Thu, 15 Aug 2013 20:04:10 -0400
8003424: Enable Class Data Sharing for CompressedOops
8016729: ObjectAlignmentInBytes=16 now forces the use of heap based compressed oops
8005933: The -Xshare:auto option is ignored for -server
Summary: Move klass metaspace above the heap and support CDS with compressed klass ptrs.
Reviewed-by: coleenp, kvn, mgerdin, tschatzl, stefank
1 /*
2 * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/bytecodeHistogram.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterGenerator.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "oops/arrayOop.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/jvmtiExport.hpp"
37 #include "prims/jvmtiThreadState.hpp"
38 #include "runtime/arguments.hpp"
39 #include "runtime/deoptimization.hpp"
40 #include "runtime/frame.inline.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/stubRoutines.hpp"
43 #include "runtime/synchronizer.hpp"
44 #include "runtime/timer.hpp"
45 #include "runtime/vframeArray.hpp"
46 #include "utilities/debug.hpp"
47 #include "utilities/macros.hpp"
49 #define __ _masm->
52 #ifndef CC_INTERP
53 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
54 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
55 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
57 //------------------------------------------------------------------------------------------------------------------------
59 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
60 address entry = __ pc();
62 // Note: There should be a minimal interpreter frame set up when stack
63 // overflow occurs since we check explicitly for it now.
64 //
65 #ifdef ASSERT
66 { Label L;
67 __ lea(rax, Address(rbp,
68 frame::interpreter_frame_monitor_block_top_offset * wordSize));
69 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp,
70 // (stack grows negative)
71 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
72 __ stop ("interpreter frame not set up");
73 __ bind(L);
74 }
75 #endif // ASSERT
76 // Restore bcp under the assumption that the current frame is still
77 // interpreted
78 __ restore_bcp();
80 // expression stack must be empty before entering the VM if an exception
81 // happened
82 __ empty_expression_stack();
83 __ empty_FPU_stack();
84 // throw exception
85 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
86 return entry;
87 }
89 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
90 address entry = __ pc();
91 // expression stack must be empty before entering the VM if an exception happened
92 __ empty_expression_stack();
93 __ empty_FPU_stack();
94 // setup parameters
95 // ??? convention: expect aberrant index in register rbx,
96 __ lea(rax, ExternalAddress((address)name));
97 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
98 return entry;
99 }
101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
102 address entry = __ pc();
103 // object is at TOS
104 __ pop(rax);
105 // expression stack must be empty before entering the VM if an exception
106 // happened
107 __ empty_expression_stack();
108 __ empty_FPU_stack();
109 __ call_VM(noreg,
110 CAST_FROM_FN_PTR(address,
111 InterpreterRuntime::throw_ClassCastException),
112 rax);
113 return entry;
114 }
116 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
117 assert(!pass_oop || message == NULL, "either oop or message but not both");
118 address entry = __ pc();
119 if (pass_oop) {
120 // object is at TOS
121 __ pop(rbx);
122 }
123 // expression stack must be empty before entering the VM if an exception happened
124 __ empty_expression_stack();
125 __ empty_FPU_stack();
126 // setup parameters
127 __ lea(rax, ExternalAddress((address)name));
128 if (pass_oop) {
129 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
130 } else {
131 if (message != NULL) {
132 __ lea(rbx, ExternalAddress((address)message));
133 } else {
134 __ movptr(rbx, NULL_WORD);
135 }
136 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
137 }
138 // throw exception
139 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
140 return entry;
141 }
144 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
145 address entry = __ pc();
146 // NULL last_sp until next java call
147 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
148 __ dispatch_next(state);
149 return entry;
150 }
153 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
154 TosState incoming_state = state;
155 address entry = __ pc();
157 #ifdef COMPILER2
158 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
159 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
160 for (int i = 1; i < 8; i++) {
161 __ ffree(i);
162 }
163 } else if (UseSSE < 2) {
164 __ empty_FPU_stack();
165 }
166 #endif
167 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
168 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
169 } else {
170 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
171 }
173 // In SSE mode, interpreter returns FP results in xmm0 but they need
174 // to end up back on the FPU so it can operate on them.
175 if (incoming_state == ftos && UseSSE >= 1) {
176 __ subptr(rsp, wordSize);
177 __ movflt(Address(rsp, 0), xmm0);
178 __ fld_s(Address(rsp, 0));
179 __ addptr(rsp, wordSize);
180 } else if (incoming_state == dtos && UseSSE >= 2) {
181 __ subptr(rsp, 2*wordSize);
182 __ movdbl(Address(rsp, 0), xmm0);
183 __ fld_d(Address(rsp, 0));
184 __ addptr(rsp, 2*wordSize);
185 }
187 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
189 // Restore stack bottom in case i2c adjusted stack
190 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
191 // and NULL it as marker that rsp is now tos until next java call
192 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
194 __ restore_bcp();
195 __ restore_locals();
197 Label L_got_cache, L_giant_index;
198 if (EnableInvokeDynamic) {
199 __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
200 __ jcc(Assembler::equal, L_giant_index);
201 }
202 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
203 __ bind(L_got_cache);
204 __ movl(rbx, Address(rbx, rcx,
205 Address::times_ptr, ConstantPoolCache::base_offset() +
206 ConstantPoolCacheEntry::flags_offset()));
207 __ andptr(rbx, 0xFF);
208 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
209 __ dispatch_next(state, step);
211 // out of the main line of code...
212 if (EnableInvokeDynamic) {
213 __ bind(L_giant_index);
214 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
215 __ jmp(L_got_cache);
216 }
218 return entry;
219 }
222 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
223 address entry = __ pc();
225 // In SSE mode, FP results are in xmm0
226 if (state == ftos && UseSSE > 0) {
227 __ subptr(rsp, wordSize);
228 __ movflt(Address(rsp, 0), xmm0);
229 __ fld_s(Address(rsp, 0));
230 __ addptr(rsp, wordSize);
231 } else if (state == dtos && UseSSE >= 2) {
232 __ subptr(rsp, 2*wordSize);
233 __ movdbl(Address(rsp, 0), xmm0);
234 __ fld_d(Address(rsp, 0));
235 __ addptr(rsp, 2*wordSize);
236 }
238 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
240 // The stack is not extended by deopt but we must NULL last_sp as this
241 // entry is like a "return".
242 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
243 __ restore_bcp();
244 __ restore_locals();
245 // handle exceptions
246 { Label L;
247 const Register thread = rcx;
248 __ get_thread(thread);
249 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
250 __ jcc(Assembler::zero, L);
251 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
252 __ should_not_reach_here();
253 __ bind(L);
254 }
255 __ dispatch_next(state, step);
256 return entry;
257 }
260 int AbstractInterpreter::BasicType_as_index(BasicType type) {
261 int i = 0;
262 switch (type) {
263 case T_BOOLEAN: i = 0; break;
264 case T_CHAR : i = 1; break;
265 case T_BYTE : i = 2; break;
266 case T_SHORT : i = 3; break;
267 case T_INT : // fall through
268 case T_LONG : // fall through
269 case T_VOID : i = 4; break;
270 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
271 case T_DOUBLE : i = 6; break;
272 case T_OBJECT : // fall through
273 case T_ARRAY : i = 7; break;
274 default : ShouldNotReachHere();
275 }
276 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
277 return i;
278 }
281 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
282 address entry = __ pc();
283 switch (type) {
284 case T_BOOLEAN: __ c2bool(rax); break;
285 case T_CHAR : __ andptr(rax, 0xFFFF); break;
286 case T_BYTE : __ sign_extend_byte (rax); break;
287 case T_SHORT : __ sign_extend_short(rax); break;
288 case T_INT : /* nothing to do */ break;
289 case T_DOUBLE :
290 case T_FLOAT :
291 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
292 __ pop(t); // remove return address first
293 // Must return a result for interpreter or compiler. In SSE
294 // mode, results are returned in xmm0 and the FPU stack must
295 // be empty.
296 if (type == T_FLOAT && UseSSE >= 1) {
297 // Load ST0
298 __ fld_d(Address(rsp, 0));
299 // Store as float and empty fpu stack
300 __ fstp_s(Address(rsp, 0));
301 // and reload
302 __ movflt(xmm0, Address(rsp, 0));
303 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
304 __ movdbl(xmm0, Address(rsp, 0));
305 } else {
306 // restore ST0
307 __ fld_d(Address(rsp, 0));
308 }
309 // and pop the temp
310 __ addptr(rsp, 2 * wordSize);
311 __ push(t); // restore return address
312 }
313 break;
314 case T_OBJECT :
315 // retrieve result from frame
316 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
317 // and verify it
318 __ verify_oop(rax);
319 break;
320 default : ShouldNotReachHere();
321 }
322 __ ret(0); // return from result handler
323 return entry;
324 }
326 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
327 address entry = __ pc();
328 __ push(state);
329 __ call_VM(noreg, runtime_entry);
330 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
331 return entry;
332 }
335 // Helpers for commoning out cases in the various type of method entries.
336 //
338 // increment invocation count & check for overflow
339 //
340 // Note: checking for negative value instead of overflow
341 // so we have a 'sticky' overflow test
342 //
343 // rbx,: method
344 // rcx: invocation counter
345 //
346 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
347 Label done;
348 // Note: In tiered we increment either counters in MethodCounters* or in MDO
349 // depending if we're profiling or not.
350 if (TieredCompilation) {
351 int increment = InvocationCounter::count_increment;
352 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
353 Label no_mdo;
354 if (ProfileInterpreter) {
355 // Are we profiling?
356 __ movptr(rax, Address(rbx, Method::method_data_offset()));
357 __ testptr(rax, rax);
358 __ jccb(Assembler::zero, no_mdo);
359 // Increment counter in the MDO
360 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
361 in_bytes(InvocationCounter::counter_offset()));
362 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
363 __ jmp(done);
364 }
365 __ bind(no_mdo);
366 // Increment counter in MethodCounters
367 const Address invocation_counter(rax,
368 MethodCounters::invocation_counter_offset() +
369 InvocationCounter::counter_offset());
371 __ get_method_counters(rbx, rax, done);
372 __ increment_mask_and_jump(invocation_counter, increment, mask,
373 rcx, false, Assembler::zero, overflow);
374 __ bind(done);
375 } else {
376 const Address backedge_counter (rax,
377 MethodCounters::backedge_counter_offset() +
378 InvocationCounter::counter_offset());
379 const Address invocation_counter(rax,
380 MethodCounters::invocation_counter_offset() +
381 InvocationCounter::counter_offset());
383 __ get_method_counters(rbx, rax, done);
385 if (ProfileInterpreter) {
386 __ incrementl(Address(rax,
387 MethodCounters::interpreter_invocation_counter_offset()));
388 }
390 // Update standard invocation counters
391 __ movl(rcx, invocation_counter);
392 __ incrementl(rcx, InvocationCounter::count_increment);
393 __ movl(invocation_counter, rcx); // save invocation count
395 __ movl(rax, backedge_counter); // load backedge counter
396 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
398 __ addl(rcx, rax); // add both counters
400 // profile_method is non-null only for interpreted method so
401 // profile_method != NULL == !native_call
402 // BytecodeInterpreter only calls for native so code is elided.
404 if (ProfileInterpreter && profile_method != NULL) {
405 // Test to see if we should create a method data oop
406 __ cmp32(rcx,
407 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
408 __ jcc(Assembler::less, *profile_method_continue);
410 // if no method data exists, go to profile_method
411 __ test_method_data_pointer(rax, *profile_method);
412 }
414 __ cmp32(rcx,
415 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
416 __ jcc(Assembler::aboveEqual, *overflow);
417 __ bind(done);
418 }
419 }
421 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
423 // Asm interpreter on entry
424 // rdi - locals
425 // rsi - bcp
426 // rbx, - method
427 // rdx - cpool
428 // rbp, - interpreter frame
430 // C++ interpreter on entry
431 // rsi - new interpreter state pointer
432 // rbp - interpreter frame pointer
433 // rbx - method
435 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
436 // rbx, - method
437 // rcx - rcvr (assuming there is one)
438 // top of stack return address of interpreter caller
439 // rsp - sender_sp
441 // C++ interpreter only
442 // rsi - previous interpreter state pointer
444 // InterpreterRuntime::frequency_counter_overflow takes one argument
445 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
446 // The call returns the address of the verified entry point for the method or NULL
447 // if the compilation did not complete (either went background or bailed out).
448 __ movptr(rax, (intptr_t)false);
449 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
451 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
453 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
454 // and jump to the interpreted entry.
455 __ jmp(*do_continue, relocInfo::none);
457 }
459 void InterpreterGenerator::generate_stack_overflow_check(void) {
460 // see if we've got enough room on the stack for locals plus overhead.
461 // the expression stack grows down incrementally, so the normal guard
462 // page mechanism will work for that.
463 //
464 // Registers live on entry:
465 //
466 // Asm interpreter
467 // rdx: number of additional locals this frame needs (what we must check)
468 // rbx,: Method*
470 // destroyed on exit
471 // rax,
473 // NOTE: since the additional locals are also always pushed (wasn't obvious in
474 // generate_method_entry) so the guard should work for them too.
475 //
477 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
478 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
480 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
481 // be sure to change this if you add/subtract anything to/from the overhead area
482 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
484 const int page_size = os::vm_page_size();
486 Label after_frame_check;
488 // see if the frame is greater than one page in size. If so,
489 // then we need to verify there is enough stack space remaining
490 // for the additional locals.
491 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
492 __ jcc(Assembler::belowEqual, after_frame_check);
494 // compute rsp as if this were going to be the last frame on
495 // the stack before the red zone
497 Label after_frame_check_pop;
499 __ push(rsi);
501 const Register thread = rsi;
503 __ get_thread(thread);
505 const Address stack_base(thread, Thread::stack_base_offset());
506 const Address stack_size(thread, Thread::stack_size_offset());
508 // locals + overhead, in bytes
509 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
511 #ifdef ASSERT
512 Label stack_base_okay, stack_size_okay;
513 // verify that thread stack base is non-zero
514 __ cmpptr(stack_base, (int32_t)NULL_WORD);
515 __ jcc(Assembler::notEqual, stack_base_okay);
516 __ stop("stack base is zero");
517 __ bind(stack_base_okay);
518 // verify that thread stack size is non-zero
519 __ cmpptr(stack_size, 0);
520 __ jcc(Assembler::notEqual, stack_size_okay);
521 __ stop("stack size is zero");
522 __ bind(stack_size_okay);
523 #endif
525 // Add stack base to locals and subtract stack size
526 __ addptr(rax, stack_base);
527 __ subptr(rax, stack_size);
529 // Use the maximum number of pages we might bang.
530 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
531 (StackRedPages+StackYellowPages);
532 __ addptr(rax, max_pages * page_size);
534 // check against the current stack bottom
535 __ cmpptr(rsp, rax);
536 __ jcc(Assembler::above, after_frame_check_pop);
538 __ pop(rsi); // get saved bcp / (c++ prev state ).
540 // Restore sender's sp as SP. This is necessary if the sender's
541 // frame is an extended compiled frame (see gen_c2i_adapter())
542 // and safer anyway in case of JSR292 adaptations.
544 __ pop(rax); // return address must be moved if SP is changed
545 __ mov(rsp, rsi);
546 __ push(rax);
548 // Note: the restored frame is not necessarily interpreted.
549 // Use the shared runtime version of the StackOverflowError.
550 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
551 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
552 // all done with frame size check
553 __ bind(after_frame_check_pop);
554 __ pop(rsi);
556 __ bind(after_frame_check);
557 }
559 // Allocate monitor and lock method (asm interpreter)
560 // rbx, - Method*
561 //
562 void InterpreterGenerator::lock_method(void) {
563 // synchronize method
564 const Address access_flags (rbx, Method::access_flags_offset());
565 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
566 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
568 #ifdef ASSERT
569 { Label L;
570 __ movl(rax, access_flags);
571 __ testl(rax, JVM_ACC_SYNCHRONIZED);
572 __ jcc(Assembler::notZero, L);
573 __ stop("method doesn't need synchronization");
574 __ bind(L);
575 }
576 #endif // ASSERT
577 // get synchronization object
578 { Label done;
579 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
580 __ movl(rax, access_flags);
581 __ testl(rax, JVM_ACC_STATIC);
582 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
583 __ jcc(Assembler::zero, done);
584 __ movptr(rax, Address(rbx, Method::const_offset()));
585 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
586 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
587 __ movptr(rax, Address(rax, mirror_offset));
588 __ bind(done);
589 }
590 // add space for monitor & lock
591 __ subptr(rsp, entry_size); // add space for a monitor entry
592 __ movptr(monitor_block_top, rsp); // set new monitor block top
593 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
594 __ mov(rdx, rsp); // object address
595 __ lock_object(rdx);
596 }
598 //
599 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
600 // and for native methods hence the shared code.
602 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
603 // initialize fixed part of activation frame
604 __ push(rax); // save return address
605 __ enter(); // save old & set new rbp,
608 __ push(rsi); // set sender sp
609 __ push((int32_t)NULL_WORD); // leave last_sp as null
610 __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod*
611 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
612 __ push(rbx); // save Method*
613 if (ProfileInterpreter) {
614 Label method_data_continue;
615 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
616 __ testptr(rdx, rdx);
617 __ jcc(Assembler::zero, method_data_continue);
618 __ addptr(rdx, in_bytes(MethodData::data_offset()));
619 __ bind(method_data_continue);
620 __ push(rdx); // set the mdp (method data pointer)
621 } else {
622 __ push(0);
623 }
625 __ movptr(rdx, Address(rbx, Method::const_offset()));
626 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
627 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
628 __ push(rdx); // set constant pool cache
629 __ push(rdi); // set locals pointer
630 if (native_call) {
631 __ push(0); // no bcp
632 } else {
633 __ push(rsi); // set bcp
634 }
635 __ push(0); // reserve word for pointer to expression stack bottom
636 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
637 }
639 // End of helpers
641 //
642 // Various method entries
643 //------------------------------------------------------------------------------------------------------------------------
644 //
645 //
647 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
649 address InterpreterGenerator::generate_accessor_entry(void) {
651 // rbx,: Method*
652 // rcx: receiver (preserve for slow entry into asm interpreter)
654 // rsi: senderSP must preserved for slow path, set SP to it on fast path
656 address entry_point = __ pc();
657 Label xreturn_path;
659 // do fastpath for resolved accessor methods
660 if (UseFastAccessorMethods) {
661 Label slow_path;
662 // If we need a safepoint check, generate full interpreter entry.
663 ExternalAddress state(SafepointSynchronize::address_of_state());
664 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
665 SafepointSynchronize::_not_synchronized);
667 __ jcc(Assembler::notEqual, slow_path);
668 // ASM/C++ Interpreter
669 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
670 // Note: We can only use this code if the getfield has been resolved
671 // and if we don't have a null-pointer exception => check for
672 // these conditions first and use slow path if necessary.
673 // rbx,: method
674 // rcx: receiver
675 __ movptr(rax, Address(rsp, wordSize));
677 // check if local 0 != NULL and read field
678 __ testptr(rax, rax);
679 __ jcc(Assembler::zero, slow_path);
681 // read first instruction word and extract bytecode @ 1 and index @ 2
682 __ movptr(rdx, Address(rbx, Method::const_offset()));
683 __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
684 __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
685 // Shift codes right to get the index on the right.
686 // The bytecode fetched looks like <index><0xb4><0x2a>
687 __ shrl(rdx, 2*BitsPerByte);
688 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
689 __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
691 // rax,: local 0
692 // rbx,: method
693 // rcx: receiver - do not destroy since it is needed for slow path!
694 // rcx: scratch
695 // rdx: constant pool cache index
696 // rdi: constant pool cache
697 // rsi: sender sp
699 // check if getfield has been resolved and read constant pool cache entry
700 // check the validity of the cache entry by testing whether _indices field
701 // contains Bytecode::_getfield in b1 byte.
702 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
703 __ movl(rcx,
704 Address(rdi,
705 rdx,
706 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
707 __ shrl(rcx, 2*BitsPerByte);
708 __ andl(rcx, 0xFF);
709 __ cmpl(rcx, Bytecodes::_getfield);
710 __ jcc(Assembler::notEqual, slow_path);
712 // Note: constant pool entry is not valid before bytecode is resolved
713 __ movptr(rcx,
714 Address(rdi,
715 rdx,
716 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
717 __ movl(rdx,
718 Address(rdi,
719 rdx,
720 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
722 Label notByte, notShort, notChar;
723 const Address field_address (rax, rcx, Address::times_1);
725 // Need to differentiate between igetfield, agetfield, bgetfield etc.
726 // because they are different sizes.
727 // Use the type from the constant pool cache
728 __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
729 // Make sure we don't need to mask rdx after the above shift
730 ConstantPoolCacheEntry::verify_tos_state_shift();
731 __ cmpl(rdx, btos);
732 __ jcc(Assembler::notEqual, notByte);
733 __ load_signed_byte(rax, field_address);
734 __ jmp(xreturn_path);
736 __ bind(notByte);
737 __ cmpl(rdx, stos);
738 __ jcc(Assembler::notEqual, notShort);
739 __ load_signed_short(rax, field_address);
740 __ jmp(xreturn_path);
742 __ bind(notShort);
743 __ cmpl(rdx, ctos);
744 __ jcc(Assembler::notEqual, notChar);
745 __ load_unsigned_short(rax, field_address);
746 __ jmp(xreturn_path);
748 __ bind(notChar);
749 #ifdef ASSERT
750 Label okay;
751 __ cmpl(rdx, atos);
752 __ jcc(Assembler::equal, okay);
753 __ cmpl(rdx, itos);
754 __ jcc(Assembler::equal, okay);
755 __ stop("what type is this?");
756 __ bind(okay);
757 #endif // ASSERT
758 // All the rest are a 32 bit wordsize
759 // This is ok for now. Since fast accessors should be going away
760 __ movptr(rax, field_address);
762 __ bind(xreturn_path);
764 // _ireturn/_areturn
765 __ pop(rdi); // get return address
766 __ mov(rsp, rsi); // set sp to sender sp
767 __ jmp(rdi);
769 // generate a vanilla interpreter entry as the slow path
770 __ bind(slow_path);
772 (void) generate_normal_entry(false);
773 return entry_point;
774 }
775 return NULL;
777 }
779 // Method entry for java.lang.ref.Reference.get.
780 address InterpreterGenerator::generate_Reference_get_entry(void) {
781 #if INCLUDE_ALL_GCS
782 // Code: _aload_0, _getfield, _areturn
783 // parameter size = 1
784 //
785 // The code that gets generated by this routine is split into 2 parts:
786 // 1. The "intrinsified" code for G1 (or any SATB based GC),
787 // 2. The slow path - which is an expansion of the regular method entry.
788 //
789 // Notes:-
790 // * In the G1 code we do not check whether we need to block for
791 // a safepoint. If G1 is enabled then we must execute the specialized
792 // code for Reference.get (except when the Reference object is null)
793 // so that we can log the value in the referent field with an SATB
794 // update buffer.
795 // If the code for the getfield template is modified so that the
796 // G1 pre-barrier code is executed when the current method is
797 // Reference.get() then going through the normal method entry
798 // will be fine.
799 // * The G1 code below can, however, check the receiver object (the instance
800 // of java.lang.Reference) and jump to the slow path if null. If the
801 // Reference object is null then we obviously cannot fetch the referent
802 // and so we don't need to call the G1 pre-barrier. Thus we can use the
803 // regular method entry code to generate the NPE.
804 //
805 // This code is based on generate_accessor_enty.
807 // rbx,: Method*
808 // rcx: receiver (preserve for slow entry into asm interpreter)
810 // rsi: senderSP must preserved for slow path, set SP to it on fast path
812 address entry = __ pc();
814 const int referent_offset = java_lang_ref_Reference::referent_offset;
815 guarantee(referent_offset > 0, "referent offset not initialized");
817 if (UseG1GC) {
818 Label slow_path;
820 // Check if local 0 != NULL
821 // If the receiver is null then it is OK to jump to the slow path.
822 __ movptr(rax, Address(rsp, wordSize));
823 __ testptr(rax, rax);
824 __ jcc(Assembler::zero, slow_path);
826 // rax: local 0 (must be preserved across the G1 barrier call)
827 //
828 // rbx: method (at this point it's scratch)
829 // rcx: receiver (at this point it's scratch)
830 // rdx: scratch
831 // rdi: scratch
832 //
833 // rsi: sender sp
835 // Preserve the sender sp in case the pre-barrier
836 // calls the runtime
837 __ push(rsi);
839 // Load the value of the referent field.
840 const Address field_address(rax, referent_offset);
841 __ movptr(rax, field_address);
843 // Generate the G1 pre-barrier code to log the value of
844 // the referent field in an SATB buffer.
845 __ get_thread(rcx);
846 __ g1_write_barrier_pre(noreg /* obj */,
847 rax /* pre_val */,
848 rcx /* thread */,
849 rbx /* tmp */,
850 true /* tosca_save */,
851 true /* expand_call */);
853 // _areturn
854 __ pop(rsi); // get sender sp
855 __ pop(rdi); // get return address
856 __ mov(rsp, rsi); // set sp to sender sp
857 __ jmp(rdi);
859 __ bind(slow_path);
860 (void) generate_normal_entry(false);
862 return entry;
863 }
864 #endif // INCLUDE_ALL_GCS
866 // If G1 is not enabled then attempt to go through the accessor entry point
867 // Reference.get is an accessor
868 return generate_accessor_entry();
869 }
871 /**
872 * Method entry for static native methods:
873 * int java.util.zip.CRC32.update(int crc, int b)
874 */
875 address InterpreterGenerator::generate_CRC32_update_entry() {
876 if (UseCRC32Intrinsics) {
877 address entry = __ pc();
879 // rbx,: Method*
880 // rsi: senderSP must preserved for slow path, set SP to it on fast path
881 // rdx: scratch
882 // rdi: scratch
884 Label slow_path;
885 // If we need a safepoint check, generate full interpreter entry.
886 ExternalAddress state(SafepointSynchronize::address_of_state());
887 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
888 SafepointSynchronize::_not_synchronized);
889 __ jcc(Assembler::notEqual, slow_path);
891 // We don't generate local frame and don't align stack because
892 // we call stub code and there is no safepoint on this path.
894 // Load parameters
895 const Register crc = rax; // crc
896 const Register val = rdx; // source java byte value
897 const Register tbl = rdi; // scratch
899 // Arguments are reversed on java expression stack
900 __ movl(val, Address(rsp, wordSize)); // byte value
901 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
903 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
904 __ notl(crc); // ~crc
905 __ update_byte_crc32(crc, val, tbl);
906 __ notl(crc); // ~crc
907 // result in rax
909 // _areturn
910 __ pop(rdi); // get return address
911 __ mov(rsp, rsi); // set sp to sender sp
912 __ jmp(rdi);
914 // generate a vanilla native entry as the slow path
915 __ bind(slow_path);
917 (void) generate_native_entry(false);
919 return entry;
920 }
921 return generate_native_entry(false);
922 }
924 /**
925 * Method entry for static native methods:
926 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
927 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
928 */
929 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
930 if (UseCRC32Intrinsics) {
931 address entry = __ pc();
933 // rbx,: Method*
934 // rsi: senderSP must preserved for slow path, set SP to it on fast path
935 // rdx: scratch
936 // rdi: scratch
938 Label slow_path;
939 // If we need a safepoint check, generate full interpreter entry.
940 ExternalAddress state(SafepointSynchronize::address_of_state());
941 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
942 SafepointSynchronize::_not_synchronized);
943 __ jcc(Assembler::notEqual, slow_path);
945 // We don't generate local frame and don't align stack because
946 // we call stub code and there is no safepoint on this path.
948 // Load parameters
949 const Register crc = rax; // crc
950 const Register buf = rdx; // source java byte array address
951 const Register len = rdi; // length
953 // Arguments are reversed on java expression stack
954 __ movl(len, Address(rsp, wordSize)); // Length
955 // Calculate address of start element
956 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
957 __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
958 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
959 __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC
960 } else {
961 __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
962 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
963 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
964 __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC
965 }
967 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
968 // result in rax
970 // _areturn
971 __ pop(rdi); // get return address
972 __ mov(rsp, rsi); // set sp to sender sp
973 __ jmp(rdi);
975 // generate a vanilla native entry as the slow path
976 __ bind(slow_path);
978 (void) generate_native_entry(false);
980 return entry;
981 }
982 return generate_native_entry(false);
983 }
985 //
986 // Interpreter stub for calling a native method. (asm interpreter)
987 // This sets up a somewhat different looking stack for calling the native method
988 // than the typical interpreter frame setup.
989 //
991 address InterpreterGenerator::generate_native_entry(bool synchronized) {
992 // determine code generation flags
993 bool inc_counter = UseCompiler || CountCompiledCalls;
995 // rbx,: Method*
996 // rsi: sender sp
997 // rsi: previous interpreter state (C++ interpreter) must preserve
998 address entry_point = __ pc();
1000 const Address constMethod (rbx, Method::const_offset());
1001 const Address access_flags (rbx, Method::access_flags_offset());
1002 const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
1004 // get parameter size (always needed)
1005 __ movptr(rcx, constMethod);
1006 __ load_unsigned_short(rcx, size_of_parameters);
1008 // native calls don't need the stack size check since they have no expression stack
1009 // and the arguments are already on the stack and we only add a handful of words
1010 // to the stack
1012 // rbx,: Method*
1013 // rcx: size of parameters
1014 // rsi: sender sp
1016 __ pop(rax); // get return address
1017 // for natives the size of locals is zero
1019 // compute beginning of parameters (rdi)
1020 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1023 // add 2 zero-initialized slots for native calls
1024 // NULL result handler
1025 __ push((int32_t)NULL_WORD);
1026 // NULL oop temp (mirror or jni oop result)
1027 __ push((int32_t)NULL_WORD);
1029 // initialize fixed part of activation frame
1030 generate_fixed_frame(true);
1032 // make sure method is native & not abstract
1033 #ifdef ASSERT
1034 __ movl(rax, access_flags);
1035 {
1036 Label L;
1037 __ testl(rax, JVM_ACC_NATIVE);
1038 __ jcc(Assembler::notZero, L);
1039 __ stop("tried to execute non-native method as native");
1040 __ bind(L);
1041 }
1042 { Label L;
1043 __ testl(rax, JVM_ACC_ABSTRACT);
1044 __ jcc(Assembler::zero, L);
1045 __ stop("tried to execute abstract method in interpreter");
1046 __ bind(L);
1047 }
1048 #endif
1050 // Since at this point in the method invocation the exception handler
1051 // would try to exit the monitor of synchronized methods which hasn't
1052 // been entered yet, we set the thread local variable
1053 // _do_not_unlock_if_synchronized to true. The remove_activation will
1054 // check this flag.
1056 __ get_thread(rax);
1057 const Address do_not_unlock_if_synchronized(rax,
1058 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1059 __ movbool(do_not_unlock_if_synchronized, true);
1061 // increment invocation count & check for overflow
1062 Label invocation_counter_overflow;
1063 if (inc_counter) {
1064 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1065 }
1067 Label continue_after_compile;
1068 __ bind(continue_after_compile);
1070 bang_stack_shadow_pages(true);
1072 // reset the _do_not_unlock_if_synchronized flag
1073 __ get_thread(rax);
1074 __ movbool(do_not_unlock_if_synchronized, false);
1076 // check for synchronized methods
1077 // Must happen AFTER invocation_counter check and stack overflow check,
1078 // so method is not locked if overflows.
1079 //
1080 if (synchronized) {
1081 lock_method();
1082 } else {
1083 // no synchronization necessary
1084 #ifdef ASSERT
1085 { Label L;
1086 __ movl(rax, access_flags);
1087 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1088 __ jcc(Assembler::zero, L);
1089 __ stop("method needs synchronization");
1090 __ bind(L);
1091 }
1092 #endif
1093 }
1095 // start execution
1096 #ifdef ASSERT
1097 { Label L;
1098 const Address monitor_block_top (rbp,
1099 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1100 __ movptr(rax, monitor_block_top);
1101 __ cmpptr(rax, rsp);
1102 __ jcc(Assembler::equal, L);
1103 __ stop("broken stack frame setup in interpreter");
1104 __ bind(L);
1105 }
1106 #endif
1108 // jvmti/dtrace support
1109 __ notify_method_entry();
1111 // work registers
1112 const Register method = rbx;
1113 const Register thread = rdi;
1114 const Register t = rcx;
1116 // allocate space for parameters
1117 __ get_method(method);
1118 __ movptr(t, Address(method, Method::const_offset()));
1119 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1121 __ shlptr(t, Interpreter::logStackElementSize);
1122 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
1123 __ subptr(rsp, t);
1124 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
1126 // get signature handler
1127 { Label L;
1128 __ movptr(t, Address(method, Method::signature_handler_offset()));
1129 __ testptr(t, t);
1130 __ jcc(Assembler::notZero, L);
1131 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1132 __ get_method(method);
1133 __ movptr(t, Address(method, Method::signature_handler_offset()));
1134 __ bind(L);
1135 }
1137 // call signature handler
1138 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
1139 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
1140 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1141 // The generated handlers do not touch RBX (the method oop).
1142 // However, large signatures cannot be cached and are generated
1143 // each time here. The slow-path generator will blow RBX
1144 // sometime, so we must reload it after the call.
1145 __ call(t);
1146 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
1148 // result handler is in rax,
1149 // set result handler
1150 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1152 // pass mirror handle if static call
1153 { Label L;
1154 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1155 __ movl(t, Address(method, Method::access_flags_offset()));
1156 __ testl(t, JVM_ACC_STATIC);
1157 __ jcc(Assembler::zero, L);
1158 // get mirror
1159 __ movptr(t, Address(method, Method:: const_offset()));
1160 __ movptr(t, Address(t, ConstMethod::constants_offset()));
1161 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1162 __ movptr(t, Address(t, mirror_offset));
1163 // copy mirror into activation frame
1164 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
1165 // pass handle to mirror
1166 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1167 __ movptr(Address(rsp, wordSize), t);
1168 __ bind(L);
1169 }
1171 // get native function entry point
1172 { Label L;
1173 __ movptr(rax, Address(method, Method::native_function_offset()));
1174 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1175 __ cmpptr(rax, unsatisfied.addr());
1176 __ jcc(Assembler::notEqual, L);
1177 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1178 __ get_method(method);
1179 __ movptr(rax, Address(method, Method::native_function_offset()));
1180 __ bind(L);
1181 }
1183 // pass JNIEnv
1184 __ get_thread(thread);
1185 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1186 __ movptr(Address(rsp, 0), t);
1188 // set_last_Java_frame_before_call
1189 // It is enough that the pc()
1190 // points into the right code segment. It does not have to be the correct return pc.
1191 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1193 // change thread state
1194 #ifdef ASSERT
1195 { Label L;
1196 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1197 __ cmpl(t, _thread_in_Java);
1198 __ jcc(Assembler::equal, L);
1199 __ stop("Wrong thread state in native stub");
1200 __ bind(L);
1201 }
1202 #endif
1204 // Change state to native
1205 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1206 __ call(rax);
1208 // result potentially in rdx:rax or ST0
1210 // Verify or restore cpu control state after JNI call
1211 __ restore_cpu_control_state_after_jni();
1213 // save potential result in ST(0) & rdx:rax
1214 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1215 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1216 // It is safe to do this push because state is _thread_in_native and return address will be found
1217 // via _last_native_pc and not via _last_jave_sp
1219 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1220 // If the order changes or anything else is added to the stack the code in
1221 // interpreter_frame_result will have to be changed.
1223 { Label L;
1224 Label push_double;
1225 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1226 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1227 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1228 float_handler.addr());
1229 __ jcc(Assembler::equal, push_double);
1230 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1231 double_handler.addr());
1232 __ jcc(Assembler::notEqual, L);
1233 __ bind(push_double);
1234 __ push(dtos);
1235 __ bind(L);
1236 }
1237 __ push(ltos);
1239 // change thread state
1240 __ get_thread(thread);
1241 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1242 if(os::is_MP()) {
1243 if (UseMembar) {
1244 // Force this write out before the read below
1245 __ membar(Assembler::Membar_mask_bits(
1246 Assembler::LoadLoad | Assembler::LoadStore |
1247 Assembler::StoreLoad | Assembler::StoreStore));
1248 } else {
1249 // Write serialization page so VM thread can do a pseudo remote membar.
1250 // We use the current thread pointer to calculate a thread specific
1251 // offset to write to within the page. This minimizes bus traffic
1252 // due to cache line collision.
1253 __ serialize_memory(thread, rcx);
1254 }
1255 }
1257 if (AlwaysRestoreFPU) {
1258 // Make sure the control word is correct.
1259 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1260 }
1262 // check for safepoint operation in progress and/or pending suspend requests
1263 { Label Continue;
1265 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1266 SafepointSynchronize::_not_synchronized);
1268 Label L;
1269 __ jcc(Assembler::notEqual, L);
1270 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1271 __ jcc(Assembler::equal, Continue);
1272 __ bind(L);
1274 // Don't use call_VM as it will see a possible pending exception and forward it
1275 // and never return here preventing us from clearing _last_native_pc down below.
1276 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1277 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1278 // by hand.
1279 //
1280 __ push(thread);
1281 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1282 JavaThread::check_special_condition_for_native_trans)));
1283 __ increment(rsp, wordSize);
1284 __ get_thread(thread);
1286 __ bind(Continue);
1287 }
1289 // change thread state
1290 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1292 __ reset_last_Java_frame(thread, true, true);
1294 // reset handle block
1295 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1296 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1298 // If result was an oop then unbox and save it in the frame
1299 { Label L;
1300 Label no_oop, store_result;
1301 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1302 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1303 handler.addr());
1304 __ jcc(Assembler::notEqual, no_oop);
1305 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1306 __ pop(ltos);
1307 __ testptr(rax, rax);
1308 __ jcc(Assembler::zero, store_result);
1309 // unbox
1310 __ movptr(rax, Address(rax, 0));
1311 __ bind(store_result);
1312 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1313 // keep stack depth as expected by pushing oop which will eventually be discarded
1314 __ push(ltos);
1315 __ bind(no_oop);
1316 }
1318 {
1319 Label no_reguard;
1320 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1321 __ jcc(Assembler::notEqual, no_reguard);
1323 __ pusha();
1324 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1325 __ popa();
1327 __ bind(no_reguard);
1328 }
1330 // restore rsi to have legal interpreter frame,
1331 // i.e., bci == 0 <=> rsi == code_base()
1332 // Can't call_VM until bcp is within reasonable.
1333 __ get_method(method); // method is junk from thread_in_native to now.
1334 __ movptr(rsi, Address(method,Method::const_offset())); // get ConstMethod*
1335 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
1337 // handle exceptions (exception handling will handle unlocking!)
1338 { Label L;
1339 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1340 __ jcc(Assembler::zero, L);
1341 // Note: At some point we may want to unify this with the code used in call_VM_base();
1342 // i.e., we should use the StubRoutines::forward_exception code. For now this
1343 // doesn't work here because the rsp is not correctly set at this point.
1344 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1345 __ should_not_reach_here();
1346 __ bind(L);
1347 }
1349 // do unlocking if necessary
1350 { Label L;
1351 __ movl(t, Address(method, Method::access_flags_offset()));
1352 __ testl(t, JVM_ACC_SYNCHRONIZED);
1353 __ jcc(Assembler::zero, L);
1354 // the code below should be shared with interpreter macro assembler implementation
1355 { Label unlock;
1356 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1357 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1358 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1360 __ lea(rdx, monitor); // address of first monitor
1362 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1363 __ testptr(t, t);
1364 __ jcc(Assembler::notZero, unlock);
1366 // Entry already unlocked, need to throw exception
1367 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1368 __ should_not_reach_here();
1370 __ bind(unlock);
1371 __ unlock_object(rdx);
1372 }
1373 __ bind(L);
1374 }
1376 // jvmti/dtrace support
1377 // Note: This must happen _after_ handling/throwing any exceptions since
1378 // the exception handler code notifies the runtime of method exits
1379 // too. If this happens before, method entry/exit notifications are
1380 // not properly paired (was bug - gri 11/22/99).
1381 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1383 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1384 __ pop(ltos);
1385 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1386 __ call(t);
1388 // remove activation
1389 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1390 __ leave(); // remove frame anchor
1391 __ pop(rdi); // get return address
1392 __ mov(rsp, t); // set sp to sender sp
1393 __ jmp(rdi);
1395 if (inc_counter) {
1396 // Handle overflow of counter and compile method
1397 __ bind(invocation_counter_overflow);
1398 generate_counter_overflow(&continue_after_compile);
1399 }
1401 return entry_point;
1402 }
1404 //
1405 // Generic interpreted method entry to (asm) interpreter
1406 //
1407 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1408 // determine code generation flags
1409 bool inc_counter = UseCompiler || CountCompiledCalls;
1411 // rbx,: Method*
1412 // rsi: sender sp
1413 address entry_point = __ pc();
1415 const Address constMethod (rbx, Method::const_offset());
1416 const Address access_flags (rbx, Method::access_flags_offset());
1417 const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
1418 const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset());
1420 // get parameter size (always needed)
1421 __ movptr(rdx, constMethod);
1422 __ load_unsigned_short(rcx, size_of_parameters);
1424 // rbx,: Method*
1425 // rcx: size of parameters
1427 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1429 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1430 __ subl(rdx, rcx); // rdx = no. of additional locals
1432 // see if we've got enough room on the stack for locals plus overhead.
1433 generate_stack_overflow_check();
1435 // get return address
1436 __ pop(rax);
1438 // compute beginning of parameters (rdi)
1439 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1441 // rdx - # of additional locals
1442 // allocate space for locals
1443 // explicitly initialize locals
1444 {
1445 Label exit, loop;
1446 __ testl(rdx, rdx);
1447 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1448 __ bind(loop);
1449 __ push((int32_t)NULL_WORD); // initialize local variables
1450 __ decrement(rdx); // until everything initialized
1451 __ jcc(Assembler::greater, loop);
1452 __ bind(exit);
1453 }
1455 // initialize fixed part of activation frame
1456 generate_fixed_frame(false);
1458 // make sure method is not native & not abstract
1459 #ifdef ASSERT
1460 __ movl(rax, access_flags);
1461 {
1462 Label L;
1463 __ testl(rax, JVM_ACC_NATIVE);
1464 __ jcc(Assembler::zero, L);
1465 __ stop("tried to execute native method as non-native");
1466 __ bind(L);
1467 }
1468 { Label L;
1469 __ testl(rax, JVM_ACC_ABSTRACT);
1470 __ jcc(Assembler::zero, L);
1471 __ stop("tried to execute abstract method in interpreter");
1472 __ bind(L);
1473 }
1474 #endif
1476 // Since at this point in the method invocation the exception handler
1477 // would try to exit the monitor of synchronized methods which hasn't
1478 // been entered yet, we set the thread local variable
1479 // _do_not_unlock_if_synchronized to true. The remove_activation will
1480 // check this flag.
1482 __ get_thread(rax);
1483 const Address do_not_unlock_if_synchronized(rax,
1484 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1485 __ movbool(do_not_unlock_if_synchronized, true);
1487 // increment invocation count & check for overflow
1488 Label invocation_counter_overflow;
1489 Label profile_method;
1490 Label profile_method_continue;
1491 if (inc_counter) {
1492 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1493 if (ProfileInterpreter) {
1494 __ bind(profile_method_continue);
1495 }
1496 }
1497 Label continue_after_compile;
1498 __ bind(continue_after_compile);
1500 bang_stack_shadow_pages(false);
1502 // reset the _do_not_unlock_if_synchronized flag
1503 __ get_thread(rax);
1504 __ movbool(do_not_unlock_if_synchronized, false);
1506 // check for synchronized methods
1507 // Must happen AFTER invocation_counter check and stack overflow check,
1508 // so method is not locked if overflows.
1509 //
1510 if (synchronized) {
1511 // Allocate monitor and lock method
1512 lock_method();
1513 } else {
1514 // no synchronization necessary
1515 #ifdef ASSERT
1516 { Label L;
1517 __ movl(rax, access_flags);
1518 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1519 __ jcc(Assembler::zero, L);
1520 __ stop("method needs synchronization");
1521 __ bind(L);
1522 }
1523 #endif
1524 }
1526 // start execution
1527 #ifdef ASSERT
1528 { Label L;
1529 const Address monitor_block_top (rbp,
1530 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1531 __ movptr(rax, monitor_block_top);
1532 __ cmpptr(rax, rsp);
1533 __ jcc(Assembler::equal, L);
1534 __ stop("broken stack frame setup in interpreter");
1535 __ bind(L);
1536 }
1537 #endif
1539 // jvmti support
1540 __ notify_method_entry();
1542 __ dispatch_next(vtos);
1544 // invocation counter overflow
1545 if (inc_counter) {
1546 if (ProfileInterpreter) {
1547 // We have decided to profile this method in the interpreter
1548 __ bind(profile_method);
1549 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1550 __ set_method_data_pointer_for_bcp();
1551 __ get_method(rbx);
1552 __ jmp(profile_method_continue);
1553 }
1554 // Handle overflow of counter and compile method
1555 __ bind(invocation_counter_overflow);
1556 generate_counter_overflow(&continue_after_compile);
1557 }
1559 return entry_point;
1560 }
1562 //------------------------------------------------------------------------------------------------------------------------
1563 // Entry points
1564 //
1565 // Here we generate the various kind of entries into the interpreter.
1566 // The two main entry type are generic bytecode methods and native call method.
1567 // These both come in synchronized and non-synchronized versions but the
1568 // frame layout they create is very similar. The other method entry
1569 // types are really just special purpose entries that are really entry
1570 // and interpretation all in one. These are for trivial methods like
1571 // accessor, empty, or special math methods.
1572 //
1573 // When control flow reaches any of the entry types for the interpreter
1574 // the following holds ->
1575 //
1576 // Arguments:
1577 //
1578 // rbx,: Method*
1579 // rcx: receiver
1580 //
1581 //
1582 // Stack layout immediately at entry
1583 //
1584 // [ return address ] <--- rsp
1585 // [ parameter n ]
1586 // ...
1587 // [ parameter 1 ]
1588 // [ expression stack ] (caller's java expression stack)
1590 // Assuming that we don't go to one of the trivial specialized
1591 // entries the stack will look like below when we are ready to execute
1592 // the first bytecode (or call the native routine). The register usage
1593 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1594 //
1595 // local variables follow incoming parameters immediately; i.e.
1596 // the return address is moved to the end of the locals).
1597 //
1598 // [ monitor entry ] <--- rsp
1599 // ...
1600 // [ monitor entry ]
1601 // [ expr. stack bottom ]
1602 // [ saved rsi ]
1603 // [ current rdi ]
1604 // [ Method* ]
1605 // [ saved rbp, ] <--- rbp,
1606 // [ return address ]
1607 // [ local variable m ]
1608 // ...
1609 // [ local variable 1 ]
1610 // [ parameter n ]
1611 // ...
1612 // [ parameter 1 ] <--- rdi
1614 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1615 // determine code generation flags
1616 bool synchronized = false;
1617 address entry_point = NULL;
1618 InterpreterGenerator* ig_this = (InterpreterGenerator*)this;
1620 switch (kind) {
1621 case Interpreter::zerolocals : break;
1622 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1623 case Interpreter::native : entry_point = ig_this->generate_native_entry(false); break;
1624 case Interpreter::native_synchronized : entry_point = ig_this->generate_native_entry(true); break;
1625 case Interpreter::empty : entry_point = ig_this->generate_empty_entry(); break;
1626 case Interpreter::accessor : entry_point = ig_this->generate_accessor_entry(); break;
1627 case Interpreter::abstract : entry_point = ig_this->generate_abstract_entry(); break;
1629 case Interpreter::java_lang_math_sin : // fall thru
1630 case Interpreter::java_lang_math_cos : // fall thru
1631 case Interpreter::java_lang_math_tan : // fall thru
1632 case Interpreter::java_lang_math_abs : // fall thru
1633 case Interpreter::java_lang_math_log : // fall thru
1634 case Interpreter::java_lang_math_log10 : // fall thru
1635 case Interpreter::java_lang_math_sqrt : // fall thru
1636 case Interpreter::java_lang_math_pow : // fall thru
1637 case Interpreter::java_lang_math_exp : entry_point = ig_this->generate_math_entry(kind); break;
1638 case Interpreter::java_lang_ref_reference_get
1639 : entry_point = ig_this->generate_Reference_get_entry(); break;
1640 case Interpreter::java_util_zip_CRC32_update
1641 : entry_point = ig_this->generate_CRC32_update_entry(); break;
1642 case Interpreter::java_util_zip_CRC32_updateBytes
1643 : // fall thru
1644 case Interpreter::java_util_zip_CRC32_updateByteBuffer
1645 : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break;
1646 default:
1647 fatal(err_msg("unexpected method kind: %d", kind));
1648 break;
1649 }
1651 if (entry_point) return entry_point;
1653 return ig_this->generate_normal_entry(synchronized);
1655 }
1657 // These should never be compiled since the interpreter will prefer
1658 // the compiled version to the intrinsic version.
1659 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1660 switch (method_kind(m)) {
1661 case Interpreter::java_lang_math_sin : // fall thru
1662 case Interpreter::java_lang_math_cos : // fall thru
1663 case Interpreter::java_lang_math_tan : // fall thru
1664 case Interpreter::java_lang_math_abs : // fall thru
1665 case Interpreter::java_lang_math_log : // fall thru
1666 case Interpreter::java_lang_math_log10 : // fall thru
1667 case Interpreter::java_lang_math_sqrt : // fall thru
1668 case Interpreter::java_lang_math_pow : // fall thru
1669 case Interpreter::java_lang_math_exp :
1670 return false;
1671 default:
1672 return true;
1673 }
1674 }
1676 // How much stack a method activation needs in words.
1677 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1679 const int stub_code = 4; // see generate_call_stub
1680 // Save space for one monitor to get into the interpreted method in case
1681 // the method is synchronized
1682 int monitor_size = method->is_synchronized() ?
1683 1*frame::interpreter_frame_monitor_size() : 0;
1685 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1686 // be sure to change this if you add/subtract anything to/from the overhead area
1687 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1689 const int method_stack = (method->max_locals() + method->max_stack()) *
1690 Interpreter::stackElementWords;
1691 return overhead_size + method_stack + stub_code;
1692 }
1694 // asm based interpreter deoptimization helpers
1696 int AbstractInterpreter::layout_activation(Method* method,
1697 int tempcount,
1698 int popframe_extra_args,
1699 int moncount,
1700 int caller_actual_parameters,
1701 int callee_param_count,
1702 int callee_locals,
1703 frame* caller,
1704 frame* interpreter_frame,
1705 bool is_top_frame,
1706 bool is_bottom_frame) {
1707 // Note: This calculation must exactly parallel the frame setup
1708 // in AbstractInterpreterGenerator::generate_method_entry.
1709 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1710 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
1711 // as determined by a previous call to this method.
1712 // It is also guaranteed to be walkable even though it is in a skeletal state
1713 // NOTE: return size is in words not bytes
1715 // fixed size of an interpreter frame:
1716 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1717 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1718 Interpreter::stackElementWords;
1720 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
1722 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
1723 // Since the callee parameters already account for the callee's params we only need to account for
1724 // the extra locals.
1727 int size = overhead +
1728 ((callee_locals - callee_param_count)*Interpreter::stackElementWords) +
1729 (moncount*frame::interpreter_frame_monitor_size()) +
1730 tempcount*Interpreter::stackElementWords + popframe_extra_args;
1732 if (interpreter_frame != NULL) {
1733 #ifdef ASSERT
1734 if (!EnableInvokeDynamic)
1735 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1736 // Probably, since deoptimization doesn't work yet.
1737 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1738 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1739 #endif
1741 interpreter_frame->interpreter_frame_set_method(method);
1742 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1743 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1744 // and sender_sp is fp+8
1745 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1747 #ifdef ASSERT
1748 if (caller->is_interpreted_frame()) {
1749 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1750 }
1751 #endif
1753 interpreter_frame->interpreter_frame_set_locals(locals);
1754 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1755 BasicObjectLock* monbot = montop - moncount;
1756 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1758 // Set last_sp
1759 intptr_t* rsp = (intptr_t*) monbot -
1760 tempcount*Interpreter::stackElementWords -
1761 popframe_extra_args;
1762 interpreter_frame->interpreter_frame_set_last_sp(rsp);
1764 // All frames but the initial (oldest) interpreter frame we fill in have a
1765 // value for sender_sp that allows walking the stack but isn't
1766 // truly correct. Correct the value here.
1768 if (extra_locals != 0 &&
1769 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1770 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1771 }
1772 *interpreter_frame->interpreter_frame_cache_addr() =
1773 method->constants()->cache();
1774 }
1775 return size;
1776 }
1779 //------------------------------------------------------------------------------------------------------------------------
1780 // Exceptions
1782 void TemplateInterpreterGenerator::generate_throw_exception() {
1783 // Entry point in previous activation (i.e., if the caller was interpreted)
1784 Interpreter::_rethrow_exception_entry = __ pc();
1785 const Register thread = rcx;
1787 // Restore sp to interpreter_frame_last_sp even though we are going
1788 // to empty the expression stack for the exception processing.
1789 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1790 // rax,: exception
1791 // rdx: return address/pc that threw exception
1792 __ restore_bcp(); // rsi points to call/send
1793 __ restore_locals();
1795 // Entry point for exceptions thrown within interpreter code
1796 Interpreter::_throw_exception_entry = __ pc();
1797 // expression stack is undefined here
1798 // rax,: exception
1799 // rsi: exception bcp
1800 __ verify_oop(rax);
1802 // expression stack must be empty before entering the VM in case of an exception
1803 __ empty_expression_stack();
1804 __ empty_FPU_stack();
1805 // find exception handler address and preserve exception oop
1806 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1807 // rax,: exception handler entry point
1808 // rdx: preserved exception oop
1809 // rsi: bcp for exception handler
1810 __ push_ptr(rdx); // push exception which is now the only value on the stack
1811 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1813 // If the exception is not handled in the current frame the frame is removed and
1814 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1815 //
1816 // Note: At this point the bci is still the bxi for the instruction which caused
1817 // the exception and the expression stack is empty. Thus, for any VM calls
1818 // at this point, GC will find a legal oop map (with empty expression stack).
1820 // In current activation
1821 // tos: exception
1822 // rsi: exception bcp
1824 //
1825 // JVMTI PopFrame support
1826 //
1828 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1829 __ empty_expression_stack();
1830 __ empty_FPU_stack();
1831 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1832 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1833 // popframe handling cycles.
1834 __ get_thread(thread);
1835 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1836 __ orl(rdx, JavaThread::popframe_processing_bit);
1837 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1839 {
1840 // Check to see whether we are returning to a deoptimized frame.
1841 // (The PopFrame call ensures that the caller of the popped frame is
1842 // either interpreted or compiled and deoptimizes it if compiled.)
1843 // In this case, we can't call dispatch_next() after the frame is
1844 // popped, but instead must save the incoming arguments and restore
1845 // them after deoptimization has occurred.
1846 //
1847 // Note that we don't compare the return PC against the
1848 // deoptimization blob's unpack entry because of the presence of
1849 // adapter frames in C2.
1850 Label caller_not_deoptimized;
1851 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1852 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1853 __ testl(rax, rax);
1854 __ jcc(Assembler::notZero, caller_not_deoptimized);
1856 // Compute size of arguments for saving when returning to deoptimized caller
1857 __ get_method(rax);
1858 __ movptr(rax, Address(rax, Method::const_offset()));
1859 __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset()));
1860 __ shlptr(rax, Interpreter::logStackElementSize);
1861 __ restore_locals();
1862 __ subptr(rdi, rax);
1863 __ addptr(rdi, wordSize);
1864 // Save these arguments
1865 __ get_thread(thread);
1866 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1868 __ remove_activation(vtos, rdx,
1869 /* throw_monitor_exception */ false,
1870 /* install_monitor_exception */ false,
1871 /* notify_jvmdi */ false);
1873 // Inform deoptimization that it is responsible for restoring these arguments
1874 __ get_thread(thread);
1875 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1877 // Continue in deoptimization handler
1878 __ jmp(rdx);
1880 __ bind(caller_not_deoptimized);
1881 }
1883 __ remove_activation(vtos, rdx,
1884 /* throw_monitor_exception */ false,
1885 /* install_monitor_exception */ false,
1886 /* notify_jvmdi */ false);
1888 // Finish with popframe handling
1889 // A previous I2C followed by a deoptimization might have moved the
1890 // outgoing arguments further up the stack. PopFrame expects the
1891 // mutations to those outgoing arguments to be preserved and other
1892 // constraints basically require this frame to look exactly as
1893 // though it had previously invoked an interpreted activation with
1894 // no space between the top of the expression stack (current
1895 // last_sp) and the top of stack. Rather than force deopt to
1896 // maintain this kind of invariant all the time we call a small
1897 // fixup routine to move the mutated arguments onto the top of our
1898 // expression stack if necessary.
1899 __ mov(rax, rsp);
1900 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1901 __ get_thread(thread);
1902 // PC must point into interpreter here
1903 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1904 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1905 __ get_thread(thread);
1906 __ reset_last_Java_frame(thread, true, true);
1907 // Restore the last_sp and null it out
1908 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1909 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1911 __ restore_bcp();
1912 __ restore_locals();
1913 // The method data pointer was incremented already during
1914 // call profiling. We have to restore the mdp for the current bcp.
1915 if (ProfileInterpreter) {
1916 __ set_method_data_pointer_for_bcp();
1917 }
1919 // Clear the popframe condition flag
1920 __ get_thread(thread);
1921 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1923 #if INCLUDE_JVMTI
1924 if (EnableInvokeDynamic) {
1925 Label L_done;
1926 const Register local0 = rdi;
1928 __ cmpb(Address(rsi, 0), Bytecodes::_invokestatic);
1929 __ jcc(Assembler::notEqual, L_done);
1931 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1932 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1934 __ get_method(rdx);
1935 __ movptr(rax, Address(local0, 0));
1936 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rsi);
1938 __ testptr(rax, rax);
1939 __ jcc(Assembler::zero, L_done);
1941 __ movptr(Address(rbx, 0), rax);
1942 __ bind(L_done);
1943 }
1944 #endif // INCLUDE_JVMTI
1946 __ dispatch_next(vtos);
1947 // end of PopFrame support
1949 Interpreter::_remove_activation_entry = __ pc();
1951 // preserve exception over this code sequence
1952 __ pop_ptr(rax);
1953 __ get_thread(thread);
1954 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1955 // remove the activation (without doing throws on illegalMonitorExceptions)
1956 __ remove_activation(vtos, rdx, false, true, false);
1957 // restore exception
1958 __ get_thread(thread);
1959 __ get_vm_result(rax, thread);
1961 // Inbetween activations - previous activation type unknown yet
1962 // compute continuation point - the continuation point expects
1963 // the following registers set up:
1964 //
1965 // rax: exception
1966 // rdx: return address/pc that threw exception
1967 // rsp: expression stack of caller
1968 // rbp: rbp, of caller
1969 __ push(rax); // save exception
1970 __ push(rdx); // save return address
1971 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1972 __ mov(rbx, rax); // save exception handler
1973 __ pop(rdx); // restore return address
1974 __ pop(rax); // restore exception
1975 // Note that an "issuing PC" is actually the next PC after the call
1976 __ jmp(rbx); // jump to exception handler of caller
1977 }
1980 //
1981 // JVMTI ForceEarlyReturn support
1982 //
1983 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1984 address entry = __ pc();
1985 const Register thread = rcx;
1987 __ restore_bcp();
1988 __ restore_locals();
1989 __ empty_expression_stack();
1990 __ empty_FPU_stack();
1991 __ load_earlyret_value(state);
1993 __ get_thread(thread);
1994 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1995 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1997 // Clear the earlyret state
1998 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
2000 __ remove_activation(state, rsi,
2001 false, /* throw_monitor_exception */
2002 false, /* install_monitor_exception */
2003 true); /* notify_jvmdi */
2004 __ jmp(rsi);
2005 return entry;
2006 } // end of ForceEarlyReturn support
2009 //------------------------------------------------------------------------------------------------------------------------
2010 // Helper for vtos entry point generation
2012 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) {
2013 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2014 Label L;
2015 fep = __ pc(); __ push(ftos); __ jmp(L);
2016 dep = __ pc(); __ push(dtos); __ jmp(L);
2017 lep = __ pc(); __ push(ltos); __ jmp(L);
2018 aep = __ pc(); __ push(atos); __ jmp(L);
2019 bep = cep = sep = // fall through
2020 iep = __ pc(); __ push(itos); // fall through
2021 vep = __ pc(); __ bind(L); // fall through
2022 generate_and_dispatch(t);
2023 }
2025 //------------------------------------------------------------------------------------------------------------------------
2026 // Generation of individual instructions
2028 // helpers for generate_and_dispatch
2032 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
2033 : TemplateInterpreterGenerator(code) {
2034 generate_all(); // down here so it can be "virtual"
2035 }
2037 //------------------------------------------------------------------------------------------------------------------------
2039 // Non-product code
2040 #ifndef PRODUCT
2041 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2042 address entry = __ pc();
2044 // prepare expression stack
2045 __ pop(rcx); // pop return address so expression stack is 'pure'
2046 __ push(state); // save tosca
2048 // pass tosca registers as arguments & call tracer
2049 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
2050 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
2051 __ pop(state); // restore tosca
2053 // return
2054 __ jmp(rcx);
2056 return entry;
2057 }
2060 void TemplateInterpreterGenerator::count_bytecode() {
2061 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
2062 }
2065 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2066 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
2067 }
2070 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2071 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
2072 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
2073 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2074 ExternalAddress table((address) BytecodePairHistogram::_counters);
2075 Address index(noreg, rbx, Address::times_4);
2076 __ incrementl(ArrayAddress(table, index));
2077 }
2080 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2081 // Call a little run-time stub to avoid blow-up for each bytecode.
2082 // The run-time runtime saves the right registers, depending on
2083 // the tosca in-state for the given template.
2084 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2085 "entry must have been generated");
2086 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2087 }
2090 void TemplateInterpreterGenerator::stop_interpreter_at() {
2091 Label L;
2092 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
2093 StopInterpreterAt);
2094 __ jcc(Assembler::notEqual, L);
2095 __ int3();
2096 __ bind(L);
2097 }
2098 #endif // !PRODUCT
2099 #endif // CC_INTERP