0
|
1 /*
|
|
2 * Copyright 2007 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
|
|
20 * CA 95054 USA or visit www.sun.com if you need additional information or
|
|
21 * have any questions.
|
|
22 *
|
|
23 */
|
|
24
|
|
25 #include "incls/_precompiled.incl"
|
|
26 #include "incls/_cppInterpreter_x86.cpp.incl"
|
|
27
|
|
28 #ifdef CC_INTERP
|
|
29
|
|
30 // Routine exists to make tracebacks look decent in debugger
|
|
31 // while we are recursed in the frame manager/c++ interpreter.
|
|
32 // We could use an address in the frame manager but having
|
|
33 // frames look natural in the debugger is a plus.
|
|
34 extern "C" void RecursiveInterpreterActivation(interpreterState istate )
|
|
35 {
|
|
36 //
|
|
37 ShouldNotReachHere();
|
|
38 }
|
|
39
|
|
40
|
|
41 #define __ _masm->
|
|
42 #define STATE(field_name) (Address(state, byte_offset_of(BytecodeInterpreter, field_name)))
|
|
43
|
|
44 Label fast_accessor_slow_entry_path; // fast accessor methods need to be able to jmp to unsynchronized
|
|
45 // c++ interpreter entry point this holds that entry point label.
|
|
46
|
|
47 // NEEDED for JVMTI?
|
|
48 // address AbstractInterpreter::_remove_activation_preserving_args_entry;
|
|
49
|
|
50 static address unctrap_frame_manager_entry = NULL;
|
|
51
|
|
52 static address deopt_frame_manager_return_atos = NULL;
|
|
53 static address deopt_frame_manager_return_btos = NULL;
|
|
54 static address deopt_frame_manager_return_itos = NULL;
|
|
55 static address deopt_frame_manager_return_ltos = NULL;
|
|
56 static address deopt_frame_manager_return_ftos = NULL;
|
|
57 static address deopt_frame_manager_return_dtos = NULL;
|
|
58 static address deopt_frame_manager_return_vtos = NULL;
|
|
59
|
|
60 int AbstractInterpreter::BasicType_as_index(BasicType type) {
|
|
61 int i = 0;
|
|
62 switch (type) {
|
|
63 case T_BOOLEAN: i = 0; break;
|
|
64 case T_CHAR : i = 1; break;
|
|
65 case T_BYTE : i = 2; break;
|
|
66 case T_SHORT : i = 3; break;
|
|
67 case T_INT : i = 4; break;
|
|
68 case T_VOID : i = 5; break;
|
|
69 case T_FLOAT : i = 8; break;
|
|
70 case T_LONG : i = 9; break;
|
|
71 case T_DOUBLE : i = 6; break;
|
|
72 case T_OBJECT : // fall through
|
|
73 case T_ARRAY : i = 7; break;
|
|
74 default : ShouldNotReachHere();
|
|
75 }
|
|
76 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
|
|
77 return i;
|
|
78 }
|
|
79
|
|
80 // Is this pc anywhere within code owned by the interpreter?
|
|
81 // This only works for pc that might possibly be exposed to frame
|
|
82 // walkers. It clearly misses all of the actual c++ interpreter
|
|
83 // implementation
|
|
84 bool CppInterpreter::contains(address pc) {
|
|
85 return (_code->contains(pc) ||
|
|
86 pc == CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation));
|
|
87 }
|
|
88
|
|
89
|
|
90 address CppInterpreterGenerator::generate_result_handler_for(BasicType type) {
|
|
91 const Register state = rsi; // current activation object, valid on entry
|
|
92 address entry = __ pc();
|
|
93 switch (type) {
|
|
94 case T_BOOLEAN: __ c2bool(rax); break;
|
|
95 case T_CHAR : __ andl(rax, 0xFFFF); break;
|
|
96 case T_BYTE : __ sign_extend_byte (rax); break;
|
|
97 case T_SHORT : __ sign_extend_short(rax); break;
|
|
98 case T_VOID : // fall thru
|
|
99 case T_LONG : // fall thru
|
|
100 case T_INT : /* nothing to do */ break;
|
|
101 case T_DOUBLE :
|
|
102 case T_FLOAT :
|
|
103 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
|
|
104 __ popl(t); // remove return address first
|
|
105 __ pop_dtos_to_rsp();
|
|
106 // Must return a result for interpreter or compiler. In SSE
|
|
107 // mode, results are returned in xmm0 and the FPU stack must
|
|
108 // be empty.
|
|
109 if (type == T_FLOAT && UseSSE >= 1) {
|
|
110 // Load ST0
|
|
111 __ fld_d(Address(rsp, 0));
|
|
112 // Store as float and empty fpu stack
|
|
113 __ fstp_s(Address(rsp, 0));
|
|
114 // and reload
|
|
115 __ movflt(xmm0, Address(rsp, 0));
|
|
116 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
|
|
117 __ movdbl(xmm0, Address(rsp, 0));
|
|
118 } else {
|
|
119 // restore ST0
|
|
120 __ fld_d(Address(rsp, 0));
|
|
121 }
|
|
122 // and pop the temp
|
|
123 __ addl(rsp, 2 * wordSize);
|
|
124 __ pushl(t); // restore return address
|
|
125 }
|
|
126 break;
|
|
127 case T_OBJECT :
|
|
128 // retrieve result from frame
|
|
129 __ movl(rax, STATE(_oop_temp));
|
|
130 // and verify it
|
|
131 __ verify_oop(rax);
|
|
132 break;
|
|
133 default : ShouldNotReachHere();
|
|
134 }
|
|
135 __ ret(0); // return from result handler
|
|
136 return entry;
|
|
137 }
|
|
138
|
|
139 // tosca based result to c++ interpreter stack based result.
|
|
140 // Result goes to top of native stack.
|
|
141
|
|
142 #undef EXTEND // SHOULD NOT BE NEEDED
|
|
143 address CppInterpreterGenerator::generate_tosca_to_stack_converter(BasicType type) {
|
|
144 // A result is in the tosca (abi result) from either a native method call or compiled
|
|
145 // code. Place this result on the java expression stack so C++ interpreter can use it.
|
|
146 address entry = __ pc();
|
|
147
|
|
148 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
|
|
149 __ popl(t); // remove return address first
|
|
150 switch (type) {
|
|
151 case T_VOID:
|
|
152 break;
|
|
153 case T_BOOLEAN:
|
|
154 #ifdef EXTEND
|
|
155 __ c2bool(rax);
|
|
156 #endif
|
|
157 __ pushl(rax);
|
|
158 break;
|
|
159 case T_CHAR :
|
|
160 #ifdef EXTEND
|
|
161 __ andl(rax, 0xFFFF);
|
|
162 #endif
|
|
163 __ pushl(rax);
|
|
164 break;
|
|
165 case T_BYTE :
|
|
166 #ifdef EXTEND
|
|
167 __ sign_extend_byte (rax);
|
|
168 #endif
|
|
169 __ pushl(rax);
|
|
170 break;
|
|
171 case T_SHORT :
|
|
172 #ifdef EXTEND
|
|
173 __ sign_extend_short(rax);
|
|
174 #endif
|
|
175 __ pushl(rax);
|
|
176 break;
|
|
177 case T_LONG :
|
|
178 __ pushl(rdx);
|
|
179 __ pushl(rax);
|
|
180 break;
|
|
181 case T_INT :
|
|
182 __ pushl(rax);
|
|
183 break;
|
|
184 case T_FLOAT :
|
|
185 // Result is in ST(0)
|
|
186 if ( UseSSE < 1) {
|
|
187 __ push(ftos); // and save it
|
|
188 } else {
|
|
189 __ subl(rsp, wordSize);
|
|
190 __ movflt(Address(rsp, 0), xmm0);
|
|
191 }
|
|
192 break;
|
|
193 case T_DOUBLE :
|
|
194 if ( UseSSE < 2 ) {
|
|
195 __ push(dtos); // put ST0 on java stack
|
|
196 } else {
|
|
197 __ subl(rsp, 2*wordSize);
|
|
198 __ movdbl(Address(rsp, 0), xmm0);
|
|
199 }
|
|
200 break;
|
|
201 case T_OBJECT :
|
|
202 __ verify_oop(rax); // verify it
|
|
203 __ pushl(rax);
|
|
204 break;
|
|
205 default : ShouldNotReachHere();
|
|
206 }
|
|
207 __ jmp(t); // return from result handler
|
|
208 return entry;
|
|
209 }
|
|
210
|
|
211 address CppInterpreterGenerator::generate_stack_to_stack_converter(BasicType type) {
|
|
212 // A result is in the java expression stack of the interpreted method that has just
|
|
213 // returned. Place this result on the java expression stack of the caller.
|
|
214 //
|
|
215 // The current interpreter activation in rsi is for the method just returning its
|
|
216 // result. So we know that the result of this method is on the top of the current
|
|
217 // execution stack (which is pre-pushed) and will be return to the top of the caller
|
|
218 // stack. The top of the callers stack is the bottom of the locals of the current
|
|
219 // activation.
|
|
220 // Because of the way activation are managed by the frame manager the value of rsp is
|
|
221 // below both the stack top of the current activation and naturally the stack top
|
|
222 // of the calling activation. This enable this routine to leave the return address
|
|
223 // to the frame manager on the stack and do a vanilla return.
|
|
224 //
|
|
225 // On entry: rsi - interpreter state of activation returning a (potential) result
|
|
226 // On Return: rsi - unchanged
|
|
227 // rax - new stack top for caller activation (i.e. activation in _prev_link)
|
|
228 //
|
|
229 // Can destroy rdx, rcx.
|
|
230 //
|
|
231
|
|
232 address entry = __ pc();
|
|
233 const Register state = rsi; // current activation object, valid on entry
|
|
234 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
|
|
235 switch (type) {
|
|
236 case T_VOID:
|
|
237 __ movl(rax, STATE(_locals)); // pop parameters get new stack value
|
|
238 __ addl(rax, wordSize); // account for prepush before we return
|
|
239 break;
|
|
240 case T_FLOAT :
|
|
241 case T_BOOLEAN:
|
|
242 case T_CHAR :
|
|
243 case T_BYTE :
|
|
244 case T_SHORT :
|
|
245 case T_INT :
|
|
246 // 1 word result
|
|
247 __ movl(rdx, STATE(_stack));
|
|
248 __ movl(rax, STATE(_locals)); // address for result
|
|
249 __ movl(rdx, Address(rdx, wordSize)); // get result
|
|
250 __ movl(Address(rax, 0), rdx); // and store it
|
|
251 break;
|
|
252 case T_LONG :
|
|
253 case T_DOUBLE :
|
|
254 // return top two words on current expression stack to caller's expression stack
|
|
255 // The caller's expression stack is adjacent to the current frame manager's intepretState
|
|
256 // except we allocated one extra word for this intepretState so we won't overwrite it
|
|
257 // when we return a two word result.
|
|
258
|
|
259 __ movl(rax, STATE(_locals)); // address for result
|
|
260 __ movl(rcx, STATE(_stack));
|
|
261 __ subl(rax, wordSize); // need addition word besides locals[0]
|
|
262 __ movl(rdx, Address(rcx, 2*wordSize)); // get result word
|
|
263 __ movl(Address(rax, wordSize), rdx); // and store it
|
|
264 __ movl(rdx, Address(rcx, wordSize)); // get result word
|
|
265 __ movl(Address(rax, 0), rdx); // and store it
|
|
266 break;
|
|
267 case T_OBJECT :
|
|
268 __ movl(rdx, STATE(_stack));
|
|
269 __ movl(rax, STATE(_locals)); // address for result
|
|
270 __ movl(rdx, Address(rdx, wordSize)); // get result
|
|
271 __ verify_oop(rdx); // verify it
|
|
272 __ movl(Address(rax, 0), rdx); // and store it
|
|
273 break;
|
|
274 default : ShouldNotReachHere();
|
|
275 }
|
|
276 __ ret(0);
|
|
277 return entry;
|
|
278 }
|
|
279
|
|
280 address CppInterpreterGenerator::generate_stack_to_native_abi_converter(BasicType type) {
|
|
281 // A result is in the java expression stack of the interpreted method that has just
|
|
282 // returned. Place this result in the native abi that the caller expects.
|
|
283 //
|
|
284 // Similar to generate_stack_to_stack_converter above. Called at a similar time from the
|
|
285 // frame manager execept in this situation the caller is native code (c1/c2/call_stub)
|
|
286 // and so rather than return result onto caller's java expression stack we return the
|
|
287 // result in the expected location based on the native abi.
|
|
288 // On entry: rsi - interpreter state of activation returning a (potential) result
|
|
289 // On Return: rsi - unchanged
|
|
290 // Other registers changed [rax/rdx/ST(0) as needed for the result returned]
|
|
291
|
|
292 address entry = __ pc();
|
|
293 const Register state = rsi; // current activation object, valid on entry
|
|
294 switch (type) {
|
|
295 case T_VOID:
|
|
296 break;
|
|
297 case T_BOOLEAN:
|
|
298 case T_CHAR :
|
|
299 case T_BYTE :
|
|
300 case T_SHORT :
|
|
301 case T_INT :
|
|
302 __ movl(rdx, STATE(_stack)); // get top of stack
|
|
303 __ movl(rax, Address(rdx, wordSize)); // get result word 1
|
|
304 break;
|
|
305 case T_LONG :
|
|
306 __ movl(rdx, STATE(_stack)); // get top of stack
|
|
307 __ movl(rax, Address(rdx, wordSize)); // get result low word
|
|
308 __ movl(rdx, Address(rdx, 2*wordSize)); // get result high word
|
|
309 break;
|
|
310 break;
|
|
311 case T_FLOAT :
|
|
312 __ movl(rdx, STATE(_stack)); // get top of stack
|
|
313 if ( UseSSE >= 1) {
|
|
314 __ movflt(xmm0, Address(rdx, wordSize));
|
|
315 } else {
|
|
316 __ fld_s(Address(rdx, wordSize)); // pushd float result
|
|
317 }
|
|
318 break;
|
|
319 case T_DOUBLE :
|
|
320 __ movl(rdx, STATE(_stack)); // get top of stack
|
|
321 if ( UseSSE > 1) {
|
|
322 __ movdbl(xmm0, Address(rdx, wordSize));
|
|
323 } else {
|
|
324 __ fld_d(Address(rdx, wordSize)); // push double result
|
|
325 }
|
|
326 break;
|
|
327 case T_OBJECT :
|
|
328 __ movl(rdx, STATE(_stack)); // get top of stack
|
|
329 __ movl(rax, Address(rdx, wordSize)); // get result word 1
|
|
330 __ verify_oop(rax); // verify it
|
|
331 break;
|
|
332 default : ShouldNotReachHere();
|
|
333 }
|
|
334 __ ret(0);
|
|
335 return entry;
|
|
336 }
|
|
337
|
|
338 address CppInterpreter::return_entry(TosState state, int length) {
|
|
339 // make it look good in the debugger
|
|
340 return CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation);
|
|
341 }
|
|
342
|
|
343 address CppInterpreter::deopt_entry(TosState state, int length) {
|
|
344 address ret = NULL;
|
|
345 if (length != 0) {
|
|
346 switch (state) {
|
|
347 case atos: ret = deopt_frame_manager_return_atos; break;
|
|
348 case btos: ret = deopt_frame_manager_return_btos; break;
|
|
349 case ctos:
|
|
350 case stos:
|
|
351 case itos: ret = deopt_frame_manager_return_itos; break;
|
|
352 case ltos: ret = deopt_frame_manager_return_ltos; break;
|
|
353 case ftos: ret = deopt_frame_manager_return_ftos; break;
|
|
354 case dtos: ret = deopt_frame_manager_return_dtos; break;
|
|
355 case vtos: ret = deopt_frame_manager_return_vtos; break;
|
|
356 }
|
|
357 } else {
|
|
358 ret = unctrap_frame_manager_entry; // re-execute the bytecode ( e.g. uncommon trap)
|
|
359 }
|
|
360 assert(ret != NULL, "Not initialized");
|
|
361 return ret;
|
|
362 }
|
|
363
|
|
364 // C++ Interpreter
|
|
365 void CppInterpreterGenerator::generate_compute_interpreter_state(const Register state,
|
|
366 const Register locals,
|
|
367 const Register sender_sp,
|
|
368 bool native) {
|
|
369
|
|
370 // On entry the "locals" argument points to locals[0] (or where it would be in case no locals in
|
|
371 // a static method). "state" contains any previous frame manager state which we must save a link
|
|
372 // to in the newly generated state object. On return "state" is a pointer to the newly allocated
|
|
373 // state object. We must allocate and initialize a new interpretState object and the method
|
|
374 // expression stack. Because the returned result (if any) of the method will be placed on the caller's
|
|
375 // expression stack and this will overlap with locals[0] (and locals[1] if double/long) we must
|
|
376 // be sure to leave space on the caller's stack so that this result will not overwrite values when
|
|
377 // locals[0] and locals[1] do not exist (and in fact are return address and saved rbp). So when
|
|
378 // we are non-native we in essence ensure that locals[0-1] exist. We play an extra trick in
|
|
379 // non-product builds and initialize this last local with the previous interpreterState as
|
|
380 // this makes things look real nice in the debugger.
|
|
381
|
|
382 // State on entry
|
|
383 // Assumes locals == &locals[0]
|
|
384 // Assumes state == any previous frame manager state (assuming call path from c++ interpreter)
|
|
385 // Assumes rax = return address
|
|
386 // rcx == senders_sp
|
|
387 // rbx == method
|
|
388 // Modifies rcx, rdx, rax
|
|
389 // Returns:
|
|
390 // state == address of new interpreterState
|
|
391 // rsp == bottom of method's expression stack.
|
|
392
|
|
393 const Address const_offset (rbx, methodOopDesc::const_offset());
|
|
394
|
|
395
|
|
396 // On entry sp is the sender's sp. This includes the space for the arguments
|
|
397 // that the sender pushed. If the sender pushed no args (a static) and the
|
|
398 // caller returns a long then we need two words on the sender's stack which
|
|
399 // are not present (although when we return a restore full size stack the
|
|
400 // space will be present). If we didn't allocate two words here then when
|
|
401 // we "push" the result of the caller's stack we would overwrite the return
|
|
402 // address and the saved rbp. Not good. So simply allocate 2 words now
|
|
403 // just to be safe. This is the "static long no_params() method" issue.
|
|
404 // See Lo.java for a testcase.
|
|
405 // We don't need this for native calls because they return result in
|
|
406 // register and the stack is expanded in the caller before we store
|
|
407 // the results on the stack.
|
|
408
|
|
409 if (!native) {
|
|
410 #ifdef PRODUCT
|
|
411 __ subl(rsp, 2*wordSize);
|
|
412 #else /* PRODUCT */
|
|
413 __ pushl((int)NULL);
|
|
414 __ pushl(state); // make it look like a real argument
|
|
415 #endif /* PRODUCT */
|
|
416 }
|
|
417
|
|
418 // Now that we are assure of space for stack result, setup typical linkage
|
|
419
|
|
420 __ pushl(rax);
|
|
421 __ enter();
|
|
422
|
|
423 __ movl(rax, state); // save current state
|
|
424
|
|
425 __ leal(rsp, Address(rsp, -(int)sizeof(BytecodeInterpreter)));
|
|
426 __ movl(state, rsp);
|
|
427
|
|
428 // rsi == state/locals rax == prevstate
|
|
429
|
|
430 // initialize the "shadow" frame so that use since C++ interpreter not directly
|
|
431 // recursive. Simpler to recurse but we can't trim expression stack as we call
|
|
432 // new methods.
|
|
433 __ movl(STATE(_locals), locals); // state->_locals = locals()
|
|
434 __ movl(STATE(_self_link), state); // point to self
|
|
435 __ movl(STATE(_prev_link), rax); // state->_link = state on entry (NULL or previous state)
|
|
436 __ movl(STATE(_sender_sp), sender_sp); // state->_sender_sp = sender_sp
|
|
437 __ get_thread(rax); // get vm's javathread*
|
|
438 __ movl(STATE(_thread), rax); // state->_bcp = codes()
|
|
439 __ movl(rdx, Address(rbx, methodOopDesc::const_offset())); // get constantMethodOop
|
|
440 __ leal(rdx, Address(rdx, constMethodOopDesc::codes_offset())); // get code base
|
|
441 if (native) {
|
|
442 __ movl(STATE(_bcp), (intptr_t)NULL); // state->_bcp = NULL
|
|
443 } else {
|
|
444 __ movl(STATE(_bcp), rdx); // state->_bcp = codes()
|
|
445 }
|
|
446 __ xorl(rdx, rdx);
|
|
447 __ movl(STATE(_oop_temp), rdx); // state->_oop_temp = NULL (only really needed for native)
|
|
448 __ movl(STATE(_mdx), rdx); // state->_mdx = NULL
|
|
449 __ movl(rdx, Address(rbx, methodOopDesc::constants_offset()));
|
|
450 __ movl(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
|
|
451 __ movl(STATE(_constants), rdx); // state->_constants = constants()
|
|
452
|
|
453 __ movl(STATE(_method), rbx); // state->_method = method()
|
|
454 __ movl(STATE(_msg), (int) BytecodeInterpreter::method_entry); // state->_msg = initial method entry
|
|
455 __ movl(STATE(_result._to_call._callee), (int) NULL); // state->_result._to_call._callee_callee = NULL
|
|
456
|
|
457
|
|
458 __ movl(STATE(_monitor_base), rsp); // set monitor block bottom (grows down) this would point to entry [0]
|
|
459 // entries run from -1..x where &monitor[x] ==
|
|
460
|
|
461 {
|
|
462 // Must not attempt to lock method until we enter interpreter as gc won't be able to find the
|
|
463 // initial frame. However we allocate a free monitor so we don't have to shuffle the expression stack
|
|
464 // immediately.
|
|
465
|
|
466 // synchronize method
|
|
467 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
|
|
468 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
|
|
469 Label not_synced;
|
|
470
|
|
471 __ movl(rax, access_flags);
|
|
472 __ testl(rax, JVM_ACC_SYNCHRONIZED);
|
|
473 __ jcc(Assembler::zero, not_synced);
|
|
474
|
|
475 // Allocate initial monitor and pre initialize it
|
|
476 // get synchronization object
|
|
477
|
|
478 Label done;
|
|
479 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
|
|
480 __ movl(rax, access_flags);
|
|
481 __ testl(rax, JVM_ACC_STATIC);
|
|
482 __ movl(rax, Address(locals, 0)); // get receiver (assume this is frequent case)
|
|
483 __ jcc(Assembler::zero, done);
|
|
484 __ movl(rax, Address(rbx, methodOopDesc::constants_offset()));
|
|
485 __ movl(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
|
|
486 __ movl(rax, Address(rax, mirror_offset));
|
|
487 __ bind(done);
|
|
488 // add space for monitor & lock
|
|
489 __ subl(rsp, entry_size); // add space for a monitor entry
|
|
490 __ movl(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
|
|
491 __ bind(not_synced);
|
|
492 }
|
|
493
|
|
494 __ movl(STATE(_stack_base), rsp); // set expression stack base ( == &monitors[-count])
|
|
495 if (native) {
|
|
496 __ movl(STATE(_stack), rsp); // set current expression stack tos
|
|
497 __ movl(STATE(_stack_limit), rsp);
|
|
498 } else {
|
|
499 __ subl(rsp, wordSize); // pre-push stack
|
|
500 __ movl(STATE(_stack), rsp); // set current expression stack tos
|
|
501
|
|
502 // compute full expression stack limit
|
|
503
|
|
504 const Address size_of_stack (rbx, methodOopDesc::max_stack_offset());
|
|
505 __ load_unsigned_word(rdx, size_of_stack); // get size of expression stack in words
|
|
506 __ negl(rdx); // so we can subtract in next step
|
|
507 // Allocate expression stack
|
|
508 __ leal(rsp, Address(rsp, rdx, Address::times_4));
|
|
509 __ movl(STATE(_stack_limit), rsp);
|
|
510 }
|
|
511
|
|
512 }
|
|
513
|
|
514 // Helpers for commoning out cases in the various type of method entries.
|
|
515 //
|
|
516
|
|
517 // increment invocation count & check for overflow
|
|
518 //
|
|
519 // Note: checking for negative value instead of overflow
|
|
520 // so we have a 'sticky' overflow test
|
|
521 //
|
|
522 // rbx,: method
|
|
523 // rcx: invocation counter
|
|
524 //
|
|
525 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
|
|
526
|
|
527 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
|
|
528 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset());
|
|
529
|
|
530 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
|
|
531 __ increment(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset()));
|
|
532 }
|
|
533 // Update standard invocation counters
|
|
534 __ movl(rax, backedge_counter); // load backedge counter
|
|
535
|
|
536 __ increment(rcx, InvocationCounter::count_increment);
|
|
537 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
|
|
538
|
|
539 __ movl(invocation_counter, rcx); // save invocation count
|
|
540 __ addl(rcx, rax); // add both counters
|
|
541
|
|
542 // profile_method is non-null only for interpreted method so
|
|
543 // profile_method != NULL == !native_call
|
|
544 // BytecodeInterpreter only calls for native so code is elided.
|
|
545
|
|
546 __ cmp32(rcx,
|
|
547 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
|
|
548 __ jcc(Assembler::aboveEqual, *overflow);
|
|
549
|
|
550 }
|
|
551
|
|
552 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
|
|
553
|
|
554 // C++ interpreter on entry
|
|
555 // rsi - new interpreter state pointer
|
|
556 // rbp - interpreter frame pointer
|
|
557 // rbx - method
|
|
558
|
|
559 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
|
|
560 // rbx, - method
|
|
561 // rcx - rcvr (assuming there is one)
|
|
562 // top of stack return address of interpreter caller
|
|
563 // rsp - sender_sp
|
|
564
|
|
565 // C++ interpreter only
|
|
566 // rsi - previous interpreter state pointer
|
|
567
|
|
568 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
|
|
569
|
|
570 // InterpreterRuntime::frequency_counter_overflow takes one argument
|
|
571 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
|
|
572 // The call returns the address of the verified entry point for the method or NULL
|
|
573 // if the compilation did not complete (either went background or bailed out).
|
|
574 __ movl(rax, (int)false);
|
|
575 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
|
|
576
|
|
577 // for c++ interpreter can rsi really be munged?
|
|
578 __ leal(rsi, Address(rbp, -sizeof(BytecodeInterpreter))); // restore state
|
|
579 __ movl(rbx, Address(rsi, byte_offset_of(BytecodeInterpreter, _method))); // restore method
|
|
580 __ movl(rdi, Address(rsi, byte_offset_of(BytecodeInterpreter, _locals))); // get locals pointer
|
|
581
|
|
582 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
|
|
583 // and jump to the interpreted entry.
|
|
584 __ jmp(*do_continue, relocInfo::none);
|
|
585
|
|
586 }
|
|
587
|
|
588 void InterpreterGenerator::generate_stack_overflow_check(void) {
|
|
589 // see if we've got enough room on the stack for locals plus overhead.
|
|
590 // the expression stack grows down incrementally, so the normal guard
|
|
591 // page mechanism will work for that.
|
|
592 //
|
|
593 // Registers live on entry:
|
|
594 //
|
|
595 // Asm interpreter
|
|
596 // rdx: number of additional locals this frame needs (what we must check)
|
|
597 // rbx,: methodOop
|
|
598
|
|
599 // C++ Interpreter
|
|
600 // rsi: previous interpreter frame state object
|
|
601 // rdi: &locals[0]
|
|
602 // rcx: # of locals
|
|
603 // rdx: number of additional locals this frame needs (what we must check)
|
|
604 // rbx: methodOop
|
|
605
|
|
606 // destroyed on exit
|
|
607 // rax,
|
|
608
|
|
609 // NOTE: since the additional locals are also always pushed (wasn't obvious in
|
|
610 // generate_method_entry) so the guard should work for them too.
|
|
611 //
|
|
612
|
|
613 // monitor entry size: see picture of stack set (generate_method_entry) and frame_i486.hpp
|
|
614 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
|
|
615
|
|
616 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
|
|
617 // be sure to change this if you add/subtract anything to/from the overhead area
|
|
618 const int overhead_size = (int)sizeof(BytecodeInterpreter);
|
|
619
|
|
620 const int page_size = os::vm_page_size();
|
|
621
|
|
622 Label after_frame_check;
|
|
623
|
|
624 // compute rsp as if this were going to be the last frame on
|
|
625 // the stack before the red zone
|
|
626
|
|
627 Label after_frame_check_pop;
|
|
628
|
|
629 // save rsi == caller's bytecode ptr (c++ previous interp. state)
|
|
630 // QQQ problem here?? rsi overload????
|
|
631 __ pushl(rsi);
|
|
632
|
|
633 const Register thread = rsi;
|
|
634
|
|
635 __ get_thread(thread);
|
|
636
|
|
637 const Address stack_base(thread, Thread::stack_base_offset());
|
|
638 const Address stack_size(thread, Thread::stack_size_offset());
|
|
639
|
|
640 // locals + overhead, in bytes
|
|
641 const Address size_of_stack (rbx, methodOopDesc::max_stack_offset());
|
|
642 // Always give one monitor to allow us to start interp if sync method.
|
|
643 // Any additional monitors need a check when moving the expression stack
|
|
644 const one_monitor = frame::interpreter_frame_monitor_size() * wordSize;
|
|
645 __ load_unsigned_word(rax, size_of_stack); // get size of expression stack in words
|
|
646 __ leal(rax, Address(noreg, rax, Interpreter::stackElementScale(), one_monitor));
|
|
647 __ leal(rax, Address(rax, rdx, Interpreter::stackElementScale(), overhead_size));
|
|
648
|
|
649 #ifdef ASSERT
|
|
650 Label stack_base_okay, stack_size_okay;
|
|
651 // verify that thread stack base is non-zero
|
|
652 __ cmpl(stack_base, 0);
|
|
653 __ jcc(Assembler::notEqual, stack_base_okay);
|
|
654 __ stop("stack base is zero");
|
|
655 __ bind(stack_base_okay);
|
|
656 // verify that thread stack size is non-zero
|
|
657 __ cmpl(stack_size, 0);
|
|
658 __ jcc(Assembler::notEqual, stack_size_okay);
|
|
659 __ stop("stack size is zero");
|
|
660 __ bind(stack_size_okay);
|
|
661 #endif
|
|
662
|
|
663 // Add stack base to locals and subtract stack size
|
|
664 __ addl(rax, stack_base);
|
|
665 __ subl(rax, stack_size);
|
|
666
|
|
667 // We should have a magic number here for the size of the c++ interpreter frame.
|
|
668 // We can't actually tell this ahead of time. The debug version size is around 3k
|
|
669 // product is 1k and fastdebug is 4k
|
|
670 const int slop = 6 * K;
|
|
671
|
|
672 // Use the maximum number of pages we might bang.
|
|
673 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
|
|
674 (StackRedPages+StackYellowPages);
|
|
675 // Only need this if we are stack banging which is temporary while
|
|
676 // we're debugging.
|
|
677 __ addl(rax, slop + 2*max_pages * page_size);
|
|
678
|
|
679 // check against the current stack bottom
|
|
680 __ cmpl(rsp, rax);
|
|
681 __ jcc(Assembler::above, after_frame_check_pop);
|
|
682
|
|
683 __ popl(rsi); // get saved bcp / (c++ prev state ).
|
|
684
|
|
685 // throw exception return address becomes throwing pc
|
|
686 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
|
|
687
|
|
688 // all done with frame size check
|
|
689 __ bind(after_frame_check_pop);
|
|
690 __ popl(rsi);
|
|
691
|
|
692 __ bind(after_frame_check);
|
|
693 }
|
|
694
|
|
695 // Find preallocated monitor and lock method (C++ interpreter)
|
|
696 // rbx - methodOop
|
|
697 //
|
|
698 void InterpreterGenerator::lock_method(void) {
|
|
699 // assumes state == rsi == pointer to current interpreterState
|
|
700 // minimally destroys rax, rdx, rdi
|
|
701 //
|
|
702 // synchronize method
|
|
703 const Register state = rsi;
|
|
704 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
|
|
705 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
|
|
706
|
|
707 // find initial monitor i.e. monitors[-1]
|
|
708 __ movl(rdx, STATE(_monitor_base)); // get monitor bottom limit
|
|
709 __ subl(rdx, entry_size); // point to initial monitor
|
|
710
|
|
711 #ifdef ASSERT
|
|
712 { Label L;
|
|
713 __ movl(rax, access_flags);
|
|
714 __ testl(rax, JVM_ACC_SYNCHRONIZED);
|
|
715 __ jcc(Assembler::notZero, L);
|
|
716 __ stop("method doesn't need synchronization");
|
|
717 __ bind(L);
|
|
718 }
|
|
719 #endif // ASSERT
|
|
720 // get synchronization object
|
|
721 { Label done;
|
|
722 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
|
|
723 __ movl(rax, access_flags);
|
|
724 __ movl(rdi, STATE(_locals)); // prepare to get receiver (assume common case)
|
|
725 __ testl(rax, JVM_ACC_STATIC);
|
|
726 __ movl(rax, Address(rdi, 0)); // get receiver (assume this is frequent case)
|
|
727 __ jcc(Assembler::zero, done);
|
|
728 __ movl(rax, Address(rbx, methodOopDesc::constants_offset()));
|
|
729 __ movl(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
|
|
730 __ movl(rax, Address(rax, mirror_offset));
|
|
731 __ bind(done);
|
|
732 }
|
|
733 #ifdef ASSERT
|
|
734 { Label L;
|
|
735 __ cmpl(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); // correct object?
|
|
736 __ jcc(Assembler::equal, L);
|
|
737 __ stop("wrong synchronization lobject");
|
|
738 __ bind(L);
|
|
739 }
|
|
740 #endif // ASSERT
|
|
741 // can destroy rax, rdx, rcx, and (via call_VM) rdi!
|
|
742 __ lock_object(rdx);
|
|
743 }
|
|
744
|
|
745 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
|
|
746
|
|
747 address InterpreterGenerator::generate_accessor_entry(void) {
|
|
748
|
|
749 // rbx,: methodOop
|
|
750 // rcx: receiver (preserve for slow entry into asm interpreter)
|
|
751
|
|
752 // rsi: senderSP must preserved for slow path, set SP to it on fast path
|
|
753
|
|
754 Label xreturn_path;
|
|
755
|
|
756 // do fastpath for resolved accessor methods
|
|
757 if (UseFastAccessorMethods) {
|
|
758
|
|
759 address entry_point = __ pc();
|
|
760
|
|
761 Label slow_path;
|
|
762 // If we need a safepoint check, generate full interpreter entry.
|
|
763 ExternalAddress state(SafepointSynchronize::address_of_state());
|
|
764 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
|
|
765 SafepointSynchronize::_not_synchronized);
|
|
766
|
|
767 __ jcc(Assembler::notEqual, slow_path);
|
|
768 // ASM/C++ Interpreter
|
|
769 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
|
|
770 // Note: We can only use this code if the getfield has been resolved
|
|
771 // and if we don't have a null-pointer exception => check for
|
|
772 // these conditions first and use slow path if necessary.
|
|
773 // rbx,: method
|
|
774 // rcx: receiver
|
|
775 __ movl(rax, Address(rsp, wordSize));
|
|
776
|
|
777 // check if local 0 != NULL and read field
|
|
778 __ testl(rax, rax);
|
|
779 __ jcc(Assembler::zero, slow_path);
|
|
780
|
|
781 __ movl(rdi, Address(rbx, methodOopDesc::constants_offset()));
|
|
782 // read first instruction word and extract bytecode @ 1 and index @ 2
|
|
783 __ movl(rdx, Address(rbx, methodOopDesc::const_offset()));
|
|
784 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
|
|
785 // Shift codes right to get the index on the right.
|
|
786 // The bytecode fetched looks like <index><0xb4><0x2a>
|
|
787 __ shrl(rdx, 2*BitsPerByte);
|
|
788 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
|
|
789 __ movl(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
|
|
790
|
|
791 // rax,: local 0
|
|
792 // rbx,: method
|
|
793 // rcx: receiver - do not destroy since it is needed for slow path!
|
|
794 // rcx: scratch
|
|
795 // rdx: constant pool cache index
|
|
796 // rdi: constant pool cache
|
|
797 // rsi: sender sp
|
|
798
|
|
799 // check if getfield has been resolved and read constant pool cache entry
|
|
800 // check the validity of the cache entry by testing whether _indices field
|
|
801 // contains Bytecode::_getfield in b1 byte.
|
|
802 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
|
|
803 __ movl(rcx,
|
|
804 Address(rdi,
|
|
805 rdx,
|
|
806 Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
|
|
807 __ shrl(rcx, 2*BitsPerByte);
|
|
808 __ andl(rcx, 0xFF);
|
|
809 __ cmpl(rcx, Bytecodes::_getfield);
|
|
810 __ jcc(Assembler::notEqual, slow_path);
|
|
811
|
|
812 // Note: constant pool entry is not valid before bytecode is resolved
|
|
813 __ movl(rcx,
|
|
814 Address(rdi,
|
|
815 rdx,
|
|
816 Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()));
|
|
817 __ movl(rdx,
|
|
818 Address(rdi,
|
|
819 rdx,
|
|
820 Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset()));
|
|
821
|
|
822 Label notByte, notShort, notChar;
|
|
823 const Address field_address (rax, rcx, Address::times_1);
|
|
824
|
|
825 // Need to differentiate between igetfield, agetfield, bgetfield etc.
|
|
826 // because they are different sizes.
|
|
827 // Use the type from the constant pool cache
|
|
828 __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
|
|
829 // Make sure we don't need to mask rdx for tosBits after the above shift
|
|
830 ConstantPoolCacheEntry::verify_tosBits();
|
|
831 __ cmpl(rdx, btos);
|
|
832 __ jcc(Assembler::notEqual, notByte);
|
|
833 __ load_signed_byte(rax, field_address);
|
|
834 __ jmp(xreturn_path);
|
|
835
|
|
836 __ bind(notByte);
|
|
837 __ cmpl(rdx, stos);
|
|
838 __ jcc(Assembler::notEqual, notShort);
|
|
839 __ load_signed_word(rax, field_address);
|
|
840 __ jmp(xreturn_path);
|
|
841
|
|
842 __ bind(notShort);
|
|
843 __ cmpl(rdx, ctos);
|
|
844 __ jcc(Assembler::notEqual, notChar);
|
|
845 __ load_unsigned_word(rax, field_address);
|
|
846 __ jmp(xreturn_path);
|
|
847
|
|
848 __ bind(notChar);
|
|
849 #ifdef ASSERT
|
|
850 Label okay;
|
|
851 __ cmpl(rdx, atos);
|
|
852 __ jcc(Assembler::equal, okay);
|
|
853 __ cmpl(rdx, itos);
|
|
854 __ jcc(Assembler::equal, okay);
|
|
855 __ stop("what type is this?");
|
|
856 __ bind(okay);
|
|
857 #endif // ASSERT
|
|
858 // All the rest are a 32 bit wordsize
|
|
859 __ movl(rax, field_address);
|
|
860
|
|
861 __ bind(xreturn_path);
|
|
862
|
|
863 // _ireturn/_areturn
|
|
864 __ popl(rdi); // get return address
|
|
865 __ movl(rsp, rsi); // set sp to sender sp
|
|
866 __ jmp(rdi);
|
|
867
|
|
868 // generate a vanilla interpreter entry as the slow path
|
|
869 __ bind(slow_path);
|
|
870 // We will enter c++ interpreter looking like it was
|
|
871 // called by the call_stub this will cause it to return
|
|
872 // a tosca result to the invoker which might have been
|
|
873 // the c++ interpreter itself.
|
|
874
|
|
875 __ jmp(fast_accessor_slow_entry_path);
|
|
876 return entry_point;
|
|
877
|
|
878 } else {
|
|
879 return NULL;
|
|
880 }
|
|
881
|
|
882 }
|
|
883
|
|
884 //
|
|
885 // C++ Interpreter stub for calling a native method.
|
|
886 // This sets up a somewhat different looking stack for calling the native method
|
|
887 // than the typical interpreter frame setup but still has the pointer to
|
|
888 // an interpreter state.
|
|
889 //
|
|
890
|
|
891 address InterpreterGenerator::generate_native_entry(bool synchronized) {
|
|
892 // determine code generation flags
|
|
893 bool inc_counter = UseCompiler || CountCompiledCalls;
|
|
894
|
|
895 // rbx: methodOop
|
|
896 // rcx: receiver (unused)
|
|
897 // rsi: previous interpreter state (if called from C++ interpreter) must preserve
|
|
898 // in any case. If called via c1/c2/call_stub rsi is junk (to use) but harmless
|
|
899 // to save/restore.
|
|
900 address entry_point = __ pc();
|
|
901
|
|
902 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
|
|
903 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset());
|
|
904 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
|
|
905 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
|
|
906
|
|
907 // rsi == state/locals rdi == prevstate
|
|
908 const Register state = rsi;
|
|
909 const Register locals = rdi;
|
|
910
|
|
911 // get parameter size (always needed)
|
|
912 __ load_unsigned_word(rcx, size_of_parameters);
|
|
913
|
|
914 // rbx: methodOop
|
|
915 // rcx: size of parameters
|
|
916 __ popl(rax); // get return address
|
|
917 // for natives the size of locals is zero
|
|
918
|
|
919 // compute beginning of parameters /locals
|
|
920 __ leal(locals, Address(rsp, rcx, Address::times_4, -wordSize));
|
|
921
|
|
922 // initialize fixed part of activation frame
|
|
923
|
|
924 // Assumes rax = return address
|
|
925
|
|
926 // allocate and initialize new interpreterState and method expression stack
|
|
927 // IN(locals) -> locals
|
|
928 // IN(state) -> previous frame manager state (NULL from stub/c1/c2)
|
|
929 // destroys rax, rcx, rdx
|
|
930 // OUT (state) -> new interpreterState
|
|
931 // OUT(rsp) -> bottom of methods expression stack
|
|
932
|
|
933 // save sender_sp
|
|
934 __ movl(rcx, rsi);
|
|
935 // start with NULL previous state
|
|
936 __ movl(state, 0);
|
|
937 generate_compute_interpreter_state(state, locals, rcx, true);
|
|
938
|
|
939 #ifdef ASSERT
|
|
940 { Label L;
|
|
941 __ movl(rax, STATE(_stack_base));
|
|
942 __ cmpl(rax, rsp);
|
|
943 __ jcc(Assembler::equal, L);
|
|
944 __ stop("broken stack frame setup in interpreter");
|
|
945 __ bind(L);
|
|
946 }
|
|
947 #endif
|
|
948
|
|
949 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
|
|
950
|
|
951 __ movl(rax, STATE(_thread)); // get thread
|
|
952 // Since at this point in the method invocation the exception handler
|
|
953 // would try to exit the monitor of synchronized methods which hasn't
|
|
954 // been entered yet, we set the thread local variable
|
|
955 // _do_not_unlock_if_synchronized to true. The remove_activation will
|
|
956 // check this flag.
|
|
957
|
|
958 const Address do_not_unlock_if_synchronized(rax,
|
|
959 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
|
|
960 __ movbool(do_not_unlock_if_synchronized, true);
|
|
961
|
|
962 // make sure method is native & not abstract
|
|
963 #ifdef ASSERT
|
|
964 __ movl(rax, access_flags);
|
|
965 {
|
|
966 Label L;
|
|
967 __ testl(rax, JVM_ACC_NATIVE);
|
|
968 __ jcc(Assembler::notZero, L);
|
|
969 __ stop("tried to execute non-native method as native");
|
|
970 __ bind(L);
|
|
971 }
|
|
972 { Label L;
|
|
973 __ testl(rax, JVM_ACC_ABSTRACT);
|
|
974 __ jcc(Assembler::zero, L);
|
|
975 __ stop("tried to execute abstract method in interpreter");
|
|
976 __ bind(L);
|
|
977 }
|
|
978 #endif
|
|
979
|
|
980
|
|
981 // increment invocation count & check for overflow
|
|
982 Label invocation_counter_overflow;
|
|
983 if (inc_counter) {
|
|
984 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
|
|
985 }
|
|
986
|
|
987 Label continue_after_compile;
|
|
988
|
|
989 __ bind(continue_after_compile);
|
|
990
|
|
991 bang_stack_shadow_pages(true);
|
|
992
|
|
993 // reset the _do_not_unlock_if_synchronized flag
|
|
994 __ movl(rax, STATE(_thread)); // get thread
|
|
995 __ movbool(do_not_unlock_if_synchronized, false);
|
|
996
|
|
997
|
|
998 // check for synchronized native methods
|
|
999 //
|
|
1000 // Note: This must happen *after* invocation counter check, since
|
|
1001 // when overflow happens, the method should not be locked.
|
|
1002 if (synchronized) {
|
|
1003 // potentially kills rax, rcx, rdx, rdi
|
|
1004 lock_method();
|
|
1005 } else {
|
|
1006 // no synchronization necessary
|
|
1007 #ifdef ASSERT
|
|
1008 { Label L;
|
|
1009 __ movl(rax, access_flags);
|
|
1010 __ testl(rax, JVM_ACC_SYNCHRONIZED);
|
|
1011 __ jcc(Assembler::zero, L);
|
|
1012 __ stop("method needs synchronization");
|
|
1013 __ bind(L);
|
|
1014 }
|
|
1015 #endif
|
|
1016 }
|
|
1017
|
|
1018 // start execution
|
|
1019
|
|
1020 // jvmti support
|
|
1021 __ notify_method_entry();
|
|
1022
|
|
1023 // work registers
|
|
1024 const Register method = rbx;
|
|
1025 const Register thread = rdi;
|
|
1026 const Register t = rcx;
|
|
1027
|
|
1028 // allocate space for parameters
|
|
1029 __ movl(method, STATE(_method));
|
|
1030 __ verify_oop(method);
|
|
1031 __ load_unsigned_word(t, Address(method, methodOopDesc::size_of_parameters_offset()));
|
|
1032 __ shll(t, 2);
|
|
1033 __ addl(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
|
|
1034 __ subl(rsp, t);
|
|
1035 __ andl(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
|
|
1036
|
|
1037 // get signature handler
|
|
1038 Label pending_exception_present;
|
|
1039
|
|
1040 { Label L;
|
|
1041 __ movl(t, Address(method, methodOopDesc::signature_handler_offset()));
|
|
1042 __ testl(t, t);
|
|
1043 __ jcc(Assembler::notZero, L);
|
|
1044 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method, false);
|
|
1045 __ movl(method, STATE(_method));
|
|
1046 __ cmpl(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
|
|
1047 __ jcc(Assembler::notEqual, pending_exception_present);
|
|
1048 __ verify_oop(method);
|
|
1049 __ movl(t, Address(method, methodOopDesc::signature_handler_offset()));
|
|
1050 __ bind(L);
|
|
1051 }
|
|
1052 #ifdef ASSERT
|
|
1053 {
|
|
1054 Label L;
|
|
1055 __ pushl(t);
|
|
1056 __ get_thread(t); // get vm's javathread*
|
|
1057 __ cmpl(t, STATE(_thread));
|
|
1058 __ jcc(Assembler::equal, L);
|
|
1059 __ int3();
|
|
1060 __ bind(L);
|
|
1061 __ popl(t);
|
|
1062 }
|
|
1063 #endif //
|
|
1064
|
|
1065 // call signature handler
|
|
1066 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
|
|
1067 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
|
|
1068 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
|
|
1069 // The generated handlers do not touch RBX (the method oop).
|
|
1070 // However, large signatures cannot be cached and are generated
|
|
1071 // each time here. The slow-path generator will blow RBX
|
|
1072 // sometime, so we must reload it after the call.
|
|
1073 __ movl(rdi, STATE(_locals)); // get the from pointer
|
|
1074 __ call(t);
|
|
1075 __ movl(method, STATE(_method));
|
|
1076 __ verify_oop(method);
|
|
1077
|
|
1078 // result handler is in rax
|
|
1079 // set result handler
|
|
1080 __ movl(STATE(_result_handler), rax);
|
|
1081
|
|
1082 // pass mirror handle if static call
|
|
1083 { Label L;
|
|
1084 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
|
|
1085 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
|
|
1086 __ testl(t, JVM_ACC_STATIC);
|
|
1087 __ jcc(Assembler::zero, L);
|
|
1088 // get mirror
|
|
1089 __ movl(t, Address(method, methodOopDesc:: constants_offset()));
|
|
1090 __ movl(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
|
|
1091 __ movl(t, Address(t, mirror_offset));
|
|
1092 // copy mirror into activation object
|
|
1093 __ movl(STATE(_oop_temp), t);
|
|
1094 // pass handle to mirror
|
|
1095 __ leal(t, STATE(_oop_temp));
|
|
1096 __ movl(Address(rsp, wordSize), t);
|
|
1097 __ bind(L);
|
|
1098 }
|
|
1099 #ifdef ASSERT
|
|
1100 {
|
|
1101 Label L;
|
|
1102 __ pushl(t);
|
|
1103 __ get_thread(t); // get vm's javathread*
|
|
1104 __ cmpl(t, STATE(_thread));
|
|
1105 __ jcc(Assembler::equal, L);
|
|
1106 __ int3();
|
|
1107 __ bind(L);
|
|
1108 __ popl(t);
|
|
1109 }
|
|
1110 #endif //
|
|
1111
|
|
1112 // get native function entry point
|
|
1113 { Label L;
|
|
1114 __ movl(rax, Address(method, methodOopDesc::native_function_offset()));
|
|
1115 __ testl(rax, rax);
|
|
1116 __ jcc(Assembler::notZero, L);
|
|
1117 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
|
|
1118 __ movl(method, STATE(_method));
|
|
1119 __ verify_oop(method);
|
|
1120 __ movl(rax, Address(method, methodOopDesc::native_function_offset()));
|
|
1121 __ bind(L);
|
|
1122 }
|
|
1123
|
|
1124 // pass JNIEnv
|
|
1125 __ movl(thread, STATE(_thread)); // get thread
|
|
1126 __ leal(t, Address(thread, JavaThread::jni_environment_offset()));
|
|
1127 __ movl(Address(rsp, 0), t);
|
|
1128 #ifdef ASSERT
|
|
1129 {
|
|
1130 Label L;
|
|
1131 __ pushl(t);
|
|
1132 __ get_thread(t); // get vm's javathread*
|
|
1133 __ cmpl(t, STATE(_thread));
|
|
1134 __ jcc(Assembler::equal, L);
|
|
1135 __ int3();
|
|
1136 __ bind(L);
|
|
1137 __ popl(t);
|
|
1138 }
|
|
1139 #endif //
|
|
1140
|
|
1141 #ifdef ASSERT
|
|
1142 { Label L;
|
|
1143 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
|
|
1144 __ cmpl(t, _thread_in_Java);
|
|
1145 __ jcc(Assembler::equal, L);
|
|
1146 __ stop("Wrong thread state in native stub");
|
|
1147 __ bind(L);
|
|
1148 }
|
|
1149 #endif
|
|
1150
|
|
1151 // Change state to native (we save the return address in the thread, since it might not
|
|
1152 // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
|
|
1153 // points into the right code segment. It does not have to be the correct return pc.
|
|
1154
|
|
1155 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
|
|
1156
|
|
1157 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
|
|
1158
|
|
1159 __ call(rax);
|
|
1160
|
|
1161 // result potentially in rdx:rax or ST0
|
|
1162 __ movl(method, STATE(_method));
|
|
1163 __ movl(thread, STATE(_thread)); // get thread
|
|
1164
|
|
1165 // The potential result is in ST(0) & rdx:rax
|
|
1166 // With C++ interpreter we leave any possible result in ST(0) until we are in result handler and then
|
|
1167 // we do the appropriate stuff for returning the result. rdx:rax must always be saved because just about
|
|
1168 // anything we do here will destroy it, st(0) is only saved if we re-enter the vm where it would
|
|
1169 // be destroyed.
|
|
1170 // It is safe to do these pushes because state is _thread_in_native and return address will be found
|
|
1171 // via _last_native_pc and not via _last_jave_sp
|
|
1172
|
|
1173 // Must save the value of ST(0) since it could be destroyed before we get to result handler
|
|
1174 { Label Lpush, Lskip;
|
|
1175 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
|
|
1176 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
|
|
1177 __ cmpptr(STATE(_result_handler), float_handler.addr());
|
|
1178 __ jcc(Assembler::equal, Lpush);
|
|
1179 __ cmpptr(STATE(_result_handler), double_handler.addr());
|
|
1180 __ jcc(Assembler::notEqual, Lskip);
|
|
1181 __ bind(Lpush);
|
|
1182 __ push(dtos);
|
|
1183 __ bind(Lskip);
|
|
1184 }
|
|
1185
|
|
1186 __ push(ltos); // save rax:rdx for potential use by result handler.
|
|
1187
|
|
1188 // Either restore the MXCSR register after returning from the JNI Call
|
|
1189 // or verify that it wasn't changed.
|
|
1190 if (VM_Version::supports_sse()) {
|
|
1191 if (RestoreMXCSROnJNICalls) {
|
|
1192 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
|
|
1193 }
|
|
1194 else if (CheckJNICalls ) {
|
|
1195 __ call(RuntimeAddress(StubRoutines::i486::verify_mxcsr_entry()));
|
|
1196 }
|
|
1197 }
|
|
1198
|
|
1199 // Either restore the x87 floating pointer control word after returning
|
|
1200 // from the JNI call or verify that it wasn't changed.
|
|
1201 if (CheckJNICalls) {
|
|
1202 __ call(RuntimeAddress(StubRoutines::i486::verify_fpu_cntrl_wrd_entry()));
|
|
1203 }
|
|
1204
|
|
1205
|
|
1206 // change thread state
|
|
1207 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
|
|
1208 if(os::is_MP()) {
|
|
1209 // Write serialization page so VM thread can do a pseudo remote membar.
|
|
1210 // We use the current thread pointer to calculate a thread specific
|
|
1211 // offset to write to within the page. This minimizes bus traffic
|
|
1212 // due to cache line collision.
|
|
1213 __ serialize_memory(thread, rcx);
|
|
1214 }
|
|
1215
|
|
1216 // check for safepoint operation in progress and/or pending suspend requests
|
|
1217 { Label Continue;
|
|
1218
|
|
1219 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
|
|
1220 SafepointSynchronize::_not_synchronized);
|
|
1221
|
|
1222 // threads running native code and they are expected to self-suspend
|
|
1223 // when leaving the _thread_in_native state. We need to check for
|
|
1224 // pending suspend requests here.
|
|
1225 Label L;
|
|
1226 __ jcc(Assembler::notEqual, L);
|
|
1227 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
|
|
1228 __ jcc(Assembler::equal, Continue);
|
|
1229 __ bind(L);
|
|
1230
|
|
1231 // Don't use call_VM as it will see a possible pending exception and forward it
|
|
1232 // and never return here preventing us from clearing _last_native_pc down below.
|
|
1233 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
|
|
1234 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
|
|
1235 // by hand.
|
|
1236 //
|
|
1237 __ pushl(thread);
|
|
1238 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
|
|
1239 JavaThread::check_special_condition_for_native_trans)));
|
|
1240 __ increment(rsp, wordSize);
|
|
1241
|
|
1242 __ movl(method, STATE(_method));
|
|
1243 __ verify_oop(method);
|
|
1244 __ movl(thread, STATE(_thread)); // get thread
|
|
1245
|
|
1246 __ bind(Continue);
|
|
1247 }
|
|
1248
|
|
1249 // change thread state
|
|
1250 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
|
|
1251
|
|
1252 __ reset_last_Java_frame(thread, true, true);
|
|
1253
|
|
1254 // reset handle block
|
|
1255 __ movl(t, Address(thread, JavaThread::active_handles_offset()));
|
|
1256 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
|
|
1257
|
|
1258 // If result was an oop then unbox and save it in the frame
|
|
1259 { Label L;
|
|
1260 Label no_oop, store_result;
|
|
1261 ExternalAddress oop_handler(AbstractInterpreter::result_handler(T_OBJECT));
|
|
1262 __ cmpptr(STATE(_result_handler), oop_handler.addr());
|
|
1263 __ jcc(Assembler::notEqual, no_oop);
|
|
1264 __ pop(ltos);
|
|
1265 __ testl(rax, rax);
|
|
1266 __ jcc(Assembler::zero, store_result);
|
|
1267 // unbox
|
|
1268 __ movl(rax, Address(rax, 0));
|
|
1269 __ bind(store_result);
|
|
1270 __ movl(STATE(_oop_temp), rax);
|
|
1271 // keep stack depth as expected by pushing oop which will eventually be discarded
|
|
1272 __ push(ltos);
|
|
1273 __ bind(no_oop);
|
|
1274 }
|
|
1275
|
|
1276 {
|
|
1277 Label no_reguard;
|
|
1278 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
|
|
1279 __ jcc(Assembler::notEqual, no_reguard);
|
|
1280
|
|
1281 __ pushad();
|
|
1282 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
|
|
1283 __ popad();
|
|
1284
|
|
1285 __ bind(no_reguard);
|
|
1286 }
|
|
1287
|
|
1288
|
|
1289 // QQQ Seems like for native methods we simply return and the caller will see the pending
|
|
1290 // exception and do the right thing. Certainly the interpreter will, don't know about
|
|
1291 // compiled methods.
|
|
1292 // Seems that the answer to above is no this is wrong. The old code would see the exception
|
|
1293 // and forward it before doing the unlocking and notifying jvmdi that method has exited.
|
|
1294 // This seems wrong need to investigate the spec.
|
|
1295
|
|
1296 // handle exceptions (exception handling will handle unlocking!)
|
|
1297 { Label L;
|
|
1298 __ cmpl(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
|
|
1299 __ jcc(Assembler::zero, L);
|
|
1300 __ bind(pending_exception_present);
|
|
1301
|
|
1302 // There are potential results on the stack (rax/rdx, ST(0)) we ignore these and simply
|
|
1303 // return and let caller deal with exception. This skips the unlocking here which
|
|
1304 // seems wrong but seems to be what asm interpreter did. Can't find this in the spec.
|
|
1305 // Note: must preverve method in rbx
|
|
1306 //
|
|
1307
|
|
1308 // remove activation
|
|
1309
|
|
1310 __ movl(t, STATE(_sender_sp));
|
|
1311 __ leave(); // remove frame anchor
|
|
1312 __ popl(rdi); // get return address
|
|
1313 __ movl(state, STATE(_prev_link)); // get previous state for return
|
|
1314 __ movl(rsp, t); // set sp to sender sp
|
|
1315 __ pushl(rdi); // [ush throwing pc
|
|
1316 // The skips unlocking!! This seems to be what asm interpreter does but seems
|
|
1317 // very wrong. Not clear if this violates the spec.
|
|
1318 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
|
|
1319 __ bind(L);
|
|
1320 }
|
|
1321
|
|
1322 // do unlocking if necessary
|
|
1323 { Label L;
|
|
1324 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
|
|
1325 __ testl(t, JVM_ACC_SYNCHRONIZED);
|
|
1326 __ jcc(Assembler::zero, L);
|
|
1327 // the code below should be shared with interpreter macro assembler implementation
|
|
1328 { Label unlock;
|
|
1329 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
|
|
1330 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
|
|
1331 __ movl(rdx, STATE(_monitor_base));
|
|
1332 __ subl(rdx, frame::interpreter_frame_monitor_size() * wordSize); // address of initial monitor
|
|
1333
|
|
1334 __ movl(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
|
|
1335 __ testl(t, t);
|
|
1336 __ jcc(Assembler::notZero, unlock);
|
|
1337
|
|
1338 // Entry already unlocked, need to throw exception
|
|
1339 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
|
|
1340 __ should_not_reach_here();
|
|
1341
|
|
1342 __ bind(unlock);
|
|
1343 __ unlock_object(rdx);
|
|
1344 // unlock can blow rbx so restore it for path that needs it below
|
|
1345 __ movl(method, STATE(_method));
|
|
1346 }
|
|
1347 __ bind(L);
|
|
1348 }
|
|
1349
|
|
1350 // jvmti support
|
|
1351 // Note: This must happen _after_ handling/throwing any exceptions since
|
|
1352 // the exception handler code notifies the runtime of method exits
|
|
1353 // too. If this happens before, method entry/exit notifications are
|
|
1354 // not properly paired (was bug - gri 11/22/99).
|
|
1355 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
|
|
1356
|
|
1357 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
|
|
1358 __ pop(ltos); // restore rax/rdx floating result if present still on stack
|
|
1359 __ movl(t, STATE(_result_handler)); // get result handler
|
|
1360 __ call(t); // call result handler to convert to tosca form
|
|
1361
|
|
1362 // remove activation
|
|
1363
|
|
1364 __ movl(t, STATE(_sender_sp));
|
|
1365
|
|
1366 __ leave(); // remove frame anchor
|
|
1367 __ popl(rdi); // get return address
|
|
1368 __ movl(state, STATE(_prev_link)); // get previous state for return (if c++ interpreter was caller)
|
|
1369 __ movl(rsp, t); // set sp to sender sp
|
|
1370 __ jmp(rdi);
|
|
1371
|
|
1372 // invocation counter overflow
|
|
1373 if (inc_counter) {
|
|
1374 // Handle overflow of counter and compile method
|
|
1375 __ bind(invocation_counter_overflow);
|
|
1376 generate_counter_overflow(&continue_after_compile);
|
|
1377 }
|
|
1378
|
|
1379 return entry_point;
|
|
1380 }
|
|
1381
|
|
1382 // Generate entries that will put a result type index into rcx
|
|
1383 void CppInterpreterGenerator::generate_deopt_handling() {
|
|
1384
|
|
1385 const Register state = rsi;
|
|
1386 Label return_from_deopt_common;
|
|
1387
|
|
1388 // Generate entries that will put a result type index into rcx
|
|
1389 // deopt needs to jump to here to enter the interpreter (return a result)
|
|
1390 deopt_frame_manager_return_atos = __ pc();
|
|
1391
|
|
1392 // rax is live here
|
|
1393 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_OBJECT)); // Result stub address array index
|
|
1394 __ jmp(return_from_deopt_common);
|
|
1395
|
|
1396
|
|
1397 // deopt needs to jump to here to enter the interpreter (return a result)
|
|
1398 deopt_frame_manager_return_btos = __ pc();
|
|
1399
|
|
1400 // rax is live here
|
|
1401 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_BOOLEAN)); // Result stub address array index
|
|
1402 __ jmp(return_from_deopt_common);
|
|
1403
|
|
1404 // deopt needs to jump to here to enter the interpreter (return a result)
|
|
1405 deopt_frame_manager_return_itos = __ pc();
|
|
1406
|
|
1407 // rax is live here
|
|
1408 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_INT)); // Result stub address array index
|
|
1409 __ jmp(return_from_deopt_common);
|
|
1410
|
|
1411 // deopt needs to jump to here to enter the interpreter (return a result)
|
|
1412
|
|
1413 deopt_frame_manager_return_ltos = __ pc();
|
|
1414 // rax,rdx are live here
|
|
1415 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_LONG)); // Result stub address array index
|
|
1416 __ jmp(return_from_deopt_common);
|
|
1417
|
|
1418 // deopt needs to jump to here to enter the interpreter (return a result)
|
|
1419
|
|
1420 deopt_frame_manager_return_ftos = __ pc();
|
|
1421 // st(0) is live here
|
|
1422 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index
|
|
1423 __ jmp(return_from_deopt_common);
|
|
1424
|
|
1425 // deopt needs to jump to here to enter the interpreter (return a result)
|
|
1426 deopt_frame_manager_return_dtos = __ pc();
|
|
1427
|
|
1428 // st(0) is live here
|
|
1429 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index
|
|
1430 __ jmp(return_from_deopt_common);
|
|
1431
|
|
1432 // deopt needs to jump to here to enter the interpreter (return a result)
|
|
1433 deopt_frame_manager_return_vtos = __ pc();
|
|
1434
|
|
1435 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_VOID));
|
|
1436
|
|
1437 // Deopt return common
|
|
1438 // an index is present in rcx that lets us move any possible result being
|
|
1439 // return to the interpreter's stack
|
|
1440 //
|
|
1441 // Because we have a full sized interpreter frame on the youngest
|
|
1442 // activation the stack is pushed too deep to share the tosca to
|
|
1443 // stack converters directly. We shrink the stack to the desired
|
|
1444 // amount and then push result and then re-extend the stack.
|
|
1445 // We could have the code in size_activation layout a short
|
|
1446 // frame for the top activation but that would look different
|
|
1447 // than say sparc (which needs a full size activation because
|
|
1448 // the windows are in the way. Really it could be short? QQQ
|
|
1449 //
|
|
1450 __ bind(return_from_deopt_common);
|
|
1451
|
|
1452 __ leal(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
|
|
1453
|
|
1454 // setup rsp so we can push the "result" as needed.
|
|
1455 __ movl(rsp, STATE(_stack)); // trim stack (is prepushed)
|
|
1456 __ addl(rsp, wordSize); // undo prepush
|
|
1457
|
|
1458 ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack);
|
|
1459 // Address index(noreg, rcx, Address::times_4);
|
|
1460 __ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_4)));
|
|
1461 // __ movl(rcx, Address(noreg, rcx, Address::times_4, int(AbstractInterpreter::_tosca_to_stack)));
|
|
1462 __ call(rcx); // call result converter
|
|
1463
|
|
1464 __ movl(STATE(_msg), (int)BytecodeInterpreter::deopt_resume);
|
|
1465 __ leal(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present)
|
|
1466 __ movl(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed,
|
|
1467 // result if any on stack already )
|
|
1468 __ movl(rsp, STATE(_stack_limit)); // restore expression stack to full depth
|
|
1469 }
|
|
1470
|
|
1471 // Generate the code to handle a more_monitors message from the c++ interpreter
|
|
1472 void CppInterpreterGenerator::generate_more_monitors() {
|
|
1473
|
|
1474 const Register state = rsi;
|
|
1475
|
|
1476 Label entry, loop;
|
|
1477 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
|
|
1478 // 1. compute new pointers // rsp: old expression stack top
|
|
1479 __ movl(rdx, STATE(_stack_base)); // rdx: old expression stack bottom
|
|
1480 __ subl(rsp, entry_size); // move expression stack top limit
|
|
1481 __ subl(STATE(_stack), entry_size); // update interpreter stack top
|
|
1482 __ movl(STATE(_stack_limit), rsp); // inform interpreter
|
|
1483 __ subl(rdx, entry_size); // move expression stack bottom
|
|
1484 __ movl(STATE(_stack_base), rdx); // inform interpreter
|
|
1485 __ movl(rcx, STATE(_stack)); // set start value for copy loop
|
|
1486 __ jmp(entry);
|
|
1487 // 2. move expression stack contents
|
|
1488 __ bind(loop);
|
|
1489 __ movl(rbx, Address(rcx, entry_size)); // load expression stack word from old location
|
|
1490 __ movl(Address(rcx, 0), rbx); // and store it at new location
|
|
1491 __ addl(rcx, wordSize); // advance to next word
|
|
1492 __ bind(entry);
|
|
1493 __ cmpl(rcx, rdx); // check if bottom reached
|
|
1494 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word
|
|
1495 // now zero the slot so we can find it.
|
|
1496 __ movl(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), (int) NULL);
|
|
1497 __ movl(STATE(_msg), (int)BytecodeInterpreter::got_monitors);
|
|
1498 }
|
|
1499
|
|
1500
|
|
1501 // Initial entry to C++ interpreter from the call_stub.
|
|
1502 // This entry point is called the frame manager since it handles the generation
|
|
1503 // of interpreter activation frames via requests directly from the vm (via call_stub)
|
|
1504 // and via requests from the interpreter. The requests from the call_stub happen
|
|
1505 // directly thru the entry point. Requests from the interpreter happen via returning
|
|
1506 // from the interpreter and examining the message the interpreter has returned to
|
|
1507 // the frame manager. The frame manager can take the following requests:
|
|
1508
|
|
1509 // NO_REQUEST - error, should never happen.
|
|
1510 // MORE_MONITORS - need a new monitor. Shuffle the expression stack on down and
|
|
1511 // allocate a new monitor.
|
|
1512 // CALL_METHOD - setup a new activation to call a new method. Very similar to what
|
|
1513 // happens during entry during the entry via the call stub.
|
|
1514 // RETURN_FROM_METHOD - remove an activation. Return to interpreter or call stub.
|
|
1515 //
|
|
1516 // Arguments:
|
|
1517 //
|
|
1518 // rbx: methodOop
|
|
1519 // rcx: receiver - unused (retrieved from stack as needed)
|
|
1520 // rsi: previous frame manager state (NULL from the call_stub/c1/c2)
|
|
1521 //
|
|
1522 //
|
|
1523 // Stack layout at entry
|
|
1524 //
|
|
1525 // [ return address ] <--- rsp
|
|
1526 // [ parameter n ]
|
|
1527 // ...
|
|
1528 // [ parameter 1 ]
|
|
1529 // [ expression stack ]
|
|
1530 //
|
|
1531 //
|
|
1532 // We are free to blow any registers we like because the call_stub which brought us here
|
|
1533 // initially has preserved the callee save registers already.
|
|
1534 //
|
|
1535 //
|
|
1536
|
|
1537 static address interpreter_frame_manager = NULL;
|
|
1538
|
|
1539 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
|
|
1540
|
|
1541 // rbx: methodOop
|
|
1542 // rsi: sender sp
|
|
1543
|
|
1544 // Because we redispatch "recursive" interpreter entries thru this same entry point
|
|
1545 // the "input" register usage is a little strange and not what you expect coming
|
|
1546 // from the call_stub. From the call stub rsi/rdi (current/previous) interpreter
|
|
1547 // state are NULL but on "recursive" dispatches they are what you'd expect.
|
|
1548 // rsi: current interpreter state (C++ interpreter) must preserve (null from call_stub/c1/c2)
|
|
1549
|
|
1550
|
|
1551 // A single frame manager is plenty as we don't specialize for synchronized. We could and
|
|
1552 // the code is pretty much ready. Would need to change the test below and for good measure
|
|
1553 // modify generate_interpreter_state to only do the (pre) sync stuff stuff for synchronized
|
|
1554 // routines. Not clear this is worth it yet.
|
|
1555
|
|
1556 if (interpreter_frame_manager) return interpreter_frame_manager;
|
|
1557
|
|
1558 address entry_point = __ pc();
|
|
1559
|
|
1560 // Fast accessor methods share this entry point.
|
|
1561 // This works because frame manager is in the same codelet
|
|
1562 if (UseFastAccessorMethods && !synchronized) __ bind(fast_accessor_slow_entry_path);
|
|
1563
|
|
1564 Label dispatch_entry_2;
|
|
1565 __ movl(rcx, rsi);
|
|
1566 __ movl(rsi, 0); // no current activation
|
|
1567
|
|
1568 __ jmp(dispatch_entry_2);
|
|
1569
|
|
1570 const Register state = rsi; // current activation object, valid on entry
|
|
1571 const Register locals = rdi;
|
|
1572
|
|
1573 Label re_dispatch;
|
|
1574
|
|
1575 __ bind(re_dispatch);
|
|
1576
|
|
1577 // save sender sp (doesn't include return address
|
|
1578 __ leal(rcx, Address(rsp, wordSize));
|
|
1579
|
|
1580 __ bind(dispatch_entry_2);
|
|
1581
|
|
1582 // save sender sp
|
|
1583 __ pushl(rcx);
|
|
1584
|
|
1585 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
|
|
1586 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset());
|
|
1587 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
|
|
1588
|
|
1589 // const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
|
|
1590 // const Address monitor_block_bot (rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
|
|
1591 // const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
|
|
1592
|
|
1593 // get parameter size (always needed)
|
|
1594 __ load_unsigned_word(rcx, size_of_parameters);
|
|
1595
|
|
1596 // rbx: methodOop
|
|
1597 // rcx: size of parameters
|
|
1598 __ load_unsigned_word(rdx, size_of_locals); // get size of locals in words
|
|
1599
|
|
1600 __ subl(rdx, rcx); // rdx = no. of additional locals
|
|
1601
|
|
1602 // see if we've got enough room on the stack for locals plus overhead.
|
|
1603 generate_stack_overflow_check(); // C++
|
|
1604
|
|
1605 // c++ interpreter does not use stack banging or any implicit exceptions
|
|
1606 // leave for now to verify that check is proper.
|
|
1607 bang_stack_shadow_pages(false);
|
|
1608
|
|
1609
|
|
1610
|
|
1611 // compute beginning of parameters (rdi)
|
|
1612 __ leal(locals, Address(rsp, rcx, Address::times_4, wordSize));
|
|
1613
|
|
1614 // save sender's sp
|
|
1615 // __ movl(rcx, rsp);
|
|
1616
|
|
1617 // get sender's sp
|
|
1618 __ popl(rcx);
|
|
1619
|
|
1620 // get return address
|
|
1621 __ popl(rax);
|
|
1622
|
|
1623 // rdx - # of additional locals
|
|
1624 // allocate space for locals
|
|
1625 // explicitly initialize locals
|
|
1626 {
|
|
1627 Label exit, loop;
|
|
1628 __ testl(rdx, rdx);
|
|
1629 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
|
|
1630 __ bind(loop);
|
|
1631 __ pushl((int)NULL); // initialize local variables
|
|
1632 __ decrement(rdx); // until everything initialized
|
|
1633 __ jcc(Assembler::greater, loop);
|
|
1634 __ bind(exit);
|
|
1635 }
|
|
1636
|
|
1637
|
|
1638 // Assumes rax = return address
|
|
1639
|
|
1640 // allocate and initialize new interpreterState and method expression stack
|
|
1641 // IN(locals) -> locals
|
|
1642 // IN(state) -> any current interpreter activation
|
|
1643 // destroys rax, rcx, rdx, rdi
|
|
1644 // OUT (state) -> new interpreterState
|
|
1645 // OUT(rsp) -> bottom of methods expression stack
|
|
1646
|
|
1647 generate_compute_interpreter_state(state, locals, rcx, false);
|
|
1648
|
|
1649 // Call interpreter
|
|
1650
|
|
1651 Label call_interpreter;
|
|
1652 __ bind(call_interpreter);
|
|
1653
|
|
1654 // c++ interpreter does not use stack banging or any implicit exceptions
|
|
1655 // leave for now to verify that check is proper.
|
|
1656 bang_stack_shadow_pages(false);
|
|
1657
|
|
1658
|
|
1659 // Call interpreter enter here if message is
|
|
1660 // set and we know stack size is valid
|
|
1661
|
|
1662 Label call_interpreter_2;
|
|
1663
|
|
1664 __ bind(call_interpreter_2);
|
|
1665
|
|
1666 {
|
|
1667 const Register thread = rcx;
|
|
1668
|
|
1669 __ pushl(state); // push arg to interpreter
|
|
1670 __ movl(thread, STATE(_thread));
|
|
1671
|
|
1672 // We can setup the frame anchor with everything we want at this point
|
|
1673 // as we are thread_in_Java and no safepoints can occur until we go to
|
|
1674 // vm mode. We do have to clear flags on return from vm but that is it
|
|
1675 //
|
|
1676 __ movl(Address(thread, JavaThread::last_Java_fp_offset()), rbp);
|
|
1677 __ movl(Address(thread, JavaThread::last_Java_sp_offset()), rsp);
|
|
1678
|
|
1679 // Call the interpreter
|
|
1680
|
|
1681 RuntimeAddress normal(CAST_FROM_FN_PTR(address, BytecodeInterpreter::run));
|
|
1682 RuntimeAddress checking(CAST_FROM_FN_PTR(address, BytecodeInterpreter::runWithChecks));
|
|
1683
|
|
1684 __ call(JvmtiExport::can_post_interpreter_events() ? checking : normal);
|
|
1685 __ popl(rax); // discard parameter to run
|
|
1686 //
|
|
1687 // state is preserved since it is callee saved
|
|
1688 //
|
|
1689
|
|
1690 // reset_last_Java_frame
|
|
1691
|
|
1692 __ movl(thread, STATE(_thread));
|
|
1693 __ reset_last_Java_frame(thread, true, true);
|
|
1694 }
|
|
1695
|
|
1696 // examine msg from interpreter to determine next action
|
|
1697
|
|
1698 __ movl(rdx, STATE(_msg)); // Get new message
|
|
1699
|
|
1700 Label call_method;
|
|
1701 Label return_from_interpreted_method;
|
|
1702 Label throw_exception;
|
|
1703 Label bad_msg;
|
|
1704 Label do_OSR;
|
|
1705
|
|
1706 __ cmpl(rdx, (int)BytecodeInterpreter::call_method);
|
|
1707 __ jcc(Assembler::equal, call_method);
|
|
1708 __ cmpl(rdx, (int)BytecodeInterpreter::return_from_method);
|
|
1709 __ jcc(Assembler::equal, return_from_interpreted_method);
|
|
1710 __ cmpl(rdx, (int)BytecodeInterpreter::do_osr);
|
|
1711 __ jcc(Assembler::equal, do_OSR);
|
|
1712 __ cmpl(rdx, (int)BytecodeInterpreter::throwing_exception);
|
|
1713 __ jcc(Assembler::equal, throw_exception);
|
|
1714 __ cmpl(rdx, (int)BytecodeInterpreter::more_monitors);
|
|
1715 __ jcc(Assembler::notEqual, bad_msg);
|
|
1716
|
|
1717 // Allocate more monitor space, shuffle expression stack....
|
|
1718
|
|
1719 generate_more_monitors();
|
|
1720
|
|
1721 __ jmp(call_interpreter);
|
|
1722
|
|
1723 // uncommon trap needs to jump to here to enter the interpreter (re-execute current bytecode)
|
|
1724 unctrap_frame_manager_entry = __ pc();
|
|
1725 //
|
|
1726 // Load the registers we need.
|
|
1727 __ leal(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
|
|
1728 __ movl(rsp, STATE(_stack_limit)); // restore expression stack to full depth
|
|
1729 __ jmp(call_interpreter_2);
|
|
1730
|
|
1731
|
|
1732
|
|
1733 //=============================================================================
|
|
1734 // Returning from a compiled method into a deopted method. The bytecode at the
|
|
1735 // bcp has completed. The result of the bytecode is in the native abi (the tosca
|
|
1736 // for the template based interpreter). Any stack space that was used by the
|
|
1737 // bytecode that has completed has been removed (e.g. parameters for an invoke)
|
|
1738 // so all that we have to do is place any pending result on the expression stack
|
|
1739 // and resume execution on the next bytecode.
|
|
1740
|
|
1741
|
|
1742 generate_deopt_handling();
|
|
1743 __ jmp(call_interpreter);
|
|
1744
|
|
1745
|
|
1746 // Current frame has caught an exception we need to dispatch to the
|
|
1747 // handler. We can get here because a native interpreter frame caught
|
|
1748 // an exception in which case there is no handler and we must rethrow
|
|
1749 // If it is a vanilla interpreted frame the we simply drop into the
|
|
1750 // interpreter and let it do the lookup.
|
|
1751
|
|
1752 Interpreter::_rethrow_exception_entry = __ pc();
|
|
1753 // rax: exception
|
|
1754 // rdx: return address/pc that threw exception
|
|
1755
|
|
1756 Label return_with_exception;
|
|
1757 Label unwind_and_forward;
|
|
1758
|
|
1759 // restore state pointer.
|
|
1760 __ leal(state, Address(rbp, -sizeof(BytecodeInterpreter)));
|
|
1761
|
|
1762 __ movl(rbx, STATE(_method)); // get method
|
|
1763 __ movl(rcx, STATE(_thread)); // get thread
|
|
1764
|
|
1765 // Store exception with interpreter will expect it
|
|
1766 __ movl(Address(rcx, Thread::pending_exception_offset()), rax);
|
|
1767
|
|
1768 // is current frame vanilla or native?
|
|
1769
|
|
1770 __ movl(rdx, access_flags);
|
|
1771 __ testl(rdx, JVM_ACC_NATIVE);
|
|
1772 __ jcc(Assembler::zero, return_with_exception); // vanilla interpreted frame, handle directly
|
|
1773
|
|
1774 // We drop thru to unwind a native interpreted frame with a pending exception
|
|
1775 // We jump here for the initial interpreter frame with exception pending
|
|
1776 // We unwind the current acivation and forward it to our caller.
|
|
1777
|
|
1778 __ bind(unwind_and_forward);
|
|
1779
|
|
1780 // unwind rbp, return stack to unextended value and re-push return address
|
|
1781
|
|
1782 __ movl(rcx, STATE(_sender_sp));
|
|
1783 __ leave();
|
|
1784 __ popl(rdx);
|
|
1785 __ movl(rsp, rcx);
|
|
1786 __ pushl(rdx);
|
|
1787 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
|
|
1788
|
|
1789 // Return point from a call which returns a result in the native abi
|
|
1790 // (c1/c2/jni-native). This result must be processed onto the java
|
|
1791 // expression stack.
|
|
1792 //
|
|
1793 // A pending exception may be present in which case there is no result present
|
|
1794
|
|
1795 Label resume_interpreter;
|
|
1796 Label do_float;
|
|
1797 Label do_double;
|
|
1798 Label done_conv;
|
|
1799
|
|
1800 address compiled_entry = __ pc();
|
|
1801
|
|
1802 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
|
|
1803 if (UseSSE < 2) {
|
|
1804 __ leal(state, Address(rbp, -sizeof(BytecodeInterpreter)));
|
|
1805 __ movl(rbx, STATE(_result._to_call._callee)); // get method just executed
|
|
1806 __ movl(rcx, Address(rbx, methodOopDesc::result_index_offset()));
|
|
1807 __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index
|
|
1808 __ jcc(Assembler::equal, do_float);
|
|
1809 __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index
|
|
1810 __ jcc(Assembler::equal, do_double);
|
|
1811 #ifdef COMPILER2
|
|
1812 __ empty_FPU_stack();
|
|
1813 #endif // COMPILER2
|
|
1814 __ jmp(done_conv);
|
|
1815
|
|
1816 __ bind(do_float);
|
|
1817 #ifdef COMPILER2
|
|
1818 for (int i = 1; i < 8; i++) {
|
|
1819 __ ffree(i);
|
|
1820 }
|
|
1821 #endif // COMPILER2
|
|
1822 __ jmp(done_conv);
|
|
1823 __ bind(do_double);
|
|
1824 #ifdef COMPILER2
|
|
1825 for (int i = 1; i < 8; i++) {
|
|
1826 __ ffree(i);
|
|
1827 }
|
|
1828 #endif // COMPILER2
|
|
1829 __ jmp(done_conv);
|
|
1830 } else {
|
|
1831 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
|
|
1832 __ jmp(done_conv);
|
|
1833 }
|
|
1834
|
|
1835 // emit a sentinel we can test for when converting an interpreter
|
|
1836 // entry point to a compiled entry point.
|
|
1837 __ a_long(Interpreter::return_sentinel);
|
|
1838 __ a_long((int)compiled_entry);
|
|
1839
|
|
1840 // Return point to interpreter from compiled/native method
|
|
1841
|
|
1842 InternalAddress return_from_native_method(__ pc());
|
|
1843
|
|
1844 __ bind(done_conv);
|
|
1845
|
|
1846
|
|
1847 // Result if any is in tosca. The java expression stack is in the state that the
|
|
1848 // calling convention left it (i.e. params may or may not be present)
|
|
1849 // Copy the result from tosca and place it on java expression stack.
|
|
1850
|
|
1851 // Restore rsi as compiled code may not preserve it
|
|
1852
|
|
1853 __ leal(state, Address(rbp, -sizeof(BytecodeInterpreter)));
|
|
1854
|
|
1855 // restore stack to what we had when we left (in case i2c extended it)
|
|
1856
|
|
1857 __ movl(rsp, STATE(_stack));
|
|
1858 __ leal(rsp, Address(rsp, wordSize));
|
|
1859
|
|
1860 // If there is a pending exception then we don't really have a result to process
|
|
1861
|
|
1862 __ movl(rcx, STATE(_thread)); // get thread
|
|
1863 __ cmpl(Address(rcx, Thread::pending_exception_offset()), (int)NULL);
|
|
1864 __ jcc(Assembler::notZero, return_with_exception);
|
|
1865
|
|
1866 // get method just executed
|
|
1867 __ movl(rbx, STATE(_result._to_call._callee));
|
|
1868
|
|
1869 // callee left args on top of expression stack, remove them
|
|
1870 __ load_unsigned_word(rcx, Address(rbx, methodOopDesc::size_of_parameters_offset()));
|
|
1871 __ leal(rsp, Address(rsp, rcx, Address::times_4));
|
|
1872
|
|
1873 __ movl(rcx, Address(rbx, methodOopDesc::result_index_offset()));
|
|
1874 ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack);
|
|
1875 // Address index(noreg, rax, Address::times_4);
|
|
1876 __ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_4)));
|
|
1877 // __ movl(rcx, Address(noreg, rcx, Address::times_4, int(AbstractInterpreter::_tosca_to_stack)));
|
|
1878 __ call(rcx); // call result converter
|
|
1879 __ jmp(resume_interpreter);
|
|
1880
|
|
1881 // An exception is being caught on return to a vanilla interpreter frame.
|
|
1882 // Empty the stack and resume interpreter
|
|
1883
|
|
1884 __ bind(return_with_exception);
|
|
1885
|
|
1886 // Exception present, empty stack
|
|
1887 __ movl(rsp, STATE(_stack_base));
|
|
1888 __ jmp(resume_interpreter);
|
|
1889
|
|
1890 // Return from interpreted method we return result appropriate to the caller (i.e. "recursive"
|
|
1891 // interpreter call, or native) and unwind this interpreter activation.
|
|
1892 // All monitors should be unlocked.
|
|
1893
|
|
1894 __ bind(return_from_interpreted_method);
|
|
1895
|
|
1896 Label return_to_initial_caller;
|
|
1897
|
|
1898 __ movl(rbx, STATE(_method)); // get method just executed
|
|
1899 __ cmpl(STATE(_prev_link), (int)NULL); // returning from "recursive" interpreter call?
|
|
1900 __ movl(rax, Address(rbx, methodOopDesc::result_index_offset())); // get result type index
|
|
1901 __ jcc(Assembler::equal, return_to_initial_caller); // back to native code (call_stub/c1/c2)
|
|
1902
|
|
1903 // Copy result to callers java stack
|
|
1904 ExternalAddress stack_to_stack((address)CppInterpreter::_stack_to_stack);
|
|
1905 // Address index(noreg, rax, Address::times_4);
|
|
1906
|
|
1907 __ movptr(rax, ArrayAddress(stack_to_stack, Address(noreg, rax, Address::times_4)));
|
|
1908 // __ movl(rax, Address(noreg, rax, Address::times_4, int(AbstractInterpreter::_stack_to_stack)));
|
|
1909 __ call(rax); // call result converter
|
|
1910
|
|
1911 Label unwind_recursive_activation;
|
|
1912 __ bind(unwind_recursive_activation);
|
|
1913
|
|
1914 // returning to interpreter method from "recursive" interpreter call
|
|
1915 // result converter left rax pointing to top of the java stack for method we are returning
|
|
1916 // to. Now all we must do is unwind the state from the completed call
|
|
1917
|
|
1918 __ movl(state, STATE(_prev_link)); // unwind state
|
|
1919 __ leave(); // pop the frame
|
|
1920 __ movl(rsp, rax); // unwind stack to remove args
|
|
1921
|
|
1922 // Resume the interpreter. The current frame contains the current interpreter
|
|
1923 // state object.
|
|
1924 //
|
|
1925
|
|
1926 __ bind(resume_interpreter);
|
|
1927
|
|
1928 // state == interpreterState object for method we are resuming
|
|
1929
|
|
1930 __ movl(STATE(_msg), (int)BytecodeInterpreter::method_resume);
|
|
1931 __ leal(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present)
|
|
1932 __ movl(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed,
|
|
1933 // result if any on stack already )
|
|
1934 __ movl(rsp, STATE(_stack_limit)); // restore expression stack to full depth
|
|
1935 __ jmp(call_interpreter_2); // No need to bang
|
|
1936
|
|
1937 // interpreter returning to native code (call_stub/c1/c2)
|
|
1938 // convert result and unwind initial activation
|
|
1939 // rax - result index
|
|
1940
|
|
1941 __ bind(return_to_initial_caller);
|
|
1942 ExternalAddress stack_to_native((address)CppInterpreter::_stack_to_native_abi);
|
|
1943 // Address index(noreg, rax, Address::times_4);
|
|
1944
|
|
1945 __ movptr(rax, ArrayAddress(stack_to_native, Address(noreg, rax, Address::times_4)));
|
|
1946 __ call(rax); // call result converter
|
|
1947
|
|
1948 Label unwind_initial_activation;
|
|
1949 __ bind(unwind_initial_activation);
|
|
1950
|
|
1951 // RETURN TO CALL_STUB/C1/C2 code (result if any in rax/rdx ST(0))
|
|
1952
|
|
1953 /* Current stack picture
|
|
1954
|
|
1955 [ incoming parameters ]
|
|
1956 [ extra locals ]
|
|
1957 [ return address to CALL_STUB/C1/C2]
|
|
1958 fp -> [ CALL_STUB/C1/C2 fp ]
|
|
1959 BytecodeInterpreter object
|
|
1960 expression stack
|
|
1961 sp ->
|
|
1962
|
|
1963 */
|
|
1964
|
|
1965 // return restoring the stack to the original sender_sp value
|
|
1966
|
|
1967 __ movl(rcx, STATE(_sender_sp));
|
|
1968 __ leave();
|
|
1969 __ popl(rdi); // get return address
|
|
1970 // set stack to sender's sp
|
|
1971 __ movl(rsp, rcx);
|
|
1972 __ jmp(rdi); // return to call_stub
|
|
1973
|
|
1974 // OSR request, adjust return address to make current frame into adapter frame
|
|
1975 // and enter OSR nmethod
|
|
1976
|
|
1977 __ bind(do_OSR);
|
|
1978
|
|
1979 Label remove_initial_frame;
|
|
1980
|
|
1981 // We are going to pop this frame. Is there another interpreter frame underneath
|
|
1982 // it or is it callstub/compiled?
|
|
1983
|
|
1984 // Move buffer to the expected parameter location
|
|
1985 __ movl(rcx, STATE(_result._osr._osr_buf));
|
|
1986
|
|
1987 __ movl(rax, STATE(_result._osr._osr_entry));
|
|
1988
|
|
1989 __ cmpl(STATE(_prev_link), (int)NULL); // returning from "recursive" interpreter call?
|
|
1990 __ jcc(Assembler::equal, remove_initial_frame); // back to native code (call_stub/c1/c2)
|
|
1991
|
|
1992 // __ movl(state, STATE(_prev_link)); // unwind state
|
|
1993 __ movl(rsi, STATE(_sender_sp)); // get sender's sp in expected register
|
|
1994 __ leave(); // pop the frame
|
|
1995 __ movl(rsp, rsi); // trim any stack expansion
|
|
1996
|
|
1997
|
|
1998 // We know we are calling compiled so push specialized return
|
|
1999 // method uses specialized entry, push a return so we look like call stub setup
|
|
2000 // this path will handle fact that result is returned in registers and not
|
|
2001 // on the java stack.
|
|
2002
|
|
2003 __ pushptr(return_from_native_method.addr());
|
|
2004
|
|
2005 __ jmp(rax);
|
|
2006
|
|
2007 __ bind(remove_initial_frame);
|
|
2008
|
|
2009 __ movl(rdx, STATE(_sender_sp));
|
|
2010 __ leave();
|
|
2011 // get real return
|
|
2012 __ popl(rsi);
|
|
2013 // set stack to sender's sp
|
|
2014 __ movl(rsp, rdx);
|
|
2015 // repush real return
|
|
2016 __ pushl(rsi);
|
|
2017 // Enter OSR nmethod
|
|
2018 __ jmp(rax);
|
|
2019
|
|
2020
|
|
2021
|
|
2022
|
|
2023 // Call a new method. All we do is (temporarily) trim the expression stack
|
|
2024 // push a return address to bring us back to here and leap to the new entry.
|
|
2025
|
|
2026 __ bind(call_method);
|
|
2027
|
|
2028 // stack points to next free location and not top element on expression stack
|
|
2029 // method expects sp to be pointing to topmost element
|
|
2030
|
|
2031 __ movl(rsp, STATE(_stack)); // pop args to c++ interpreter, set sp to java stack top
|
|
2032 __ leal(rsp, Address(rsp, wordSize));
|
|
2033
|
|
2034 __ movl(rbx, STATE(_result._to_call._callee)); // get method to execute
|
|
2035
|
|
2036 // don't need a return address if reinvoking interpreter
|
|
2037
|
|
2038 // Make it look like call_stub calling conventions
|
|
2039
|
|
2040 // Get (potential) receiver
|
|
2041 __ load_unsigned_word(rcx, size_of_parameters); // get size of parameters in words
|
|
2042
|
|
2043 ExternalAddress recursive(CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation));
|
|
2044 __ pushptr(recursive.addr()); // make it look good in the debugger
|
|
2045
|
|
2046 InternalAddress entry(entry_point);
|
|
2047 __ cmpptr(STATE(_result._to_call._callee_entry_point), entry.addr()); // returning to interpreter?
|
|
2048 __ jcc(Assembler::equal, re_dispatch); // yes
|
|
2049
|
|
2050 __ popl(rax); // pop dummy address
|
|
2051
|
|
2052
|
|
2053 // get specialized entry
|
|
2054 __ movl(rax, STATE(_result._to_call._callee_entry_point));
|
|
2055 // set sender SP
|
|
2056 __ movl(rsi, rsp);
|
|
2057
|
|
2058 // method uses specialized entry, push a return so we look like call stub setup
|
|
2059 // this path will handle fact that result is returned in registers and not
|
|
2060 // on the java stack.
|
|
2061
|
|
2062 __ pushptr(return_from_native_method.addr());
|
|
2063
|
|
2064 __ jmp(rax);
|
|
2065
|
|
2066 __ bind(bad_msg);
|
|
2067 __ stop("Bad message from interpreter");
|
|
2068
|
|
2069 // Interpreted method "returned" with an exception pass it on...
|
|
2070 // Pass result, unwind activation and continue/return to interpreter/call_stub
|
|
2071 // We handle result (if any) differently based on return to interpreter or call_stub
|
|
2072
|
|
2073 Label unwind_initial_with_pending_exception;
|
|
2074
|
|
2075 __ bind(throw_exception);
|
|
2076 __ cmpl(STATE(_prev_link), (int)NULL); // returning from recursive interpreter call?
|
|
2077 __ jcc(Assembler::equal, unwind_initial_with_pending_exception); // no, back to native code (call_stub/c1/c2)
|
|
2078 __ movl(rax, STATE(_locals)); // pop parameters get new stack value
|
|
2079 __ addl(rax, wordSize); // account for prepush before we return
|
|
2080 __ jmp(unwind_recursive_activation);
|
|
2081
|
|
2082 __ bind(unwind_initial_with_pending_exception);
|
|
2083
|
|
2084 // We will unwind the current (initial) interpreter frame and forward
|
|
2085 // the exception to the caller. We must put the exception in the
|
|
2086 // expected register and clear pending exception and then forward.
|
|
2087
|
|
2088 __ jmp(unwind_and_forward);
|
|
2089
|
|
2090 interpreter_frame_manager = entry_point;
|
|
2091 return entry_point;
|
|
2092 }
|
|
2093
|
|
2094 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
|
|
2095 // determine code generation flags
|
|
2096 bool synchronized = false;
|
|
2097 address entry_point = NULL;
|
|
2098
|
|
2099 switch (kind) {
|
|
2100 case Interpreter::zerolocals : break;
|
|
2101 case Interpreter::zerolocals_synchronized: synchronized = true; break;
|
|
2102 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
|
|
2103 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
|
|
2104 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
|
|
2105 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
|
|
2106 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
|
|
2107
|
|
2108 case Interpreter::java_lang_math_sin : // fall thru
|
|
2109 case Interpreter::java_lang_math_cos : // fall thru
|
|
2110 case Interpreter::java_lang_math_tan : // fall thru
|
|
2111 case Interpreter::java_lang_math_abs : // fall thru
|
|
2112 case Interpreter::java_lang_math_log : // fall thru
|
|
2113 case Interpreter::java_lang_math_log10 : // fall thru
|
|
2114 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
|
|
2115 default : ShouldNotReachHere(); break;
|
|
2116 }
|
|
2117
|
|
2118 if (entry_point) return entry_point;
|
|
2119
|
|
2120 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
|
|
2121
|
|
2122 }
|
|
2123
|
|
2124 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
|
|
2125 : CppInterpreterGenerator(code) {
|
|
2126 generate_all(); // down here so it can be "virtual"
|
|
2127 }
|
|
2128
|
|
2129 // Deoptimization helpers for C++ interpreter
|
|
2130
|
|
2131 // How much stack a method activation needs in words.
|
|
2132 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
|
|
2133
|
|
2134 const int stub_code = 4; // see generate_call_stub
|
|
2135 // Save space for one monitor to get into the interpreted method in case
|
|
2136 // the method is synchronized
|
|
2137 int monitor_size = method->is_synchronized() ?
|
|
2138 1*frame::interpreter_frame_monitor_size() : 0;
|
|
2139
|
|
2140 // total static overhead size. Account for interpreter state object, return
|
|
2141 // address, saved rbp and 2 words for a "static long no_params() method" issue.
|
|
2142
|
|
2143 const int overhead_size = sizeof(BytecodeInterpreter)/wordSize +
|
|
2144 ( frame::sender_sp_offset - frame::link_offset) + 2;
|
|
2145
|
|
2146 const int method_stack = (method->max_locals() + method->max_stack()) *
|
|
2147 Interpreter::stackElementWords();
|
|
2148 return overhead_size + method_stack + stub_code;
|
|
2149 }
|
|
2150
|
|
2151 // returns the activation size.
|
|
2152 static int size_activation_helper(int extra_locals_size, int monitor_size) {
|
|
2153 return (extra_locals_size + // the addition space for locals
|
|
2154 2*BytesPerWord + // return address and saved rbp
|
|
2155 2*BytesPerWord + // "static long no_params() method" issue
|
|
2156 sizeof(BytecodeInterpreter) + // interpreterState
|
|
2157 monitor_size); // monitors
|
|
2158 }
|
|
2159
|
|
2160 void BytecodeInterpreter::layout_interpreterState(interpreterState to_fill,
|
|
2161 frame* caller,
|
|
2162 frame* current,
|
|
2163 methodOop method,
|
|
2164 intptr_t* locals,
|
|
2165 intptr_t* stack,
|
|
2166 intptr_t* stack_base,
|
|
2167 intptr_t* monitor_base,
|
|
2168 intptr_t* frame_bottom,
|
|
2169 bool is_top_frame
|
|
2170 )
|
|
2171 {
|
|
2172 // What about any vtable?
|
|
2173 //
|
|
2174 to_fill->_thread = JavaThread::current();
|
|
2175 // This gets filled in later but make it something recognizable for now
|
|
2176 to_fill->_bcp = method->code_base();
|
|
2177 to_fill->_locals = locals;
|
|
2178 to_fill->_constants = method->constants()->cache();
|
|
2179 to_fill->_method = method;
|
|
2180 to_fill->_mdx = NULL;
|
|
2181 to_fill->_stack = stack;
|
|
2182 if (is_top_frame && JavaThread::current()->popframe_forcing_deopt_reexecution() ) {
|
|
2183 to_fill->_msg = deopt_resume2;
|
|
2184 } else {
|
|
2185 to_fill->_msg = method_resume;
|
|
2186 }
|
|
2187 to_fill->_result._to_call._bcp_advance = 0;
|
|
2188 to_fill->_result._to_call._callee_entry_point = NULL; // doesn't matter to anyone
|
|
2189 to_fill->_result._to_call._callee = NULL; // doesn't matter to anyone
|
|
2190 to_fill->_prev_link = NULL;
|
|
2191
|
|
2192 to_fill->_sender_sp = caller->unextended_sp();
|
|
2193
|
|
2194 if (caller->is_interpreted_frame()) {
|
|
2195 interpreterState prev = caller->get_interpreterState();
|
|
2196 to_fill->_prev_link = prev;
|
|
2197 // *current->register_addr(GR_Iprev_state) = (intptr_t) prev;
|
|
2198 // Make the prev callee look proper
|
|
2199 prev->_result._to_call._callee = method;
|
|
2200 if (*prev->_bcp == Bytecodes::_invokeinterface) {
|
|
2201 prev->_result._to_call._bcp_advance = 5;
|
|
2202 } else {
|
|
2203 prev->_result._to_call._bcp_advance = 3;
|
|
2204 }
|
|
2205 }
|
|
2206 to_fill->_oop_temp = NULL;
|
|
2207 to_fill->_stack_base = stack_base;
|
|
2208 // Need +1 here because stack_base points to the word just above the first expr stack entry
|
|
2209 // and stack_limit is supposed to point to the word just below the last expr stack entry.
|
|
2210 // See generate_compute_interpreter_state.
|
|
2211 to_fill->_stack_limit = stack_base - (method->max_stack() + 1);
|
|
2212 to_fill->_monitor_base = (BasicObjectLock*) monitor_base;
|
|
2213
|
|
2214 to_fill->_self_link = to_fill;
|
|
2215 assert(stack >= to_fill->_stack_limit && stack < to_fill->_stack_base,
|
|
2216 "Stack top out of range");
|
|
2217 }
|
|
2218
|
|
2219 int AbstractInterpreter::layout_activation(methodOop method,
|
|
2220 int tempcount, //
|
|
2221 int popframe_extra_args,
|
|
2222 int moncount,
|
|
2223 int callee_param_count,
|
|
2224 int callee_locals,
|
|
2225 frame* caller,
|
|
2226 frame* interpreter_frame,
|
|
2227 bool is_top_frame) {
|
|
2228
|
|
2229 assert(popframe_extra_args == 0, "FIX ME");
|
|
2230 // NOTE this code must exactly mimic what InterpreterGenerator::generate_compute_interpreter_state()
|
|
2231 // does as far as allocating an interpreter frame.
|
|
2232 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
|
|
2233 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
|
|
2234 // as determined by a previous call to this method.
|
|
2235 // It is also guaranteed to be walkable even though it is in a skeletal state
|
|
2236 // NOTE: return size is in words not bytes
|
|
2237 // NOTE: tempcount is the current size of the java expression stack. For top most
|
|
2238 // frames we will allocate a full sized expression stack and not the curback
|
|
2239 // version that non-top frames have.
|
|
2240
|
|
2241 // Calculate the amount our frame will be adjust by the callee. For top frame
|
|
2242 // this is zero.
|
|
2243
|
|
2244 // NOTE: ia64 seems to do this wrong (or at least backwards) in that it
|
|
2245 // calculates the extra locals based on itself. Not what the callee does
|
|
2246 // to it. So it ignores last_frame_adjust value. Seems suspicious as far
|
|
2247 // as getting sender_sp correct.
|
|
2248
|
|
2249 int extra_locals_size = (callee_locals - callee_param_count) * BytesPerWord;
|
|
2250 int monitor_size = sizeof(BasicObjectLock) * moncount;
|
|
2251
|
|
2252 // First calculate the frame size without any java expression stack
|
|
2253 int short_frame_size = size_activation_helper(extra_locals_size,
|
|
2254 monitor_size);
|
|
2255
|
|
2256 // Now with full size expression stack
|
|
2257 int full_frame_size = short_frame_size + method->max_stack() * BytesPerWord;
|
|
2258
|
|
2259 // and now with only live portion of the expression stack
|
|
2260 short_frame_size = short_frame_size + tempcount * BytesPerWord;
|
|
2261
|
|
2262 // the size the activation is right now. Only top frame is full size
|
|
2263 int frame_size = (is_top_frame ? full_frame_size : short_frame_size);
|
|
2264
|
|
2265 if (interpreter_frame != NULL) {
|
|
2266 #ifdef ASSERT
|
|
2267 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
|
|
2268 #endif
|
|
2269
|
|
2270 // MUCHO HACK
|
|
2271
|
|
2272 intptr_t* frame_bottom = (intptr_t*) ((intptr_t)interpreter_frame->sp() - (full_frame_size - frame_size));
|
|
2273
|
|
2274 /* Now fillin the interpreterState object */
|
|
2275
|
|
2276 // The state object is the first thing on the frame and easily located
|
|
2277
|
|
2278 interpreterState cur_state = (interpreterState) ((intptr_t)interpreter_frame->fp() - sizeof(BytecodeInterpreter));
|
|
2279
|
|
2280
|
|
2281 // Find the locals pointer. This is rather simple on x86 because there is no
|
|
2282 // confusing rounding at the callee to account for. We can trivially locate
|
|
2283 // our locals based on the current fp().
|
|
2284 // Note: the + 2 is for handling the "static long no_params() method" issue.
|
|
2285 // (too bad I don't really remember that issue well...)
|
|
2286
|
|
2287 intptr_t* locals;
|
|
2288 // If the caller is interpreted we need to make sure that locals points to the first
|
|
2289 // argument that the caller passed and not in an area where the stack might have been extended.
|
|
2290 // because the stack to stack to converter needs a proper locals value in order to remove the
|
|
2291 // arguments from the caller and place the result in the proper location. Hmm maybe it'd be
|
|
2292 // simpler if we simply stored the result in the BytecodeInterpreter object and let the c++ code
|
|
2293 // adjust the stack?? HMMM QQQ
|
|
2294 //
|
|
2295 if (caller->is_interpreted_frame()) {
|
|
2296 // locals must agree with the caller because it will be used to set the
|
|
2297 // caller's tos when we return.
|
|
2298 interpreterState prev = caller->get_interpreterState();
|
|
2299 // stack() is prepushed.
|
|
2300 locals = prev->stack() + method->size_of_parameters();
|
|
2301 // locals = caller->unextended_sp() + (method->size_of_parameters() - 1);
|
|
2302 if (locals != interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2) {
|
|
2303 // os::breakpoint();
|
|
2304 }
|
|
2305 } else {
|
|
2306 // this is where a c2i would have placed locals (except for the +2)
|
|
2307 locals = interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2;
|
|
2308 }
|
|
2309
|
|
2310 intptr_t* monitor_base = (intptr_t*) cur_state;
|
|
2311 intptr_t* stack_base = (intptr_t*) ((intptr_t) monitor_base - monitor_size);
|
|
2312 /* +1 because stack is always prepushed */
|
|
2313 intptr_t* stack = (intptr_t*) ((intptr_t) stack_base - (tempcount + 1) * BytesPerWord);
|
|
2314
|
|
2315
|
|
2316 BytecodeInterpreter::layout_interpreterState(cur_state,
|
|
2317 caller,
|
|
2318 interpreter_frame,
|
|
2319 method,
|
|
2320 locals,
|
|
2321 stack,
|
|
2322 stack_base,
|
|
2323 monitor_base,
|
|
2324 frame_bottom,
|
|
2325 is_top_frame);
|
|
2326
|
|
2327 // BytecodeInterpreter::pd_layout_interpreterState(cur_state, interpreter_return_address, interpreter_frame->fp());
|
|
2328 }
|
|
2329 return frame_size/BytesPerWord;
|
|
2330 }
|
|
2331
|
|
2332 #endif // CC_INTERP (all)
|