annotate src/cpu/x86/vm/templateInterpreter_x86_32.cpp @ 5604:bd3237e0e18d

8026328: Setting a breakpoint on invokedynamic crashes the JVM Reviewed-by: jrose, roland
author twisti
date Thu, 24 Oct 2013 16:23:07 -0700
parents 5ccbab1c69f3
children
rev   line source
duke@0 1 /*
jiangli@4501 2 * Copyright (c) 1997, 2013, 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
stefank@1879 25 #include "precompiled.hpp"
twisti@3883 26 #include "asm/macroAssembler.hpp"
stefank@1879 27 #include "interpreter/bytecodeHistogram.hpp"
stefank@1879 28 #include "interpreter/interpreter.hpp"
stefank@1879 29 #include "interpreter/interpreterGenerator.hpp"
stefank@1879 30 #include "interpreter/interpreterRuntime.hpp"
stefank@1879 31 #include "interpreter/templateTable.hpp"
stefank@1879 32 #include "oops/arrayOop.hpp"
coleenp@3602 33 #include "oops/methodData.hpp"
coleenp@3602 34 #include "oops/method.hpp"
stefank@1879 35 #include "oops/oop.inline.hpp"
stefank@1879 36 #include "prims/jvmtiExport.hpp"
stefank@1879 37 #include "prims/jvmtiThreadState.hpp"
stefank@1879 38 #include "runtime/arguments.hpp"
stefank@1879 39 #include "runtime/deoptimization.hpp"
stefank@1879 40 #include "runtime/frame.inline.hpp"
stefank@1879 41 #include "runtime/sharedRuntime.hpp"
stefank@1879 42 #include "runtime/stubRoutines.hpp"
stefank@1879 43 #include "runtime/synchronizer.hpp"
stefank@1879 44 #include "runtime/timer.hpp"
stefank@1879 45 #include "runtime/vframeArray.hpp"
stefank@1879 46 #include "utilities/debug.hpp"
jprovino@4107 47 #include "utilities/macros.hpp"
duke@0 48
duke@0 49 #define __ _masm->
duke@0 50
duke@0 51
duke@0 52 #ifndef CC_INTERP
duke@0 53 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
duke@0 54 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
duke@0 55 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
duke@0 56
duke@0 57 //------------------------------------------------------------------------------------------------------------------------
duke@0 58
duke@0 59 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
duke@0 60 address entry = __ pc();
duke@0 61
duke@0 62 // Note: There should be a minimal interpreter frame set up when stack
duke@0 63 // overflow occurs since we check explicitly for it now.
duke@0 64 //
duke@0 65 #ifdef ASSERT
duke@0 66 { Label L;
never@304 67 __ lea(rax, Address(rbp,
duke@0 68 frame::interpreter_frame_monitor_block_top_offset * wordSize));
never@304 69 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp,
duke@0 70 // (stack grows negative)
duke@0 71 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
duke@0 72 __ stop ("interpreter frame not set up");
duke@0 73 __ bind(L);
duke@0 74 }
duke@0 75 #endif // ASSERT
duke@0 76 // Restore bcp under the assumption that the current frame is still
duke@0 77 // interpreted
duke@0 78 __ restore_bcp();
duke@0 79
duke@0 80 // expression stack must be empty before entering the VM if an exception
duke@0 81 // happened
duke@0 82 __ empty_expression_stack();
duke@0 83 __ empty_FPU_stack();
duke@0 84 // throw exception
duke@0 85 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
duke@0 86 return entry;
duke@0 87 }
duke@0 88
duke@0 89 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
duke@0 90 address entry = __ pc();
duke@0 91 // expression stack must be empty before entering the VM if an exception happened
duke@0 92 __ empty_expression_stack();
duke@0 93 __ empty_FPU_stack();
duke@0 94 // setup parameters
duke@0 95 // ??? convention: expect aberrant index in register rbx,
duke@0 96 __ lea(rax, ExternalAddress((address)name));
duke@0 97 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
duke@0 98 return entry;
duke@0 99 }
duke@0 100
duke@0 101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
duke@0 102 address entry = __ pc();
duke@0 103 // object is at TOS
never@304 104 __ pop(rax);
duke@0 105 // expression stack must be empty before entering the VM if an exception
duke@0 106 // happened
duke@0 107 __ empty_expression_stack();
duke@0 108 __ empty_FPU_stack();
duke@0 109 __ call_VM(noreg,
duke@0 110 CAST_FROM_FN_PTR(address,
duke@0 111 InterpreterRuntime::throw_ClassCastException),
duke@0 112 rax);
duke@0 113 return entry;
duke@0 114 }
duke@0 115
duke@0 116 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
duke@0 117 assert(!pass_oop || message == NULL, "either oop or message but not both");
duke@0 118 address entry = __ pc();
duke@0 119 if (pass_oop) {
duke@0 120 // object is at TOS
never@304 121 __ pop(rbx);
duke@0 122 }
duke@0 123 // expression stack must be empty before entering the VM if an exception happened
duke@0 124 __ empty_expression_stack();
duke@0 125 __ empty_FPU_stack();
duke@0 126 // setup parameters
duke@0 127 __ lea(rax, ExternalAddress((address)name));
duke@0 128 if (pass_oop) {
duke@0 129 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
duke@0 130 } else {
duke@0 131 if (message != NULL) {
duke@0 132 __ lea(rbx, ExternalAddress((address)message));
duke@0 133 } else {
xlu@512 134 __ movptr(rbx, NULL_WORD);
duke@0 135 }
duke@0 136 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
duke@0 137 }
duke@0 138 // throw exception
duke@0 139 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
duke@0 140 return entry;
duke@0 141 }
duke@0 142
duke@0 143
duke@0 144 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
duke@0 145 address entry = __ pc();
duke@0 146 // NULL last_sp until next java call
xlu@512 147 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
duke@0 148 __ dispatch_next(state);
duke@0 149 return entry;
duke@0 150 }
duke@0 151
duke@0 152
twisti@5604 153 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
twisti@2117 154 address entry = __ pc();
duke@0 155
duke@0 156 #ifdef COMPILER2
duke@0 157 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
twisti@5604 158 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
duke@0 159 for (int i = 1; i < 8; i++) {
duke@0 160 __ ffree(i);
duke@0 161 }
duke@0 162 } else if (UseSSE < 2) {
duke@0 163 __ empty_FPU_stack();
duke@0 164 }
duke@0 165 #endif
twisti@5604 166 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
duke@0 167 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
duke@0 168 } else {
duke@0 169 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
duke@0 170 }
duke@0 171
duke@0 172 // In SSE mode, interpreter returns FP results in xmm0 but they need
duke@0 173 // to end up back on the FPU so it can operate on them.
twisti@5604 174 if (state == ftos && UseSSE >= 1) {
never@304 175 __ subptr(rsp, wordSize);
duke@0 176 __ movflt(Address(rsp, 0), xmm0);
duke@0 177 __ fld_s(Address(rsp, 0));
never@304 178 __ addptr(rsp, wordSize);
twisti@5604 179 } else if (state == dtos && UseSSE >= 2) {
never@304 180 __ subptr(rsp, 2*wordSize);
duke@0 181 __ movdbl(Address(rsp, 0), xmm0);
duke@0 182 __ fld_d(Address(rsp, 0));
never@304 183 __ addptr(rsp, 2*wordSize);
duke@0 184 }
duke@0 185
duke@0 186 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
duke@0 187
duke@0 188 // Restore stack bottom in case i2c adjusted stack
never@304 189 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
duke@0 190 // and NULL it as marker that rsp is now tos until next java call
xlu@512 191 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
duke@0 192
duke@0 193 __ restore_bcp();
duke@0 194 __ restore_locals();
jrose@726 195
twisti@5604 196 if (state == atos) {
roland@5486 197 Register mdp = rbx;
roland@5486 198 Register tmp = rcx;
roland@5486 199 __ profile_return_type(mdp, rax, tmp);
roland@5486 200 }
roland@5486 201
twisti@5604 202 const Register cache = rbx;
twisti@5604 203 const Register index = rcx;
twisti@5604 204 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
twisti@5604 205
twisti@5604 206 const Register flags = cache;
twisti@5604 207 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
twisti@5604 208 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
twisti@5604 209 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
duke@0 210 __ dispatch_next(state, step);
jrose@726 211
duke@0 212 return entry;
duke@0 213 }
duke@0 214
duke@0 215
duke@0 216 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
duke@0 217 address entry = __ pc();
duke@0 218
duke@0 219 // In SSE mode, FP results are in xmm0
duke@0 220 if (state == ftos && UseSSE > 0) {
never@304 221 __ subptr(rsp, wordSize);
duke@0 222 __ movflt(Address(rsp, 0), xmm0);
duke@0 223 __ fld_s(Address(rsp, 0));
never@304 224 __ addptr(rsp, wordSize);
duke@0 225 } else if (state == dtos && UseSSE >= 2) {
never@304 226 __ subptr(rsp, 2*wordSize);
duke@0 227 __ movdbl(Address(rsp, 0), xmm0);
duke@0 228 __ fld_d(Address(rsp, 0));
never@304 229 __ addptr(rsp, 2*wordSize);
duke@0 230 }
duke@0 231
duke@0 232 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
duke@0 233
duke@0 234 // The stack is not extended by deopt but we must NULL last_sp as this
duke@0 235 // entry is like a "return".
xlu@512 236 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
duke@0 237 __ restore_bcp();
duke@0 238 __ restore_locals();
duke@0 239 // handle exceptions
duke@0 240 { Label L;
duke@0 241 const Register thread = rcx;
duke@0 242 __ get_thread(thread);
never@304 243 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
duke@0 244 __ jcc(Assembler::zero, L);
duke@0 245 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
duke@0 246 __ should_not_reach_here();
duke@0 247 __ bind(L);
duke@0 248 }
duke@0 249 __ dispatch_next(state, step);
duke@0 250 return entry;
duke@0 251 }
duke@0 252
duke@0 253
duke@0 254 int AbstractInterpreter::BasicType_as_index(BasicType type) {
duke@0 255 int i = 0;
duke@0 256 switch (type) {
duke@0 257 case T_BOOLEAN: i = 0; break;
duke@0 258 case T_CHAR : i = 1; break;
duke@0 259 case T_BYTE : i = 2; break;
duke@0 260 case T_SHORT : i = 3; break;
duke@0 261 case T_INT : // fall through
duke@0 262 case T_LONG : // fall through
duke@0 263 case T_VOID : i = 4; break;
duke@0 264 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
duke@0 265 case T_DOUBLE : i = 6; break;
duke@0 266 case T_OBJECT : // fall through
duke@0 267 case T_ARRAY : i = 7; break;
duke@0 268 default : ShouldNotReachHere();
duke@0 269 }
duke@0 270 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
duke@0 271 return i;
duke@0 272 }
duke@0 273
duke@0 274
duke@0 275 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
duke@0 276 address entry = __ pc();
duke@0 277 switch (type) {
duke@0 278 case T_BOOLEAN: __ c2bool(rax); break;
never@304 279 case T_CHAR : __ andptr(rax, 0xFFFF); break;
duke@0 280 case T_BYTE : __ sign_extend_byte (rax); break;
duke@0 281 case T_SHORT : __ sign_extend_short(rax); break;
duke@0 282 case T_INT : /* nothing to do */ break;
duke@0 283 case T_DOUBLE :
duke@0 284 case T_FLOAT :
duke@0 285 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
never@304 286 __ pop(t); // remove return address first
duke@0 287 // Must return a result for interpreter or compiler. In SSE
duke@0 288 // mode, results are returned in xmm0 and the FPU stack must
duke@0 289 // be empty.
duke@0 290 if (type == T_FLOAT && UseSSE >= 1) {
duke@0 291 // Load ST0
duke@0 292 __ fld_d(Address(rsp, 0));
duke@0 293 // Store as float and empty fpu stack
duke@0 294 __ fstp_s(Address(rsp, 0));
duke@0 295 // and reload
duke@0 296 __ movflt(xmm0, Address(rsp, 0));
duke@0 297 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
duke@0 298 __ movdbl(xmm0, Address(rsp, 0));
duke@0 299 } else {
duke@0 300 // restore ST0
duke@0 301 __ fld_d(Address(rsp, 0));
duke@0 302 }
duke@0 303 // and pop the temp
never@304 304 __ addptr(rsp, 2 * wordSize);
never@304 305 __ push(t); // restore return address
duke@0 306 }
duke@0 307 break;
duke@0 308 case T_OBJECT :
duke@0 309 // retrieve result from frame
never@304 310 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
duke@0 311 // and verify it
duke@0 312 __ verify_oop(rax);
duke@0 313 break;
duke@0 314 default : ShouldNotReachHere();
duke@0 315 }
duke@0 316 __ ret(0); // return from result handler
duke@0 317 return entry;
duke@0 318 }
duke@0 319
duke@0 320 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
duke@0 321 address entry = __ pc();
duke@0 322 __ push(state);
duke@0 323 __ call_VM(noreg, runtime_entry);
duke@0 324 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
duke@0 325 return entry;
duke@0 326 }
duke@0 327
duke@0 328
duke@0 329 // Helpers for commoning out cases in the various type of method entries.
duke@0 330 //
duke@0 331
duke@0 332 // increment invocation count & check for overflow
duke@0 333 //
duke@0 334 // Note: checking for negative value instead of overflow
duke@0 335 // so we have a 'sticky' overflow test
duke@0 336 //
duke@0 337 // rbx,: method
duke@0 338 // rcx: invocation counter
duke@0 339 //
duke@0 340 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
jiangli@4501 341 Label done;
jiangli@4501 342 // Note: In tiered we increment either counters in MethodCounters* or in MDO
jiangli@4501 343 // depending if we're profiling or not.
iveresov@1703 344 if (TieredCompilation) {
iveresov@1703 345 int increment = InvocationCounter::count_increment;
iveresov@1703 346 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
jiangli@4501 347 Label no_mdo;
iveresov@1703 348 if (ProfileInterpreter) {
iveresov@1703 349 // Are we profiling?
coleenp@3602 350 __ movptr(rax, Address(rbx, Method::method_data_offset()));
iveresov@1703 351 __ testptr(rax, rax);
iveresov@1703 352 __ jccb(Assembler::zero, no_mdo);
iveresov@1703 353 // Increment counter in the MDO
coleenp@3602 354 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
iveresov@1703 355 in_bytes(InvocationCounter::counter_offset()));
iveresov@1703 356 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
jiangli@4538 357 __ jmp(done);
iveresov@1703 358 }
iveresov@1703 359 __ bind(no_mdo);
jiangli@4501 360 // Increment counter in MethodCounters
jiangli@4501 361 const Address invocation_counter(rax,
jiangli@4501 362 MethodCounters::invocation_counter_offset() +
jiangli@4501 363 InvocationCounter::counter_offset());
jiangli@4501 364
jiangli@4501 365 __ get_method_counters(rbx, rax, done);
jiangli@4501 366 __ increment_mask_and_jump(invocation_counter, increment, mask,
jiangli@4501 367 rcx, false, Assembler::zero, overflow);
iveresov@1703 368 __ bind(done);
iveresov@1703 369 } else {
jiangli@4501 370 const Address backedge_counter (rax,
jiangli@4501 371 MethodCounters::backedge_counter_offset() +
jiangli@4501 372 InvocationCounter::counter_offset());
jiangli@4501 373 const Address invocation_counter(rax,
jiangli@4501 374 MethodCounters::invocation_counter_offset() +
jiangli@4501 375 InvocationCounter::counter_offset());
duke@0 376
jiangli@4501 377 __ get_method_counters(rbx, rax, done);
jiangli@4501 378
jiangli@4501 379 if (ProfileInterpreter) {
jiangli@4501 380 __ incrementl(Address(rax,
jiangli@4501 381 MethodCounters::interpreter_invocation_counter_offset()));
iveresov@1703 382 }
jiangli@4501 383
iveresov@1703 384 // Update standard invocation counters
jiangli@4501 385 __ movl(rcx, invocation_counter);
jiangli@4501 386 __ incrementl(rcx, InvocationCounter::count_increment);
jiangli@4501 387 __ movl(invocation_counter, rcx); // save invocation count
jiangli@4501 388
iveresov@1703 389 __ movl(rax, backedge_counter); // load backedge counter
iveresov@1703 390 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
iveresov@1703 391
iveresov@1703 392 __ addl(rcx, rax); // add both counters
iveresov@1703 393
iveresov@1703 394 // profile_method is non-null only for interpreted method so
iveresov@1703 395 // profile_method != NULL == !native_call
iveresov@1703 396 // BytecodeInterpreter only calls for native so code is elided.
iveresov@1703 397
iveresov@1703 398 if (ProfileInterpreter && profile_method != NULL) {
iveresov@1703 399 // Test to see if we should create a method data oop
iveresov@1703 400 __ cmp32(rcx,
iveresov@1703 401 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
iveresov@1703 402 __ jcc(Assembler::less, *profile_method_continue);
iveresov@1703 403
iveresov@1703 404 // if no method data exists, go to profile_method
iveresov@1703 405 __ test_method_data_pointer(rax, *profile_method);
iveresov@1703 406 }
iveresov@1703 407
iveresov@1703 408 __ cmp32(rcx,
iveresov@1703 409 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
iveresov@1703 410 __ jcc(Assembler::aboveEqual, *overflow);
jiangli@4501 411 __ bind(done);
duke@0 412 }
duke@0 413 }
duke@0 414
duke@0 415 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
duke@0 416
duke@0 417 // Asm interpreter on entry
duke@0 418 // rdi - locals
duke@0 419 // rsi - bcp
duke@0 420 // rbx, - method
duke@0 421 // rdx - cpool
duke@0 422 // rbp, - interpreter frame
duke@0 423
duke@0 424 // C++ interpreter on entry
duke@0 425 // rsi - new interpreter state pointer
duke@0 426 // rbp - interpreter frame pointer
duke@0 427 // rbx - method
duke@0 428
duke@0 429 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
duke@0 430 // rbx, - method
duke@0 431 // rcx - rcvr (assuming there is one)
duke@0 432 // top of stack return address of interpreter caller
duke@0 433 // rsp - sender_sp
duke@0 434
duke@0 435 // C++ interpreter only
duke@0 436 // rsi - previous interpreter state pointer
duke@0 437
duke@0 438 // InterpreterRuntime::frequency_counter_overflow takes one argument
duke@0 439 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
duke@0 440 // The call returns the address of the verified entry point for the method or NULL
duke@0 441 // if the compilation did not complete (either went background or bailed out).
xlu@533 442 __ movptr(rax, (intptr_t)false);
duke@0 443 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
duke@0 444
coleenp@3602 445 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
duke@0 446
duke@0 447 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
duke@0 448 // and jump to the interpreted entry.
duke@0 449 __ jmp(*do_continue, relocInfo::none);
duke@0 450
duke@0 451 }
duke@0 452
duke@0 453 void InterpreterGenerator::generate_stack_overflow_check(void) {
duke@0 454 // see if we've got enough room on the stack for locals plus overhead.
duke@0 455 // the expression stack grows down incrementally, so the normal guard
duke@0 456 // page mechanism will work for that.
duke@0 457 //
duke@0 458 // Registers live on entry:
duke@0 459 //
duke@0 460 // Asm interpreter
duke@0 461 // rdx: number of additional locals this frame needs (what we must check)
coleenp@3602 462 // rbx,: Method*
duke@0 463
duke@0 464 // destroyed on exit
duke@0 465 // rax,
duke@0 466
duke@0 467 // NOTE: since the additional locals are also always pushed (wasn't obvious in
duke@0 468 // generate_method_entry) so the guard should work for them too.
duke@0 469 //
duke@0 470
duke@0 471 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
duke@0 472 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
duke@0 473
duke@0 474 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
duke@0 475 // be sure to change this if you add/subtract anything to/from the overhead area
duke@0 476 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
duke@0 477
duke@0 478 const int page_size = os::vm_page_size();
duke@0 479
duke@0 480 Label after_frame_check;
duke@0 481
duke@0 482 // see if the frame is greater than one page in size. If so,
duke@0 483 // then we need to verify there is enough stack space remaining
duke@0 484 // for the additional locals.
twisti@1426 485 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
duke@0 486 __ jcc(Assembler::belowEqual, after_frame_check);
duke@0 487
duke@0 488 // compute rsp as if this were going to be the last frame on
duke@0 489 // the stack before the red zone
duke@0 490
duke@0 491 Label after_frame_check_pop;
duke@0 492
never@304 493 __ push(rsi);
duke@0 494
duke@0 495 const Register thread = rsi;
duke@0 496
duke@0 497 __ get_thread(thread);
duke@0 498
duke@0 499 const Address stack_base(thread, Thread::stack_base_offset());
duke@0 500 const Address stack_size(thread, Thread::stack_size_offset());
duke@0 501
duke@0 502 // locals + overhead, in bytes
never@304 503 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
duke@0 504
duke@0 505 #ifdef ASSERT
duke@0 506 Label stack_base_okay, stack_size_okay;
duke@0 507 // verify that thread stack base is non-zero
never@304 508 __ cmpptr(stack_base, (int32_t)NULL_WORD);
duke@0 509 __ jcc(Assembler::notEqual, stack_base_okay);
duke@0 510 __ stop("stack base is zero");
duke@0 511 __ bind(stack_base_okay);
duke@0 512 // verify that thread stack size is non-zero
never@304 513 __ cmpptr(stack_size, 0);
duke@0 514 __ jcc(Assembler::notEqual, stack_size_okay);
duke@0 515 __ stop("stack size is zero");
duke@0 516 __ bind(stack_size_okay);
duke@0 517 #endif
duke@0 518
duke@0 519 // Add stack base to locals and subtract stack size
never@304 520 __ addptr(rax, stack_base);
never@304 521 __ subptr(rax, stack_size);
duke@0 522
duke@0 523 // Use the maximum number of pages we might bang.
duke@0 524 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
duke@0 525 (StackRedPages+StackYellowPages);
never@304 526 __ addptr(rax, max_pages * page_size);
duke@0 527
duke@0 528 // check against the current stack bottom
never@304 529 __ cmpptr(rsp, rax);
duke@0 530 __ jcc(Assembler::above, after_frame_check_pop);
duke@0 531
never@304 532 __ pop(rsi); // get saved bcp / (c++ prev state ).
duke@0 533
bdelsart@2937 534 // Restore sender's sp as SP. This is necessary if the sender's
bdelsart@2937 535 // frame is an extended compiled frame (see gen_c2i_adapter())
bdelsart@2937 536 // and safer anyway in case of JSR292 adaptations.
duke@0 537
bdelsart@2937 538 __ pop(rax); // return address must be moved if SP is changed
bdelsart@2937 539 __ mov(rsp, rsi);
bdelsart@2937 540 __ push(rax);
bdelsart@2937 541
bdelsart@2937 542 // Note: the restored frame is not necessarily interpreted.
bdelsart@2937 543 // Use the shared runtime version of the StackOverflowError.
bdelsart@2937 544 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
bdelsart@2937 545 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
duke@0 546 // all done with frame size check
duke@0 547 __ bind(after_frame_check_pop);
never@304 548 __ pop(rsi);
duke@0 549
duke@0 550 __ bind(after_frame_check);
duke@0 551 }
duke@0 552
duke@0 553 // Allocate monitor and lock method (asm interpreter)
coleenp@3602 554 // rbx, - Method*
duke@0 555 //
duke@0 556 void InterpreterGenerator::lock_method(void) {
duke@0 557 // synchronize method
coleenp@3602 558 const Address access_flags (rbx, Method::access_flags_offset());
duke@0 559 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
duke@0 560 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
duke@0 561
duke@0 562 #ifdef ASSERT
duke@0 563 { Label L;
duke@0 564 __ movl(rax, access_flags);
duke@0 565 __ testl(rax, JVM_ACC_SYNCHRONIZED);
duke@0 566 __ jcc(Assembler::notZero, L);
duke@0 567 __ stop("method doesn't need synchronization");
duke@0 568 __ bind(L);
duke@0 569 }
duke@0 570 #endif // ASSERT
duke@0 571 // get synchronization object
duke@0 572 { Label done;
stefank@2956 573 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
duke@0 574 __ movl(rax, access_flags);
duke@0 575 __ testl(rax, JVM_ACC_STATIC);
never@304 576 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
duke@0 577 __ jcc(Assembler::zero, done);
coleenp@3602 578 __ movptr(rax, Address(rbx, Method::const_offset()));
coleenp@3602 579 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
coleenp@3602 580 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
never@304 581 __ movptr(rax, Address(rax, mirror_offset));
duke@0 582 __ bind(done);
duke@0 583 }
duke@0 584 // add space for monitor & lock
never@304 585 __ subptr(rsp, entry_size); // add space for a monitor entry
never@304 586 __ movptr(monitor_block_top, rsp); // set new monitor block top
never@304 587 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
never@304 588 __ mov(rdx, rsp); // object address
duke@0 589 __ lock_object(rdx);
duke@0 590 }
duke@0 591
duke@0 592 //
duke@0 593 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
duke@0 594 // and for native methods hence the shared code.
duke@0 595
duke@0 596 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
duke@0 597 // initialize fixed part of activation frame
never@304 598 __ push(rax); // save return address
duke@0 599 __ enter(); // save old & set new rbp,
duke@0 600
duke@0 601
never@304 602 __ push(rsi); // set sender sp
never@304 603 __ push((int32_t)NULL_WORD); // leave last_sp as null
coleenp@3602 604 __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod*
coleenp@3602 605 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
coleenp@3602 606 __ push(rbx); // save Method*
duke@0 607 if (ProfileInterpreter) {
duke@0 608 Label method_data_continue;
coleenp@3602 609 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
never@304 610 __ testptr(rdx, rdx);
duke@0 611 __ jcc(Assembler::zero, method_data_continue);
coleenp@3602 612 __ addptr(rdx, in_bytes(MethodData::data_offset()));
duke@0 613 __ bind(method_data_continue);
never@304 614 __ push(rdx); // set the mdp (method data pointer)
duke@0 615 } else {
never@304 616 __ push(0);
duke@0 617 }
duke@0 618
coleenp@3602 619 __ movptr(rdx, Address(rbx, Method::const_offset()));
coleenp@3602 620 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
coleenp@3602 621 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
never@304 622 __ push(rdx); // set constant pool cache
never@304 623 __ push(rdi); // set locals pointer
duke@0 624 if (native_call) {
never@304 625 __ push(0); // no bcp
duke@0 626 } else {
never@304 627 __ push(rsi); // set bcp
duke@0 628 }
never@304 629 __ push(0); // reserve word for pointer to expression stack bottom
never@304 630 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
duke@0 631 }
duke@0 632
duke@0 633 // End of helpers
duke@0 634
duke@0 635 //
duke@0 636 // Various method entries
duke@0 637 //------------------------------------------------------------------------------------------------------------------------
duke@0 638 //
duke@0 639 //
duke@0 640
duke@0 641 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
duke@0 642
duke@0 643 address InterpreterGenerator::generate_accessor_entry(void) {
duke@0 644
coleenp@3602 645 // rbx,: Method*
duke@0 646 // rcx: receiver (preserve for slow entry into asm interpreter)
duke@0 647
duke@0 648 // rsi: senderSP must preserved for slow path, set SP to it on fast path
duke@0 649
duke@0 650 address entry_point = __ pc();
duke@0 651 Label xreturn_path;
duke@0 652
duke@0 653 // do fastpath for resolved accessor methods
duke@0 654 if (UseFastAccessorMethods) {
duke@0 655 Label slow_path;
duke@0 656 // If we need a safepoint check, generate full interpreter entry.
duke@0 657 ExternalAddress state(SafepointSynchronize::address_of_state());
duke@0 658 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
duke@0 659 SafepointSynchronize::_not_synchronized);
duke@0 660
duke@0 661 __ jcc(Assembler::notEqual, slow_path);
duke@0 662 // ASM/C++ Interpreter
duke@0 663 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
duke@0 664 // Note: We can only use this code if the getfield has been resolved
duke@0 665 // and if we don't have a null-pointer exception => check for
duke@0 666 // these conditions first and use slow path if necessary.
duke@0 667 // rbx,: method
duke@0 668 // rcx: receiver
never@304 669 __ movptr(rax, Address(rsp, wordSize));
duke@0 670
duke@0 671 // check if local 0 != NULL and read field
never@304 672 __ testptr(rax, rax);
duke@0 673 __ jcc(Assembler::zero, slow_path);
duke@0 674
duke@0 675 // read first instruction word and extract bytecode @ 1 and index @ 2
coleenp@3602 676 __ movptr(rdx, Address(rbx, Method::const_offset()));
coleenp@3602 677 __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
coleenp@3602 678 __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
duke@0 679 // Shift codes right to get the index on the right.
duke@0 680 // The bytecode fetched looks like <index><0xb4><0x2a>
duke@0 681 __ shrl(rdx, 2*BitsPerByte);
duke@0 682 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
coleenp@3602 683 __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
duke@0 684
duke@0 685 // rax,: local 0
duke@0 686 // rbx,: method
duke@0 687 // rcx: receiver - do not destroy since it is needed for slow path!
duke@0 688 // rcx: scratch
duke@0 689 // rdx: constant pool cache index
duke@0 690 // rdi: constant pool cache
duke@0 691 // rsi: sender sp
duke@0 692
duke@0 693 // check if getfield has been resolved and read constant pool cache entry
duke@0 694 // check the validity of the cache entry by testing whether _indices field
duke@0 695 // contains Bytecode::_getfield in b1 byte.
duke@0 696 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
duke@0 697 __ movl(rcx,
duke@0 698 Address(rdi,
duke@0 699 rdx,
coleenp@3602 700 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
duke@0 701 __ shrl(rcx, 2*BitsPerByte);
duke@0 702 __ andl(rcx, 0xFF);
duke@0 703 __ cmpl(rcx, Bytecodes::_getfield);
duke@0 704 __ jcc(Assembler::notEqual, slow_path);
duke@0 705
duke@0 706 // Note: constant pool entry is not valid before bytecode is resolved
never@304 707 __ movptr(rcx,
never@304 708 Address(rdi,
never@304 709 rdx,
coleenp@3602 710 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
duke@0 711 __ movl(rdx,
duke@0 712 Address(rdi,
duke@0 713 rdx,
coleenp@3602 714 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
duke@0 715
duke@0 716 Label notByte, notShort, notChar;
duke@0 717 const Address field_address (rax, rcx, Address::times_1);
duke@0 718
duke@0 719 // Need to differentiate between igetfield, agetfield, bgetfield etc.
duke@0 720 // because they are different sizes.
duke@0 721 // Use the type from the constant pool cache
twisti@3534 722 __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
twisti@3534 723 // Make sure we don't need to mask rdx after the above shift
twisti@3534 724 ConstantPoolCacheEntry::verify_tos_state_shift();
duke@0 725 __ cmpl(rdx, btos);
duke@0 726 __ jcc(Assembler::notEqual, notByte);
duke@0 727 __ load_signed_byte(rax, field_address);
duke@0 728 __ jmp(xreturn_path);
duke@0 729
duke@0 730 __ bind(notByte);
duke@0 731 __ cmpl(rdx, stos);
duke@0 732 __ jcc(Assembler::notEqual, notShort);
jrose@622 733 __ load_signed_short(rax, field_address);
duke@0 734 __ jmp(xreturn_path);
duke@0 735
duke@0 736 __ bind(notShort);
duke@0 737 __ cmpl(rdx, ctos);
duke@0 738 __ jcc(Assembler::notEqual, notChar);
jrose@622 739 __ load_unsigned_short(rax, field_address);
duke@0 740 __ jmp(xreturn_path);
duke@0 741
duke@0 742 __ bind(notChar);
duke@0 743 #ifdef ASSERT
duke@0 744 Label okay;
duke@0 745 __ cmpl(rdx, atos);
duke@0 746 __ jcc(Assembler::equal, okay);
duke@0 747 __ cmpl(rdx, itos);
duke@0 748 __ jcc(Assembler::equal, okay);
duke@0 749 __ stop("what type is this?");
duke@0 750 __ bind(okay);
duke@0 751 #endif // ASSERT
duke@0 752 // All the rest are a 32 bit wordsize
never@304 753 // This is ok for now. Since fast accessors should be going away
never@304 754 __ movptr(rax, field_address);
duke@0 755
duke@0 756 __ bind(xreturn_path);
duke@0 757
duke@0 758 // _ireturn/_areturn
never@304 759 __ pop(rdi); // get return address
never@304 760 __ mov(rsp, rsi); // set sp to sender sp
duke@0 761 __ jmp(rdi);
duke@0 762
duke@0 763 // generate a vanilla interpreter entry as the slow path
duke@0 764 __ bind(slow_path);
duke@0 765
duke@0 766 (void) generate_normal_entry(false);
duke@0 767 return entry_point;
duke@0 768 }
duke@0 769 return NULL;
duke@0 770
duke@0 771 }
duke@0 772
johnc@2346 773 // Method entry for java.lang.ref.Reference.get.
johnc@2346 774 address InterpreterGenerator::generate_Reference_get_entry(void) {
jprovino@4107 775 #if INCLUDE_ALL_GCS
johnc@2346 776 // Code: _aload_0, _getfield, _areturn
johnc@2346 777 // parameter size = 1
johnc@2346 778 //
johnc@2346 779 // The code that gets generated by this routine is split into 2 parts:
johnc@2346 780 // 1. The "intrinsified" code for G1 (or any SATB based GC),
johnc@2346 781 // 2. The slow path - which is an expansion of the regular method entry.
johnc@2346 782 //
johnc@2346 783 // Notes:-
johnc@2346 784 // * In the G1 code we do not check whether we need to block for
johnc@2346 785 // a safepoint. If G1 is enabled then we must execute the specialized
johnc@2346 786 // code for Reference.get (except when the Reference object is null)
johnc@2346 787 // so that we can log the value in the referent field with an SATB
johnc@2346 788 // update buffer.
johnc@2346 789 // If the code for the getfield template is modified so that the
johnc@2346 790 // G1 pre-barrier code is executed when the current method is
johnc@2346 791 // Reference.get() then going through the normal method entry
johnc@2346 792 // will be fine.
johnc@2346 793 // * The G1 code below can, however, check the receiver object (the instance
johnc@2346 794 // of java.lang.Reference) and jump to the slow path if null. If the
johnc@2346 795 // Reference object is null then we obviously cannot fetch the referent
johnc@2346 796 // and so we don't need to call the G1 pre-barrier. Thus we can use the
johnc@2346 797 // regular method entry code to generate the NPE.
johnc@2346 798 //
johnc@2346 799 // This code is based on generate_accessor_enty.
johnc@2346 800
coleenp@3602 801 // rbx,: Method*
johnc@2346 802 // rcx: receiver (preserve for slow entry into asm interpreter)
johnc@2346 803
johnc@2346 804 // rsi: senderSP must preserved for slow path, set SP to it on fast path
johnc@2346 805
johnc@2346 806 address entry = __ pc();
johnc@2346 807
johnc@2346 808 const int referent_offset = java_lang_ref_Reference::referent_offset;
johnc@2346 809 guarantee(referent_offset > 0, "referent offset not initialized");
johnc@2346 810
johnc@2346 811 if (UseG1GC) {
johnc@2346 812 Label slow_path;
johnc@2346 813
johnc@2346 814 // Check if local 0 != NULL
johnc@2346 815 // If the receiver is null then it is OK to jump to the slow path.
johnc@2346 816 __ movptr(rax, Address(rsp, wordSize));
johnc@2346 817 __ testptr(rax, rax);
johnc@2346 818 __ jcc(Assembler::zero, slow_path);
johnc@2346 819
johnc@2346 820 // rax: local 0 (must be preserved across the G1 barrier call)
johnc@2346 821 //
johnc@2346 822 // rbx: method (at this point it's scratch)
johnc@2346 823 // rcx: receiver (at this point it's scratch)
johnc@2346 824 // rdx: scratch
johnc@2346 825 // rdi: scratch
johnc@2346 826 //
johnc@2346 827 // rsi: sender sp
johnc@2346 828
johnc@2346 829 // Preserve the sender sp in case the pre-barrier
johnc@2346 830 // calls the runtime
johnc@2346 831 __ push(rsi);
johnc@2346 832
johnc@2346 833 // Load the value of the referent field.
johnc@2346 834 const Address field_address(rax, referent_offset);
johnc@2346 835 __ movptr(rax, field_address);
johnc@2346 836
johnc@2346 837 // Generate the G1 pre-barrier code to log the value of
johnc@2346 838 // the referent field in an SATB buffer.
johnc@2346 839 __ get_thread(rcx);
johnc@2346 840 __ g1_write_barrier_pre(noreg /* obj */,
johnc@2346 841 rax /* pre_val */,
johnc@2346 842 rcx /* thread */,
johnc@2346 843 rbx /* tmp */,
johnc@2346 844 true /* tosca_save */,
johnc@2346 845 true /* expand_call */);
johnc@2346 846
johnc@2346 847 // _areturn
johnc@2346 848 __ pop(rsi); // get sender sp
johnc@2346 849 __ pop(rdi); // get return address
johnc@2346 850 __ mov(rsp, rsi); // set sp to sender sp
johnc@2346 851 __ jmp(rdi);
johnc@2346 852
johnc@2346 853 __ bind(slow_path);
johnc@2346 854 (void) generate_normal_entry(false);
johnc@2346 855
johnc@2346 856 return entry;
johnc@2346 857 }
jprovino@4107 858 #endif // INCLUDE_ALL_GCS
johnc@2346 859
johnc@2346 860 // If G1 is not enabled then attempt to go through the accessor entry point
johnc@2346 861 // Reference.get is an accessor
johnc@2346 862 return generate_accessor_entry();
johnc@2346 863 }
johnc@2346 864
drchase@4918 865 /**
drchase@4918 866 * Method entry for static native methods:
drchase@4918 867 * int java.util.zip.CRC32.update(int crc, int b)
drchase@4918 868 */
drchase@4918 869 address InterpreterGenerator::generate_CRC32_update_entry() {
drchase@4918 870 if (UseCRC32Intrinsics) {
drchase@4918 871 address entry = __ pc();
drchase@4918 872
drchase@4918 873 // rbx,: Method*
drchase@4918 874 // rsi: senderSP must preserved for slow path, set SP to it on fast path
drchase@4918 875 // rdx: scratch
drchase@4918 876 // rdi: scratch
drchase@4918 877
drchase@4918 878 Label slow_path;
drchase@4918 879 // If we need a safepoint check, generate full interpreter entry.
drchase@4918 880 ExternalAddress state(SafepointSynchronize::address_of_state());
drchase@4918 881 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
drchase@4918 882 SafepointSynchronize::_not_synchronized);
drchase@4918 883 __ jcc(Assembler::notEqual, slow_path);
drchase@4918 884
drchase@4918 885 // We don't generate local frame and don't align stack because
drchase@4918 886 // we call stub code and there is no safepoint on this path.
drchase@4918 887
drchase@4918 888 // Load parameters
drchase@4918 889 const Register crc = rax; // crc
drchase@4918 890 const Register val = rdx; // source java byte value
drchase@4918 891 const Register tbl = rdi; // scratch
drchase@4918 892
drchase@4918 893 // Arguments are reversed on java expression stack
drchase@4918 894 __ movl(val, Address(rsp, wordSize)); // byte value
drchase@4918 895 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
drchase@4918 896
drchase@4918 897 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
drchase@4918 898 __ notl(crc); // ~crc
drchase@4918 899 __ update_byte_crc32(crc, val, tbl);
drchase@4918 900 __ notl(crc); // ~crc
drchase@4918 901 // result in rax
drchase@4918 902
drchase@4918 903 // _areturn
drchase@4918 904 __ pop(rdi); // get return address
drchase@4918 905 __ mov(rsp, rsi); // set sp to sender sp
drchase@4918 906 __ jmp(rdi);
drchase@4918 907
drchase@4918 908 // generate a vanilla native entry as the slow path
drchase@4918 909 __ bind(slow_path);
drchase@4918 910
drchase@4918 911 (void) generate_native_entry(false);
drchase@4918 912
drchase@4918 913 return entry;
drchase@4918 914 }
drchase@4918 915 return generate_native_entry(false);
drchase@4918 916 }
drchase@4918 917
drchase@4918 918 /**
drchase@4918 919 * Method entry for static native methods:
drchase@4918 920 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
drchase@4918 921 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
drchase@4918 922 */
drchase@4918 923 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
drchase@4918 924 if (UseCRC32Intrinsics) {
drchase@4918 925 address entry = __ pc();
drchase@4918 926
drchase@4918 927 // rbx,: Method*
drchase@4918 928 // rsi: senderSP must preserved for slow path, set SP to it on fast path
drchase@4918 929 // rdx: scratch
drchase@4918 930 // rdi: scratch
drchase@4918 931
drchase@4918 932 Label slow_path;
drchase@4918 933 // If we need a safepoint check, generate full interpreter entry.
drchase@4918 934 ExternalAddress state(SafepointSynchronize::address_of_state());
drchase@4918 935 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
drchase@4918 936 SafepointSynchronize::_not_synchronized);
drchase@4918 937 __ jcc(Assembler::notEqual, slow_path);
drchase@4918 938
drchase@4918 939 // We don't generate local frame and don't align stack because
drchase@4918 940 // we call stub code and there is no safepoint on this path.
drchase@4918 941
drchase@4918 942 // Load parameters
drchase@4918 943 const Register crc = rax; // crc
drchase@4918 944 const Register buf = rdx; // source java byte array address
drchase@4918 945 const Register len = rdi; // length
drchase@4918 946
drchase@4918 947 // Arguments are reversed on java expression stack
drchase@4918 948 __ movl(len, Address(rsp, wordSize)); // Length
drchase@4918 949 // Calculate address of start element
drchase@4918 950 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
drchase@4918 951 __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
drchase@4918 952 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
drchase@4918 953 __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC
drchase@4918 954 } else {
drchase@4918 955 __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
drchase@4918 956 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
drchase@4918 957 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
drchase@4918 958 __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC
drchase@4918 959 }
drchase@4918 960
drchase@4918 961 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
drchase@4918 962 // result in rax
drchase@4918 963
drchase@4918 964 // _areturn
drchase@4918 965 __ pop(rdi); // get return address
drchase@4918 966 __ mov(rsp, rsi); // set sp to sender sp
drchase@4918 967 __ jmp(rdi);
drchase@4918 968
drchase@4918 969 // generate a vanilla native entry as the slow path
drchase@4918 970 __ bind(slow_path);
drchase@4918 971
drchase@4918 972 (void) generate_native_entry(false);
drchase@4918 973
drchase@4918 974 return entry;
drchase@4918 975 }
drchase@4918 976 return generate_native_entry(false);
drchase@4918 977 }
drchase@4918 978
duke@0 979 //
duke@0 980 // Interpreter stub for calling a native method. (asm interpreter)
duke@0 981 // This sets up a somewhat different looking stack for calling the native method
duke@0 982 // than the typical interpreter frame setup.
duke@0 983 //
duke@0 984
duke@0 985 address InterpreterGenerator::generate_native_entry(bool synchronized) {
duke@0 986 // determine code generation flags
duke@0 987 bool inc_counter = UseCompiler || CountCompiledCalls;
duke@0 988
coleenp@3602 989 // rbx,: Method*
duke@0 990 // rsi: sender sp
duke@0 991 // rsi: previous interpreter state (C++ interpreter) must preserve
duke@0 992 address entry_point = __ pc();
duke@0 993
jiangli@3903 994 const Address constMethod (rbx, Method::const_offset());
coleenp@3602 995 const Address access_flags (rbx, Method::access_flags_offset());
jiangli@3903 996 const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
duke@0 997
duke@0 998 // get parameter size (always needed)
jiangli@3903 999 __ movptr(rcx, constMethod);
jrose@622 1000 __ load_unsigned_short(rcx, size_of_parameters);
duke@0 1001
duke@0 1002 // native calls don't need the stack size check since they have no expression stack
duke@0 1003 // and the arguments are already on the stack and we only add a handful of words
duke@0 1004 // to the stack
duke@0 1005
coleenp@3602 1006 // rbx,: Method*
duke@0 1007 // rcx: size of parameters
duke@0 1008 // rsi: sender sp
duke@0 1009
never@304 1010 __ pop(rax); // get return address
duke@0 1011 // for natives the size of locals is zero
duke@0 1012
duke@0 1013 // compute beginning of parameters (rdi)
never@304 1014 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
duke@0 1015
duke@0 1016
duke@0 1017 // add 2 zero-initialized slots for native calls
duke@0 1018 // NULL result handler
never@304 1019 __ push((int32_t)NULL_WORD);
duke@0 1020 // NULL oop temp (mirror or jni oop result)
never@304 1021 __ push((int32_t)NULL_WORD);
duke@0 1022
duke@0 1023 // initialize fixed part of activation frame
duke@0 1024 generate_fixed_frame(true);
duke@0 1025
duke@0 1026 // make sure method is native & not abstract
duke@0 1027 #ifdef ASSERT
duke@0 1028 __ movl(rax, access_flags);
duke@0 1029 {
duke@0 1030 Label L;
duke@0 1031 __ testl(rax, JVM_ACC_NATIVE);
duke@0 1032 __ jcc(Assembler::notZero, L);
duke@0 1033 __ stop("tried to execute non-native method as native");
duke@0 1034 __ bind(L);
duke@0 1035 }
duke@0 1036 { Label L;
duke@0 1037 __ testl(rax, JVM_ACC_ABSTRACT);
duke@0 1038 __ jcc(Assembler::zero, L);
duke@0 1039 __ stop("tried to execute abstract method in interpreter");
duke@0 1040 __ bind(L);
duke@0 1041 }
duke@0 1042 #endif
duke@0 1043
duke@0 1044 // Since at this point in the method invocation the exception handler
duke@0 1045 // would try to exit the monitor of synchronized methods which hasn't
duke@0 1046 // been entered yet, we set the thread local variable
duke@0 1047 // _do_not_unlock_if_synchronized to true. The remove_activation will
duke@0 1048 // check this flag.
duke@0 1049
duke@0 1050 __ get_thread(rax);
duke@0 1051 const Address do_not_unlock_if_synchronized(rax,
duke@0 1052 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
duke@0 1053 __ movbool(do_not_unlock_if_synchronized, true);
duke@0 1054
duke@0 1055 // increment invocation count & check for overflow
duke@0 1056 Label invocation_counter_overflow;
duke@0 1057 if (inc_counter) {
duke@0 1058 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
duke@0 1059 }
duke@0 1060
duke@0 1061 Label continue_after_compile;
duke@0 1062 __ bind(continue_after_compile);
duke@0 1063
duke@0 1064 bang_stack_shadow_pages(true);
duke@0 1065
duke@0 1066 // reset the _do_not_unlock_if_synchronized flag
duke@0 1067 __ get_thread(rax);
duke@0 1068 __ movbool(do_not_unlock_if_synchronized, false);
duke@0 1069
duke@0 1070 // check for synchronized methods
duke@0 1071 // Must happen AFTER invocation_counter check and stack overflow check,
duke@0 1072 // so method is not locked if overflows.
duke@0 1073 //
duke@0 1074 if (synchronized) {
duke@0 1075 lock_method();
duke@0 1076 } else {
duke@0 1077 // no synchronization necessary
duke@0 1078 #ifdef ASSERT
duke@0 1079 { Label L;
duke@0 1080 __ movl(rax, access_flags);
duke@0 1081 __ testl(rax, JVM_ACC_SYNCHRONIZED);
duke@0 1082 __ jcc(Assembler::zero, L);
duke@0 1083 __ stop("method needs synchronization");
duke@0 1084 __ bind(L);
duke@0 1085 }
duke@0 1086 #endif
duke@0 1087 }
duke@0 1088
duke@0 1089 // start execution
duke@0 1090 #ifdef ASSERT
duke@0 1091 { Label L;
duke@0 1092 const Address monitor_block_top (rbp,
duke@0 1093 frame::interpreter_frame_monitor_block_top_offset * wordSize);
never@304 1094 __ movptr(rax, monitor_block_top);
never@304 1095 __ cmpptr(rax, rsp);
duke@0 1096 __ jcc(Assembler::equal, L);
duke@0 1097 __ stop("broken stack frame setup in interpreter");
duke@0 1098 __ bind(L);
duke@0 1099 }
duke@0 1100 #endif
duke@0 1101
duke@0 1102 // jvmti/dtrace support
duke@0 1103 __ notify_method_entry();
duke@0 1104
duke@0 1105 // work registers
duke@0 1106 const Register method = rbx;
duke@0 1107 const Register thread = rdi;
duke@0 1108 const Register t = rcx;
duke@0 1109
duke@0 1110 // allocate space for parameters
duke@0 1111 __ get_method(method);
jiangli@3903 1112 __ movptr(t, Address(method, Method::const_offset()));
jiangli@3903 1113 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
jiangli@3903 1114
twisti@1426 1115 __ shlptr(t, Interpreter::logStackElementSize);
never@304 1116 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
never@304 1117 __ subptr(rsp, t);
never@304 1118 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
duke@0 1119
duke@0 1120 // get signature handler
duke@0 1121 { Label L;
coleenp@3602 1122 __ movptr(t, Address(method, Method::signature_handler_offset()));
never@304 1123 __ testptr(t, t);
duke@0 1124 __ jcc(Assembler::notZero, L);
duke@0 1125 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
duke@0 1126 __ get_method(method);
coleenp@3602 1127 __ movptr(t, Address(method, Method::signature_handler_offset()));
duke@0 1128 __ bind(L);
duke@0 1129 }
duke@0 1130
duke@0 1131 // call signature handler
duke@0 1132 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
duke@0 1133 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
duke@0 1134 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
duke@0 1135 // The generated handlers do not touch RBX (the method oop).
duke@0 1136 // However, large signatures cannot be cached and are generated
duke@0 1137 // each time here. The slow-path generator will blow RBX
duke@0 1138 // sometime, so we must reload it after the call.
duke@0 1139 __ call(t);
duke@0 1140 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
duke@0 1141
duke@0 1142 // result handler is in rax,
duke@0 1143 // set result handler
never@304 1144 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
duke@0 1145
duke@0 1146 // pass mirror handle if static call
duke@0 1147 { Label L;
stefank@2956 1148 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
coleenp@3602 1149 __ movl(t, Address(method, Method::access_flags_offset()));
duke@0 1150 __ testl(t, JVM_ACC_STATIC);
duke@0 1151 __ jcc(Assembler::zero, L);
duke@0 1152 // get mirror
coleenp@3602 1153 __ movptr(t, Address(method, Method:: const_offset()));
coleenp@3602 1154 __ movptr(t, Address(t, ConstMethod::constants_offset()));
coleenp@3602 1155 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
never@304 1156 __ movptr(t, Address(t, mirror_offset));
duke@0 1157 // copy mirror into activation frame
never@304 1158 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
duke@0 1159 // pass handle to mirror
never@304 1160 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
never@304 1161 __ movptr(Address(rsp, wordSize), t);
duke@0 1162 __ bind(L);
duke@0 1163 }
duke@0 1164
duke@0 1165 // get native function entry point
duke@0 1166 { Label L;
coleenp@3602 1167 __ movptr(rax, Address(method, Method::native_function_offset()));
duke@0 1168 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
never@304 1169 __ cmpptr(rax, unsatisfied.addr());
duke@0 1170 __ jcc(Assembler::notEqual, L);
duke@0 1171 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
duke@0 1172 __ get_method(method);
coleenp@3602 1173 __ movptr(rax, Address(method, Method::native_function_offset()));
duke@0 1174 __ bind(L);
duke@0 1175 }
duke@0 1176
duke@0 1177 // pass JNIEnv
duke@0 1178 __ get_thread(thread);
never@304 1179 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
never@304 1180 __ movptr(Address(rsp, 0), t);
duke@0 1181
duke@0 1182 // set_last_Java_frame_before_call
duke@0 1183 // It is enough that the pc()
duke@0 1184 // points into the right code segment. It does not have to be the correct return pc.
duke@0 1185 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
duke@0 1186
duke@0 1187 // change thread state
duke@0 1188 #ifdef ASSERT
duke@0 1189 { Label L;
duke@0 1190 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
duke@0 1191 __ cmpl(t, _thread_in_Java);
duke@0 1192 __ jcc(Assembler::equal, L);
duke@0 1193 __ stop("Wrong thread state in native stub");
duke@0 1194 __ bind(L);
duke@0 1195 }
duke@0 1196 #endif
duke@0 1197
duke@0 1198 // Change state to native
duke@0 1199 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
duke@0 1200 __ call(rax);
duke@0 1201
duke@0 1202 // result potentially in rdx:rax or ST0
duke@0 1203
kvn@4438 1204 // Verify or restore cpu control state after JNI call
kvn@4438 1205 __ restore_cpu_control_state_after_jni();
duke@0 1206
duke@0 1207 // save potential result in ST(0) & rdx:rax
duke@0 1208 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
duke@0 1209 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
duke@0 1210 // It is safe to do this push because state is _thread_in_native and return address will be found
duke@0 1211 // via _last_native_pc and not via _last_jave_sp
duke@0 1212
duke@0 1213 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
duke@0 1214 // If the order changes or anything else is added to the stack the code in
duke@0 1215 // interpreter_frame_result will have to be changed.
duke@0 1216
duke@0 1217 { Label L;
duke@0 1218 Label push_double;
duke@0 1219 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
duke@0 1220 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
duke@0 1221 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
duke@0 1222 float_handler.addr());
duke@0 1223 __ jcc(Assembler::equal, push_double);
duke@0 1224 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
duke@0 1225 double_handler.addr());
duke@0 1226 __ jcc(Assembler::notEqual, L);
duke@0 1227 __ bind(push_double);
duke@0 1228 __ push(dtos);
duke@0 1229 __ bind(L);
duke@0 1230 }
duke@0 1231 __ push(ltos);
duke@0 1232
duke@0 1233 // change thread state
duke@0 1234 __ get_thread(thread);
duke@0 1235 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
duke@0 1236 if(os::is_MP()) {
duke@0 1237 if (UseMembar) {
never@304 1238 // Force this write out before the read below
never@304 1239 __ membar(Assembler::Membar_mask_bits(
never@304 1240 Assembler::LoadLoad | Assembler::LoadStore |
never@304 1241 Assembler::StoreLoad | Assembler::StoreStore));
duke@0 1242 } else {
duke@0 1243 // Write serialization page so VM thread can do a pseudo remote membar.
duke@0 1244 // We use the current thread pointer to calculate a thread specific
duke@0 1245 // offset to write to within the page. This minimizes bus traffic
duke@0 1246 // due to cache line collision.
duke@0 1247 __ serialize_memory(thread, rcx);
duke@0 1248 }
duke@0 1249 }
duke@0 1250
duke@0 1251 if (AlwaysRestoreFPU) {
duke@0 1252 // Make sure the control word is correct.
duke@0 1253 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
duke@0 1254 }
duke@0 1255
duke@0 1256 // check for safepoint operation in progress and/or pending suspend requests
duke@0 1257 { Label Continue;
duke@0 1258
duke@0 1259 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
duke@0 1260 SafepointSynchronize::_not_synchronized);
duke@0 1261
duke@0 1262 Label L;
duke@0 1263 __ jcc(Assembler::notEqual, L);
duke@0 1264 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
duke@0 1265 __ jcc(Assembler::equal, Continue);
duke@0 1266 __ bind(L);
duke@0 1267
duke@0 1268 // Don't use call_VM as it will see a possible pending exception and forward it
duke@0 1269 // and never return here preventing us from clearing _last_native_pc down below.
duke@0 1270 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
duke@0 1271 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
duke@0 1272 // by hand.
duke@0 1273 //
never@304 1274 __ push(thread);
duke@0 1275 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
duke@0 1276 JavaThread::check_special_condition_for_native_trans)));
duke@0 1277 __ increment(rsp, wordSize);
duke@0 1278 __ get_thread(thread);
duke@0 1279
duke@0 1280 __ bind(Continue);
duke@0 1281 }
duke@0 1282
duke@0 1283 // change thread state
duke@0 1284 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
duke@0 1285
duke@0 1286 __ reset_last_Java_frame(thread, true, true);
duke@0 1287
duke@0 1288 // reset handle block
never@304 1289 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
xlu@512 1290 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
duke@0 1291
duke@0 1292 // If result was an oop then unbox and save it in the frame
duke@0 1293 { Label L;
duke@0 1294 Label no_oop, store_result;
duke@0 1295 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
duke@0 1296 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
duke@0 1297 handler.addr());
duke@0 1298 __ jcc(Assembler::notEqual, no_oop);
never@304 1299 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
duke@0 1300 __ pop(ltos);
never@304 1301 __ testptr(rax, rax);
duke@0 1302 __ jcc(Assembler::zero, store_result);
duke@0 1303 // unbox
never@304 1304 __ movptr(rax, Address(rax, 0));
duke@0 1305 __ bind(store_result);
never@304 1306 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
duke@0 1307 // keep stack depth as expected by pushing oop which will eventually be discarded
duke@0 1308 __ push(ltos);
duke@0 1309 __ bind(no_oop);
duke@0 1310 }
duke@0 1311
duke@0 1312 {
duke@0 1313 Label no_reguard;
duke@0 1314 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
duke@0 1315 __ jcc(Assembler::notEqual, no_reguard);
duke@0 1316
never@304 1317 __ pusha();
duke@0 1318 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
never@304 1319 __ popa();
duke@0 1320
duke@0 1321 __ bind(no_reguard);
duke@0 1322 }
duke@0 1323
duke@0 1324 // restore rsi to have legal interpreter frame,
duke@0 1325 // i.e., bci == 0 <=> rsi == code_base()
duke@0 1326 // Can't call_VM until bcp is within reasonable.
duke@0 1327 __ get_method(method); // method is junk from thread_in_native to now.
coleenp@3602 1328 __ movptr(rsi, Address(method,Method::const_offset())); // get ConstMethod*
coleenp@3602 1329 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
duke@0 1330
duke@0 1331 // handle exceptions (exception handling will handle unlocking!)
duke@0 1332 { Label L;
never@304 1333 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
duke@0 1334 __ jcc(Assembler::zero, L);
duke@0 1335 // Note: At some point we may want to unify this with the code used in call_VM_base();
duke@0 1336 // i.e., we should use the StubRoutines::forward_exception code. For now this
duke@0 1337 // doesn't work here because the rsp is not correctly set at this point.
duke@0 1338 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
duke@0 1339 __ should_not_reach_here();
duke@0 1340 __ bind(L);
duke@0 1341 }
duke@0 1342
duke@0 1343 // do unlocking if necessary
duke@0 1344 { Label L;
coleenp@3602 1345 __ movl(t, Address(method, Method::access_flags_offset()));
duke@0 1346 __ testl(t, JVM_ACC_SYNCHRONIZED);
duke@0 1347 __ jcc(Assembler::zero, L);
duke@0 1348 // the code below should be shared with interpreter macro assembler implementation
duke@0 1349 { Label unlock;
duke@0 1350 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
duke@0 1351 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
duke@0 1352 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
duke@0 1353
never@304 1354 __ lea(rdx, monitor); // address of first monitor
duke@0 1355
never@304 1356 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
never@304 1357 __ testptr(t, t);
duke@0 1358 __ jcc(Assembler::notZero, unlock);
duke@0 1359
duke@0 1360 // Entry already unlocked, need to throw exception
duke@0 1361 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
duke@0 1362 __ should_not_reach_here();
duke@0 1363
duke@0 1364 __ bind(unlock);
duke@0 1365 __ unlock_object(rdx);
duke@0 1366 }
duke@0 1367 __ bind(L);
duke@0 1368 }
duke@0 1369
duke@0 1370 // jvmti/dtrace support
duke@0 1371 // Note: This must happen _after_ handling/throwing any exceptions since
duke@0 1372 // the exception handler code notifies the runtime of method exits
duke@0 1373 // too. If this happens before, method entry/exit notifications are
duke@0 1374 // not properly paired (was bug - gri 11/22/99).
duke@0 1375 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
duke@0 1376
duke@0 1377 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
duke@0 1378 __ pop(ltos);
never@304 1379 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
duke@0 1380 __ call(t);
duke@0 1381
duke@0 1382 // remove activation
never@304 1383 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
duke@0 1384 __ leave(); // remove frame anchor
never@304 1385 __ pop(rdi); // get return address
never@304 1386 __ mov(rsp, t); // set sp to sender sp
duke@0 1387 __ jmp(rdi);
duke@0 1388
duke@0 1389 if (inc_counter) {
duke@0 1390 // Handle overflow of counter and compile method
duke@0 1391 __ bind(invocation_counter_overflow);
duke@0 1392 generate_counter_overflow(&continue_after_compile);
duke@0 1393 }
duke@0 1394
duke@0 1395 return entry_point;
duke@0 1396 }
duke@0 1397
duke@0 1398 //
duke@0 1399 // Generic interpreted method entry to (asm) interpreter
duke@0 1400 //
duke@0 1401 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
duke@0 1402 // determine code generation flags
duke@0 1403 bool inc_counter = UseCompiler || CountCompiledCalls;
duke@0 1404
coleenp@3602 1405 // rbx,: Method*
duke@0 1406 // rsi: sender sp
duke@0 1407 address entry_point = __ pc();
duke@0 1408
jiangli@3903 1409 const Address constMethod (rbx, Method::const_offset());
coleenp@3602 1410 const Address access_flags (rbx, Method::access_flags_offset());
jiangli@3903 1411 const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
jiangli@3903 1412 const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset());
duke@0 1413
duke@0 1414 // get parameter size (always needed)
jiangli@3903 1415 __ movptr(rdx, constMethod);
jrose@622 1416 __ load_unsigned_short(rcx, size_of_parameters);
duke@0 1417
coleenp@3602 1418 // rbx,: Method*
duke@0 1419 // rcx: size of parameters
duke@0 1420
duke@0 1421 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
duke@0 1422
jrose@622 1423 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
duke@0 1424 __ subl(rdx, rcx); // rdx = no. of additional locals
duke@0 1425
duke@0 1426 // see if we've got enough room on the stack for locals plus overhead.
duke@0 1427 generate_stack_overflow_check();
duke@0 1428
duke@0 1429 // get return address
never@304 1430 __ pop(rax);
duke@0 1431
duke@0 1432 // compute beginning of parameters (rdi)
never@304 1433 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
duke@0 1434
duke@0 1435 // rdx - # of additional locals
duke@0 1436 // allocate space for locals
duke@0 1437 // explicitly initialize locals
duke@0 1438 {
duke@0 1439 Label exit, loop;
duke@0 1440 __ testl(rdx, rdx);
duke@0 1441 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
duke@0 1442 __ bind(loop);
never@304 1443 __ push((int32_t)NULL_WORD); // initialize local variables
duke@0 1444 __ decrement(rdx); // until everything initialized
duke@0 1445 __ jcc(Assembler::greater, loop);
duke@0 1446 __ bind(exit);
duke@0 1447 }
duke@0 1448
duke@0 1449 // initialize fixed part of activation frame
duke@0 1450 generate_fixed_frame(false);
duke@0 1451
duke@0 1452 // make sure method is not native & not abstract
duke@0 1453 #ifdef ASSERT
duke@0 1454 __ movl(rax, access_flags);
duke@0 1455 {
duke@0 1456 Label L;
duke@0 1457 __ testl(rax, JVM_ACC_NATIVE);
duke@0 1458 __ jcc(Assembler::zero, L);
duke@0 1459 __ stop("tried to execute native method as non-native");
duke@0 1460 __ bind(L);
duke@0 1461 }
duke@0 1462 { Label L;
duke@0 1463 __ testl(rax, JVM_ACC_ABSTRACT);
duke@0 1464 __ jcc(Assembler::zero, L);
duke@0 1465 __ stop("tried to execute abstract method in interpreter");
duke@0 1466 __ bind(L);
duke@0 1467 }
duke@0 1468 #endif
duke@0 1469
duke@0 1470 // Since at this point in the method invocation the exception handler
duke@0 1471 // would try to exit the monitor of synchronized methods which hasn't
duke@0 1472 // been entered yet, we set the thread local variable
duke@0 1473 // _do_not_unlock_if_synchronized to true. The remove_activation will
duke@0 1474 // check this flag.
duke@0 1475
duke@0 1476 __ get_thread(rax);
duke@0 1477 const Address do_not_unlock_if_synchronized(rax,
duke@0 1478 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
duke@0 1479 __ movbool(do_not_unlock_if_synchronized, true);
duke@0 1480
roland@5552 1481 __ profile_parameters_type(rax, rcx, rdx);
duke@0 1482 // increment invocation count & check for overflow
duke@0 1483 Label invocation_counter_overflow;
duke@0 1484 Label profile_method;
duke@0 1485 Label profile_method_continue;
duke@0 1486 if (inc_counter) {
duke@0 1487 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
duke@0 1488 if (ProfileInterpreter) {
duke@0 1489 __ bind(profile_method_continue);
duke@0 1490 }
duke@0 1491 }
duke@0 1492 Label continue_after_compile;
duke@0 1493 __ bind(continue_after_compile);
duke@0 1494
duke@0 1495 bang_stack_shadow_pages(false);
duke@0 1496
duke@0 1497 // reset the _do_not_unlock_if_synchronized flag
duke@0 1498 __ get_thread(rax);
duke@0 1499 __ movbool(do_not_unlock_if_synchronized, false);
duke@0 1500
duke@0 1501 // check for synchronized methods
duke@0 1502 // Must happen AFTER invocation_counter check and stack overflow check,
duke@0 1503 // so method is not locked if overflows.
duke@0 1504 //
duke@0 1505 if (synchronized) {
duke@0 1506 // Allocate monitor and lock method
duke@0 1507 lock_method();
duke@0 1508 } else {
duke@0 1509 // no synchronization necessary
duke@0 1510 #ifdef ASSERT
duke@0 1511 { Label L;
duke@0 1512 __ movl(rax, access_flags);
duke@0 1513 __ testl(rax, JVM_ACC_SYNCHRONIZED);
duke@0 1514 __ jcc(Assembler::zero, L);
duke@0 1515 __ stop("method needs synchronization");
duke@0 1516 __ bind(L);
duke@0 1517 }
duke@0 1518 #endif
duke@0 1519 }
duke@0 1520
duke@0 1521 // start execution
duke@0 1522 #ifdef ASSERT
duke@0 1523 { Label L;
duke@0 1524 const Address monitor_block_top (rbp,
duke@0 1525 frame::interpreter_frame_monitor_block_top_offset * wordSize);
never@304 1526 __ movptr(rax, monitor_block_top);
never@304 1527 __ cmpptr(rax, rsp);
duke@0 1528 __ jcc(Assembler::equal, L);
duke@0 1529 __ stop("broken stack frame setup in interpreter");
duke@0 1530 __ bind(L);
duke@0 1531 }
duke@0 1532 #endif
duke@0 1533
duke@0 1534 // jvmti support
duke@0 1535 __ notify_method_entry();
duke@0 1536
duke@0 1537 __ dispatch_next(vtos);
duke@0 1538
duke@0 1539 // invocation counter overflow
duke@0 1540 if (inc_counter) {
duke@0 1541 if (ProfileInterpreter) {
duke@0 1542 // We have decided to profile this method in the interpreter
duke@0 1543 __ bind(profile_method);
iveresov@2003 1544 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
iveresov@2003 1545 __ set_method_data_pointer_for_bcp();
iveresov@2026 1546 __ get_method(rbx);
duke@0 1547 __ jmp(profile_method_continue);
duke@0 1548 }
duke@0 1549 // Handle overflow of counter and compile method
duke@0 1550 __ bind(invocation_counter_overflow);
duke@0 1551 generate_counter_overflow(&continue_after_compile);
duke@0 1552 }
duke@0 1553
duke@0 1554 return entry_point;
duke@0 1555 }
duke@0 1556
duke@0 1557 //------------------------------------------------------------------------------------------------------------------------
duke@0 1558 // Entry points
duke@0 1559 //
duke@0 1560 // Here we generate the various kind of entries into the interpreter.
duke@0 1561 // The two main entry type are generic bytecode methods and native call method.
duke@0 1562 // These both come in synchronized and non-synchronized versions but the
duke@0 1563 // frame layout they create is very similar. The other method entry
duke@0 1564 // types are really just special purpose entries that are really entry
duke@0 1565 // and interpretation all in one. These are for trivial methods like
duke@0 1566 // accessor, empty, or special math methods.
duke@0 1567 //
duke@0 1568 // When control flow reaches any of the entry types for the interpreter
duke@0 1569 // the following holds ->
duke@0 1570 //
duke@0 1571 // Arguments:
duke@0 1572 //
coleenp@3602 1573 // rbx,: Method*
duke@0 1574 // rcx: receiver
duke@0 1575 //
duke@0 1576 //
duke@0 1577 // Stack layout immediately at entry
duke@0 1578 //
duke@0 1579 // [ return address ] <--- rsp
duke@0 1580 // [ parameter n ]
duke@0 1581 // ...
duke@0 1582 // [ parameter 1 ]
duke@0 1583 // [ expression stack ] (caller's java expression stack)
duke@0 1584
duke@0 1585 // Assuming that we don't go to one of the trivial specialized
duke@0 1586 // entries the stack will look like below when we are ready to execute
duke@0 1587 // the first bytecode (or call the native routine). The register usage
duke@0 1588 // will be as the template based interpreter expects (see interpreter_x86.hpp).
duke@0 1589 //
duke@0 1590 // local variables follow incoming parameters immediately; i.e.
duke@0 1591 // the return address is moved to the end of the locals).
duke@0 1592 //
duke@0 1593 // [ monitor entry ] <--- rsp
duke@0 1594 // ...
duke@0 1595 // [ monitor entry ]
duke@0 1596 // [ expr. stack bottom ]
duke@0 1597 // [ saved rsi ]
duke@0 1598 // [ current rdi ]
coleenp@3602 1599 // [ Method* ]
duke@0 1600 // [ saved rbp, ] <--- rbp,
duke@0 1601 // [ return address ]
duke@0 1602 // [ local variable m ]
duke@0 1603 // ...
duke@0 1604 // [ local variable 1 ]
duke@0 1605 // [ parameter n ]
duke@0 1606 // ...
duke@0 1607 // [ parameter 1 ] <--- rdi
duke@0 1608
duke@0 1609 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
duke@0 1610 // determine code generation flags
duke@0 1611 bool synchronized = false;
duke@0 1612 address entry_point = NULL;
drchase@4918 1613 InterpreterGenerator* ig_this = (InterpreterGenerator*)this;
duke@0 1614
duke@0 1615 switch (kind) {
drchase@4918 1616 case Interpreter::zerolocals : break;
drchase@4918 1617 case Interpreter::zerolocals_synchronized: synchronized = true; break;
drchase@4918 1618 case Interpreter::native : entry_point = ig_this->generate_native_entry(false); break;
drchase@4918 1619 case Interpreter::native_synchronized : entry_point = ig_this->generate_native_entry(true); break;
drchase@4918 1620 case Interpreter::empty : entry_point = ig_this->generate_empty_entry(); break;
drchase@4918 1621 case Interpreter::accessor : entry_point = ig_this->generate_accessor_entry(); break;
drchase@4918 1622 case Interpreter::abstract : entry_point = ig_this->generate_abstract_entry(); break;
duke@0 1623
duke@0 1624 case Interpreter::java_lang_math_sin : // fall thru
duke@0 1625 case Interpreter::java_lang_math_cos : // fall thru
duke@0 1626 case Interpreter::java_lang_math_tan : // fall thru
duke@0 1627 case Interpreter::java_lang_math_abs : // fall thru
duke@0 1628 case Interpreter::java_lang_math_log : // fall thru
duke@0 1629 case Interpreter::java_lang_math_log10 : // fall thru
roland@3352 1630 case Interpreter::java_lang_math_sqrt : // fall thru
roland@3352 1631 case Interpreter::java_lang_math_pow : // fall thru
drchase@4918 1632 case Interpreter::java_lang_math_exp : entry_point = ig_this->generate_math_entry(kind); break;
johnc@2346 1633 case Interpreter::java_lang_ref_reference_get
drchase@4918 1634 : entry_point = ig_this->generate_Reference_get_entry(); break;
drchase@4918 1635 case Interpreter::java_util_zip_CRC32_update
drchase@4918 1636 : entry_point = ig_this->generate_CRC32_update_entry(); break;
drchase@4918 1637 case Interpreter::java_util_zip_CRC32_updateBytes
drchase@4918 1638 : // fall thru
drchase@4918 1639 case Interpreter::java_util_zip_CRC32_updateByteBuffer
drchase@4918 1640 : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break;
twisti@3534 1641 default:
twisti@3534 1642 fatal(err_msg("unexpected method kind: %d", kind));
twisti@3534 1643 break;
duke@0 1644 }
duke@0 1645
duke@0 1646 if (entry_point) return entry_point;
duke@0 1647
drchase@4918 1648 return ig_this->generate_normal_entry(synchronized);
duke@0 1649
duke@0 1650 }
duke@0 1651
never@1174 1652 // These should never be compiled since the interpreter will prefer
never@1174 1653 // the compiled version to the intrinsic version.
never@1174 1654 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
never@1174 1655 switch (method_kind(m)) {
never@1174 1656 case Interpreter::java_lang_math_sin : // fall thru
never@1174 1657 case Interpreter::java_lang_math_cos : // fall thru
never@1174 1658 case Interpreter::java_lang_math_tan : // fall thru
never@1174 1659 case Interpreter::java_lang_math_abs : // fall thru
never@1174 1660 case Interpreter::java_lang_math_log : // fall thru
never@1174 1661 case Interpreter::java_lang_math_log10 : // fall thru
roland@3352 1662 case Interpreter::java_lang_math_sqrt : // fall thru
roland@3352 1663 case Interpreter::java_lang_math_pow : // fall thru
roland@3352 1664 case Interpreter::java_lang_math_exp :
never@1174 1665 return false;
never@1174 1666 default:
never@1174 1667 return true;
never@1174 1668 }
never@1174 1669 }
never@1174 1670
duke@0 1671 // How much stack a method activation needs in words.
coleenp@3602 1672 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
duke@0 1673
duke@0 1674 const int stub_code = 4; // see generate_call_stub
duke@0 1675 // Save space for one monitor to get into the interpreted method in case
duke@0 1676 // the method is synchronized
duke@0 1677 int monitor_size = method->is_synchronized() ?
duke@0 1678 1*frame::interpreter_frame_monitor_size() : 0;
duke@0 1679
duke@0 1680 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
duke@0 1681 // be sure to change this if you add/subtract anything to/from the overhead area
duke@0 1682 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
duke@0 1683
roland@4790 1684 const int method_stack = (method->max_locals() + method->max_stack()) *
twisti@1426 1685 Interpreter::stackElementWords;
duke@0 1686 return overhead_size + method_stack + stub_code;
duke@0 1687 }
duke@0 1688
duke@0 1689 // asm based interpreter deoptimization helpers
duke@0 1690
coleenp@3602 1691 int AbstractInterpreter::layout_activation(Method* method,
duke@0 1692 int tempcount,
duke@0 1693 int popframe_extra_args,
duke@0 1694 int moncount,
never@2466 1695 int caller_actual_parameters,
duke@0 1696 int callee_param_count,
duke@0 1697 int callee_locals,
duke@0 1698 frame* caller,
duke@0 1699 frame* interpreter_frame,
roland@4292 1700 bool is_top_frame,
roland@4292 1701 bool is_bottom_frame) {
duke@0 1702 // Note: This calculation must exactly parallel the frame setup
duke@0 1703 // in AbstractInterpreterGenerator::generate_method_entry.
duke@0 1704 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
duke@0 1705 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
duke@0 1706 // as determined by a previous call to this method.
duke@0 1707 // It is also guaranteed to be walkable even though it is in a skeletal state
duke@0 1708 // NOTE: return size is in words not bytes
duke@0 1709
duke@0 1710 // fixed size of an interpreter frame:
twisti@1426 1711 int max_locals = method->max_locals() * Interpreter::stackElementWords;
duke@0 1712 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
twisti@1426 1713 Interpreter::stackElementWords;
duke@0 1714
duke@0 1715 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
duke@0 1716
duke@0 1717 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
duke@0 1718 // Since the callee parameters already account for the callee's params we only need to account for
duke@0 1719 // the extra locals.
duke@0 1720
duke@0 1721
duke@0 1722 int size = overhead +
twisti@1426 1723 ((callee_locals - callee_param_count)*Interpreter::stackElementWords) +
duke@0 1724 (moncount*frame::interpreter_frame_monitor_size()) +
twisti@1426 1725 tempcount*Interpreter::stackElementWords + popframe_extra_args;
duke@0 1726
duke@0 1727 if (interpreter_frame != NULL) {
duke@0 1728 #ifdef ASSERT
twisti@2263 1729 if (!EnableInvokeDynamic)
twisti@1135 1730 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
twisti@1135 1731 // Probably, since deoptimization doesn't work yet.
twisti@1135 1732 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
duke@0 1733 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
duke@0 1734 #endif
duke@0 1735
duke@0 1736 interpreter_frame->interpreter_frame_set_method(method);
duke@0 1737 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
duke@0 1738 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
duke@0 1739 // and sender_sp is fp+8
duke@0 1740 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
duke@0 1741
twisti@2803 1742 #ifdef ASSERT
twisti@2803 1743 if (caller->is_interpreted_frame()) {
twisti@2803 1744 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
twisti@2803 1745 }
twisti@2803 1746 #endif
twisti@2803 1747
duke@0 1748 interpreter_frame->interpreter_frame_set_locals(locals);
duke@0 1749 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
duke@0 1750 BasicObjectLock* monbot = montop - moncount;
duke@0 1751 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
duke@0 1752
duke@0 1753 // Set last_sp
duke@0 1754 intptr_t* rsp = (intptr_t*) monbot -
twisti@1426 1755 tempcount*Interpreter::stackElementWords -
duke@0 1756 popframe_extra_args;
duke@0 1757 interpreter_frame->interpreter_frame_set_last_sp(rsp);
duke@0 1758
duke@0 1759 // All frames but the initial (oldest) interpreter frame we fill in have a
duke@0 1760 // value for sender_sp that allows walking the stack but isn't
duke@0 1761 // truly correct. Correct the value here.
duke@0 1762
duke@0 1763 if (extra_locals != 0 &&
duke@0 1764 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
duke@0 1765 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
duke@0 1766 }
duke@0 1767 *interpreter_frame->interpreter_frame_cache_addr() =
duke@0 1768 method->constants()->cache();
duke@0 1769 }
duke@0 1770 return size;
duke@0 1771 }
duke@0 1772
duke@0 1773
duke@0 1774 //------------------------------------------------------------------------------------------------------------------------
duke@0 1775 // Exceptions
duke@0 1776
duke@0 1777 void TemplateInterpreterGenerator::generate_throw_exception() {
duke@0 1778 // Entry point in previous activation (i.e., if the caller was interpreted)
duke@0 1779 Interpreter::_rethrow_exception_entry = __ pc();
twisti@1295 1780 const Register thread = rcx;
duke@0 1781
duke@0 1782 // Restore sp to interpreter_frame_last_sp even though we are going
duke@0 1783 // to empty the expression stack for the exception processing.
xlu@512 1784 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
duke@0 1785 // rax,: exception
duke@0 1786 // rdx: return address/pc that threw exception
duke@0 1787 __ restore_bcp(); // rsi points to call/send
duke@0 1788 __ restore_locals();
duke@0 1789
duke@0 1790 // Entry point for exceptions thrown within interpreter code
duke@0 1791 Interpreter::_throw_exception_entry = __ pc();
duke@0 1792 // expression stack is undefined here
duke@0 1793 // rax,: exception
duke@0 1794 // rsi: exception bcp
duke@0 1795 __ verify_oop(rax);
duke@0 1796
duke@0 1797 // expression stack must be empty before entering the VM in case of an exception
duke@0 1798 __ empty_expression_stack();
duke@0 1799 __ empty_FPU_stack();
duke@0 1800 // find exception handler address and preserve exception oop
duke@0 1801 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
duke@0 1802 // rax,: exception handler entry point
duke@0 1803 // rdx: preserved exception oop
duke@0 1804 // rsi: bcp for exception handler
duke@0 1805 __ push_ptr(rdx); // push exception which is now the only value on the stack
duke@0 1806 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
duke@0 1807
duke@0 1808 // If the exception is not handled in the current frame the frame is removed and
duke@0 1809 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
duke@0 1810 //
duke@0 1811 // Note: At this point the bci is still the bxi for the instruction which caused
duke@0 1812 // the exception and the expression stack is empty. Thus, for any VM calls
duke@0 1813 // at this point, GC will find a legal oop map (with empty expression stack).
duke@0 1814
duke@0 1815 // In current activation
duke@0 1816 // tos: exception
duke@0 1817 // rsi: exception bcp
duke@0 1818
duke@0 1819 //
duke@0 1820 // JVMTI PopFrame support
duke@0 1821 //
duke@0 1822
duke@0 1823 Interpreter::_remove_activation_preserving_args_entry = __ pc();
duke@0 1824 __ empty_expression_stack();
duke@0 1825 __ empty_FPU_stack();
duke@0 1826 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
duke@0 1827 // currently handling popframe, so that call_VMs that may happen later do not trigger new
duke@0 1828 // popframe handling cycles.
twisti@1295 1829 __ get_thread(thread);
twisti@1295 1830 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
duke@0 1831 __ orl(rdx, JavaThread::popframe_processing_bit);
twisti@1295 1832 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
duke@0 1833
duke@0 1834 {
duke@0 1835 // Check to see whether we are returning to a deoptimized frame.
duke@0 1836 // (The PopFrame call ensures that the caller of the popped frame is
duke@0 1837 // either interpreted or compiled and deoptimizes it if compiled.)
duke@0 1838 // In this case, we can't call dispatch_next() after the frame is
duke@0 1839 // popped, but instead must save the incoming arguments and restore
duke@0 1840 // them after deoptimization has occurred.
duke@0 1841 //
duke@0 1842 // Note that we don't compare the return PC against the
duke@0 1843 // deoptimization blob's unpack entry because of the presence of
duke@0 1844 // adapter frames in C2.
duke@0 1845 Label caller_not_deoptimized;
never@304 1846 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
duke@0 1847 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
duke@0 1848 __ testl(rax, rax);
duke@0 1849 __ jcc(Assembler::notZero, caller_not_deoptimized);
duke@0 1850
duke@0 1851 // Compute size of arguments for saving when returning to deoptimized caller
duke@0 1852 __ get_method(rax);
jiangli@3903 1853 __ movptr(rax, Address(rax, Method::const_offset()));
jiangli@3903 1854 __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset()));
twisti@1426 1855 __ shlptr(rax, Interpreter::logStackElementSize);
duke@0 1856 __ restore_locals();
never@304 1857 __ subptr(rdi, rax);
never@304 1858 __ addptr(rdi, wordSize);
duke@0 1859 // Save these arguments
twisti@1295 1860 __ get_thread(thread);
twisti@1295 1861 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
duke@0 1862
duke@0 1863 __ remove_activation(vtos, rdx,
duke@0 1864 /* throw_monitor_exception */ false,
duke@0 1865 /* install_monitor_exception */ false,
duke@0 1866 /* notify_jvmdi */ false);
duke@0 1867
duke@0 1868 // Inform deoptimization that it is responsible for restoring these arguments
twisti@1295 1869 __ get_thread(thread);
twisti@1295 1870 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
duke@0 1871
duke@0 1872 // Continue in deoptimization handler
duke@0 1873 __ jmp(rdx);
duke@0 1874
duke@0 1875 __ bind(caller_not_deoptimized);
duke@0 1876 }
duke@0 1877
duke@0 1878 __ remove_activation(vtos, rdx,
duke@0 1879 /* throw_monitor_exception */ false,
duke@0 1880 /* install_monitor_exception */ false,
duke@0 1881 /* notify_jvmdi */ false);
duke@0 1882
duke@0 1883 // Finish with popframe handling
duke@0 1884 // A previous I2C followed by a deoptimization might have moved the
duke@0 1885 // outgoing arguments further up the stack. PopFrame expects the
duke@0 1886 // mutations to those outgoing arguments to be preserved and other
duke@0 1887 // constraints basically require this frame to look exactly as
duke@0 1888 // though it had previously invoked an interpreted activation with
duke@0 1889 // no space between the top of the expression stack (current
duke@0 1890 // last_sp) and the top of stack. Rather than force deopt to
duke@0 1891 // maintain this kind of invariant all the time we call a small
duke@0 1892 // fixup routine to move the mutated arguments onto the top of our
duke@0 1893 // expression stack if necessary.
never@304 1894 __ mov(rax, rsp);
never@304 1895 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
twisti@1295 1896 __ get_thread(thread);
duke@0 1897 // PC must point into interpreter here
twisti@1295 1898 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
twisti@1295 1899 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
twisti@1295 1900 __ get_thread(thread);
twisti@1295 1901 __ reset_last_Java_frame(thread, true, true);
duke@0 1902 // Restore the last_sp and null it out
never@304 1903 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
xlu@512 1904 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
duke@0 1905
duke@0 1906 __ restore_bcp();
duke@0 1907 __ restore_locals();
duke@0 1908 // The method data pointer was incremented already during
duke@0 1909 // call profiling. We have to restore the mdp for the current bcp.
duke@0 1910 if (ProfileInterpreter) {
duke@0 1911 __ set_method_data_pointer_for_bcp();
duke@0 1912 }
duke@0 1913
duke@0 1914 // Clear the popframe condition flag
twisti@1295 1915 __ get_thread(thread);
twisti@1295 1916 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
duke@0 1917
sspitsyn@5061 1918 #if INCLUDE_JVMTI
sspitsyn@5061 1919 if (EnableInvokeDynamic) {
sspitsyn@5061 1920 Label L_done;
sspitsyn@5061 1921 const Register local0 = rdi;
sspitsyn@5061 1922
sspitsyn@5061 1923 __ cmpb(Address(rsi, 0), Bytecodes::_invokestatic);
sspitsyn@5061 1924 __ jcc(Assembler::notEqual, L_done);
sspitsyn@5061 1925
sspitsyn@5061 1926 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
sspitsyn@5061 1927 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
sspitsyn@5061 1928
sspitsyn@5061 1929 __ get_method(rdx);
sspitsyn@5061 1930 __ movptr(rax, Address(local0, 0));
sspitsyn@5061 1931 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rsi);
sspitsyn@5061 1932
sspitsyn@5061 1933 __ testptr(rax, rax);
sspitsyn@5061 1934 __ jcc(Assembler::zero, L_done);
sspitsyn@5061 1935
sspitsyn@5061 1936 __ movptr(Address(rbx, 0), rax);
sspitsyn@5061 1937 __ bind(L_done);
sspitsyn@5061 1938 }
sspitsyn@5061 1939 #endif // INCLUDE_JVMTI
sspitsyn@5061 1940
duke@0 1941 __ dispatch_next(vtos);
duke@0 1942 // end of PopFrame support
duke@0 1943
duke@0 1944 Interpreter::_remove_activation_entry = __ pc();
duke@0 1945
duke@0 1946 // preserve exception over this code sequence
duke@0 1947 __ pop_ptr(rax);
twisti@1295 1948 __ get_thread(thread);
twisti@1295 1949 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
duke@0 1950 // remove the activation (without doing throws on illegalMonitorExceptions)
duke@0 1951 __ remove_activation(vtos, rdx, false, true, false);
duke@0 1952 // restore exception
twisti@1295 1953 __ get_thread(thread);
coleenp@3602 1954 __ get_vm_result(rax, thread);
duke@0 1955
duke@0 1956 // Inbetween activations - previous activation type unknown yet
duke@0 1957 // compute continuation point - the continuation point expects
duke@0 1958 // the following registers set up:
duke@0 1959 //
twisti@1295 1960 // rax: exception
duke@0 1961 // rdx: return address/pc that threw exception
duke@0 1962 // rsp: expression stack of caller
twisti@1295 1963 // rbp: rbp, of caller
never@304 1964 __ push(rax); // save exception
never@304 1965 __ push(rdx); // save return address
twisti@1295 1966 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
never@304 1967 __ mov(rbx, rax); // save exception handler
never@304 1968 __ pop(rdx); // restore return address
never@304 1969 __ pop(rax); // restore exception
duke@0 1970 // Note that an "issuing PC" is actually the next PC after the call
duke@0 1971 __ jmp(rbx); // jump to exception handler of caller
duke@0 1972 }
duke@0 1973
duke@0 1974
duke@0 1975 //
duke@0 1976 // JVMTI ForceEarlyReturn support
duke@0 1977 //
duke@0 1978 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
duke@0 1979 address entry = __ pc();
twisti@1295 1980 const Register thread = rcx;
duke@0 1981
duke@0 1982 __ restore_bcp();
duke@0 1983 __ restore_locals();
duke@0 1984 __ empty_expression_stack();
duke@0 1985 __ empty_FPU_stack();
duke@0 1986 __ load_earlyret_value(state);
duke@0 1987
twisti@1295 1988 __ get_thread(thread);
twisti@1295 1989 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
duke@0 1990 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
duke@0 1991
duke@0 1992 // Clear the earlyret state
duke@0 1993 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
duke@0 1994
duke@0 1995 __ remove_activation(state, rsi,
duke@0 1996 false, /* throw_monitor_exception */
duke@0 1997 false, /* install_monitor_exception */
duke@0 1998 true); /* notify_jvmdi */
duke@0 1999 __ jmp(rsi);
duke@0 2000 return entry;
duke@0 2001 } // end of ForceEarlyReturn support
duke@0 2002
duke@0 2003
duke@0 2004 //------------------------------------------------------------------------------------------------------------------------
duke@0 2005 // Helper for vtos entry point generation
duke@0 2006
duke@0 2007 void TemplateInterpreterGenerator::set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) {
duke@0 2008 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
duke@0 2009 Label L;
duke@0 2010 fep = __ pc(); __ push(ftos); __ jmp(L);
duke@0 2011 dep = __ pc(); __ push(dtos); __ jmp(L);
duke@0 2012 lep = __ pc(); __ push(ltos); __ jmp(L);
duke@0 2013 aep = __ pc(); __ push(atos); __ jmp(L);
duke@0 2014 bep = cep = sep = // fall through
duke@0 2015 iep = __ pc(); __ push(itos); // fall through
duke@0 2016 vep = __ pc(); __ bind(L); // fall through
duke@0 2017 generate_and_dispatch(t);
duke@0 2018 }
duke@0 2019
duke@0 2020 //------------------------------------------------------------------------------------------------------------------------
duke@0 2021 // Generation of individual instructions
duke@0 2022
duke@0 2023 // helpers for generate_and_dispatch
duke@0 2024
duke@0 2025
duke@0 2026
duke@0 2027 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
duke@0 2028 : TemplateInterpreterGenerator(code) {
duke@0 2029 generate_all(); // down here so it can be "virtual"
duke@0 2030 }
duke@0 2031
duke@0 2032 //------------------------------------------------------------------------------------------------------------------------
duke@0 2033
duke@0 2034 // Non-product code
duke@0 2035 #ifndef PRODUCT
duke@0 2036 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
duke@0 2037 address entry = __ pc();
duke@0 2038
duke@0 2039 // prepare expression stack
never@304 2040 __ pop(rcx); // pop return address so expression stack is 'pure'
duke@0 2041 __ push(state); // save tosca
duke@0 2042
duke@0 2043 // pass tosca registers as arguments & call tracer
duke@0 2044 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
never@304 2045 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
duke@0 2046 __ pop(state); // restore tosca
duke@0 2047
duke@0 2048 // return
duke@0 2049 __ jmp(rcx);
duke@0 2050
duke@0 2051 return entry;
duke@0 2052 }
duke@0 2053
duke@0 2054
duke@0 2055 void TemplateInterpreterGenerator::count_bytecode() {
never@304 2056 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
duke@0 2057 }
duke@0 2058
duke@0 2059
duke@0 2060 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
never@304 2061 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
duke@0 2062 }
duke@0 2063
duke@0 2064
duke@0 2065 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
duke@0 2066 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
duke@0 2067 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
duke@0 2068 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
duke@0 2069 ExternalAddress table((address) BytecodePairHistogram::_counters);
duke@0 2070 Address index(noreg, rbx, Address::times_4);
never@304 2071 __ incrementl(ArrayAddress(table, index));
duke@0 2072 }
duke@0 2073
duke@0 2074
duke@0 2075 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
duke@0 2076 // Call a little run-time stub to avoid blow-up for each bytecode.
duke@0 2077 // The run-time runtime saves the right registers, depending on
duke@0 2078 // the tosca in-state for the given template.
duke@0 2079 assert(Interpreter::trace_code(t->tos_in()) != NULL,
duke@0 2080 "entry must have been generated");
duke@0 2081 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
duke@0 2082 }
duke@0 2083
duke@0 2084
duke@0 2085 void TemplateInterpreterGenerator::stop_interpreter_at() {
duke@0 2086 Label L;
duke@0 2087 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
duke@0 2088 StopInterpreterAt);
duke@0 2089 __ jcc(Assembler::notEqual, L);
duke@0 2090 __ int3();
duke@0 2091 __ bind(L);
duke@0 2092 }
duke@0 2093 #endif // !PRODUCT
duke@0 2094 #endif // CC_INTERP