0
|
1 /*
|
|
2 * Copyright 1997-2007 Sun Microsystems, Inc. All Rights Reserved.
|
|
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
|
|
4 *
|
|
5 * This code is free software; you can redistribute it and/or modify it
|
|
6 * under the terms of the GNU General Public License version 2 only, as
|
|
7 * published by the Free Software Foundation.
|
|
8 *
|
|
9 * This code is distributed in the hope that it will be useful, but WITHOUT
|
|
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
|
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
|
12 * version 2 for more details (a copy is included in the LICENSE file that
|
|
13 * accompanied this code).
|
|
14 *
|
|
15 * You should have received a copy of the GNU General Public License version
|
|
16 * 2 along with this work; if not, write to the Free Software Foundation,
|
|
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
|
|
18 *
|
|
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
|
|
20 * CA 95054 USA or visit www.sun.com if you need additional information or
|
|
21 * have any questions.
|
|
22 *
|
|
23 */
|
|
24
|
|
25 #include "incls/_precompiled.incl"
|
|
26 #include "incls/_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;
|
|
46 __ leal(rax, Address(rbp,
|
|
47 frame::interpreter_frame_monitor_block_top_offset * wordSize));
|
|
48 __ cmpl(rax, rsp); // rax, = maximal rsp for current rbp,
|
|
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
|
|
83 __ popl(rax);
|
|
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
|
|
95 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
|
|
96 assert(!pass_oop || message == NULL, "either oop or message but not both");
|
|
97 address entry = __ pc();
|
|
98 if (pass_oop) {
|
|
99 // object is at TOS
|
|
100 __ popl(rbx);
|
|
101 }
|
|
102 // expression stack must be empty before entering the VM if an exception happened
|
|
103 __ empty_expression_stack();
|
|
104 __ empty_FPU_stack();
|
|
105 // setup parameters
|
|
106 __ lea(rax, ExternalAddress((address)name));
|
|
107 if (pass_oop) {
|
|
108 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
|
|
109 } else {
|
|
110 if (message != NULL) {
|
|
111 __ lea(rbx, ExternalAddress((address)message));
|
|
112 } else {
|
|
113 __ movl(rbx, NULL_WORD);
|
|
114 }
|
|
115 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
|
|
116 }
|
|
117 // throw exception
|
|
118 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
|
|
119 return entry;
|
|
120 }
|
|
121
|
|
122
|
|
123 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
|
|
124 address entry = __ pc();
|
|
125 // NULL last_sp until next java call
|
|
126 __ movl(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
|
|
127 __ dispatch_next(state);
|
|
128 return entry;
|
|
129 }
|
|
130
|
|
131
|
|
132 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
|
|
133 Label interpreter_entry;
|
|
134 address compiled_entry = __ pc();
|
|
135
|
|
136 #ifdef COMPILER2
|
|
137 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
|
|
138 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
|
|
139 for (int i = 1; i < 8; i++) {
|
|
140 __ ffree(i);
|
|
141 }
|
|
142 } else if (UseSSE < 2) {
|
|
143 __ empty_FPU_stack();
|
|
144 }
|
|
145 #endif
|
|
146 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
|
|
147 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
|
|
148 } else {
|
|
149 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
|
|
150 }
|
|
151
|
|
152 __ jmp(interpreter_entry, relocInfo::none);
|
|
153 // emit a sentinel we can test for when converting an interpreter
|
|
154 // entry point to a compiled entry point.
|
|
155 __ a_long(Interpreter::return_sentinel);
|
|
156 __ a_long((int)compiled_entry);
|
|
157 address entry = __ pc();
|
|
158 __ bind(interpreter_entry);
|
|
159
|
|
160 // In SSE mode, interpreter returns FP results in xmm0 but they need
|
|
161 // to end up back on the FPU so it can operate on them.
|
|
162 if (state == ftos && UseSSE >= 1) {
|
|
163 __ subl(rsp, wordSize);
|
|
164 __ movflt(Address(rsp, 0), xmm0);
|
|
165 __ fld_s(Address(rsp, 0));
|
|
166 __ addl(rsp, wordSize);
|
|
167 } else if (state == dtos && UseSSE >= 2) {
|
|
168 __ subl(rsp, 2*wordSize);
|
|
169 __ movdbl(Address(rsp, 0), xmm0);
|
|
170 __ fld_d(Address(rsp, 0));
|
|
171 __ addl(rsp, 2*wordSize);
|
|
172 }
|
|
173
|
|
174 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
|
|
175
|
|
176 // Restore stack bottom in case i2c adjusted stack
|
|
177 __ movl(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
|
|
178 // and NULL it as marker that rsp is now tos until next java call
|
|
179 __ movl(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
|
|
180
|
|
181 __ restore_bcp();
|
|
182 __ restore_locals();
|
|
183 __ get_cache_and_index_at_bcp(rbx, rcx, 1);
|
|
184 __ movl(rbx, Address(rbx, rcx,
|
|
185 Address::times_4, constantPoolCacheOopDesc::base_offset() +
|
|
186 ConstantPoolCacheEntry::flags_offset()));
|
|
187 __ andl(rbx, 0xFF);
|
|
188 __ leal(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
|
|
189 __ dispatch_next(state, step);
|
|
190 return entry;
|
|
191 }
|
|
192
|
|
193
|
|
194 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
|
|
195 address entry = __ pc();
|
|
196
|
|
197 // In SSE mode, FP results are in xmm0
|
|
198 if (state == ftos && UseSSE > 0) {
|
|
199 __ subl(rsp, wordSize);
|
|
200 __ movflt(Address(rsp, 0), xmm0);
|
|
201 __ fld_s(Address(rsp, 0));
|
|
202 __ addl(rsp, wordSize);
|
|
203 } else if (state == dtos && UseSSE >= 2) {
|
|
204 __ subl(rsp, 2*wordSize);
|
|
205 __ movdbl(Address(rsp, 0), xmm0);
|
|
206 __ fld_d(Address(rsp, 0));
|
|
207 __ addl(rsp, 2*wordSize);
|
|
208 }
|
|
209
|
|
210 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
|
|
211
|
|
212 // The stack is not extended by deopt but we must NULL last_sp as this
|
|
213 // entry is like a "return".
|
|
214 __ movl(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
|
|
215 __ restore_bcp();
|
|
216 __ restore_locals();
|
|
217 // handle exceptions
|
|
218 { Label L;
|
|
219 const Register thread = rcx;
|
|
220 __ get_thread(thread);
|
|
221 __ cmpl(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
|
|
222 __ jcc(Assembler::zero, L);
|
|
223 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
|
|
224 __ should_not_reach_here();
|
|
225 __ bind(L);
|
|
226 }
|
|
227 __ dispatch_next(state, step);
|
|
228 return entry;
|
|
229 }
|
|
230
|
|
231
|
|
232 int AbstractInterpreter::BasicType_as_index(BasicType type) {
|
|
233 int i = 0;
|
|
234 switch (type) {
|
|
235 case T_BOOLEAN: i = 0; break;
|
|
236 case T_CHAR : i = 1; break;
|
|
237 case T_BYTE : i = 2; break;
|
|
238 case T_SHORT : i = 3; break;
|
|
239 case T_INT : // fall through
|
|
240 case T_LONG : // fall through
|
|
241 case T_VOID : i = 4; break;
|
|
242 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
|
|
243 case T_DOUBLE : i = 6; break;
|
|
244 case T_OBJECT : // fall through
|
|
245 case T_ARRAY : i = 7; break;
|
|
246 default : ShouldNotReachHere();
|
|
247 }
|
|
248 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
|
|
249 return i;
|
|
250 }
|
|
251
|
|
252
|
|
253 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
|
|
254 address entry = __ pc();
|
|
255 switch (type) {
|
|
256 case T_BOOLEAN: __ c2bool(rax); break;
|
|
257 case T_CHAR : __ andl(rax, 0xFFFF); break;
|
|
258 case T_BYTE : __ sign_extend_byte (rax); break;
|
|
259 case T_SHORT : __ sign_extend_short(rax); break;
|
|
260 case T_INT : /* nothing to do */ break;
|
|
261 case T_DOUBLE :
|
|
262 case T_FLOAT :
|
|
263 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
|
|
264 __ popl(t); // remove return address first
|
|
265 __ pop_dtos_to_rsp();
|
|
266 // Must return a result for interpreter or compiler. In SSE
|
|
267 // mode, results are returned in xmm0 and the FPU stack must
|
|
268 // be empty.
|
|
269 if (type == T_FLOAT && UseSSE >= 1) {
|
|
270 // Load ST0
|
|
271 __ fld_d(Address(rsp, 0));
|
|
272 // Store as float and empty fpu stack
|
|
273 __ fstp_s(Address(rsp, 0));
|
|
274 // and reload
|
|
275 __ movflt(xmm0, Address(rsp, 0));
|
|
276 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
|
|
277 __ movdbl(xmm0, Address(rsp, 0));
|
|
278 } else {
|
|
279 // restore ST0
|
|
280 __ fld_d(Address(rsp, 0));
|
|
281 }
|
|
282 // and pop the temp
|
|
283 __ addl(rsp, 2 * wordSize);
|
|
284 __ pushl(t); // restore return address
|
|
285 }
|
|
286 break;
|
|
287 case T_OBJECT :
|
|
288 // retrieve result from frame
|
|
289 __ movl(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
|
|
290 // and verify it
|
|
291 __ verify_oop(rax);
|
|
292 break;
|
|
293 default : ShouldNotReachHere();
|
|
294 }
|
|
295 __ ret(0); // return from result handler
|
|
296 return entry;
|
|
297 }
|
|
298
|
|
299 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
|
|
300 address entry = __ pc();
|
|
301 __ push(state);
|
|
302 __ call_VM(noreg, runtime_entry);
|
|
303 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
|
|
304 return entry;
|
|
305 }
|
|
306
|
|
307
|
|
308 // Helpers for commoning out cases in the various type of method entries.
|
|
309 //
|
|
310
|
|
311 // increment invocation count & check for overflow
|
|
312 //
|
|
313 // Note: checking for negative value instead of overflow
|
|
314 // so we have a 'sticky' overflow test
|
|
315 //
|
|
316 // rbx,: method
|
|
317 // rcx: invocation counter
|
|
318 //
|
|
319 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
|
|
320
|
|
321 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
|
|
322 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset());
|
|
323
|
|
324 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
|
|
325 __ increment(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset()));
|
|
326 }
|
|
327 // Update standard invocation counters
|
|
328 __ movl(rax, backedge_counter); // load backedge counter
|
|
329
|
|
330 __ increment(rcx, InvocationCounter::count_increment);
|
|
331 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
|
|
332
|
|
333 __ movl(invocation_counter, rcx); // save invocation count
|
|
334 __ addl(rcx, rax); // add both counters
|
|
335
|
|
336 // profile_method is non-null only for interpreted method so
|
|
337 // profile_method != NULL == !native_call
|
|
338 // BytecodeInterpreter only calls for native so code is elided.
|
|
339
|
|
340 if (ProfileInterpreter && profile_method != NULL) {
|
|
341 // Test to see if we should create a method data oop
|
|
342 __ cmp32(rcx,
|
|
343 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
|
|
344 __ jcc(Assembler::less, *profile_method_continue);
|
|
345
|
|
346 // if no method data exists, go to profile_method
|
|
347 __ test_method_data_pointer(rax, *profile_method);
|
|
348 }
|
|
349
|
|
350 __ cmp32(rcx,
|
|
351 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
|
|
352 __ jcc(Assembler::aboveEqual, *overflow);
|
|
353
|
|
354 }
|
|
355
|
|
356 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
|
|
357
|
|
358 // Asm interpreter on entry
|
|
359 // rdi - locals
|
|
360 // rsi - bcp
|
|
361 // rbx, - method
|
|
362 // rdx - cpool
|
|
363 // rbp, - interpreter frame
|
|
364
|
|
365 // C++ interpreter on entry
|
|
366 // rsi - new interpreter state pointer
|
|
367 // rbp - interpreter frame pointer
|
|
368 // rbx - method
|
|
369
|
|
370 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
|
|
371 // rbx, - method
|
|
372 // rcx - rcvr (assuming there is one)
|
|
373 // top of stack return address of interpreter caller
|
|
374 // rsp - sender_sp
|
|
375
|
|
376 // C++ interpreter only
|
|
377 // rsi - previous interpreter state pointer
|
|
378
|
|
379 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
|
|
380
|
|
381 // InterpreterRuntime::frequency_counter_overflow takes one argument
|
|
382 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
|
|
383 // The call returns the address of the verified entry point for the method or NULL
|
|
384 // if the compilation did not complete (either went background or bailed out).
|
|
385 __ movl(rax, (int)false);
|
|
386 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
|
|
387
|
|
388 __ movl(rbx, Address(rbp, method_offset)); // restore methodOop
|
|
389
|
|
390 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
|
|
391 // and jump to the interpreted entry.
|
|
392 __ jmp(*do_continue, relocInfo::none);
|
|
393
|
|
394 }
|
|
395
|
|
396 void InterpreterGenerator::generate_stack_overflow_check(void) {
|
|
397 // see if we've got enough room on the stack for locals plus overhead.
|
|
398 // the expression stack grows down incrementally, so the normal guard
|
|
399 // page mechanism will work for that.
|
|
400 //
|
|
401 // Registers live on entry:
|
|
402 //
|
|
403 // Asm interpreter
|
|
404 // rdx: number of additional locals this frame needs (what we must check)
|
|
405 // rbx,: methodOop
|
|
406
|
|
407 // destroyed on exit
|
|
408 // rax,
|
|
409
|
|
410 // NOTE: since the additional locals are also always pushed (wasn't obvious in
|
|
411 // generate_method_entry) so the guard should work for them too.
|
|
412 //
|
|
413
|
|
414 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
|
|
415 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
|
|
416
|
|
417 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
|
|
418 // be sure to change this if you add/subtract anything to/from the overhead area
|
|
419 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
|
|
420
|
|
421 const int page_size = os::vm_page_size();
|
|
422
|
|
423 Label after_frame_check;
|
|
424
|
|
425 // see if the frame is greater than one page in size. If so,
|
|
426 // then we need to verify there is enough stack space remaining
|
|
427 // for the additional locals.
|
|
428 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize());
|
|
429 __ jcc(Assembler::belowEqual, after_frame_check);
|
|
430
|
|
431 // compute rsp as if this were going to be the last frame on
|
|
432 // the stack before the red zone
|
|
433
|
|
434 Label after_frame_check_pop;
|
|
435
|
|
436 __ pushl(rsi);
|
|
437
|
|
438 const Register thread = rsi;
|
|
439
|
|
440 __ get_thread(thread);
|
|
441
|
|
442 const Address stack_base(thread, Thread::stack_base_offset());
|
|
443 const Address stack_size(thread, Thread::stack_size_offset());
|
|
444
|
|
445 // locals + overhead, in bytes
|
|
446 __ leal(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
|
|
447
|
|
448 #ifdef ASSERT
|
|
449 Label stack_base_okay, stack_size_okay;
|
|
450 // verify that thread stack base is non-zero
|
|
451 __ cmpl(stack_base, 0);
|
|
452 __ jcc(Assembler::notEqual, stack_base_okay);
|
|
453 __ stop("stack base is zero");
|
|
454 __ bind(stack_base_okay);
|
|
455 // verify that thread stack size is non-zero
|
|
456 __ cmpl(stack_size, 0);
|
|
457 __ jcc(Assembler::notEqual, stack_size_okay);
|
|
458 __ stop("stack size is zero");
|
|
459 __ bind(stack_size_okay);
|
|
460 #endif
|
|
461
|
|
462 // Add stack base to locals and subtract stack size
|
|
463 __ addl(rax, stack_base);
|
|
464 __ subl(rax, stack_size);
|
|
465
|
|
466 // Use the maximum number of pages we might bang.
|
|
467 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
|
|
468 (StackRedPages+StackYellowPages);
|
|
469 __ addl(rax, max_pages * page_size);
|
|
470
|
|
471 // check against the current stack bottom
|
|
472 __ cmpl(rsp, rax);
|
|
473 __ jcc(Assembler::above, after_frame_check_pop);
|
|
474
|
|
475 __ popl(rsi); // get saved bcp / (c++ prev state ).
|
|
476
|
|
477 __ popl(rax); // get return address
|
|
478 __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
|
|
479
|
|
480 // all done with frame size check
|
|
481 __ bind(after_frame_check_pop);
|
|
482 __ popl(rsi);
|
|
483
|
|
484 __ bind(after_frame_check);
|
|
485 }
|
|
486
|
|
487 // Allocate monitor and lock method (asm interpreter)
|
|
488 // rbx, - methodOop
|
|
489 //
|
|
490 void InterpreterGenerator::lock_method(void) {
|
|
491 // synchronize method
|
|
492 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
|
|
493 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
|
|
494 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
|
|
495
|
|
496 #ifdef ASSERT
|
|
497 { Label L;
|
|
498 __ movl(rax, access_flags);
|
|
499 __ testl(rax, JVM_ACC_SYNCHRONIZED);
|
|
500 __ jcc(Assembler::notZero, L);
|
|
501 __ stop("method doesn't need synchronization");
|
|
502 __ bind(L);
|
|
503 }
|
|
504 #endif // ASSERT
|
|
505 // get synchronization object
|
|
506 { Label done;
|
|
507 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
|
|
508 __ movl(rax, access_flags);
|
|
509 __ testl(rax, JVM_ACC_STATIC);
|
|
510 __ movl(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
|
|
511 __ jcc(Assembler::zero, done);
|
|
512 __ movl(rax, Address(rbx, methodOopDesc::constants_offset()));
|
|
513 __ movl(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
|
|
514 __ movl(rax, Address(rax, mirror_offset));
|
|
515 __ bind(done);
|
|
516 }
|
|
517 // add space for monitor & lock
|
|
518 __ subl(rsp, entry_size); // add space for a monitor entry
|
|
519 __ movl(monitor_block_top, rsp); // set new monitor block top
|
|
520 __ movl(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
|
|
521 __ movl(rdx, rsp); // object address
|
|
522 __ lock_object(rdx);
|
|
523 }
|
|
524
|
|
525 //
|
|
526 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
|
|
527 // and for native methods hence the shared code.
|
|
528
|
|
529 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
|
|
530 // initialize fixed part of activation frame
|
|
531 __ pushl(rax); // save return address
|
|
532 __ enter(); // save old & set new rbp,
|
|
533
|
|
534
|
|
535 __ pushl(rsi); // set sender sp
|
|
536 __ pushl(NULL_WORD); // leave last_sp as null
|
|
537 __ movl(rsi, Address(rbx,methodOopDesc::const_offset())); // get constMethodOop
|
|
538 __ leal(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
|
|
539 __ pushl(rbx); // save methodOop
|
|
540 if (ProfileInterpreter) {
|
|
541 Label method_data_continue;
|
|
542 __ movl(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
|
|
543 __ testl(rdx, rdx);
|
|
544 __ jcc(Assembler::zero, method_data_continue);
|
|
545 __ addl(rdx, in_bytes(methodDataOopDesc::data_offset()));
|
|
546 __ bind(method_data_continue);
|
|
547 __ pushl(rdx); // set the mdp (method data pointer)
|
|
548 } else {
|
|
549 __ pushl(0);
|
|
550 }
|
|
551
|
|
552 __ movl(rdx, Address(rbx, methodOopDesc::constants_offset()));
|
|
553 __ movl(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
|
|
554 __ pushl(rdx); // set constant pool cache
|
|
555 __ pushl(rdi); // set locals pointer
|
|
556 if (native_call) {
|
|
557 __ pushl(0); // no bcp
|
|
558 } else {
|
|
559 __ pushl(rsi); // set bcp
|
|
560 }
|
|
561 __ pushl(0); // reserve word for pointer to expression stack bottom
|
|
562 __ movl(Address(rsp, 0), rsp); // set expression stack bottom
|
|
563 }
|
|
564
|
|
565 // End of helpers
|
|
566
|
|
567 //
|
|
568 // Various method entries
|
|
569 //------------------------------------------------------------------------------------------------------------------------
|
|
570 //
|
|
571 //
|
|
572
|
|
573 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
|
|
574
|
|
575 address InterpreterGenerator::generate_accessor_entry(void) {
|
|
576
|
|
577 // rbx,: methodOop
|
|
578 // rcx: receiver (preserve for slow entry into asm interpreter)
|
|
579
|
|
580 // rsi: senderSP must preserved for slow path, set SP to it on fast path
|
|
581
|
|
582 address entry_point = __ pc();
|
|
583 Label xreturn_path;
|
|
584
|
|
585 // do fastpath for resolved accessor methods
|
|
586 if (UseFastAccessorMethods) {
|
|
587 Label slow_path;
|
|
588 // If we need a safepoint check, generate full interpreter entry.
|
|
589 ExternalAddress state(SafepointSynchronize::address_of_state());
|
|
590 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
|
|
591 SafepointSynchronize::_not_synchronized);
|
|
592
|
|
593 __ jcc(Assembler::notEqual, slow_path);
|
|
594 // ASM/C++ Interpreter
|
|
595 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
|
|
596 // Note: We can only use this code if the getfield has been resolved
|
|
597 // and if we don't have a null-pointer exception => check for
|
|
598 // these conditions first and use slow path if necessary.
|
|
599 // rbx,: method
|
|
600 // rcx: receiver
|
|
601 __ movl(rax, Address(rsp, wordSize));
|
|
602
|
|
603 // check if local 0 != NULL and read field
|
|
604 __ testl(rax, rax);
|
|
605 __ jcc(Assembler::zero, slow_path);
|
|
606
|
|
607 __ movl(rdi, Address(rbx, methodOopDesc::constants_offset()));
|
|
608 // read first instruction word and extract bytecode @ 1 and index @ 2
|
|
609 __ movl(rdx, Address(rbx, methodOopDesc::const_offset()));
|
|
610 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
|
|
611 // Shift codes right to get the index on the right.
|
|
612 // The bytecode fetched looks like <index><0xb4><0x2a>
|
|
613 __ shrl(rdx, 2*BitsPerByte);
|
|
614 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
|
|
615 __ movl(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
|
|
616
|
|
617 // rax,: local 0
|
|
618 // rbx,: method
|
|
619 // rcx: receiver - do not destroy since it is needed for slow path!
|
|
620 // rcx: scratch
|
|
621 // rdx: constant pool cache index
|
|
622 // rdi: constant pool cache
|
|
623 // rsi: sender sp
|
|
624
|
|
625 // check if getfield has been resolved and read constant pool cache entry
|
|
626 // check the validity of the cache entry by testing whether _indices field
|
|
627 // contains Bytecode::_getfield in b1 byte.
|
|
628 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
|
|
629 __ movl(rcx,
|
|
630 Address(rdi,
|
|
631 rdx,
|
|
632 Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
|
|
633 __ shrl(rcx, 2*BitsPerByte);
|
|
634 __ andl(rcx, 0xFF);
|
|
635 __ cmpl(rcx, Bytecodes::_getfield);
|
|
636 __ jcc(Assembler::notEqual, slow_path);
|
|
637
|
|
638 // Note: constant pool entry is not valid before bytecode is resolved
|
|
639 __ movl(rcx,
|
|
640 Address(rdi,
|
|
641 rdx,
|
|
642 Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()));
|
|
643 __ movl(rdx,
|
|
644 Address(rdi,
|
|
645 rdx,
|
|
646 Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset()));
|
|
647
|
|
648 Label notByte, notShort, notChar;
|
|
649 const Address field_address (rax, rcx, Address::times_1);
|
|
650
|
|
651 // Need to differentiate between igetfield, agetfield, bgetfield etc.
|
|
652 // because they are different sizes.
|
|
653 // Use the type from the constant pool cache
|
|
654 __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
|
|
655 // Make sure we don't need to mask rdx for tosBits after the above shift
|
|
656 ConstantPoolCacheEntry::verify_tosBits();
|
|
657 __ cmpl(rdx, btos);
|
|
658 __ jcc(Assembler::notEqual, notByte);
|
|
659 __ load_signed_byte(rax, field_address);
|
|
660 __ jmp(xreturn_path);
|
|
661
|
|
662 __ bind(notByte);
|
|
663 __ cmpl(rdx, stos);
|
|
664 __ jcc(Assembler::notEqual, notShort);
|
|
665 __ load_signed_word(rax, field_address);
|
|
666 __ jmp(xreturn_path);
|
|
667
|
|
668 __ bind(notShort);
|
|
669 __ cmpl(rdx, ctos);
|
|
670 __ jcc(Assembler::notEqual, notChar);
|
|
671 __ load_unsigned_word(rax, field_address);
|
|
672 __ jmp(xreturn_path);
|
|
673
|
|
674 __ bind(notChar);
|
|
675 #ifdef ASSERT
|
|
676 Label okay;
|
|
677 __ cmpl(rdx, atos);
|
|
678 __ jcc(Assembler::equal, okay);
|
|
679 __ cmpl(rdx, itos);
|
|
680 __ jcc(Assembler::equal, okay);
|
|
681 __ stop("what type is this?");
|
|
682 __ bind(okay);
|
|
683 #endif // ASSERT
|
|
684 // All the rest are a 32 bit wordsize
|
|
685 __ movl(rax, field_address);
|
|
686
|
|
687 __ bind(xreturn_path);
|
|
688
|
|
689 // _ireturn/_areturn
|
|
690 __ popl(rdi); // get return address
|
|
691 __ movl(rsp, rsi); // set sp to sender sp
|
|
692 __ jmp(rdi);
|
|
693
|
|
694 // generate a vanilla interpreter entry as the slow path
|
|
695 __ bind(slow_path);
|
|
696
|
|
697 (void) generate_normal_entry(false);
|
|
698 return entry_point;
|
|
699 }
|
|
700 return NULL;
|
|
701
|
|
702 }
|
|
703
|
|
704 //
|
|
705 // Interpreter stub for calling a native method. (asm interpreter)
|
|
706 // This sets up a somewhat different looking stack for calling the native method
|
|
707 // than the typical interpreter frame setup.
|
|
708 //
|
|
709
|
|
710 address InterpreterGenerator::generate_native_entry(bool synchronized) {
|
|
711 // determine code generation flags
|
|
712 bool inc_counter = UseCompiler || CountCompiledCalls;
|
|
713
|
|
714 // rbx,: methodOop
|
|
715 // rsi: sender sp
|
|
716 // rsi: previous interpreter state (C++ interpreter) must preserve
|
|
717 address entry_point = __ pc();
|
|
718
|
|
719
|
|
720 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
|
|
721 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
|
|
722 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
|
|
723
|
|
724 // get parameter size (always needed)
|
|
725 __ load_unsigned_word(rcx, size_of_parameters);
|
|
726
|
|
727 // native calls don't need the stack size check since they have no expression stack
|
|
728 // and the arguments are already on the stack and we only add a handful of words
|
|
729 // to the stack
|
|
730
|
|
731 // rbx,: methodOop
|
|
732 // rcx: size of parameters
|
|
733 // rsi: sender sp
|
|
734
|
|
735 __ popl(rax); // get return address
|
|
736 // for natives the size of locals is zero
|
|
737
|
|
738 // compute beginning of parameters (rdi)
|
|
739 __ leal(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
|
|
740
|
|
741
|
|
742 // add 2 zero-initialized slots for native calls
|
|
743 // NULL result handler
|
|
744 __ pushl(NULL_WORD);
|
|
745 // NULL oop temp (mirror or jni oop result)
|
|
746 __ pushl(NULL_WORD);
|
|
747
|
|
748 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
|
|
749 // initialize fixed part of activation frame
|
|
750
|
|
751 generate_fixed_frame(true);
|
|
752
|
|
753 // make sure method is native & not abstract
|
|
754 #ifdef ASSERT
|
|
755 __ movl(rax, access_flags);
|
|
756 {
|
|
757 Label L;
|
|
758 __ testl(rax, JVM_ACC_NATIVE);
|
|
759 __ jcc(Assembler::notZero, L);
|
|
760 __ stop("tried to execute non-native method as native");
|
|
761 __ bind(L);
|
|
762 }
|
|
763 { Label L;
|
|
764 __ testl(rax, JVM_ACC_ABSTRACT);
|
|
765 __ jcc(Assembler::zero, L);
|
|
766 __ stop("tried to execute abstract method in interpreter");
|
|
767 __ bind(L);
|
|
768 }
|
|
769 #endif
|
|
770
|
|
771 // Since at this point in the method invocation the exception handler
|
|
772 // would try to exit the monitor of synchronized methods which hasn't
|
|
773 // been entered yet, we set the thread local variable
|
|
774 // _do_not_unlock_if_synchronized to true. The remove_activation will
|
|
775 // check this flag.
|
|
776
|
|
777 __ get_thread(rax);
|
|
778 const Address do_not_unlock_if_synchronized(rax,
|
|
779 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
|
|
780 __ movbool(do_not_unlock_if_synchronized, true);
|
|
781
|
|
782 // increment invocation count & check for overflow
|
|
783 Label invocation_counter_overflow;
|
|
784 if (inc_counter) {
|
|
785 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
|
|
786 }
|
|
787
|
|
788 Label continue_after_compile;
|
|
789 __ bind(continue_after_compile);
|
|
790
|
|
791 bang_stack_shadow_pages(true);
|
|
792
|
|
793 // reset the _do_not_unlock_if_synchronized flag
|
|
794 __ get_thread(rax);
|
|
795 __ movbool(do_not_unlock_if_synchronized, false);
|
|
796
|
|
797 // check for synchronized methods
|
|
798 // Must happen AFTER invocation_counter check and stack overflow check,
|
|
799 // so method is not locked if overflows.
|
|
800 //
|
|
801 if (synchronized) {
|
|
802 lock_method();
|
|
803 } else {
|
|
804 // no synchronization necessary
|
|
805 #ifdef ASSERT
|
|
806 { Label L;
|
|
807 __ movl(rax, access_flags);
|
|
808 __ testl(rax, JVM_ACC_SYNCHRONIZED);
|
|
809 __ jcc(Assembler::zero, L);
|
|
810 __ stop("method needs synchronization");
|
|
811 __ bind(L);
|
|
812 }
|
|
813 #endif
|
|
814 }
|
|
815
|
|
816 // start execution
|
|
817 #ifdef ASSERT
|
|
818 { Label L;
|
|
819 const Address monitor_block_top (rbp,
|
|
820 frame::interpreter_frame_monitor_block_top_offset * wordSize);
|
|
821 __ movl(rax, monitor_block_top);
|
|
822 __ cmpl(rax, rsp);
|
|
823 __ jcc(Assembler::equal, L);
|
|
824 __ stop("broken stack frame setup in interpreter");
|
|
825 __ bind(L);
|
|
826 }
|
|
827 #endif
|
|
828
|
|
829 // jvmti/dtrace support
|
|
830 __ notify_method_entry();
|
|
831
|
|
832 // work registers
|
|
833 const Register method = rbx;
|
|
834 const Register thread = rdi;
|
|
835 const Register t = rcx;
|
|
836
|
|
837 // allocate space for parameters
|
|
838 __ get_method(method);
|
|
839 __ verify_oop(method);
|
|
840 __ load_unsigned_word(t, Address(method, methodOopDesc::size_of_parameters_offset()));
|
|
841 __ shll(t, Interpreter::logStackElementSize());
|
|
842 __ addl(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
|
|
843 __ subl(rsp, t);
|
|
844 __ andl(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
|
|
845
|
|
846 // get signature handler
|
|
847 { Label L;
|
|
848 __ movl(t, Address(method, methodOopDesc::signature_handler_offset()));
|
|
849 __ testl(t, t);
|
|
850 __ jcc(Assembler::notZero, L);
|
|
851 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
|
|
852 __ get_method(method);
|
|
853 __ movl(t, Address(method, methodOopDesc::signature_handler_offset()));
|
|
854 __ bind(L);
|
|
855 }
|
|
856
|
|
857 // call signature handler
|
|
858 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
|
|
859 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
|
|
860 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
|
|
861 // The generated handlers do not touch RBX (the method oop).
|
|
862 // However, large signatures cannot be cached and are generated
|
|
863 // each time here. The slow-path generator will blow RBX
|
|
864 // sometime, so we must reload it after the call.
|
|
865 __ call(t);
|
|
866 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
|
|
867
|
|
868 // result handler is in rax,
|
|
869 // set result handler
|
|
870 __ movl(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
|
|
871
|
|
872 // pass mirror handle if static call
|
|
873 { Label L;
|
|
874 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
|
|
875 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
|
|
876 __ testl(t, JVM_ACC_STATIC);
|
|
877 __ jcc(Assembler::zero, L);
|
|
878 // get mirror
|
|
879 __ movl(t, Address(method, methodOopDesc:: constants_offset()));
|
|
880 __ movl(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
|
|
881 __ movl(t, Address(t, mirror_offset));
|
|
882 // copy mirror into activation frame
|
|
883 __ movl(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
|
|
884 // pass handle to mirror
|
|
885 __ leal(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
|
|
886 __ movl(Address(rsp, wordSize), t);
|
|
887 __ bind(L);
|
|
888 }
|
|
889
|
|
890 // get native function entry point
|
|
891 { Label L;
|
|
892 __ movl(rax, Address(method, methodOopDesc::native_function_offset()));
|
|
893 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
|
|
894 __ cmp32(rax, unsatisfied.addr());
|
|
895 __ jcc(Assembler::notEqual, L);
|
|
896 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
|
|
897 __ get_method(method);
|
|
898 __ verify_oop(method);
|
|
899 __ movl(rax, Address(method, methodOopDesc::native_function_offset()));
|
|
900 __ bind(L);
|
|
901 }
|
|
902
|
|
903 // pass JNIEnv
|
|
904 __ get_thread(thread);
|
|
905 __ leal(t, Address(thread, JavaThread::jni_environment_offset()));
|
|
906 __ movl(Address(rsp, 0), t);
|
|
907
|
|
908 // set_last_Java_frame_before_call
|
|
909 // It is enough that the pc()
|
|
910 // points into the right code segment. It does not have to be the correct return pc.
|
|
911 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
|
|
912
|
|
913 // change thread state
|
|
914 #ifdef ASSERT
|
|
915 { Label L;
|
|
916 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
|
|
917 __ cmpl(t, _thread_in_Java);
|
|
918 __ jcc(Assembler::equal, L);
|
|
919 __ stop("Wrong thread state in native stub");
|
|
920 __ bind(L);
|
|
921 }
|
|
922 #endif
|
|
923
|
|
924 // Change state to native
|
|
925 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
|
|
926 __ call(rax);
|
|
927
|
|
928 // result potentially in rdx:rax or ST0
|
|
929
|
|
930 // Either restore the MXCSR register after returning from the JNI Call
|
|
931 // or verify that it wasn't changed.
|
|
932 if (VM_Version::supports_sse()) {
|
|
933 if (RestoreMXCSROnJNICalls) {
|
|
934 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
|
|
935 }
|
|
936 else if (CheckJNICalls ) {
|
|
937 __ call(RuntimeAddress(StubRoutines::i486::verify_mxcsr_entry()));
|
|
938 }
|
|
939 }
|
|
940
|
|
941 // Either restore the x87 floating pointer control word after returning
|
|
942 // from the JNI call or verify that it wasn't changed.
|
|
943 if (CheckJNICalls) {
|
|
944 __ call(RuntimeAddress(StubRoutines::i486::verify_fpu_cntrl_wrd_entry()));
|
|
945 }
|
|
946
|
|
947 // save potential result in ST(0) & rdx:rax
|
|
948 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
|
|
949 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
|
|
950 // It is safe to do this push because state is _thread_in_native and return address will be found
|
|
951 // via _last_native_pc and not via _last_jave_sp
|
|
952
|
|
953 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
|
|
954 // If the order changes or anything else is added to the stack the code in
|
|
955 // interpreter_frame_result will have to be changed.
|
|
956
|
|
957 { Label L;
|
|
958 Label push_double;
|
|
959 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
|
|
960 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
|
|
961 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
|
|
962 float_handler.addr());
|
|
963 __ jcc(Assembler::equal, push_double);
|
|
964 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
|
|
965 double_handler.addr());
|
|
966 __ jcc(Assembler::notEqual, L);
|
|
967 __ bind(push_double);
|
|
968 __ push(dtos);
|
|
969 __ bind(L);
|
|
970 }
|
|
971 __ push(ltos);
|
|
972
|
|
973 // change thread state
|
|
974 __ get_thread(thread);
|
|
975 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
|
|
976 if(os::is_MP()) {
|
|
977 if (UseMembar) {
|
|
978 __ membar(); // Force this write out before the read below
|
|
979 } else {
|
|
980 // Write serialization page so VM thread can do a pseudo remote membar.
|
|
981 // We use the current thread pointer to calculate a thread specific
|
|
982 // offset to write to within the page. This minimizes bus traffic
|
|
983 // due to cache line collision.
|
|
984 __ serialize_memory(thread, rcx);
|
|
985 }
|
|
986 }
|
|
987
|
|
988 if (AlwaysRestoreFPU) {
|
|
989 // Make sure the control word is correct.
|
|
990 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
|
|
991 }
|
|
992
|
|
993 // check for safepoint operation in progress and/or pending suspend requests
|
|
994 { Label Continue;
|
|
995
|
|
996 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
|
|
997 SafepointSynchronize::_not_synchronized);
|
|
998
|
|
999 Label L;
|
|
1000 __ jcc(Assembler::notEqual, L);
|
|
1001 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
|
|
1002 __ jcc(Assembler::equal, Continue);
|
|
1003 __ bind(L);
|
|
1004
|
|
1005 // Don't use call_VM as it will see a possible pending exception and forward it
|
|
1006 // and never return here preventing us from clearing _last_native_pc down below.
|
|
1007 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
|
|
1008 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
|
|
1009 // by hand.
|
|
1010 //
|
|
1011 __ pushl(thread);
|
|
1012 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
|
|
1013 JavaThread::check_special_condition_for_native_trans)));
|
|
1014 __ increment(rsp, wordSize);
|
|
1015 __ get_thread(thread);
|
|
1016
|
|
1017 __ bind(Continue);
|
|
1018 }
|
|
1019
|
|
1020 // change thread state
|
|
1021 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
|
|
1022
|
|
1023 __ reset_last_Java_frame(thread, true, true);
|
|
1024
|
|
1025 // reset handle block
|
|
1026 __ movl(t, Address(thread, JavaThread::active_handles_offset()));
|
|
1027 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), 0);
|
|
1028
|
|
1029 // If result was an oop then unbox and save it in the frame
|
|
1030 { Label L;
|
|
1031 Label no_oop, store_result;
|
|
1032 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
|
|
1033 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
|
|
1034 handler.addr());
|
|
1035 __ jcc(Assembler::notEqual, no_oop);
|
|
1036 __ cmpl(Address(rsp, 0), NULL_WORD);
|
|
1037 __ pop(ltos);
|
|
1038 __ testl(rax, rax);
|
|
1039 __ jcc(Assembler::zero, store_result);
|
|
1040 // unbox
|
|
1041 __ movl(rax, Address(rax, 0));
|
|
1042 __ bind(store_result);
|
|
1043 __ movl(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
|
|
1044 // keep stack depth as expected by pushing oop which will eventually be discarded
|
|
1045 __ push(ltos);
|
|
1046 __ bind(no_oop);
|
|
1047 }
|
|
1048
|
|
1049 {
|
|
1050 Label no_reguard;
|
|
1051 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
|
|
1052 __ jcc(Assembler::notEqual, no_reguard);
|
|
1053
|
|
1054 __ pushad();
|
|
1055 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
|
|
1056 __ popad();
|
|
1057
|
|
1058 __ bind(no_reguard);
|
|
1059 }
|
|
1060
|
|
1061 // restore rsi to have legal interpreter frame,
|
|
1062 // i.e., bci == 0 <=> rsi == code_base()
|
|
1063 // Can't call_VM until bcp is within reasonable.
|
|
1064 __ get_method(method); // method is junk from thread_in_native to now.
|
|
1065 __ verify_oop(method);
|
|
1066 __ movl(rsi, Address(method,methodOopDesc::const_offset())); // get constMethodOop
|
|
1067 __ leal(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
|
|
1068
|
|
1069 // handle exceptions (exception handling will handle unlocking!)
|
|
1070 { Label L;
|
|
1071 __ cmpl(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
|
|
1072 __ jcc(Assembler::zero, L);
|
|
1073 // Note: At some point we may want to unify this with the code used in call_VM_base();
|
|
1074 // i.e., we should use the StubRoutines::forward_exception code. For now this
|
|
1075 // doesn't work here because the rsp is not correctly set at this point.
|
|
1076 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
|
|
1077 __ should_not_reach_here();
|
|
1078 __ bind(L);
|
|
1079 }
|
|
1080
|
|
1081 // do unlocking if necessary
|
|
1082 { Label L;
|
|
1083 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
|
|
1084 __ testl(t, JVM_ACC_SYNCHRONIZED);
|
|
1085 __ jcc(Assembler::zero, L);
|
|
1086 // the code below should be shared with interpreter macro assembler implementation
|
|
1087 { Label unlock;
|
|
1088 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
|
|
1089 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
|
|
1090 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
|
|
1091
|
|
1092 __ leal(rdx, monitor); // address of first monitor
|
|
1093
|
|
1094 __ movl(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
|
|
1095 __ testl(t, t);
|
|
1096 __ jcc(Assembler::notZero, unlock);
|
|
1097
|
|
1098 // Entry already unlocked, need to throw exception
|
|
1099 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
|
|
1100 __ should_not_reach_here();
|
|
1101
|
|
1102 __ bind(unlock);
|
|
1103 __ unlock_object(rdx);
|
|
1104 }
|
|
1105 __ bind(L);
|
|
1106 }
|
|
1107
|
|
1108 // jvmti/dtrace support
|
|
1109 // Note: This must happen _after_ handling/throwing any exceptions since
|
|
1110 // the exception handler code notifies the runtime of method exits
|
|
1111 // too. If this happens before, method entry/exit notifications are
|
|
1112 // not properly paired (was bug - gri 11/22/99).
|
|
1113 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
|
|
1114
|
|
1115 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
|
|
1116 __ pop(ltos);
|
|
1117 __ movl(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
|
|
1118 __ call(t);
|
|
1119
|
|
1120 // remove activation
|
|
1121 __ movl(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
|
|
1122 __ leave(); // remove frame anchor
|
|
1123 __ popl(rdi); // get return address
|
|
1124 __ movl(rsp, t); // set sp to sender sp
|
|
1125 __ jmp(rdi);
|
|
1126
|
|
1127 if (inc_counter) {
|
|
1128 // Handle overflow of counter and compile method
|
|
1129 __ bind(invocation_counter_overflow);
|
|
1130 generate_counter_overflow(&continue_after_compile);
|
|
1131 }
|
|
1132
|
|
1133 return entry_point;
|
|
1134 }
|
|
1135
|
|
1136 //
|
|
1137 // Generic interpreted method entry to (asm) interpreter
|
|
1138 //
|
|
1139 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
|
|
1140 // determine code generation flags
|
|
1141 bool inc_counter = UseCompiler || CountCompiledCalls;
|
|
1142
|
|
1143 // rbx,: methodOop
|
|
1144 // rsi: sender sp
|
|
1145 address entry_point = __ pc();
|
|
1146
|
|
1147
|
|
1148 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
|
|
1149 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset());
|
|
1150 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
|
|
1151 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
|
|
1152
|
|
1153 // get parameter size (always needed)
|
|
1154 __ load_unsigned_word(rcx, size_of_parameters);
|
|
1155
|
|
1156 // rbx,: methodOop
|
|
1157 // rcx: size of parameters
|
|
1158
|
|
1159 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
|
|
1160
|
|
1161 __ load_unsigned_word(rdx, size_of_locals); // get size of locals in words
|
|
1162 __ subl(rdx, rcx); // rdx = no. of additional locals
|
|
1163
|
|
1164 // see if we've got enough room on the stack for locals plus overhead.
|
|
1165 generate_stack_overflow_check();
|
|
1166
|
|
1167 // get return address
|
|
1168 __ popl(rax);
|
|
1169
|
|
1170 // compute beginning of parameters (rdi)
|
|
1171 __ leal(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
|
|
1172
|
|
1173 // rdx - # of additional locals
|
|
1174 // allocate space for locals
|
|
1175 // explicitly initialize locals
|
|
1176 {
|
|
1177 Label exit, loop;
|
|
1178 __ testl(rdx, rdx);
|
|
1179 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
|
|
1180 __ bind(loop);
|
|
1181 if (TaggedStackInterpreter) __ pushl(NULL_WORD); // push tag
|
|
1182 __ pushl(NULL_WORD); // initialize local variables
|
|
1183 __ decrement(rdx); // until everything initialized
|
|
1184 __ jcc(Assembler::greater, loop);
|
|
1185 __ bind(exit);
|
|
1186 }
|
|
1187
|
|
1188 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
|
|
1189 // initialize fixed part of activation frame
|
|
1190 generate_fixed_frame(false);
|
|
1191
|
|
1192 // make sure method is not native & not abstract
|
|
1193 #ifdef ASSERT
|
|
1194 __ movl(rax, access_flags);
|
|
1195 {
|
|
1196 Label L;
|
|
1197 __ testl(rax, JVM_ACC_NATIVE);
|
|
1198 __ jcc(Assembler::zero, L);
|
|
1199 __ stop("tried to execute native method as non-native");
|
|
1200 __ bind(L);
|
|
1201 }
|
|
1202 { Label L;
|
|
1203 __ testl(rax, JVM_ACC_ABSTRACT);
|
|
1204 __ jcc(Assembler::zero, L);
|
|
1205 __ stop("tried to execute abstract method in interpreter");
|
|
1206 __ bind(L);
|
|
1207 }
|
|
1208 #endif
|
|
1209
|
|
1210 // Since at this point in the method invocation the exception handler
|
|
1211 // would try to exit the monitor of synchronized methods which hasn't
|
|
1212 // been entered yet, we set the thread local variable
|
|
1213 // _do_not_unlock_if_synchronized to true. The remove_activation will
|
|
1214 // check this flag.
|
|
1215
|
|
1216 __ get_thread(rax);
|
|
1217 const Address do_not_unlock_if_synchronized(rax,
|
|
1218 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
|
|
1219 __ movbool(do_not_unlock_if_synchronized, true);
|
|
1220
|
|
1221 // increment invocation count & check for overflow
|
|
1222 Label invocation_counter_overflow;
|
|
1223 Label profile_method;
|
|
1224 Label profile_method_continue;
|
|
1225 if (inc_counter) {
|
|
1226 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
|
|
1227 if (ProfileInterpreter) {
|
|
1228 __ bind(profile_method_continue);
|
|
1229 }
|
|
1230 }
|
|
1231 Label continue_after_compile;
|
|
1232 __ bind(continue_after_compile);
|
|
1233
|
|
1234 bang_stack_shadow_pages(false);
|
|
1235
|
|
1236 // reset the _do_not_unlock_if_synchronized flag
|
|
1237 __ get_thread(rax);
|
|
1238 __ movbool(do_not_unlock_if_synchronized, false);
|
|
1239
|
|
1240 // check for synchronized methods
|
|
1241 // Must happen AFTER invocation_counter check and stack overflow check,
|
|
1242 // so method is not locked if overflows.
|
|
1243 //
|
|
1244 if (synchronized) {
|
|
1245 // Allocate monitor and lock method
|
|
1246 lock_method();
|
|
1247 } else {
|
|
1248 // no synchronization necessary
|
|
1249 #ifdef ASSERT
|
|
1250 { Label L;
|
|
1251 __ movl(rax, access_flags);
|
|
1252 __ testl(rax, JVM_ACC_SYNCHRONIZED);
|
|
1253 __ jcc(Assembler::zero, L);
|
|
1254 __ stop("method needs synchronization");
|
|
1255 __ bind(L);
|
|
1256 }
|
|
1257 #endif
|
|
1258 }
|
|
1259
|
|
1260 // start execution
|
|
1261 #ifdef ASSERT
|
|
1262 { Label L;
|
|
1263 const Address monitor_block_top (rbp,
|
|
1264 frame::interpreter_frame_monitor_block_top_offset * wordSize);
|
|
1265 __ movl(rax, monitor_block_top);
|
|
1266 __ cmpl(rax, rsp);
|
|
1267 __ jcc(Assembler::equal, L);
|
|
1268 __ stop("broken stack frame setup in interpreter");
|
|
1269 __ bind(L);
|
|
1270 }
|
|
1271 #endif
|
|
1272
|
|
1273 // jvmti support
|
|
1274 __ notify_method_entry();
|
|
1275
|
|
1276 __ dispatch_next(vtos);
|
|
1277
|
|
1278 // invocation counter overflow
|
|
1279 if (inc_counter) {
|
|
1280 if (ProfileInterpreter) {
|
|
1281 // We have decided to profile this method in the interpreter
|
|
1282 __ bind(profile_method);
|
|
1283
|
|
1284 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi, true);
|
|
1285
|
|
1286 __ movl(rbx, Address(rbp, method_offset)); // restore methodOop
|
|
1287 __ movl(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
|
|
1288 __ movl(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
|
|
1289 __ test_method_data_pointer(rax, profile_method_continue);
|
|
1290 __ addl(rax, in_bytes(methodDataOopDesc::data_offset()));
|
|
1291 __ movl(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
|
|
1292 __ jmp(profile_method_continue);
|
|
1293 }
|
|
1294 // Handle overflow of counter and compile method
|
|
1295 __ bind(invocation_counter_overflow);
|
|
1296 generate_counter_overflow(&continue_after_compile);
|
|
1297 }
|
|
1298
|
|
1299 return entry_point;
|
|
1300 }
|
|
1301
|
|
1302 //------------------------------------------------------------------------------------------------------------------------
|
|
1303 // Entry points
|
|
1304 //
|
|
1305 // Here we generate the various kind of entries into the interpreter.
|
|
1306 // The two main entry type are generic bytecode methods and native call method.
|
|
1307 // These both come in synchronized and non-synchronized versions but the
|
|
1308 // frame layout they create is very similar. The other method entry
|
|
1309 // types are really just special purpose entries that are really entry
|
|
1310 // and interpretation all in one. These are for trivial methods like
|
|
1311 // accessor, empty, or special math methods.
|
|
1312 //
|
|
1313 // When control flow reaches any of the entry types for the interpreter
|
|
1314 // the following holds ->
|
|
1315 //
|
|
1316 // Arguments:
|
|
1317 //
|
|
1318 // rbx,: methodOop
|
|
1319 // rcx: receiver
|
|
1320 //
|
|
1321 //
|
|
1322 // Stack layout immediately at entry
|
|
1323 //
|
|
1324 // [ return address ] <--- rsp
|
|
1325 // [ parameter n ]
|
|
1326 // ...
|
|
1327 // [ parameter 1 ]
|
|
1328 // [ expression stack ] (caller's java expression stack)
|
|
1329
|
|
1330 // Assuming that we don't go to one of the trivial specialized
|
|
1331 // entries the stack will look like below when we are ready to execute
|
|
1332 // the first bytecode (or call the native routine). The register usage
|
|
1333 // will be as the template based interpreter expects (see interpreter_x86.hpp).
|
|
1334 //
|
|
1335 // local variables follow incoming parameters immediately; i.e.
|
|
1336 // the return address is moved to the end of the locals).
|
|
1337 //
|
|
1338 // [ monitor entry ] <--- rsp
|
|
1339 // ...
|
|
1340 // [ monitor entry ]
|
|
1341 // [ expr. stack bottom ]
|
|
1342 // [ saved rsi ]
|
|
1343 // [ current rdi ]
|
|
1344 // [ methodOop ]
|
|
1345 // [ saved rbp, ] <--- rbp,
|
|
1346 // [ return address ]
|
|
1347 // [ local variable m ]
|
|
1348 // ...
|
|
1349 // [ local variable 1 ]
|
|
1350 // [ parameter n ]
|
|
1351 // ...
|
|
1352 // [ parameter 1 ] <--- rdi
|
|
1353
|
|
1354 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
|
|
1355 // determine code generation flags
|
|
1356 bool synchronized = false;
|
|
1357 address entry_point = NULL;
|
|
1358
|
|
1359 switch (kind) {
|
|
1360 case Interpreter::zerolocals : break;
|
|
1361 case Interpreter::zerolocals_synchronized: synchronized = true; break;
|
|
1362 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
|
|
1363 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
|
|
1364 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
|
|
1365 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
|
|
1366 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
|
|
1367
|
|
1368 case Interpreter::java_lang_math_sin : // fall thru
|
|
1369 case Interpreter::java_lang_math_cos : // fall thru
|
|
1370 case Interpreter::java_lang_math_tan : // fall thru
|
|
1371 case Interpreter::java_lang_math_abs : // fall thru
|
|
1372 case Interpreter::java_lang_math_log : // fall thru
|
|
1373 case Interpreter::java_lang_math_log10 : // fall thru
|
|
1374 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
|
|
1375 default : ShouldNotReachHere(); break;
|
|
1376 }
|
|
1377
|
|
1378 if (entry_point) return entry_point;
|
|
1379
|
|
1380 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
|
|
1381
|
|
1382 }
|
|
1383
|
|
1384 // How much stack a method activation needs in words.
|
|
1385 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
|
|
1386
|
|
1387 const int stub_code = 4; // see generate_call_stub
|
|
1388 // Save space for one monitor to get into the interpreted method in case
|
|
1389 // the method is synchronized
|
|
1390 int monitor_size = method->is_synchronized() ?
|
|
1391 1*frame::interpreter_frame_monitor_size() : 0;
|
|
1392
|
|
1393 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
|
|
1394 // be sure to change this if you add/subtract anything to/from the overhead area
|
|
1395 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
|
|
1396
|
|
1397 const int method_stack = (method->max_locals() + method->max_stack()) *
|
|
1398 Interpreter::stackElementWords();
|
|
1399 return overhead_size + method_stack + stub_code;
|
|
1400 }
|
|
1401
|
|
1402 // asm based interpreter deoptimization helpers
|
|
1403
|
|
1404 int AbstractInterpreter::layout_activation(methodOop method,
|
|
1405 int tempcount,
|
|
1406 int popframe_extra_args,
|
|
1407 int moncount,
|
|
1408 int callee_param_count,
|
|
1409 int callee_locals,
|
|
1410 frame* caller,
|
|
1411 frame* interpreter_frame,
|
|
1412 bool is_top_frame) {
|
|
1413 // Note: This calculation must exactly parallel the frame setup
|
|
1414 // in AbstractInterpreterGenerator::generate_method_entry.
|
|
1415 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
|
|
1416 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
|
|
1417 // as determined by a previous call to this method.
|
|
1418 // It is also guaranteed to be walkable even though it is in a skeletal state
|
|
1419 // NOTE: return size is in words not bytes
|
|
1420
|
|
1421 // fixed size of an interpreter frame:
|
|
1422 int max_locals = method->max_locals() * Interpreter::stackElementWords();
|
|
1423 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
|
|
1424 Interpreter::stackElementWords();
|
|
1425
|
|
1426 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
|
|
1427
|
|
1428 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
|
|
1429 // Since the callee parameters already account for the callee's params we only need to account for
|
|
1430 // the extra locals.
|
|
1431
|
|
1432
|
|
1433 int size = overhead +
|
|
1434 ((callee_locals - callee_param_count)*Interpreter::stackElementWords()) +
|
|
1435 (moncount*frame::interpreter_frame_monitor_size()) +
|
|
1436 tempcount*Interpreter::stackElementWords() + popframe_extra_args;
|
|
1437
|
|
1438 if (interpreter_frame != NULL) {
|
|
1439 #ifdef ASSERT
|
|
1440 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
|
|
1441 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
|
|
1442 #endif
|
|
1443
|
|
1444 interpreter_frame->interpreter_frame_set_method(method);
|
|
1445 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
|
|
1446 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
|
|
1447 // and sender_sp is fp+8
|
|
1448 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
|
|
1449
|
|
1450 interpreter_frame->interpreter_frame_set_locals(locals);
|
|
1451 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
|
|
1452 BasicObjectLock* monbot = montop - moncount;
|
|
1453 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
|
|
1454
|
|
1455 // Set last_sp
|
|
1456 intptr_t* rsp = (intptr_t*) monbot -
|
|
1457 tempcount*Interpreter::stackElementWords() -
|
|
1458 popframe_extra_args;
|
|
1459 interpreter_frame->interpreter_frame_set_last_sp(rsp);
|
|
1460
|
|
1461 // All frames but the initial (oldest) interpreter frame we fill in have a
|
|
1462 // value for sender_sp that allows walking the stack but isn't
|
|
1463 // truly correct. Correct the value here.
|
|
1464
|
|
1465 if (extra_locals != 0 &&
|
|
1466 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
|
|
1467 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
|
|
1468 }
|
|
1469 *interpreter_frame->interpreter_frame_cache_addr() =
|
|
1470 method->constants()->cache();
|
|
1471 }
|
|
1472 return size;
|
|
1473 }
|
|
1474
|
|
1475
|
|
1476 //------------------------------------------------------------------------------------------------------------------------
|
|
1477 // Exceptions
|
|
1478
|
|
1479 void TemplateInterpreterGenerator::generate_throw_exception() {
|
|
1480 // Entry point in previous activation (i.e., if the caller was interpreted)
|
|
1481 Interpreter::_rethrow_exception_entry = __ pc();
|
|
1482
|
|
1483 // Restore sp to interpreter_frame_last_sp even though we are going
|
|
1484 // to empty the expression stack for the exception processing.
|
|
1485 __ movl(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
|
|
1486 // rax,: exception
|
|
1487 // rdx: return address/pc that threw exception
|
|
1488 __ restore_bcp(); // rsi points to call/send
|
|
1489 __ restore_locals();
|
|
1490
|
|
1491 // Entry point for exceptions thrown within interpreter code
|
|
1492 Interpreter::_throw_exception_entry = __ pc();
|
|
1493 // expression stack is undefined here
|
|
1494 // rax,: exception
|
|
1495 // rsi: exception bcp
|
|
1496 __ verify_oop(rax);
|
|
1497
|
|
1498 // expression stack must be empty before entering the VM in case of an exception
|
|
1499 __ empty_expression_stack();
|
|
1500 __ empty_FPU_stack();
|
|
1501 // find exception handler address and preserve exception oop
|
|
1502 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
|
|
1503 // rax,: exception handler entry point
|
|
1504 // rdx: preserved exception oop
|
|
1505 // rsi: bcp for exception handler
|
|
1506 __ push_ptr(rdx); // push exception which is now the only value on the stack
|
|
1507 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
|
|
1508
|
|
1509 // If the exception is not handled in the current frame the frame is removed and
|
|
1510 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
|
|
1511 //
|
|
1512 // Note: At this point the bci is still the bxi for the instruction which caused
|
|
1513 // the exception and the expression stack is empty. Thus, for any VM calls
|
|
1514 // at this point, GC will find a legal oop map (with empty expression stack).
|
|
1515
|
|
1516 // In current activation
|
|
1517 // tos: exception
|
|
1518 // rsi: exception bcp
|
|
1519
|
|
1520 //
|
|
1521 // JVMTI PopFrame support
|
|
1522 //
|
|
1523
|
|
1524 Interpreter::_remove_activation_preserving_args_entry = __ pc();
|
|
1525 __ empty_expression_stack();
|
|
1526 __ empty_FPU_stack();
|
|
1527 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
|
|
1528 // currently handling popframe, so that call_VMs that may happen later do not trigger new
|
|
1529 // popframe handling cycles.
|
|
1530 __ get_thread(rcx);
|
|
1531 __ movl(rdx, Address(rcx, JavaThread::popframe_condition_offset()));
|
|
1532 __ orl(rdx, JavaThread::popframe_processing_bit);
|
|
1533 __ movl(Address(rcx, JavaThread::popframe_condition_offset()), rdx);
|
|
1534
|
|
1535 {
|
|
1536 // Check to see whether we are returning to a deoptimized frame.
|
|
1537 // (The PopFrame call ensures that the caller of the popped frame is
|
|
1538 // either interpreted or compiled and deoptimizes it if compiled.)
|
|
1539 // In this case, we can't call dispatch_next() after the frame is
|
|
1540 // popped, but instead must save the incoming arguments and restore
|
|
1541 // them after deoptimization has occurred.
|
|
1542 //
|
|
1543 // Note that we don't compare the return PC against the
|
|
1544 // deoptimization blob's unpack entry because of the presence of
|
|
1545 // adapter frames in C2.
|
|
1546 Label caller_not_deoptimized;
|
|
1547 __ movl(rdx, Address(rbp, frame::return_addr_offset * wordSize));
|
|
1548 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
|
|
1549 __ testl(rax, rax);
|
|
1550 __ jcc(Assembler::notZero, caller_not_deoptimized);
|
|
1551
|
|
1552 // Compute size of arguments for saving when returning to deoptimized caller
|
|
1553 __ get_method(rax);
|
|
1554 __ verify_oop(rax);
|
|
1555 __ load_unsigned_word(rax, Address(rax, in_bytes(methodOopDesc::size_of_parameters_offset())));
|
|
1556 __ shll(rax, Interpreter::logStackElementSize());
|
|
1557 __ restore_locals();
|
|
1558 __ subl(rdi, rax);
|
|
1559 __ addl(rdi, wordSize);
|
|
1560 // Save these arguments
|
|
1561 __ get_thread(rcx);
|
|
1562 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), rcx, rax, rdi);
|
|
1563
|
|
1564 __ remove_activation(vtos, rdx,
|
|
1565 /* throw_monitor_exception */ false,
|
|
1566 /* install_monitor_exception */ false,
|
|
1567 /* notify_jvmdi */ false);
|
|
1568
|
|
1569 // Inform deoptimization that it is responsible for restoring these arguments
|
|
1570 __ get_thread(rcx);
|
|
1571 __ movl(Address(rcx, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
|
|
1572
|
|
1573 // Continue in deoptimization handler
|
|
1574 __ jmp(rdx);
|
|
1575
|
|
1576 __ bind(caller_not_deoptimized);
|
|
1577 }
|
|
1578
|
|
1579 __ remove_activation(vtos, rdx,
|
|
1580 /* throw_monitor_exception */ false,
|
|
1581 /* install_monitor_exception */ false,
|
|
1582 /* notify_jvmdi */ false);
|
|
1583
|
|
1584 // Finish with popframe handling
|
|
1585 // A previous I2C followed by a deoptimization might have moved the
|
|
1586 // outgoing arguments further up the stack. PopFrame expects the
|
|
1587 // mutations to those outgoing arguments to be preserved and other
|
|
1588 // constraints basically require this frame to look exactly as
|
|
1589 // though it had previously invoked an interpreted activation with
|
|
1590 // no space between the top of the expression stack (current
|
|
1591 // last_sp) and the top of stack. Rather than force deopt to
|
|
1592 // maintain this kind of invariant all the time we call a small
|
|
1593 // fixup routine to move the mutated arguments onto the top of our
|
|
1594 // expression stack if necessary.
|
|
1595 __ movl(rax, rsp);
|
|
1596 __ movl(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
|
|
1597 __ get_thread(rcx);
|
|
1598 // PC must point into interpreter here
|
|
1599 __ set_last_Java_frame(rcx, noreg, rbp, __ pc());
|
|
1600 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), rcx, rax, rbx);
|
|
1601 __ get_thread(rcx);
|
|
1602 __ reset_last_Java_frame(rcx, true, true);
|
|
1603 // Restore the last_sp and null it out
|
|
1604 __ movl(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
|
|
1605 __ movl(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
|
|
1606
|
|
1607 __ restore_bcp();
|
|
1608 __ restore_locals();
|
|
1609 // The method data pointer was incremented already during
|
|
1610 // call profiling. We have to restore the mdp for the current bcp.
|
|
1611 if (ProfileInterpreter) {
|
|
1612 __ set_method_data_pointer_for_bcp();
|
|
1613 }
|
|
1614
|
|
1615 // Clear the popframe condition flag
|
|
1616 __ get_thread(rcx);
|
|
1617 __ movl(Address(rcx, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
|
|
1618
|
|
1619 __ dispatch_next(vtos);
|
|
1620 // end of PopFrame support
|
|
1621
|
|
1622 Interpreter::_remove_activation_entry = __ pc();
|
|
1623
|
|
1624 // preserve exception over this code sequence
|
|
1625 __ pop_ptr(rax);
|
|
1626 __ get_thread(rcx);
|
|
1627 __ movl(Address(rcx, JavaThread::vm_result_offset()), rax);
|
|
1628 // remove the activation (without doing throws on illegalMonitorExceptions)
|
|
1629 __ remove_activation(vtos, rdx, false, true, false);
|
|
1630 // restore exception
|
|
1631 __ get_thread(rcx);
|
|
1632 __ movl(rax, Address(rcx, JavaThread::vm_result_offset()));
|
|
1633 __ movl(Address(rcx, JavaThread::vm_result_offset()), NULL_WORD);
|
|
1634 __ verify_oop(rax);
|
|
1635
|
|
1636 // Inbetween activations - previous activation type unknown yet
|
|
1637 // compute continuation point - the continuation point expects
|
|
1638 // the following registers set up:
|
|
1639 //
|
|
1640 // rax,: exception
|
|
1641 // rdx: return address/pc that threw exception
|
|
1642 // rsp: expression stack of caller
|
|
1643 // rbp,: rbp, of caller
|
|
1644 __ pushl(rax); // save exception
|
|
1645 __ pushl(rdx); // save return address
|
|
1646 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rdx);
|
|
1647 __ movl(rbx, rax); // save exception handler
|
|
1648 __ popl(rdx); // restore return address
|
|
1649 __ popl(rax); // restore exception
|
|
1650 // Note that an "issuing PC" is actually the next PC after the call
|
|
1651 __ jmp(rbx); // jump to exception handler of caller
|
|
1652 }
|
|
1653
|
|
1654
|
|
1655 //
|
|
1656 // JVMTI ForceEarlyReturn support
|
|
1657 //
|
|
1658 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
|
|
1659 address entry = __ pc();
|
|
1660
|
|
1661 __ restore_bcp();
|
|
1662 __ restore_locals();
|
|
1663 __ empty_expression_stack();
|
|
1664 __ empty_FPU_stack();
|
|
1665 __ load_earlyret_value(state);
|
|
1666
|
|
1667 __ get_thread(rcx);
|
|
1668 __ movl(rcx, Address(rcx, JavaThread::jvmti_thread_state_offset()));
|
|
1669 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
|
|
1670
|
|
1671 // Clear the earlyret state
|
|
1672 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
|
|
1673
|
|
1674 __ remove_activation(state, rsi,
|
|
1675 false, /* throw_monitor_exception */
|
|
1676 false, /* install_monitor_exception */
|
|
1677 true); /* notify_jvmdi */
|
|
1678 __ jmp(rsi);
|
|
1679 return entry;
|
|
1680 } // end of ForceEarlyReturn support
|
|
1681
|
|
1682
|
|
1683 //------------------------------------------------------------------------------------------------------------------------
|
|
1684 // Helper for vtos entry point generation
|
|
1685
|
|
1686 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) {
|
|
1687 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
|
|
1688 Label L;
|
|
1689 fep = __ pc(); __ push(ftos); __ jmp(L);
|
|
1690 dep = __ pc(); __ push(dtos); __ jmp(L);
|
|
1691 lep = __ pc(); __ push(ltos); __ jmp(L);
|
|
1692 aep = __ pc(); __ push(atos); __ jmp(L);
|
|
1693 bep = cep = sep = // fall through
|
|
1694 iep = __ pc(); __ push(itos); // fall through
|
|
1695 vep = __ pc(); __ bind(L); // fall through
|
|
1696 generate_and_dispatch(t);
|
|
1697 }
|
|
1698
|
|
1699 //------------------------------------------------------------------------------------------------------------------------
|
|
1700 // Generation of individual instructions
|
|
1701
|
|
1702 // helpers for generate_and_dispatch
|
|
1703
|
|
1704
|
|
1705
|
|
1706 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
|
|
1707 : TemplateInterpreterGenerator(code) {
|
|
1708 generate_all(); // down here so it can be "virtual"
|
|
1709 }
|
|
1710
|
|
1711 //------------------------------------------------------------------------------------------------------------------------
|
|
1712
|
|
1713 // Non-product code
|
|
1714 #ifndef PRODUCT
|
|
1715 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
|
|
1716 address entry = __ pc();
|
|
1717
|
|
1718 // prepare expression stack
|
|
1719 __ popl(rcx); // pop return address so expression stack is 'pure'
|
|
1720 __ push(state); // save tosca
|
|
1721
|
|
1722 // pass tosca registers as arguments & call tracer
|
|
1723 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
|
|
1724 __ movl(rcx, rax); // make sure return address is not destroyed by pop(state)
|
|
1725 __ pop(state); // restore tosca
|
|
1726
|
|
1727 // return
|
|
1728 __ jmp(rcx);
|
|
1729
|
|
1730 return entry;
|
|
1731 }
|
|
1732
|
|
1733
|
|
1734 void TemplateInterpreterGenerator::count_bytecode() {
|
|
1735 __ increment(ExternalAddress((address) &BytecodeCounter::_counter_value));
|
|
1736 }
|
|
1737
|
|
1738
|
|
1739 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
|
|
1740 __ increment(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
|
|
1741 }
|
|
1742
|
|
1743
|
|
1744 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
|
|
1745 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
|
|
1746 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
|
|
1747 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
|
|
1748 ExternalAddress table((address) BytecodePairHistogram::_counters);
|
|
1749 Address index(noreg, rbx, Address::times_4);
|
|
1750 __ increment(ArrayAddress(table, index));
|
|
1751 }
|
|
1752
|
|
1753
|
|
1754 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
|
|
1755 // Call a little run-time stub to avoid blow-up for each bytecode.
|
|
1756 // The run-time runtime saves the right registers, depending on
|
|
1757 // the tosca in-state for the given template.
|
|
1758 assert(Interpreter::trace_code(t->tos_in()) != NULL,
|
|
1759 "entry must have been generated");
|
|
1760 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
|
|
1761 }
|
|
1762
|
|
1763
|
|
1764 void TemplateInterpreterGenerator::stop_interpreter_at() {
|
|
1765 Label L;
|
|
1766 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
|
|
1767 StopInterpreterAt);
|
|
1768 __ jcc(Assembler::notEqual, L);
|
|
1769 __ int3();
|
|
1770 __ bind(L);
|
|
1771 }
|
|
1772 #endif // !PRODUCT
|
|
1773 #endif // CC_INTERP
|