Mercurial > hg > graal-compiler
annotate src/cpu/x86/vm/templateInterpreter_x86_32.cpp @ 710:e5b0439ef4ae
6655638: dynamic languages need method handles
Summary: initial implementation, with known omissions (x86/64, sparc, compiler optim., c-oops, C++ interp.)
Reviewed-by: kvn, twisti, never
author | jrose |
---|---|
date | Wed, 08 Apr 2009 10:56:49 -0700 |
parents | 7bb995fbd3c0 |
children | be93aad57795 |
rev | line source |
---|---|
0 | 1 /* |
579 | 2 * Copyright 1997-2009 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/_templateInterpreter_x86_32.cpp.incl" | |
27 | |
28 #define __ _masm-> | |
29 | |
30 | |
31 #ifndef CC_INTERP | |
32 const int method_offset = frame::interpreter_frame_method_offset * wordSize; | |
33 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize; | |
34 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize; | |
35 | |
36 //------------------------------------------------------------------------------------------------------------------------ | |
37 | |
38 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { | |
39 address entry = __ pc(); | |
40 | |
41 // Note: There should be a minimal interpreter frame set up when stack | |
42 // overflow occurs since we check explicitly for it now. | |
43 // | |
44 #ifdef ASSERT | |
45 { Label L; | |
304 | 46 __ lea(rax, Address(rbp, |
0 | 47 frame::interpreter_frame_monitor_block_top_offset * wordSize)); |
304 | 48 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp, |
0 | 49 // (stack grows negative) |
50 __ jcc(Assembler::aboveEqual, L); // check if frame is complete | |
51 __ stop ("interpreter frame not set up"); | |
52 __ bind(L); | |
53 } | |
54 #endif // ASSERT | |
55 // Restore bcp under the assumption that the current frame is still | |
56 // interpreted | |
57 __ restore_bcp(); | |
58 | |
59 // expression stack must be empty before entering the VM if an exception | |
60 // happened | |
61 __ empty_expression_stack(); | |
62 __ empty_FPU_stack(); | |
63 // throw exception | |
64 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError)); | |
65 return entry; | |
66 } | |
67 | |
68 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) { | |
69 address entry = __ pc(); | |
70 // expression stack must be empty before entering the VM if an exception happened | |
71 __ empty_expression_stack(); | |
72 __ empty_FPU_stack(); | |
73 // setup parameters | |
74 // ??? convention: expect aberrant index in register rbx, | |
75 __ lea(rax, ExternalAddress((address)name)); | |
76 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx); | |
77 return entry; | |
78 } | |
79 | |
80 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { | |
81 address entry = __ pc(); | |
82 // object is at TOS | |
304 | 83 __ pop(rax); |
0 | 84 // expression stack must be empty before entering the VM if an exception |
85 // happened | |
86 __ empty_expression_stack(); | |
87 __ empty_FPU_stack(); | |
88 __ call_VM(noreg, | |
89 CAST_FROM_FN_PTR(address, | |
90 InterpreterRuntime::throw_ClassCastException), | |
91 rax); | |
92 return entry; | |
93 } | |
94 | |
710 | 95 // Arguments are: required type at TOS+8, failing object (or NULL) at TOS+4. |
96 // pc at TOS (just for debugging) | |
97 address TemplateInterpreterGenerator::generate_WrongMethodType_handler() { | |
98 address entry = __ pc(); | |
99 | |
100 __ pop(rbx); // actual failing object is at TOS | |
101 __ pop(rax); // required type is at TOS+4 | |
102 | |
103 __ verify_oop(rbx); | |
104 __ verify_oop(rax); | |
105 | |
106 // Various method handle types use interpreter registers as temps. | |
107 __ restore_bcp(); | |
108 __ restore_locals(); | |
109 | |
110 // Expression stack must be empty before entering the VM for an exception. | |
111 __ empty_expression_stack(); | |
112 __ empty_FPU_stack(); | |
113 __ call_VM(noreg, | |
114 CAST_FROM_FN_PTR(address, | |
115 InterpreterRuntime::throw_WrongMethodTypeException), | |
116 // pass required type, failing object (or NULL) | |
117 rax, rbx); | |
118 return entry; | |
119 } | |
120 | |
121 | |
0 | 122 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) { |
123 assert(!pass_oop || message == NULL, "either oop or message but not both"); | |
124 address entry = __ pc(); | |
125 if (pass_oop) { | |
126 // object is at TOS | |
304 | 127 __ pop(rbx); |
0 | 128 } |
129 // expression stack must be empty before entering the VM if an exception happened | |
130 __ empty_expression_stack(); | |
131 __ empty_FPU_stack(); | |
132 // setup parameters | |
133 __ lea(rax, ExternalAddress((address)name)); | |
134 if (pass_oop) { | |
135 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx); | |
136 } else { | |
137 if (message != NULL) { | |
138 __ lea(rbx, ExternalAddress((address)message)); | |
139 } else { | |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
337
diff
changeset
|
140 __ movptr(rbx, NULL_WORD); |
0 | 141 } |
142 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx); | |
143 } | |
144 // throw exception | |
145 __ jump(ExternalAddress(Interpreter::throw_exception_entry())); | |
146 return entry; | |
147 } | |
148 | |
149 | |
150 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { | |
151 address entry = __ pc(); | |
152 // NULL last_sp until next java call | |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
337
diff
changeset
|
153 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); |
0 | 154 __ dispatch_next(state); |
155 return entry; | |
156 } | |
157 | |
158 | |
159 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) { | |
160 Label interpreter_entry; | |
161 address compiled_entry = __ pc(); | |
162 | |
163 #ifdef COMPILER2 | |
164 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases | |
165 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) { | |
166 for (int i = 1; i < 8; i++) { | |
167 __ ffree(i); | |
168 } | |
169 } else if (UseSSE < 2) { | |
170 __ empty_FPU_stack(); | |
171 } | |
172 #endif | |
173 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) { | |
174 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled"); | |
175 } else { | |
176 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled"); | |
177 } | |
178 | |
179 __ jmp(interpreter_entry, relocInfo::none); | |
180 // emit a sentinel we can test for when converting an interpreter | |
181 // entry point to a compiled entry point. | |
182 __ a_long(Interpreter::return_sentinel); | |
183 __ a_long((int)compiled_entry); | |
184 address entry = __ pc(); | |
185 __ bind(interpreter_entry); | |
186 | |
187 // In SSE mode, interpreter returns FP results in xmm0 but they need | |
188 // to end up back on the FPU so it can operate on them. | |
189 if (state == ftos && UseSSE >= 1) { | |
304 | 190 __ subptr(rsp, wordSize); |
0 | 191 __ movflt(Address(rsp, 0), xmm0); |
192 __ fld_s(Address(rsp, 0)); | |
304 | 193 __ addptr(rsp, wordSize); |
0 | 194 } else if (state == dtos && UseSSE >= 2) { |
304 | 195 __ subptr(rsp, 2*wordSize); |
0 | 196 __ movdbl(Address(rsp, 0), xmm0); |
197 __ fld_d(Address(rsp, 0)); | |
304 | 198 __ addptr(rsp, 2*wordSize); |
0 | 199 } |
200 | |
201 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter"); | |
202 | |
203 // Restore stack bottom in case i2c adjusted stack | |
304 | 204 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); |
0 | 205 // and NULL it as marker that rsp is now tos until next java call |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
337
diff
changeset
|
206 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); |
0 | 207 |
208 __ restore_bcp(); | |
209 __ restore_locals(); | |
210 __ get_cache_and_index_at_bcp(rbx, rcx, 1); | |
211 __ movl(rbx, Address(rbx, rcx, | |
304 | 212 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + |
0 | 213 ConstantPoolCacheEntry::flags_offset())); |
304 | 214 __ andptr(rbx, 0xFF); |
215 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); | |
0 | 216 __ dispatch_next(state, step); |
217 return entry; | |
218 } | |
219 | |
220 | |
221 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) { | |
222 address entry = __ pc(); | |
223 | |
224 // In SSE mode, FP results are in xmm0 | |
225 if (state == ftos && UseSSE > 0) { | |
304 | 226 __ subptr(rsp, wordSize); |
0 | 227 __ movflt(Address(rsp, 0), xmm0); |
228 __ fld_s(Address(rsp, 0)); | |
304 | 229 __ addptr(rsp, wordSize); |
0 | 230 } else if (state == dtos && UseSSE >= 2) { |
304 | 231 __ subptr(rsp, 2*wordSize); |
0 | 232 __ movdbl(Address(rsp, 0), xmm0); |
233 __ fld_d(Address(rsp, 0)); | |
304 | 234 __ addptr(rsp, 2*wordSize); |
0 | 235 } |
236 | |
237 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter"); | |
238 | |
239 // The stack is not extended by deopt but we must NULL last_sp as this | |
240 // entry is like a "return". | |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
337
diff
changeset
|
241 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); |
0 | 242 __ restore_bcp(); |
243 __ restore_locals(); | |
244 // handle exceptions | |
245 { Label L; | |
246 const Register thread = rcx; | |
247 __ get_thread(thread); | |
304 | 248 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
0 | 249 __ jcc(Assembler::zero, L); |
250 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception)); | |
251 __ should_not_reach_here(); | |
252 __ bind(L); | |
253 } | |
254 __ dispatch_next(state, step); | |
255 return entry; | |
256 } | |
257 | |
258 | |
259 int AbstractInterpreter::BasicType_as_index(BasicType type) { | |
260 int i = 0; | |
261 switch (type) { | |
262 case T_BOOLEAN: i = 0; break; | |
263 case T_CHAR : i = 1; break; | |
264 case T_BYTE : i = 2; break; | |
265 case T_SHORT : i = 3; break; | |
266 case T_INT : // fall through | |
267 case T_LONG : // fall through | |
268 case T_VOID : i = 4; break; | |
269 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE | |
270 case T_DOUBLE : i = 6; break; | |
271 case T_OBJECT : // fall through | |
272 case T_ARRAY : i = 7; break; | |
273 default : ShouldNotReachHere(); | |
274 } | |
275 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds"); | |
276 return i; | |
277 } | |
278 | |
279 | |
280 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) { | |
281 address entry = __ pc(); | |
282 switch (type) { | |
283 case T_BOOLEAN: __ c2bool(rax); break; | |
304 | 284 case T_CHAR : __ andptr(rax, 0xFFFF); break; |
0 | 285 case T_BYTE : __ sign_extend_byte (rax); break; |
286 case T_SHORT : __ sign_extend_short(rax); break; | |
287 case T_INT : /* nothing to do */ break; | |
288 case T_DOUBLE : | |
289 case T_FLOAT : | |
290 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); | |
304 | 291 __ pop(t); // remove return address first |
0 | 292 __ pop_dtos_to_rsp(); |
293 // Must return a result for interpreter or compiler. In SSE | |
294 // mode, results are returned in xmm0 and the FPU stack must | |
295 // be empty. | |
296 if (type == T_FLOAT && UseSSE >= 1) { | |
297 // Load ST0 | |
298 __ fld_d(Address(rsp, 0)); | |
299 // Store as float and empty fpu stack | |
300 __ fstp_s(Address(rsp, 0)); | |
301 // and reload | |
302 __ movflt(xmm0, Address(rsp, 0)); | |
303 } else if (type == T_DOUBLE && UseSSE >= 2 ) { | |
304 __ movdbl(xmm0, Address(rsp, 0)); | |
305 } else { | |
306 // restore ST0 | |
307 __ fld_d(Address(rsp, 0)); | |
308 } | |
309 // and pop the temp | |
304 | 310 __ addptr(rsp, 2 * wordSize); |
311 __ push(t); // restore return address | |
0 | 312 } |
313 break; | |
314 case T_OBJECT : | |
315 // retrieve result from frame | |
304 | 316 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize)); |
0 | 317 // and verify it |
318 __ verify_oop(rax); | |
319 break; | |
320 default : ShouldNotReachHere(); | |
321 } | |
322 __ ret(0); // return from result handler | |
323 return entry; | |
324 } | |
325 | |
326 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) { | |
327 address entry = __ pc(); | |
328 __ push(state); | |
329 __ call_VM(noreg, runtime_entry); | |
330 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); | |
331 return entry; | |
332 } | |
333 | |
334 | |
335 // Helpers for commoning out cases in the various type of method entries. | |
336 // | |
337 | |
338 // increment invocation count & check for overflow | |
339 // | |
340 // Note: checking for negative value instead of overflow | |
341 // so we have a 'sticky' overflow test | |
342 // | |
343 // rbx,: method | |
344 // rcx: invocation counter | |
345 // | |
346 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { | |
347 | |
348 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset()); | |
349 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset()); | |
350 | |
351 if (ProfileInterpreter) { // %%% Merge this into methodDataOop | |
304 | 352 __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset())); |
0 | 353 } |
354 // Update standard invocation counters | |
355 __ movl(rax, backedge_counter); // load backedge counter | |
356 | |
304 | 357 __ incrementl(rcx, InvocationCounter::count_increment); |
0 | 358 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits |
359 | |
360 __ movl(invocation_counter, rcx); // save invocation count | |
361 __ addl(rcx, rax); // add both counters | |
362 | |
363 // profile_method is non-null only for interpreted method so | |
364 // profile_method != NULL == !native_call | |
365 // BytecodeInterpreter only calls for native so code is elided. | |
366 | |
367 if (ProfileInterpreter && profile_method != NULL) { | |
368 // Test to see if we should create a method data oop | |
369 __ cmp32(rcx, | |
370 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit)); | |
371 __ jcc(Assembler::less, *profile_method_continue); | |
372 | |
373 // if no method data exists, go to profile_method | |
374 __ test_method_data_pointer(rax, *profile_method); | |
375 } | |
376 | |
377 __ cmp32(rcx, | |
378 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit)); | |
379 __ jcc(Assembler::aboveEqual, *overflow); | |
380 | |
381 } | |
382 | |
383 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) { | |
384 | |
385 // Asm interpreter on entry | |
386 // rdi - locals | |
387 // rsi - bcp | |
388 // rbx, - method | |
389 // rdx - cpool | |
390 // rbp, - interpreter frame | |
391 | |
392 // C++ interpreter on entry | |
393 // rsi - new interpreter state pointer | |
394 // rbp - interpreter frame pointer | |
395 // rbx - method | |
396 | |
397 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ] | |
398 // rbx, - method | |
399 // rcx - rcvr (assuming there is one) | |
400 // top of stack return address of interpreter caller | |
401 // rsp - sender_sp | |
402 | |
403 // C++ interpreter only | |
404 // rsi - previous interpreter state pointer | |
405 | |
406 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); | |
407 | |
408 // InterpreterRuntime::frequency_counter_overflow takes one argument | |
409 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp). | |
410 // The call returns the address of the verified entry point for the method or NULL | |
411 // if the compilation did not complete (either went background or bailed out). | |
533
dc3ad84615cf
6795913: A few remaining wrong casts need to be fixed for building hotspot successfully on Mac OS.
xlu
parents:
512
diff
changeset
|
412 __ movptr(rax, (intptr_t)false); |
0 | 413 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax); |
414 | |
304 | 415 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop |
0 | 416 |
417 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame | |
418 // and jump to the interpreted entry. | |
419 __ jmp(*do_continue, relocInfo::none); | |
420 | |
421 } | |
422 | |
423 void InterpreterGenerator::generate_stack_overflow_check(void) { | |
424 // see if we've got enough room on the stack for locals plus overhead. | |
425 // the expression stack grows down incrementally, so the normal guard | |
426 // page mechanism will work for that. | |
427 // | |
428 // Registers live on entry: | |
429 // | |
430 // Asm interpreter | |
431 // rdx: number of additional locals this frame needs (what we must check) | |
432 // rbx,: methodOop | |
433 | |
434 // destroyed on exit | |
435 // rax, | |
436 | |
437 // NOTE: since the additional locals are also always pushed (wasn't obvious in | |
438 // generate_method_entry) so the guard should work for them too. | |
439 // | |
440 | |
441 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp | |
442 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; | |
443 | |
444 // total overhead size: entry_size + (saved rbp, thru expr stack bottom). | |
445 // be sure to change this if you add/subtract anything to/from the overhead area | |
446 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size; | |
447 | |
448 const int page_size = os::vm_page_size(); | |
449 | |
450 Label after_frame_check; | |
451 | |
452 // see if the frame is greater than one page in size. If so, | |
453 // then we need to verify there is enough stack space remaining | |
454 // for the additional locals. | |
455 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize()); | |
456 __ jcc(Assembler::belowEqual, after_frame_check); | |
457 | |
458 // compute rsp as if this were going to be the last frame on | |
459 // the stack before the red zone | |
460 | |
461 Label after_frame_check_pop; | |
462 | |
304 | 463 __ push(rsi); |
0 | 464 |
465 const Register thread = rsi; | |
466 | |
467 __ get_thread(thread); | |
468 | |
469 const Address stack_base(thread, Thread::stack_base_offset()); | |
470 const Address stack_size(thread, Thread::stack_size_offset()); | |
471 | |
472 // locals + overhead, in bytes | |
304 | 473 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size)); |
0 | 474 |
475 #ifdef ASSERT | |
476 Label stack_base_okay, stack_size_okay; | |
477 // verify that thread stack base is non-zero | |
304 | 478 __ cmpptr(stack_base, (int32_t)NULL_WORD); |
0 | 479 __ jcc(Assembler::notEqual, stack_base_okay); |
480 __ stop("stack base is zero"); | |
481 __ bind(stack_base_okay); | |
482 // verify that thread stack size is non-zero | |
304 | 483 __ cmpptr(stack_size, 0); |
0 | 484 __ jcc(Assembler::notEqual, stack_size_okay); |
485 __ stop("stack size is zero"); | |
486 __ bind(stack_size_okay); | |
487 #endif | |
488 | |
489 // Add stack base to locals and subtract stack size | |
304 | 490 __ addptr(rax, stack_base); |
491 __ subptr(rax, stack_size); | |
0 | 492 |
493 // Use the maximum number of pages we might bang. | |
494 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages : | |
495 (StackRedPages+StackYellowPages); | |
304 | 496 __ addptr(rax, max_pages * page_size); |
0 | 497 |
498 // check against the current stack bottom | |
304 | 499 __ cmpptr(rsp, rax); |
0 | 500 __ jcc(Assembler::above, after_frame_check_pop); |
501 | |
304 | 502 __ pop(rsi); // get saved bcp / (c++ prev state ). |
0 | 503 |
304 | 504 __ pop(rax); // get return address |
0 | 505 __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry())); |
506 | |
507 // all done with frame size check | |
508 __ bind(after_frame_check_pop); | |
304 | 509 __ pop(rsi); |
0 | 510 |
511 __ bind(after_frame_check); | |
512 } | |
513 | |
514 // Allocate monitor and lock method (asm interpreter) | |
515 // rbx, - methodOop | |
516 // | |
517 void InterpreterGenerator::lock_method(void) { | |
518 // synchronize method | |
519 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); | |
520 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); | |
521 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; | |
522 | |
523 #ifdef ASSERT | |
524 { Label L; | |
525 __ movl(rax, access_flags); | |
526 __ testl(rax, JVM_ACC_SYNCHRONIZED); | |
527 __ jcc(Assembler::notZero, L); | |
528 __ stop("method doesn't need synchronization"); | |
529 __ bind(L); | |
530 } | |
531 #endif // ASSERT | |
532 // get synchronization object | |
533 { Label done; | |
534 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes(); | |
535 __ movl(rax, access_flags); | |
536 __ testl(rax, JVM_ACC_STATIC); | |
304 | 537 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case) |
0 | 538 __ jcc(Assembler::zero, done); |
304 | 539 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset())); |
540 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes())); | |
541 __ movptr(rax, Address(rax, mirror_offset)); | |
0 | 542 __ bind(done); |
543 } | |
544 // add space for monitor & lock | |
304 | 545 __ subptr(rsp, entry_size); // add space for a monitor entry |
546 __ movptr(monitor_block_top, rsp); // set new monitor block top | |
547 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object | |
548 __ mov(rdx, rsp); // object address | |
0 | 549 __ lock_object(rdx); |
550 } | |
551 | |
552 // | |
553 // Generate a fixed interpreter frame. This is identical setup for interpreted methods | |
554 // and for native methods hence the shared code. | |
555 | |
556 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { | |
557 // initialize fixed part of activation frame | |
304 | 558 __ push(rax); // save return address |
0 | 559 __ enter(); // save old & set new rbp, |
560 | |
561 | |
304 | 562 __ push(rsi); // set sender sp |
563 __ push((int32_t)NULL_WORD); // leave last_sp as null | |
564 __ movptr(rsi, Address(rbx,methodOopDesc::const_offset())); // get constMethodOop | |
565 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase | |
566 __ push(rbx); // save methodOop | |
0 | 567 if (ProfileInterpreter) { |
568 Label method_data_continue; | |
304 | 569 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset()))); |
570 __ testptr(rdx, rdx); | |
0 | 571 __ jcc(Assembler::zero, method_data_continue); |
304 | 572 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset())); |
0 | 573 __ bind(method_data_continue); |
304 | 574 __ push(rdx); // set the mdp (method data pointer) |
0 | 575 } else { |
304 | 576 __ push(0); |
0 | 577 } |
578 | |
304 | 579 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset())); |
580 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes())); | |
581 __ push(rdx); // set constant pool cache | |
582 __ push(rdi); // set locals pointer | |
0 | 583 if (native_call) { |
304 | 584 __ push(0); // no bcp |
0 | 585 } else { |
304 | 586 __ push(rsi); // set bcp |
0 | 587 } |
304 | 588 __ push(0); // reserve word for pointer to expression stack bottom |
589 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom | |
0 | 590 } |
591 | |
592 // End of helpers | |
593 | |
594 // | |
595 // Various method entries | |
596 //------------------------------------------------------------------------------------------------------------------------ | |
597 // | |
598 // | |
599 | |
600 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry | |
601 | |
602 address InterpreterGenerator::generate_accessor_entry(void) { | |
603 | |
604 // rbx,: methodOop | |
605 // rcx: receiver (preserve for slow entry into asm interpreter) | |
606 | |
607 // rsi: senderSP must preserved for slow path, set SP to it on fast path | |
608 | |
609 address entry_point = __ pc(); | |
610 Label xreturn_path; | |
611 | |
612 // do fastpath for resolved accessor methods | |
613 if (UseFastAccessorMethods) { | |
614 Label slow_path; | |
615 // If we need a safepoint check, generate full interpreter entry. | |
616 ExternalAddress state(SafepointSynchronize::address_of_state()); | |
617 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), | |
618 SafepointSynchronize::_not_synchronized); | |
619 | |
620 __ jcc(Assembler::notEqual, slow_path); | |
621 // ASM/C++ Interpreter | |
622 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1 | |
623 // Note: We can only use this code if the getfield has been resolved | |
624 // and if we don't have a null-pointer exception => check for | |
625 // these conditions first and use slow path if necessary. | |
626 // rbx,: method | |
627 // rcx: receiver | |
304 | 628 __ movptr(rax, Address(rsp, wordSize)); |
0 | 629 |
630 // check if local 0 != NULL and read field | |
304 | 631 __ testptr(rax, rax); |
0 | 632 __ jcc(Assembler::zero, slow_path); |
633 | |
304 | 634 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset())); |
0 | 635 // read first instruction word and extract bytecode @ 1 and index @ 2 |
304 | 636 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset())); |
0 | 637 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset())); |
638 // Shift codes right to get the index on the right. | |
639 // The bytecode fetched looks like <index><0xb4><0x2a> | |
640 __ shrl(rdx, 2*BitsPerByte); | |
641 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); | |
304 | 642 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes())); |
0 | 643 |
644 // rax,: local 0 | |
645 // rbx,: method | |
646 // rcx: receiver - do not destroy since it is needed for slow path! | |
647 // rcx: scratch | |
648 // rdx: constant pool cache index | |
649 // rdi: constant pool cache | |
650 // rsi: sender sp | |
651 | |
652 // check if getfield has been resolved and read constant pool cache entry | |
653 // check the validity of the cache entry by testing whether _indices field | |
654 // contains Bytecode::_getfield in b1 byte. | |
655 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below"); | |
656 __ movl(rcx, | |
657 Address(rdi, | |
658 rdx, | |
304 | 659 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset())); |
0 | 660 __ shrl(rcx, 2*BitsPerByte); |
661 __ andl(rcx, 0xFF); | |
662 __ cmpl(rcx, Bytecodes::_getfield); | |
663 __ jcc(Assembler::notEqual, slow_path); | |
664 | |
665 // Note: constant pool entry is not valid before bytecode is resolved | |
304 | 666 __ movptr(rcx, |
667 Address(rdi, | |
668 rdx, | |
669 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())); | |
0 | 670 __ movl(rdx, |
671 Address(rdi, | |
672 rdx, | |
304 | 673 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset())); |
0 | 674 |
675 Label notByte, notShort, notChar; | |
676 const Address field_address (rax, rcx, Address::times_1); | |
677 | |
678 // Need to differentiate between igetfield, agetfield, bgetfield etc. | |
679 // because they are different sizes. | |
680 // Use the type from the constant pool cache | |
681 __ shrl(rdx, ConstantPoolCacheEntry::tosBits); | |
682 // Make sure we don't need to mask rdx for tosBits after the above shift | |
683 ConstantPoolCacheEntry::verify_tosBits(); | |
684 __ cmpl(rdx, btos); | |
685 __ jcc(Assembler::notEqual, notByte); | |
686 __ load_signed_byte(rax, field_address); | |
687 __ jmp(xreturn_path); | |
688 | |
689 __ bind(notByte); | |
690 __ cmpl(rdx, stos); | |
691 __ jcc(Assembler::notEqual, notShort); | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
533
diff
changeset
|
692 __ load_signed_short(rax, field_address); |
0 | 693 __ jmp(xreturn_path); |
694 | |
695 __ bind(notShort); | |
696 __ cmpl(rdx, ctos); | |
697 __ jcc(Assembler::notEqual, notChar); | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
533
diff
changeset
|
698 __ load_unsigned_short(rax, field_address); |
0 | 699 __ jmp(xreturn_path); |
700 | |
701 __ bind(notChar); | |
702 #ifdef ASSERT | |
703 Label okay; | |
704 __ cmpl(rdx, atos); | |
705 __ jcc(Assembler::equal, okay); | |
706 __ cmpl(rdx, itos); | |
707 __ jcc(Assembler::equal, okay); | |
708 __ stop("what type is this?"); | |
709 __ bind(okay); | |
710 #endif // ASSERT | |
711 // All the rest are a 32 bit wordsize | |
304 | 712 // This is ok for now. Since fast accessors should be going away |
713 __ movptr(rax, field_address); | |
0 | 714 |
715 __ bind(xreturn_path); | |
716 | |
717 // _ireturn/_areturn | |
304 | 718 __ pop(rdi); // get return address |
719 __ mov(rsp, rsi); // set sp to sender sp | |
0 | 720 __ jmp(rdi); |
721 | |
722 // generate a vanilla interpreter entry as the slow path | |
723 __ bind(slow_path); | |
724 | |
725 (void) generate_normal_entry(false); | |
726 return entry_point; | |
727 } | |
728 return NULL; | |
729 | |
730 } | |
731 | |
732 // | |
733 // Interpreter stub for calling a native method. (asm interpreter) | |
734 // This sets up a somewhat different looking stack for calling the native method | |
735 // than the typical interpreter frame setup. | |
736 // | |
737 | |
738 address InterpreterGenerator::generate_native_entry(bool synchronized) { | |
739 // determine code generation flags | |
740 bool inc_counter = UseCompiler || CountCompiledCalls; | |
741 | |
742 // rbx,: methodOop | |
743 // rsi: sender sp | |
744 // rsi: previous interpreter state (C++ interpreter) must preserve | |
745 address entry_point = __ pc(); | |
746 | |
747 | |
748 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); | |
749 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset()); | |
750 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); | |
751 | |
752 // get parameter size (always needed) | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
533
diff
changeset
|
753 __ load_unsigned_short(rcx, size_of_parameters); |
0 | 754 |
755 // native calls don't need the stack size check since they have no expression stack | |
756 // and the arguments are already on the stack and we only add a handful of words | |
757 // to the stack | |
758 | |
759 // rbx,: methodOop | |
760 // rcx: size of parameters | |
761 // rsi: sender sp | |
762 | |
304 | 763 __ pop(rax); // get return address |
0 | 764 // for natives the size of locals is zero |
765 | |
766 // compute beginning of parameters (rdi) | |
304 | 767 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize)); |
0 | 768 |
769 | |
770 // add 2 zero-initialized slots for native calls | |
771 // NULL result handler | |
304 | 772 __ push((int32_t)NULL_WORD); |
0 | 773 // NULL oop temp (mirror or jni oop result) |
304 | 774 __ push((int32_t)NULL_WORD); |
0 | 775 |
776 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count | |
777 // initialize fixed part of activation frame | |
778 | |
779 generate_fixed_frame(true); | |
780 | |
781 // make sure method is native & not abstract | |
782 #ifdef ASSERT | |
783 __ movl(rax, access_flags); | |
784 { | |
785 Label L; | |
786 __ testl(rax, JVM_ACC_NATIVE); | |
787 __ jcc(Assembler::notZero, L); | |
788 __ stop("tried to execute non-native method as native"); | |
789 __ bind(L); | |
790 } | |
791 { Label L; | |
792 __ testl(rax, JVM_ACC_ABSTRACT); | |
793 __ jcc(Assembler::zero, L); | |
794 __ stop("tried to execute abstract method in interpreter"); | |
795 __ bind(L); | |
796 } | |
797 #endif | |
798 | |
799 // Since at this point in the method invocation the exception handler | |
800 // would try to exit the monitor of synchronized methods which hasn't | |
801 // been entered yet, we set the thread local variable | |
802 // _do_not_unlock_if_synchronized to true. The remove_activation will | |
803 // check this flag. | |
804 | |
805 __ get_thread(rax); | |
806 const Address do_not_unlock_if_synchronized(rax, | |
807 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); | |
808 __ movbool(do_not_unlock_if_synchronized, true); | |
809 | |
810 // increment invocation count & check for overflow | |
811 Label invocation_counter_overflow; | |
812 if (inc_counter) { | |
813 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); | |
814 } | |
815 | |
816 Label continue_after_compile; | |
817 __ bind(continue_after_compile); | |
818 | |
819 bang_stack_shadow_pages(true); | |
820 | |
821 // reset the _do_not_unlock_if_synchronized flag | |
822 __ get_thread(rax); | |
823 __ movbool(do_not_unlock_if_synchronized, false); | |
824 | |
825 // check for synchronized methods | |
826 // Must happen AFTER invocation_counter check and stack overflow check, | |
827 // so method is not locked if overflows. | |
828 // | |
829 if (synchronized) { | |
830 lock_method(); | |
831 } else { | |
832 // no synchronization necessary | |
833 #ifdef ASSERT | |
834 { Label L; | |
835 __ movl(rax, access_flags); | |
836 __ testl(rax, JVM_ACC_SYNCHRONIZED); | |
837 __ jcc(Assembler::zero, L); | |
838 __ stop("method needs synchronization"); | |
839 __ bind(L); | |
840 } | |
841 #endif | |
842 } | |
843 | |
844 // start execution | |
845 #ifdef ASSERT | |
846 { Label L; | |
847 const Address monitor_block_top (rbp, | |
848 frame::interpreter_frame_monitor_block_top_offset * wordSize); | |
304 | 849 __ movptr(rax, monitor_block_top); |
850 __ cmpptr(rax, rsp); | |
0 | 851 __ jcc(Assembler::equal, L); |
852 __ stop("broken stack frame setup in interpreter"); | |
853 __ bind(L); | |
854 } | |
855 #endif | |
856 | |
857 // jvmti/dtrace support | |
858 __ notify_method_entry(); | |
859 | |
860 // work registers | |
861 const Register method = rbx; | |
862 const Register thread = rdi; | |
863 const Register t = rcx; | |
864 | |
865 // allocate space for parameters | |
866 __ get_method(method); | |
867 __ verify_oop(method); | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
533
diff
changeset
|
868 __ load_unsigned_short(t, Address(method, methodOopDesc::size_of_parameters_offset())); |
304 | 869 __ shlptr(t, Interpreter::logStackElementSize()); |
870 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror | |
871 __ subptr(rsp, t); | |
872 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics | |
0 | 873 |
874 // get signature handler | |
875 { Label L; | |
304 | 876 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); |
877 __ testptr(t, t); | |
0 | 878 __ jcc(Assembler::notZero, L); |
879 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method); | |
880 __ get_method(method); | |
304 | 881 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset())); |
0 | 882 __ bind(L); |
883 } | |
884 | |
885 // call signature handler | |
886 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code"); | |
887 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code"); | |
888 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code"); | |
889 // The generated handlers do not touch RBX (the method oop). | |
890 // However, large signatures cannot be cached and are generated | |
891 // each time here. The slow-path generator will blow RBX | |
892 // sometime, so we must reload it after the call. | |
893 __ call(t); | |
894 __ get_method(method); // slow path call blows RBX on DevStudio 5.0 | |
895 | |
896 // result handler is in rax, | |
897 // set result handler | |
304 | 898 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax); |
0 | 899 |
900 // pass mirror handle if static call | |
901 { Label L; | |
902 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes(); | |
903 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); | |
904 __ testl(t, JVM_ACC_STATIC); | |
905 __ jcc(Assembler::zero, L); | |
906 // get mirror | |
304 | 907 __ movptr(t, Address(method, methodOopDesc:: constants_offset())); |
908 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes())); | |
909 __ movptr(t, Address(t, mirror_offset)); | |
0 | 910 // copy mirror into activation frame |
304 | 911 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t); |
0 | 912 // pass handle to mirror |
304 | 913 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize)); |
914 __ movptr(Address(rsp, wordSize), t); | |
0 | 915 __ bind(L); |
916 } | |
917 | |
918 // get native function entry point | |
919 { Label L; | |
304 | 920 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); |
0 | 921 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry()); |
304 | 922 __ cmpptr(rax, unsatisfied.addr()); |
0 | 923 __ jcc(Assembler::notEqual, L); |
924 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method); | |
925 __ get_method(method); | |
926 __ verify_oop(method); | |
304 | 927 __ movptr(rax, Address(method, methodOopDesc::native_function_offset())); |
0 | 928 __ bind(L); |
929 } | |
930 | |
931 // pass JNIEnv | |
932 __ get_thread(thread); | |
304 | 933 __ lea(t, Address(thread, JavaThread::jni_environment_offset())); |
934 __ movptr(Address(rsp, 0), t); | |
0 | 935 |
936 // set_last_Java_frame_before_call | |
937 // It is enough that the pc() | |
938 // points into the right code segment. It does not have to be the correct return pc. | |
939 __ set_last_Java_frame(thread, noreg, rbp, __ pc()); | |
940 | |
941 // change thread state | |
942 #ifdef ASSERT | |
943 { Label L; | |
944 __ movl(t, Address(thread, JavaThread::thread_state_offset())); | |
945 __ cmpl(t, _thread_in_Java); | |
946 __ jcc(Assembler::equal, L); | |
947 __ stop("Wrong thread state in native stub"); | |
948 __ bind(L); | |
949 } | |
950 #endif | |
951 | |
952 // Change state to native | |
953 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native); | |
954 __ call(rax); | |
955 | |
956 // result potentially in rdx:rax or ST0 | |
957 | |
958 // Either restore the MXCSR register after returning from the JNI Call | |
959 // or verify that it wasn't changed. | |
960 if (VM_Version::supports_sse()) { | |
961 if (RestoreMXCSROnJNICalls) { | |
962 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std())); | |
963 } | |
964 else if (CheckJNICalls ) { | |
304 | 965 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry())); |
0 | 966 } |
967 } | |
968 | |
969 // Either restore the x87 floating pointer control word after returning | |
970 // from the JNI call or verify that it wasn't changed. | |
971 if (CheckJNICalls) { | |
304 | 972 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry())); |
0 | 973 } |
974 | |
975 // save potential result in ST(0) & rdx:rax | |
976 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 - | |
977 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers) | |
978 // It is safe to do this push because state is _thread_in_native and return address will be found | |
979 // via _last_native_pc and not via _last_jave_sp | |
980 | |
981 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result. | |
982 // If the order changes or anything else is added to the stack the code in | |
983 // interpreter_frame_result will have to be changed. | |
984 | |
985 { Label L; | |
986 Label push_double; | |
987 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT)); | |
988 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE)); | |
989 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize), | |
990 float_handler.addr()); | |
991 __ jcc(Assembler::equal, push_double); | |
992 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize), | |
993 double_handler.addr()); | |
994 __ jcc(Assembler::notEqual, L); | |
995 __ bind(push_double); | |
996 __ push(dtos); | |
997 __ bind(L); | |
998 } | |
999 __ push(ltos); | |
1000 | |
1001 // change thread state | |
1002 __ get_thread(thread); | |
1003 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans); | |
1004 if(os::is_MP()) { | |
1005 if (UseMembar) { | |
304 | 1006 // Force this write out before the read below |
1007 __ membar(Assembler::Membar_mask_bits( | |
1008 Assembler::LoadLoad | Assembler::LoadStore | | |
1009 Assembler::StoreLoad | Assembler::StoreStore)); | |
0 | 1010 } else { |
1011 // Write serialization page so VM thread can do a pseudo remote membar. | |
1012 // We use the current thread pointer to calculate a thread specific | |
1013 // offset to write to within the page. This minimizes bus traffic | |
1014 // due to cache line collision. | |
1015 __ serialize_memory(thread, rcx); | |
1016 } | |
1017 } | |
1018 | |
1019 if (AlwaysRestoreFPU) { | |
1020 // Make sure the control word is correct. | |
1021 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); | |
1022 } | |
1023 | |
1024 // check for safepoint operation in progress and/or pending suspend requests | |
1025 { Label Continue; | |
1026 | |
1027 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), | |
1028 SafepointSynchronize::_not_synchronized); | |
1029 | |
1030 Label L; | |
1031 __ jcc(Assembler::notEqual, L); | |
1032 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0); | |
1033 __ jcc(Assembler::equal, Continue); | |
1034 __ bind(L); | |
1035 | |
1036 // Don't use call_VM as it will see a possible pending exception and forward it | |
1037 // and never return here preventing us from clearing _last_native_pc down below. | |
1038 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are | |
1039 // preserved and correspond to the bcp/locals pointers. So we do a runtime call | |
1040 // by hand. | |
1041 // | |
304 | 1042 __ push(thread); |
0 | 1043 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, |
1044 JavaThread::check_special_condition_for_native_trans))); | |
1045 __ increment(rsp, wordSize); | |
1046 __ get_thread(thread); | |
1047 | |
1048 __ bind(Continue); | |
1049 } | |
1050 | |
1051 // change thread state | |
1052 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java); | |
1053 | |
1054 __ reset_last_Java_frame(thread, true, true); | |
1055 | |
1056 // reset handle block | |
304 | 1057 __ movptr(t, Address(thread, JavaThread::active_handles_offset())); |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
337
diff
changeset
|
1058 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD); |
0 | 1059 |
1060 // If result was an oop then unbox and save it in the frame | |
1061 { Label L; | |
1062 Label no_oop, store_result; | |
1063 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT)); | |
1064 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), | |
1065 handler.addr()); | |
1066 __ jcc(Assembler::notEqual, no_oop); | |
304 | 1067 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD); |
0 | 1068 __ pop(ltos); |
304 | 1069 __ testptr(rax, rax); |
0 | 1070 __ jcc(Assembler::zero, store_result); |
1071 // unbox | |
304 | 1072 __ movptr(rax, Address(rax, 0)); |
0 | 1073 __ bind(store_result); |
304 | 1074 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax); |
0 | 1075 // keep stack depth as expected by pushing oop which will eventually be discarded |
1076 __ push(ltos); | |
1077 __ bind(no_oop); | |
1078 } | |
1079 | |
1080 { | |
1081 Label no_reguard; | |
1082 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled); | |
1083 __ jcc(Assembler::notEqual, no_reguard); | |
1084 | |
304 | 1085 __ pusha(); |
0 | 1086 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages))); |
304 | 1087 __ popa(); |
0 | 1088 |
1089 __ bind(no_reguard); | |
1090 } | |
1091 | |
1092 // restore rsi to have legal interpreter frame, | |
1093 // i.e., bci == 0 <=> rsi == code_base() | |
1094 // Can't call_VM until bcp is within reasonable. | |
1095 __ get_method(method); // method is junk from thread_in_native to now. | |
1096 __ verify_oop(method); | |
304 | 1097 __ movptr(rsi, Address(method,methodOopDesc::const_offset())); // get constMethodOop |
1098 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase | |
0 | 1099 |
1100 // handle exceptions (exception handling will handle unlocking!) | |
1101 { Label L; | |
304 | 1102 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
0 | 1103 __ jcc(Assembler::zero, L); |
1104 // Note: At some point we may want to unify this with the code used in call_VM_base(); | |
1105 // i.e., we should use the StubRoutines::forward_exception code. For now this | |
1106 // doesn't work here because the rsp is not correctly set at this point. | |
1107 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception)); | |
1108 __ should_not_reach_here(); | |
1109 __ bind(L); | |
1110 } | |
1111 | |
1112 // do unlocking if necessary | |
1113 { Label L; | |
1114 __ movl(t, Address(method, methodOopDesc::access_flags_offset())); | |
1115 __ testl(t, JVM_ACC_SYNCHRONIZED); | |
1116 __ jcc(Assembler::zero, L); | |
1117 // the code below should be shared with interpreter macro assembler implementation | |
1118 { Label unlock; | |
1119 // BasicObjectLock will be first in list, since this is a synchronized method. However, need | |
1120 // to check that the object has not been unlocked by an explicit monitorexit bytecode. | |
1121 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock)); | |
1122 | |
304 | 1123 __ lea(rdx, monitor); // address of first monitor |
0 | 1124 |
304 | 1125 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); |
1126 __ testptr(t, t); | |
0 | 1127 __ jcc(Assembler::notZero, unlock); |
1128 | |
1129 // Entry already unlocked, need to throw exception | |
1130 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); | |
1131 __ should_not_reach_here(); | |
1132 | |
1133 __ bind(unlock); | |
1134 __ unlock_object(rdx); | |
1135 } | |
1136 __ bind(L); | |
1137 } | |
1138 | |
1139 // jvmti/dtrace support | |
1140 // Note: This must happen _after_ handling/throwing any exceptions since | |
1141 // the exception handler code notifies the runtime of method exits | |
1142 // too. If this happens before, method entry/exit notifications are | |
1143 // not properly paired (was bug - gri 11/22/99). | |
1144 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI); | |
1145 | |
1146 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result | |
1147 __ pop(ltos); | |
304 | 1148 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize)); |
0 | 1149 __ call(t); |
1150 | |
1151 // remove activation | |
304 | 1152 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp |
0 | 1153 __ leave(); // remove frame anchor |
304 | 1154 __ pop(rdi); // get return address |
1155 __ mov(rsp, t); // set sp to sender sp | |
0 | 1156 __ jmp(rdi); |
1157 | |
1158 if (inc_counter) { | |
1159 // Handle overflow of counter and compile method | |
1160 __ bind(invocation_counter_overflow); | |
1161 generate_counter_overflow(&continue_after_compile); | |
1162 } | |
1163 | |
1164 return entry_point; | |
1165 } | |
1166 | |
1167 // | |
1168 // Generic interpreted method entry to (asm) interpreter | |
1169 // | |
1170 address InterpreterGenerator::generate_normal_entry(bool synchronized) { | |
1171 // determine code generation flags | |
1172 bool inc_counter = UseCompiler || CountCompiledCalls; | |
1173 | |
1174 // rbx,: methodOop | |
1175 // rsi: sender sp | |
1176 address entry_point = __ pc(); | |
1177 | |
1178 | |
1179 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset()); | |
1180 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset()); | |
1181 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset()); | |
1182 const Address access_flags (rbx, methodOopDesc::access_flags_offset()); | |
1183 | |
1184 // get parameter size (always needed) | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
533
diff
changeset
|
1185 __ load_unsigned_short(rcx, size_of_parameters); |
0 | 1186 |
1187 // rbx,: methodOop | |
1188 // rcx: size of parameters | |
1189 | |
1190 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i ) | |
1191 | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
533
diff
changeset
|
1192 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words |
0 | 1193 __ subl(rdx, rcx); // rdx = no. of additional locals |
1194 | |
1195 // see if we've got enough room on the stack for locals plus overhead. | |
1196 generate_stack_overflow_check(); | |
1197 | |
1198 // get return address | |
304 | 1199 __ pop(rax); |
0 | 1200 |
1201 // compute beginning of parameters (rdi) | |
304 | 1202 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize)); |
0 | 1203 |
1204 // rdx - # of additional locals | |
1205 // allocate space for locals | |
1206 // explicitly initialize locals | |
1207 { | |
1208 Label exit, loop; | |
1209 __ testl(rdx, rdx); | |
1210 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0 | |
1211 __ bind(loop); | |
304 | 1212 if (TaggedStackInterpreter) { |
1213 __ push((int32_t)NULL_WORD); // push tag | |
1214 } | |
1215 __ push((int32_t)NULL_WORD); // initialize local variables | |
0 | 1216 __ decrement(rdx); // until everything initialized |
1217 __ jcc(Assembler::greater, loop); | |
1218 __ bind(exit); | |
1219 } | |
1220 | |
1221 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count | |
1222 // initialize fixed part of activation frame | |
1223 generate_fixed_frame(false); | |
1224 | |
1225 // make sure method is not native & not abstract | |
1226 #ifdef ASSERT | |
1227 __ movl(rax, access_flags); | |
1228 { | |
1229 Label L; | |
1230 __ testl(rax, JVM_ACC_NATIVE); | |
1231 __ jcc(Assembler::zero, L); | |
1232 __ stop("tried to execute native method as non-native"); | |
1233 __ bind(L); | |
1234 } | |
1235 { Label L; | |
1236 __ testl(rax, JVM_ACC_ABSTRACT); | |
1237 __ jcc(Assembler::zero, L); | |
1238 __ stop("tried to execute abstract method in interpreter"); | |
1239 __ bind(L); | |
1240 } | |
1241 #endif | |
1242 | |
1243 // Since at this point in the method invocation the exception handler | |
1244 // would try to exit the monitor of synchronized methods which hasn't | |
1245 // been entered yet, we set the thread local variable | |
1246 // _do_not_unlock_if_synchronized to true. The remove_activation will | |
1247 // check this flag. | |
1248 | |
1249 __ get_thread(rax); | |
1250 const Address do_not_unlock_if_synchronized(rax, | |
1251 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); | |
1252 __ movbool(do_not_unlock_if_synchronized, true); | |
1253 | |
1254 // increment invocation count & check for overflow | |
1255 Label invocation_counter_overflow; | |
1256 Label profile_method; | |
1257 Label profile_method_continue; | |
1258 if (inc_counter) { | |
1259 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue); | |
1260 if (ProfileInterpreter) { | |
1261 __ bind(profile_method_continue); | |
1262 } | |
1263 } | |
1264 Label continue_after_compile; | |
1265 __ bind(continue_after_compile); | |
1266 | |
1267 bang_stack_shadow_pages(false); | |
1268 | |
1269 // reset the _do_not_unlock_if_synchronized flag | |
1270 __ get_thread(rax); | |
1271 __ movbool(do_not_unlock_if_synchronized, false); | |
1272 | |
1273 // check for synchronized methods | |
1274 // Must happen AFTER invocation_counter check and stack overflow check, | |
1275 // so method is not locked if overflows. | |
1276 // | |
1277 if (synchronized) { | |
1278 // Allocate monitor and lock method | |
1279 lock_method(); | |
1280 } else { | |
1281 // no synchronization necessary | |
1282 #ifdef ASSERT | |
1283 { Label L; | |
1284 __ movl(rax, access_flags); | |
1285 __ testl(rax, JVM_ACC_SYNCHRONIZED); | |
1286 __ jcc(Assembler::zero, L); | |
1287 __ stop("method needs synchronization"); | |
1288 __ bind(L); | |
1289 } | |
1290 #endif | |
1291 } | |
1292 | |
1293 // start execution | |
1294 #ifdef ASSERT | |
1295 { Label L; | |
1296 const Address monitor_block_top (rbp, | |
1297 frame::interpreter_frame_monitor_block_top_offset * wordSize); | |
304 | 1298 __ movptr(rax, monitor_block_top); |
1299 __ cmpptr(rax, rsp); | |
0 | 1300 __ jcc(Assembler::equal, L); |
1301 __ stop("broken stack frame setup in interpreter"); | |
1302 __ bind(L); | |
1303 } | |
1304 #endif | |
1305 | |
1306 // jvmti support | |
1307 __ notify_method_entry(); | |
1308 | |
1309 __ dispatch_next(vtos); | |
1310 | |
1311 // invocation counter overflow | |
1312 if (inc_counter) { | |
1313 if (ProfileInterpreter) { | |
1314 // We have decided to profile this method in the interpreter | |
1315 __ bind(profile_method); | |
1316 | |
1317 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi, true); | |
1318 | |
304 | 1319 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop |
1320 __ movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset()))); | |
1321 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax); | |
0 | 1322 __ test_method_data_pointer(rax, profile_method_continue); |
304 | 1323 __ addptr(rax, in_bytes(methodDataOopDesc::data_offset())); |
1324 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax); | |
0 | 1325 __ jmp(profile_method_continue); |
1326 } | |
1327 // Handle overflow of counter and compile method | |
1328 __ bind(invocation_counter_overflow); | |
1329 generate_counter_overflow(&continue_after_compile); | |
1330 } | |
1331 | |
1332 return entry_point; | |
1333 } | |
1334 | |
1335 //------------------------------------------------------------------------------------------------------------------------ | |
1336 // Entry points | |
1337 // | |
1338 // Here we generate the various kind of entries into the interpreter. | |
1339 // The two main entry type are generic bytecode methods and native call method. | |
1340 // These both come in synchronized and non-synchronized versions but the | |
1341 // frame layout they create is very similar. The other method entry | |
1342 // types are really just special purpose entries that are really entry | |
1343 // and interpretation all in one. These are for trivial methods like | |
1344 // accessor, empty, or special math methods. | |
1345 // | |
1346 // When control flow reaches any of the entry types for the interpreter | |
1347 // the following holds -> | |
1348 // | |
1349 // Arguments: | |
1350 // | |
1351 // rbx,: methodOop | |
1352 // rcx: receiver | |
1353 // | |
1354 // | |
1355 // Stack layout immediately at entry | |
1356 // | |
1357 // [ return address ] <--- rsp | |
1358 // [ parameter n ] | |
1359 // ... | |
1360 // [ parameter 1 ] | |
1361 // [ expression stack ] (caller's java expression stack) | |
1362 | |
1363 // Assuming that we don't go to one of the trivial specialized | |
1364 // entries the stack will look like below when we are ready to execute | |
1365 // the first bytecode (or call the native routine). The register usage | |
1366 // will be as the template based interpreter expects (see interpreter_x86.hpp). | |
1367 // | |
1368 // local variables follow incoming parameters immediately; i.e. | |
1369 // the return address is moved to the end of the locals). | |
1370 // | |
1371 // [ monitor entry ] <--- rsp | |
1372 // ... | |
1373 // [ monitor entry ] | |
1374 // [ expr. stack bottom ] | |
1375 // [ saved rsi ] | |
1376 // [ current rdi ] | |
1377 // [ methodOop ] | |
1378 // [ saved rbp, ] <--- rbp, | |
1379 // [ return address ] | |
1380 // [ local variable m ] | |
1381 // ... | |
1382 // [ local variable 1 ] | |
1383 // [ parameter n ] | |
1384 // ... | |
1385 // [ parameter 1 ] <--- rdi | |
1386 | |
1387 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) { | |
1388 // determine code generation flags | |
1389 bool synchronized = false; | |
1390 address entry_point = NULL; | |
1391 | |
1392 switch (kind) { | |
1393 case Interpreter::zerolocals : break; | |
1394 case Interpreter::zerolocals_synchronized: synchronized = true; break; | |
1395 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break; | |
1396 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break; | |
1397 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break; | |
1398 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break; | |
1399 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break; | |
710 | 1400 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break; |
0 | 1401 |
1402 case Interpreter::java_lang_math_sin : // fall thru | |
1403 case Interpreter::java_lang_math_cos : // fall thru | |
1404 case Interpreter::java_lang_math_tan : // fall thru | |
1405 case Interpreter::java_lang_math_abs : // fall thru | |
1406 case Interpreter::java_lang_math_log : // fall thru | |
1407 case Interpreter::java_lang_math_log10 : // fall thru | |
1408 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break; | |
1409 default : ShouldNotReachHere(); break; | |
1410 } | |
1411 | |
1412 if (entry_point) return entry_point; | |
1413 | |
1414 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized); | |
1415 | |
1416 } | |
1417 | |
1418 // How much stack a method activation needs in words. | |
1419 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) { | |
1420 | |
1421 const int stub_code = 4; // see generate_call_stub | |
1422 // Save space for one monitor to get into the interpreted method in case | |
1423 // the method is synchronized | |
1424 int monitor_size = method->is_synchronized() ? | |
1425 1*frame::interpreter_frame_monitor_size() : 0; | |
1426 | |
1427 // total overhead size: entry_size + (saved rbp, thru expr stack bottom). | |
1428 // be sure to change this if you add/subtract anything to/from the overhead area | |
1429 const int overhead_size = -frame::interpreter_frame_initial_sp_offset; | |
1430 | |
710 | 1431 const int extra_stack = methodOopDesc::extra_stack_entries(); |
1432 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) * | |
0 | 1433 Interpreter::stackElementWords(); |
1434 return overhead_size + method_stack + stub_code; | |
1435 } | |
1436 | |
1437 // asm based interpreter deoptimization helpers | |
1438 | |
1439 int AbstractInterpreter::layout_activation(methodOop method, | |
1440 int tempcount, | |
1441 int popframe_extra_args, | |
1442 int moncount, | |
1443 int callee_param_count, | |
1444 int callee_locals, | |
1445 frame* caller, | |
1446 frame* interpreter_frame, | |
1447 bool is_top_frame) { | |
1448 // Note: This calculation must exactly parallel the frame setup | |
1449 // in AbstractInterpreterGenerator::generate_method_entry. | |
1450 // If interpreter_frame!=NULL, set up the method, locals, and monitors. | |
1451 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size, | |
1452 // as determined by a previous call to this method. | |
1453 // It is also guaranteed to be walkable even though it is in a skeletal state | |
1454 // NOTE: return size is in words not bytes | |
1455 | |
1456 // fixed size of an interpreter frame: | |
1457 int max_locals = method->max_locals() * Interpreter::stackElementWords(); | |
1458 int extra_locals = (method->max_locals() - method->size_of_parameters()) * | |
1459 Interpreter::stackElementWords(); | |
1460 | |
1461 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset; | |
1462 | |
1463 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion) | |
1464 // Since the callee parameters already account for the callee's params we only need to account for | |
1465 // the extra locals. | |
1466 | |
1467 | |
1468 int size = overhead + | |
1469 ((callee_locals - callee_param_count)*Interpreter::stackElementWords()) + | |
1470 (moncount*frame::interpreter_frame_monitor_size()) + | |
1471 tempcount*Interpreter::stackElementWords() + popframe_extra_args; | |
1472 | |
1473 if (interpreter_frame != NULL) { | |
1474 #ifdef ASSERT | |
1475 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable"); | |
1476 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)"); | |
1477 #endif | |
1478 | |
1479 interpreter_frame->interpreter_frame_set_method(method); | |
1480 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp | |
1481 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp) | |
1482 // and sender_sp is fp+8 | |
1483 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1; | |
1484 | |
1485 interpreter_frame->interpreter_frame_set_locals(locals); | |
1486 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin(); | |
1487 BasicObjectLock* monbot = montop - moncount; | |
1488 interpreter_frame->interpreter_frame_set_monitor_end(monbot); | |
1489 | |
1490 // Set last_sp | |
1491 intptr_t* rsp = (intptr_t*) monbot - | |
1492 tempcount*Interpreter::stackElementWords() - | |
1493 popframe_extra_args; | |
1494 interpreter_frame->interpreter_frame_set_last_sp(rsp); | |
1495 | |
1496 // All frames but the initial (oldest) interpreter frame we fill in have a | |
1497 // value for sender_sp that allows walking the stack but isn't | |
1498 // truly correct. Correct the value here. | |
1499 | |
1500 if (extra_locals != 0 && | |
1501 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) { | |
1502 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals); | |
1503 } | |
1504 *interpreter_frame->interpreter_frame_cache_addr() = | |
1505 method->constants()->cache(); | |
1506 } | |
1507 return size; | |
1508 } | |
1509 | |
1510 | |
1511 //------------------------------------------------------------------------------------------------------------------------ | |
1512 // Exceptions | |
1513 | |
1514 void TemplateInterpreterGenerator::generate_throw_exception() { | |
1515 // Entry point in previous activation (i.e., if the caller was interpreted) | |
1516 Interpreter::_rethrow_exception_entry = __ pc(); | |
1517 | |
1518 // Restore sp to interpreter_frame_last_sp even though we are going | |
1519 // to empty the expression stack for the exception processing. | |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
337
diff
changeset
|
1520 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); |
0 | 1521 // rax,: exception |
1522 // rdx: return address/pc that threw exception | |
1523 __ restore_bcp(); // rsi points to call/send | |
1524 __ restore_locals(); | |
1525 | |
1526 // Entry point for exceptions thrown within interpreter code | |
1527 Interpreter::_throw_exception_entry = __ pc(); | |
1528 // expression stack is undefined here | |
1529 // rax,: exception | |
1530 // rsi: exception bcp | |
1531 __ verify_oop(rax); | |
1532 | |
1533 // expression stack must be empty before entering the VM in case of an exception | |
1534 __ empty_expression_stack(); | |
1535 __ empty_FPU_stack(); | |
1536 // find exception handler address and preserve exception oop | |
1537 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax); | |
1538 // rax,: exception handler entry point | |
1539 // rdx: preserved exception oop | |
1540 // rsi: bcp for exception handler | |
1541 __ push_ptr(rdx); // push exception which is now the only value on the stack | |
1542 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!) | |
1543 | |
1544 // If the exception is not handled in the current frame the frame is removed and | |
1545 // the exception is rethrown (i.e. exception continuation is _rethrow_exception). | |
1546 // | |
1547 // Note: At this point the bci is still the bxi for the instruction which caused | |
1548 // the exception and the expression stack is empty. Thus, for any VM calls | |
1549 // at this point, GC will find a legal oop map (with empty expression stack). | |
1550 | |
1551 // In current activation | |
1552 // tos: exception | |
1553 // rsi: exception bcp | |
1554 | |
1555 // | |
1556 // JVMTI PopFrame support | |
1557 // | |
1558 | |
1559 Interpreter::_remove_activation_preserving_args_entry = __ pc(); | |
1560 __ empty_expression_stack(); | |
1561 __ empty_FPU_stack(); | |
1562 // Set the popframe_processing bit in pending_popframe_condition indicating that we are | |
1563 // currently handling popframe, so that call_VMs that may happen later do not trigger new | |
1564 // popframe handling cycles. | |
1565 __ get_thread(rcx); | |
1566 __ movl(rdx, Address(rcx, JavaThread::popframe_condition_offset())); | |
1567 __ orl(rdx, JavaThread::popframe_processing_bit); | |
1568 __ movl(Address(rcx, JavaThread::popframe_condition_offset()), rdx); | |
1569 | |
1570 { | |
1571 // Check to see whether we are returning to a deoptimized frame. | |
1572 // (The PopFrame call ensures that the caller of the popped frame is | |
1573 // either interpreted or compiled and deoptimizes it if compiled.) | |
1574 // In this case, we can't call dispatch_next() after the frame is | |
1575 // popped, but instead must save the incoming arguments and restore | |
1576 // them after deoptimization has occurred. | |
1577 // | |
1578 // Note that we don't compare the return PC against the | |
1579 // deoptimization blob's unpack entry because of the presence of | |
1580 // adapter frames in C2. | |
1581 Label caller_not_deoptimized; | |
304 | 1582 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize)); |
0 | 1583 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx); |
1584 __ testl(rax, rax); | |
1585 __ jcc(Assembler::notZero, caller_not_deoptimized); | |
1586 | |
1587 // Compute size of arguments for saving when returning to deoptimized caller | |
1588 __ get_method(rax); | |
1589 __ verify_oop(rax); | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
533
diff
changeset
|
1590 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::size_of_parameters_offset()))); |
304 | 1591 __ shlptr(rax, Interpreter::logStackElementSize()); |
0 | 1592 __ restore_locals(); |
304 | 1593 __ subptr(rdi, rax); |
1594 __ addptr(rdi, wordSize); | |
0 | 1595 // Save these arguments |
1596 __ get_thread(rcx); | |
1597 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), rcx, rax, rdi); | |
1598 | |
1599 __ remove_activation(vtos, rdx, | |
1600 /* throw_monitor_exception */ false, | |
1601 /* install_monitor_exception */ false, | |
1602 /* notify_jvmdi */ false); | |
1603 | |
1604 // Inform deoptimization that it is responsible for restoring these arguments | |
1605 __ get_thread(rcx); | |
1606 __ movl(Address(rcx, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit); | |
1607 | |
1608 // Continue in deoptimization handler | |
1609 __ jmp(rdx); | |
1610 | |
1611 __ bind(caller_not_deoptimized); | |
1612 } | |
1613 | |
1614 __ remove_activation(vtos, rdx, | |
1615 /* throw_monitor_exception */ false, | |
1616 /* install_monitor_exception */ false, | |
1617 /* notify_jvmdi */ false); | |
1618 | |
1619 // Finish with popframe handling | |
1620 // A previous I2C followed by a deoptimization might have moved the | |
1621 // outgoing arguments further up the stack. PopFrame expects the | |
1622 // mutations to those outgoing arguments to be preserved and other | |
1623 // constraints basically require this frame to look exactly as | |
1624 // though it had previously invoked an interpreted activation with | |
1625 // no space between the top of the expression stack (current | |
1626 // last_sp) and the top of stack. Rather than force deopt to | |
1627 // maintain this kind of invariant all the time we call a small | |
1628 // fixup routine to move the mutated arguments onto the top of our | |
1629 // expression stack if necessary. | |
304 | 1630 __ mov(rax, rsp); |
1631 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); | |
0 | 1632 __ get_thread(rcx); |
1633 // PC must point into interpreter here | |
1634 __ set_last_Java_frame(rcx, noreg, rbp, __ pc()); | |
1635 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), rcx, rax, rbx); | |
1636 __ get_thread(rcx); | |
1637 __ reset_last_Java_frame(rcx, true, true); | |
1638 // Restore the last_sp and null it out | |
304 | 1639 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize)); |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
337
diff
changeset
|
1640 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); |
0 | 1641 |
1642 __ restore_bcp(); | |
1643 __ restore_locals(); | |
1644 // The method data pointer was incremented already during | |
1645 // call profiling. We have to restore the mdp for the current bcp. | |
1646 if (ProfileInterpreter) { | |
1647 __ set_method_data_pointer_for_bcp(); | |
1648 } | |
1649 | |
1650 // Clear the popframe condition flag | |
1651 __ get_thread(rcx); | |
1652 __ movl(Address(rcx, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive); | |
1653 | |
1654 __ dispatch_next(vtos); | |
1655 // end of PopFrame support | |
1656 | |
1657 Interpreter::_remove_activation_entry = __ pc(); | |
1658 | |
1659 // preserve exception over this code sequence | |
1660 __ pop_ptr(rax); | |
1661 __ get_thread(rcx); | |
304 | 1662 __ movptr(Address(rcx, JavaThread::vm_result_offset()), rax); |
0 | 1663 // remove the activation (without doing throws on illegalMonitorExceptions) |
1664 __ remove_activation(vtos, rdx, false, true, false); | |
1665 // restore exception | |
1666 __ get_thread(rcx); | |
304 | 1667 __ movptr(rax, Address(rcx, JavaThread::vm_result_offset())); |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
337
diff
changeset
|
1668 __ movptr(Address(rcx, JavaThread::vm_result_offset()), NULL_WORD); |
0 | 1669 __ verify_oop(rax); |
1670 | |
1671 // Inbetween activations - previous activation type unknown yet | |
1672 // compute continuation point - the continuation point expects | |
1673 // the following registers set up: | |
1674 // | |
1675 // rax,: exception | |
1676 // rdx: return address/pc that threw exception | |
1677 // rsp: expression stack of caller | |
1678 // rbp,: rbp, of caller | |
304 | 1679 __ push(rax); // save exception |
1680 __ push(rdx); // save return address | |
0 | 1681 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rdx); |
304 | 1682 __ mov(rbx, rax); // save exception handler |
1683 __ pop(rdx); // restore return address | |
1684 __ pop(rax); // restore exception | |
0 | 1685 // Note that an "issuing PC" is actually the next PC after the call |
1686 __ jmp(rbx); // jump to exception handler of caller | |
1687 } | |
1688 | |
1689 | |
1690 // | |
1691 // JVMTI ForceEarlyReturn support | |
1692 // | |
1693 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { | |
1694 address entry = __ pc(); | |
1695 | |
1696 __ restore_bcp(); | |
1697 __ restore_locals(); | |
1698 __ empty_expression_stack(); | |
1699 __ empty_FPU_stack(); | |
1700 __ load_earlyret_value(state); | |
1701 | |
1702 __ get_thread(rcx); | |
304 | 1703 __ movptr(rcx, Address(rcx, JavaThread::jvmti_thread_state_offset())); |
0 | 1704 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset()); |
1705 | |
1706 // Clear the earlyret state | |
1707 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive); | |
1708 | |
1709 __ remove_activation(state, rsi, | |
1710 false, /* throw_monitor_exception */ | |
1711 false, /* install_monitor_exception */ | |
1712 true); /* notify_jvmdi */ | |
1713 __ jmp(rsi); | |
1714 return entry; | |
1715 } // end of ForceEarlyReturn support | |
1716 | |
1717 | |
1718 //------------------------------------------------------------------------------------------------------------------------ | |
1719 // Helper for vtos entry point generation | |
1720 | |
1721 void TemplateInterpreterGenerator::set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) { | |
1722 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); | |
1723 Label L; | |
1724 fep = __ pc(); __ push(ftos); __ jmp(L); | |
1725 dep = __ pc(); __ push(dtos); __ jmp(L); | |
1726 lep = __ pc(); __ push(ltos); __ jmp(L); | |
1727 aep = __ pc(); __ push(atos); __ jmp(L); | |
1728 bep = cep = sep = // fall through | |
1729 iep = __ pc(); __ push(itos); // fall through | |
1730 vep = __ pc(); __ bind(L); // fall through | |
1731 generate_and_dispatch(t); | |
1732 } | |
1733 | |
1734 //------------------------------------------------------------------------------------------------------------------------ | |
1735 // Generation of individual instructions | |
1736 | |
1737 // helpers for generate_and_dispatch | |
1738 | |
1739 | |
1740 | |
1741 InterpreterGenerator::InterpreterGenerator(StubQueue* code) | |
1742 : TemplateInterpreterGenerator(code) { | |
1743 generate_all(); // down here so it can be "virtual" | |
1744 } | |
1745 | |
1746 //------------------------------------------------------------------------------------------------------------------------ | |
1747 | |
1748 // Non-product code | |
1749 #ifndef PRODUCT | |
1750 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { | |
1751 address entry = __ pc(); | |
1752 | |
1753 // prepare expression stack | |
304 | 1754 __ pop(rcx); // pop return address so expression stack is 'pure' |
0 | 1755 __ push(state); // save tosca |
1756 | |
1757 // pass tosca registers as arguments & call tracer | |
1758 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx); | |
304 | 1759 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state) |
0 | 1760 __ pop(state); // restore tosca |
1761 | |
1762 // return | |
1763 __ jmp(rcx); | |
1764 | |
1765 return entry; | |
1766 } | |
1767 | |
1768 | |
1769 void TemplateInterpreterGenerator::count_bytecode() { | |
304 | 1770 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value)); |
0 | 1771 } |
1772 | |
1773 | |
1774 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { | |
304 | 1775 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()])); |
0 | 1776 } |
1777 | |
1778 | |
1779 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { | |
1780 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx); | |
1781 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes); | |
1782 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); | |
1783 ExternalAddress table((address) BytecodePairHistogram::_counters); | |
1784 Address index(noreg, rbx, Address::times_4); | |
304 | 1785 __ incrementl(ArrayAddress(table, index)); |
0 | 1786 } |
1787 | |
1788 | |
1789 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { | |
1790 // Call a little run-time stub to avoid blow-up for each bytecode. | |
1791 // The run-time runtime saves the right registers, depending on | |
1792 // the tosca in-state for the given template. | |
1793 assert(Interpreter::trace_code(t->tos_in()) != NULL, | |
1794 "entry must have been generated"); | |
1795 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in()))); | |
1796 } | |
1797 | |
1798 | |
1799 void TemplateInterpreterGenerator::stop_interpreter_at() { | |
1800 Label L; | |
1801 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value), | |
1802 StopInterpreterAt); | |
1803 __ jcc(Assembler::notEqual, L); | |
1804 __ int3(); | |
1805 __ bind(L); | |
1806 } | |
1807 #endif // !PRODUCT | |
1808 #endif // CC_INTERP |