Mercurial > hg > truffle
annotate src/cpu/x86/vm/cppInterpreter_x86.cpp @ 955:26fbe81d30cf
6880052: SIGSEGV in GraphKit::null_check_common()
Summary: Check that a klass is not NULL before the is_loaded() call.
Reviewed-by: never
author | kvn |
---|---|
date | Tue, 08 Sep 2009 16:56:31 -0700 |
parents | e5b0439ef4ae |
children | c18cbe5936b8 |
rev | line source |
---|---|
0 | 1 /* |
579 | 2 * Copyright 2007-2009 Sun Microsystems, Inc. All Rights Reserved. |
0 | 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 | |
304 | 47 // default registers for state and sender_sp |
48 // state and sender_sp are the same on 32bit because we have no choice. | |
49 // state could be rsi on 64bit but it is an arg reg and not callee save | |
50 // so r13 is better choice. | |
51 | |
52 const Register state = NOT_LP64(rsi) LP64_ONLY(r13); | |
53 const Register sender_sp_on_entry = NOT_LP64(rsi) LP64_ONLY(r13); | |
54 | |
0 | 55 // NEEDED for JVMTI? |
56 // address AbstractInterpreter::_remove_activation_preserving_args_entry; | |
57 | |
58 static address unctrap_frame_manager_entry = NULL; | |
59 | |
60 static address deopt_frame_manager_return_atos = NULL; | |
61 static address deopt_frame_manager_return_btos = NULL; | |
62 static address deopt_frame_manager_return_itos = NULL; | |
63 static address deopt_frame_manager_return_ltos = NULL; | |
64 static address deopt_frame_manager_return_ftos = NULL; | |
65 static address deopt_frame_manager_return_dtos = NULL; | |
66 static address deopt_frame_manager_return_vtos = NULL; | |
67 | |
68 int AbstractInterpreter::BasicType_as_index(BasicType type) { | |
69 int i = 0; | |
70 switch (type) { | |
71 case T_BOOLEAN: i = 0; break; | |
72 case T_CHAR : i = 1; break; | |
73 case T_BYTE : i = 2; break; | |
74 case T_SHORT : i = 3; break; | |
75 case T_INT : i = 4; break; | |
76 case T_VOID : i = 5; break; | |
77 case T_FLOAT : i = 8; break; | |
78 case T_LONG : i = 9; break; | |
79 case T_DOUBLE : i = 6; break; | |
80 case T_OBJECT : // fall through | |
81 case T_ARRAY : i = 7; break; | |
82 default : ShouldNotReachHere(); | |
83 } | |
84 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds"); | |
85 return i; | |
86 } | |
87 | |
88 // Is this pc anywhere within code owned by the interpreter? | |
89 // This only works for pc that might possibly be exposed to frame | |
90 // walkers. It clearly misses all of the actual c++ interpreter | |
91 // implementation | |
92 bool CppInterpreter::contains(address pc) { | |
93 return (_code->contains(pc) || | |
94 pc == CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation)); | |
95 } | |
96 | |
97 | |
98 address CppInterpreterGenerator::generate_result_handler_for(BasicType type) { | |
99 address entry = __ pc(); | |
100 switch (type) { | |
101 case T_BOOLEAN: __ c2bool(rax); break; | |
102 case T_CHAR : __ andl(rax, 0xFFFF); break; | |
103 case T_BYTE : __ sign_extend_byte (rax); break; | |
104 case T_SHORT : __ sign_extend_short(rax); break; | |
105 case T_VOID : // fall thru | |
106 case T_LONG : // fall thru | |
107 case T_INT : /* nothing to do */ break; | |
304 | 108 |
0 | 109 case T_DOUBLE : |
110 case T_FLOAT : | |
304 | 111 { |
112 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); | |
113 __ pop(t); // remove return address first | |
0 | 114 // Must return a result for interpreter or compiler. In SSE |
115 // mode, results are returned in xmm0 and the FPU stack must | |
116 // be empty. | |
117 if (type == T_FLOAT && UseSSE >= 1) { | |
304 | 118 #ifndef _LP64 |
0 | 119 // Load ST0 |
120 __ fld_d(Address(rsp, 0)); | |
121 // Store as float and empty fpu stack | |
122 __ fstp_s(Address(rsp, 0)); | |
304 | 123 #endif // !_LP64 |
0 | 124 // and reload |
125 __ movflt(xmm0, Address(rsp, 0)); | |
126 } else if (type == T_DOUBLE && UseSSE >= 2 ) { | |
127 __ movdbl(xmm0, Address(rsp, 0)); | |
128 } else { | |
129 // restore ST0 | |
130 __ fld_d(Address(rsp, 0)); | |
131 } | |
132 // and pop the temp | |
304 | 133 __ addptr(rsp, 2 * wordSize); |
134 __ push(t); // restore return address | |
0 | 135 } |
136 break; | |
137 case T_OBJECT : | |
138 // retrieve result from frame | |
304 | 139 __ movptr(rax, STATE(_oop_temp)); |
0 | 140 // and verify it |
141 __ verify_oop(rax); | |
142 break; | |
143 default : ShouldNotReachHere(); | |
144 } | |
145 __ ret(0); // return from result handler | |
146 return entry; | |
147 } | |
148 | |
149 // tosca based result to c++ interpreter stack based result. | |
150 // Result goes to top of native stack. | |
151 | |
152 #undef EXTEND // SHOULD NOT BE NEEDED | |
153 address CppInterpreterGenerator::generate_tosca_to_stack_converter(BasicType type) { | |
154 // A result is in the tosca (abi result) from either a native method call or compiled | |
155 // code. Place this result on the java expression stack so C++ interpreter can use it. | |
156 address entry = __ pc(); | |
157 | |
158 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); | |
304 | 159 __ pop(t); // remove return address first |
0 | 160 switch (type) { |
161 case T_VOID: | |
162 break; | |
163 case T_BOOLEAN: | |
164 #ifdef EXTEND | |
165 __ c2bool(rax); | |
166 #endif | |
304 | 167 __ push(rax); |
0 | 168 break; |
169 case T_CHAR : | |
170 #ifdef EXTEND | |
171 __ andl(rax, 0xFFFF); | |
172 #endif | |
304 | 173 __ push(rax); |
0 | 174 break; |
175 case T_BYTE : | |
176 #ifdef EXTEND | |
177 __ sign_extend_byte (rax); | |
178 #endif | |
304 | 179 __ push(rax); |
0 | 180 break; |
181 case T_SHORT : | |
182 #ifdef EXTEND | |
183 __ sign_extend_short(rax); | |
184 #endif | |
304 | 185 __ push(rax); |
0 | 186 break; |
187 case T_LONG : | |
304 | 188 __ push(rdx); // pushes useless junk on 64bit |
189 __ push(rax); | |
0 | 190 break; |
191 case T_INT : | |
304 | 192 __ push(rax); |
0 | 193 break; |
194 case T_FLOAT : | |
304 | 195 // Result is in ST(0)/xmm0 |
196 __ subptr(rsp, wordSize); | |
0 | 197 if ( UseSSE < 1) { |
304 | 198 __ fstp_s(Address(rsp, 0)); |
0 | 199 } else { |
200 __ movflt(Address(rsp, 0), xmm0); | |
201 } | |
202 break; | |
203 case T_DOUBLE : | |
304 | 204 __ subptr(rsp, 2*wordSize); |
0 | 205 if ( UseSSE < 2 ) { |
304 | 206 __ fstp_d(Address(rsp, 0)); |
0 | 207 } else { |
208 __ movdbl(Address(rsp, 0), xmm0); | |
209 } | |
210 break; | |
211 case T_OBJECT : | |
212 __ verify_oop(rax); // verify it | |
304 | 213 __ push(rax); |
0 | 214 break; |
215 default : ShouldNotReachHere(); | |
216 } | |
217 __ jmp(t); // return from result handler | |
218 return entry; | |
219 } | |
220 | |
221 address CppInterpreterGenerator::generate_stack_to_stack_converter(BasicType type) { | |
222 // A result is in the java expression stack of the interpreted method that has just | |
223 // returned. Place this result on the java expression stack of the caller. | |
224 // | |
304 | 225 // The current interpreter activation in rsi/r13 is for the method just returning its |
0 | 226 // result. So we know that the result of this method is on the top of the current |
227 // execution stack (which is pre-pushed) and will be return to the top of the caller | |
228 // stack. The top of the callers stack is the bottom of the locals of the current | |
229 // activation. | |
230 // Because of the way activation are managed by the frame manager the value of rsp is | |
231 // below both the stack top of the current activation and naturally the stack top | |
232 // of the calling activation. This enable this routine to leave the return address | |
233 // to the frame manager on the stack and do a vanilla return. | |
234 // | |
304 | 235 // On entry: rsi/r13 - interpreter state of activation returning a (potential) result |
236 // On Return: rsi/r13 - unchanged | |
0 | 237 // rax - new stack top for caller activation (i.e. activation in _prev_link) |
238 // | |
239 // Can destroy rdx, rcx. | |
240 // | |
241 | |
242 address entry = __ pc(); | |
243 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); | |
244 switch (type) { | |
245 case T_VOID: | |
304 | 246 __ movptr(rax, STATE(_locals)); // pop parameters get new stack value |
247 __ addptr(rax, wordSize); // account for prepush before we return | |
0 | 248 break; |
249 case T_FLOAT : | |
250 case T_BOOLEAN: | |
251 case T_CHAR : | |
252 case T_BYTE : | |
253 case T_SHORT : | |
254 case T_INT : | |
255 // 1 word result | |
304 | 256 __ movptr(rdx, STATE(_stack)); |
257 __ movptr(rax, STATE(_locals)); // address for result | |
0 | 258 __ movl(rdx, Address(rdx, wordSize)); // get result |
304 | 259 __ movptr(Address(rax, 0), rdx); // and store it |
0 | 260 break; |
261 case T_LONG : | |
262 case T_DOUBLE : | |
263 // return top two words on current expression stack to caller's expression stack | |
264 // The caller's expression stack is adjacent to the current frame manager's intepretState | |
265 // except we allocated one extra word for this intepretState so we won't overwrite it | |
266 // when we return a two word result. | |
267 | |
304 | 268 __ movptr(rax, STATE(_locals)); // address for result |
269 __ movptr(rcx, STATE(_stack)); | |
270 __ subptr(rax, wordSize); // need addition word besides locals[0] | |
271 __ movptr(rdx, Address(rcx, 2*wordSize)); // get result word (junk in 64bit) | |
272 __ movptr(Address(rax, wordSize), rdx); // and store it | |
273 __ movptr(rdx, Address(rcx, wordSize)); // get result word | |
274 __ movptr(Address(rax, 0), rdx); // and store it | |
0 | 275 break; |
276 case T_OBJECT : | |
304 | 277 __ movptr(rdx, STATE(_stack)); |
278 __ movptr(rax, STATE(_locals)); // address for result | |
279 __ movptr(rdx, Address(rdx, wordSize)); // get result | |
0 | 280 __ verify_oop(rdx); // verify it |
304 | 281 __ movptr(Address(rax, 0), rdx); // and store it |
0 | 282 break; |
283 default : ShouldNotReachHere(); | |
284 } | |
285 __ ret(0); | |
286 return entry; | |
287 } | |
288 | |
289 address CppInterpreterGenerator::generate_stack_to_native_abi_converter(BasicType type) { | |
290 // A result is in the java expression stack of the interpreted method that has just | |
291 // returned. Place this result in the native abi that the caller expects. | |
292 // | |
293 // Similar to generate_stack_to_stack_converter above. Called at a similar time from the | |
294 // frame manager execept in this situation the caller is native code (c1/c2/call_stub) | |
295 // and so rather than return result onto caller's java expression stack we return the | |
296 // result in the expected location based on the native abi. | |
304 | 297 // On entry: rsi/r13 - interpreter state of activation returning a (potential) result |
298 // On Return: rsi/r13 - unchanged | |
0 | 299 // Other registers changed [rax/rdx/ST(0) as needed for the result returned] |
300 | |
301 address entry = __ pc(); | |
302 switch (type) { | |
303 case T_VOID: | |
304 break; | |
305 case T_BOOLEAN: | |
306 case T_CHAR : | |
307 case T_BYTE : | |
308 case T_SHORT : | |
309 case T_INT : | |
304 | 310 __ movptr(rdx, STATE(_stack)); // get top of stack |
0 | 311 __ movl(rax, Address(rdx, wordSize)); // get result word 1 |
312 break; | |
313 case T_LONG : | |
304 | 314 __ movptr(rdx, STATE(_stack)); // get top of stack |
315 __ movptr(rax, Address(rdx, wordSize)); // get result low word | |
316 NOT_LP64(__ movl(rdx, Address(rdx, 2*wordSize));) // get result high word | |
0 | 317 break; |
318 case T_FLOAT : | |
304 | 319 __ movptr(rdx, STATE(_stack)); // get top of stack |
0 | 320 if ( UseSSE >= 1) { |
321 __ movflt(xmm0, Address(rdx, wordSize)); | |
322 } else { | |
323 __ fld_s(Address(rdx, wordSize)); // pushd float result | |
324 } | |
325 break; | |
326 case T_DOUBLE : | |
304 | 327 __ movptr(rdx, STATE(_stack)); // get top of stack |
0 | 328 if ( UseSSE > 1) { |
329 __ movdbl(xmm0, Address(rdx, wordSize)); | |
330 } else { | |
331 __ fld_d(Address(rdx, wordSize)); // push double result | |
332 } | |
333 break; | |
334 case T_OBJECT : | |
304 | 335 __ movptr(rdx, STATE(_stack)); // get top of stack |
336 __ movptr(rax, Address(rdx, wordSize)); // get result word 1 | |
0 | 337 __ verify_oop(rax); // verify it |
338 break; | |
339 default : ShouldNotReachHere(); | |
340 } | |
341 __ ret(0); | |
342 return entry; | |
343 } | |
344 | |
345 address CppInterpreter::return_entry(TosState state, int length) { | |
346 // make it look good in the debugger | |
347 return CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation); | |
348 } | |
349 | |
350 address CppInterpreter::deopt_entry(TosState state, int length) { | |
351 address ret = NULL; | |
352 if (length != 0) { | |
353 switch (state) { | |
354 case atos: ret = deopt_frame_manager_return_atos; break; | |
355 case btos: ret = deopt_frame_manager_return_btos; break; | |
356 case ctos: | |
357 case stos: | |
358 case itos: ret = deopt_frame_manager_return_itos; break; | |
359 case ltos: ret = deopt_frame_manager_return_ltos; break; | |
360 case ftos: ret = deopt_frame_manager_return_ftos; break; | |
361 case dtos: ret = deopt_frame_manager_return_dtos; break; | |
362 case vtos: ret = deopt_frame_manager_return_vtos; break; | |
363 } | |
364 } else { | |
365 ret = unctrap_frame_manager_entry; // re-execute the bytecode ( e.g. uncommon trap) | |
366 } | |
367 assert(ret != NULL, "Not initialized"); | |
368 return ret; | |
369 } | |
370 | |
371 // C++ Interpreter | |
372 void CppInterpreterGenerator::generate_compute_interpreter_state(const Register state, | |
373 const Register locals, | |
374 const Register sender_sp, | |
375 bool native) { | |
376 | |
377 // On entry the "locals" argument points to locals[0] (or where it would be in case no locals in | |
378 // a static method). "state" contains any previous frame manager state which we must save a link | |
379 // to in the newly generated state object. On return "state" is a pointer to the newly allocated | |
380 // state object. We must allocate and initialize a new interpretState object and the method | |
381 // expression stack. Because the returned result (if any) of the method will be placed on the caller's | |
382 // expression stack and this will overlap with locals[0] (and locals[1] if double/long) we must | |
383 // be sure to leave space on the caller's stack so that this result will not overwrite values when | |
384 // locals[0] and locals[1] do not exist (and in fact are return address and saved rbp). So when | |
385 // we are non-native we in essence ensure that locals[0-1] exist. We play an extra trick in | |
386 // non-product builds and initialize this last local with the previous interpreterState as | |
387 // this makes things look real nice in the debugger. | |
388 | |
389 // State on entry | |
390 // Assumes locals == &locals[0] | |
391 // Assumes state == any previous frame manager state (assuming call path from c++ interpreter) | |
392 // Assumes rax = return address | |
393 // rcx == senders_sp | |
394 // rbx == method | |
395 // Modifies rcx, rdx, rax | |
396 // Returns: | |
397 // state == address of new interpreterState | |
398 // rsp == bottom of method's expression stack. | |
399 | |
400 const Address const_offset (rbx, methodOopDesc::const_offset()); | |
401 | |
402 | |
403 // On entry sp is the sender's sp. This includes the space for the arguments | |
404 // that the sender pushed. If the sender pushed no args (a static) and the | |
405 // caller returns a long then we need two words on the sender's stack which | |
406 // are not present (although when we return a restore full size stack the | |
407 // space will be present). If we didn't allocate two words here then when | |
408 // we "push" the result of the caller's stack we would overwrite the return | |
409 // address and the saved rbp. Not good. So simply allocate 2 words now | |
410 // just to be safe. This is the "static long no_params() method" issue. | |
411 // See Lo.java for a testcase. | |
412 // We don't need this for native calls because they return result in | |
413 // register and the stack is expanded in the caller before we store | |
414 // the results on the stack. | |
415 | |
416 if (!native) { | |
417 #ifdef PRODUCT | |
304 | 418 __ subptr(rsp, 2*wordSize); |
0 | 419 #else /* PRODUCT */ |
304 | 420 __ push((int32_t)NULL_WORD); |
421 __ push(state); // make it look like a real argument | |
0 | 422 #endif /* PRODUCT */ |
423 } | |
424 | |
425 // Now that we are assure of space for stack result, setup typical linkage | |
426 | |
304 | 427 __ push(rax); |
0 | 428 __ enter(); |
429 | |
304 | 430 __ mov(rax, state); // save current state |
431 | |
432 __ lea(rsp, Address(rsp, -(int)sizeof(BytecodeInterpreter))); | |
433 __ mov(state, rsp); | |
434 | |
435 // rsi/r13 == state/locals rax == prevstate | |
0 | 436 |
437 // initialize the "shadow" frame so that use since C++ interpreter not directly | |
438 // recursive. Simpler to recurse but we can't trim expression stack as we call | |
439 // new methods. | |
304 | 440 __ movptr(STATE(_locals), locals); // state->_locals = locals() |
441 __ movptr(STATE(_self_link), state); // point to self | |
442 __ movptr(STATE(_prev_link), rax); // state->_link = state on entry (NULL or previous state) | |
443 __ movptr(STATE(_sender_sp), sender_sp); // state->_sender_sp = sender_sp | |
444 #ifdef _LP64 | |
445 __ movptr(STATE(_thread), r15_thread); // state->_bcp = codes() | |
446 #else | |
0 | 447 __ get_thread(rax); // get vm's javathread* |
304 | 448 __ movptr(STATE(_thread), rax); // state->_bcp = codes() |
449 #endif // _LP64 | |
450 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset())); // get constantMethodOop | |
451 __ lea(rdx, Address(rdx, constMethodOopDesc::codes_offset())); // get code base | |
0 | 452 if (native) { |
304 | 453 __ movptr(STATE(_bcp), (int32_t)NULL_WORD); // state->_bcp = NULL |
0 | 454 } else { |
304 | 455 __ movptr(STATE(_bcp), rdx); // state->_bcp = codes() |
0 | 456 } |
304 | 457 __ xorptr(rdx, rdx); |
458 __ movptr(STATE(_oop_temp), rdx); // state->_oop_temp = NULL (only really needed for native) | |
459 __ movptr(STATE(_mdx), rdx); // state->_mdx = NULL | |
460 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset())); | |
461 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes())); | |
462 __ movptr(STATE(_constants), rdx); // state->_constants = constants() | |
463 | |
464 __ movptr(STATE(_method), rbx); // state->_method = method() | |
465 __ movl(STATE(_msg), (int32_t) BytecodeInterpreter::method_entry); // state->_msg = initial method entry | |
466 __ movptr(STATE(_result._to_call._callee), (int32_t) NULL_WORD); // state->_result._to_call._callee_callee = NULL | |
467 | |
468 | |
469 __ movptr(STATE(_monitor_base), rsp); // set monitor block bottom (grows down) this would point to entry [0] | |
0 | 470 // entries run from -1..x where &monitor[x] == |
471 | |
472 { | |
473 // Must not attempt to lock method until we enter interpreter as gc won't be able to find the | |
474 // initial frame. However we allocate a free monitor so we don't have to shuffle the expression stack | |
475 // immediately. | |
476 | |
477 // synchronize method | |
478 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); | |
479 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; | |
480 Label not_synced; | |
481 | |
482 __ movl(rax, access_flags); | |
483 __ testl(rax, JVM_ACC_SYNCHRONIZED); | |
484 __ jcc(Assembler::zero, not_synced); | |
485 | |
486 // Allocate initial monitor and pre initialize it | |
487 // get synchronization object | |
488 | |
489 Label done; | |
490 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes(); | |
491 __ movl(rax, access_flags); | |
492 __ testl(rax, JVM_ACC_STATIC); | |
304 | 493 __ movptr(rax, Address(locals, 0)); // get receiver (assume this is frequent case) |
0 | 494 __ jcc(Assembler::zero, done); |
304 | 495 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset())); |
496 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes())); | |
497 __ movptr(rax, Address(rax, mirror_offset)); | |
0 | 498 __ bind(done); |
499 // add space for monitor & lock | |
304 | 500 __ subptr(rsp, entry_size); // add space for a monitor entry |
501 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object | |
0 | 502 __ bind(not_synced); |
503 } | |
504 | |
304 | 505 __ movptr(STATE(_stack_base), rsp); // set expression stack base ( == &monitors[-count]) |
0 | 506 if (native) { |
304 | 507 __ movptr(STATE(_stack), rsp); // set current expression stack tos |
508 __ movptr(STATE(_stack_limit), rsp); | |
0 | 509 } else { |
304 | 510 __ subptr(rsp, wordSize); // pre-push stack |
511 __ movptr(STATE(_stack), rsp); // set current expression stack tos | |
0 | 512 |
513 // compute full expression stack limit | |
514 | |
515 const Address size_of_stack (rbx, methodOopDesc::max_stack_offset()); | |
710 | 516 const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_words(); |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
517 __ load_unsigned_short(rdx, size_of_stack); // get size of expression stack in words |
304 | 518 __ negptr(rdx); // so we can subtract in next step |
0 | 519 // Allocate expression stack |
710 | 520 __ lea(rsp, Address(rsp, rdx, Address::times_ptr, -extra_stack)); |
304 | 521 __ movptr(STATE(_stack_limit), rsp); |
0 | 522 } |
523 | |
304 | 524 #ifdef _LP64 |
525 // Make sure stack is properly aligned and sized for the abi | |
526 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows | |
605 | 527 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI) |
304 | 528 #endif // _LP64 |
529 | |
530 | |
531 | |
0 | 532 } |
533 | |
534 // Helpers for commoning out cases in the various type of method entries. | |
535 // | |
536 | |
537 // increment invocation count & check for overflow | |
538 // | |
539 // Note: checking for negative value instead of overflow | |
540 // so we have a 'sticky' overflow test | |
541 // | |
542 // rbx,: method | |
543 // rcx: invocation counter | |
544 // | |
545 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { | |
546 | |
547 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset()); | |
548 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset()); | |
549 | |
550 if (ProfileInterpreter) { // %%% Merge this into methodDataOop | |
304 | 551 __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset())); |
0 | 552 } |
553 // Update standard invocation counters | |
554 __ movl(rax, backedge_counter); // load backedge counter | |
555 | |
556 __ increment(rcx, InvocationCounter::count_increment); | |
557 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits | |
558 | |
559 __ movl(invocation_counter, rcx); // save invocation count | |
560 __ addl(rcx, rax); // add both counters | |
561 | |
562 // profile_method is non-null only for interpreted method so | |
563 // profile_method != NULL == !native_call | |
564 // BytecodeInterpreter only calls for native so code is elided. | |
565 | |
566 __ cmp32(rcx, | |
567 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit)); | |
568 __ jcc(Assembler::aboveEqual, *overflow); | |
569 | |
570 } | |
571 | |
572 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) { | |
573 | |
574 // C++ interpreter on entry | |
304 | 575 // rsi/r13 - new interpreter state pointer |
0 | 576 // rbp - interpreter frame pointer |
577 // rbx - method | |
578 | |
579 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] | |
580 // rbx, - method | |
581 // rcx - rcvr (assuming there is one) | |
582 // top of stack return address of interpreter caller | |
583 // rsp - sender_sp | |
584 | |
585 // C++ interpreter only | |
304 | 586 // rsi/r13 - previous interpreter state pointer |
0 | 587 |
588 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); | |
589 | |
590 // InterpreterRuntime::frequency_counter_overflow takes one argument | |
591 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp). | |
592 // The call returns the address of the verified entry point for the method or NULL | |
593 // if the compilation did not complete (either went background or bailed out). | |
304 | 594 __ movptr(rax, (int32_t)false); |
0 | 595 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax); |
596 | |
597 // for c++ interpreter can rsi really be munged? | |
520
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
337
diff
changeset
|
598 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); // restore state |
304 | 599 __ movptr(rbx, Address(state, byte_offset_of(BytecodeInterpreter, _method))); // restore method |
600 __ movptr(rdi, Address(state, byte_offset_of(BytecodeInterpreter, _locals))); // get locals pointer | |
601 | |
0 | 602 __ jmp(*do_continue, relocInfo::none); |
603 | |
604 } | |
605 | |
606 void InterpreterGenerator::generate_stack_overflow_check(void) { | |
607 // see if we've got enough room on the stack for locals plus overhead. | |
608 // the expression stack grows down incrementally, so the normal guard | |
609 // page mechanism will work for that. | |
610 // | |
611 // Registers live on entry: | |
612 // | |
613 // Asm interpreter | |
614 // rdx: number of additional locals this frame needs (what we must check) | |
615 // rbx,: methodOop | |
616 | |
617 // C++ Interpreter | |
304 | 618 // rsi/r13: previous interpreter frame state object |
0 | 619 // rdi: &locals[0] |
620 // rcx: # of locals | |
621 // rdx: number of additional locals this frame needs (what we must check) | |
622 // rbx: methodOop | |
623 | |
624 // destroyed on exit | |
625 // rax, | |
626 | |
627 // NOTE: since the additional locals are also always pushed (wasn't obvious in | |
628 // generate_method_entry) so the guard should work for them too. | |
629 // | |
630 | |
631 // monitor entry size: see picture of stack set (generate_method_entry) and frame_i486.hpp | |
632 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; | |
633 | |
634 // total overhead size: entry_size + (saved rbp, thru expr stack bottom). | |
635 // be sure to change this if you add/subtract anything to/from the overhead area | |
636 const int overhead_size = (int)sizeof(BytecodeInterpreter); | |
637 | |
638 const int page_size = os::vm_page_size(); | |
639 | |
640 Label after_frame_check; | |
641 | |
642 // compute rsp as if this were going to be the last frame on | |
643 // the stack before the red zone | |
644 | |
645 Label after_frame_check_pop; | |
646 | |
647 // save rsi == caller's bytecode ptr (c++ previous interp. state) | |
648 // QQQ problem here?? rsi overload???? | |
304 | 649 __ push(state); |
650 | |
651 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rsi); | |
652 | |
653 NOT_LP64(__ get_thread(thread)); | |
0 | 654 |
655 const Address stack_base(thread, Thread::stack_base_offset()); | |
656 const Address stack_size(thread, Thread::stack_size_offset()); | |
657 | |
658 // locals + overhead, in bytes | |
659 const Address size_of_stack (rbx, methodOopDesc::max_stack_offset()); | |
660 // Always give one monitor to allow us to start interp if sync method. | |
661 // Any additional monitors need a check when moving the expression stack | |
520
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
337
diff
changeset
|
662 const int one_monitor = frame::interpreter_frame_monitor_size() * wordSize; |
710 | 663 const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries(); |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
664 __ load_unsigned_short(rax, size_of_stack); // get size of expression stack in words |
710 | 665 __ lea(rax, Address(noreg, rax, Interpreter::stackElementScale(), extra_stack + one_monitor)); |
304 | 666 __ lea(rax, Address(rax, rdx, Interpreter::stackElementScale(), overhead_size)); |
0 | 667 |
668 #ifdef ASSERT | |
669 Label stack_base_okay, stack_size_okay; | |
670 // verify that thread stack base is non-zero | |
304 | 671 __ cmpptr(stack_base, (int32_t)0); |
0 | 672 __ jcc(Assembler::notEqual, stack_base_okay); |
673 __ stop("stack base is zero"); | |
674 __ bind(stack_base_okay); | |
675 // verify that thread stack size is non-zero | |
304 | 676 __ cmpptr(stack_size, (int32_t)0); |
0 | 677 __ jcc(Assembler::notEqual, stack_size_okay); |
678 __ stop("stack size is zero"); | |
679 __ bind(stack_size_okay); | |
680 #endif | |
681 | |
682 // Add stack base to locals and subtract stack size | |
304 | 683 __ addptr(rax, stack_base); |
684 __ subptr(rax, stack_size); | |
0 | 685 |
686 // We should have a magic number here for the size of the c++ interpreter frame. | |
687 // We can't actually tell this ahead of time. The debug version size is around 3k | |
688 // product is 1k and fastdebug is 4k | |
689 const int slop = 6 * K; | |
690 | |
691 // Use the maximum number of pages we might bang. | |
692 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages : | |
693 (StackRedPages+StackYellowPages); | |
694 // Only need this if we are stack banging which is temporary while | |
695 // we're debugging. | |
304 | 696 __ addptr(rax, slop + 2*max_pages * page_size); |
0 | 697 |
698 // check against the current stack bottom | |
304 | 699 __ cmpptr(rsp, rax); |
0 | 700 __ jcc(Assembler::above, after_frame_check_pop); |
701 | |
304 | 702 __ pop(state); // get c++ prev state. |
0 | 703 |
704 // throw exception return address becomes throwing pc | |
705 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError)); | |
706 | |
707 // all done with frame size check | |
708 __ bind(after_frame_check_pop); | |
304 | 709 __ pop(state); |
0 | 710 |
711 __ bind(after_frame_check); | |
712 } | |
713 | |
714 // Find preallocated monitor and lock method (C++ interpreter) | |
715 // rbx - methodOop | |
716 // | |
717 void InterpreterGenerator::lock_method(void) { | |
304 | 718 // assumes state == rsi/r13 == pointer to current interpreterState |
719 // minimally destroys rax, rdx|c_rarg1, rdi | |
0 | 720 // |
721 // synchronize method | |
722 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; | |
723 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); | |
724 | |
304 | 725 const Register monitor = NOT_LP64(rdx) LP64_ONLY(c_rarg1); |
726 | |
0 | 727 // find initial monitor i.e. monitors[-1] |
304 | 728 __ movptr(monitor, STATE(_monitor_base)); // get monitor bottom limit |
729 __ subptr(monitor, entry_size); // point to initial monitor | |
0 | 730 |
731 #ifdef ASSERT | |
732 { Label L; | |
733 __ movl(rax, access_flags); | |
734 __ testl(rax, JVM_ACC_SYNCHRONIZED); | |
735 __ jcc(Assembler::notZero, L); | |
736 __ stop("method doesn't need synchronization"); | |
737 __ bind(L); | |
738 } | |
739 #endif // ASSERT | |
740 // get synchronization object | |
741 { Label done; | |
742 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes(); | |
743 __ movl(rax, access_flags); | |
304 | 744 __ movptr(rdi, STATE(_locals)); // prepare to get receiver (assume common case) |
0 | 745 __ testl(rax, JVM_ACC_STATIC); |
304 | 746 __ movptr(rax, Address(rdi, 0)); // get receiver (assume this is frequent case) |
0 | 747 __ jcc(Assembler::zero, done); |
304 | 748 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset())); |
749 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes())); | |
750 __ movptr(rax, Address(rax, mirror_offset)); | |
0 | 751 __ bind(done); |
752 } | |
753 #ifdef ASSERT | |
754 { Label L; | |
304 | 755 __ cmpptr(rax, Address(monitor, BasicObjectLock::obj_offset_in_bytes())); // correct object? |
0 | 756 __ jcc(Assembler::equal, L); |
757 __ stop("wrong synchronization lobject"); | |
758 __ bind(L); | |
759 } | |
760 #endif // ASSERT | |
304 | 761 // can destroy rax, rdx|c_rarg1, rcx, and (via call_VM) rdi! |
762 __ lock_object(monitor); | |
0 | 763 } |
764 | |
765 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry | |
766 | |
767 address InterpreterGenerator::generate_accessor_entry(void) { | |
768 | |
304 | 769 // rbx: methodOop |
770 | |
771 // rsi/r13: senderSP must preserved for slow path, set SP to it on fast path | |
0 | 772 |
773 Label xreturn_path; | |
774 | |
775 // do fastpath for resolved accessor methods | |
776 if (UseFastAccessorMethods) { | |
777 | |
778 address entry_point = __ pc(); | |
779 | |
780 Label slow_path; | |
781 // If we need a safepoint check, generate full interpreter entry. | |
782 ExternalAddress state(SafepointSynchronize::address_of_state()); | |
783 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), | |
784 SafepointSynchronize::_not_synchronized); | |
785 | |
786 __ jcc(Assembler::notEqual, slow_path); | |
787 // ASM/C++ Interpreter | |
788 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1 | |
789 // Note: We can only use this code if the getfield has been resolved | |
790 // and if we don't have a null-pointer exception => check for | |
791 // these conditions first and use slow path if necessary. | |
792 // rbx,: method | |
793 // rcx: receiver | |
304 | 794 __ movptr(rax, Address(rsp, wordSize)); |
0 | 795 |
796 // check if local 0 != NULL and read field | |
304 | 797 __ testptr(rax, rax); |
0 | 798 __ jcc(Assembler::zero, slow_path); |
799 | |
304 | 800 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset())); |
0 | 801 // read first instruction word and extract bytecode @ 1 and index @ 2 |
304 | 802 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset())); |
0 | 803 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset())); |
804 // Shift codes right to get the index on the right. | |
805 // The bytecode fetched looks like <index><0xb4><0x2a> | |
806 __ shrl(rdx, 2*BitsPerByte); | |
807 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); | |
304 | 808 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes())); |
0 | 809 |
810 // rax,: local 0 | |
811 // rbx,: method | |
812 // rcx: receiver - do not destroy since it is needed for slow path! | |
813 // rcx: scratch | |
814 // rdx: constant pool cache index | |
815 // rdi: constant pool cache | |
304 | 816 // rsi/r13: sender sp |
0 | 817 |
818 // check if getfield has been resolved and read constant pool cache entry | |
819 // check the validity of the cache entry by testing whether _indices field | |
820 // contains Bytecode::_getfield in b1 byte. | |
821 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below"); | |
822 __ movl(rcx, | |
823 Address(rdi, | |
824 rdx, | |
304 | 825 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset())); |
0 | 826 __ shrl(rcx, 2*BitsPerByte); |
827 __ andl(rcx, 0xFF); | |
828 __ cmpl(rcx, Bytecodes::_getfield); | |
829 __ jcc(Assembler::notEqual, slow_path); | |
830 | |
831 // Note: constant pool entry is not valid before bytecode is resolved | |
304 | 832 __ movptr(rcx, |
0 | 833 Address(rdi, |
834 rdx, | |
304 | 835 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())); |
0 | 836 __ movl(rdx, |
837 Address(rdi, | |
838 rdx, | |
304 | 839 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset())); |
0 | 840 |
841 Label notByte, notShort, notChar; | |
842 const Address field_address (rax, rcx, Address::times_1); | |
843 | |
844 // Need to differentiate between igetfield, agetfield, bgetfield etc. | |
845 // because they are different sizes. | |
846 // Use the type from the constant pool cache | |
847 __ shrl(rdx, ConstantPoolCacheEntry::tosBits); | |
848 // Make sure we don't need to mask rdx for tosBits after the above shift | |
849 ConstantPoolCacheEntry::verify_tosBits(); | |
304 | 850 #ifdef _LP64 |
851 Label notObj; | |
852 __ cmpl(rdx, atos); | |
853 __ jcc(Assembler::notEqual, notObj); | |
854 // atos | |
855 __ movptr(rax, field_address); | |
856 __ jmp(xreturn_path); | |
857 | |
858 __ bind(notObj); | |
859 #endif // _LP64 | |
0 | 860 __ cmpl(rdx, btos); |
861 __ jcc(Assembler::notEqual, notByte); | |
862 __ load_signed_byte(rax, field_address); | |
863 __ jmp(xreturn_path); | |
864 | |
865 __ bind(notByte); | |
866 __ cmpl(rdx, stos); | |
867 __ jcc(Assembler::notEqual, notShort); | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
868 __ load_signed_short(rax, field_address); |
0 | 869 __ jmp(xreturn_path); |
870 | |
871 __ bind(notShort); | |
872 __ cmpl(rdx, ctos); | |
873 __ jcc(Assembler::notEqual, notChar); | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
874 __ load_unsigned_short(rax, field_address); |
0 | 875 __ jmp(xreturn_path); |
876 | |
877 __ bind(notChar); | |
878 #ifdef ASSERT | |
879 Label okay; | |
304 | 880 #ifndef _LP64 |
0 | 881 __ cmpl(rdx, atos); |
882 __ jcc(Assembler::equal, okay); | |
304 | 883 #endif // _LP64 |
0 | 884 __ cmpl(rdx, itos); |
885 __ jcc(Assembler::equal, okay); | |
886 __ stop("what type is this?"); | |
887 __ bind(okay); | |
888 #endif // ASSERT | |
889 // All the rest are a 32 bit wordsize | |
890 __ movl(rax, field_address); | |
891 | |
892 __ bind(xreturn_path); | |
893 | |
894 // _ireturn/_areturn | |
304 | 895 __ pop(rdi); // get return address |
896 __ mov(rsp, sender_sp_on_entry); // set sp to sender sp | |
0 | 897 __ jmp(rdi); |
898 | |
899 // generate a vanilla interpreter entry as the slow path | |
900 __ bind(slow_path); | |
901 // We will enter c++ interpreter looking like it was | |
902 // called by the call_stub this will cause it to return | |
903 // a tosca result to the invoker which might have been | |
904 // the c++ interpreter itself. | |
905 | |
906 __ jmp(fast_accessor_slow_entry_path); | |
907 return entry_point; | |
908 | |
909 } else { | |
910 return NULL; | |
911 } | |
912 | |
913 } | |
914 | |
915 // | |
916 // C++ Interpreter stub for calling a native method. | |
917 // This sets up a somewhat different looking stack for calling the native method | |
918 // than the typical interpreter frame setup but still has the pointer to | |
919 // an interpreter state. | |
920 // | |
921 | |
922 address InterpreterGenerator::generate_native_entry(bool synchronized) { | |
923 // determine code generation flags | |
924 bool inc_counter = UseCompiler || CountCompiledCalls; | |
925 | |
926 // rbx: methodOop | |
927 // rcx: receiver (unused) | |
304 | 928 // rsi/r13: previous interpreter state (if called from C++ interpreter) must preserve |
929 // in any case. If called via c1/c2/call_stub rsi/r13 is junk (to use) but harmless | |
0 | 930 // to save/restore. |
931 address entry_point = __ pc(); | |
932 | |
933 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); | |
934 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset()); | |
935 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset()); | |
936 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); | |
937 | |
304 | 938 // rsi/r13 == state/locals rdi == prevstate |
0 | 939 const Register locals = rdi; |
940 | |
941 // get parameter size (always needed) | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
942 __ load_unsigned_short(rcx, size_of_parameters); |
0 | 943 |
944 // rbx: methodOop | |
945 // rcx: size of parameters | |
304 | 946 __ pop(rax); // get return address |
0 | 947 // for natives the size of locals is zero |
948 | |
949 // compute beginning of parameters /locals | |
304 | 950 __ lea(locals, Address(rsp, rcx, Address::times_ptr, -wordSize)); |
0 | 951 |
952 // initialize fixed part of activation frame | |
953 | |
954 // Assumes rax = return address | |
955 | |
956 // allocate and initialize new interpreterState and method expression stack | |
957 // IN(locals) -> locals | |
958 // IN(state) -> previous frame manager state (NULL from stub/c1/c2) | |
959 // destroys rax, rcx, rdx | |
960 // OUT (state) -> new interpreterState | |
961 // OUT(rsp) -> bottom of methods expression stack | |
962 | |
963 // save sender_sp | |
304 | 964 __ mov(rcx, sender_sp_on_entry); |
0 | 965 // start with NULL previous state |
304 | 966 __ movptr(state, (int32_t)NULL_WORD); |
0 | 967 generate_compute_interpreter_state(state, locals, rcx, true); |
968 | |
969 #ifdef ASSERT | |
970 { Label L; | |
304 | 971 __ movptr(rax, STATE(_stack_base)); |
972 #ifdef _LP64 | |
973 // duplicate the alignment rsp got after setting stack_base | |
974 __ subptr(rax, frame::arg_reg_save_area_bytes); // windows | |
605 | 975 __ andptr(rax, -16); // must be 16 byte boundary (see amd64 ABI) |
304 | 976 #endif // _LP64 |
977 __ cmpptr(rax, rsp); | |
0 | 978 __ jcc(Assembler::equal, L); |
979 __ stop("broken stack frame setup in interpreter"); | |
980 __ bind(L); | |
981 } | |
982 #endif | |
983 | |
984 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count | |
985 | |
304 | 986 const Register unlock_thread = LP64_ONLY(r15_thread) NOT_LP64(rax); |
987 NOT_LP64(__ movptr(unlock_thread, STATE(_thread));) // get thread | |
0 | 988 // Since at this point in the method invocation the exception handler |
989 // would try to exit the monitor of synchronized methods which hasn't | |
990 // been entered yet, we set the thread local variable | |
991 // _do_not_unlock_if_synchronized to true. The remove_activation will | |
992 // check this flag. | |
993 | |
304 | 994 const Address do_not_unlock_if_synchronized(unlock_thread, |
0 | 995 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); |
996 __ movbool(do_not_unlock_if_synchronized, true); | |
997 | |
998 // make sure method is native & not abstract | |
999 #ifdef ASSERT | |
1000 __ movl(rax, access_flags); | |
1001 { | |
1002 Label L; | |
1003 __ testl(rax, JVM_ACC_NATIVE); | |
1004 __ jcc(Assembler::notZero, L); | |
1005 __ stop("tried to execute non-native method as native"); | |
1006 __ bind(L); | |
1007 } | |
1008 { Label L; | |
1009 __ testl(rax, JVM_ACC_ABSTRACT); | |
1010 __ jcc(Assembler::zero, L); | |
1011 __ stop("tried to execute abstract method in interpreter"); | |
1012 __ bind(L); | |
1013 } | |
1014 #endif | |
1015 | |
1016 | |
1017 // increment invocation count & check for overflow | |
1018 Label invocation_counter_overflow; | |
1019 if (inc_counter) { | |
1020 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); | |
1021 } | |
1022 | |
1023 Label continue_after_compile; | |
1024 | |
1025 __ bind(continue_after_compile); | |
1026 | |
1027 bang_stack_shadow_pages(true); | |
1028 | |
1029 // reset the _do_not_unlock_if_synchronized flag | |
304 | 1030 NOT_LP64(__ movl(rax, STATE(_thread));) // get thread |
0 | 1031 __ movbool(do_not_unlock_if_synchronized, false); |
1032 | |
1033 | |
1034 // check for synchronized native methods | |
1035 // | |
1036 // Note: This must happen *after* invocation counter check, since | |
1037 // when overflow happens, the method should not be locked. | |
1038 if (synchronized) { | |
1039 // potentially kills rax, rcx, rdx, rdi | |
1040 lock_method(); | |
1041 } else { | |
1042 // no synchronization necessary | |
1043 #ifdef ASSERT | |
1044 { Label L; | |
1045 __ movl(rax, access_flags); | |
1046 __ testl(rax, JVM_ACC_SYNCHRONIZED); | |
1047 __ jcc(Assembler::zero, L); | |
1048 __ stop("method needs synchronization"); | |
1049 __ bind(L); | |
1050 } | |
1051 #endif | |
1052 } | |
1053 | |
1054 // start execution | |
1055 | |
1056 // jvmti support | |
1057 __ notify_method_entry(); | |
1058 | |
1059 // work registers | |
1060 const Register method = rbx; | |
304 | 1061 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rdi); |
1062 const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); // rcx|rscratch1 | |
0 | 1063 |
1064 // allocate space for parameters | |
304 | 1065 __ movptr(method, STATE(_method)); |
0 | 1066 __ verify_oop(method); |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
1067 __ load_unsigned_short(t, Address(method, methodOopDesc::size_of_parameters_offset())); |
0 | 1068 __ shll(t, 2); |
304 | 1069 #ifdef _LP64 |
1070 __ subptr(rsp, t); | |
1071 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows | |
605 | 1072 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI) |
304 | 1073 #else |
1074 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror | |
1075 __ subptr(rsp, t); | |
1076 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics | |
1077 #endif // _LP64 | |
0 | 1078 |
1079 // get signature handler | |
1080 Label pending_exception_present; | |
1081 | |
1082 { Label L; | |
304 | 1083 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); |
1084 __ testptr(t, t); | |
0 | 1085 __ jcc(Assembler::notZero, L); |
1086 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method, false); | |
304 | 1087 __ movptr(method, STATE(_method)); |
1088 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); | |
0 | 1089 __ jcc(Assembler::notEqual, pending_exception_present); |
1090 __ verify_oop(method); | |
304 | 1091 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); |
0 | 1092 __ bind(L); |
1093 } | |
1094 #ifdef ASSERT | |
1095 { | |
1096 Label L; | |
304 | 1097 __ push(t); |
0 | 1098 __ get_thread(t); // get vm's javathread* |
304 | 1099 __ cmpptr(t, STATE(_thread)); |
0 | 1100 __ jcc(Assembler::equal, L); |
1101 __ int3(); | |
1102 __ bind(L); | |
304 | 1103 __ pop(t); |
0 | 1104 } |
1105 #endif // | |
1106 | |
304 | 1107 const Register from_ptr = InterpreterRuntime::SignatureHandlerGenerator::from(); |
0 | 1108 // call signature handler |
1109 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code"); | |
304 | 1110 |
0 | 1111 // The generated handlers do not touch RBX (the method oop). |
1112 // However, large signatures cannot be cached and are generated | |
1113 // each time here. The slow-path generator will blow RBX | |
1114 // sometime, so we must reload it after the call. | |
304 | 1115 __ movptr(from_ptr, STATE(_locals)); // get the from pointer |
0 | 1116 __ call(t); |
304 | 1117 __ movptr(method, STATE(_method)); |
0 | 1118 __ verify_oop(method); |
1119 | |
1120 // result handler is in rax | |
1121 // set result handler | |
304 | 1122 __ movptr(STATE(_result_handler), rax); |
1123 | |
1124 | |
1125 // get native function entry point | |
1126 { Label L; | |
1127 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); | |
1128 __ testptr(rax, rax); | |
1129 __ jcc(Assembler::notZero, L); | |
1130 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method); | |
1131 __ movptr(method, STATE(_method)); | |
1132 __ verify_oop(method); | |
1133 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); | |
1134 __ bind(L); | |
1135 } | |
0 | 1136 |
1137 // pass mirror handle if static call | |
1138 { Label L; | |
1139 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes(); | |
1140 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); | |
1141 __ testl(t, JVM_ACC_STATIC); | |
1142 __ jcc(Assembler::zero, L); | |
1143 // get mirror | |
304 | 1144 __ movptr(t, Address(method, methodOopDesc:: constants_offset())); |
1145 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes())); | |
1146 __ movptr(t, Address(t, mirror_offset)); | |
0 | 1147 // copy mirror into activation object |
304 | 1148 __ movptr(STATE(_oop_temp), t); |
0 | 1149 // pass handle to mirror |
304 | 1150 #ifdef _LP64 |
1151 __ lea(c_rarg1, STATE(_oop_temp)); | |
1152 #else | |
1153 __ lea(t, STATE(_oop_temp)); | |
1154 __ movptr(Address(rsp, wordSize), t); | |
1155 #endif // _LP64 | |
0 | 1156 __ bind(L); |
1157 } | |
1158 #ifdef ASSERT | |
1159 { | |
1160 Label L; | |
304 | 1161 __ push(t); |
0 | 1162 __ get_thread(t); // get vm's javathread* |
304 | 1163 __ cmpptr(t, STATE(_thread)); |
0 | 1164 __ jcc(Assembler::equal, L); |
1165 __ int3(); | |
1166 __ bind(L); | |
304 | 1167 __ pop(t); |
0 | 1168 } |
1169 #endif // | |
1170 | |
1171 // pass JNIEnv | |
304 | 1172 #ifdef _LP64 |
1173 __ lea(c_rarg0, Address(thread, JavaThread::jni_environment_offset())); | |
1174 #else | |
1175 __ movptr(thread, STATE(_thread)); // get thread | |
1176 __ lea(t, Address(thread, JavaThread::jni_environment_offset())); | |
1177 | |
1178 __ movptr(Address(rsp, 0), t); | |
1179 #endif // _LP64 | |
1180 | |
0 | 1181 #ifdef ASSERT |
1182 { | |
1183 Label L; | |
304 | 1184 __ push(t); |
0 | 1185 __ get_thread(t); // get vm's javathread* |
304 | 1186 __ cmpptr(t, STATE(_thread)); |
0 | 1187 __ jcc(Assembler::equal, L); |
1188 __ int3(); | |
1189 __ bind(L); | |
304 | 1190 __ pop(t); |
0 | 1191 } |
1192 #endif // | |
1193 | |
1194 #ifdef ASSERT | |
1195 { Label L; | |
1196 __ movl(t, Address(thread, JavaThread::thread_state_offset())); | |
1197 __ cmpl(t, _thread_in_Java); | |
1198 __ jcc(Assembler::equal, L); | |
1199 __ stop("Wrong thread state in native stub"); | |
1200 __ bind(L); | |
1201 } | |
1202 #endif | |
1203 | |
1204 // Change state to native (we save the return address in the thread, since it might not | |
1205 // be pushed on the stack when we do a a stack traversal). It is enough that the pc() | |
1206 // points into the right code segment. It does not have to be the correct return pc. | |
1207 | |
1208 __ set_last_Java_frame(thread, noreg, rbp, __ pc()); | |
1209 | |
1210 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native); | |
1211 | |
1212 __ call(rax); | |
1213 | |
1214 // result potentially in rdx:rax or ST0 | |
304 | 1215 __ movptr(method, STATE(_method)); |
1216 NOT_LP64(__ movptr(thread, STATE(_thread));) // get thread | |
0 | 1217 |
1218 // The potential result is in ST(0) & rdx:rax | |
1219 // With C++ interpreter we leave any possible result in ST(0) until we are in result handler and then | |
1220 // we do the appropriate stuff for returning the result. rdx:rax must always be saved because just about | |
1221 // anything we do here will destroy it, st(0) is only saved if we re-enter the vm where it would | |
1222 // be destroyed. | |
1223 // It is safe to do these pushes because state is _thread_in_native and return address will be found | |
1224 // via _last_native_pc and not via _last_jave_sp | |
1225 | |
304 | 1226 // Must save the value of ST(0)/xmm0 since it could be destroyed before we get to result handler |
0 | 1227 { Label Lpush, Lskip; |
1228 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT)); | |
1229 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE)); | |
1230 __ cmpptr(STATE(_result_handler), float_handler.addr()); | |
1231 __ jcc(Assembler::equal, Lpush); | |
1232 __ cmpptr(STATE(_result_handler), double_handler.addr()); | |
1233 __ jcc(Assembler::notEqual, Lskip); | |
1234 __ bind(Lpush); | |
304 | 1235 __ subptr(rsp, 2*wordSize); |
1236 if ( UseSSE < 2 ) { | |
1237 __ fstp_d(Address(rsp, 0)); | |
1238 } else { | |
1239 __ movdbl(Address(rsp, 0), xmm0); | |
1240 } | |
0 | 1241 __ bind(Lskip); |
1242 } | |
1243 | |
304 | 1244 // save rax:rdx for potential use by result handler. |
1245 __ push(rax); | |
1246 #ifndef _LP64 | |
1247 __ push(rdx); | |
1248 #endif // _LP64 | |
0 | 1249 |
1250 // Either restore the MXCSR register after returning from the JNI Call | |
1251 // or verify that it wasn't changed. | |
1252 if (VM_Version::supports_sse()) { | |
1253 if (RestoreMXCSROnJNICalls) { | |
1254 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std())); | |
1255 } | |
1256 else if (CheckJNICalls ) { | |
304 | 1257 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry())); |
0 | 1258 } |
1259 } | |
1260 | |
304 | 1261 #ifndef _LP64 |
0 | 1262 // Either restore the x87 floating pointer control word after returning |
1263 // from the JNI call or verify that it wasn't changed. | |
1264 if (CheckJNICalls) { | |
304 | 1265 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry())); |
0 | 1266 } |
304 | 1267 #endif // _LP64 |
0 | 1268 |
1269 | |
1270 // change thread state | |
1271 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans); | |
1272 if(os::is_MP()) { | |
1273 // Write serialization page so VM thread can do a pseudo remote membar. | |
1274 // We use the current thread pointer to calculate a thread specific | |
1275 // offset to write to within the page. This minimizes bus traffic | |
1276 // due to cache line collision. | |
1277 __ serialize_memory(thread, rcx); | |
1278 } | |
1279 | |
1280 // check for safepoint operation in progress and/or pending suspend requests | |
1281 { Label Continue; | |
1282 | |
1283 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), | |
1284 SafepointSynchronize::_not_synchronized); | |
1285 | |
1286 // threads running native code and they are expected to self-suspend | |
1287 // when leaving the _thread_in_native state. We need to check for | |
1288 // pending suspend requests here. | |
1289 Label L; | |
1290 __ jcc(Assembler::notEqual, L); | |
1291 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0); | |
1292 __ jcc(Assembler::equal, Continue); | |
1293 __ bind(L); | |
1294 | |
1295 // Don't use call_VM as it will see a possible pending exception and forward it | |
1296 // and never return here preventing us from clearing _last_native_pc down below. | |
1297 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are | |
304 | 1298 // preserved and correspond to the bcp/locals pointers. |
0 | 1299 // |
304 | 1300 |
1301 ((MacroAssembler*)_masm)->call_VM_leaf(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), | |
1302 thread); | |
0 | 1303 __ increment(rsp, wordSize); |
1304 | |
304 | 1305 __ movptr(method, STATE(_method)); |
0 | 1306 __ verify_oop(method); |
304 | 1307 __ movptr(thread, STATE(_thread)); // get thread |
0 | 1308 |
1309 __ bind(Continue); | |
1310 } | |
1311 | |
1312 // change thread state | |
1313 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java); | |
1314 | |
1315 __ reset_last_Java_frame(thread, true, true); | |
1316 | |
1317 // reset handle block | |
304 | 1318 __ movptr(t, Address(thread, JavaThread::active_handles_offset())); |
1319 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD); | |
0 | 1320 |
1321 // If result was an oop then unbox and save it in the frame | |
1322 { Label L; | |
1323 Label no_oop, store_result; | |
1324 ExternalAddress oop_handler(AbstractInterpreter::result_handler(T_OBJECT)); | |
1325 __ cmpptr(STATE(_result_handler), oop_handler.addr()); | |
1326 __ jcc(Assembler::notEqual, no_oop); | |
304 | 1327 #ifndef _LP64 |
1328 __ pop(rdx); | |
1329 #endif // _LP64 | |
1330 __ pop(rax); | |
1331 __ testptr(rax, rax); | |
0 | 1332 __ jcc(Assembler::zero, store_result); |
1333 // unbox | |
304 | 1334 __ movptr(rax, Address(rax, 0)); |
0 | 1335 __ bind(store_result); |
304 | 1336 __ movptr(STATE(_oop_temp), rax); |
0 | 1337 // keep stack depth as expected by pushing oop which will eventually be discarded |
304 | 1338 __ push(rax); |
1339 #ifndef _LP64 | |
1340 __ push(rdx); | |
1341 #endif // _LP64 | |
0 | 1342 __ bind(no_oop); |
1343 } | |
1344 | |
1345 { | |
1346 Label no_reguard; | |
1347 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled); | |
1348 __ jcc(Assembler::notEqual, no_reguard); | |
1349 | |
304 | 1350 __ pusha(); |
0 | 1351 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); |
304 | 1352 __ popa(); |
0 | 1353 |
1354 __ bind(no_reguard); | |
1355 } | |
1356 | |
1357 | |
1358 // QQQ Seems like for native methods we simply return and the caller will see the pending | |
1359 // exception and do the right thing. Certainly the interpreter will, don't know about | |
1360 // compiled methods. | |
1361 // Seems that the answer to above is no this is wrong. The old code would see the exception | |
1362 // and forward it before doing the unlocking and notifying jvmdi that method has exited. | |
1363 // This seems wrong need to investigate the spec. | |
1364 | |
1365 // handle exceptions (exception handling will handle unlocking!) | |
1366 { Label L; | |
304 | 1367 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
0 | 1368 __ jcc(Assembler::zero, L); |
1369 __ bind(pending_exception_present); | |
1370 | |
1371 // There are potential results on the stack (rax/rdx, ST(0)) we ignore these and simply | |
1372 // return and let caller deal with exception. This skips the unlocking here which | |
1373 // seems wrong but seems to be what asm interpreter did. Can't find this in the spec. | |
1374 // Note: must preverve method in rbx | |
1375 // | |
1376 | |
1377 // remove activation | |
1378 | |
304 | 1379 __ movptr(t, STATE(_sender_sp)); |
0 | 1380 __ leave(); // remove frame anchor |
304 | 1381 __ pop(rdi); // get return address |
1382 __ movptr(state, STATE(_prev_link)); // get previous state for return | |
1383 __ mov(rsp, t); // set sp to sender sp | |
1384 __ push(rdi); // push throwing pc | |
0 | 1385 // The skips unlocking!! This seems to be what asm interpreter does but seems |
1386 // very wrong. Not clear if this violates the spec. | |
1387 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); | |
1388 __ bind(L); | |
1389 } | |
1390 | |
1391 // do unlocking if necessary | |
1392 { Label L; | |
1393 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); | |
1394 __ testl(t, JVM_ACC_SYNCHRONIZED); | |
1395 __ jcc(Assembler::zero, L); | |
1396 // the code below should be shared with interpreter macro assembler implementation | |
1397 { Label unlock; | |
304 | 1398 const Register monitor = NOT_LP64(rdx) LP64_ONLY(c_rarg1); |
0 | 1399 // BasicObjectLock will be first in list, since this is a synchronized method. However, need |
1400 // to check that the object has not been unlocked by an explicit monitorexit bytecode. | |
304 | 1401 __ movptr(monitor, STATE(_monitor_base)); |
1402 __ subptr(monitor, frame::interpreter_frame_monitor_size() * wordSize); // address of initial monitor | |
1403 | |
1404 __ movptr(t, Address(monitor, BasicObjectLock::obj_offset_in_bytes())); | |
1405 __ testptr(t, t); | |
0 | 1406 __ jcc(Assembler::notZero, unlock); |
1407 | |
1408 // Entry already unlocked, need to throw exception | |
1409 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); | |
1410 __ should_not_reach_here(); | |
1411 | |
1412 __ bind(unlock); | |
304 | 1413 __ unlock_object(monitor); |
0 | 1414 // unlock can blow rbx so restore it for path that needs it below |
304 | 1415 __ movptr(method, STATE(_method)); |
0 | 1416 } |
1417 __ bind(L); | |
1418 } | |
1419 | |
1420 // jvmti support | |
1421 // Note: This must happen _after_ handling/throwing any exceptions since | |
1422 // the exception handler code notifies the runtime of method exits | |
1423 // too. If this happens before, method entry/exit notifications are | |
1424 // not properly paired (was bug - gri 11/22/99). | |
1425 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); | |
1426 | |
1427 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result | |
304 | 1428 #ifndef _LP64 |
1429 __ pop(rdx); | |
1430 #endif // _LP64 | |
1431 __ pop(rax); | |
1432 __ movptr(t, STATE(_result_handler)); // get result handler | |
0 | 1433 __ call(t); // call result handler to convert to tosca form |
1434 | |
1435 // remove activation | |
1436 | |
304 | 1437 __ movptr(t, STATE(_sender_sp)); |
0 | 1438 |
1439 __ leave(); // remove frame anchor | |
304 | 1440 __ pop(rdi); // get return address |
1441 __ movptr(state, STATE(_prev_link)); // get previous state for return (if c++ interpreter was caller) | |
1442 __ mov(rsp, t); // set sp to sender sp | |
0 | 1443 __ jmp(rdi); |
1444 | |
1445 // invocation counter overflow | |
1446 if (inc_counter) { | |
1447 // Handle overflow of counter and compile method | |
1448 __ bind(invocation_counter_overflow); | |
1449 generate_counter_overflow(&continue_after_compile); | |
1450 } | |
1451 | |
1452 return entry_point; | |
1453 } | |
1454 | |
1455 // Generate entries that will put a result type index into rcx | |
1456 void CppInterpreterGenerator::generate_deopt_handling() { | |
1457 | |
1458 Label return_from_deopt_common; | |
1459 | |
1460 // Generate entries that will put a result type index into rcx | |
1461 // deopt needs to jump to here to enter the interpreter (return a result) | |
1462 deopt_frame_manager_return_atos = __ pc(); | |
1463 | |
1464 // rax is live here | |
1465 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_OBJECT)); // Result stub address array index | |
1466 __ jmp(return_from_deopt_common); | |
1467 | |
1468 | |
1469 // deopt needs to jump to here to enter the interpreter (return a result) | |
1470 deopt_frame_manager_return_btos = __ pc(); | |
1471 | |
1472 // rax is live here | |
1473 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_BOOLEAN)); // Result stub address array index | |
1474 __ jmp(return_from_deopt_common); | |
1475 | |
1476 // deopt needs to jump to here to enter the interpreter (return a result) | |
1477 deopt_frame_manager_return_itos = __ pc(); | |
1478 | |
1479 // rax is live here | |
1480 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_INT)); // Result stub address array index | |
1481 __ jmp(return_from_deopt_common); | |
1482 | |
1483 // deopt needs to jump to here to enter the interpreter (return a result) | |
1484 | |
1485 deopt_frame_manager_return_ltos = __ pc(); | |
1486 // rax,rdx are live here | |
1487 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_LONG)); // Result stub address array index | |
1488 __ jmp(return_from_deopt_common); | |
1489 | |
1490 // deopt needs to jump to here to enter the interpreter (return a result) | |
1491 | |
1492 deopt_frame_manager_return_ftos = __ pc(); | |
1493 // st(0) is live here | |
1494 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index | |
1495 __ jmp(return_from_deopt_common); | |
1496 | |
1497 // deopt needs to jump to here to enter the interpreter (return a result) | |
1498 deopt_frame_manager_return_dtos = __ pc(); | |
1499 | |
1500 // st(0) is live here | |
1501 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index | |
1502 __ jmp(return_from_deopt_common); | |
1503 | |
1504 // deopt needs to jump to here to enter the interpreter (return a result) | |
1505 deopt_frame_manager_return_vtos = __ pc(); | |
1506 | |
1507 __ movl(rcx, AbstractInterpreter::BasicType_as_index(T_VOID)); | |
1508 | |
1509 // Deopt return common | |
1510 // an index is present in rcx that lets us move any possible result being | |
1511 // return to the interpreter's stack | |
1512 // | |
1513 // Because we have a full sized interpreter frame on the youngest | |
1514 // activation the stack is pushed too deep to share the tosca to | |
1515 // stack converters directly. We shrink the stack to the desired | |
1516 // amount and then push result and then re-extend the stack. | |
1517 // We could have the code in size_activation layout a short | |
1518 // frame for the top activation but that would look different | |
1519 // than say sparc (which needs a full size activation because | |
1520 // the windows are in the way. Really it could be short? QQQ | |
1521 // | |
1522 __ bind(return_from_deopt_common); | |
1523 | |
304 | 1524 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); |
0 | 1525 |
1526 // setup rsp so we can push the "result" as needed. | |
304 | 1527 __ movptr(rsp, STATE(_stack)); // trim stack (is prepushed) |
1528 __ addptr(rsp, wordSize); // undo prepush | |
0 | 1529 |
1530 ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack); | |
304 | 1531 // Address index(noreg, rcx, Address::times_ptr); |
1532 __ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_ptr))); | |
1533 // __ movl(rcx, Address(noreg, rcx, Address::times_ptr, int(AbstractInterpreter::_tosca_to_stack))); | |
0 | 1534 __ call(rcx); // call result converter |
1535 | |
1536 __ movl(STATE(_msg), (int)BytecodeInterpreter::deopt_resume); | |
304 | 1537 __ lea(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present) |
1538 __ movptr(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed, | |
0 | 1539 // result if any on stack already ) |
304 | 1540 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth |
0 | 1541 } |
1542 | |
1543 // Generate the code to handle a more_monitors message from the c++ interpreter | |
1544 void CppInterpreterGenerator::generate_more_monitors() { | |
1545 | |
1546 | |
1547 Label entry, loop; | |
1548 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; | |
304 | 1549 // 1. compute new pointers // rsp: old expression stack top |
1550 __ movptr(rdx, STATE(_stack_base)); // rdx: old expression stack bottom | |
1551 __ subptr(rsp, entry_size); // move expression stack top limit | |
1552 __ subptr(STATE(_stack), entry_size); // update interpreter stack top | |
1553 __ subptr(STATE(_stack_limit), entry_size); // inform interpreter | |
1554 __ subptr(rdx, entry_size); // move expression stack bottom | |
1555 __ movptr(STATE(_stack_base), rdx); // inform interpreter | |
1556 __ movptr(rcx, STATE(_stack)); // set start value for copy loop | |
0 | 1557 __ jmp(entry); |
1558 // 2. move expression stack contents | |
1559 __ bind(loop); | |
304 | 1560 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location |
1561 __ movptr(Address(rcx, 0), rbx); // and store it at new location | |
1562 __ addptr(rcx, wordSize); // advance to next word | |
0 | 1563 __ bind(entry); |
304 | 1564 __ cmpptr(rcx, rdx); // check if bottom reached |
1565 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word | |
0 | 1566 // now zero the slot so we can find it. |
304 | 1567 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD); |
0 | 1568 __ movl(STATE(_msg), (int)BytecodeInterpreter::got_monitors); |
1569 } | |
1570 | |
1571 | |
1572 // Initial entry to C++ interpreter from the call_stub. | |
1573 // This entry point is called the frame manager since it handles the generation | |
1574 // of interpreter activation frames via requests directly from the vm (via call_stub) | |
1575 // and via requests from the interpreter. The requests from the call_stub happen | |
1576 // directly thru the entry point. Requests from the interpreter happen via returning | |
1577 // from the interpreter and examining the message the interpreter has returned to | |
1578 // the frame manager. The frame manager can take the following requests: | |
1579 | |
1580 // NO_REQUEST - error, should never happen. | |
1581 // MORE_MONITORS - need a new monitor. Shuffle the expression stack on down and | |
1582 // allocate a new monitor. | |
1583 // CALL_METHOD - setup a new activation to call a new method. Very similar to what | |
1584 // happens during entry during the entry via the call stub. | |
1585 // RETURN_FROM_METHOD - remove an activation. Return to interpreter or call stub. | |
1586 // | |
1587 // Arguments: | |
1588 // | |
1589 // rbx: methodOop | |
1590 // rcx: receiver - unused (retrieved from stack as needed) | |
304 | 1591 // rsi/r13: previous frame manager state (NULL from the call_stub/c1/c2) |
0 | 1592 // |
1593 // | |
1594 // Stack layout at entry | |
1595 // | |
1596 // [ return address ] <--- rsp | |
1597 // [ parameter n ] | |
1598 // ... | |
1599 // [ parameter 1 ] | |
1600 // [ expression stack ] | |
1601 // | |
1602 // | |
1603 // We are free to blow any registers we like because the call_stub which brought us here | |
1604 // initially has preserved the callee save registers already. | |
1605 // | |
1606 // | |
1607 | |
1608 static address interpreter_frame_manager = NULL; | |
1609 | |
1610 address InterpreterGenerator::generate_normal_entry(bool synchronized) { | |
1611 | |
1612 // rbx: methodOop | |
304 | 1613 // rsi/r13: sender sp |
0 | 1614 |
1615 // Because we redispatch "recursive" interpreter entries thru this same entry point | |
1616 // the "input" register usage is a little strange and not what you expect coming | |
1617 // from the call_stub. From the call stub rsi/rdi (current/previous) interpreter | |
1618 // state are NULL but on "recursive" dispatches they are what you'd expect. | |
1619 // rsi: current interpreter state (C++ interpreter) must preserve (null from call_stub/c1/c2) | |
1620 | |
1621 | |
1622 // A single frame manager is plenty as we don't specialize for synchronized. We could and | |
1623 // the code is pretty much ready. Would need to change the test below and for good measure | |
1624 // modify generate_interpreter_state to only do the (pre) sync stuff stuff for synchronized | |
1625 // routines. Not clear this is worth it yet. | |
1626 | |
1627 if (interpreter_frame_manager) return interpreter_frame_manager; | |
1628 | |
1629 address entry_point = __ pc(); | |
1630 | |
1631 // Fast accessor methods share this entry point. | |
1632 // This works because frame manager is in the same codelet | |
1633 if (UseFastAccessorMethods && !synchronized) __ bind(fast_accessor_slow_entry_path); | |
1634 | |
1635 Label dispatch_entry_2; | |
304 | 1636 __ movptr(rcx, sender_sp_on_entry); |
1637 __ movptr(state, (int32_t)NULL_WORD); // no current activation | |
0 | 1638 |
1639 __ jmp(dispatch_entry_2); | |
1640 | |
1641 const Register locals = rdi; | |
1642 | |
1643 Label re_dispatch; | |
1644 | |
1645 __ bind(re_dispatch); | |
1646 | |
1647 // save sender sp (doesn't include return address | |
304 | 1648 __ lea(rcx, Address(rsp, wordSize)); |
0 | 1649 |
1650 __ bind(dispatch_entry_2); | |
1651 | |
1652 // save sender sp | |
304 | 1653 __ push(rcx); |
0 | 1654 |
1655 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); | |
1656 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset()); | |
1657 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); | |
1658 | |
1659 // const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); | |
1660 // const Address monitor_block_bot (rbp, frame::interpreter_frame_initial_sp_offset * wordSize); | |
1661 // const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock)); | |
1662 | |
1663 // get parameter size (always needed) | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
1664 __ load_unsigned_short(rcx, size_of_parameters); |
0 | 1665 |
1666 // rbx: methodOop | |
1667 // rcx: size of parameters | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
1668 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words |
0 | 1669 |
304 | 1670 __ subptr(rdx, rcx); // rdx = no. of additional locals |
0 | 1671 |
1672 // see if we've got enough room on the stack for locals plus overhead. | |
1673 generate_stack_overflow_check(); // C++ | |
1674 | |
1675 // c++ interpreter does not use stack banging or any implicit exceptions | |
1676 // leave for now to verify that check is proper. | |
1677 bang_stack_shadow_pages(false); | |
1678 | |
1679 | |
1680 | |
1681 // compute beginning of parameters (rdi) | |
304 | 1682 __ lea(locals, Address(rsp, rcx, Address::times_ptr, wordSize)); |
0 | 1683 |
1684 // save sender's sp | |
1685 // __ movl(rcx, rsp); | |
1686 | |
1687 // get sender's sp | |
304 | 1688 __ pop(rcx); |
0 | 1689 |
1690 // get return address | |
304 | 1691 __ pop(rax); |
0 | 1692 |
1693 // rdx - # of additional locals | |
1694 // allocate space for locals | |
1695 // explicitly initialize locals | |
1696 { | |
1697 Label exit, loop; | |
304 | 1698 __ testl(rdx, rdx); // (32bit ok) |
0 | 1699 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0 |
1700 __ bind(loop); | |
304 | 1701 __ push((int32_t)NULL_WORD); // initialize local variables |
0 | 1702 __ decrement(rdx); // until everything initialized |
1703 __ jcc(Assembler::greater, loop); | |
1704 __ bind(exit); | |
1705 } | |
1706 | |
1707 | |
1708 // Assumes rax = return address | |
1709 | |
1710 // allocate and initialize new interpreterState and method expression stack | |
1711 // IN(locals) -> locals | |
1712 // IN(state) -> any current interpreter activation | |
1713 // destroys rax, rcx, rdx, rdi | |
1714 // OUT (state) -> new interpreterState | |
1715 // OUT(rsp) -> bottom of methods expression stack | |
1716 | |
1717 generate_compute_interpreter_state(state, locals, rcx, false); | |
1718 | |
1719 // Call interpreter | |
1720 | |
1721 Label call_interpreter; | |
1722 __ bind(call_interpreter); | |
1723 | |
1724 // c++ interpreter does not use stack banging or any implicit exceptions | |
1725 // leave for now to verify that check is proper. | |
1726 bang_stack_shadow_pages(false); | |
1727 | |
1728 | |
1729 // Call interpreter enter here if message is | |
1730 // set and we know stack size is valid | |
1731 | |
1732 Label call_interpreter_2; | |
1733 | |
1734 __ bind(call_interpreter_2); | |
1735 | |
1736 { | |
304 | 1737 const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread); |
1738 | |
1739 #ifdef _LP64 | |
1740 __ mov(c_rarg0, state); | |
1741 #else | |
1742 __ push(state); // push arg to interpreter | |
1743 __ movptr(thread, STATE(_thread)); | |
1744 #endif // _LP64 | |
0 | 1745 |
1746 // We can setup the frame anchor with everything we want at this point | |
1747 // as we are thread_in_Java and no safepoints can occur until we go to | |
1748 // vm mode. We do have to clear flags on return from vm but that is it | |
1749 // | |
304 | 1750 __ movptr(Address(thread, JavaThread::last_Java_fp_offset()), rbp); |
1751 __ movptr(Address(thread, JavaThread::last_Java_sp_offset()), rsp); | |
0 | 1752 |
1753 // Call the interpreter | |
1754 | |
1755 RuntimeAddress normal(CAST_FROM_FN_PTR(address, BytecodeInterpreter::run)); | |
1756 RuntimeAddress checking(CAST_FROM_FN_PTR(address, BytecodeInterpreter::runWithChecks)); | |
1757 | |
1758 __ call(JvmtiExport::can_post_interpreter_events() ? checking : normal); | |
304 | 1759 NOT_LP64(__ pop(rax);) // discard parameter to run |
0 | 1760 // |
1761 // state is preserved since it is callee saved | |
1762 // | |
1763 | |
1764 // reset_last_Java_frame | |
1765 | |
304 | 1766 NOT_LP64(__ movl(thread, STATE(_thread));) |
0 | 1767 __ reset_last_Java_frame(thread, true, true); |
1768 } | |
1769 | |
1770 // examine msg from interpreter to determine next action | |
1771 | |
1772 __ movl(rdx, STATE(_msg)); // Get new message | |
1773 | |
1774 Label call_method; | |
1775 Label return_from_interpreted_method; | |
1776 Label throw_exception; | |
1777 Label bad_msg; | |
1778 Label do_OSR; | |
1779 | |
304 | 1780 __ cmpl(rdx, (int32_t)BytecodeInterpreter::call_method); |
0 | 1781 __ jcc(Assembler::equal, call_method); |
304 | 1782 __ cmpl(rdx, (int32_t)BytecodeInterpreter::return_from_method); |
0 | 1783 __ jcc(Assembler::equal, return_from_interpreted_method); |
304 | 1784 __ cmpl(rdx, (int32_t)BytecodeInterpreter::do_osr); |
0 | 1785 __ jcc(Assembler::equal, do_OSR); |
304 | 1786 __ cmpl(rdx, (int32_t)BytecodeInterpreter::throwing_exception); |
0 | 1787 __ jcc(Assembler::equal, throw_exception); |
304 | 1788 __ cmpl(rdx, (int32_t)BytecodeInterpreter::more_monitors); |
0 | 1789 __ jcc(Assembler::notEqual, bad_msg); |
1790 | |
1791 // Allocate more monitor space, shuffle expression stack.... | |
1792 | |
1793 generate_more_monitors(); | |
1794 | |
1795 __ jmp(call_interpreter); | |
1796 | |
1797 // uncommon trap needs to jump to here to enter the interpreter (re-execute current bytecode) | |
1798 unctrap_frame_manager_entry = __ pc(); | |
1799 // | |
1800 // Load the registers we need. | |
304 | 1801 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); |
1802 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth | |
0 | 1803 __ jmp(call_interpreter_2); |
1804 | |
1805 | |
1806 | |
1807 //============================================================================= | |
1808 // Returning from a compiled method into a deopted method. The bytecode at the | |
1809 // bcp has completed. The result of the bytecode is in the native abi (the tosca | |
1810 // for the template based interpreter). Any stack space that was used by the | |
1811 // bytecode that has completed has been removed (e.g. parameters for an invoke) | |
1812 // so all that we have to do is place any pending result on the expression stack | |
1813 // and resume execution on the next bytecode. | |
1814 | |
1815 | |
1816 generate_deopt_handling(); | |
1817 __ jmp(call_interpreter); | |
1818 | |
1819 | |
1820 // Current frame has caught an exception we need to dispatch to the | |
1821 // handler. We can get here because a native interpreter frame caught | |
1822 // an exception in which case there is no handler and we must rethrow | |
1823 // If it is a vanilla interpreted frame the we simply drop into the | |
1824 // interpreter and let it do the lookup. | |
1825 | |
1826 Interpreter::_rethrow_exception_entry = __ pc(); | |
1827 // rax: exception | |
1828 // rdx: return address/pc that threw exception | |
1829 | |
1830 Label return_with_exception; | |
1831 Label unwind_and_forward; | |
1832 | |
1833 // restore state pointer. | |
520
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
337
diff
changeset
|
1834 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); |
304 | 1835 |
1836 __ movptr(rbx, STATE(_method)); // get method | |
1837 #ifdef _LP64 | |
1838 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax); | |
1839 #else | |
0 | 1840 __ movl(rcx, STATE(_thread)); // get thread |
1841 | |
1842 // Store exception with interpreter will expect it | |
304 | 1843 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax); |
1844 #endif // _LP64 | |
0 | 1845 |
1846 // is current frame vanilla or native? | |
1847 | |
1848 __ movl(rdx, access_flags); | |
1849 __ testl(rdx, JVM_ACC_NATIVE); | |
1850 __ jcc(Assembler::zero, return_with_exception); // vanilla interpreted frame, handle directly | |
1851 | |
1852 // We drop thru to unwind a native interpreted frame with a pending exception | |
1853 // We jump here for the initial interpreter frame with exception pending | |
1854 // We unwind the current acivation and forward it to our caller. | |
1855 | |
1856 __ bind(unwind_and_forward); | |
1857 | |
1858 // unwind rbp, return stack to unextended value and re-push return address | |
1859 | |
304 | 1860 __ movptr(rcx, STATE(_sender_sp)); |
0 | 1861 __ leave(); |
304 | 1862 __ pop(rdx); |
1863 __ mov(rsp, rcx); | |
1864 __ push(rdx); | |
0 | 1865 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
1866 | |
1867 // Return point from a call which returns a result in the native abi | |
1868 // (c1/c2/jni-native). This result must be processed onto the java | |
1869 // expression stack. | |
1870 // | |
1871 // A pending exception may be present in which case there is no result present | |
1872 | |
1873 Label resume_interpreter; | |
1874 Label do_float; | |
1875 Label do_double; | |
1876 Label done_conv; | |
1877 | |
1878 address compiled_entry = __ pc(); | |
1879 | |
1880 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases | |
1881 if (UseSSE < 2) { | |
520
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
337
diff
changeset
|
1882 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); |
304 | 1883 __ movptr(rbx, STATE(_result._to_call._callee)); // get method just executed |
0 | 1884 __ movl(rcx, Address(rbx, methodOopDesc::result_index_offset())); |
1885 __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index | |
1886 __ jcc(Assembler::equal, do_float); | |
1887 __ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index | |
1888 __ jcc(Assembler::equal, do_double); | |
520
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
337
diff
changeset
|
1889 #if !defined(_LP64) || defined(COMPILER1) || !defined(COMPILER2) |
0 | 1890 __ empty_FPU_stack(); |
1891 #endif // COMPILER2 | |
1892 __ jmp(done_conv); | |
1893 | |
1894 __ bind(do_float); | |
1895 #ifdef COMPILER2 | |
1896 for (int i = 1; i < 8; i++) { | |
1897 __ ffree(i); | |
1898 } | |
1899 #endif // COMPILER2 | |
1900 __ jmp(done_conv); | |
1901 __ bind(do_double); | |
1902 #ifdef COMPILER2 | |
1903 for (int i = 1; i < 8; i++) { | |
1904 __ ffree(i); | |
1905 } | |
1906 #endif // COMPILER2 | |
1907 __ jmp(done_conv); | |
1908 } else { | |
1909 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled"); | |
1910 __ jmp(done_conv); | |
1911 } | |
1912 | |
304 | 1913 #if 0 |
0 | 1914 // emit a sentinel we can test for when converting an interpreter |
1915 // entry point to a compiled entry point. | |
1916 __ a_long(Interpreter::return_sentinel); | |
1917 __ a_long((int)compiled_entry); | |
304 | 1918 #endif |
0 | 1919 |
1920 // Return point to interpreter from compiled/native method | |
1921 | |
1922 InternalAddress return_from_native_method(__ pc()); | |
1923 | |
1924 __ bind(done_conv); | |
1925 | |
1926 | |
1927 // Result if any is in tosca. The java expression stack is in the state that the | |
1928 // calling convention left it (i.e. params may or may not be present) | |
1929 // Copy the result from tosca and place it on java expression stack. | |
1930 | |
304 | 1931 // Restore rsi/r13 as compiled code may not preserve it |
1932 | |
520
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
337
diff
changeset
|
1933 __ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); |
0 | 1934 |
1935 // restore stack to what we had when we left (in case i2c extended it) | |
1936 | |
304 | 1937 __ movptr(rsp, STATE(_stack)); |
1938 __ lea(rsp, Address(rsp, wordSize)); | |
0 | 1939 |
1940 // If there is a pending exception then we don't really have a result to process | |
1941 | |
304 | 1942 #ifdef _LP64 |
1943 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); | |
1944 #else | |
1945 __ movptr(rcx, STATE(_thread)); // get thread | |
1946 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD); | |
520
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
337
diff
changeset
|
1947 #endif // _LP64 |
0 | 1948 __ jcc(Assembler::notZero, return_with_exception); |
1949 | |
1950 // get method just executed | |
304 | 1951 __ movptr(rbx, STATE(_result._to_call._callee)); |
0 | 1952 |
1953 // callee left args on top of expression stack, remove them | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
1954 __ load_unsigned_short(rcx, Address(rbx, methodOopDesc::size_of_parameters_offset())); |
304 | 1955 __ lea(rsp, Address(rsp, rcx, Address::times_ptr)); |
0 | 1956 |
1957 __ movl(rcx, Address(rbx, methodOopDesc::result_index_offset())); | |
1958 ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack); | |
304 | 1959 // Address index(noreg, rax, Address::times_ptr); |
1960 __ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_ptr))); | |
1961 // __ movl(rcx, Address(noreg, rcx, Address::times_ptr, int(AbstractInterpreter::_tosca_to_stack))); | |
0 | 1962 __ call(rcx); // call result converter |
1963 __ jmp(resume_interpreter); | |
1964 | |
1965 // An exception is being caught on return to a vanilla interpreter frame. | |
1966 // Empty the stack and resume interpreter | |
1967 | |
1968 __ bind(return_with_exception); | |
1969 | |
1970 // Exception present, empty stack | |
304 | 1971 __ movptr(rsp, STATE(_stack_base)); |
0 | 1972 __ jmp(resume_interpreter); |
1973 | |
1974 // Return from interpreted method we return result appropriate to the caller (i.e. "recursive" | |
1975 // interpreter call, or native) and unwind this interpreter activation. | |
1976 // All monitors should be unlocked. | |
1977 | |
1978 __ bind(return_from_interpreted_method); | |
1979 | |
1980 Label return_to_initial_caller; | |
1981 | |
304 | 1982 __ movptr(rbx, STATE(_method)); // get method just executed |
1983 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from "recursive" interpreter call? | |
0 | 1984 __ movl(rax, Address(rbx, methodOopDesc::result_index_offset())); // get result type index |
1985 __ jcc(Assembler::equal, return_to_initial_caller); // back to native code (call_stub/c1/c2) | |
1986 | |
1987 // Copy result to callers java stack | |
1988 ExternalAddress stack_to_stack((address)CppInterpreter::_stack_to_stack); | |
304 | 1989 // Address index(noreg, rax, Address::times_ptr); |
1990 | |
1991 __ movptr(rax, ArrayAddress(stack_to_stack, Address(noreg, rax, Address::times_ptr))); | |
1992 // __ movl(rax, Address(noreg, rax, Address::times_ptr, int(AbstractInterpreter::_stack_to_stack))); | |
0 | 1993 __ call(rax); // call result converter |
1994 | |
1995 Label unwind_recursive_activation; | |
1996 __ bind(unwind_recursive_activation); | |
1997 | |
1998 // returning to interpreter method from "recursive" interpreter call | |
1999 // result converter left rax pointing to top of the java stack for method we are returning | |
2000 // to. Now all we must do is unwind the state from the completed call | |
2001 | |
304 | 2002 __ movptr(state, STATE(_prev_link)); // unwind state |
0 | 2003 __ leave(); // pop the frame |
304 | 2004 __ mov(rsp, rax); // unwind stack to remove args |
0 | 2005 |
2006 // Resume the interpreter. The current frame contains the current interpreter | |
2007 // state object. | |
2008 // | |
2009 | |
2010 __ bind(resume_interpreter); | |
2011 | |
2012 // state == interpreterState object for method we are resuming | |
2013 | |
2014 __ movl(STATE(_msg), (int)BytecodeInterpreter::method_resume); | |
304 | 2015 __ lea(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present) |
2016 __ movptr(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed, | |
0 | 2017 // result if any on stack already ) |
304 | 2018 __ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth |
0 | 2019 __ jmp(call_interpreter_2); // No need to bang |
2020 | |
2021 // interpreter returning to native code (call_stub/c1/c2) | |
2022 // convert result and unwind initial activation | |
2023 // rax - result index | |
2024 | |
2025 __ bind(return_to_initial_caller); | |
2026 ExternalAddress stack_to_native((address)CppInterpreter::_stack_to_native_abi); | |
304 | 2027 // Address index(noreg, rax, Address::times_ptr); |
2028 | |
2029 __ movptr(rax, ArrayAddress(stack_to_native, Address(noreg, rax, Address::times_ptr))); | |
0 | 2030 __ call(rax); // call result converter |
2031 | |
2032 Label unwind_initial_activation; | |
2033 __ bind(unwind_initial_activation); | |
2034 | |
2035 // RETURN TO CALL_STUB/C1/C2 code (result if any in rax/rdx ST(0)) | |
2036 | |
2037 /* Current stack picture | |
2038 | |
2039 [ incoming parameters ] | |
2040 [ extra locals ] | |
2041 [ return address to CALL_STUB/C1/C2] | |
2042 fp -> [ CALL_STUB/C1/C2 fp ] | |
2043 BytecodeInterpreter object | |
2044 expression stack | |
2045 sp -> | |
2046 | |
2047 */ | |
2048 | |
2049 // return restoring the stack to the original sender_sp value | |
2050 | |
304 | 2051 __ movptr(rcx, STATE(_sender_sp)); |
0 | 2052 __ leave(); |
304 | 2053 __ pop(rdi); // get return address |
0 | 2054 // set stack to sender's sp |
304 | 2055 __ mov(rsp, rcx); |
0 | 2056 __ jmp(rdi); // return to call_stub |
2057 | |
2058 // OSR request, adjust return address to make current frame into adapter frame | |
2059 // and enter OSR nmethod | |
2060 | |
2061 __ bind(do_OSR); | |
2062 | |
2063 Label remove_initial_frame; | |
2064 | |
2065 // We are going to pop this frame. Is there another interpreter frame underneath | |
2066 // it or is it callstub/compiled? | |
2067 | |
2068 // Move buffer to the expected parameter location | |
304 | 2069 __ movptr(rcx, STATE(_result._osr._osr_buf)); |
2070 | |
2071 __ movptr(rax, STATE(_result._osr._osr_entry)); | |
2072 | |
2073 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from "recursive" interpreter call? | |
0 | 2074 __ jcc(Assembler::equal, remove_initial_frame); // back to native code (call_stub/c1/c2) |
2075 | |
304 | 2076 __ movptr(sender_sp_on_entry, STATE(_sender_sp)); // get sender's sp in expected register |
0 | 2077 __ leave(); // pop the frame |
304 | 2078 __ mov(rsp, sender_sp_on_entry); // trim any stack expansion |
0 | 2079 |
2080 | |
2081 // We know we are calling compiled so push specialized return | |
2082 // method uses specialized entry, push a return so we look like call stub setup | |
2083 // this path will handle fact that result is returned in registers and not | |
2084 // on the java stack. | |
2085 | |
2086 __ pushptr(return_from_native_method.addr()); | |
2087 | |
2088 __ jmp(rax); | |
2089 | |
2090 __ bind(remove_initial_frame); | |
2091 | |
304 | 2092 __ movptr(rdx, STATE(_sender_sp)); |
0 | 2093 __ leave(); |
2094 // get real return | |
304 | 2095 __ pop(rsi); |
0 | 2096 // set stack to sender's sp |
304 | 2097 __ mov(rsp, rdx); |
0 | 2098 // repush real return |
304 | 2099 __ push(rsi); |
0 | 2100 // Enter OSR nmethod |
2101 __ jmp(rax); | |
2102 | |
2103 | |
2104 | |
2105 | |
2106 // Call a new method. All we do is (temporarily) trim the expression stack | |
2107 // push a return address to bring us back to here and leap to the new entry. | |
2108 | |
2109 __ bind(call_method); | |
2110 | |
2111 // stack points to next free location and not top element on expression stack | |
2112 // method expects sp to be pointing to topmost element | |
2113 | |
304 | 2114 __ movptr(rsp, STATE(_stack)); // pop args to c++ interpreter, set sp to java stack top |
2115 __ lea(rsp, Address(rsp, wordSize)); | |
2116 | |
2117 __ movptr(rbx, STATE(_result._to_call._callee)); // get method to execute | |
0 | 2118 |
2119 // don't need a return address if reinvoking interpreter | |
2120 | |
2121 // Make it look like call_stub calling conventions | |
2122 | |
2123 // Get (potential) receiver | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
605
diff
changeset
|
2124 __ load_unsigned_short(rcx, size_of_parameters); // get size of parameters in words |
0 | 2125 |
2126 ExternalAddress recursive(CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation)); | |
2127 __ pushptr(recursive.addr()); // make it look good in the debugger | |
2128 | |
2129 InternalAddress entry(entry_point); | |
2130 __ cmpptr(STATE(_result._to_call._callee_entry_point), entry.addr()); // returning to interpreter? | |
2131 __ jcc(Assembler::equal, re_dispatch); // yes | |
2132 | |
304 | 2133 __ pop(rax); // pop dummy address |
0 | 2134 |
2135 | |
2136 // get specialized entry | |
304 | 2137 __ movptr(rax, STATE(_result._to_call._callee_entry_point)); |
0 | 2138 // set sender SP |
304 | 2139 __ mov(sender_sp_on_entry, rsp); |
0 | 2140 |
2141 // method uses specialized entry, push a return so we look like call stub setup | |
2142 // this path will handle fact that result is returned in registers and not | |
2143 // on the java stack. | |
2144 | |
2145 __ pushptr(return_from_native_method.addr()); | |
2146 | |
2147 __ jmp(rax); | |
2148 | |
2149 __ bind(bad_msg); | |
2150 __ stop("Bad message from interpreter"); | |
2151 | |
2152 // Interpreted method "returned" with an exception pass it on... | |
2153 // Pass result, unwind activation and continue/return to interpreter/call_stub | |
2154 // We handle result (if any) differently based on return to interpreter or call_stub | |
2155 | |
2156 Label unwind_initial_with_pending_exception; | |
2157 | |
2158 __ bind(throw_exception); | |
304 | 2159 __ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from recursive interpreter call? |
0 | 2160 __ jcc(Assembler::equal, unwind_initial_with_pending_exception); // no, back to native code (call_stub/c1/c2) |
304 | 2161 __ movptr(rax, STATE(_locals)); // pop parameters get new stack value |
2162 __ addptr(rax, wordSize); // account for prepush before we return | |
0 | 2163 __ jmp(unwind_recursive_activation); |
2164 | |
2165 __ bind(unwind_initial_with_pending_exception); | |
2166 | |
2167 // We will unwind the current (initial) interpreter frame and forward | |
2168 // the exception to the caller. We must put the exception in the | |
2169 // expected register and clear pending exception and then forward. | |
2170 | |
2171 __ jmp(unwind_and_forward); | |
2172 | |
2173 interpreter_frame_manager = entry_point; | |
2174 return entry_point; | |
2175 } | |
2176 | |
2177 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) { | |
2178 // determine code generation flags | |
2179 bool synchronized = false; | |
2180 address entry_point = NULL; | |
2181 | |
2182 switch (kind) { | |
2183 case Interpreter::zerolocals : break; | |
2184 case Interpreter::zerolocals_synchronized: synchronized = true; break; | |
2185 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break; | |
2186 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break; | |
2187 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break; | |
2188 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break; | |
2189 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break; | |
710 | 2190 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break; |
0 | 2191 |
2192 case Interpreter::java_lang_math_sin : // fall thru | |
2193 case Interpreter::java_lang_math_cos : // fall thru | |
2194 case Interpreter::java_lang_math_tan : // fall thru | |
2195 case Interpreter::java_lang_math_abs : // fall thru | |
2196 case Interpreter::java_lang_math_log : // fall thru | |
2197 case Interpreter::java_lang_math_log10 : // fall thru | |
2198 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break; | |
2199 default : ShouldNotReachHere(); break; | |
2200 } | |
2201 | |
2202 if (entry_point) return entry_point; | |
2203 | |
2204 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized); | |
2205 | |
2206 } | |
2207 | |
2208 InterpreterGenerator::InterpreterGenerator(StubQueue* code) | |
2209 : CppInterpreterGenerator(code) { | |
2210 generate_all(); // down here so it can be "virtual" | |
2211 } | |
2212 | |
2213 // Deoptimization helpers for C++ interpreter | |
2214 | |
2215 // How much stack a method activation needs in words. | |
2216 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) { | |
2217 | |
2218 const int stub_code = 4; // see generate_call_stub | |
2219 // Save space for one monitor to get into the interpreted method in case | |
2220 // the method is synchronized | |
2221 int monitor_size = method->is_synchronized() ? | |
2222 1*frame::interpreter_frame_monitor_size() : 0; | |
2223 | |
2224 // total static overhead size. Account for interpreter state object, return | |
2225 // address, saved rbp and 2 words for a "static long no_params() method" issue. | |
2226 | |
2227 const int overhead_size = sizeof(BytecodeInterpreter)/wordSize + | |
2228 ( frame::sender_sp_offset - frame::link_offset) + 2; | |
2229 | |
710 | 2230 const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries(); |
2231 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) * | |
0 | 2232 Interpreter::stackElementWords(); |
2233 return overhead_size + method_stack + stub_code; | |
2234 } | |
2235 | |
2236 // returns the activation size. | |
2237 static int size_activation_helper(int extra_locals_size, int monitor_size) { | |
2238 return (extra_locals_size + // the addition space for locals | |
2239 2*BytesPerWord + // return address and saved rbp | |
2240 2*BytesPerWord + // "static long no_params() method" issue | |
2241 sizeof(BytecodeInterpreter) + // interpreterState | |
2242 monitor_size); // monitors | |
2243 } | |
2244 | |
2245 void BytecodeInterpreter::layout_interpreterState(interpreterState to_fill, | |
2246 frame* caller, | |
2247 frame* current, | |
2248 methodOop method, | |
2249 intptr_t* locals, | |
2250 intptr_t* stack, | |
2251 intptr_t* stack_base, | |
2252 intptr_t* monitor_base, | |
2253 intptr_t* frame_bottom, | |
2254 bool is_top_frame | |
2255 ) | |
2256 { | |
2257 // What about any vtable? | |
2258 // | |
2259 to_fill->_thread = JavaThread::current(); | |
2260 // This gets filled in later but make it something recognizable for now | |
2261 to_fill->_bcp = method->code_base(); | |
2262 to_fill->_locals = locals; | |
2263 to_fill->_constants = method->constants()->cache(); | |
2264 to_fill->_method = method; | |
2265 to_fill->_mdx = NULL; | |
2266 to_fill->_stack = stack; | |
2267 if (is_top_frame && JavaThread::current()->popframe_forcing_deopt_reexecution() ) { | |
2268 to_fill->_msg = deopt_resume2; | |
2269 } else { | |
2270 to_fill->_msg = method_resume; | |
2271 } | |
2272 to_fill->_result._to_call._bcp_advance = 0; | |
2273 to_fill->_result._to_call._callee_entry_point = NULL; // doesn't matter to anyone | |
2274 to_fill->_result._to_call._callee = NULL; // doesn't matter to anyone | |
2275 to_fill->_prev_link = NULL; | |
2276 | |
2277 to_fill->_sender_sp = caller->unextended_sp(); | |
2278 | |
2279 if (caller->is_interpreted_frame()) { | |
2280 interpreterState prev = caller->get_interpreterState(); | |
2281 to_fill->_prev_link = prev; | |
2282 // *current->register_addr(GR_Iprev_state) = (intptr_t) prev; | |
2283 // Make the prev callee look proper | |
2284 prev->_result._to_call._callee = method; | |
2285 if (*prev->_bcp == Bytecodes::_invokeinterface) { | |
2286 prev->_result._to_call._bcp_advance = 5; | |
2287 } else { | |
2288 prev->_result._to_call._bcp_advance = 3; | |
2289 } | |
2290 } | |
2291 to_fill->_oop_temp = NULL; | |
2292 to_fill->_stack_base = stack_base; | |
2293 // Need +1 here because stack_base points to the word just above the first expr stack entry | |
2294 // and stack_limit is supposed to point to the word just below the last expr stack entry. | |
2295 // See generate_compute_interpreter_state. | |
710 | 2296 int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries(); |
2297 to_fill->_stack_limit = stack_base - (method->max_stack() + extra_stack + 1); | |
0 | 2298 to_fill->_monitor_base = (BasicObjectLock*) monitor_base; |
2299 | |
2300 to_fill->_self_link = to_fill; | |
2301 assert(stack >= to_fill->_stack_limit && stack < to_fill->_stack_base, | |
2302 "Stack top out of range"); | |
2303 } | |
2304 | |
2305 int AbstractInterpreter::layout_activation(methodOop method, | |
2306 int tempcount, // | |
2307 int popframe_extra_args, | |
2308 int moncount, | |
2309 int callee_param_count, | |
2310 int callee_locals, | |
2311 frame* caller, | |
2312 frame* interpreter_frame, | |
2313 bool is_top_frame) { | |
2314 | |
2315 assert(popframe_extra_args == 0, "FIX ME"); | |
2316 // NOTE this code must exactly mimic what InterpreterGenerator::generate_compute_interpreter_state() | |
2317 // does as far as allocating an interpreter frame. | |
2318 // If interpreter_frame!=NULL, set up the method, locals, and monitors. | |
2319 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size, | |
2320 // as determined by a previous call to this method. | |
2321 // It is also guaranteed to be walkable even though it is in a skeletal state | |
2322 // NOTE: return size is in words not bytes | |
2323 // NOTE: tempcount is the current size of the java expression stack. For top most | |
2324 // frames we will allocate a full sized expression stack and not the curback | |
2325 // version that non-top frames have. | |
2326 | |
2327 // Calculate the amount our frame will be adjust by the callee. For top frame | |
2328 // this is zero. | |
2329 | |
2330 // NOTE: ia64 seems to do this wrong (or at least backwards) in that it | |
2331 // calculates the extra locals based on itself. Not what the callee does | |
2332 // to it. So it ignores last_frame_adjust value. Seems suspicious as far | |
2333 // as getting sender_sp correct. | |
2334 | |
2335 int extra_locals_size = (callee_locals - callee_param_count) * BytesPerWord; | |
2336 int monitor_size = sizeof(BasicObjectLock) * moncount; | |
2337 | |
2338 // First calculate the frame size without any java expression stack | |
2339 int short_frame_size = size_activation_helper(extra_locals_size, | |
2340 monitor_size); | |
2341 | |
2342 // Now with full size expression stack | |
710 | 2343 int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries(); |
2344 int full_frame_size = short_frame_size + (method->max_stack() + extra_stack) * BytesPerWord; | |
0 | 2345 |
2346 // and now with only live portion of the expression stack | |
2347 short_frame_size = short_frame_size + tempcount * BytesPerWord; | |
2348 | |
2349 // the size the activation is right now. Only top frame is full size | |
2350 int frame_size = (is_top_frame ? full_frame_size : short_frame_size); | |
2351 | |
2352 if (interpreter_frame != NULL) { | |
2353 #ifdef ASSERT | |
2354 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable"); | |
2355 #endif | |
2356 | |
2357 // MUCHO HACK | |
2358 | |
2359 intptr_t* frame_bottom = (intptr_t*) ((intptr_t)interpreter_frame->sp() - (full_frame_size - frame_size)); | |
2360 | |
2361 /* Now fillin the interpreterState object */ | |
2362 | |
2363 // The state object is the first thing on the frame and easily located | |
2364 | |
2365 interpreterState cur_state = (interpreterState) ((intptr_t)interpreter_frame->fp() - sizeof(BytecodeInterpreter)); | |
2366 | |
2367 | |
2368 // Find the locals pointer. This is rather simple on x86 because there is no | |
2369 // confusing rounding at the callee to account for. We can trivially locate | |
2370 // our locals based on the current fp(). | |
2371 // Note: the + 2 is for handling the "static long no_params() method" issue. | |
2372 // (too bad I don't really remember that issue well...) | |
2373 | |
2374 intptr_t* locals; | |
2375 // If the caller is interpreted we need to make sure that locals points to the first | |
2376 // argument that the caller passed and not in an area where the stack might have been extended. | |
2377 // because the stack to stack to converter needs a proper locals value in order to remove the | |
2378 // arguments from the caller and place the result in the proper location. Hmm maybe it'd be | |
2379 // simpler if we simply stored the result in the BytecodeInterpreter object and let the c++ code | |
2380 // adjust the stack?? HMMM QQQ | |
2381 // | |
2382 if (caller->is_interpreted_frame()) { | |
2383 // locals must agree with the caller because it will be used to set the | |
2384 // caller's tos when we return. | |
2385 interpreterState prev = caller->get_interpreterState(); | |
2386 // stack() is prepushed. | |
2387 locals = prev->stack() + method->size_of_parameters(); | |
2388 // locals = caller->unextended_sp() + (method->size_of_parameters() - 1); | |
2389 if (locals != interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2) { | |
2390 // os::breakpoint(); | |
2391 } | |
2392 } else { | |
2393 // this is where a c2i would have placed locals (except for the +2) | |
2394 locals = interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2; | |
2395 } | |
2396 | |
2397 intptr_t* monitor_base = (intptr_t*) cur_state; | |
2398 intptr_t* stack_base = (intptr_t*) ((intptr_t) monitor_base - monitor_size); | |
2399 /* +1 because stack is always prepushed */ | |
2400 intptr_t* stack = (intptr_t*) ((intptr_t) stack_base - (tempcount + 1) * BytesPerWord); | |
2401 | |
2402 | |
2403 BytecodeInterpreter::layout_interpreterState(cur_state, | |
2404 caller, | |
2405 interpreter_frame, | |
2406 method, | |
2407 locals, | |
2408 stack, | |
2409 stack_base, | |
2410 monitor_base, | |
2411 frame_bottom, | |
2412 is_top_frame); | |
2413 | |
2414 // BytecodeInterpreter::pd_layout_interpreterState(cur_state, interpreter_return_address, interpreter_frame->fp()); | |
2415 } | |
2416 return frame_size/BytesPerWord; | |
2417 } | |
2418 | |
2419 #endif // CC_INTERP (all) |