1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/cpu/x86/vm/templateInterpreter_x86_64.cpp Wed Apr 27 01:25:04 2016 +0800
1.3 @@ -0,0 +1,2046 @@
1.4 +/*
1.5 + * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved.
1.6 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
1.7 + *
1.8 + * This code is free software; you can redistribute it and/or modify it
1.9 + * under the terms of the GNU General Public License version 2 only, as
1.10 + * published by the Free Software Foundation.
1.11 + *
1.12 + * This code is distributed in the hope that it will be useful, but WITHOUT
1.13 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
1.14 + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
1.15 + * version 2 for more details (a copy is included in the LICENSE file that
1.16 + * accompanied this code).
1.17 + *
1.18 + * You should have received a copy of the GNU General Public License version
1.19 + * 2 along with this work; if not, write to the Free Software Foundation,
1.20 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
1.21 + *
1.22 + * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
1.23 + * or visit www.oracle.com if you need additional information or have any
1.24 + * questions.
1.25 + *
1.26 + */
1.27 +
1.28 +#include "precompiled.hpp"
1.29 +#include "asm/macroAssembler.hpp"
1.30 +#include "interpreter/bytecodeHistogram.hpp"
1.31 +#include "interpreter/interpreter.hpp"
1.32 +#include "interpreter/interpreterGenerator.hpp"
1.33 +#include "interpreter/interpreterRuntime.hpp"
1.34 +#include "interpreter/templateTable.hpp"
1.35 +#include "oops/arrayOop.hpp"
1.36 +#include "oops/methodData.hpp"
1.37 +#include "oops/method.hpp"
1.38 +#include "oops/oop.inline.hpp"
1.39 +#include "prims/jvmtiExport.hpp"
1.40 +#include "prims/jvmtiThreadState.hpp"
1.41 +#include "runtime/arguments.hpp"
1.42 +#include "runtime/deoptimization.hpp"
1.43 +#include "runtime/frame.inline.hpp"
1.44 +#include "runtime/sharedRuntime.hpp"
1.45 +#include "runtime/stubRoutines.hpp"
1.46 +#include "runtime/synchronizer.hpp"
1.47 +#include "runtime/timer.hpp"
1.48 +#include "runtime/vframeArray.hpp"
1.49 +#include "utilities/debug.hpp"
1.50 +#include "utilities/macros.hpp"
1.51 +
1.52 +#define __ _masm->
1.53 +
1.54 +#ifndef CC_INTERP
1.55 +
1.56 +const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1.57 +const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
1.58 +const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
1.59 +
1.60 +//-----------------------------------------------------------------------------
1.61 +
1.62 +address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
1.63 + address entry = __ pc();
1.64 +
1.65 +#ifdef ASSERT
1.66 + {
1.67 + Label L;
1.68 + __ lea(rax, Address(rbp,
1.69 + frame::interpreter_frame_monitor_block_top_offset *
1.70 + wordSize));
1.71 + __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
1.72 + // grows negative)
1.73 + __ jcc(Assembler::aboveEqual, L); // check if frame is complete
1.74 + __ stop ("interpreter frame not set up");
1.75 + __ bind(L);
1.76 + }
1.77 +#endif // ASSERT
1.78 + // Restore bcp under the assumption that the current frame is still
1.79 + // interpreted
1.80 + __ restore_bcp();
1.81 +
1.82 + // expression stack must be empty before entering the VM if an
1.83 + // exception happened
1.84 + __ empty_expression_stack();
1.85 + // throw exception
1.86 + __ call_VM(noreg,
1.87 + CAST_FROM_FN_PTR(address,
1.88 + InterpreterRuntime::throw_StackOverflowError));
1.89 + return entry;
1.90 +}
1.91 +
1.92 +address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
1.93 + const char* name) {
1.94 + address entry = __ pc();
1.95 + // expression stack must be empty before entering the VM if an
1.96 + // exception happened
1.97 + __ empty_expression_stack();
1.98 + // setup parameters
1.99 + // ??? convention: expect aberrant index in register ebx
1.100 + __ lea(c_rarg1, ExternalAddress((address)name));
1.101 + __ call_VM(noreg,
1.102 + CAST_FROM_FN_PTR(address,
1.103 + InterpreterRuntime::
1.104 + throw_ArrayIndexOutOfBoundsException),
1.105 + c_rarg1, rbx);
1.106 + return entry;
1.107 +}
1.108 +
1.109 +address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
1.110 + address entry = __ pc();
1.111 +
1.112 + // object is at TOS
1.113 + __ pop(c_rarg1);
1.114 +
1.115 + // expression stack must be empty before entering the VM if an
1.116 + // exception happened
1.117 + __ empty_expression_stack();
1.118 +
1.119 + __ call_VM(noreg,
1.120 + CAST_FROM_FN_PTR(address,
1.121 + InterpreterRuntime::
1.122 + throw_ClassCastException),
1.123 + c_rarg1);
1.124 + return entry;
1.125 +}
1.126 +
1.127 +address TemplateInterpreterGenerator::generate_exception_handler_common(
1.128 + const char* name, const char* message, bool pass_oop) {
1.129 + assert(!pass_oop || message == NULL, "either oop or message but not both");
1.130 + address entry = __ pc();
1.131 + if (pass_oop) {
1.132 + // object is at TOS
1.133 + __ pop(c_rarg2);
1.134 + }
1.135 + // expression stack must be empty before entering the VM if an
1.136 + // exception happened
1.137 + __ empty_expression_stack();
1.138 + // setup parameters
1.139 + __ lea(c_rarg1, ExternalAddress((address)name));
1.140 + if (pass_oop) {
1.141 + __ call_VM(rax, CAST_FROM_FN_PTR(address,
1.142 + InterpreterRuntime::
1.143 + create_klass_exception),
1.144 + c_rarg1, c_rarg2);
1.145 + } else {
1.146 + // kind of lame ExternalAddress can't take NULL because
1.147 + // external_word_Relocation will assert.
1.148 + if (message != NULL) {
1.149 + __ lea(c_rarg2, ExternalAddress((address)message));
1.150 + } else {
1.151 + __ movptr(c_rarg2, NULL_WORD);
1.152 + }
1.153 + __ call_VM(rax,
1.154 + CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
1.155 + c_rarg1, c_rarg2);
1.156 + }
1.157 + // throw exception
1.158 + __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
1.159 + return entry;
1.160 +}
1.161 +
1.162 +
1.163 +address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
1.164 + address entry = __ pc();
1.165 + // NULL last_sp until next java call
1.166 + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1.167 + __ dispatch_next(state);
1.168 + return entry;
1.169 +}
1.170 +
1.171 +
1.172 +address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
1.173 + address entry = __ pc();
1.174 +
1.175 + // Restore stack bottom in case i2c adjusted stack
1.176 + __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1.177 + // and NULL it as marker that esp is now tos until next java call
1.178 + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1.179 +
1.180 + __ restore_bcp();
1.181 + __ restore_locals();
1.182 +
1.183 + if (state == atos) {
1.184 + Register mdp = rbx;
1.185 + Register tmp = rcx;
1.186 + __ profile_return_type(mdp, rax, tmp);
1.187 + }
1.188 +
1.189 + const Register cache = rbx;
1.190 + const Register index = rcx;
1.191 + __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
1.192 +
1.193 + const Register flags = cache;
1.194 + __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
1.195 + __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
1.196 + __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
1.197 + __ dispatch_next(state, step);
1.198 +
1.199 + return entry;
1.200 +}
1.201 +
1.202 +
1.203 +address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
1.204 + int step) {
1.205 + address entry = __ pc();
1.206 + // NULL last_sp until next java call
1.207 + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1.208 + __ restore_bcp();
1.209 + __ restore_locals();
1.210 + // handle exceptions
1.211 + {
1.212 + Label L;
1.213 + __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1.214 + __ jcc(Assembler::zero, L);
1.215 + __ call_VM(noreg,
1.216 + CAST_FROM_FN_PTR(address,
1.217 + InterpreterRuntime::throw_pending_exception));
1.218 + __ should_not_reach_here();
1.219 + __ bind(L);
1.220 + }
1.221 + __ dispatch_next(state, step);
1.222 + return entry;
1.223 +}
1.224 +
1.225 +int AbstractInterpreter::BasicType_as_index(BasicType type) {
1.226 + int i = 0;
1.227 + switch (type) {
1.228 + case T_BOOLEAN: i = 0; break;
1.229 + case T_CHAR : i = 1; break;
1.230 + case T_BYTE : i = 2; break;
1.231 + case T_SHORT : i = 3; break;
1.232 + case T_INT : i = 4; break;
1.233 + case T_LONG : i = 5; break;
1.234 + case T_VOID : i = 6; break;
1.235 + case T_FLOAT : i = 7; break;
1.236 + case T_DOUBLE : i = 8; break;
1.237 + case T_OBJECT : i = 9; break;
1.238 + case T_ARRAY : i = 9; break;
1.239 + default : ShouldNotReachHere();
1.240 + }
1.241 + assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
1.242 + "index out of bounds");
1.243 + return i;
1.244 +}
1.245 +
1.246 +
1.247 +address TemplateInterpreterGenerator::generate_result_handler_for(
1.248 + BasicType type) {
1.249 + address entry = __ pc();
1.250 + switch (type) {
1.251 + case T_BOOLEAN: __ c2bool(rax); break;
1.252 + case T_CHAR : __ movzwl(rax, rax); break;
1.253 + case T_BYTE : __ sign_extend_byte(rax); break;
1.254 + case T_SHORT : __ sign_extend_short(rax); break;
1.255 + case T_INT : /* nothing to do */ break;
1.256 + case T_LONG : /* nothing to do */ break;
1.257 + case T_VOID : /* nothing to do */ break;
1.258 + case T_FLOAT : /* nothing to do */ break;
1.259 + case T_DOUBLE : /* nothing to do */ break;
1.260 + case T_OBJECT :
1.261 + // retrieve result from frame
1.262 + __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
1.263 + // and verify it
1.264 + __ verify_oop(rax);
1.265 + break;
1.266 + default : ShouldNotReachHere();
1.267 + }
1.268 + __ ret(0); // return from result handler
1.269 + return entry;
1.270 +}
1.271 +
1.272 +address TemplateInterpreterGenerator::generate_safept_entry_for(
1.273 + TosState state,
1.274 + address runtime_entry) {
1.275 + address entry = __ pc();
1.276 + __ push(state);
1.277 + __ call_VM(noreg, runtime_entry);
1.278 + __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
1.279 + return entry;
1.280 +}
1.281 +
1.282 +
1.283 +
1.284 +// Helpers for commoning out cases in the various type of method entries.
1.285 +//
1.286 +
1.287 +
1.288 +// increment invocation count & check for overflow
1.289 +//
1.290 +// Note: checking for negative value instead of overflow
1.291 +// so we have a 'sticky' overflow test
1.292 +//
1.293 +// rbx: method
1.294 +// ecx: invocation counter
1.295 +//
1.296 +void InterpreterGenerator::generate_counter_incr(
1.297 + Label* overflow,
1.298 + Label* profile_method,
1.299 + Label* profile_method_continue) {
1.300 + Label done;
1.301 + // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
1.302 + if (TieredCompilation) {
1.303 + int increment = InvocationCounter::count_increment;
1.304 + int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1.305 + Label no_mdo;
1.306 + if (ProfileInterpreter) {
1.307 + // Are we profiling?
1.308 + __ movptr(rax, Address(rbx, Method::method_data_offset()));
1.309 + __ testptr(rax, rax);
1.310 + __ jccb(Assembler::zero, no_mdo);
1.311 + // Increment counter in the MDO
1.312 + const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
1.313 + in_bytes(InvocationCounter::counter_offset()));
1.314 + __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
1.315 + __ jmp(done);
1.316 + }
1.317 + __ bind(no_mdo);
1.318 + // Increment counter in MethodCounters
1.319 + const Address invocation_counter(rax,
1.320 + MethodCounters::invocation_counter_offset() +
1.321 + InvocationCounter::counter_offset());
1.322 + __ get_method_counters(rbx, rax, done);
1.323 + __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
1.324 + false, Assembler::zero, overflow);
1.325 + __ bind(done);
1.326 + } else {
1.327 + const Address backedge_counter(rax,
1.328 + MethodCounters::backedge_counter_offset() +
1.329 + InvocationCounter::counter_offset());
1.330 + const Address invocation_counter(rax,
1.331 + MethodCounters::invocation_counter_offset() +
1.332 + InvocationCounter::counter_offset());
1.333 +
1.334 + __ get_method_counters(rbx, rax, done);
1.335 +
1.336 + if (ProfileInterpreter) {
1.337 + __ incrementl(Address(rax,
1.338 + MethodCounters::interpreter_invocation_counter_offset()));
1.339 + }
1.340 + // Update standard invocation counters
1.341 + __ movl(rcx, invocation_counter);
1.342 + __ incrementl(rcx, InvocationCounter::count_increment);
1.343 + __ movl(invocation_counter, rcx); // save invocation count
1.344 +
1.345 + __ movl(rax, backedge_counter); // load backedge counter
1.346 + __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
1.347 +
1.348 + __ addl(rcx, rax); // add both counters
1.349 +
1.350 + // profile_method is non-null only for interpreted method so
1.351 + // profile_method != NULL == !native_call
1.352 +
1.353 + if (ProfileInterpreter && profile_method != NULL) {
1.354 + // Test to see if we should create a method data oop
1.355 + __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
1.356 + __ jcc(Assembler::less, *profile_method_continue);
1.357 +
1.358 + // if no method data exists, go to profile_method
1.359 + __ test_method_data_pointer(rax, *profile_method);
1.360 + }
1.361 +
1.362 + __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
1.363 + __ jcc(Assembler::aboveEqual, *overflow);
1.364 + __ bind(done);
1.365 + }
1.366 +}
1.367 +
1.368 +void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
1.369 +
1.370 + // Asm interpreter on entry
1.371 + // r14 - locals
1.372 + // r13 - bcp
1.373 + // rbx - method
1.374 + // edx - cpool --- DOES NOT APPEAR TO BE TRUE
1.375 + // rbp - interpreter frame
1.376 +
1.377 + // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
1.378 + // Everything as it was on entry
1.379 + // rdx is not restored. Doesn't appear to really be set.
1.380 +
1.381 + // InterpreterRuntime::frequency_counter_overflow takes two
1.382 + // arguments, the first (thread) is passed by call_VM, the second
1.383 + // indicates if the counter overflow occurs at a backwards branch
1.384 + // (NULL bcp). We pass zero for it. The call returns the address
1.385 + // of the verified entry point for the method or NULL if the
1.386 + // compilation did not complete (either went background or bailed
1.387 + // out).
1.388 + __ movl(c_rarg1, 0);
1.389 + __ call_VM(noreg,
1.390 + CAST_FROM_FN_PTR(address,
1.391 + InterpreterRuntime::frequency_counter_overflow),
1.392 + c_rarg1);
1.393 +
1.394 + __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
1.395 + // Preserve invariant that r13/r14 contain bcp/locals of sender frame
1.396 + // and jump to the interpreted entry.
1.397 + __ jmp(*do_continue, relocInfo::none);
1.398 +}
1.399 +
1.400 +// See if we've got enough room on the stack for locals plus overhead.
1.401 +// The expression stack grows down incrementally, so the normal guard
1.402 +// page mechanism will work for that.
1.403 +//
1.404 +// NOTE: Since the additional locals are also always pushed (wasn't
1.405 +// obvious in generate_method_entry) so the guard should work for them
1.406 +// too.
1.407 +//
1.408 +// Args:
1.409 +// rdx: number of additional locals this frame needs (what we must check)
1.410 +// rbx: Method*
1.411 +//
1.412 +// Kills:
1.413 +// rax
1.414 +void InterpreterGenerator::generate_stack_overflow_check(void) {
1.415 +
1.416 + // monitor entry size: see picture of stack set
1.417 + // (generate_method_entry) and frame_amd64.hpp
1.418 + const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1.419 +
1.420 + // total overhead size: entry_size + (saved rbp through expr stack
1.421 + // bottom). be sure to change this if you add/subtract anything
1.422 + // to/from the overhead area
1.423 + const int overhead_size =
1.424 + -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
1.425 +
1.426 + const int page_size = os::vm_page_size();
1.427 +
1.428 + Label after_frame_check;
1.429 +
1.430 + // see if the frame is greater than one page in size. If so,
1.431 + // then we need to verify there is enough stack space remaining
1.432 + // for the additional locals.
1.433 + __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
1.434 + __ jcc(Assembler::belowEqual, after_frame_check);
1.435 +
1.436 + // compute rsp as if this were going to be the last frame on
1.437 + // the stack before the red zone
1.438 +
1.439 + const Address stack_base(r15_thread, Thread::stack_base_offset());
1.440 + const Address stack_size(r15_thread, Thread::stack_size_offset());
1.441 +
1.442 + // locals + overhead, in bytes
1.443 + __ mov(rax, rdx);
1.444 + __ shlptr(rax, Interpreter::logStackElementSize); // 2 slots per parameter.
1.445 + __ addptr(rax, overhead_size);
1.446 +
1.447 +#ifdef ASSERT
1.448 + Label stack_base_okay, stack_size_okay;
1.449 + // verify that thread stack base is non-zero
1.450 + __ cmpptr(stack_base, (int32_t)NULL_WORD);
1.451 + __ jcc(Assembler::notEqual, stack_base_okay);
1.452 + __ stop("stack base is zero");
1.453 + __ bind(stack_base_okay);
1.454 + // verify that thread stack size is non-zero
1.455 + __ cmpptr(stack_size, 0);
1.456 + __ jcc(Assembler::notEqual, stack_size_okay);
1.457 + __ stop("stack size is zero");
1.458 + __ bind(stack_size_okay);
1.459 +#endif
1.460 +
1.461 + // Add stack base to locals and subtract stack size
1.462 + __ addptr(rax, stack_base);
1.463 + __ subptr(rax, stack_size);
1.464 +
1.465 + // Use the maximum number of pages we might bang.
1.466 + const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
1.467 + (StackRedPages+StackYellowPages);
1.468 +
1.469 + // add in the red and yellow zone sizes
1.470 + __ addptr(rax, max_pages * page_size);
1.471 +
1.472 + // check against the current stack bottom
1.473 + __ cmpptr(rsp, rax);
1.474 + __ jcc(Assembler::above, after_frame_check);
1.475 +
1.476 + // Restore sender's sp as SP. This is necessary if the sender's
1.477 + // frame is an extended compiled frame (see gen_c2i_adapter())
1.478 + // and safer anyway in case of JSR292 adaptations.
1.479 +
1.480 + __ pop(rax); // return address must be moved if SP is changed
1.481 + __ mov(rsp, r13);
1.482 + __ push(rax);
1.483 +
1.484 + // Note: the restored frame is not necessarily interpreted.
1.485 + // Use the shared runtime version of the StackOverflowError.
1.486 + assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
1.487 + __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
1.488 +
1.489 + // all done with frame size check
1.490 + __ bind(after_frame_check);
1.491 +}
1.492 +
1.493 +// Allocate monitor and lock method (asm interpreter)
1.494 +//
1.495 +// Args:
1.496 +// rbx: Method*
1.497 +// r14: locals
1.498 +//
1.499 +// Kills:
1.500 +// rax
1.501 +// c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
1.502 +// rscratch1, rscratch2 (scratch regs)
1.503 +void InterpreterGenerator::lock_method(void) {
1.504 + // synchronize method
1.505 + const Address access_flags(rbx, Method::access_flags_offset());
1.506 + const Address monitor_block_top(
1.507 + rbp,
1.508 + frame::interpreter_frame_monitor_block_top_offset * wordSize);
1.509 + const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1.510 +
1.511 +#ifdef ASSERT
1.512 + {
1.513 + Label L;
1.514 + __ movl(rax, access_flags);
1.515 + __ testl(rax, JVM_ACC_SYNCHRONIZED);
1.516 + __ jcc(Assembler::notZero, L);
1.517 + __ stop("method doesn't need synchronization");
1.518 + __ bind(L);
1.519 + }
1.520 +#endif // ASSERT
1.521 +
1.522 + // get synchronization object
1.523 + {
1.524 + const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1.525 + Label done;
1.526 + __ movl(rax, access_flags);
1.527 + __ testl(rax, JVM_ACC_STATIC);
1.528 + // get receiver (assume this is frequent case)
1.529 + __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
1.530 + __ jcc(Assembler::zero, done);
1.531 + __ movptr(rax, Address(rbx, Method::const_offset()));
1.532 + __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
1.533 + __ movptr(rax, Address(rax,
1.534 + ConstantPool::pool_holder_offset_in_bytes()));
1.535 + __ movptr(rax, Address(rax, mirror_offset));
1.536 +
1.537 +#ifdef ASSERT
1.538 + {
1.539 + Label L;
1.540 + __ testptr(rax, rax);
1.541 + __ jcc(Assembler::notZero, L);
1.542 + __ stop("synchronization object is NULL");
1.543 + __ bind(L);
1.544 + }
1.545 +#endif // ASSERT
1.546 +
1.547 + __ bind(done);
1.548 + }
1.549 +
1.550 + // add space for monitor & lock
1.551 + __ subptr(rsp, entry_size); // add space for a monitor entry
1.552 + __ movptr(monitor_block_top, rsp); // set new monitor block top
1.553 + // store object
1.554 + __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
1.555 + __ movptr(c_rarg1, rsp); // object address
1.556 + __ lock_object(c_rarg1);
1.557 +}
1.558 +
1.559 +// Generate a fixed interpreter frame. This is identical setup for
1.560 +// interpreted methods and for native methods hence the shared code.
1.561 +//
1.562 +// Args:
1.563 +// rax: return address
1.564 +// rbx: Method*
1.565 +// r14: pointer to locals
1.566 +// r13: sender sp
1.567 +// rdx: cp cache
1.568 +void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
1.569 + // initialize fixed part of activation frame
1.570 + __ push(rax); // save return address
1.571 + __ enter(); // save old & set new rbp
1.572 + __ push(r13); // set sender sp
1.573 + __ push((int)NULL_WORD); // leave last_sp as null
1.574 + __ movptr(r13, Address(rbx, Method::const_offset())); // get ConstMethod*
1.575 + __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase
1.576 + __ push(rbx); // save Method*
1.577 + if (ProfileInterpreter) {
1.578 + Label method_data_continue;
1.579 + __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
1.580 + __ testptr(rdx, rdx);
1.581 + __ jcc(Assembler::zero, method_data_continue);
1.582 + __ addptr(rdx, in_bytes(MethodData::data_offset()));
1.583 + __ bind(method_data_continue);
1.584 + __ push(rdx); // set the mdp (method data pointer)
1.585 + } else {
1.586 + __ push(0);
1.587 + }
1.588 +
1.589 + __ movptr(rdx, Address(rbx, Method::const_offset()));
1.590 + __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
1.591 + __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
1.592 + __ push(rdx); // set constant pool cache
1.593 + __ push(r14); // set locals pointer
1.594 + if (native_call) {
1.595 + __ push(0); // no bcp
1.596 + } else {
1.597 + __ push(r13); // set bcp
1.598 + }
1.599 + __ push(0); // reserve word for pointer to expression stack bottom
1.600 + __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
1.601 +}
1.602 +
1.603 +// End of helpers
1.604 +
1.605 +// Various method entries
1.606 +//------------------------------------------------------------------------------------------------------------------------
1.607 +//
1.608 +//
1.609 +
1.610 +// Call an accessor method (assuming it is resolved, otherwise drop
1.611 +// into vanilla (slow path) entry
1.612 +address InterpreterGenerator::generate_accessor_entry(void) {
1.613 + // rbx: Method*
1.614 +
1.615 + // r13: senderSP must preserver for slow path, set SP to it on fast path
1.616 +
1.617 + address entry_point = __ pc();
1.618 + Label xreturn_path;
1.619 +
1.620 + // do fastpath for resolved accessor methods
1.621 + if (UseFastAccessorMethods) {
1.622 + // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
1.623 + // thereof; parameter size = 1
1.624 + // Note: We can only use this code if the getfield has been resolved
1.625 + // and if we don't have a null-pointer exception => check for
1.626 + // these conditions first and use slow path if necessary.
1.627 + Label slow_path;
1.628 + // If we need a safepoint check, generate full interpreter entry.
1.629 + __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1.630 + SafepointSynchronize::_not_synchronized);
1.631 +
1.632 + __ jcc(Assembler::notEqual, slow_path);
1.633 + // rbx: method
1.634 + __ movptr(rax, Address(rsp, wordSize));
1.635 +
1.636 + // check if local 0 != NULL and read field
1.637 + __ testptr(rax, rax);
1.638 + __ jcc(Assembler::zero, slow_path);
1.639 +
1.640 + // read first instruction word and extract bytecode @ 1 and index @ 2
1.641 + __ movptr(rdx, Address(rbx, Method::const_offset()));
1.642 + __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
1.643 + __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
1.644 + // Shift codes right to get the index on the right.
1.645 + // The bytecode fetched looks like <index><0xb4><0x2a>
1.646 + __ shrl(rdx, 2 * BitsPerByte);
1.647 + __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
1.648 + __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
1.649 +
1.650 + // rax: local 0
1.651 + // rbx: method
1.652 + // rdx: constant pool cache index
1.653 + // rdi: constant pool cache
1.654 +
1.655 + // check if getfield has been resolved and read constant pool cache entry
1.656 + // check the validity of the cache entry by testing whether _indices field
1.657 + // contains Bytecode::_getfield in b1 byte.
1.658 + assert(in_words(ConstantPoolCacheEntry::size()) == 4,
1.659 + "adjust shift below");
1.660 + __ movl(rcx,
1.661 + Address(rdi,
1.662 + rdx,
1.663 + Address::times_8,
1.664 + ConstantPoolCache::base_offset() +
1.665 + ConstantPoolCacheEntry::indices_offset()));
1.666 + __ shrl(rcx, 2 * BitsPerByte);
1.667 + __ andl(rcx, 0xFF);
1.668 + __ cmpl(rcx, Bytecodes::_getfield);
1.669 + __ jcc(Assembler::notEqual, slow_path);
1.670 +
1.671 + // Note: constant pool entry is not valid before bytecode is resolved
1.672 + __ movptr(rcx,
1.673 + Address(rdi,
1.674 + rdx,
1.675 + Address::times_8,
1.676 + ConstantPoolCache::base_offset() +
1.677 + ConstantPoolCacheEntry::f2_offset()));
1.678 + // edx: flags
1.679 + __ movl(rdx,
1.680 + Address(rdi,
1.681 + rdx,
1.682 + Address::times_8,
1.683 + ConstantPoolCache::base_offset() +
1.684 + ConstantPoolCacheEntry::flags_offset()));
1.685 +
1.686 + Label notObj, notInt, notByte, notShort;
1.687 + const Address field_address(rax, rcx, Address::times_1);
1.688 +
1.689 + // Need to differentiate between igetfield, agetfield, bgetfield etc.
1.690 + // because they are different sizes.
1.691 + // Use the type from the constant pool cache
1.692 + __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
1.693 + // Make sure we don't need to mask edx after the above shift
1.694 + ConstantPoolCacheEntry::verify_tos_state_shift();
1.695 +
1.696 + __ cmpl(rdx, atos);
1.697 + __ jcc(Assembler::notEqual, notObj);
1.698 + // atos
1.699 + __ load_heap_oop(rax, field_address);
1.700 + __ jmp(xreturn_path);
1.701 +
1.702 + __ bind(notObj);
1.703 + __ cmpl(rdx, itos);
1.704 + __ jcc(Assembler::notEqual, notInt);
1.705 + // itos
1.706 + __ movl(rax, field_address);
1.707 + __ jmp(xreturn_path);
1.708 +
1.709 + __ bind(notInt);
1.710 + __ cmpl(rdx, btos);
1.711 + __ jcc(Assembler::notEqual, notByte);
1.712 + // btos
1.713 + __ load_signed_byte(rax, field_address);
1.714 + __ jmp(xreturn_path);
1.715 +
1.716 + __ bind(notByte);
1.717 + __ cmpl(rdx, stos);
1.718 + __ jcc(Assembler::notEqual, notShort);
1.719 + // stos
1.720 + __ load_signed_short(rax, field_address);
1.721 + __ jmp(xreturn_path);
1.722 +
1.723 + __ bind(notShort);
1.724 +#ifdef ASSERT
1.725 + Label okay;
1.726 + __ cmpl(rdx, ctos);
1.727 + __ jcc(Assembler::equal, okay);
1.728 + __ stop("what type is this?");
1.729 + __ bind(okay);
1.730 +#endif
1.731 + // ctos
1.732 + __ load_unsigned_short(rax, field_address);
1.733 +
1.734 + __ bind(xreturn_path);
1.735 +
1.736 + // _ireturn/_areturn
1.737 + __ pop(rdi);
1.738 + __ mov(rsp, r13);
1.739 + __ jmp(rdi);
1.740 + __ ret(0);
1.741 +
1.742 + // generate a vanilla interpreter entry as the slow path
1.743 + __ bind(slow_path);
1.744 + (void) generate_normal_entry(false);
1.745 + } else {
1.746 + (void) generate_normal_entry(false);
1.747 + }
1.748 +
1.749 + return entry_point;
1.750 +}
1.751 +
1.752 +// Method entry for java.lang.ref.Reference.get.
1.753 +address InterpreterGenerator::generate_Reference_get_entry(void) {
1.754 +#if INCLUDE_ALL_GCS
1.755 + // Code: _aload_0, _getfield, _areturn
1.756 + // parameter size = 1
1.757 + //
1.758 + // The code that gets generated by this routine is split into 2 parts:
1.759 + // 1. The "intrinsified" code for G1 (or any SATB based GC),
1.760 + // 2. The slow path - which is an expansion of the regular method entry.
1.761 + //
1.762 + // Notes:-
1.763 + // * In the G1 code we do not check whether we need to block for
1.764 + // a safepoint. If G1 is enabled then we must execute the specialized
1.765 + // code for Reference.get (except when the Reference object is null)
1.766 + // so that we can log the value in the referent field with an SATB
1.767 + // update buffer.
1.768 + // If the code for the getfield template is modified so that the
1.769 + // G1 pre-barrier code is executed when the current method is
1.770 + // Reference.get() then going through the normal method entry
1.771 + // will be fine.
1.772 + // * The G1 code can, however, check the receiver object (the instance
1.773 + // of java.lang.Reference) and jump to the slow path if null. If the
1.774 + // Reference object is null then we obviously cannot fetch the referent
1.775 + // and so we don't need to call the G1 pre-barrier. Thus we can use the
1.776 + // regular method entry code to generate the NPE.
1.777 + //
1.778 + // This code is based on generate_accessor_enty.
1.779 + //
1.780 + // rbx: Method*
1.781 +
1.782 + // r13: senderSP must preserve for slow path, set SP to it on fast path
1.783 +
1.784 + address entry = __ pc();
1.785 +
1.786 + const int referent_offset = java_lang_ref_Reference::referent_offset;
1.787 + guarantee(referent_offset > 0, "referent offset not initialized");
1.788 +
1.789 + if (UseG1GC) {
1.790 + Label slow_path;
1.791 + // rbx: method
1.792 +
1.793 + // Check if local 0 != NULL
1.794 + // If the receiver is null then it is OK to jump to the slow path.
1.795 + __ movptr(rax, Address(rsp, wordSize));
1.796 +
1.797 + __ testptr(rax, rax);
1.798 + __ jcc(Assembler::zero, slow_path);
1.799 +
1.800 + // rax: local 0
1.801 + // rbx: method (but can be used as scratch now)
1.802 + // rdx: scratch
1.803 + // rdi: scratch
1.804 +
1.805 + // Generate the G1 pre-barrier code to log the value of
1.806 + // the referent field in an SATB buffer.
1.807 +
1.808 + // Load the value of the referent field.
1.809 + const Address field_address(rax, referent_offset);
1.810 + __ load_heap_oop(rax, field_address);
1.811 +
1.812 + // Generate the G1 pre-barrier code to log the value of
1.813 + // the referent field in an SATB buffer.
1.814 + __ g1_write_barrier_pre(noreg /* obj */,
1.815 + rax /* pre_val */,
1.816 + r15_thread /* thread */,
1.817 + rbx /* tmp */,
1.818 + true /* tosca_live */,
1.819 + true /* expand_call */);
1.820 +
1.821 + // _areturn
1.822 + __ pop(rdi); // get return address
1.823 + __ mov(rsp, r13); // set sp to sender sp
1.824 + __ jmp(rdi);
1.825 + __ ret(0);
1.826 +
1.827 + // generate a vanilla interpreter entry as the slow path
1.828 + __ bind(slow_path);
1.829 + (void) generate_normal_entry(false);
1.830 +
1.831 + return entry;
1.832 + }
1.833 +#endif // INCLUDE_ALL_GCS
1.834 +
1.835 + // If G1 is not enabled then attempt to go through the accessor entry point
1.836 + // Reference.get is an accessor
1.837 + return generate_accessor_entry();
1.838 +}
1.839 +
1.840 +/**
1.841 + * Method entry for static native methods:
1.842 + * int java.util.zip.CRC32.update(int crc, int b)
1.843 + */
1.844 +address InterpreterGenerator::generate_CRC32_update_entry() {
1.845 + if (UseCRC32Intrinsics) {
1.846 + address entry = __ pc();
1.847 +
1.848 + // rbx,: Method*
1.849 + // r13: senderSP must preserved for slow path, set SP to it on fast path
1.850 + // c_rarg0: scratch (rdi on non-Win64, rcx on Win64)
1.851 + // c_rarg1: scratch (rsi on non-Win64, rdx on Win64)
1.852 +
1.853 + Label slow_path;
1.854 + // If we need a safepoint check, generate full interpreter entry.
1.855 + ExternalAddress state(SafepointSynchronize::address_of_state());
1.856 + __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1.857 + SafepointSynchronize::_not_synchronized);
1.858 + __ jcc(Assembler::notEqual, slow_path);
1.859 +
1.860 + // We don't generate local frame and don't align stack because
1.861 + // we call stub code and there is no safepoint on this path.
1.862 +
1.863 + // Load parameters
1.864 + const Register crc = rax; // crc
1.865 + const Register val = c_rarg0; // source java byte value
1.866 + const Register tbl = c_rarg1; // scratch
1.867 +
1.868 + // Arguments are reversed on java expression stack
1.869 + __ movl(val, Address(rsp, wordSize)); // byte value
1.870 + __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
1.871 +
1.872 + __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
1.873 + __ notl(crc); // ~crc
1.874 + __ update_byte_crc32(crc, val, tbl);
1.875 + __ notl(crc); // ~crc
1.876 + // result in rax
1.877 +
1.878 + // _areturn
1.879 + __ pop(rdi); // get return address
1.880 + __ mov(rsp, r13); // set sp to sender sp
1.881 + __ jmp(rdi);
1.882 +
1.883 + // generate a vanilla native entry as the slow path
1.884 + __ bind(slow_path);
1.885 +
1.886 + (void) generate_native_entry(false);
1.887 +
1.888 + return entry;
1.889 + }
1.890 + return generate_native_entry(false);
1.891 +}
1.892 +
1.893 +/**
1.894 + * Method entry for static native methods:
1.895 + * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1.896 + * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1.897 + */
1.898 +address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1.899 + if (UseCRC32Intrinsics) {
1.900 + address entry = __ pc();
1.901 +
1.902 + // rbx,: Method*
1.903 + // r13: senderSP must preserved for slow path, set SP to it on fast path
1.904 +
1.905 + Label slow_path;
1.906 + // If we need a safepoint check, generate full interpreter entry.
1.907 + ExternalAddress state(SafepointSynchronize::address_of_state());
1.908 + __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1.909 + SafepointSynchronize::_not_synchronized);
1.910 + __ jcc(Assembler::notEqual, slow_path);
1.911 +
1.912 + // We don't generate local frame and don't align stack because
1.913 + // we call stub code and there is no safepoint on this path.
1.914 +
1.915 + // Load parameters
1.916 + const Register crc = c_rarg0; // crc
1.917 + const Register buf = c_rarg1; // source java byte array address
1.918 + const Register len = c_rarg2; // length
1.919 + const Register off = len; // offset (never overlaps with 'len')
1.920 +
1.921 + // Arguments are reversed on java expression stack
1.922 + // Calculate address of start element
1.923 + if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1.924 + __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
1.925 + __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
1.926 + __ addq(buf, off); // + offset
1.927 + __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC
1.928 + } else {
1.929 + __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
1.930 + __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1.931 + __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
1.932 + __ addq(buf, off); // + offset
1.933 + __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC
1.934 + }
1.935 + // Can now load 'len' since we're finished with 'off'
1.936 + __ movl(len, Address(rsp, wordSize)); // Length
1.937 +
1.938 + __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
1.939 + // result in rax
1.940 +
1.941 + // _areturn
1.942 + __ pop(rdi); // get return address
1.943 + __ mov(rsp, r13); // set sp to sender sp
1.944 + __ jmp(rdi);
1.945 +
1.946 + // generate a vanilla native entry as the slow path
1.947 + __ bind(slow_path);
1.948 +
1.949 + (void) generate_native_entry(false);
1.950 +
1.951 + return entry;
1.952 + }
1.953 + return generate_native_entry(false);
1.954 +}
1.955 +
1.956 +// Interpreter stub for calling a native method. (asm interpreter)
1.957 +// This sets up a somewhat different looking stack for calling the
1.958 +// native method than the typical interpreter frame setup.
1.959 +address InterpreterGenerator::generate_native_entry(bool synchronized) {
1.960 + // determine code generation flags
1.961 + bool inc_counter = UseCompiler || CountCompiledCalls;
1.962 +
1.963 + // rbx: Method*
1.964 + // r13: sender sp
1.965 +
1.966 + address entry_point = __ pc();
1.967 +
1.968 + const Address constMethod (rbx, Method::const_offset());
1.969 + const Address access_flags (rbx, Method::access_flags_offset());
1.970 + const Address size_of_parameters(rcx, ConstMethod::
1.971 + size_of_parameters_offset());
1.972 +
1.973 +
1.974 + // get parameter size (always needed)
1.975 + __ movptr(rcx, constMethod);
1.976 + __ load_unsigned_short(rcx, size_of_parameters);
1.977 +
1.978 + // native calls don't need the stack size check since they have no
1.979 + // expression stack and the arguments are already on the stack and
1.980 + // we only add a handful of words to the stack
1.981 +
1.982 + // rbx: Method*
1.983 + // rcx: size of parameters
1.984 + // r13: sender sp
1.985 + __ pop(rax); // get return address
1.986 +
1.987 + // for natives the size of locals is zero
1.988 +
1.989 + // compute beginning of parameters (r14)
1.990 + __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1.991 +
1.992 + // add 2 zero-initialized slots for native calls
1.993 + // initialize result_handler slot
1.994 + __ push((int) NULL_WORD);
1.995 + // slot for oop temp
1.996 + // (static native method holder mirror/jni oop result)
1.997 + __ push((int) NULL_WORD);
1.998 +
1.999 + // initialize fixed part of activation frame
1.1000 + generate_fixed_frame(true);
1.1001 +
1.1002 + // make sure method is native & not abstract
1.1003 +#ifdef ASSERT
1.1004 + __ movl(rax, access_flags);
1.1005 + {
1.1006 + Label L;
1.1007 + __ testl(rax, JVM_ACC_NATIVE);
1.1008 + __ jcc(Assembler::notZero, L);
1.1009 + __ stop("tried to execute non-native method as native");
1.1010 + __ bind(L);
1.1011 + }
1.1012 + {
1.1013 + Label L;
1.1014 + __ testl(rax, JVM_ACC_ABSTRACT);
1.1015 + __ jcc(Assembler::zero, L);
1.1016 + __ stop("tried to execute abstract method in interpreter");
1.1017 + __ bind(L);
1.1018 + }
1.1019 +#endif
1.1020 +
1.1021 + // Since at this point in the method invocation the exception handler
1.1022 + // would try to exit the monitor of synchronized methods which hasn't
1.1023 + // been entered yet, we set the thread local variable
1.1024 + // _do_not_unlock_if_synchronized to true. The remove_activation will
1.1025 + // check this flag.
1.1026 +
1.1027 + const Address do_not_unlock_if_synchronized(r15_thread,
1.1028 + in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1.1029 + __ movbool(do_not_unlock_if_synchronized, true);
1.1030 +
1.1031 + // increment invocation count & check for overflow
1.1032 + Label invocation_counter_overflow;
1.1033 + if (inc_counter) {
1.1034 + generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1.1035 + }
1.1036 +
1.1037 + Label continue_after_compile;
1.1038 + __ bind(continue_after_compile);
1.1039 +
1.1040 + bang_stack_shadow_pages(true);
1.1041 +
1.1042 + // reset the _do_not_unlock_if_synchronized flag
1.1043 + __ movbool(do_not_unlock_if_synchronized, false);
1.1044 +
1.1045 + // check for synchronized methods
1.1046 + // Must happen AFTER invocation_counter check and stack overflow check,
1.1047 + // so method is not locked if overflows.
1.1048 + if (synchronized) {
1.1049 + lock_method();
1.1050 + } else {
1.1051 + // no synchronization necessary
1.1052 +#ifdef ASSERT
1.1053 + {
1.1054 + Label L;
1.1055 + __ movl(rax, access_flags);
1.1056 + __ testl(rax, JVM_ACC_SYNCHRONIZED);
1.1057 + __ jcc(Assembler::zero, L);
1.1058 + __ stop("method needs synchronization");
1.1059 + __ bind(L);
1.1060 + }
1.1061 +#endif
1.1062 + }
1.1063 +
1.1064 + // start execution
1.1065 +#ifdef ASSERT
1.1066 + {
1.1067 + Label L;
1.1068 + const Address monitor_block_top(rbp,
1.1069 + frame::interpreter_frame_monitor_block_top_offset * wordSize);
1.1070 + __ movptr(rax, monitor_block_top);
1.1071 + __ cmpptr(rax, rsp);
1.1072 + __ jcc(Assembler::equal, L);
1.1073 + __ stop("broken stack frame setup in interpreter");
1.1074 + __ bind(L);
1.1075 + }
1.1076 +#endif
1.1077 +
1.1078 + // jvmti support
1.1079 + __ notify_method_entry();
1.1080 +
1.1081 + // work registers
1.1082 + const Register method = rbx;
1.1083 + const Register t = r11;
1.1084 +
1.1085 + // allocate space for parameters
1.1086 + __ get_method(method);
1.1087 + __ movptr(t, Address(method, Method::const_offset()));
1.1088 + __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1.1089 + __ shll(t, Interpreter::logStackElementSize);
1.1090 +
1.1091 + __ subptr(rsp, t);
1.1092 + __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1.1093 + __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
1.1094 +
1.1095 + // get signature handler
1.1096 + {
1.1097 + Label L;
1.1098 + __ movptr(t, Address(method, Method::signature_handler_offset()));
1.1099 + __ testptr(t, t);
1.1100 + __ jcc(Assembler::notZero, L);
1.1101 + __ call_VM(noreg,
1.1102 + CAST_FROM_FN_PTR(address,
1.1103 + InterpreterRuntime::prepare_native_call),
1.1104 + method);
1.1105 + __ get_method(method);
1.1106 + __ movptr(t, Address(method, Method::signature_handler_offset()));
1.1107 + __ bind(L);
1.1108 + }
1.1109 +
1.1110 + // call signature handler
1.1111 + assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
1.1112 + "adjust this code");
1.1113 + assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
1.1114 + "adjust this code");
1.1115 + assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1.1116 + "adjust this code");
1.1117 +
1.1118 + // The generated handlers do not touch RBX (the method oop).
1.1119 + // However, large signatures cannot be cached and are generated
1.1120 + // each time here. The slow-path generator can do a GC on return,
1.1121 + // so we must reload it after the call.
1.1122 + __ call(t);
1.1123 + __ get_method(method); // slow path can do a GC, reload RBX
1.1124 +
1.1125 +
1.1126 + // result handler is in rax
1.1127 + // set result handler
1.1128 + __ movptr(Address(rbp,
1.1129 + (frame::interpreter_frame_result_handler_offset) * wordSize),
1.1130 + rax);
1.1131 +
1.1132 + // pass mirror handle if static call
1.1133 + {
1.1134 + Label L;
1.1135 + const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1.1136 + __ movl(t, Address(method, Method::access_flags_offset()));
1.1137 + __ testl(t, JVM_ACC_STATIC);
1.1138 + __ jcc(Assembler::zero, L);
1.1139 + // get mirror
1.1140 + __ movptr(t, Address(method, Method::const_offset()));
1.1141 + __ movptr(t, Address(t, ConstMethod::constants_offset()));
1.1142 + __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1.1143 + __ movptr(t, Address(t, mirror_offset));
1.1144 + // copy mirror into activation frame
1.1145 + __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1.1146 + t);
1.1147 + // pass handle to mirror
1.1148 + __ lea(c_rarg1,
1.1149 + Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1.1150 + __ bind(L);
1.1151 + }
1.1152 +
1.1153 + // get native function entry point
1.1154 + {
1.1155 + Label L;
1.1156 + __ movptr(rax, Address(method, Method::native_function_offset()));
1.1157 + ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1.1158 + __ movptr(rscratch2, unsatisfied.addr());
1.1159 + __ cmpptr(rax, rscratch2);
1.1160 + __ jcc(Assembler::notEqual, L);
1.1161 + __ call_VM(noreg,
1.1162 + CAST_FROM_FN_PTR(address,
1.1163 + InterpreterRuntime::prepare_native_call),
1.1164 + method);
1.1165 + __ get_method(method);
1.1166 + __ movptr(rax, Address(method, Method::native_function_offset()));
1.1167 + __ bind(L);
1.1168 + }
1.1169 +
1.1170 + // pass JNIEnv
1.1171 + __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1.1172 +
1.1173 + // It is enough that the pc() points into the right code
1.1174 + // segment. It does not have to be the correct return pc.
1.1175 + __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1.1176 +
1.1177 + // change thread state
1.1178 +#ifdef ASSERT
1.1179 + {
1.1180 + Label L;
1.1181 + __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
1.1182 + __ cmpl(t, _thread_in_Java);
1.1183 + __ jcc(Assembler::equal, L);
1.1184 + __ stop("Wrong thread state in native stub");
1.1185 + __ bind(L);
1.1186 + }
1.1187 +#endif
1.1188 +
1.1189 + // Change state to native
1.1190 +
1.1191 + __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1.1192 + _thread_in_native);
1.1193 +
1.1194 + // Call the native method.
1.1195 + __ call(rax);
1.1196 + // result potentially in rax or xmm0
1.1197 +
1.1198 + // Verify or restore cpu control state after JNI call
1.1199 + __ restore_cpu_control_state_after_jni();
1.1200 +
1.1201 + // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1.1202 + // in order to extract the result of a method call. If the order of these
1.1203 + // pushes change or anything else is added to the stack then the code in
1.1204 + // interpreter_frame_result must also change.
1.1205 +
1.1206 + __ push(dtos);
1.1207 + __ push(ltos);
1.1208 +
1.1209 + // change thread state
1.1210 + __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1.1211 + _thread_in_native_trans);
1.1212 +
1.1213 + if (os::is_MP()) {
1.1214 + if (UseMembar) {
1.1215 + // Force this write out before the read below
1.1216 + __ membar(Assembler::Membar_mask_bits(
1.1217 + Assembler::LoadLoad | Assembler::LoadStore |
1.1218 + Assembler::StoreLoad | Assembler::StoreStore));
1.1219 + } else {
1.1220 + // Write serialization page so VM thread can do a pseudo remote membar.
1.1221 + // We use the current thread pointer to calculate a thread specific
1.1222 + // offset to write to within the page. This minimizes bus traffic
1.1223 + // due to cache line collision.
1.1224 + __ serialize_memory(r15_thread, rscratch2);
1.1225 + }
1.1226 + }
1.1227 +
1.1228 + // check for safepoint operation in progress and/or pending suspend requests
1.1229 + {
1.1230 + Label Continue;
1.1231 + __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1.1232 + SafepointSynchronize::_not_synchronized);
1.1233 +
1.1234 + Label L;
1.1235 + __ jcc(Assembler::notEqual, L);
1.1236 + __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1.1237 + __ jcc(Assembler::equal, Continue);
1.1238 + __ bind(L);
1.1239 +
1.1240 + // Don't use call_VM as it will see a possible pending exception
1.1241 + // and forward it and never return here preventing us from
1.1242 + // clearing _last_native_pc down below. Also can't use
1.1243 + // call_VM_leaf either as it will check to see if r13 & r14 are
1.1244 + // preserved and correspond to the bcp/locals pointers. So we do a
1.1245 + // runtime call by hand.
1.1246 + //
1.1247 + __ mov(c_rarg0, r15_thread);
1.1248 + __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1.1249 + __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1.1250 + __ andptr(rsp, -16); // align stack as required by ABI
1.1251 + __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1.1252 + __ mov(rsp, r12); // restore sp
1.1253 + __ reinit_heapbase();
1.1254 + __ bind(Continue);
1.1255 + }
1.1256 +
1.1257 + // change thread state
1.1258 + __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1.1259 +
1.1260 + // reset_last_Java_frame
1.1261 + __ reset_last_Java_frame(true, true);
1.1262 +
1.1263 + // reset handle block
1.1264 + __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1.1265 + __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1.1266 +
1.1267 + // If result is an oop unbox and store it in frame where gc will see it
1.1268 + // and result handler will pick it up
1.1269 +
1.1270 + {
1.1271 + Label no_oop, store_result;
1.1272 + __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1.1273 + __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1.1274 + __ jcc(Assembler::notEqual, no_oop);
1.1275 + // retrieve result
1.1276 + __ pop(ltos);
1.1277 + __ testptr(rax, rax);
1.1278 + __ jcc(Assembler::zero, store_result);
1.1279 + __ movptr(rax, Address(rax, 0));
1.1280 + __ bind(store_result);
1.1281 + __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1.1282 + // keep stack depth as expected by pushing oop which will eventually be discarde
1.1283 + __ push(ltos);
1.1284 + __ bind(no_oop);
1.1285 + }
1.1286 +
1.1287 +
1.1288 + {
1.1289 + Label no_reguard;
1.1290 + __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1.1291 + JavaThread::stack_guard_yellow_disabled);
1.1292 + __ jcc(Assembler::notEqual, no_reguard);
1.1293 +
1.1294 + __ pusha(); // XXX only save smashed registers
1.1295 + __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1.1296 + __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1.1297 + __ andptr(rsp, -16); // align stack as required by ABI
1.1298 + __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1.1299 + __ mov(rsp, r12); // restore sp
1.1300 + __ popa(); // XXX only restore smashed registers
1.1301 + __ reinit_heapbase();
1.1302 +
1.1303 + __ bind(no_reguard);
1.1304 + }
1.1305 +
1.1306 +
1.1307 + // The method register is junk from after the thread_in_native transition
1.1308 + // until here. Also can't call_VM until the bcp has been
1.1309 + // restored. Need bcp for throwing exception below so get it now.
1.1310 + __ get_method(method);
1.1311 +
1.1312 + // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1.1313 + // r13 == code_base()
1.1314 + __ movptr(r13, Address(method, Method::const_offset())); // get ConstMethod*
1.1315 + __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase
1.1316 + // handle exceptions (exception handling will handle unlocking!)
1.1317 + {
1.1318 + Label L;
1.1319 + __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1.1320 + __ jcc(Assembler::zero, L);
1.1321 + // Note: At some point we may want to unify this with the code
1.1322 + // used in call_VM_base(); i.e., we should use the
1.1323 + // StubRoutines::forward_exception code. For now this doesn't work
1.1324 + // here because the rsp is not correctly set at this point.
1.1325 + __ MacroAssembler::call_VM(noreg,
1.1326 + CAST_FROM_FN_PTR(address,
1.1327 + InterpreterRuntime::throw_pending_exception));
1.1328 + __ should_not_reach_here();
1.1329 + __ bind(L);
1.1330 + }
1.1331 +
1.1332 + // do unlocking if necessary
1.1333 + {
1.1334 + Label L;
1.1335 + __ movl(t, Address(method, Method::access_flags_offset()));
1.1336 + __ testl(t, JVM_ACC_SYNCHRONIZED);
1.1337 + __ jcc(Assembler::zero, L);
1.1338 + // the code below should be shared with interpreter macro
1.1339 + // assembler implementation
1.1340 + {
1.1341 + Label unlock;
1.1342 + // BasicObjectLock will be first in list, since this is a
1.1343 + // synchronized method. However, need to check that the object
1.1344 + // has not been unlocked by an explicit monitorexit bytecode.
1.1345 + const Address monitor(rbp,
1.1346 + (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1.1347 + wordSize - sizeof(BasicObjectLock)));
1.1348 +
1.1349 + // monitor expect in c_rarg1 for slow unlock path
1.1350 + __ lea(c_rarg1, monitor); // address of first monitor
1.1351 +
1.1352 + __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1.1353 + __ testptr(t, t);
1.1354 + __ jcc(Assembler::notZero, unlock);
1.1355 +
1.1356 + // Entry already unlocked, need to throw exception
1.1357 + __ MacroAssembler::call_VM(noreg,
1.1358 + CAST_FROM_FN_PTR(address,
1.1359 + InterpreterRuntime::throw_illegal_monitor_state_exception));
1.1360 + __ should_not_reach_here();
1.1361 +
1.1362 + __ bind(unlock);
1.1363 + __ unlock_object(c_rarg1);
1.1364 + }
1.1365 + __ bind(L);
1.1366 + }
1.1367 +
1.1368 + // jvmti support
1.1369 + // Note: This must happen _after_ handling/throwing any exceptions since
1.1370 + // the exception handler code notifies the runtime of method exits
1.1371 + // too. If this happens before, method entry/exit notifications are
1.1372 + // not properly paired (was bug - gri 11/22/99).
1.1373 + __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1.1374 +
1.1375 + // restore potential result in edx:eax, call result handler to
1.1376 + // restore potential result in ST0 & handle result
1.1377 +
1.1378 + __ pop(ltos);
1.1379 + __ pop(dtos);
1.1380 +
1.1381 + __ movptr(t, Address(rbp,
1.1382 + (frame::interpreter_frame_result_handler_offset) * wordSize));
1.1383 + __ call(t);
1.1384 +
1.1385 + // remove activation
1.1386 + __ movptr(t, Address(rbp,
1.1387 + frame::interpreter_frame_sender_sp_offset *
1.1388 + wordSize)); // get sender sp
1.1389 + __ leave(); // remove frame anchor
1.1390 + __ pop(rdi); // get return address
1.1391 + __ mov(rsp, t); // set sp to sender sp
1.1392 + __ jmp(rdi);
1.1393 +
1.1394 + if (inc_counter) {
1.1395 + // Handle overflow of counter and compile method
1.1396 + __ bind(invocation_counter_overflow);
1.1397 + generate_counter_overflow(&continue_after_compile);
1.1398 + }
1.1399 +
1.1400 + return entry_point;
1.1401 +}
1.1402 +
1.1403 +//
1.1404 +// Generic interpreted method entry to (asm) interpreter
1.1405 +//
1.1406 +address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1.1407 + // determine code generation flags
1.1408 + bool inc_counter = UseCompiler || CountCompiledCalls;
1.1409 +
1.1410 + // ebx: Method*
1.1411 + // r13: sender sp
1.1412 + address entry_point = __ pc();
1.1413 +
1.1414 + const Address constMethod(rbx, Method::const_offset());
1.1415 + const Address access_flags(rbx, Method::access_flags_offset());
1.1416 + const Address size_of_parameters(rdx,
1.1417 + ConstMethod::size_of_parameters_offset());
1.1418 + const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1.1419 +
1.1420 +
1.1421 + // get parameter size (always needed)
1.1422 + __ movptr(rdx, constMethod);
1.1423 + __ load_unsigned_short(rcx, size_of_parameters);
1.1424 +
1.1425 + // rbx: Method*
1.1426 + // rcx: size of parameters
1.1427 + // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1.1428 +
1.1429 + __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1.1430 + __ subl(rdx, rcx); // rdx = no. of additional locals
1.1431 +
1.1432 + // YYY
1.1433 +// __ incrementl(rdx);
1.1434 +// __ andl(rdx, -2);
1.1435 +
1.1436 + // see if we've got enough room on the stack for locals plus overhead.
1.1437 + generate_stack_overflow_check();
1.1438 +
1.1439 + // get return address
1.1440 + __ pop(rax);
1.1441 +
1.1442 + // compute beginning of parameters (r14)
1.1443 + __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1.1444 +
1.1445 + // rdx - # of additional locals
1.1446 + // allocate space for locals
1.1447 + // explicitly initialize locals
1.1448 + {
1.1449 + Label exit, loop;
1.1450 + __ testl(rdx, rdx);
1.1451 + __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1.1452 + __ bind(loop);
1.1453 + __ push((int) NULL_WORD); // initialize local variables
1.1454 + __ decrementl(rdx); // until everything initialized
1.1455 + __ jcc(Assembler::greater, loop);
1.1456 + __ bind(exit);
1.1457 + }
1.1458 +
1.1459 + // initialize fixed part of activation frame
1.1460 + generate_fixed_frame(false);
1.1461 +
1.1462 + // make sure method is not native & not abstract
1.1463 +#ifdef ASSERT
1.1464 + __ movl(rax, access_flags);
1.1465 + {
1.1466 + Label L;
1.1467 + __ testl(rax, JVM_ACC_NATIVE);
1.1468 + __ jcc(Assembler::zero, L);
1.1469 + __ stop("tried to execute native method as non-native");
1.1470 + __ bind(L);
1.1471 + }
1.1472 + {
1.1473 + Label L;
1.1474 + __ testl(rax, JVM_ACC_ABSTRACT);
1.1475 + __ jcc(Assembler::zero, L);
1.1476 + __ stop("tried to execute abstract method in interpreter");
1.1477 + __ bind(L);
1.1478 + }
1.1479 +#endif
1.1480 +
1.1481 + // Since at this point in the method invocation the exception
1.1482 + // handler would try to exit the monitor of synchronized methods
1.1483 + // which hasn't been entered yet, we set the thread local variable
1.1484 + // _do_not_unlock_if_synchronized to true. The remove_activation
1.1485 + // will check this flag.
1.1486 +
1.1487 + const Address do_not_unlock_if_synchronized(r15_thread,
1.1488 + in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1.1489 + __ movbool(do_not_unlock_if_synchronized, true);
1.1490 +
1.1491 + __ profile_parameters_type(rax, rcx, rdx);
1.1492 + // increment invocation count & check for overflow
1.1493 + Label invocation_counter_overflow;
1.1494 + Label profile_method;
1.1495 + Label profile_method_continue;
1.1496 + if (inc_counter) {
1.1497 + generate_counter_incr(&invocation_counter_overflow,
1.1498 + &profile_method,
1.1499 + &profile_method_continue);
1.1500 + if (ProfileInterpreter) {
1.1501 + __ bind(profile_method_continue);
1.1502 + }
1.1503 + }
1.1504 +
1.1505 + Label continue_after_compile;
1.1506 + __ bind(continue_after_compile);
1.1507 +
1.1508 + // check for synchronized interpreted methods
1.1509 + bang_stack_shadow_pages(false);
1.1510 +
1.1511 + // reset the _do_not_unlock_if_synchronized flag
1.1512 + __ movbool(do_not_unlock_if_synchronized, false);
1.1513 +
1.1514 + // check for synchronized methods
1.1515 + // Must happen AFTER invocation_counter check and stack overflow check,
1.1516 + // so method is not locked if overflows.
1.1517 + if (synchronized) {
1.1518 + // Allocate monitor and lock method
1.1519 + lock_method();
1.1520 + } else {
1.1521 + // no synchronization necessary
1.1522 +#ifdef ASSERT
1.1523 + {
1.1524 + Label L;
1.1525 + __ movl(rax, access_flags);
1.1526 + __ testl(rax, JVM_ACC_SYNCHRONIZED);
1.1527 + __ jcc(Assembler::zero, L);
1.1528 + __ stop("method needs synchronization");
1.1529 + __ bind(L);
1.1530 + }
1.1531 +#endif
1.1532 + }
1.1533 +
1.1534 + // start execution
1.1535 +#ifdef ASSERT
1.1536 + {
1.1537 + Label L;
1.1538 + const Address monitor_block_top (rbp,
1.1539 + frame::interpreter_frame_monitor_block_top_offset * wordSize);
1.1540 + __ movptr(rax, monitor_block_top);
1.1541 + __ cmpptr(rax, rsp);
1.1542 + __ jcc(Assembler::equal, L);
1.1543 + __ stop("broken stack frame setup in interpreter");
1.1544 + __ bind(L);
1.1545 + }
1.1546 +#endif
1.1547 +
1.1548 + // jvmti support
1.1549 + __ notify_method_entry();
1.1550 +
1.1551 + __ dispatch_next(vtos);
1.1552 +
1.1553 + // invocation counter overflow
1.1554 + if (inc_counter) {
1.1555 + if (ProfileInterpreter) {
1.1556 + // We have decided to profile this method in the interpreter
1.1557 + __ bind(profile_method);
1.1558 + __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1.1559 + __ set_method_data_pointer_for_bcp();
1.1560 + __ get_method(rbx);
1.1561 + __ jmp(profile_method_continue);
1.1562 + }
1.1563 + // Handle overflow of counter and compile method
1.1564 + __ bind(invocation_counter_overflow);
1.1565 + generate_counter_overflow(&continue_after_compile);
1.1566 + }
1.1567 +
1.1568 + return entry_point;
1.1569 +}
1.1570 +
1.1571 +// Entry points
1.1572 +//
1.1573 +// Here we generate the various kind of entries into the interpreter.
1.1574 +// The two main entry type are generic bytecode methods and native
1.1575 +// call method. These both come in synchronized and non-synchronized
1.1576 +// versions but the frame layout they create is very similar. The
1.1577 +// other method entry types are really just special purpose entries
1.1578 +// that are really entry and interpretation all in one. These are for
1.1579 +// trivial methods like accessor, empty, or special math methods.
1.1580 +//
1.1581 +// When control flow reaches any of the entry types for the interpreter
1.1582 +// the following holds ->
1.1583 +//
1.1584 +// Arguments:
1.1585 +//
1.1586 +// rbx: Method*
1.1587 +//
1.1588 +// Stack layout immediately at entry
1.1589 +//
1.1590 +// [ return address ] <--- rsp
1.1591 +// [ parameter n ]
1.1592 +// ...
1.1593 +// [ parameter 1 ]
1.1594 +// [ expression stack ] (caller's java expression stack)
1.1595 +
1.1596 +// Assuming that we don't go to one of the trivial specialized entries
1.1597 +// the stack will look like below when we are ready to execute the
1.1598 +// first bytecode (or call the native routine). The register usage
1.1599 +// will be as the template based interpreter expects (see
1.1600 +// interpreter_amd64.hpp).
1.1601 +//
1.1602 +// local variables follow incoming parameters immediately; i.e.
1.1603 +// the return address is moved to the end of the locals).
1.1604 +//
1.1605 +// [ monitor entry ] <--- rsp
1.1606 +// ...
1.1607 +// [ monitor entry ]
1.1608 +// [ expr. stack bottom ]
1.1609 +// [ saved r13 ]
1.1610 +// [ current r14 ]
1.1611 +// [ Method* ]
1.1612 +// [ saved ebp ] <--- rbp
1.1613 +// [ return address ]
1.1614 +// [ local variable m ]
1.1615 +// ...
1.1616 +// [ local variable 1 ]
1.1617 +// [ parameter n ]
1.1618 +// ...
1.1619 +// [ parameter 1 ] <--- r14
1.1620 +
1.1621 +address AbstractInterpreterGenerator::generate_method_entry(
1.1622 + AbstractInterpreter::MethodKind kind) {
1.1623 + // determine code generation flags
1.1624 + bool synchronized = false;
1.1625 + address entry_point = NULL;
1.1626 + InterpreterGenerator* ig_this = (InterpreterGenerator*)this;
1.1627 +
1.1628 + switch (kind) {
1.1629 + case Interpreter::zerolocals : break;
1.1630 + case Interpreter::zerolocals_synchronized: synchronized = true; break;
1.1631 + case Interpreter::native : entry_point = ig_this->generate_native_entry(false); break;
1.1632 + case Interpreter::native_synchronized : entry_point = ig_this->generate_native_entry(true); break;
1.1633 + case Interpreter::empty : entry_point = ig_this->generate_empty_entry(); break;
1.1634 + case Interpreter::accessor : entry_point = ig_this->generate_accessor_entry(); break;
1.1635 + case Interpreter::abstract : entry_point = ig_this->generate_abstract_entry(); break;
1.1636 +
1.1637 + case Interpreter::java_lang_math_sin : // fall thru
1.1638 + case Interpreter::java_lang_math_cos : // fall thru
1.1639 + case Interpreter::java_lang_math_tan : // fall thru
1.1640 + case Interpreter::java_lang_math_abs : // fall thru
1.1641 + case Interpreter::java_lang_math_log : // fall thru
1.1642 + case Interpreter::java_lang_math_log10 : // fall thru
1.1643 + case Interpreter::java_lang_math_sqrt : // fall thru
1.1644 + case Interpreter::java_lang_math_pow : // fall thru
1.1645 + case Interpreter::java_lang_math_exp : entry_point = ig_this->generate_math_entry(kind); break;
1.1646 + case Interpreter::java_lang_ref_reference_get
1.1647 + : entry_point = ig_this->generate_Reference_get_entry(); break;
1.1648 + case Interpreter::java_util_zip_CRC32_update
1.1649 + : entry_point = ig_this->generate_CRC32_update_entry(); break;
1.1650 + case Interpreter::java_util_zip_CRC32_updateBytes
1.1651 + : // fall thru
1.1652 + case Interpreter::java_util_zip_CRC32_updateByteBuffer
1.1653 + : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break;
1.1654 + default:
1.1655 + fatal(err_msg("unexpected method kind: %d", kind));
1.1656 + break;
1.1657 + }
1.1658 +
1.1659 + if (entry_point) {
1.1660 + return entry_point;
1.1661 + }
1.1662 +
1.1663 + return ig_this->generate_normal_entry(synchronized);
1.1664 +}
1.1665 +
1.1666 +// These should never be compiled since the interpreter will prefer
1.1667 +// the compiled version to the intrinsic version.
1.1668 +bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1.1669 + switch (method_kind(m)) {
1.1670 + case Interpreter::java_lang_math_sin : // fall thru
1.1671 + case Interpreter::java_lang_math_cos : // fall thru
1.1672 + case Interpreter::java_lang_math_tan : // fall thru
1.1673 + case Interpreter::java_lang_math_abs : // fall thru
1.1674 + case Interpreter::java_lang_math_log : // fall thru
1.1675 + case Interpreter::java_lang_math_log10 : // fall thru
1.1676 + case Interpreter::java_lang_math_sqrt : // fall thru
1.1677 + case Interpreter::java_lang_math_pow : // fall thru
1.1678 + case Interpreter::java_lang_math_exp :
1.1679 + return false;
1.1680 + default:
1.1681 + return true;
1.1682 + }
1.1683 +}
1.1684 +
1.1685 +// How much stack a method activation needs in words.
1.1686 +int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1.1687 + const int entry_size = frame::interpreter_frame_monitor_size();
1.1688 +
1.1689 + // total overhead size: entry_size + (saved rbp thru expr stack
1.1690 + // bottom). be sure to change this if you add/subtract anything
1.1691 + // to/from the overhead area
1.1692 + const int overhead_size =
1.1693 + -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1.1694 +
1.1695 + const int stub_code = frame::entry_frame_after_call_words;
1.1696 + const int method_stack = (method->max_locals() + method->max_stack()) *
1.1697 + Interpreter::stackElementWords;
1.1698 + return (overhead_size + method_stack + stub_code);
1.1699 +}
1.1700 +
1.1701 +//-----------------------------------------------------------------------------
1.1702 +// Exceptions
1.1703 +
1.1704 +void TemplateInterpreterGenerator::generate_throw_exception() {
1.1705 + // Entry point in previous activation (i.e., if the caller was
1.1706 + // interpreted)
1.1707 + Interpreter::_rethrow_exception_entry = __ pc();
1.1708 + // Restore sp to interpreter_frame_last_sp even though we are going
1.1709 + // to empty the expression stack for the exception processing.
1.1710 + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1.1711 + // rax: exception
1.1712 + // rdx: return address/pc that threw exception
1.1713 + __ restore_bcp(); // r13 points to call/send
1.1714 + __ restore_locals();
1.1715 + __ reinit_heapbase(); // restore r12 as heapbase.
1.1716 + // Entry point for exceptions thrown within interpreter code
1.1717 + Interpreter::_throw_exception_entry = __ pc();
1.1718 + // expression stack is undefined here
1.1719 + // rax: exception
1.1720 + // r13: exception bcp
1.1721 + __ verify_oop(rax);
1.1722 + __ mov(c_rarg1, rax);
1.1723 +
1.1724 + // expression stack must be empty before entering the VM in case of
1.1725 + // an exception
1.1726 + __ empty_expression_stack();
1.1727 + // find exception handler address and preserve exception oop
1.1728 + __ call_VM(rdx,
1.1729 + CAST_FROM_FN_PTR(address,
1.1730 + InterpreterRuntime::exception_handler_for_exception),
1.1731 + c_rarg1);
1.1732 + // rax: exception handler entry point
1.1733 + // rdx: preserved exception oop
1.1734 + // r13: bcp for exception handler
1.1735 + __ push_ptr(rdx); // push exception which is now the only value on the stack
1.1736 + __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1.1737 +
1.1738 + // If the exception is not handled in the current frame the frame is
1.1739 + // removed and the exception is rethrown (i.e. exception
1.1740 + // continuation is _rethrow_exception).
1.1741 + //
1.1742 + // Note: At this point the bci is still the bxi for the instruction
1.1743 + // which caused the exception and the expression stack is
1.1744 + // empty. Thus, for any VM calls at this point, GC will find a legal
1.1745 + // oop map (with empty expression stack).
1.1746 +
1.1747 + // In current activation
1.1748 + // tos: exception
1.1749 + // esi: exception bcp
1.1750 +
1.1751 + //
1.1752 + // JVMTI PopFrame support
1.1753 + //
1.1754 +
1.1755 + Interpreter::_remove_activation_preserving_args_entry = __ pc();
1.1756 + __ empty_expression_stack();
1.1757 + // Set the popframe_processing bit in pending_popframe_condition
1.1758 + // indicating that we are currently handling popframe, so that
1.1759 + // call_VMs that may happen later do not trigger new popframe
1.1760 + // handling cycles.
1.1761 + __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1.1762 + __ orl(rdx, JavaThread::popframe_processing_bit);
1.1763 + __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1.1764 +
1.1765 + {
1.1766 + // Check to see whether we are returning to a deoptimized frame.
1.1767 + // (The PopFrame call ensures that the caller of the popped frame is
1.1768 + // either interpreted or compiled and deoptimizes it if compiled.)
1.1769 + // In this case, we can't call dispatch_next() after the frame is
1.1770 + // popped, but instead must save the incoming arguments and restore
1.1771 + // them after deoptimization has occurred.
1.1772 + //
1.1773 + // Note that we don't compare the return PC against the
1.1774 + // deoptimization blob's unpack entry because of the presence of
1.1775 + // adapter frames in C2.
1.1776 + Label caller_not_deoptimized;
1.1777 + __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1.1778 + __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1.1779 + InterpreterRuntime::interpreter_contains), c_rarg1);
1.1780 + __ testl(rax, rax);
1.1781 + __ jcc(Assembler::notZero, caller_not_deoptimized);
1.1782 +
1.1783 + // Compute size of arguments for saving when returning to
1.1784 + // deoptimized caller
1.1785 + __ get_method(rax);
1.1786 + __ movptr(rax, Address(rax, Method::const_offset()));
1.1787 + __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1.1788 + size_of_parameters_offset())));
1.1789 + __ shll(rax, Interpreter::logStackElementSize);
1.1790 + __ restore_locals(); // XXX do we need this?
1.1791 + __ subptr(r14, rax);
1.1792 + __ addptr(r14, wordSize);
1.1793 + // Save these arguments
1.1794 + __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1.1795 + Deoptimization::
1.1796 + popframe_preserve_args),
1.1797 + r15_thread, rax, r14);
1.1798 +
1.1799 + __ remove_activation(vtos, rdx,
1.1800 + /* throw_monitor_exception */ false,
1.1801 + /* install_monitor_exception */ false,
1.1802 + /* notify_jvmdi */ false);
1.1803 +
1.1804 + // Inform deoptimization that it is responsible for restoring
1.1805 + // these arguments
1.1806 + __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1.1807 + JavaThread::popframe_force_deopt_reexecution_bit);
1.1808 +
1.1809 + // Continue in deoptimization handler
1.1810 + __ jmp(rdx);
1.1811 +
1.1812 + __ bind(caller_not_deoptimized);
1.1813 + }
1.1814 +
1.1815 + __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1.1816 + /* throw_monitor_exception */ false,
1.1817 + /* install_monitor_exception */ false,
1.1818 + /* notify_jvmdi */ false);
1.1819 +
1.1820 + // Finish with popframe handling
1.1821 + // A previous I2C followed by a deoptimization might have moved the
1.1822 + // outgoing arguments further up the stack. PopFrame expects the
1.1823 + // mutations to those outgoing arguments to be preserved and other
1.1824 + // constraints basically require this frame to look exactly as
1.1825 + // though it had previously invoked an interpreted activation with
1.1826 + // no space between the top of the expression stack (current
1.1827 + // last_sp) and the top of stack. Rather than force deopt to
1.1828 + // maintain this kind of invariant all the time we call a small
1.1829 + // fixup routine to move the mutated arguments onto the top of our
1.1830 + // expression stack if necessary.
1.1831 + __ mov(c_rarg1, rsp);
1.1832 + __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1.1833 + // PC must point into interpreter here
1.1834 + __ set_last_Java_frame(noreg, rbp, __ pc());
1.1835 + __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1.1836 + __ reset_last_Java_frame(true, true);
1.1837 + // Restore the last_sp and null it out
1.1838 + __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1.1839 + __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1.1840 +
1.1841 + __ restore_bcp(); // XXX do we need this?
1.1842 + __ restore_locals(); // XXX do we need this?
1.1843 + // The method data pointer was incremented already during
1.1844 + // call profiling. We have to restore the mdp for the current bcp.
1.1845 + if (ProfileInterpreter) {
1.1846 + __ set_method_data_pointer_for_bcp();
1.1847 + }
1.1848 +
1.1849 + // Clear the popframe condition flag
1.1850 + __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1.1851 + JavaThread::popframe_inactive);
1.1852 +
1.1853 +#if INCLUDE_JVMTI
1.1854 + if (EnableInvokeDynamic) {
1.1855 + Label L_done;
1.1856 + const Register local0 = r14;
1.1857 +
1.1858 + __ cmpb(Address(r13, 0), Bytecodes::_invokestatic);
1.1859 + __ jcc(Assembler::notEqual, L_done);
1.1860 +
1.1861 + // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1.1862 + // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1.1863 +
1.1864 + __ get_method(rdx);
1.1865 + __ movptr(rax, Address(local0, 0));
1.1866 + __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, r13);
1.1867 +
1.1868 + __ testptr(rax, rax);
1.1869 + __ jcc(Assembler::zero, L_done);
1.1870 +
1.1871 + __ movptr(Address(rbx, 0), rax);
1.1872 + __ bind(L_done);
1.1873 + }
1.1874 +#endif // INCLUDE_JVMTI
1.1875 +
1.1876 + __ dispatch_next(vtos);
1.1877 + // end of PopFrame support
1.1878 +
1.1879 + Interpreter::_remove_activation_entry = __ pc();
1.1880 +
1.1881 + // preserve exception over this code sequence
1.1882 + __ pop_ptr(rax);
1.1883 + __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1.1884 + // remove the activation (without doing throws on illegalMonitorExceptions)
1.1885 + __ remove_activation(vtos, rdx, false, true, false);
1.1886 + // restore exception
1.1887 + __ get_vm_result(rax, r15_thread);
1.1888 +
1.1889 + // In between activations - previous activation type unknown yet
1.1890 + // compute continuation point - the continuation point expects the
1.1891 + // following registers set up:
1.1892 + //
1.1893 + // rax: exception
1.1894 + // rdx: return address/pc that threw exception
1.1895 + // rsp: expression stack of caller
1.1896 + // rbp: ebp of caller
1.1897 + __ push(rax); // save exception
1.1898 + __ push(rdx); // save return address
1.1899 + __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1.1900 + SharedRuntime::exception_handler_for_return_address),
1.1901 + r15_thread, rdx);
1.1902 + __ mov(rbx, rax); // save exception handler
1.1903 + __ pop(rdx); // restore return address
1.1904 + __ pop(rax); // restore exception
1.1905 + // Note that an "issuing PC" is actually the next PC after the call
1.1906 + __ jmp(rbx); // jump to exception
1.1907 + // handler of caller
1.1908 +}
1.1909 +
1.1910 +
1.1911 +//
1.1912 +// JVMTI ForceEarlyReturn support
1.1913 +//
1.1914 +address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1.1915 + address entry = __ pc();
1.1916 +
1.1917 + __ restore_bcp();
1.1918 + __ restore_locals();
1.1919 + __ empty_expression_stack();
1.1920 + __ load_earlyret_value(state);
1.1921 +
1.1922 + __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1.1923 + Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1.1924 +
1.1925 + // Clear the earlyret state
1.1926 + __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1.1927 +
1.1928 + __ remove_activation(state, rsi,
1.1929 + false, /* throw_monitor_exception */
1.1930 + false, /* install_monitor_exception */
1.1931 + true); /* notify_jvmdi */
1.1932 + __ jmp(rsi);
1.1933 +
1.1934 + return entry;
1.1935 +} // end of ForceEarlyReturn support
1.1936 +
1.1937 +
1.1938 +//-----------------------------------------------------------------------------
1.1939 +// Helper for vtos entry point generation
1.1940 +
1.1941 +void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1.1942 + address& bep,
1.1943 + address& cep,
1.1944 + address& sep,
1.1945 + address& aep,
1.1946 + address& iep,
1.1947 + address& lep,
1.1948 + address& fep,
1.1949 + address& dep,
1.1950 + address& vep) {
1.1951 + assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1.1952 + Label L;
1.1953 + aep = __ pc(); __ push_ptr(); __ jmp(L);
1.1954 + fep = __ pc(); __ push_f(); __ jmp(L);
1.1955 + dep = __ pc(); __ push_d(); __ jmp(L);
1.1956 + lep = __ pc(); __ push_l(); __ jmp(L);
1.1957 + bep = cep = sep =
1.1958 + iep = __ pc(); __ push_i();
1.1959 + vep = __ pc();
1.1960 + __ bind(L);
1.1961 + generate_and_dispatch(t);
1.1962 +}
1.1963 +
1.1964 +
1.1965 +//-----------------------------------------------------------------------------
1.1966 +// Generation of individual instructions
1.1967 +
1.1968 +// helpers for generate_and_dispatch
1.1969 +
1.1970 +
1.1971 +InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1.1972 + : TemplateInterpreterGenerator(code) {
1.1973 + generate_all(); // down here so it can be "virtual"
1.1974 +}
1.1975 +
1.1976 +//-----------------------------------------------------------------------------
1.1977 +
1.1978 +// Non-product code
1.1979 +#ifndef PRODUCT
1.1980 +address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1.1981 + address entry = __ pc();
1.1982 +
1.1983 + __ push(state);
1.1984 + __ push(c_rarg0);
1.1985 + __ push(c_rarg1);
1.1986 + __ push(c_rarg2);
1.1987 + __ push(c_rarg3);
1.1988 + __ mov(c_rarg2, rax); // Pass itos
1.1989 +#ifdef _WIN64
1.1990 + __ movflt(xmm3, xmm0); // Pass ftos
1.1991 +#endif
1.1992 + __ call_VM(noreg,
1.1993 + CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1.1994 + c_rarg1, c_rarg2, c_rarg3);
1.1995 + __ pop(c_rarg3);
1.1996 + __ pop(c_rarg2);
1.1997 + __ pop(c_rarg1);
1.1998 + __ pop(c_rarg0);
1.1999 + __ pop(state);
1.2000 + __ ret(0); // return from result handler
1.2001 +
1.2002 + return entry;
1.2003 +}
1.2004 +
1.2005 +void TemplateInterpreterGenerator::count_bytecode() {
1.2006 + __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1.2007 +}
1.2008 +
1.2009 +void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1.2010 + __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1.2011 +}
1.2012 +
1.2013 +void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1.2014 + __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1.2015 + __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1.2016 + __ orl(rbx,
1.2017 + ((int) t->bytecode()) <<
1.2018 + BytecodePairHistogram::log2_number_of_codes);
1.2019 + __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1.2020 + __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1.2021 + __ incrementl(Address(rscratch1, rbx, Address::times_4));
1.2022 +}
1.2023 +
1.2024 +
1.2025 +void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1.2026 + // Call a little run-time stub to avoid blow-up for each bytecode.
1.2027 + // The run-time runtime saves the right registers, depending on
1.2028 + // the tosca in-state for the given template.
1.2029 +
1.2030 + assert(Interpreter::trace_code(t->tos_in()) != NULL,
1.2031 + "entry must have been generated");
1.2032 + __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1.2033 + __ andptr(rsp, -16); // align stack as required by ABI
1.2034 + __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1.2035 + __ mov(rsp, r12); // restore sp
1.2036 + __ reinit_heapbase();
1.2037 +}
1.2038 +
1.2039 +
1.2040 +void TemplateInterpreterGenerator::stop_interpreter_at() {
1.2041 + Label L;
1.2042 + __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1.2043 + StopInterpreterAt);
1.2044 + __ jcc(Assembler::notEqual, L);
1.2045 + __ int3();
1.2046 + __ bind(L);
1.2047 +}
1.2048 +#endif // !PRODUCT
1.2049 +#endif // ! CC_INTERP
