Wed, 13 Mar 2013 09:44:45 +0100
8009761: Deoptimization on sparc doesn't set Llast_SP correctly in the interpreter frames it creates
Summary: deoptimization doesn't set up callee frames so that they restore caller frames correctly.
Reviewed-by: kvn
1 /*
2 * Copyright (c) 2007, 2012, 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/cppInterpreter.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterGenerator.hpp"
31 #include "interpreter/interpreterRuntime.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/interfaceSupport.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/synchronizer.hpp"
45 #include "runtime/timer.hpp"
46 #include "runtime/vframeArray.hpp"
47 #include "utilities/debug.hpp"
48 #include "utilities/macros.hpp"
49 #ifdef SHARK
50 #include "shark/shark_globals.hpp"
51 #endif
53 #ifdef CC_INTERP
55 // Routine exists to make tracebacks look decent in debugger
56 // while we are recursed in the frame manager/c++ interpreter.
57 // We could use an address in the frame manager but having
58 // frames look natural in the debugger is a plus.
59 extern "C" void RecursiveInterpreterActivation(interpreterState istate )
60 {
61 //
62 ShouldNotReachHere();
63 }
66 #define __ _masm->
67 #define STATE(field_name) (Address(state, byte_offset_of(BytecodeInterpreter, field_name)))
69 Label fast_accessor_slow_entry_path; // fast accessor methods need to be able to jmp to unsynchronized
70 // c++ interpreter entry point this holds that entry point label.
72 // default registers for state and sender_sp
73 // state and sender_sp are the same on 32bit because we have no choice.
74 // state could be rsi on 64bit but it is an arg reg and not callee save
75 // so r13 is better choice.
77 const Register state = NOT_LP64(rsi) LP64_ONLY(r13);
78 const Register sender_sp_on_entry = NOT_LP64(rsi) LP64_ONLY(r13);
80 // NEEDED for JVMTI?
81 // address AbstractInterpreter::_remove_activation_preserving_args_entry;
83 static address unctrap_frame_manager_entry = NULL;
85 static address deopt_frame_manager_return_atos = NULL;
86 static address deopt_frame_manager_return_btos = NULL;
87 static address deopt_frame_manager_return_itos = NULL;
88 static address deopt_frame_manager_return_ltos = NULL;
89 static address deopt_frame_manager_return_ftos = NULL;
90 static address deopt_frame_manager_return_dtos = NULL;
91 static address deopt_frame_manager_return_vtos = NULL;
93 int AbstractInterpreter::BasicType_as_index(BasicType type) {
94 int i = 0;
95 switch (type) {
96 case T_BOOLEAN: i = 0; break;
97 case T_CHAR : i = 1; break;
98 case T_BYTE : i = 2; break;
99 case T_SHORT : i = 3; break;
100 case T_INT : i = 4; break;
101 case T_VOID : i = 5; break;
102 case T_FLOAT : i = 8; break;
103 case T_LONG : i = 9; break;
104 case T_DOUBLE : i = 6; break;
105 case T_OBJECT : // fall through
106 case T_ARRAY : i = 7; break;
107 default : ShouldNotReachHere();
108 }
109 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
110 return i;
111 }
113 // Is this pc anywhere within code owned by the interpreter?
114 // This only works for pc that might possibly be exposed to frame
115 // walkers. It clearly misses all of the actual c++ interpreter
116 // implementation
117 bool CppInterpreter::contains(address pc) {
118 return (_code->contains(pc) ||
119 pc == CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation));
120 }
123 address CppInterpreterGenerator::generate_result_handler_for(BasicType type) {
124 address entry = __ pc();
125 switch (type) {
126 case T_BOOLEAN: __ c2bool(rax); break;
127 case T_CHAR : __ andl(rax, 0xFFFF); break;
128 case T_BYTE : __ sign_extend_byte (rax); break;
129 case T_SHORT : __ sign_extend_short(rax); break;
130 case T_VOID : // fall thru
131 case T_LONG : // fall thru
132 case T_INT : /* nothing to do */ break;
134 case T_DOUBLE :
135 case T_FLOAT :
136 {
137 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
138 __ pop(t); // remove return address first
139 // Must return a result for interpreter or compiler. In SSE
140 // mode, results are returned in xmm0 and the FPU stack must
141 // be empty.
142 if (type == T_FLOAT && UseSSE >= 1) {
143 #ifndef _LP64
144 // Load ST0
145 __ fld_d(Address(rsp, 0));
146 // Store as float and empty fpu stack
147 __ fstp_s(Address(rsp, 0));
148 #endif // !_LP64
149 // and reload
150 __ movflt(xmm0, Address(rsp, 0));
151 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
152 __ movdbl(xmm0, Address(rsp, 0));
153 } else {
154 // restore ST0
155 __ fld_d(Address(rsp, 0));
156 }
157 // and pop the temp
158 __ addptr(rsp, 2 * wordSize);
159 __ push(t); // restore return address
160 }
161 break;
162 case T_OBJECT :
163 // retrieve result from frame
164 __ movptr(rax, STATE(_oop_temp));
165 // and verify it
166 __ verify_oop(rax);
167 break;
168 default : ShouldNotReachHere();
169 }
170 __ ret(0); // return from result handler
171 return entry;
172 }
174 // tosca based result to c++ interpreter stack based result.
175 // Result goes to top of native stack.
177 #undef EXTEND // SHOULD NOT BE NEEDED
178 address CppInterpreterGenerator::generate_tosca_to_stack_converter(BasicType type) {
179 // A result is in the tosca (abi result) from either a native method call or compiled
180 // code. Place this result on the java expression stack so C++ interpreter can use it.
181 address entry = __ pc();
183 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
184 __ pop(t); // remove return address first
185 switch (type) {
186 case T_VOID:
187 break;
188 case T_BOOLEAN:
189 #ifdef EXTEND
190 __ c2bool(rax);
191 #endif
192 __ push(rax);
193 break;
194 case T_CHAR :
195 #ifdef EXTEND
196 __ andl(rax, 0xFFFF);
197 #endif
198 __ push(rax);
199 break;
200 case T_BYTE :
201 #ifdef EXTEND
202 __ sign_extend_byte (rax);
203 #endif
204 __ push(rax);
205 break;
206 case T_SHORT :
207 #ifdef EXTEND
208 __ sign_extend_short(rax);
209 #endif
210 __ push(rax);
211 break;
212 case T_LONG :
213 __ push(rdx); // pushes useless junk on 64bit
214 __ push(rax);
215 break;
216 case T_INT :
217 __ push(rax);
218 break;
219 case T_FLOAT :
220 // Result is in ST(0)/xmm0
221 __ subptr(rsp, wordSize);
222 if ( UseSSE < 1) {
223 __ fstp_s(Address(rsp, 0));
224 } else {
225 __ movflt(Address(rsp, 0), xmm0);
226 }
227 break;
228 case T_DOUBLE :
229 __ subptr(rsp, 2*wordSize);
230 if ( UseSSE < 2 ) {
231 __ fstp_d(Address(rsp, 0));
232 } else {
233 __ movdbl(Address(rsp, 0), xmm0);
234 }
235 break;
236 case T_OBJECT :
237 __ verify_oop(rax); // verify it
238 __ push(rax);
239 break;
240 default : ShouldNotReachHere();
241 }
242 __ jmp(t); // return from result handler
243 return entry;
244 }
246 address CppInterpreterGenerator::generate_stack_to_stack_converter(BasicType type) {
247 // A result is in the java expression stack of the interpreted method that has just
248 // returned. Place this result on the java expression stack of the caller.
249 //
250 // The current interpreter activation in rsi/r13 is for the method just returning its
251 // result. So we know that the result of this method is on the top of the current
252 // execution stack (which is pre-pushed) and will be return to the top of the caller
253 // stack. The top of the callers stack is the bottom of the locals of the current
254 // activation.
255 // Because of the way activation are managed by the frame manager the value of rsp is
256 // below both the stack top of the current activation and naturally the stack top
257 // of the calling activation. This enable this routine to leave the return address
258 // to the frame manager on the stack and do a vanilla return.
259 //
260 // On entry: rsi/r13 - interpreter state of activation returning a (potential) result
261 // On Return: rsi/r13 - unchanged
262 // rax - new stack top for caller activation (i.e. activation in _prev_link)
263 //
264 // Can destroy rdx, rcx.
265 //
267 address entry = __ pc();
268 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
269 switch (type) {
270 case T_VOID:
271 __ movptr(rax, STATE(_locals)); // pop parameters get new stack value
272 __ addptr(rax, wordSize); // account for prepush before we return
273 break;
274 case T_FLOAT :
275 case T_BOOLEAN:
276 case T_CHAR :
277 case T_BYTE :
278 case T_SHORT :
279 case T_INT :
280 // 1 word result
281 __ movptr(rdx, STATE(_stack));
282 __ movptr(rax, STATE(_locals)); // address for result
283 __ movl(rdx, Address(rdx, wordSize)); // get result
284 __ movptr(Address(rax, 0), rdx); // and store it
285 break;
286 case T_LONG :
287 case T_DOUBLE :
288 // return top two words on current expression stack to caller's expression stack
289 // The caller's expression stack is adjacent to the current frame manager's intepretState
290 // except we allocated one extra word for this intepretState so we won't overwrite it
291 // when we return a two word result.
293 __ movptr(rax, STATE(_locals)); // address for result
294 __ movptr(rcx, STATE(_stack));
295 __ subptr(rax, wordSize); // need addition word besides locals[0]
296 __ movptr(rdx, Address(rcx, 2*wordSize)); // get result word (junk in 64bit)
297 __ movptr(Address(rax, wordSize), rdx); // and store it
298 __ movptr(rdx, Address(rcx, wordSize)); // get result word
299 __ movptr(Address(rax, 0), rdx); // and store it
300 break;
301 case T_OBJECT :
302 __ movptr(rdx, STATE(_stack));
303 __ movptr(rax, STATE(_locals)); // address for result
304 __ movptr(rdx, Address(rdx, wordSize)); // get result
305 __ verify_oop(rdx); // verify it
306 __ movptr(Address(rax, 0), rdx); // and store it
307 break;
308 default : ShouldNotReachHere();
309 }
310 __ ret(0);
311 return entry;
312 }
314 address CppInterpreterGenerator::generate_stack_to_native_abi_converter(BasicType type) {
315 // A result is in the java expression stack of the interpreted method that has just
316 // returned. Place this result in the native abi that the caller expects.
317 //
318 // Similar to generate_stack_to_stack_converter above. Called at a similar time from the
319 // frame manager execept in this situation the caller is native code (c1/c2/call_stub)
320 // and so rather than return result onto caller's java expression stack we return the
321 // result in the expected location based on the native abi.
322 // On entry: rsi/r13 - interpreter state of activation returning a (potential) result
323 // On Return: rsi/r13 - unchanged
324 // Other registers changed [rax/rdx/ST(0) as needed for the result returned]
326 address entry = __ pc();
327 switch (type) {
328 case T_VOID:
329 break;
330 case T_BOOLEAN:
331 case T_CHAR :
332 case T_BYTE :
333 case T_SHORT :
334 case T_INT :
335 __ movptr(rdx, STATE(_stack)); // get top of stack
336 __ movl(rax, Address(rdx, wordSize)); // get result word 1
337 break;
338 case T_LONG :
339 __ movptr(rdx, STATE(_stack)); // get top of stack
340 __ movptr(rax, Address(rdx, wordSize)); // get result low word
341 NOT_LP64(__ movl(rdx, Address(rdx, 2*wordSize));) // get result high word
342 break;
343 case T_FLOAT :
344 __ movptr(rdx, STATE(_stack)); // get top of stack
345 if ( UseSSE >= 1) {
346 __ movflt(xmm0, Address(rdx, wordSize));
347 } else {
348 __ fld_s(Address(rdx, wordSize)); // pushd float result
349 }
350 break;
351 case T_DOUBLE :
352 __ movptr(rdx, STATE(_stack)); // get top of stack
353 if ( UseSSE > 1) {
354 __ movdbl(xmm0, Address(rdx, wordSize));
355 } else {
356 __ fld_d(Address(rdx, wordSize)); // push double result
357 }
358 break;
359 case T_OBJECT :
360 __ movptr(rdx, STATE(_stack)); // get top of stack
361 __ movptr(rax, Address(rdx, wordSize)); // get result word 1
362 __ verify_oop(rax); // verify it
363 break;
364 default : ShouldNotReachHere();
365 }
366 __ ret(0);
367 return entry;
368 }
370 address CppInterpreter::return_entry(TosState state, int length) {
371 // make it look good in the debugger
372 return CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation);
373 }
375 address CppInterpreter::deopt_entry(TosState state, int length) {
376 address ret = NULL;
377 if (length != 0) {
378 switch (state) {
379 case atos: ret = deopt_frame_manager_return_atos; break;
380 case btos: ret = deopt_frame_manager_return_btos; break;
381 case ctos:
382 case stos:
383 case itos: ret = deopt_frame_manager_return_itos; break;
384 case ltos: ret = deopt_frame_manager_return_ltos; break;
385 case ftos: ret = deopt_frame_manager_return_ftos; break;
386 case dtos: ret = deopt_frame_manager_return_dtos; break;
387 case vtos: ret = deopt_frame_manager_return_vtos; break;
388 }
389 } else {
390 ret = unctrap_frame_manager_entry; // re-execute the bytecode ( e.g. uncommon trap)
391 }
392 assert(ret != NULL, "Not initialized");
393 return ret;
394 }
396 // C++ Interpreter
397 void CppInterpreterGenerator::generate_compute_interpreter_state(const Register state,
398 const Register locals,
399 const Register sender_sp,
400 bool native) {
402 // On entry the "locals" argument points to locals[0] (or where it would be in case no locals in
403 // a static method). "state" contains any previous frame manager state which we must save a link
404 // to in the newly generated state object. On return "state" is a pointer to the newly allocated
405 // state object. We must allocate and initialize a new interpretState object and the method
406 // expression stack. Because the returned result (if any) of the method will be placed on the caller's
407 // expression stack and this will overlap with locals[0] (and locals[1] if double/long) we must
408 // be sure to leave space on the caller's stack so that this result will not overwrite values when
409 // locals[0] and locals[1] do not exist (and in fact are return address and saved rbp). So when
410 // we are non-native we in essence ensure that locals[0-1] exist. We play an extra trick in
411 // non-product builds and initialize this last local with the previous interpreterState as
412 // this makes things look real nice in the debugger.
414 // State on entry
415 // Assumes locals == &locals[0]
416 // Assumes state == any previous frame manager state (assuming call path from c++ interpreter)
417 // Assumes rax = return address
418 // rcx == senders_sp
419 // rbx == method
420 // Modifies rcx, rdx, rax
421 // Returns:
422 // state == address of new interpreterState
423 // rsp == bottom of method's expression stack.
425 const Address const_offset (rbx, Method::const_offset());
428 // On entry sp is the sender's sp. This includes the space for the arguments
429 // that the sender pushed. If the sender pushed no args (a static) and the
430 // caller returns a long then we need two words on the sender's stack which
431 // are not present (although when we return a restore full size stack the
432 // space will be present). If we didn't allocate two words here then when
433 // we "push" the result of the caller's stack we would overwrite the return
434 // address and the saved rbp. Not good. So simply allocate 2 words now
435 // just to be safe. This is the "static long no_params() method" issue.
436 // See Lo.java for a testcase.
437 // We don't need this for native calls because they return result in
438 // register and the stack is expanded in the caller before we store
439 // the results on the stack.
441 if (!native) {
442 #ifdef PRODUCT
443 __ subptr(rsp, 2*wordSize);
444 #else /* PRODUCT */
445 __ push((int32_t)NULL_WORD);
446 __ push(state); // make it look like a real argument
447 #endif /* PRODUCT */
448 }
450 // Now that we are assure of space for stack result, setup typical linkage
452 __ push(rax);
453 __ enter();
455 __ mov(rax, state); // save current state
457 __ lea(rsp, Address(rsp, -(int)sizeof(BytecodeInterpreter)));
458 __ mov(state, rsp);
460 // rsi/r13 == state/locals rax == prevstate
462 // initialize the "shadow" frame so that use since C++ interpreter not directly
463 // recursive. Simpler to recurse but we can't trim expression stack as we call
464 // new methods.
465 __ movptr(STATE(_locals), locals); // state->_locals = locals()
466 __ movptr(STATE(_self_link), state); // point to self
467 __ movptr(STATE(_prev_link), rax); // state->_link = state on entry (NULL or previous state)
468 __ movptr(STATE(_sender_sp), sender_sp); // state->_sender_sp = sender_sp
469 #ifdef _LP64
470 __ movptr(STATE(_thread), r15_thread); // state->_bcp = codes()
471 #else
472 __ get_thread(rax); // get vm's javathread*
473 __ movptr(STATE(_thread), rax); // state->_bcp = codes()
474 #endif // _LP64
475 __ movptr(rdx, Address(rbx, Method::const_offset())); // get constantMethodOop
476 __ lea(rdx, Address(rdx, ConstMethod::codes_offset())); // get code base
477 if (native) {
478 __ movptr(STATE(_bcp), (int32_t)NULL_WORD); // state->_bcp = NULL
479 } else {
480 __ movptr(STATE(_bcp), rdx); // state->_bcp = codes()
481 }
482 __ xorptr(rdx, rdx);
483 __ movptr(STATE(_oop_temp), rdx); // state->_oop_temp = NULL (only really needed for native)
484 __ movptr(STATE(_mdx), rdx); // state->_mdx = NULL
485 __ movptr(rdx, Address(rbx, Method::const_offset()));
486 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
487 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
488 __ movptr(STATE(_constants), rdx); // state->_constants = constants()
490 __ movptr(STATE(_method), rbx); // state->_method = method()
491 __ movl(STATE(_msg), (int32_t) BytecodeInterpreter::method_entry); // state->_msg = initial method entry
492 __ movptr(STATE(_result._to_call._callee), (int32_t) NULL_WORD); // state->_result._to_call._callee_callee = NULL
495 __ movptr(STATE(_monitor_base), rsp); // set monitor block bottom (grows down) this would point to entry [0]
496 // entries run from -1..x where &monitor[x] ==
498 {
499 // Must not attempt to lock method until we enter interpreter as gc won't be able to find the
500 // initial frame. However we allocate a free monitor so we don't have to shuffle the expression stack
501 // immediately.
503 // synchronize method
504 const Address access_flags (rbx, Method::access_flags_offset());
505 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
506 Label not_synced;
508 __ movl(rax, access_flags);
509 __ testl(rax, JVM_ACC_SYNCHRONIZED);
510 __ jcc(Assembler::zero, not_synced);
512 // Allocate initial monitor and pre initialize it
513 // get synchronization object
515 Label done;
516 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
517 __ movl(rax, access_flags);
518 __ testl(rax, JVM_ACC_STATIC);
519 __ movptr(rax, Address(locals, 0)); // get receiver (assume this is frequent case)
520 __ jcc(Assembler::zero, done);
521 __ movptr(rax, Address(rbx, Method::const_offset()));
522 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
523 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
524 __ movptr(rax, Address(rax, mirror_offset));
525 __ bind(done);
526 // add space for monitor & lock
527 __ subptr(rsp, entry_size); // add space for a monitor entry
528 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
529 __ bind(not_synced);
530 }
532 __ movptr(STATE(_stack_base), rsp); // set expression stack base ( == &monitors[-count])
533 if (native) {
534 __ movptr(STATE(_stack), rsp); // set current expression stack tos
535 __ movptr(STATE(_stack_limit), rsp);
536 } else {
537 __ subptr(rsp, wordSize); // pre-push stack
538 __ movptr(STATE(_stack), rsp); // set current expression stack tos
540 // compute full expression stack limit
542 const int extra_stack = 0; //6815692//Method::extra_stack_words();
543 __ movptr(rdx, Address(rbx, Method::const_offset()));
544 __ load_unsigned_short(rdx, Address(rdx, ConstMethod::max_stack_offset())); // get size of expression stack in words
545 __ negptr(rdx); // so we can subtract in next step
546 // Allocate expression stack
547 __ lea(rsp, Address(rsp, rdx, Address::times_ptr, -extra_stack));
548 __ movptr(STATE(_stack_limit), rsp);
549 }
551 #ifdef _LP64
552 // Make sure stack is properly aligned and sized for the abi
553 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
554 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
555 #endif // _LP64
559 }
561 // Helpers for commoning out cases in the various type of method entries.
562 //
564 // increment invocation count & check for overflow
565 //
566 // Note: checking for negative value instead of overflow
567 // so we have a 'sticky' overflow test
568 //
569 // rbx,: method
570 // rcx: invocation counter
571 //
572 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
574 const Address invocation_counter(rbx, Method::invocation_counter_offset() + InvocationCounter::counter_offset());
575 const Address backedge_counter (rbx, Method::backedge_counter_offset() + InvocationCounter::counter_offset());
577 if (ProfileInterpreter) { // %%% Merge this into MethodData*
578 __ incrementl(Address(rbx,Method::interpreter_invocation_counter_offset()));
579 }
580 // Update standard invocation counters
581 __ movl(rax, backedge_counter); // load backedge counter
583 __ increment(rcx, InvocationCounter::count_increment);
584 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
586 __ movl(invocation_counter, rcx); // save invocation count
587 __ addl(rcx, rax); // add both counters
589 // profile_method is non-null only for interpreted method so
590 // profile_method != NULL == !native_call
591 // BytecodeInterpreter only calls for native so code is elided.
593 __ cmp32(rcx,
594 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
595 __ jcc(Assembler::aboveEqual, *overflow);
597 }
599 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
601 // C++ interpreter on entry
602 // rsi/r13 - new interpreter state pointer
603 // rbp - interpreter frame pointer
604 // rbx - method
606 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
607 // rbx, - method
608 // rcx - rcvr (assuming there is one)
609 // top of stack return address of interpreter caller
610 // rsp - sender_sp
612 // C++ interpreter only
613 // rsi/r13 - previous interpreter state pointer
615 // InterpreterRuntime::frequency_counter_overflow takes one argument
616 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
617 // The call returns the address of the verified entry point for the method or NULL
618 // if the compilation did not complete (either went background or bailed out).
619 __ movptr(rax, (int32_t)false);
620 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
622 // for c++ interpreter can rsi really be munged?
623 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); // restore state
624 __ movptr(rbx, Address(state, byte_offset_of(BytecodeInterpreter, _method))); // restore method
625 __ movptr(rdi, Address(state, byte_offset_of(BytecodeInterpreter, _locals))); // get locals pointer
627 __ jmp(*do_continue, relocInfo::none);
629 }
631 void InterpreterGenerator::generate_stack_overflow_check(void) {
632 // see if we've got enough room on the stack for locals plus overhead.
633 // the expression stack grows down incrementally, so the normal guard
634 // page mechanism will work for that.
635 //
636 // Registers live on entry:
637 //
638 // Asm interpreter
639 // rdx: number of additional locals this frame needs (what we must check)
640 // rbx,: Method*
642 // C++ Interpreter
643 // rsi/r13: previous interpreter frame state object
644 // rdi: &locals[0]
645 // rcx: # of locals
646 // rdx: number of additional locals this frame needs (what we must check)
647 // rbx: Method*
649 // destroyed on exit
650 // rax,
652 // NOTE: since the additional locals are also always pushed (wasn't obvious in
653 // generate_method_entry) so the guard should work for them too.
654 //
656 // monitor entry size: see picture of stack set (generate_method_entry) and frame_i486.hpp
657 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
659 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
660 // be sure to change this if you add/subtract anything to/from the overhead area
661 const int overhead_size = (int)sizeof(BytecodeInterpreter);
663 const int page_size = os::vm_page_size();
665 Label after_frame_check;
667 // compute rsp as if this were going to be the last frame on
668 // the stack before the red zone
670 Label after_frame_check_pop;
672 // save rsi == caller's bytecode ptr (c++ previous interp. state)
673 // QQQ problem here?? rsi overload????
674 __ push(state);
676 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rsi);
678 NOT_LP64(__ get_thread(thread));
680 const Address stack_base(thread, Thread::stack_base_offset());
681 const Address stack_size(thread, Thread::stack_size_offset());
683 // locals + overhead, in bytes
684 // Always give one monitor to allow us to start interp if sync method.
685 // Any additional monitors need a check when moving the expression stack
686 const int one_monitor = frame::interpreter_frame_monitor_size() * wordSize;
687 const int extra_stack = 0; //6815692//Method::extra_stack_entries();
688 __ movptr(rax, Address(rbx, Method::const_offset()));
689 __ load_unsigned_short(rax, Address(rax, ConstMethod::max_stack_offset())); // get size of expression stack in words
690 __ lea(rax, Address(noreg, rax, Interpreter::stackElementScale(), extra_stack + one_monitor));
691 __ lea(rax, Address(rax, rdx, Interpreter::stackElementScale(), overhead_size));
693 #ifdef ASSERT
694 Label stack_base_okay, stack_size_okay;
695 // verify that thread stack base is non-zero
696 __ cmpptr(stack_base, (int32_t)0);
697 __ jcc(Assembler::notEqual, stack_base_okay);
698 __ stop("stack base is zero");
699 __ bind(stack_base_okay);
700 // verify that thread stack size is non-zero
701 __ cmpptr(stack_size, (int32_t)0);
702 __ jcc(Assembler::notEqual, stack_size_okay);
703 __ stop("stack size is zero");
704 __ bind(stack_size_okay);
705 #endif
707 // Add stack base to locals and subtract stack size
708 __ addptr(rax, stack_base);
709 __ subptr(rax, stack_size);
711 // We should have a magic number here for the size of the c++ interpreter frame.
712 // We can't actually tell this ahead of time. The debug version size is around 3k
713 // product is 1k and fastdebug is 4k
714 const int slop = 6 * K;
716 // Use the maximum number of pages we might bang.
717 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
718 (StackRedPages+StackYellowPages);
719 // Only need this if we are stack banging which is temporary while
720 // we're debugging.
721 __ addptr(rax, slop + 2*max_pages * page_size);
723 // check against the current stack bottom
724 __ cmpptr(rsp, rax);
725 __ jcc(Assembler::above, after_frame_check_pop);
727 __ pop(state); // get c++ prev state.
729 // throw exception return address becomes throwing pc
730 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
732 // all done with frame size check
733 __ bind(after_frame_check_pop);
734 __ pop(state);
736 __ bind(after_frame_check);
737 }
739 // Find preallocated monitor and lock method (C++ interpreter)
740 // rbx - Method*
741 //
742 void InterpreterGenerator::lock_method(void) {
743 // assumes state == rsi/r13 == pointer to current interpreterState
744 // minimally destroys rax, rdx|c_rarg1, rdi
745 //
746 // synchronize method
747 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
748 const Address access_flags (rbx, Method::access_flags_offset());
750 const Register monitor = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
752 // find initial monitor i.e. monitors[-1]
753 __ movptr(monitor, STATE(_monitor_base)); // get monitor bottom limit
754 __ subptr(monitor, entry_size); // point to initial monitor
756 #ifdef ASSERT
757 { Label L;
758 __ movl(rax, access_flags);
759 __ testl(rax, JVM_ACC_SYNCHRONIZED);
760 __ jcc(Assembler::notZero, L);
761 __ stop("method doesn't need synchronization");
762 __ bind(L);
763 }
764 #endif // ASSERT
765 // get synchronization object
766 { Label done;
767 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
768 __ movl(rax, access_flags);
769 __ movptr(rdi, STATE(_locals)); // prepare to get receiver (assume common case)
770 __ testl(rax, JVM_ACC_STATIC);
771 __ movptr(rax, Address(rdi, 0)); // get receiver (assume this is frequent case)
772 __ jcc(Assembler::zero, done);
773 __ movptr(rax, Address(rbx, Method::const_offset()));
774 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
775 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
776 __ movptr(rax, Address(rax, mirror_offset));
777 __ bind(done);
778 }
779 #ifdef ASSERT
780 { Label L;
781 __ cmpptr(rax, Address(monitor, BasicObjectLock::obj_offset_in_bytes())); // correct object?
782 __ jcc(Assembler::equal, L);
783 __ stop("wrong synchronization lobject");
784 __ bind(L);
785 }
786 #endif // ASSERT
787 // can destroy rax, rdx|c_rarg1, rcx, and (via call_VM) rdi!
788 __ lock_object(monitor);
789 }
791 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
793 address InterpreterGenerator::generate_accessor_entry(void) {
795 // rbx: Method*
797 // rsi/r13: senderSP must preserved for slow path, set SP to it on fast path
799 Label xreturn_path;
801 // do fastpath for resolved accessor methods
802 if (UseFastAccessorMethods) {
804 address entry_point = __ pc();
806 Label slow_path;
807 // If we need a safepoint check, generate full interpreter entry.
808 ExternalAddress state(SafepointSynchronize::address_of_state());
809 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
810 SafepointSynchronize::_not_synchronized);
812 __ jcc(Assembler::notEqual, slow_path);
813 // ASM/C++ Interpreter
814 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
815 // Note: We can only use this code if the getfield has been resolved
816 // and if we don't have a null-pointer exception => check for
817 // these conditions first and use slow path if necessary.
818 // rbx,: method
819 // rcx: receiver
820 __ movptr(rax, Address(rsp, wordSize));
822 // check if local 0 != NULL and read field
823 __ testptr(rax, rax);
824 __ jcc(Assembler::zero, slow_path);
826 // read first instruction word and extract bytecode @ 1 and index @ 2
827 __ movptr(rdx, Address(rbx, Method::const_offset()));
828 __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
829 __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
830 // Shift codes right to get the index on the right.
831 // The bytecode fetched looks like <index><0xb4><0x2a>
832 __ shrl(rdx, 2*BitsPerByte);
833 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
834 __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
836 // rax,: local 0
837 // rbx,: method
838 // rcx: receiver - do not destroy since it is needed for slow path!
839 // rcx: scratch
840 // rdx: constant pool cache index
841 // rdi: constant pool cache
842 // rsi/r13: sender sp
844 // check if getfield has been resolved and read constant pool cache entry
845 // check the validity of the cache entry by testing whether _indices field
846 // contains Bytecode::_getfield in b1 byte.
847 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
848 __ movl(rcx,
849 Address(rdi,
850 rdx,
851 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
852 __ shrl(rcx, 2*BitsPerByte);
853 __ andl(rcx, 0xFF);
854 __ cmpl(rcx, Bytecodes::_getfield);
855 __ jcc(Assembler::notEqual, slow_path);
857 // Note: constant pool entry is not valid before bytecode is resolved
858 __ movptr(rcx,
859 Address(rdi,
860 rdx,
861 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
862 __ movl(rdx,
863 Address(rdi,
864 rdx,
865 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
867 Label notByte, notShort, notChar;
868 const Address field_address (rax, rcx, Address::times_1);
870 // Need to differentiate between igetfield, agetfield, bgetfield etc.
871 // because they are different sizes.
872 // Use the type from the constant pool cache
873 __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
874 // Make sure we don't need to mask rdx after the above shift
875 ConstantPoolCacheEntry::verify_tos_state_shift();
876 #ifdef _LP64
877 Label notObj;
878 __ cmpl(rdx, atos);
879 __ jcc(Assembler::notEqual, notObj);
880 // atos
881 __ movptr(rax, field_address);
882 __ jmp(xreturn_path);
884 __ bind(notObj);
885 #endif // _LP64
886 __ cmpl(rdx, btos);
887 __ jcc(Assembler::notEqual, notByte);
888 __ load_signed_byte(rax, field_address);
889 __ jmp(xreturn_path);
891 __ bind(notByte);
892 __ cmpl(rdx, stos);
893 __ jcc(Assembler::notEqual, notShort);
894 __ load_signed_short(rax, field_address);
895 __ jmp(xreturn_path);
897 __ bind(notShort);
898 __ cmpl(rdx, ctos);
899 __ jcc(Assembler::notEqual, notChar);
900 __ load_unsigned_short(rax, field_address);
901 __ jmp(xreturn_path);
903 __ bind(notChar);
904 #ifdef ASSERT
905 Label okay;
906 #ifndef _LP64
907 __ cmpl(rdx, atos);
908 __ jcc(Assembler::equal, okay);
909 #endif // _LP64
910 __ cmpl(rdx, itos);
911 __ jcc(Assembler::equal, okay);
912 __ stop("what type is this?");
913 __ bind(okay);
914 #endif // ASSERT
915 // All the rest are a 32 bit wordsize
916 __ movl(rax, field_address);
918 __ bind(xreturn_path);
920 // _ireturn/_areturn
921 __ pop(rdi); // get return address
922 __ mov(rsp, sender_sp_on_entry); // set sp to sender sp
923 __ jmp(rdi);
925 // generate a vanilla interpreter entry as the slow path
926 __ bind(slow_path);
927 // We will enter c++ interpreter looking like it was
928 // called by the call_stub this will cause it to return
929 // a tosca result to the invoker which might have been
930 // the c++ interpreter itself.
932 __ jmp(fast_accessor_slow_entry_path);
933 return entry_point;
935 } else {
936 return NULL;
937 }
939 }
941 address InterpreterGenerator::generate_Reference_get_entry(void) {
942 #if INCLUDE_ALL_GCS
943 if (UseG1GC) {
944 // We need to generate have a routine that generates code to:
945 // * load the value in the referent field
946 // * passes that value to the pre-barrier.
947 //
948 // In the case of G1 this will record the value of the
949 // referent in an SATB buffer if marking is active.
950 // This will cause concurrent marking to mark the referent
951 // field as live.
952 Unimplemented();
953 }
954 #endif // INCLUDE_ALL_GCS
956 // If G1 is not enabled then attempt to go through the accessor entry point
957 // Reference.get is an accessor
958 return generate_accessor_entry();
959 }
961 //
962 // C++ Interpreter stub for calling a native method.
963 // This sets up a somewhat different looking stack for calling the native method
964 // than the typical interpreter frame setup but still has the pointer to
965 // an interpreter state.
966 //
968 address InterpreterGenerator::generate_native_entry(bool synchronized) {
969 // determine code generation flags
970 bool inc_counter = UseCompiler || CountCompiledCalls;
972 // rbx: Method*
973 // rcx: receiver (unused)
974 // rsi/r13: previous interpreter state (if called from C++ interpreter) must preserve
975 // in any case. If called via c1/c2/call_stub rsi/r13 is junk (to use) but harmless
976 // to save/restore.
977 address entry_point = __ pc();
979 const Address constMethod (rbx, Method::const_offset());
980 const Address invocation_counter(rbx, Method::invocation_counter_offset() + InvocationCounter::counter_offset());
981 const Address access_flags (rbx, Method::access_flags_offset());
982 const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
984 // rsi/r13 == state/locals rdi == prevstate
985 const Register locals = rdi;
987 // get parameter size (always needed)
988 __ movptr(rcx, constMethod);
989 __ load_unsigned_short(rcx, size_of_parameters);
991 // rbx: Method*
992 // rcx: size of parameters
993 __ pop(rax); // get return address
994 // for natives the size of locals is zero
996 // compute beginning of parameters /locals
998 __ lea(locals, Address(rsp, rcx, Address::times_ptr, -wordSize));
1000 // initialize fixed part of activation frame
1002 // Assumes rax = return address
1004 // allocate and initialize new interpreterState and method expression stack
1005 // IN(locals) -> locals
1006 // IN(state) -> previous frame manager state (NULL from stub/c1/c2)
1007 // destroys rax, rcx, rdx
1008 // OUT (state) -> new interpreterState
1009 // OUT(rsp) -> bottom of methods expression stack
1011 // save sender_sp
1012 __ mov(rcx, sender_sp_on_entry);
1013 // start with NULL previous state
1014 __ movptr(state, (int32_t)NULL_WORD);
1015 generate_compute_interpreter_state(state, locals, rcx, true);
1017 #ifdef ASSERT
1018 { Label L;
1019 __ movptr(rax, STATE(_stack_base));
1020 #ifdef _LP64
1021 // duplicate the alignment rsp got after setting stack_base
1022 __ subptr(rax, frame::arg_reg_save_area_bytes); // windows
1023 __ andptr(rax, -16); // must be 16 byte boundary (see amd64 ABI)
1024 #endif // _LP64
1025 __ cmpptr(rax, rsp);
1026 __ jcc(Assembler::equal, L);
1027 __ stop("broken stack frame setup in interpreter");
1028 __ bind(L);
1029 }
1030 #endif
1032 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
1034 const Register unlock_thread = LP64_ONLY(r15_thread) NOT_LP64(rax);
1035 NOT_LP64(__ movptr(unlock_thread, STATE(_thread));) // get thread
1036 // Since at this point in the method invocation the exception handler
1037 // would try to exit the monitor of synchronized methods which hasn't
1038 // been entered yet, we set the thread local variable
1039 // _do_not_unlock_if_synchronized to true. The remove_activation will
1040 // check this flag.
1042 const Address do_not_unlock_if_synchronized(unlock_thread,
1043 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1044 __ movbool(do_not_unlock_if_synchronized, true);
1046 // make sure method is native & not abstract
1047 #ifdef ASSERT
1048 __ movl(rax, access_flags);
1049 {
1050 Label L;
1051 __ testl(rax, JVM_ACC_NATIVE);
1052 __ jcc(Assembler::notZero, L);
1053 __ stop("tried to execute non-native method as native");
1054 __ bind(L);
1055 }
1056 { Label L;
1057 __ testl(rax, JVM_ACC_ABSTRACT);
1058 __ jcc(Assembler::zero, L);
1059 __ stop("tried to execute abstract method in interpreter");
1060 __ bind(L);
1061 }
1062 #endif
1065 // increment invocation count & check for overflow
1066 Label invocation_counter_overflow;
1067 if (inc_counter) {
1068 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1069 }
1071 Label continue_after_compile;
1073 __ bind(continue_after_compile);
1075 bang_stack_shadow_pages(true);
1077 // reset the _do_not_unlock_if_synchronized flag
1078 NOT_LP64(__ movl(rax, STATE(_thread));) // get thread
1079 __ movbool(do_not_unlock_if_synchronized, false);
1082 // check for synchronized native methods
1083 //
1084 // Note: This must happen *after* invocation counter check, since
1085 // when overflow happens, the method should not be locked.
1086 if (synchronized) {
1087 // potentially kills rax, rcx, rdx, rdi
1088 lock_method();
1089 } else {
1090 // no synchronization necessary
1091 #ifdef ASSERT
1092 { Label L;
1093 __ movl(rax, access_flags);
1094 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1095 __ jcc(Assembler::zero, L);
1096 __ stop("method needs synchronization");
1097 __ bind(L);
1098 }
1099 #endif
1100 }
1102 // start execution
1104 // jvmti support
1105 __ notify_method_entry();
1107 // work registers
1108 const Register method = rbx;
1109 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rdi);
1110 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); // rcx|rscratch1
1111 const Address constMethod (method, Method::const_offset());
1112 const Address size_of_parameters(t, ConstMethod::size_of_parameters_offset());
1114 // allocate space for parameters
1115 __ movptr(method, STATE(_method));
1116 __ verify_method_ptr(method);
1117 __ movptr(t, constMethod);
1118 __ load_unsigned_short(t, size_of_parameters);
1119 __ shll(t, 2);
1120 #ifdef _LP64
1121 __ subptr(rsp, t);
1122 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1123 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
1124 #else
1125 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
1126 __ subptr(rsp, t);
1127 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
1128 #endif // _LP64
1130 // get signature handler
1131 Label pending_exception_present;
1133 { Label L;
1134 __ movptr(t, Address(method, Method::signature_handler_offset()));
1135 __ testptr(t, t);
1136 __ jcc(Assembler::notZero, L);
1137 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method, false);
1138 __ movptr(method, STATE(_method));
1139 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1140 __ jcc(Assembler::notEqual, pending_exception_present);
1141 __ verify_method_ptr(method);
1142 __ movptr(t, Address(method, Method::signature_handler_offset()));
1143 __ bind(L);
1144 }
1145 #ifdef ASSERT
1146 {
1147 Label L;
1148 __ push(t);
1149 __ get_thread(t); // get vm's javathread*
1150 __ cmpptr(t, STATE(_thread));
1151 __ jcc(Assembler::equal, L);
1152 __ int3();
1153 __ bind(L);
1154 __ pop(t);
1155 }
1156 #endif //
1158 const Register from_ptr = InterpreterRuntime::SignatureHandlerGenerator::from();
1159 // call signature handler
1160 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
1162 // The generated handlers do not touch RBX (the method oop).
1163 // However, large signatures cannot be cached and are generated
1164 // each time here. The slow-path generator will blow RBX
1165 // sometime, so we must reload it after the call.
1166 __ movptr(from_ptr, STATE(_locals)); // get the from pointer
1167 __ call(t);
1168 __ movptr(method, STATE(_method));
1169 __ verify_method_ptr(method);
1171 // result handler is in rax
1172 // set result handler
1173 __ movptr(STATE(_result_handler), rax);
1176 // get native function entry point
1177 { Label L;
1178 __ movptr(rax, Address(method, Method::native_function_offset()));
1179 __ testptr(rax, rax);
1180 __ jcc(Assembler::notZero, L);
1181 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1182 __ movptr(method, STATE(_method));
1183 __ verify_method_ptr(method);
1184 __ movptr(rax, Address(method, Method::native_function_offset()));
1185 __ bind(L);
1186 }
1188 // pass mirror handle if static call
1189 { Label L;
1190 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1191 __ movl(t, Address(method, Method::access_flags_offset()));
1192 __ testl(t, JVM_ACC_STATIC);
1193 __ jcc(Assembler::zero, L);
1194 // get mirror
1195 __ movptr(t, Address(method, Method:: const_offset()));
1196 __ movptr(t, Address(t, ConstMethod::constants_offset()));
1197 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1198 __ movptr(t, Address(t, mirror_offset));
1199 // copy mirror into activation object
1200 __ movptr(STATE(_oop_temp), t);
1201 // pass handle to mirror
1202 #ifdef _LP64
1203 __ lea(c_rarg1, STATE(_oop_temp));
1204 #else
1205 __ lea(t, STATE(_oop_temp));
1206 __ movptr(Address(rsp, wordSize), t);
1207 #endif // _LP64
1208 __ bind(L);
1209 }
1210 #ifdef ASSERT
1211 {
1212 Label L;
1213 __ push(t);
1214 __ get_thread(t); // get vm's javathread*
1215 __ cmpptr(t, STATE(_thread));
1216 __ jcc(Assembler::equal, L);
1217 __ int3();
1218 __ bind(L);
1219 __ pop(t);
1220 }
1221 #endif //
1223 // pass JNIEnv
1224 #ifdef _LP64
1225 __ lea(c_rarg0, Address(thread, JavaThread::jni_environment_offset()));
1226 #else
1227 __ movptr(thread, STATE(_thread)); // get thread
1228 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1230 __ movptr(Address(rsp, 0), t);
1231 #endif // _LP64
1233 #ifdef ASSERT
1234 {
1235 Label L;
1236 __ push(t);
1237 __ get_thread(t); // get vm's javathread*
1238 __ cmpptr(t, STATE(_thread));
1239 __ jcc(Assembler::equal, L);
1240 __ int3();
1241 __ bind(L);
1242 __ pop(t);
1243 }
1244 #endif //
1246 #ifdef ASSERT
1247 { Label L;
1248 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1249 __ cmpl(t, _thread_in_Java);
1250 __ jcc(Assembler::equal, L);
1251 __ stop("Wrong thread state in native stub");
1252 __ bind(L);
1253 }
1254 #endif
1256 // Change state to native (we save the return address in the thread, since it might not
1257 // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
1258 // points into the right code segment. It does not have to be the correct return pc.
1260 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1262 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1264 __ call(rax);
1266 // result potentially in rdx:rax or ST0
1267 __ movptr(method, STATE(_method));
1268 NOT_LP64(__ movptr(thread, STATE(_thread));) // get thread
1270 // The potential result is in ST(0) & rdx:rax
1271 // With C++ interpreter we leave any possible result in ST(0) until we are in result handler and then
1272 // we do the appropriate stuff for returning the result. rdx:rax must always be saved because just about
1273 // anything we do here will destroy it, st(0) is only saved if we re-enter the vm where it would
1274 // be destroyed.
1275 // It is safe to do these pushes because state is _thread_in_native and return address will be found
1276 // via _last_native_pc and not via _last_jave_sp
1278 // Must save the value of ST(0)/xmm0 since it could be destroyed before we get to result handler
1279 { Label Lpush, Lskip;
1280 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1281 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1282 __ cmpptr(STATE(_result_handler), float_handler.addr());
1283 __ jcc(Assembler::equal, Lpush);
1284 __ cmpptr(STATE(_result_handler), double_handler.addr());
1285 __ jcc(Assembler::notEqual, Lskip);
1286 __ bind(Lpush);
1287 __ subptr(rsp, 2*wordSize);
1288 if ( UseSSE < 2 ) {
1289 __ fstp_d(Address(rsp, 0));
1290 } else {
1291 __ movdbl(Address(rsp, 0), xmm0);
1292 }
1293 __ bind(Lskip);
1294 }
1296 // save rax:rdx for potential use by result handler.
1297 __ push(rax);
1298 #ifndef _LP64
1299 __ push(rdx);
1300 #endif // _LP64
1302 // Either restore the MXCSR register after returning from the JNI Call
1303 // or verify that it wasn't changed.
1304 if (VM_Version::supports_sse()) {
1305 if (RestoreMXCSROnJNICalls) {
1306 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
1307 }
1308 else if (CheckJNICalls ) {
1309 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
1310 }
1311 }
1313 #ifndef _LP64
1314 // Either restore the x87 floating pointer control word after returning
1315 // from the JNI call or verify that it wasn't changed.
1316 if (CheckJNICalls) {
1317 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
1318 }
1319 #endif // _LP64
1322 // change thread state
1323 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1324 if(os::is_MP()) {
1325 // Write serialization page so VM thread can do a pseudo remote membar.
1326 // We use the current thread pointer to calculate a thread specific
1327 // offset to write to within the page. This minimizes bus traffic
1328 // due to cache line collision.
1329 __ serialize_memory(thread, rcx);
1330 }
1332 // check for safepoint operation in progress and/or pending suspend requests
1333 { Label Continue;
1335 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1336 SafepointSynchronize::_not_synchronized);
1338 // threads running native code and they are expected to self-suspend
1339 // when leaving the _thread_in_native state. We need to check for
1340 // pending suspend requests here.
1341 Label L;
1342 __ jcc(Assembler::notEqual, L);
1343 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1344 __ jcc(Assembler::equal, Continue);
1345 __ bind(L);
1347 // Don't use call_VM as it will see a possible pending exception and forward it
1348 // and never return here preventing us from clearing _last_native_pc down below.
1349 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1350 // preserved and correspond to the bcp/locals pointers.
1351 //
1353 ((MacroAssembler*)_masm)->call_VM_leaf(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
1354 thread);
1355 __ increment(rsp, wordSize);
1357 __ movptr(method, STATE(_method));
1358 __ verify_method_ptr(method);
1359 __ movptr(thread, STATE(_thread)); // get thread
1361 __ bind(Continue);
1362 }
1364 // change thread state
1365 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1367 __ reset_last_Java_frame(thread, true, true);
1369 // reset handle block
1370 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1371 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1373 // If result was an oop then unbox and save it in the frame
1374 { Label L;
1375 Label no_oop, store_result;
1376 ExternalAddress oop_handler(AbstractInterpreter::result_handler(T_OBJECT));
1377 __ cmpptr(STATE(_result_handler), oop_handler.addr());
1378 __ jcc(Assembler::notEqual, no_oop);
1379 #ifndef _LP64
1380 __ pop(rdx);
1381 #endif // _LP64
1382 __ pop(rax);
1383 __ testptr(rax, rax);
1384 __ jcc(Assembler::zero, store_result);
1385 // unbox
1386 __ movptr(rax, Address(rax, 0));
1387 __ bind(store_result);
1388 __ movptr(STATE(_oop_temp), rax);
1389 // keep stack depth as expected by pushing oop which will eventually be discarded
1390 __ push(rax);
1391 #ifndef _LP64
1392 __ push(rdx);
1393 #endif // _LP64
1394 __ bind(no_oop);
1395 }
1397 {
1398 Label no_reguard;
1399 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1400 __ jcc(Assembler::notEqual, no_reguard);
1402 __ pusha();
1403 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1404 __ popa();
1406 __ bind(no_reguard);
1407 }
1410 // QQQ Seems like for native methods we simply return and the caller will see the pending
1411 // exception and do the right thing. Certainly the interpreter will, don't know about
1412 // compiled methods.
1413 // Seems that the answer to above is no this is wrong. The old code would see the exception
1414 // and forward it before doing the unlocking and notifying jvmdi that method has exited.
1415 // This seems wrong need to investigate the spec.
1417 // handle exceptions (exception handling will handle unlocking!)
1418 { Label L;
1419 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1420 __ jcc(Assembler::zero, L);
1421 __ bind(pending_exception_present);
1423 // There are potential results on the stack (rax/rdx, ST(0)) we ignore these and simply
1424 // return and let caller deal with exception. This skips the unlocking here which
1425 // seems wrong but seems to be what asm interpreter did. Can't find this in the spec.
1426 // Note: must preverve method in rbx
1427 //
1429 // remove activation
1431 __ movptr(t, STATE(_sender_sp));
1432 __ leave(); // remove frame anchor
1433 __ pop(rdi); // get return address
1434 __ movptr(state, STATE(_prev_link)); // get previous state for return
1435 __ mov(rsp, t); // set sp to sender sp
1436 __ push(rdi); // push throwing pc
1437 // The skips unlocking!! This seems to be what asm interpreter does but seems
1438 // very wrong. Not clear if this violates the spec.
1439 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1440 __ bind(L);
1441 }
1443 // do unlocking if necessary
1444 { Label L;
1445 __ movl(t, Address(method, Method::access_flags_offset()));
1446 __ testl(t, JVM_ACC_SYNCHRONIZED);
1447 __ jcc(Assembler::zero, L);
1448 // the code below should be shared with interpreter macro assembler implementation
1449 { Label unlock;
1450 const Register monitor = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1451 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1452 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1453 __ movptr(monitor, STATE(_monitor_base));
1454 __ subptr(monitor, frame::interpreter_frame_monitor_size() * wordSize); // address of initial monitor
1456 __ movptr(t, Address(monitor, BasicObjectLock::obj_offset_in_bytes()));
1457 __ testptr(t, t);
1458 __ jcc(Assembler::notZero, unlock);
1460 // Entry already unlocked, need to throw exception
1461 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1462 __ should_not_reach_here();
1464 __ bind(unlock);
1465 __ unlock_object(monitor);
1466 // unlock can blow rbx so restore it for path that needs it below
1467 __ movptr(method, STATE(_method));
1468 }
1469 __ bind(L);
1470 }
1472 // jvmti support
1473 // Note: This must happen _after_ handling/throwing any exceptions since
1474 // the exception handler code notifies the runtime of method exits
1475 // too. If this happens before, method entry/exit notifications are
1476 // not properly paired (was bug - gri 11/22/99).
1477 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1479 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1480 #ifndef _LP64
1481 __ pop(rdx);
1482 #endif // _LP64
1483 __ pop(rax);
1484 __ movptr(t, STATE(_result_handler)); // get result handler
1485 __ call(t); // call result handler to convert to tosca form
1487 // remove activation
1489 __ movptr(t, STATE(_sender_sp));
1491 __ leave(); // remove frame anchor
1492 __ pop(rdi); // get return address
1493 __ movptr(state, STATE(_prev_link)); // get previous state for return (if c++ interpreter was caller)
1494 __ mov(rsp, t); // set sp to sender sp
1495 __ jmp(rdi);
1497 // invocation counter overflow
1498 if (inc_counter) {
1499 // Handle overflow of counter and compile method
1500 __ bind(invocation_counter_overflow);
1501 generate_counter_overflow(&continue_after_compile);
1502 }
1504 return entry_point;
1505 }
1507 // Generate entries that will put a result type index into rcx
1508 void CppInterpreterGenerator::generate_deopt_handling() {
1510 Label return_from_deopt_common;
1512 // Generate entries that will put a result type index into rcx
1513 // deopt needs to jump to here to enter the interpreter (return a result)
1514 deopt_frame_manager_return_atos = __ pc();
1516 // rax is live here
1517 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_OBJECT)); // Result stub address array index
1518 __ jmp(return_from_deopt_common);
1521 // deopt needs to jump to here to enter the interpreter (return a result)
1522 deopt_frame_manager_return_btos = __ pc();
1524 // rax is live here
1525 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_BOOLEAN)); // Result stub address array index
1526 __ jmp(return_from_deopt_common);
1528 // deopt needs to jump to here to enter the interpreter (return a result)
1529 deopt_frame_manager_return_itos = __ pc();
1531 // rax is live here
1532 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_INT)); // Result stub address array index
1533 __ jmp(return_from_deopt_common);
1535 // deopt needs to jump to here to enter the interpreter (return a result)
1537 deopt_frame_manager_return_ltos = __ pc();
1538 // rax,rdx are live here
1539 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_LONG)); // Result stub address array index
1540 __ jmp(return_from_deopt_common);
1542 // deopt needs to jump to here to enter the interpreter (return a result)
1544 deopt_frame_manager_return_ftos = __ pc();
1545 // st(0) is live here
1546 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index
1547 __ jmp(return_from_deopt_common);
1549 // deopt needs to jump to here to enter the interpreter (return a result)
1550 deopt_frame_manager_return_dtos = __ pc();
1552 // st(0) is live here
1553 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index
1554 __ jmp(return_from_deopt_common);
1556 // deopt needs to jump to here to enter the interpreter (return a result)
1557 deopt_frame_manager_return_vtos = __ pc();
1559 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_VOID));
1561 // Deopt return common
1562 // an index is present in rcx that lets us move any possible result being
1563 // return to the interpreter's stack
1564 //
1565 // Because we have a full sized interpreter frame on the youngest
1566 // activation the stack is pushed too deep to share the tosca to
1567 // stack converters directly. We shrink the stack to the desired
1568 // amount and then push result and then re-extend the stack.
1569 // We could have the code in size_activation layout a short
1570 // frame for the top activation but that would look different
1571 // than say sparc (which needs a full size activation because
1572 // the windows are in the way. Really it could be short? QQQ
1573 //
1574 __ bind(return_from_deopt_common);
1576 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
1578 // setup rsp so we can push the "result" as needed.
1579 __ movptr(rsp, STATE(_stack)); // trim stack (is prepushed)
1580 __ addptr(rsp, wordSize); // undo prepush
1582 ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack);
1583 // Address index(noreg, rcx, Address::times_ptr);
1584 __ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_ptr)));
1585 // __ movl(rcx, Address(noreg, rcx, Address::times_ptr, int(AbstractInterpreter::_tosca_to_stack)));
1586 __ call(rcx); // call result converter
1588 __ movl(STATE(_msg), (int)BytecodeInterpreter::deopt_resume);
1589 __ lea(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present)
1590 __ movptr(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed,
1591 // result if any on stack already )
1592 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth
1593 }
1595 // Generate the code to handle a more_monitors message from the c++ interpreter
1596 void CppInterpreterGenerator::generate_more_monitors() {
1599 Label entry, loop;
1600 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1601 // 1. compute new pointers // rsp: old expression stack top
1602 __ movptr(rdx, STATE(_stack_base)); // rdx: old expression stack bottom
1603 __ subptr(rsp, entry_size); // move expression stack top limit
1604 __ subptr(STATE(_stack), entry_size); // update interpreter stack top
1605 __ subptr(STATE(_stack_limit), entry_size); // inform interpreter
1606 __ subptr(rdx, entry_size); // move expression stack bottom
1607 __ movptr(STATE(_stack_base), rdx); // inform interpreter
1608 __ movptr(rcx, STATE(_stack)); // set start value for copy loop
1609 __ jmp(entry);
1610 // 2. move expression stack contents
1611 __ bind(loop);
1612 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location
1613 __ movptr(Address(rcx, 0), rbx); // and store it at new location
1614 __ addptr(rcx, wordSize); // advance to next word
1615 __ bind(entry);
1616 __ cmpptr(rcx, rdx); // check if bottom reached
1617 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word
1618 // now zero the slot so we can find it.
1619 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
1620 __ movl(STATE(_msg), (int)BytecodeInterpreter::got_monitors);
1621 }
1624 // Initial entry to C++ interpreter from the call_stub.
1625 // This entry point is called the frame manager since it handles the generation
1626 // of interpreter activation frames via requests directly from the vm (via call_stub)
1627 // and via requests from the interpreter. The requests from the call_stub happen
1628 // directly thru the entry point. Requests from the interpreter happen via returning
1629 // from the interpreter and examining the message the interpreter has returned to
1630 // the frame manager. The frame manager can take the following requests:
1632 // NO_REQUEST - error, should never happen.
1633 // MORE_MONITORS - need a new monitor. Shuffle the expression stack on down and
1634 // allocate a new monitor.
1635 // CALL_METHOD - setup a new activation to call a new method. Very similar to what
1636 // happens during entry during the entry via the call stub.
1637 // RETURN_FROM_METHOD - remove an activation. Return to interpreter or call stub.
1638 //
1639 // Arguments:
1640 //
1641 // rbx: Method*
1642 // rcx: receiver - unused (retrieved from stack as needed)
1643 // rsi/r13: previous frame manager state (NULL from the call_stub/c1/c2)
1644 //
1645 //
1646 // Stack layout at entry
1647 //
1648 // [ return address ] <--- rsp
1649 // [ parameter n ]
1650 // ...
1651 // [ parameter 1 ]
1652 // [ expression stack ]
1653 //
1654 //
1655 // We are free to blow any registers we like because the call_stub which brought us here
1656 // initially has preserved the callee save registers already.
1657 //
1658 //
1660 static address interpreter_frame_manager = NULL;
1662 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1664 // rbx: Method*
1665 // rsi/r13: sender sp
1667 // Because we redispatch "recursive" interpreter entries thru this same entry point
1668 // the "input" register usage is a little strange and not what you expect coming
1669 // from the call_stub. From the call stub rsi/rdi (current/previous) interpreter
1670 // state are NULL but on "recursive" dispatches they are what you'd expect.
1671 // rsi: current interpreter state (C++ interpreter) must preserve (null from call_stub/c1/c2)
1674 // A single frame manager is plenty as we don't specialize for synchronized. We could and
1675 // the code is pretty much ready. Would need to change the test below and for good measure
1676 // modify generate_interpreter_state to only do the (pre) sync stuff stuff for synchronized
1677 // routines. Not clear this is worth it yet.
1679 if (interpreter_frame_manager) return interpreter_frame_manager;
1681 address entry_point = __ pc();
1683 // Fast accessor methods share this entry point.
1684 // This works because frame manager is in the same codelet
1685 if (UseFastAccessorMethods && !synchronized) __ bind(fast_accessor_slow_entry_path);
1687 Label dispatch_entry_2;
1688 __ movptr(rcx, sender_sp_on_entry);
1689 __ movptr(state, (int32_t)NULL_WORD); // no current activation
1691 __ jmp(dispatch_entry_2);
1693 const Register locals = rdi;
1695 Label re_dispatch;
1697 __ bind(re_dispatch);
1699 // save sender sp (doesn't include return address
1700 __ lea(rcx, Address(rsp, wordSize));
1702 __ bind(dispatch_entry_2);
1704 // save sender sp
1705 __ push(rcx);
1707 const Address constMethod (rbx, Method::const_offset());
1708 const Address access_flags (rbx, Method::access_flags_offset());
1709 const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
1710 const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset());
1712 // const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1713 // const Address monitor_block_bot (rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1714 // const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1716 // get parameter size (always needed)
1717 __ movptr(rdx, constMethod);
1718 __ load_unsigned_short(rcx, size_of_parameters);
1720 // rbx: Method*
1721 // rcx: size of parameters
1722 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1724 __ subptr(rdx, rcx); // rdx = no. of additional locals
1726 // see if we've got enough room on the stack for locals plus overhead.
1727 generate_stack_overflow_check(); // C++
1729 // c++ interpreter does not use stack banging or any implicit exceptions
1730 // leave for now to verify that check is proper.
1731 bang_stack_shadow_pages(false);
1735 // compute beginning of parameters (rdi)
1736 __ lea(locals, Address(rsp, rcx, Address::times_ptr, wordSize));
1738 // save sender's sp
1739 // __ movl(rcx, rsp);
1741 // get sender's sp
1742 __ pop(rcx);
1744 // get return address
1745 __ pop(rax);
1747 // rdx - # of additional locals
1748 // allocate space for locals
1749 // explicitly initialize locals
1750 {
1751 Label exit, loop;
1752 __ testl(rdx, rdx); // (32bit ok)
1753 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1754 __ bind(loop);
1755 __ push((int32_t)NULL_WORD); // initialize local variables
1756 __ decrement(rdx); // until everything initialized
1757 __ jcc(Assembler::greater, loop);
1758 __ bind(exit);
1759 }
1762 // Assumes rax = return address
1764 // allocate and initialize new interpreterState and method expression stack
1765 // IN(locals) -> locals
1766 // IN(state) -> any current interpreter activation
1767 // destroys rax, rcx, rdx, rdi
1768 // OUT (state) -> new interpreterState
1769 // OUT(rsp) -> bottom of methods expression stack
1771 generate_compute_interpreter_state(state, locals, rcx, false);
1773 // Call interpreter
1775 Label call_interpreter;
1776 __ bind(call_interpreter);
1778 // c++ interpreter does not use stack banging or any implicit exceptions
1779 // leave for now to verify that check is proper.
1780 bang_stack_shadow_pages(false);
1783 // Call interpreter enter here if message is
1784 // set and we know stack size is valid
1786 Label call_interpreter_2;
1788 __ bind(call_interpreter_2);
1790 {
1791 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1793 #ifdef _LP64
1794 __ mov(c_rarg0, state);
1795 #else
1796 __ push(state); // push arg to interpreter
1797 __ movptr(thread, STATE(_thread));
1798 #endif // _LP64
1800 // We can setup the frame anchor with everything we want at this point
1801 // as we are thread_in_Java and no safepoints can occur until we go to
1802 // vm mode. We do have to clear flags on return from vm but that is it
1803 //
1804 __ movptr(Address(thread, JavaThread::last_Java_fp_offset()), rbp);
1805 __ movptr(Address(thread, JavaThread::last_Java_sp_offset()), rsp);
1807 // Call the interpreter
1809 RuntimeAddress normal(CAST_FROM_FN_PTR(address, BytecodeInterpreter::run));
1810 RuntimeAddress checking(CAST_FROM_FN_PTR(address, BytecodeInterpreter::runWithChecks));
1812 __ call(JvmtiExport::can_post_interpreter_events() ? checking : normal);
1813 NOT_LP64(__ pop(rax);) // discard parameter to run
1814 //
1815 // state is preserved since it is callee saved
1816 //
1818 // reset_last_Java_frame
1820 NOT_LP64(__ movl(thread, STATE(_thread));)
1821 __ reset_last_Java_frame(thread, true, true);
1822 }
1824 // examine msg from interpreter to determine next action
1826 __ movl(rdx, STATE(_msg)); // Get new message
1828 Label call_method;
1829 Label return_from_interpreted_method;
1830 Label throw_exception;
1831 Label bad_msg;
1832 Label do_OSR;
1834 __ cmpl(rdx, (int32_t)BytecodeInterpreter::call_method);
1835 __ jcc(Assembler::equal, call_method);
1836 __ cmpl(rdx, (int32_t)BytecodeInterpreter::return_from_method);
1837 __ jcc(Assembler::equal, return_from_interpreted_method);
1838 __ cmpl(rdx, (int32_t)BytecodeInterpreter::do_osr);
1839 __ jcc(Assembler::equal, do_OSR);
1840 __ cmpl(rdx, (int32_t)BytecodeInterpreter::throwing_exception);
1841 __ jcc(Assembler::equal, throw_exception);
1842 __ cmpl(rdx, (int32_t)BytecodeInterpreter::more_monitors);
1843 __ jcc(Assembler::notEqual, bad_msg);
1845 // Allocate more monitor space, shuffle expression stack....
1847 generate_more_monitors();
1849 __ jmp(call_interpreter);
1851 // uncommon trap needs to jump to here to enter the interpreter (re-execute current bytecode)
1852 unctrap_frame_manager_entry = __ pc();
1853 //
1854 // Load the registers we need.
1855 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
1856 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth
1857 __ jmp(call_interpreter_2);
1861 //=============================================================================
1862 // Returning from a compiled method into a deopted method. The bytecode at the
1863 // bcp has completed. The result of the bytecode is in the native abi (the tosca
1864 // for the template based interpreter). Any stack space that was used by the
1865 // bytecode that has completed has been removed (e.g. parameters for an invoke)
1866 // so all that we have to do is place any pending result on the expression stack
1867 // and resume execution on the next bytecode.
1870 generate_deopt_handling();
1871 __ jmp(call_interpreter);
1874 // Current frame has caught an exception we need to dispatch to the
1875 // handler. We can get here because a native interpreter frame caught
1876 // an exception in which case there is no handler and we must rethrow
1877 // If it is a vanilla interpreted frame the we simply drop into the
1878 // interpreter and let it do the lookup.
1880 Interpreter::_rethrow_exception_entry = __ pc();
1881 // rax: exception
1882 // rdx: return address/pc that threw exception
1884 Label return_with_exception;
1885 Label unwind_and_forward;
1887 // restore state pointer.
1888 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
1890 __ movptr(rbx, STATE(_method)); // get method
1891 #ifdef _LP64
1892 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
1893 #else
1894 __ movl(rcx, STATE(_thread)); // get thread
1896 // Store exception with interpreter will expect it
1897 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax);
1898 #endif // _LP64
1900 // is current frame vanilla or native?
1902 __ movl(rdx, access_flags);
1903 __ testl(rdx, JVM_ACC_NATIVE);
1904 __ jcc(Assembler::zero, return_with_exception); // vanilla interpreted frame, handle directly
1906 // We drop thru to unwind a native interpreted frame with a pending exception
1907 // We jump here for the initial interpreter frame with exception pending
1908 // We unwind the current acivation and forward it to our caller.
1910 __ bind(unwind_and_forward);
1912 // unwind rbp, return stack to unextended value and re-push return address
1914 __ movptr(rcx, STATE(_sender_sp));
1915 __ leave();
1916 __ pop(rdx);
1917 __ mov(rsp, rcx);
1918 __ push(rdx);
1919 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1921 // Return point from a call which returns a result in the native abi
1922 // (c1/c2/jni-native). This result must be processed onto the java
1923 // expression stack.
1924 //
1925 // A pending exception may be present in which case there is no result present
1927 Label resume_interpreter;
1928 Label do_float;
1929 Label do_double;
1930 Label done_conv;
1932 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
1933 if (UseSSE < 2) {
1934 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
1935 __ movptr(rbx, STATE(_result._to_call._callee)); // get method just executed
1936 __ movl(rcx, Address(rbx, Method::result_index_offset()));
1937 __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index
1938 __ jcc(Assembler::equal, do_float);
1939 __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index
1940 __ jcc(Assembler::equal, do_double);
1941 #if !defined(_LP64) || defined(COMPILER1) || !defined(COMPILER2)
1942 __ empty_FPU_stack();
1943 #endif // COMPILER2
1944 __ jmp(done_conv);
1946 __ bind(do_float);
1947 #ifdef COMPILER2
1948 for (int i = 1; i < 8; i++) {
1949 __ ffree(i);
1950 }
1951 #endif // COMPILER2
1952 __ jmp(done_conv);
1953 __ bind(do_double);
1954 #ifdef COMPILER2
1955 for (int i = 1; i < 8; i++) {
1956 __ ffree(i);
1957 }
1958 #endif // COMPILER2
1959 __ jmp(done_conv);
1960 } else {
1961 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
1962 __ jmp(done_conv);
1963 }
1965 // Return point to interpreter from compiled/native method
1966 InternalAddress return_from_native_method(__ pc());
1968 __ bind(done_conv);
1971 // Result if any is in tosca. The java expression stack is in the state that the
1972 // calling convention left it (i.e. params may or may not be present)
1973 // Copy the result from tosca and place it on java expression stack.
1975 // Restore rsi/r13 as compiled code may not preserve it
1977 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
1979 // restore stack to what we had when we left (in case i2c extended it)
1981 __ movptr(rsp, STATE(_stack));
1982 __ lea(rsp, Address(rsp, wordSize));
1984 // If there is a pending exception then we don't really have a result to process
1986 #ifdef _LP64
1987 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1988 #else
1989 __ movptr(rcx, STATE(_thread)); // get thread
1990 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1991 #endif // _LP64
1992 __ jcc(Assembler::notZero, return_with_exception);
1994 // get method just executed
1995 __ movptr(rbx, STATE(_result._to_call._callee));
1997 // callee left args on top of expression stack, remove them
1998 __ movptr(rcx, constMethod);
1999 __ load_unsigned_short(rcx, Address(rcx, ConstMethod::size_of_parameters_offset()));
2001 __ lea(rsp, Address(rsp, rcx, Address::times_ptr));
2003 __ movl(rcx, Address(rbx, Method::result_index_offset()));
2004 ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack);
2005 // Address index(noreg, rax, Address::times_ptr);
2006 __ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_ptr)));
2007 // __ movl(rcx, Address(noreg, rcx, Address::times_ptr, int(AbstractInterpreter::_tosca_to_stack)));
2008 __ call(rcx); // call result converter
2009 __ jmp(resume_interpreter);
2011 // An exception is being caught on return to a vanilla interpreter frame.
2012 // Empty the stack and resume interpreter
2014 __ bind(return_with_exception);
2016 // Exception present, empty stack
2017 __ movptr(rsp, STATE(_stack_base));
2018 __ jmp(resume_interpreter);
2020 // Return from interpreted method we return result appropriate to the caller (i.e. "recursive"
2021 // interpreter call, or native) and unwind this interpreter activation.
2022 // All monitors should be unlocked.
2024 __ bind(return_from_interpreted_method);
2026 Label return_to_initial_caller;
2028 __ movptr(rbx, STATE(_method)); // get method just executed
2029 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from "recursive" interpreter call?
2030 __ movl(rax, Address(rbx, Method::result_index_offset())); // get result type index
2031 __ jcc(Assembler::equal, return_to_initial_caller); // back to native code (call_stub/c1/c2)
2033 // Copy result to callers java stack
2034 ExternalAddress stack_to_stack((address)CppInterpreter::_stack_to_stack);
2035 // Address index(noreg, rax, Address::times_ptr);
2037 __ movptr(rax, ArrayAddress(stack_to_stack, Address(noreg, rax, Address::times_ptr)));
2038 // __ movl(rax, Address(noreg, rax, Address::times_ptr, int(AbstractInterpreter::_stack_to_stack)));
2039 __ call(rax); // call result converter
2041 Label unwind_recursive_activation;
2042 __ bind(unwind_recursive_activation);
2044 // returning to interpreter method from "recursive" interpreter call
2045 // result converter left rax pointing to top of the java stack for method we are returning
2046 // to. Now all we must do is unwind the state from the completed call
2048 __ movptr(state, STATE(_prev_link)); // unwind state
2049 __ leave(); // pop the frame
2050 __ mov(rsp, rax); // unwind stack to remove args
2052 // Resume the interpreter. The current frame contains the current interpreter
2053 // state object.
2054 //
2056 __ bind(resume_interpreter);
2058 // state == interpreterState object for method we are resuming
2060 __ movl(STATE(_msg), (int)BytecodeInterpreter::method_resume);
2061 __ lea(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present)
2062 __ movptr(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed,
2063 // result if any on stack already )
2064 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth
2065 __ jmp(call_interpreter_2); // No need to bang
2067 // interpreter returning to native code (call_stub/c1/c2)
2068 // convert result and unwind initial activation
2069 // rax - result index
2071 __ bind(return_to_initial_caller);
2072 ExternalAddress stack_to_native((address)CppInterpreter::_stack_to_native_abi);
2073 // Address index(noreg, rax, Address::times_ptr);
2075 __ movptr(rax, ArrayAddress(stack_to_native, Address(noreg, rax, Address::times_ptr)));
2076 __ call(rax); // call result converter
2078 Label unwind_initial_activation;
2079 __ bind(unwind_initial_activation);
2081 // RETURN TO CALL_STUB/C1/C2 code (result if any in rax/rdx ST(0))
2083 /* Current stack picture
2085 [ incoming parameters ]
2086 [ extra locals ]
2087 [ return address to CALL_STUB/C1/C2]
2088 fp -> [ CALL_STUB/C1/C2 fp ]
2089 BytecodeInterpreter object
2090 expression stack
2091 sp ->
2093 */
2095 // return restoring the stack to the original sender_sp value
2097 __ movptr(rcx, STATE(_sender_sp));
2098 __ leave();
2099 __ pop(rdi); // get return address
2100 // set stack to sender's sp
2101 __ mov(rsp, rcx);
2102 __ jmp(rdi); // return to call_stub
2104 // OSR request, adjust return address to make current frame into adapter frame
2105 // and enter OSR nmethod
2107 __ bind(do_OSR);
2109 Label remove_initial_frame;
2111 // We are going to pop this frame. Is there another interpreter frame underneath
2112 // it or is it callstub/compiled?
2114 // Move buffer to the expected parameter location
2115 __ movptr(rcx, STATE(_result._osr._osr_buf));
2117 __ movptr(rax, STATE(_result._osr._osr_entry));
2119 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from "recursive" interpreter call?
2120 __ jcc(Assembler::equal, remove_initial_frame); // back to native code (call_stub/c1/c2)
2122 __ movptr(sender_sp_on_entry, STATE(_sender_sp)); // get sender's sp in expected register
2123 __ leave(); // pop the frame
2124 __ mov(rsp, sender_sp_on_entry); // trim any stack expansion
2127 // We know we are calling compiled so push specialized return
2128 // method uses specialized entry, push a return so we look like call stub setup
2129 // this path will handle fact that result is returned in registers and not
2130 // on the java stack.
2132 __ pushptr(return_from_native_method.addr());
2134 __ jmp(rax);
2136 __ bind(remove_initial_frame);
2138 __ movptr(rdx, STATE(_sender_sp));
2139 __ leave();
2140 // get real return
2141 __ pop(rsi);
2142 // set stack to sender's sp
2143 __ mov(rsp, rdx);
2144 // repush real return
2145 __ push(rsi);
2146 // Enter OSR nmethod
2147 __ jmp(rax);
2152 // Call a new method. All we do is (temporarily) trim the expression stack
2153 // push a return address to bring us back to here and leap to the new entry.
2155 __ bind(call_method);
2157 // stack points to next free location and not top element on expression stack
2158 // method expects sp to be pointing to topmost element
2160 __ movptr(rsp, STATE(_stack)); // pop args to c++ interpreter, set sp to java stack top
2161 __ lea(rsp, Address(rsp, wordSize));
2163 __ movptr(rbx, STATE(_result._to_call._callee)); // get method to execute
2165 // don't need a return address if reinvoking interpreter
2167 // Make it look like call_stub calling conventions
2169 // Get (potential) receiver
2170 // get size of parameters in words
2171 __ movptr(rcx, constMethod);
2172 __ load_unsigned_short(rcx, Address(rcx, ConstMethod::size_of_parameters_offset()));
2174 ExternalAddress recursive(CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation));
2175 __ pushptr(recursive.addr()); // make it look good in the debugger
2177 InternalAddress entry(entry_point);
2178 __ cmpptr(STATE(_result._to_call._callee_entry_point), entry.addr()); // returning to interpreter?
2179 __ jcc(Assembler::equal, re_dispatch); // yes
2181 __ pop(rax); // pop dummy address
2184 // get specialized entry
2185 __ movptr(rax, STATE(_result._to_call._callee_entry_point));
2186 // set sender SP
2187 __ mov(sender_sp_on_entry, rsp);
2189 // method uses specialized entry, push a return so we look like call stub setup
2190 // this path will handle fact that result is returned in registers and not
2191 // on the java stack.
2193 __ pushptr(return_from_native_method.addr());
2195 __ jmp(rax);
2197 __ bind(bad_msg);
2198 __ stop("Bad message from interpreter");
2200 // Interpreted method "returned" with an exception pass it on...
2201 // Pass result, unwind activation and continue/return to interpreter/call_stub
2202 // We handle result (if any) differently based on return to interpreter or call_stub
2204 Label unwind_initial_with_pending_exception;
2206 __ bind(throw_exception);
2207 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from recursive interpreter call?
2208 __ jcc(Assembler::equal, unwind_initial_with_pending_exception); // no, back to native code (call_stub/c1/c2)
2209 __ movptr(rax, STATE(_locals)); // pop parameters get new stack value
2210 __ addptr(rax, wordSize); // account for prepush before we return
2211 __ jmp(unwind_recursive_activation);
2213 __ bind(unwind_initial_with_pending_exception);
2215 // We will unwind the current (initial) interpreter frame and forward
2216 // the exception to the caller. We must put the exception in the
2217 // expected register and clear pending exception and then forward.
2219 __ jmp(unwind_and_forward);
2221 interpreter_frame_manager = entry_point;
2222 return entry_point;
2223 }
2225 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
2226 // determine code generation flags
2227 bool synchronized = false;
2228 address entry_point = NULL;
2230 switch (kind) {
2231 case Interpreter::zerolocals : break;
2232 case Interpreter::zerolocals_synchronized: synchronized = true; break;
2233 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
2234 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
2235 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
2236 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
2237 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
2238 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
2240 case Interpreter::java_lang_math_sin : // fall thru
2241 case Interpreter::java_lang_math_cos : // fall thru
2242 case Interpreter::java_lang_math_tan : // fall thru
2243 case Interpreter::java_lang_math_abs : // fall thru
2244 case Interpreter::java_lang_math_log : // fall thru
2245 case Interpreter::java_lang_math_log10 : // fall thru
2246 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
2247 case Interpreter::java_lang_ref_reference_get
2248 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
2249 default : ShouldNotReachHere(); break;
2250 }
2252 if (entry_point) return entry_point;
2254 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
2256 }
2258 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
2259 : CppInterpreterGenerator(code) {
2260 generate_all(); // down here so it can be "virtual"
2261 }
2263 // Deoptimization helpers for C++ interpreter
2265 // How much stack a method activation needs in words.
2266 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
2268 const int stub_code = 4; // see generate_call_stub
2269 // Save space for one monitor to get into the interpreted method in case
2270 // the method is synchronized
2271 int monitor_size = method->is_synchronized() ?
2272 1*frame::interpreter_frame_monitor_size() : 0;
2274 // total static overhead size. Account for interpreter state object, return
2275 // address, saved rbp and 2 words for a "static long no_params() method" issue.
2277 const int overhead_size = sizeof(BytecodeInterpreter)/wordSize +
2278 ( frame::sender_sp_offset - frame::link_offset) + 2;
2280 const int extra_stack = 0; //6815692//Method::extra_stack_entries();
2281 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
2282 Interpreter::stackElementWords;
2283 return overhead_size + method_stack + stub_code;
2284 }
2286 // returns the activation size.
2287 static int size_activation_helper(int extra_locals_size, int monitor_size) {
2288 return (extra_locals_size + // the addition space for locals
2289 2*BytesPerWord + // return address and saved rbp
2290 2*BytesPerWord + // "static long no_params() method" issue
2291 sizeof(BytecodeInterpreter) + // interpreterState
2292 monitor_size); // monitors
2293 }
2295 void BytecodeInterpreter::layout_interpreterState(interpreterState to_fill,
2296 frame* caller,
2297 frame* current,
2298 Method* method,
2299 intptr_t* locals,
2300 intptr_t* stack,
2301 intptr_t* stack_base,
2302 intptr_t* monitor_base,
2303 intptr_t* frame_bottom,
2304 bool is_top_frame
2305 )
2306 {
2307 // What about any vtable?
2308 //
2309 to_fill->_thread = JavaThread::current();
2310 // This gets filled in later but make it something recognizable for now
2311 to_fill->_bcp = method->code_base();
2312 to_fill->_locals = locals;
2313 to_fill->_constants = method->constants()->cache();
2314 to_fill->_method = method;
2315 to_fill->_mdx = NULL;
2316 to_fill->_stack = stack;
2317 if (is_top_frame && JavaThread::current()->popframe_forcing_deopt_reexecution() ) {
2318 to_fill->_msg = deopt_resume2;
2319 } else {
2320 to_fill->_msg = method_resume;
2321 }
2322 to_fill->_result._to_call._bcp_advance = 0;
2323 to_fill->_result._to_call._callee_entry_point = NULL; // doesn't matter to anyone
2324 to_fill->_result._to_call._callee = NULL; // doesn't matter to anyone
2325 to_fill->_prev_link = NULL;
2327 to_fill->_sender_sp = caller->unextended_sp();
2329 if (caller->is_interpreted_frame()) {
2330 interpreterState prev = caller->get_interpreterState();
2331 to_fill->_prev_link = prev;
2332 // *current->register_addr(GR_Iprev_state) = (intptr_t) prev;
2333 // Make the prev callee look proper
2334 prev->_result._to_call._callee = method;
2335 if (*prev->_bcp == Bytecodes::_invokeinterface) {
2336 prev->_result._to_call._bcp_advance = 5;
2337 } else {
2338 prev->_result._to_call._bcp_advance = 3;
2339 }
2340 }
2341 to_fill->_oop_temp = NULL;
2342 to_fill->_stack_base = stack_base;
2343 // Need +1 here because stack_base points to the word just above the first expr stack entry
2344 // and stack_limit is supposed to point to the word just below the last expr stack entry.
2345 // See generate_compute_interpreter_state.
2346 int extra_stack = 0; //6815692//Method::extra_stack_entries();
2347 to_fill->_stack_limit = stack_base - (method->max_stack() + extra_stack + 1);
2348 to_fill->_monitor_base = (BasicObjectLock*) monitor_base;
2350 to_fill->_self_link = to_fill;
2351 assert(stack >= to_fill->_stack_limit && stack < to_fill->_stack_base,
2352 "Stack top out of range");
2353 }
2355 int AbstractInterpreter::layout_activation(Method* method,
2356 int tempcount, //
2357 int popframe_extra_args,
2358 int moncount,
2359 int caller_actual_parameters,
2360 int callee_param_count,
2361 int callee_locals,
2362 frame* caller,
2363 frame* interpreter_frame,
2364 bool is_top_frame,
2365 bool is_bottom_frame) {
2367 assert(popframe_extra_args == 0, "FIX ME");
2368 // NOTE this code must exactly mimic what InterpreterGenerator::generate_compute_interpreter_state()
2369 // does as far as allocating an interpreter frame.
2370 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
2371 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
2372 // as determined by a previous call to this method.
2373 // It is also guaranteed to be walkable even though it is in a skeletal state
2374 // NOTE: return size is in words not bytes
2375 // NOTE: tempcount is the current size of the java expression stack. For top most
2376 // frames we will allocate a full sized expression stack and not the curback
2377 // version that non-top frames have.
2379 // Calculate the amount our frame will be adjust by the callee. For top frame
2380 // this is zero.
2382 // NOTE: ia64 seems to do this wrong (or at least backwards) in that it
2383 // calculates the extra locals based on itself. Not what the callee does
2384 // to it. So it ignores last_frame_adjust value. Seems suspicious as far
2385 // as getting sender_sp correct.
2387 int extra_locals_size = (callee_locals - callee_param_count) * BytesPerWord;
2388 int monitor_size = sizeof(BasicObjectLock) * moncount;
2390 // First calculate the frame size without any java expression stack
2391 int short_frame_size = size_activation_helper(extra_locals_size,
2392 monitor_size);
2394 // Now with full size expression stack
2395 int extra_stack = 0; //6815692//Method::extra_stack_entries();
2396 int full_frame_size = short_frame_size + (method->max_stack() + extra_stack) * BytesPerWord;
2398 // and now with only live portion of the expression stack
2399 short_frame_size = short_frame_size + tempcount * BytesPerWord;
2401 // the size the activation is right now. Only top frame is full size
2402 int frame_size = (is_top_frame ? full_frame_size : short_frame_size);
2404 if (interpreter_frame != NULL) {
2405 #ifdef ASSERT
2406 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
2407 #endif
2409 // MUCHO HACK
2411 intptr_t* frame_bottom = (intptr_t*) ((intptr_t)interpreter_frame->sp() - (full_frame_size - frame_size));
2413 /* Now fillin the interpreterState object */
2415 // The state object is the first thing on the frame and easily located
2417 interpreterState cur_state = (interpreterState) ((intptr_t)interpreter_frame->fp() - sizeof(BytecodeInterpreter));
2420 // Find the locals pointer. This is rather simple on x86 because there is no
2421 // confusing rounding at the callee to account for. We can trivially locate
2422 // our locals based on the current fp().
2423 // Note: the + 2 is for handling the "static long no_params() method" issue.
2424 // (too bad I don't really remember that issue well...)
2426 intptr_t* locals;
2427 // If the caller is interpreted we need to make sure that locals points to the first
2428 // argument that the caller passed and not in an area where the stack might have been extended.
2429 // because the stack to stack to converter needs a proper locals value in order to remove the
2430 // arguments from the caller and place the result in the proper location. Hmm maybe it'd be
2431 // simpler if we simply stored the result in the BytecodeInterpreter object and let the c++ code
2432 // adjust the stack?? HMMM QQQ
2433 //
2434 if (caller->is_interpreted_frame()) {
2435 // locals must agree with the caller because it will be used to set the
2436 // caller's tos when we return.
2437 interpreterState prev = caller->get_interpreterState();
2438 // stack() is prepushed.
2439 locals = prev->stack() + method->size_of_parameters();
2440 // locals = caller->unextended_sp() + (method->size_of_parameters() - 1);
2441 if (locals != interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2) {
2442 // os::breakpoint();
2443 }
2444 } else {
2445 // this is where a c2i would have placed locals (except for the +2)
2446 locals = interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2;
2447 }
2449 intptr_t* monitor_base = (intptr_t*) cur_state;
2450 intptr_t* stack_base = (intptr_t*) ((intptr_t) monitor_base - monitor_size);
2451 /* +1 because stack is always prepushed */
2452 intptr_t* stack = (intptr_t*) ((intptr_t) stack_base - (tempcount + 1) * BytesPerWord);
2455 BytecodeInterpreter::layout_interpreterState(cur_state,
2456 caller,
2457 interpreter_frame,
2458 method,
2459 locals,
2460 stack,
2461 stack_base,
2462 monitor_base,
2463 frame_bottom,
2464 is_top_frame);
2466 // BytecodeInterpreter::pd_layout_interpreterState(cur_state, interpreter_return_address, interpreter_frame->fp());
2467 }
2468 return frame_size/BytesPerWord;
2469 }
2471 #endif // CC_INTERP (all)