696 {
duke@0 1697 Label exit, loop;
never@304 1698 __ testl(rdx, rdx); // (32bit ok)
duke@0 1699 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
duke@0 1700 __ bind(loop);
never@304 1701 __ push((int32_t)NULL_WORD); // initialize local variables
duke@0 1702 __ decrement(rdx); // until everything initialized
duke@0 1703 __ jcc(Assembler::greater, loop);
duke@0 1704 __ bind(exit);
duke@0 1705 }
duke@0 1706
duke@0 1707
duke@0 1708 // Assumes rax = return address
duke@0 1709
duke@0 1710 // allocate and initialize new interpreterState and method expression stack
duke@0 1711 // IN(locals) -> locals
duke@0 1712 // IN(state) -> any current interpreter activation
duke@0 1713 // destroys rax, rcx, rdx, rdi
duke@0 1714 // OUT (state) -> new interpreterState
duke@0 1715 // OUT(rsp) -> bottom of methods expression stack
duke@0 1716
duke@0 1717 generate_compute_interpreter_state(state, locals, rcx, false);
duke@0 1718
duke@0 1719 // Call interpreter
duke@0 1720
duke@0 1721 Label call_interpreter;
duke@0 1722 __ bind(call_interpreter);
duke@0 1723
duke@0 1724 // c++ interpreter does not use stack banging or any implicit exceptions
duke@0 1725 // leave for now to verify that check is proper.
duke@0 1726 bang_stack_shadow_pages(false);
duke@0 1727
duke@0 1728
duke@0 1729 // Call interpreter enter here if message is
duke@0 1730 // set and we know stack size is valid
duke@0 1731
duke@0 1732 Label call_interpreter_2;
duke@0 1733
duke@0 1734 __ bind(call_interpreter_2);
duke@0 1735
duke@0 1736 {
never@304 1737 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
never@304 1738
never@304 1739 #ifdef _LP64
never@304 1740 __ mov(c_rarg0, state);
never@304 1741 #else
never@304 1742 __ push(state); // push arg to interpreter
never@304 1743 __ movptr(thread, STATE(_thread));
never@304 1744 #endif // _LP64
duke@0 1745
duke@0 1746 // We can setup the frame anchor with everything we want at this point
duke@0 1747 // as we are thread_in_Java and no safepoints can occur until we go to
duke@0 1748 // vm mode. We do have to clear flags on return from vm but that is it
duke@0 1749 //
never@304 1750 __ movptr(Address(thread, JavaThread::last_Java_fp_offset()), rbp);
never@304 1751 __ movptr(Address(thread, JavaThread::last_Java_sp_offset()), rsp);
duke@0 1752
duke@0 1753 // Call the interpreter
duke@0 1754
duke@0 1755 RuntimeAddress normal(CAST_FROM_FN_PTR(address, BytecodeInterpreter::run));
duke@0 1756 RuntimeAddress checking(CAST_FROM_FN_PTR(address, BytecodeInterpreter::runWithChecks));
duke@0 1757
duke@0 1758 __ call(JvmtiExport::can_post_interpreter_events() ? checking : normal);
never@304 1759 NOT_LP64(__ pop(rax);) // discard parameter to run
duke@0 1760 //
duke@0 1761 // state is preserved since it is callee saved
duke@0 1762 //
duke@0 1763
duke@0 1764 // reset_last_Java_frame
duke@0 1765
never@304 1766 NOT_LP64(__ movl(thread, STATE(_thread));)
duke@0 1767 __ reset_last_Java_frame(thread, true, true);
duke@0 1768 }
duke@0 1769
duke@0 1770 // examine msg from interpreter to determine next action
duke@0 1771
duke@0 1772 __ movl(rdx, STATE(_msg)); // Get new message
duke@0 1773
duke@0 1774 Label call_method;
duke@0 1775 Label return_from_interpreted_method;
duke@0 1776 Label throw_exception;
duke@0 1777 Label bad_msg;
duke@0 1778 Label do_OSR;
duke@0 1779
never@304 1780 __ cmpl(rdx, (int32_t)BytecodeInterpreter::call_method);
duke@0 1781 __ jcc(Assembler::equal, call_method);
never@304 1782 __ cmpl(rdx, (int32_t)BytecodeInterpreter::return_from_method);
duke@0 1783 __ jcc(Assembler::equal, return_from_interpreted_method);
never@304 1784 __ cmpl(rdx, (int32_t)BytecodeInterpreter::do_osr);
duke@0 1785 __ jcc(Assembler::equal, do_OSR);
never@304 1786 __ cmpl(rdx, (int32_t)BytecodeInterpreter::throwing_exception);
duke@0 1787 __ jcc(Assembler::equal, throw_exception);
never@304 1788 __ cmpl(rdx, (int32_t)BytecodeInterpreter::more_monitors);
duke@0 1789 __ jcc(Assembler::notEqual, bad_msg);
duke@0 1790
duke@0 1791 // Allocate more monitor space, shuffle expression stack....
duke@0 1792
duke@0 1793 generate_more_monitors();
duke@0 1794
duke@0 1795 __ jmp(call_interpreter);
duke@0 1796
duke@0 1797 // uncommon trap needs to jump to here to enter the interpreter (re-execute current bytecode)
duke@0 1798 unctrap_frame_manager_entry = __ pc();
duke@0 1799 //
duke@0 1800 // Load the registers we need.
never@304 1801 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
never@304 1802 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth
duke@0 1803 __ jmp(call_interpreter_2);
duke@0 1804
duke@0 1805
duke@0 1806
duke@0 1807 //=============================================================================
duke@0 1808 // Returning from a compiled method into a deopted method. The bytecode at the
duke@0 1809 // bcp has completed. The result of the bytecode is in the native abi (the tosca
duke@0 1810 // for the template based interpreter). Any stack space that was used by the
duke@0 1811 // bytecode that has completed has been removed (e.g. parameters for an invoke)
duke@0 1812 // so all that we have to do is place any pending result on the expression stack
duke@0 1813 // and resume execution on the next bytecode.
duke@0 1814
duke@0 1815
duke@0 1816 generate_deopt_handling();
duke@0 1817 __ jmp(call_interpreter);
duke@0 1818
duke@0 1819
duke@0 1820 // Current frame has caught an exception we need to dispatch to the
duke@0 1821 // handler. We can get here because a native interpreter frame caught
duke@0 1822 // an exception in which case there is no handler and we must rethrow
duke@0 1823 // If it is a vanilla interpreted frame the we simply drop into the
duke@0 1824 // interpreter and let it do the lookup.
duke@0 1825
duke@0 1826 Interpreter::_rethrow_exception_entry = __ pc();
duke@0 1827 // rax: exception
duke@0 1828 // rdx: return address/pc that threw exception
duke@0 1829
duke@0 1830 Label return_with_exception;
duke@0 1831 Label unwind_and_forward;
duke@0 1832
duke@0 1833 // restore state pointer.
coleenp@520 1834 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
never@304 1835
never@304 1836 __ movptr(rbx, STATE(_method)); // get method
never@304 1837 #ifdef _LP64
never@304 1838 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
never@304 1839 #else
duke@0 1840 __ movl(rcx, STATE(_thread)); // get thread
duke@0 1841
duke@0 1842 // Store exception with interpreter will expect it
never@304 1843 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax);
never@304 1844 #endif // _LP64
duke@0 1845
duke@0 1846 // is current frame vanilla or native?
duke@0 1847
duke@0 1848 __ movl(rdx, access_flags);
duke@0 1849 __ testl(rdx, JVM_ACC_NATIVE);
duke@0 1850 __ jcc(Assembler::zero, return_with_exception); // vanilla interpreted frame, handle directly
duke@0 1851
duke@0 1852 // We drop thru to unwind a native interpreted frame with a pending exception
duke@0 1853 // We jump here for the initial interpreter frame with exception pending
duke@0 1854 // We unwind the current acivation and forward it to our caller.
duke@0 1855
duke@0 1856 __ bind(unwind_and_forward);
duke@0 1857
duke@0 1858 // unwind rbp, return stack to unextended value and re-push return address
duke@0 1859
never@304 1860 __ movptr(rcx, STATE(_sender_sp));
duke@0 1861 __ leave();
never@304 1862 __ pop(rdx);
never@304 1863 __ mov(rsp, rcx);
never@304 1864 __ push(rdx);
duke@0 1865 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
duke@0 1866
duke@0 1867 // Return point from a call which returns a result in the native abi
duke@0 1868 // (c1/c2/jni-native). This result must be processed onto the java
duke@0 1869 // expression stack.
duke@0 1870 //
duke@0 1871 // A pending exception may be present in which case there is no result present
duke@0 1872
duke@0 1873 Label resume_interpreter;
duke@0 1874 Label do_float;
duke@0 1875 Label do_double;
duke@0 1876 Label done_conv;
duke@0 1877
duke@0 1878 address compiled_entry = __ pc();
duke@0 1879
duke@0 1880 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
duke@0 1881 if (UseSSE < 2) {
coleenp@520 1882 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
never@304 1883 __ movptr(rbx, STATE(_result._to_call._callee)); // get method just executed
duke@0 1884 __ movl(rcx, Address(rbx, methodOopDesc::result_index_offset()));
duke@0 1885 __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index
duke@0 1886 __ jcc(Assembler::equal, do_float);
duke@0 1887 __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index
duke@0 1888 __ jcc(Assembler::equal, do_double);
coleenp@520 1889 #if !defined(_LP64) || defined(COMPILER1) || !defined(COMPILER2)
duke@0 1890 __ empty_FPU_stack();
duke@0 1891 #endif // COMPILER2
duke@0 1892 __ jmp(done_conv);
duke@0 1893
duke@0 1894 __ bind(do_float);
duke@0 1895 #ifdef COMPILER2
duke@0 1896 for (int i = 1; i < 8; i++) {
duke@0 1897 __ ffree(i);
duke@0 1898 }
duke@0 1899 #endif // COMPILER2
duke@0 1900 __ jmp(done_conv);
duke@0 1901 __ bind(do_double);
duke@0 1902 #ifdef COMPILER2
duke@0 1903 for (int i = 1; i < 8; i++) {
duke@0 1904 __ ffree(i);
duke@0 1905 }
duke@0 1906 #endif // COMPILER2
duke@0 1907 __ jmp(done_conv);
duke@0 1908 } else {
duke@0 1909 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
duke@0 1910 __ jmp(done_conv);
duke@0 1911 }
duke@0 1912
never@304 1913 #if 0
duke@0 1914 // emit a sentinel we can test for when converting an interpreter
duke@0 1915 // entry point to a compiled entry point.
duke@0 1916 __ a_long(Interpreter::return_sentinel);
duke@0 1917 __ a_long((int)compiled_entry);
never@304 1918 #endif
duke@0 1919
duke@0 1920 // Return point to interpreter from compiled/native method
duke@0 1921
duke@0 1922 InternalAddress return_from_native_method(__ pc());
duke@0 1923
duke@0 1924 __ bind(done_conv);
duke@0 1925
duke@0 1926
duke@0 1927 // Result if any is in tosca. The java expression stack is in the state that the
duke@0 1928 // calling convention left it (i.e. params may or may not be present)
duke@0 1929 // Copy the result from tosca and place it on java expression stack.
duke@0 1930
never@304 1931 // Restore rsi/r13 as compiled code may not preserve it
never@304 1932
coleenp@520 1933 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
duke@0 1934
duke@0 1935 // restore stack to what we had when we left (in case i2c extended it)
duke@0 1936
never@304 1937 __ movptr(rsp, STATE(_stack));
never@304 1938 __ lea(rsp, Address(rsp, wordSize));
duke@0 1939
duke@0 1940 // If there is a pending exception then we don't really have a result to process
duke@0 1941
never@304 1942 #ifdef _LP64
never@304 1943 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
never@304 1944 #else
never@304 1945 __ movptr(rcx, STATE(_thread)); // get thread
never@304 1946 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
coleenp@520 1947 #endif // _LP64
duke@0 1948 __ jcc(Assembler::notZero, return_with_exception);
duke@0 1949
duke@0 1950 // get method just executed
never@304 1951 __ movptr(rbx, STATE(_result._to_call._callee));
duke@0 1952
duke@0 1953 // callee left args on top of expression stack, remove them
jrose@622 1954 __ load_unsigned_short(rcx, Address(rbx, methodOopDesc::size_of_parameters_offset()));
never@304 1955 __ lea(rsp, Address(rsp, rcx, Address::times_ptr));
duke@0 1956
duke@0 1957 __ movl(rcx, Address(rbx, methodOopDesc::result_index_offset()));
duke@0 1958 ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack);
never@304 1959 // Address index(noreg, rax, Address::times_ptr);
never@304 1960 __ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_ptr)));
never@304 1961 // __ movl(rcx, Address(noreg, rcx, Address::times_ptr, int(AbstractInterpreter::_tosca_to_stack)));
duke@0 1962 __ call(rcx); // call result converter
duke@0 1963 __ jmp(resume_interpreter);
duke@0 1964
duke@0 1965 // An exception is being caught on return to a vanilla interpreter frame.
duke@0 1966 // Empty the stack and resume interpreter
duke@0 1967
duke@0 1968 __ bind(return_with_exception);
duke@0 1969
duke@0 1970 // Exception present, empty stack
never@304 1971 __ movptr(rsp, STATE(_stack_base));
duke@0 1972 __ jmp(resume_interpreter);
duke@0 1973
duke@0 1974 // Return from interpreted method we return result appropriate to the caller (i.e. "recursive"
duke@0 1975 // interpreter call, or native) and unwind this interpreter activation.
duke@0 1976 // All monitors should be unlocked.
duke@0 1977
duke@0 1978 __ bind(return_from_interpreted_method);
duke@0 1979
duke@0 1980 Label return_to_initial_caller;
duke@0 1981
never@304 1982 __ movptr(rbx, STATE(_method)); // get method just executed
never@304 1983 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from "recursive" interpreter call?
duke@0 1984 __ movl(rax, Address(rbx, methodOopDesc::result_index_offset())); // get result type index
duke@0 1985 __ jcc(Assembler::equal, return_to_initial_caller); // back to native code (call_stub/c1/c2)
duke@0 1986
duke@0 1987 // Copy result to callers java stack
duke@0 1988 ExternalAddress stack_to_stack((address)CppInterpreter::_stack_to_stack);
never@304 1989 // Address index(noreg, rax, Address::times_ptr);
never@304 1990
never@304 1991 __ movptr(rax, ArrayAddress(stack_to_stack, Address(noreg, rax, Address::times_ptr)));
never@304 1992 // __ movl(rax, Address(noreg, rax, Address::times_ptr, int(AbstractInterpreter::_stack_to_stack)));
duke@0 1993 __ call(rax); // call result converter
duke@0 1994
duke@0 1995 Label unwind_recursive_activation;
duke@0 1996 __ bind(unwind_recursive_activation);
duke@0 1997
duke@0 1998 // returning to interpreter method from "recursive" interpreter call
duke@0 1999 // result converter left rax pointing to top of the java stack for method we are returning
duke@0 2000 // to. Now all we must do is unwind the state from the completed call
duke@0 2001
never@304 2002 __ movptr(state, STATE(_prev_link)); // unwind state
duke@0 2003 __ leave(); // pop the frame
never@304 2004 __ mov(rsp, rax); // unwind stack to remove args
duke@0 2005
duke@0 2006 // Resume the interpreter. The current frame contains the current interpreter
duke@0 2007 // state object.
duke@0 2008 //
duke@0 2009
duke@0 2010 __ bind(resume_interpreter);
duke@0 2011
duke@0 2012 // state == interpreterState object for method we are resuming
duke@0 2013
duke@0 2014 __ movl(STATE(_msg), (int)BytecodeInterpreter::method_resume);
never@304 2015 __ lea(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present)
never@304 2016 __ movptr(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed,
duke@0 2017 // result if any on stack already )
never@304 2018 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth
duke@0 2019 __ jmp(call_interpreter_2); // No need to bang
duke@0 2020
duke@0 2021 // interpreter returning to native code (call_stub/c1/c2)
duke@0 2022 // convert result and unwind initial activation
duke@0 2023 // rax - result index
duke@0 2024
duke@0 2025 __ bind(return_to_initial_caller);
duke@0 2026 ExternalAddress stack_to_native((address)CppInterpreter::_stack_to_native_abi);
never@304 2027 // Address index(noreg, rax, Address::times_ptr);
never@304 2028
never@304 2029 __ movptr(rax, ArrayAddress(stack_to_native, Address(noreg, rax, Address::times_ptr)));
duke@0 2030 __ call(rax); // call result converter
duke@0 2031
duke@0 2032 Label unwind_initial_activation;
duke@0 2033 __ bind(unwind_initial_activation);
duke@0 2034
duke@0 2035 // RETURN TO CALL_STUB/C1/C2 code (result if any in rax/rdx ST(0))
duke@0 2036
duke@0 2037 /* Current stack picture
duke@0 2038
duke@0 2039 [ incoming parameters ]
duke@0 2040 [ extra locals ]
duke@0 2041 [ return address to CALL_STUB/C1/C2]
duke@0 2042 fp -> [ CALL_STUB/C1/C2 fp ]
duke@0 2043 BytecodeInterpreter object
duke@0 2044 expression stack
duke@0 2045 sp ->
duke@0 2046
duke@0 2047 */
duke@0 2048
duke@0 2049 // return restoring the stack to the original sender_sp value
duke@0 2050
never@304 2051 __ movptr(rcx, STATE(_sender_sp));
duke@0 2052 __ leave();
never@304 2053 __ pop(rdi); // get return address
duke@0 2054 // set stack to sender's sp
never@304 2055 __ mov(rsp, rcx);
duke@0 2056 __ jmp(rdi); // return to call_stub
duke@0 2057
duke@0 2058 // OSR request, adjust return address to make current frame into adapter frame
duke@0 2059 // and enter OSR nmethod
duke@0 2060
duke@0 2061 __ bind(do_OSR);
duke@0 2062
duke@0 2063 Label remove_initial_frame;
duke@0 2064
duke@0 2065 // We are going to pop this frame. Is there another interpreter frame underneath
duke@0 2066 // it or is it callstub/compiled?
duke@0 2067
duke@0 2068 // Move buffer to the expected parameter location
never@304 2069 __ movptr(rcx, STATE(_result._osr._osr_buf));
never@304 2070
never@304 2071 __ movptr(rax, STATE(_result._osr._osr_entry));
never@304 2072
never@304 2073 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from "recursive" interpreter call?
duke@0 2074 __ jcc(Assembler::equal, remove_initial_frame); // back to native code (call_stub/c1/c2)
duke@0 2075
never@304 2076 __ movptr(sender_sp_on_entry, STATE(_sender_sp)); // get sender's sp in expected register
duke@0 2077 __ leave(); // pop the frame
never@304 2078 __ mov(rsp, sender_sp_on_entry); // trim any stack expansion
duke@0 2079
duke@0 2080
duke@0 2081 // We know we are calling compiled so push specialized return
duke@0 2082 // method uses specialized entry, push a return so we look like call stub setup
duke@0 2083 // this path will handle fact that result is returned in registers and not
duke@0 2084 // on the java stack.
duke@0 2085
duke@0 2086 __ pushptr(return_from_native_method.addr());
duke@0 2087
duke@0 2088 __ jmp(rax);
duke@0 2089
duke@0 2090 __ bind(remove_initial_frame);
duke@0 2091
never@304 2092 __ movptr(rdx, STATE(_sender_sp));
duke@0 2093 __ leave();
duke@0 2094 // get real return
never@304 2095 __ pop(rsi);
duke@0 2096 // set stack to sender's sp
never@304 2097 __ mov(rsp, rdx);
duke@0 2098 // repush real return
never@304 2099 __ push(rsi);
duke@0 2100 // Enter OSR nmethod
duke@0 2101 __ jmp(rax);
duke@0 2102
duke@0 2103
duke@0 2104
duke@0 2105
duke@0 2106 // Call a new method. All we do is (temporarily) trim the expression stack
duke@0 2107 // push a return address to bring us back to here and leap to the new entry.
duke@0 2108
duke@0 2109 __ bind(call_method);
duke@0 2110
duke@0 2111 // stack points to next free location and not top element on expression stack
duke@0 2112 // method expects sp to be pointing to topmost element
duke@0 2113
never@304 2114 __ movptr(rsp, STATE(_stack)); // pop args to c++ interpreter, set sp to java stack top
never@304 2115 __ lea(rsp, Address(rsp, wordSize));
never@304 2116
never@304 2117 __ movptr(rbx, STATE(_result._to_call._callee)); // get method to execute
duke@0 2118
duke@0 2119 // don't need a return address if reinvoking interpreter
duke@0 2120
duke@0 2121 // Make it look like call_stub calling conventions
duke@0 2122
duke@0 2123 // Get (potential) receiver
jrose@622 2124 __ load_unsigned_short(rcx, size_of_parameters); // get size of parameters in words
duke@0 2125
duke@0 2126 ExternalAddress recursive(CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation));
duke@0 2127 __ pushptr(recursive.addr()); // make it look good in the debugger
duke@0 2128
duke@0 2129 InternalAddress entry(entry_point);
duke@0 2130 __ cmpptr(STATE(_result._to_call._callee_entry_point), entry.addr()); // returning to interpreter?
duke@0 2131 __ jcc(Assembler::equal, re_dispatch); // yes
duke@0 2132
never@304 2133 __ pop(rax); // pop dummy address
duke@0 2134
duke@0 2135
duke@0 2136 // get specialized entry
never@304 2137 __ movptr(rax, STATE(_result._to_call._callee_entry_point));
duke@0 2138 // set sender SP
never@304 2139 __ mov(sender_sp_on_entry, rsp);
duke@0 2140
duke@0 2141 // method uses specialized entry, push a return so we look like call stub setup
duke@0 2142 // this path will handle fact that result is returned in registers and not
duke@0 2143 // on the java stack.
duke@0 2144
duke@0 2145 __ pushptr(return_from_native_method.addr());
duke@0 2146
duke@0 2147 __ jmp(rax);
duke@0 2148
duke@0 2149 __ bind(bad_msg);
duke@0 2150 __ stop("Bad message from interpreter");
duke@0 2151
duke@0 2152 // Interpreted method "returned" with an exception pass it on...
duke@0 2153 // Pass result, unwind activation and continue/return to interpreter/call_stub
duke@0 2154 // We handle result (if any) differently based on return to interpreter or call_stub
duke@0 2155
duke@0 2156 Label unwind_initial_with_pending_exception;
duke@0 2157
duke@0 2158 __ bind(throw_exception);
never@304 2159 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from recursive interpreter call?
duke@0 2160 __ jcc(Assembler::equal, unwind_initial_with_pending_exception); // no, back to native code (call_stub/c1/c2)
never@304 2161 __ movptr(rax, STATE(_locals)); // pop parameters get new stack value
never@304 2162 __ addptr(rax, wordSize); // account for prepush before we return
duke@0 2163 __ jmp(unwind_recursive_activation);
duke@0 2164
duke@0 2165 __ bind(unwind_initial_with_pending_exception);
duke@0 2166
duke@0 2167 // We will unwind the current (initial) interpreter frame and forward
duke@0 2168 // the exception to the caller. We must put the exception in the
duke@0 2169 // expected register and clear pending exception and then forward.
duke@0 2170
duke@0 2171 __ jmp(unwind_and_forward);
duke@0 2172
duke@0 2173 interpreter_frame_manager = entry_point;
duke@0 2174 return entry_point;
duke@0 2175 }
duke@0 2176
duke@0 2177 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
duke@0 2178 // determine code generation flags
duke@0 2179 bool synchronized = false;
duke@0 2180 address entry_point = NULL;
duke@0 2181
duke@0 2182 switch (kind) {
duke@0 2183 case Interpreter::zerolocals : break;
duke@0 2184 case Interpreter::zerolocals_synchronized: synchronized = true; break;
duke@0 2185 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
duke@0 2186 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
duke@0 2187 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
duke@0 2188 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
duke@0 2189 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
jrose@710 2190 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
duke@0 2191
duke@0 2192 case Interpreter::java_lang_math_sin : // fall thru
duke@0 2193 case Interpreter::java_lang_math_cos : // fall thru
duke@0 2194 case Interpreter::java_lang_math_tan : // fall thru
duke@0 2195 case Interpreter::java_lang_math_abs : // fall thru
duke@0 2196 case Interpreter::java_lang_math_log : // fall thru
duke@0 2197 case Interpreter::java_lang_math_log10 : // fall thru
duke@0 2198 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
duke@0 2199 default : ShouldNotReachHere(); break;
duke@0 2200 }
duke@0 2201
duke@0 2202 if (entry_point) return entry_point;
duke@0 2203
duke@0 2204 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
duke@0 2205
duke@0 2206 }
duke@0 2207
duke@0 2208 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
duke@0 2209 : CppInterpreterGenerator(code) {
duke@0 2210 generate_all(); // down here so it can be "virtual"
duke@0 2211 }
duke@0 2212
duke@0 2213 // Deoptimization helpers for C++ interpreter
duke@0 2214
duke@0 2215 // How much stack a method activation needs in words.
duke@0 2216 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
duke@0 2217
duke@0 2218 const int stub_code = 4; // see generate_call_stub
duke@0 2219 // Save space for one monitor to get into the interpreted method in case
duke@0 2220 // the method is synchronized
duke@0 2221 int monitor_size = method->is_synchronized() ?
duke@0 2222 1*frame::interpreter_frame_monitor_size() : 0;
duke@0 2223
duke@0 2224 // total static overhead size. Account for interpreter state object, return
duke@0 2225 // address, saved rbp and 2 words for a "static long no_params() method" issue.
duke@0 2226
duke@0 2227 const int overhead_size = sizeof(BytecodeInterpreter)/wordSize +
duke@0 2228 ( frame::sender_sp_offset - frame::link_offset) + 2;
duke@0 2229
jrose@710 2230 const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries();
jrose@710 2231 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
duke@0 2232 Interpreter::stackElementWords();
duke@0 2233 return overhead_size + method_stack + stub_code;
duke@0 2234 }
duke@0 2235
duke@0 2236 // returns the activation size.
duke@0 2237 static int size_activation_helper(int extra_locals_size, int monitor_size) {
duke@0 2238 return (extra_locals_size + // the addition space for locals
duke@0 2239 2*BytesPerWord + // return address and saved rbp
duke@0 2240 2*BytesPerWord + // "static long no_params() method" issue
duke@0 2241 sizeof(BytecodeInterpreter) + // interpreterState
duke@0 2242 monitor_size); // monitors
duke@0 2243 }
duke@0 2244
duke@0 2245 void BytecodeInterpreter::layout_interpreterState(interpreterState to_fill,
duke@0 2246 frame* caller,
duke@0 2247 frame* current,
duke@0 2248 methodOop method,
duke@0 2249 intptr_t* locals,
duke@0 2250 intptr_t* stack,
duke@0 2251 intptr_t* stack_base,
duke@0 2252 intptr_t* monitor_base,
duke@0 2253 intptr_t* frame_bottom,
duke@0 2254 bool is_top_frame
duke@0 2255 )
duke@0 2256 {
duke@0 2257 // What about any vtable?
duke@0 2258 //
duke@0 2259 to_fill->_thread = JavaThread::current();
duke@0 2260 // This gets filled in later but make it something recognizable for now
duke@0 2261 to_fill->_bcp = method->code_base();
duke@0 2262 to_fill->_locals = locals;
duke@0 2263 to_fill->_constants = method->constants()->cache();
duke@0 2264 to_fill->_method = method;
duke@0 2265 to_fill->_mdx = NULL;
duke@0 2266 to_fill->_stack = stack;
duke@0 2267 if (is_top_frame && JavaThread::current()->popframe_forcing_deopt_reexecution() ) {
duke@0 2268 to_fill->_msg = deopt_resume2;
duke@0 2269 } else {
duke@0 2270 to_fill->_msg = method_resume;
duke@0 2271 }
duke@0 2272 to_fill->_result._to_call._bcp_advance = 0;
duke@0 2273 to_fill->_result._to_call._callee_entry_point = NULL; // doesn't matter to anyone
duke@0 2274 to_fill->_result._to_call._callee = NULL; // doesn't matter to anyone
duke@0 2275 to_fill->_prev_link = NULL;
duke@0 2276
duke@0 2277 to_fill->_sender_sp = caller->unextended_sp();
duke@0 2278
duke@0 2279 if (caller->is_interpreted_frame()) {
duke@0 2280 interpreterState prev = caller->get_interpreterState();
duke@0 2281 to_fill->_prev_link = prev;
duke@0 2282 // *current->register_addr(GR_Iprev_state) = (intptr_t) prev;
duke@0 2283 // Make the prev callee look proper
duke@0 2284 prev->_result._to_call._callee = method;
duke@0 2285 if (*prev->_bcp == Bytecodes::_invokeinterface) {
duke@0 2286 prev->_result._to_call._bcp_advance = 5;
duke@0 2287 } else {
duke@0 2288 prev->_result._to_call._bcp_advance = 3;
duke@0 2289 }
duke@0 2290 }
duke@0 2291 to_fill->_oop_temp = NULL;
duke@0 2292 to_fill->_stack_base = stack_base;
duke@0 2293 // Need +1 here because stack_base points to the word just above the first expr stack entry
duke@0 2294 // and stack_limit is supposed to point to the word just below the last expr stack entry.
duke@0 2295 // See generate_compute_interpreter_state.
jrose@710 2296 int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries();
jrose@710 2297 to_fill->_stack_limit = stack_base - (method->max_stack() + extra_stack + 1);
duke@0 2298 to_fill->_monitor_base = (BasicObjectLock*) monitor_base;
duke@0 2299
duke@0 2300 to_fill->_self_link = to_fill;
duke@0 2301 assert(stack >= to_fill->_stack_limit && stack < to_fill->_stack_base,
duke@0 2302 "Stack top out of range");
duke@0 2303 }
duke@0 2304
duke@0 2305 int AbstractInterpreter::layout_activation(methodOop method,
duke@0 2306 int tempcount, //
duke@0 2307 int popframe_extra_args,
duke@0 2308 int moncount,
duke@0 2309 int callee_param_count,
duke@0 2310 int callee_locals,
duke@0 2311 frame* caller,
duke@0 2312 frame* interpreter_frame,
duke@0 2313 bool is_top_frame) {
duke@0 2314
duke@0 2315 assert(popframe_extra_args == 0, "FIX ME");
duke@0 2316 // NOTE this code must exactly mimic what InterpreterGenerator::generate_compute_interpreter_state()
duke@0 2317 // does as far as allocating an interpreter frame.
duke@0 2318 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
duke@0 2319 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
duke@0 2320 // as determined by a previous call to this method.
duke@0 2321 // It is also guaranteed to be walkable even though it is in a skeletal state
duke@0 2322 // NOTE: return size is in words not bytes
duke@0 2323 // NOTE: tempcount is the current size of the java expression stack. For top most
duke@0 2324 // frames we will allocate a full sized expression stack and not the curback
duke@0 2325 // version that non-top frames have.
duke@0 2326
duke@0 2327 // Calculate the amount our frame will be adjust by the callee. For top frame
duke@0 2328 // this is zero.
duke@0 2329
duke@0 2330 // NOTE: ia64 seems to do this wrong (or at least backwards) in that it
duke@0 2331 // calculates the extra locals based on itself. Not what the callee does
duke@0 2332 // to it. So it ignores last_frame_adjust value. Seems suspicious as far
duke@0 2333 // as getting sender_sp correct.
duke@0 2334
duke@0 2335 int extra_locals_size = (callee_locals - callee_param_count) * BytesPerWord;
duke@0 2336 int monitor_size = sizeof(BasicObjectLock) * moncount;
duke@0 2337
duke@0 2338 // First calculate the frame size without any java expression stack
duke@0 2339 int short_frame_size = size_activation_helper(extra_locals_size,
duke@0 2340 monitor_size);
duke@0 2341
duke@0 2342 // Now with full size expression stack
jrose@710 2343 int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries();
jrose@710 2344 int full_frame_size = short_frame_size + (method->max_stack() + extra_stack) * BytesPerWord;
duke@0 2345
duke@0 2346 // and now with only live portion of the expression stack
duke@0 2347 short_frame_size = short_frame_size + tempcount * BytesPerWord;
duke@0 2348
duke@0 2349 // the size the activation is right now. Only top frame is full size
duke@0 2350 int frame_size = (is_top_frame ? full_frame_size : short_frame_size);
duke@0 2351
duke@0 2352 if (interpreter_frame != NULL) {
duke@0 2353 #ifdef ASSERT
duke@0 2354 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
duke@0 2355 #endif
duke@0 2356
duke@0 2357 // MUCHO HACK
duke@0 2358
duke@0 2359 intptr_t* frame_bottom = (intptr_t*) ((intptr_t)interpreter_frame->sp() - (full_frame_size - frame_size));
duke@0 2360
duke@0 2361 /* Now fillin the interpreterState object */
duke@0 2362
duke@0 2363 // The state object is the first thing on the frame and easily located
duke@0 2364
duke@0 2365 interpreterState cur_state = (interpreterState) ((intptr_t)interpreter_frame->fp() - sizeof(BytecodeInterpreter));
duke@0 2366
duke@0 2367
duke@0 2368 // Find the locals pointer. This is rather simple on x86 because there is no
duke@0 2369 // confusing rounding at the callee to account for. We can trivially locate
duke@0 2370 // our locals based on the current fp().
duke@0 2371 // Note: the + 2 is for handling the "static long no_params() method" issue.
duke@0 2372 // (too bad I don't really remember that issue well...)
duke@0 2373
duke@0 2374 intptr_t* locals;
duke@0 2375 // If the caller is interpreted we need to make sure that locals points to the first
duke@0 2376 // argument that the caller passed and not in an area where the stack might have been extended.
duke@0 2377 // because the stack to stack to converter needs a proper locals value in order to remove the
duke@0 2378 // arguments from the caller and place the result in the proper location. Hmm maybe it'd be
duke@0 2379 // simpler if we simply stored the result in the BytecodeInterpreter object and let the c++ code
duke@0 2380 // adjust the stack?? HMMM QQQ
duke@0 2381 //
duke@0 2382 if (caller->is_interpreted_frame()) {
duke@0 2383 // locals must agree with the caller because it will be used to set the
duke@0 2384 // caller's tos when we return.
duke@0 2385 interpreterState prev = caller->get_interpreterState();
duke@0 2386 // stack() is prepushed.
duke@0 2387 locals = prev->stack() + method->size_of_parameters();
duke@0 2388 // locals = caller->unextended_sp() + (method->size_of_parameters() - 1);
duke@0 2389 if (locals != interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2) {
duke@0 2390 // os::breakpoint();
duke@0 2391 }
duke@0 2392 } else {
duke@0 2393 // this is where a c2i would have placed locals (except for the +2)
duke@0 2394 locals = interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2;
duke@0 2395 }
duke@0 2396
duke@0 2397 intptr_t* monitor_base = (intptr_t*) cur_state;
duke@0 2398 intptr_t* stack_base = (intptr_t*) ((intptr_t) monitor_base - monitor_size);
duke@0 2399 /* +1 because stack is always prepushed */
duke@0 2400 intptr_t* stack = (intptr_t*) ((intptr_t) stack_base - (tempcount + 1) * BytesPerWord);
duke@0 2401
duke@0 2402
duke@0 2403 BytecodeInterpreter::layout_interpreterState(cur_state,
duke@0 2404 caller,
duke@0 2405 interpreter_frame,
duke@0 2406 method,
duke@0 2407 locals,
duke@0 2408 stack,
duke@0 2409 stack_base,
duke@0 2410 monitor_base,
duke@0 2411 frame_bottom,
duke@0 2412 is_top_frame);
duke@0 2413
duke@0 2414 // BytecodeInterpreter::pd_layout_interpreterState(cur_state, interpreter_return_address, interpreter_frame->fp());
duke@0 2415 }
duke@0 2416 return frame_size/BytesPerWord;
duke@0 2417 }
duke@0 2418
duke@0 2419 #endif // CC_INTERP (all)