|
1 /* |
|
2 * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved. |
|
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
|
4 * |
|
5 * This code is free software; you can redistribute it and/or modify it |
|
6 * under the terms of the GNU General Public License version 2 only, as |
|
7 * published by the Free Software Foundation. |
|
8 * |
|
9 * This code is distributed in the hope that it will be useful, but WITHOUT |
|
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
12 * version 2 for more details (a copy is included in the LICENSE file that |
|
13 * accompanied this code). |
|
14 * |
|
15 * You should have received a copy of the GNU General Public License version |
|
16 * 2 along with this work; if not, write to the Free Software Foundation, |
|
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
|
18 * |
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
20 * or visit www.oracle.com if you need additional information or have any |
|
21 * questions. |
|
22 * |
|
23 */ |
|
24 |
|
25 #include "precompiled.hpp" |
|
26 #include "asm/macroAssembler.hpp" |
|
27 #include "interpreter/bytecodeHistogram.hpp" |
|
28 #include "interpreter/interpreter.hpp" |
|
29 #include "interpreter/interpreterGenerator.hpp" |
|
30 #include "interpreter/interpreterRuntime.hpp" |
|
31 #include "interpreter/interp_masm.hpp" |
|
32 #include "interpreter/templateTable.hpp" |
|
33 #include "oops/arrayOop.hpp" |
|
34 #include "oops/methodData.hpp" |
|
35 #include "oops/method.hpp" |
|
36 #include "oops/oop.inline.hpp" |
|
37 #include "prims/jvmtiExport.hpp" |
|
38 #include "prims/jvmtiThreadState.hpp" |
|
39 #include "runtime/arguments.hpp" |
|
40 #include "runtime/deoptimization.hpp" |
|
41 #include "runtime/frame.inline.hpp" |
|
42 #include "runtime/sharedRuntime.hpp" |
|
43 #include "runtime/stubRoutines.hpp" |
|
44 #include "runtime/synchronizer.hpp" |
|
45 #include "runtime/timer.hpp" |
|
46 #include "runtime/vframeArray.hpp" |
|
47 #include "utilities/debug.hpp" |
|
48 #include "utilities/macros.hpp" |
|
49 |
|
50 #ifndef CC_INTERP |
|
51 #ifndef FAST_DISPATCH |
|
52 #define FAST_DISPATCH 1 |
|
53 #endif |
|
54 #undef FAST_DISPATCH |
|
55 |
|
56 |
|
57 // Generation of Interpreter |
|
58 // |
|
59 // The InterpreterGenerator generates the interpreter into Interpreter::_code. |
|
60 |
|
61 |
|
62 #define __ _masm-> |
|
63 |
|
64 |
|
65 //---------------------------------------------------------------------------------------------------- |
|
66 |
|
67 |
|
68 void InterpreterGenerator::save_native_result(void) { |
|
69 // result potentially in O0/O1: save it across calls |
|
70 const Address& l_tmp = InterpreterMacroAssembler::l_tmp; |
|
71 |
|
72 // result potentially in F0/F1: save it across calls |
|
73 const Address& d_tmp = InterpreterMacroAssembler::d_tmp; |
|
74 |
|
75 // save and restore any potential method result value around the unlocking operation |
|
76 __ stf(FloatRegisterImpl::D, F0, d_tmp); |
|
77 #ifdef _LP64 |
|
78 __ stx(O0, l_tmp); |
|
79 #else |
|
80 __ std(O0, l_tmp); |
|
81 #endif |
|
82 } |
|
83 |
|
84 void InterpreterGenerator::restore_native_result(void) { |
|
85 const Address& l_tmp = InterpreterMacroAssembler::l_tmp; |
|
86 const Address& d_tmp = InterpreterMacroAssembler::d_tmp; |
|
87 |
|
88 // Restore any method result value |
|
89 __ ldf(FloatRegisterImpl::D, d_tmp, F0); |
|
90 #ifdef _LP64 |
|
91 __ ldx(l_tmp, O0); |
|
92 #else |
|
93 __ ldd(l_tmp, O0); |
|
94 #endif |
|
95 } |
|
96 |
|
97 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) { |
|
98 assert(!pass_oop || message == NULL, "either oop or message but not both"); |
|
99 address entry = __ pc(); |
|
100 // expression stack must be empty before entering the VM if an exception happened |
|
101 __ empty_expression_stack(); |
|
102 // load exception object |
|
103 __ set((intptr_t)name, G3_scratch); |
|
104 if (pass_oop) { |
|
105 __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), G3_scratch, Otos_i); |
|
106 } else { |
|
107 __ set((intptr_t)message, G4_scratch); |
|
108 __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), G3_scratch, G4_scratch); |
|
109 } |
|
110 // throw exception |
|
111 assert(Interpreter::throw_exception_entry() != NULL, "generate it first"); |
|
112 AddressLiteral thrower(Interpreter::throw_exception_entry()); |
|
113 __ jump_to(thrower, G3_scratch); |
|
114 __ delayed()->nop(); |
|
115 return entry; |
|
116 } |
|
117 |
|
118 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { |
|
119 address entry = __ pc(); |
|
120 // expression stack must be empty before entering the VM if an exception |
|
121 // happened |
|
122 __ empty_expression_stack(); |
|
123 // load exception object |
|
124 __ call_VM(Oexception, |
|
125 CAST_FROM_FN_PTR(address, |
|
126 InterpreterRuntime::throw_ClassCastException), |
|
127 Otos_i); |
|
128 __ should_not_reach_here(); |
|
129 return entry; |
|
130 } |
|
131 |
|
132 |
|
133 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) { |
|
134 address entry = __ pc(); |
|
135 // expression stack must be empty before entering the VM if an exception happened |
|
136 __ empty_expression_stack(); |
|
137 // convention: expect aberrant index in register G3_scratch, then shuffle the |
|
138 // index to G4_scratch for the VM call |
|
139 __ mov(G3_scratch, G4_scratch); |
|
140 __ set((intptr_t)name, G3_scratch); |
|
141 __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), G3_scratch, G4_scratch); |
|
142 __ should_not_reach_here(); |
|
143 return entry; |
|
144 } |
|
145 |
|
146 |
|
147 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { |
|
148 address entry = __ pc(); |
|
149 // expression stack must be empty before entering the VM if an exception happened |
|
150 __ empty_expression_stack(); |
|
151 __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError)); |
|
152 __ should_not_reach_here(); |
|
153 return entry; |
|
154 } |
|
155 |
|
156 |
|
157 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) { |
|
158 address entry = __ pc(); |
|
159 |
|
160 if (state == atos) { |
|
161 __ profile_return_type(O0, G3_scratch, G1_scratch); |
|
162 } |
|
163 |
|
164 #if !defined(_LP64) && defined(COMPILER2) |
|
165 // All return values are where we want them, except for Longs. C2 returns |
|
166 // longs in G1 in the 32-bit build whereas the interpreter wants them in O0/O1. |
|
167 // Since the interpreter will return longs in G1 and O0/O1 in the 32bit |
|
168 // build even if we are returning from interpreted we just do a little |
|
169 // stupid shuffing. |
|
170 // Note: I tried to make c2 return longs in O0/O1 and G1 so we wouldn't have to |
|
171 // do this here. Unfortunately if we did a rethrow we'd see an machepilog node |
|
172 // first which would move g1 -> O0/O1 and destroy the exception we were throwing. |
|
173 |
|
174 if (state == ltos) { |
|
175 __ srl (G1, 0, O1); |
|
176 __ srlx(G1, 32, O0); |
|
177 } |
|
178 #endif // !_LP64 && COMPILER2 |
|
179 |
|
180 // The callee returns with the stack possibly adjusted by adapter transition |
|
181 // We remove that possible adjustment here. |
|
182 // All interpreter local registers are untouched. Any result is passed back |
|
183 // in the O0/O1 or float registers. Before continuing, the arguments must be |
|
184 // popped from the java expression stack; i.e., Lesp must be adjusted. |
|
185 |
|
186 __ mov(Llast_SP, SP); // Remove any adapter added stack space. |
|
187 |
|
188 const Register cache = G3_scratch; |
|
189 const Register index = G1_scratch; |
|
190 __ get_cache_and_index_at_bcp(cache, index, 1, index_size); |
|
191 |
|
192 const Register flags = cache; |
|
193 __ ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset(), flags); |
|
194 const Register parameter_size = flags; |
|
195 __ and3(flags, ConstantPoolCacheEntry::parameter_size_mask, parameter_size); // argument size in words |
|
196 __ sll(parameter_size, Interpreter::logStackElementSize, parameter_size); // each argument size in bytes |
|
197 __ add(Lesp, parameter_size, Lesp); // pop arguments |
|
198 __ dispatch_next(state, step); |
|
199 |
|
200 return entry; |
|
201 } |
|
202 |
|
203 |
|
204 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) { |
|
205 address entry = __ pc(); |
|
206 __ get_constant_pool_cache(LcpoolCache); // load LcpoolCache |
|
207 #if INCLUDE_JVMCI |
|
208 // Check if we need to take lock at entry of synchronized method. |
|
209 if (UseJVMCICompiler) { |
|
210 Label L; |
|
211 Address pending_monitor_enter_addr(G2_thread, JavaThread::pending_monitorenter_offset()); |
|
212 __ ldbool(pending_monitor_enter_addr, Gtemp); // Load if pending monitor enter |
|
213 __ cmp_and_br_short(Gtemp, G0, Assembler::equal, Assembler::pn, L); |
|
214 // Clear flag. |
|
215 __ stbool(G0, pending_monitor_enter_addr); |
|
216 // Take lock. |
|
217 lock_method(); |
|
218 __ bind(L); |
|
219 } |
|
220 #endif |
|
221 { Label L; |
|
222 Address exception_addr(G2_thread, Thread::pending_exception_offset()); |
|
223 __ ld_ptr(exception_addr, Gtemp); // Load pending exception. |
|
224 __ br_null_short(Gtemp, Assembler::pt, L); |
|
225 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception)); |
|
226 __ should_not_reach_here(); |
|
227 __ bind(L); |
|
228 } |
|
229 __ dispatch_next(state, step); |
|
230 return entry; |
|
231 } |
|
232 |
|
233 // A result handler converts/unboxes a native call result into |
|
234 // a java interpreter/compiler result. The current frame is an |
|
235 // interpreter frame. The activation frame unwind code must be |
|
236 // consistent with that of TemplateTable::_return(...). In the |
|
237 // case of native methods, the caller's SP was not modified. |
|
238 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) { |
|
239 address entry = __ pc(); |
|
240 Register Itos_i = Otos_i ->after_save(); |
|
241 Register Itos_l = Otos_l ->after_save(); |
|
242 Register Itos_l1 = Otos_l1->after_save(); |
|
243 Register Itos_l2 = Otos_l2->after_save(); |
|
244 switch (type) { |
|
245 case T_BOOLEAN: __ subcc(G0, O0, G0); __ addc(G0, 0, Itos_i); break; // !0 => true; 0 => false |
|
246 case T_CHAR : __ sll(O0, 16, O0); __ srl(O0, 16, Itos_i); break; // cannot use and3, 0xFFFF too big as immediate value! |
|
247 case T_BYTE : __ sll(O0, 24, O0); __ sra(O0, 24, Itos_i); break; |
|
248 case T_SHORT : __ sll(O0, 16, O0); __ sra(O0, 16, Itos_i); break; |
|
249 case T_LONG : |
|
250 #ifndef _LP64 |
|
251 __ mov(O1, Itos_l2); // move other half of long |
|
252 #endif // ifdef or no ifdef, fall through to the T_INT case |
|
253 case T_INT : __ mov(O0, Itos_i); break; |
|
254 case T_VOID : /* nothing to do */ break; |
|
255 case T_FLOAT : assert(F0 == Ftos_f, "fix this code" ); break; |
|
256 case T_DOUBLE : assert(F0 == Ftos_d, "fix this code" ); break; |
|
257 case T_OBJECT : |
|
258 __ ld_ptr(FP, (frame::interpreter_frame_oop_temp_offset*wordSize) + STACK_BIAS, Itos_i); |
|
259 __ verify_oop(Itos_i); |
|
260 break; |
|
261 default : ShouldNotReachHere(); |
|
262 } |
|
263 __ ret(); // return from interpreter activation |
|
264 __ delayed()->restore(I5_savedSP, G0, SP); // remove interpreter frame |
|
265 NOT_PRODUCT(__ emit_int32(0);) // marker for disassembly |
|
266 return entry; |
|
267 } |
|
268 |
|
269 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) { |
|
270 address entry = __ pc(); |
|
271 __ push(state); |
|
272 __ call_VM(noreg, runtime_entry); |
|
273 __ dispatch_via(vtos, Interpreter::normal_table(vtos)); |
|
274 return entry; |
|
275 } |
|
276 |
|
277 |
|
278 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) { |
|
279 address entry = __ pc(); |
|
280 __ dispatch_next(state); |
|
281 return entry; |
|
282 } |
|
283 |
|
284 // |
|
285 // Helpers for commoning out cases in the various type of method entries. |
|
286 // |
|
287 |
|
288 // increment invocation count & check for overflow |
|
289 // |
|
290 // Note: checking for negative value instead of overflow |
|
291 // so we have a 'sticky' overflow test |
|
292 // |
|
293 // Lmethod: method |
|
294 // ??: invocation counter |
|
295 // |
|
296 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { |
|
297 // Note: In tiered we increment either counters in MethodCounters* or in |
|
298 // MDO depending if we're profiling or not. |
|
299 const Register G3_method_counters = G3_scratch; |
|
300 Label done; |
|
301 |
|
302 if (TieredCompilation) { |
|
303 const int increment = InvocationCounter::count_increment; |
|
304 Label no_mdo; |
|
305 if (ProfileInterpreter) { |
|
306 // If no method data exists, go to profile_continue. |
|
307 __ ld_ptr(Lmethod, Method::method_data_offset(), G4_scratch); |
|
308 __ br_null_short(G4_scratch, Assembler::pn, no_mdo); |
|
309 // Increment counter |
|
310 Address mdo_invocation_counter(G4_scratch, |
|
311 in_bytes(MethodData::invocation_counter_offset()) + |
|
312 in_bytes(InvocationCounter::counter_offset())); |
|
313 Address mask(G4_scratch, in_bytes(MethodData::invoke_mask_offset())); |
|
314 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, |
|
315 G3_scratch, Lscratch, |
|
316 Assembler::zero, overflow); |
|
317 __ ba_short(done); |
|
318 } |
|
319 |
|
320 // Increment counter in MethodCounters* |
|
321 __ bind(no_mdo); |
|
322 Address invocation_counter(G3_method_counters, |
|
323 in_bytes(MethodCounters::invocation_counter_offset()) + |
|
324 in_bytes(InvocationCounter::counter_offset())); |
|
325 __ get_method_counters(Lmethod, G3_method_counters, done); |
|
326 Address mask(G3_method_counters, in_bytes(MethodCounters::invoke_mask_offset())); |
|
327 __ increment_mask_and_jump(invocation_counter, increment, mask, |
|
328 G4_scratch, Lscratch, |
|
329 Assembler::zero, overflow); |
|
330 __ bind(done); |
|
331 } else { // not TieredCompilation |
|
332 // Update standard invocation counters |
|
333 __ get_method_counters(Lmethod, G3_method_counters, done); |
|
334 __ increment_invocation_counter(G3_method_counters, O0, G4_scratch); |
|
335 if (ProfileInterpreter) { |
|
336 Address interpreter_invocation_counter(G3_method_counters, |
|
337 in_bytes(MethodCounters::interpreter_invocation_counter_offset())); |
|
338 __ ld(interpreter_invocation_counter, G4_scratch); |
|
339 __ inc(G4_scratch); |
|
340 __ st(G4_scratch, interpreter_invocation_counter); |
|
341 } |
|
342 |
|
343 if (ProfileInterpreter && profile_method != NULL) { |
|
344 // Test to see if we should create a method data oop |
|
345 Address profile_limit(G3_method_counters, in_bytes(MethodCounters::interpreter_profile_limit_offset())); |
|
346 __ ld(profile_limit, G1_scratch); |
|
347 __ cmp_and_br_short(O0, G1_scratch, Assembler::lessUnsigned, Assembler::pn, *profile_method_continue); |
|
348 |
|
349 // if no method data exists, go to profile_method |
|
350 __ test_method_data_pointer(*profile_method); |
|
351 } |
|
352 |
|
353 Address invocation_limit(G3_method_counters, in_bytes(MethodCounters::interpreter_invocation_limit_offset())); |
|
354 __ ld(invocation_limit, G3_scratch); |
|
355 __ cmp(O0, G3_scratch); |
|
356 __ br(Assembler::greaterEqualUnsigned, false, Assembler::pn, *overflow); // Far distance |
|
357 __ delayed()->nop(); |
|
358 __ bind(done); |
|
359 } |
|
360 |
|
361 } |
|
362 |
|
363 // Allocate monitor and lock method (asm interpreter) |
|
364 // ebx - Method* |
|
365 // |
|
366 void TemplateInterpreterGenerator::lock_method() { |
|
367 __ ld(Lmethod, in_bytes(Method::access_flags_offset()), O0); // Load access flags. |
|
368 |
|
369 #ifdef ASSERT |
|
370 { Label ok; |
|
371 __ btst(JVM_ACC_SYNCHRONIZED, O0); |
|
372 __ br( Assembler::notZero, false, Assembler::pt, ok); |
|
373 __ delayed()->nop(); |
|
374 __ stop("method doesn't need synchronization"); |
|
375 __ bind(ok); |
|
376 } |
|
377 #endif // ASSERT |
|
378 |
|
379 // get synchronization object to O0 |
|
380 { Label done; |
|
381 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); |
|
382 __ btst(JVM_ACC_STATIC, O0); |
|
383 __ br( Assembler::zero, true, Assembler::pt, done); |
|
384 __ delayed()->ld_ptr(Llocals, Interpreter::local_offset_in_bytes(0), O0); // get receiver for not-static case |
|
385 |
|
386 __ ld_ptr( Lmethod, in_bytes(Method::const_offset()), O0); |
|
387 __ ld_ptr( O0, in_bytes(ConstMethod::constants_offset()), O0); |
|
388 __ ld_ptr( O0, ConstantPool::pool_holder_offset_in_bytes(), O0); |
|
389 |
|
390 // lock the mirror, not the Klass* |
|
391 __ ld_ptr( O0, mirror_offset, O0); |
|
392 |
|
393 #ifdef ASSERT |
|
394 __ tst(O0); |
|
395 __ breakpoint_trap(Assembler::zero, Assembler::ptr_cc); |
|
396 #endif // ASSERT |
|
397 |
|
398 __ bind(done); |
|
399 } |
|
400 |
|
401 __ add_monitor_to_stack(true, noreg, noreg); // allocate monitor elem |
|
402 __ st_ptr( O0, Lmonitors, BasicObjectLock::obj_offset_in_bytes()); // store object |
|
403 // __ untested("lock_object from method entry"); |
|
404 __ lock_object(Lmonitors, O0); |
|
405 } |
|
406 |
|
407 |
|
408 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rframe_size, |
|
409 Register Rscratch, |
|
410 Register Rscratch2) { |
|
411 const int page_size = os::vm_page_size(); |
|
412 Label after_frame_check; |
|
413 |
|
414 assert_different_registers(Rframe_size, Rscratch, Rscratch2); |
|
415 |
|
416 __ set(page_size, Rscratch); |
|
417 __ cmp_and_br_short(Rframe_size, Rscratch, Assembler::lessEqual, Assembler::pt, after_frame_check); |
|
418 |
|
419 // get the stack base, and in debug, verify it is non-zero |
|
420 __ ld_ptr( G2_thread, Thread::stack_base_offset(), Rscratch ); |
|
421 #ifdef ASSERT |
|
422 Label base_not_zero; |
|
423 __ br_notnull_short(Rscratch, Assembler::pn, base_not_zero); |
|
424 __ stop("stack base is zero in generate_stack_overflow_check"); |
|
425 __ bind(base_not_zero); |
|
426 #endif |
|
427 |
|
428 // get the stack size, and in debug, verify it is non-zero |
|
429 assert( sizeof(size_t) == sizeof(intptr_t), "wrong load size" ); |
|
430 __ ld_ptr( G2_thread, Thread::stack_size_offset(), Rscratch2 ); |
|
431 #ifdef ASSERT |
|
432 Label size_not_zero; |
|
433 __ br_notnull_short(Rscratch2, Assembler::pn, size_not_zero); |
|
434 __ stop("stack size is zero in generate_stack_overflow_check"); |
|
435 __ bind(size_not_zero); |
|
436 #endif |
|
437 |
|
438 // compute the beginning of the protected zone minus the requested frame size |
|
439 __ sub( Rscratch, Rscratch2, Rscratch ); |
|
440 __ set( (StackRedPages+StackYellowPages) * page_size, Rscratch2 ); |
|
441 __ add( Rscratch, Rscratch2, Rscratch ); |
|
442 |
|
443 // Add in the size of the frame (which is the same as subtracting it from the |
|
444 // SP, which would take another register |
|
445 __ add( Rscratch, Rframe_size, Rscratch ); |
|
446 |
|
447 // the frame is greater than one page in size, so check against |
|
448 // the bottom of the stack |
|
449 __ cmp_and_brx_short(SP, Rscratch, Assembler::greaterUnsigned, Assembler::pt, after_frame_check); |
|
450 |
|
451 // the stack will overflow, throw an exception |
|
452 |
|
453 // Note that SP is restored to sender's sp (in the delay slot). This |
|
454 // is necessary if the sender's frame is an extended compiled frame |
|
455 // (see gen_c2i_adapter()) and safer anyway in case of JSR292 |
|
456 // adaptations. |
|
457 |
|
458 // Note also that the restored frame is not necessarily interpreted. |
|
459 // Use the shared runtime version of the StackOverflowError. |
|
460 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated"); |
|
461 AddressLiteral stub(StubRoutines::throw_StackOverflowError_entry()); |
|
462 __ jump_to(stub, Rscratch); |
|
463 __ delayed()->mov(O5_savedSP, SP); |
|
464 |
|
465 // if you get to here, then there is enough stack space |
|
466 __ bind( after_frame_check ); |
|
467 } |
|
468 |
|
469 |
|
470 // |
|
471 // Generate a fixed interpreter frame. This is identical setup for interpreted |
|
472 // methods and for native methods hence the shared code. |
|
473 |
|
474 |
|
475 //---------------------------------------------------------------------------------------------------- |
|
476 // Stack frame layout |
|
477 // |
|
478 // When control flow reaches any of the entry types for the interpreter |
|
479 // the following holds -> |
|
480 // |
|
481 // C2 Calling Conventions: |
|
482 // |
|
483 // The entry code below assumes that the following registers are set |
|
484 // when coming in: |
|
485 // G5_method: holds the Method* of the method to call |
|
486 // Lesp: points to the TOS of the callers expression stack |
|
487 // after having pushed all the parameters |
|
488 // |
|
489 // The entry code does the following to setup an interpreter frame |
|
490 // pop parameters from the callers stack by adjusting Lesp |
|
491 // set O0 to Lesp |
|
492 // compute X = (max_locals - num_parameters) |
|
493 // bump SP up by X to accomadate the extra locals |
|
494 // compute X = max_expression_stack |
|
495 // + vm_local_words |
|
496 // + 16 words of register save area |
|
497 // save frame doing a save sp, -X, sp growing towards lower addresses |
|
498 // set Lbcp, Lmethod, LcpoolCache |
|
499 // set Llocals to i0 |
|
500 // set Lmonitors to FP - rounded_vm_local_words |
|
501 // set Lesp to Lmonitors - 4 |
|
502 // |
|
503 // The frame has now been setup to do the rest of the entry code |
|
504 |
|
505 // Try this optimization: Most method entries could live in a |
|
506 // "one size fits all" stack frame without all the dynamic size |
|
507 // calculations. It might be profitable to do all this calculation |
|
508 // statically and approximately for "small enough" methods. |
|
509 |
|
510 //----------------------------------------------------------------------------------------------- |
|
511 |
|
512 // C1 Calling conventions |
|
513 // |
|
514 // Upon method entry, the following registers are setup: |
|
515 // |
|
516 // g2 G2_thread: current thread |
|
517 // g5 G5_method: method to activate |
|
518 // g4 Gargs : pointer to last argument |
|
519 // |
|
520 // |
|
521 // Stack: |
|
522 // |
|
523 // +---------------+ <--- sp |
|
524 // | | |
|
525 // : reg save area : |
|
526 // | | |
|
527 // +---------------+ <--- sp + 0x40 |
|
528 // | | |
|
529 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) |
|
530 // | | |
|
531 // +---------------+ <--- sp + 0x5c |
|
532 // | | |
|
533 // : free : |
|
534 // | | |
|
535 // +---------------+ <--- Gargs |
|
536 // | | |
|
537 // : arguments : |
|
538 // | | |
|
539 // +---------------+ |
|
540 // | | |
|
541 // |
|
542 // |
|
543 // |
|
544 // AFTER FRAME HAS BEEN SETUP for method interpretation the stack looks like: |
|
545 // |
|
546 // +---------------+ <--- sp |
|
547 // | | |
|
548 // : reg save area : |
|
549 // | | |
|
550 // +---------------+ <--- sp + 0x40 |
|
551 // | | |
|
552 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) |
|
553 // | | |
|
554 // +---------------+ <--- sp + 0x5c |
|
555 // | | |
|
556 // : : |
|
557 // | | <--- Lesp |
|
558 // +---------------+ <--- Lmonitors (fp - 0x18) |
|
559 // | VM locals | |
|
560 // +---------------+ <--- fp |
|
561 // | | |
|
562 // : reg save area : |
|
563 // | | |
|
564 // +---------------+ <--- fp + 0x40 |
|
565 // | | |
|
566 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later) |
|
567 // | | |
|
568 // +---------------+ <--- fp + 0x5c |
|
569 // | | |
|
570 // : free : |
|
571 // | | |
|
572 // +---------------+ |
|
573 // | | |
|
574 // : nonarg locals : |
|
575 // | | |
|
576 // +---------------+ |
|
577 // | | |
|
578 // : arguments : |
|
579 // | | <--- Llocals |
|
580 // +---------------+ <--- Gargs |
|
581 // | | |
|
582 |
|
583 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) { |
|
584 // |
|
585 // |
|
586 // The entry code sets up a new interpreter frame in 4 steps: |
|
587 // |
|
588 // 1) Increase caller's SP by for the extra local space needed: |
|
589 // (check for overflow) |
|
590 // Efficient implementation of xload/xstore bytecodes requires |
|
591 // that arguments and non-argument locals are in a contigously |
|
592 // addressable memory block => non-argument locals must be |
|
593 // allocated in the caller's frame. |
|
594 // |
|
595 // 2) Create a new stack frame and register window: |
|
596 // The new stack frame must provide space for the standard |
|
597 // register save area, the maximum java expression stack size, |
|
598 // the monitor slots (0 slots initially), and some frame local |
|
599 // scratch locations. |
|
600 // |
|
601 // 3) The following interpreter activation registers must be setup: |
|
602 // Lesp : expression stack pointer |
|
603 // Lbcp : bytecode pointer |
|
604 // Lmethod : method |
|
605 // Llocals : locals pointer |
|
606 // Lmonitors : monitor pointer |
|
607 // LcpoolCache: constant pool cache |
|
608 // |
|
609 // 4) Initialize the non-argument locals if necessary: |
|
610 // Non-argument locals may need to be initialized to NULL |
|
611 // for GC to work. If the oop-map information is accurate |
|
612 // (in the absence of the JSR problem), no initialization |
|
613 // is necessary. |
|
614 // |
|
615 // (gri - 2/25/2000) |
|
616 |
|
617 |
|
618 int rounded_vm_local_words = round_to( frame::interpreter_frame_vm_local_words, WordsPerLong ); |
|
619 |
|
620 const int extra_space = |
|
621 rounded_vm_local_words + // frame local scratch space |
|
622 Method::extra_stack_entries() + // extra stack for jsr 292 |
|
623 frame::memory_parameter_word_sp_offset + // register save area |
|
624 (native_call ? frame::interpreter_frame_extra_outgoing_argument_words : 0); |
|
625 |
|
626 const Register Glocals_size = G3; |
|
627 const Register RconstMethod = Glocals_size; |
|
628 const Register Otmp1 = O3; |
|
629 const Register Otmp2 = O4; |
|
630 // Lscratch can't be used as a temporary because the call_stub uses |
|
631 // it to assert that the stack frame was setup correctly. |
|
632 const Address constMethod (G5_method, Method::const_offset()); |
|
633 const Address size_of_parameters(RconstMethod, ConstMethod::size_of_parameters_offset()); |
|
634 |
|
635 __ ld_ptr( constMethod, RconstMethod ); |
|
636 __ lduh( size_of_parameters, Glocals_size); |
|
637 |
|
638 // Gargs points to first local + BytesPerWord |
|
639 // Set the saved SP after the register window save |
|
640 // |
|
641 assert_different_registers(Gargs, Glocals_size, Gframe_size, O5_savedSP); |
|
642 __ sll(Glocals_size, Interpreter::logStackElementSize, Otmp1); |
|
643 __ add(Gargs, Otmp1, Gargs); |
|
644 |
|
645 if (native_call) { |
|
646 __ calc_mem_param_words( Glocals_size, Gframe_size ); |
|
647 __ add( Gframe_size, extra_space, Gframe_size); |
|
648 __ round_to( Gframe_size, WordsPerLong ); |
|
649 __ sll( Gframe_size, LogBytesPerWord, Gframe_size ); |
|
650 } else { |
|
651 |
|
652 // |
|
653 // Compute number of locals in method apart from incoming parameters |
|
654 // |
|
655 const Address size_of_locals (Otmp1, ConstMethod::size_of_locals_offset()); |
|
656 __ ld_ptr( constMethod, Otmp1 ); |
|
657 __ lduh( size_of_locals, Otmp1 ); |
|
658 __ sub( Otmp1, Glocals_size, Glocals_size ); |
|
659 __ round_to( Glocals_size, WordsPerLong ); |
|
660 __ sll( Glocals_size, Interpreter::logStackElementSize, Glocals_size ); |
|
661 |
|
662 // see if the frame is greater than one page in size. If so, |
|
663 // then we need to verify there is enough stack space remaining |
|
664 // Frame_size = (max_stack + extra_space) * BytesPerWord; |
|
665 __ ld_ptr( constMethod, Gframe_size ); |
|
666 __ lduh( Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size ); |
|
667 __ add( Gframe_size, extra_space, Gframe_size ); |
|
668 __ round_to( Gframe_size, WordsPerLong ); |
|
669 __ sll( Gframe_size, Interpreter::logStackElementSize, Gframe_size); |
|
670 |
|
671 // Add in java locals size for stack overflow check only |
|
672 __ add( Gframe_size, Glocals_size, Gframe_size ); |
|
673 |
|
674 const Register Otmp2 = O4; |
|
675 assert_different_registers(Otmp1, Otmp2, O5_savedSP); |
|
676 generate_stack_overflow_check(Gframe_size, Otmp1, Otmp2); |
|
677 |
|
678 __ sub( Gframe_size, Glocals_size, Gframe_size); |
|
679 |
|
680 // |
|
681 // bump SP to accomodate the extra locals |
|
682 // |
|
683 __ sub( SP, Glocals_size, SP ); |
|
684 } |
|
685 |
|
686 // |
|
687 // now set up a stack frame with the size computed above |
|
688 // |
|
689 __ neg( Gframe_size ); |
|
690 __ save( SP, Gframe_size, SP ); |
|
691 |
|
692 // |
|
693 // now set up all the local cache registers |
|
694 // |
|
695 // NOTE: At this point, Lbyte_code/Lscratch has been modified. Note |
|
696 // that all present references to Lbyte_code initialize the register |
|
697 // immediately before use |
|
698 if (native_call) { |
|
699 __ mov(G0, Lbcp); |
|
700 } else { |
|
701 __ ld_ptr(G5_method, Method::const_offset(), Lbcp); |
|
702 __ add(Lbcp, in_bytes(ConstMethod::codes_offset()), Lbcp); |
|
703 } |
|
704 __ mov( G5_method, Lmethod); // set Lmethod |
|
705 __ get_constant_pool_cache( LcpoolCache ); // set LcpoolCache |
|
706 __ sub(FP, rounded_vm_local_words * BytesPerWord, Lmonitors ); // set Lmonitors |
|
707 #ifdef _LP64 |
|
708 __ add( Lmonitors, STACK_BIAS, Lmonitors ); // Account for 64 bit stack bias |
|
709 #endif |
|
710 __ sub(Lmonitors, BytesPerWord, Lesp); // set Lesp |
|
711 |
|
712 // setup interpreter activation registers |
|
713 __ sub(Gargs, BytesPerWord, Llocals); // set Llocals |
|
714 |
|
715 if (ProfileInterpreter) { |
|
716 #ifdef FAST_DISPATCH |
|
717 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since |
|
718 // they both use I2. |
|
719 assert(0, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive"); |
|
720 #endif // FAST_DISPATCH |
|
721 __ set_method_data_pointer(); |
|
722 } |
|
723 |
|
724 } |
|
725 |
|
726 // Method entry for java.lang.ref.Reference.get. |
|
727 address InterpreterGenerator::generate_Reference_get_entry(void) { |
|
728 #if INCLUDE_ALL_GCS |
|
729 // Code: _aload_0, _getfield, _areturn |
|
730 // parameter size = 1 |
|
731 // |
|
732 // The code that gets generated by this routine is split into 2 parts: |
|
733 // 1. The "intrinsified" code for G1 (or any SATB based GC), |
|
734 // 2. The slow path - which is an expansion of the regular method entry. |
|
735 // |
|
736 // Notes:- |
|
737 // * In the G1 code we do not check whether we need to block for |
|
738 // a safepoint. If G1 is enabled then we must execute the specialized |
|
739 // code for Reference.get (except when the Reference object is null) |
|
740 // so that we can log the value in the referent field with an SATB |
|
741 // update buffer. |
|
742 // If the code for the getfield template is modified so that the |
|
743 // G1 pre-barrier code is executed when the current method is |
|
744 // Reference.get() then going through the normal method entry |
|
745 // will be fine. |
|
746 // * The G1 code can, however, check the receiver object (the instance |
|
747 // of java.lang.Reference) and jump to the slow path if null. If the |
|
748 // Reference object is null then we obviously cannot fetch the referent |
|
749 // and so we don't need to call the G1 pre-barrier. Thus we can use the |
|
750 // regular method entry code to generate the NPE. |
|
751 // |
|
752 // This code is based on generate_accessor_enty. |
|
753 |
|
754 address entry = __ pc(); |
|
755 |
|
756 const int referent_offset = java_lang_ref_Reference::referent_offset; |
|
757 guarantee(referent_offset > 0, "referent offset not initialized"); |
|
758 |
|
759 if (UseG1GC) { |
|
760 Label slow_path; |
|
761 |
|
762 // In the G1 code we don't check if we need to reach a safepoint. We |
|
763 // continue and the thread will safepoint at the next bytecode dispatch. |
|
764 |
|
765 // Check if local 0 != NULL |
|
766 // If the receiver is null then it is OK to jump to the slow path. |
|
767 __ ld_ptr(Gargs, G0, Otos_i ); // get local 0 |
|
768 // check if local 0 == NULL and go the slow path |
|
769 __ cmp_and_brx_short(Otos_i, 0, Assembler::equal, Assembler::pn, slow_path); |
|
770 |
|
771 |
|
772 // Load the value of the referent field. |
|
773 if (Assembler::is_simm13(referent_offset)) { |
|
774 __ load_heap_oop(Otos_i, referent_offset, Otos_i); |
|
775 } else { |
|
776 __ set(referent_offset, G3_scratch); |
|
777 __ load_heap_oop(Otos_i, G3_scratch, Otos_i); |
|
778 } |
|
779 |
|
780 // Generate the G1 pre-barrier code to log the value of |
|
781 // the referent field in an SATB buffer. Note with |
|
782 // these parameters the pre-barrier does not generate |
|
783 // the load of the previous value |
|
784 |
|
785 __ g1_write_barrier_pre(noreg /* obj */, noreg /* index */, 0 /* offset */, |
|
786 Otos_i /* pre_val */, |
|
787 G3_scratch /* tmp */, |
|
788 true /* preserve_o_regs */); |
|
789 |
|
790 // _areturn |
|
791 __ retl(); // return from leaf routine |
|
792 __ delayed()->mov(O5_savedSP, SP); |
|
793 |
|
794 // Generate regular method entry |
|
795 __ bind(slow_path); |
|
796 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals)); |
|
797 return entry; |
|
798 } |
|
799 #endif // INCLUDE_ALL_GCS |
|
800 |
|
801 // If G1 is not enabled then attempt to go through the accessor entry point |
|
802 // Reference.get is an accessor |
|
803 return NULL; |
|
804 } |
|
805 |
|
806 /** |
|
807 * Method entry for static native methods: |
|
808 * int java.util.zip.CRC32.update(int crc, int b) |
|
809 */ |
|
810 address InterpreterGenerator::generate_CRC32_update_entry() { |
|
811 |
|
812 if (UseCRC32Intrinsics) { |
|
813 address entry = __ pc(); |
|
814 |
|
815 Label L_slow_path; |
|
816 // If we need a safepoint check, generate full interpreter entry. |
|
817 ExternalAddress state(SafepointSynchronize::address_of_state()); |
|
818 __ set(ExternalAddress(SafepointSynchronize::address_of_state()), O2); |
|
819 __ set(SafepointSynchronize::_not_synchronized, O3); |
|
820 __ cmp_and_br_short(O2, O3, Assembler::notEqual, Assembler::pt, L_slow_path); |
|
821 |
|
822 // Load parameters |
|
823 const Register crc = O0; // initial crc |
|
824 const Register val = O1; // byte to update with |
|
825 const Register table = O2; // address of 256-entry lookup table |
|
826 |
|
827 __ ldub(Gargs, 3, val); |
|
828 __ lduw(Gargs, 8, crc); |
|
829 |
|
830 __ set(ExternalAddress(StubRoutines::crc_table_addr()), table); |
|
831 |
|
832 __ not1(crc); // ~crc |
|
833 __ clruwu(crc); |
|
834 __ update_byte_crc32(crc, val, table); |
|
835 __ not1(crc); // ~crc |
|
836 |
|
837 // result in O0 |
|
838 __ retl(); |
|
839 __ delayed()->nop(); |
|
840 |
|
841 // generate a vanilla native entry as the slow path |
|
842 __ bind(L_slow_path); |
|
843 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native)); |
|
844 return entry; |
|
845 } |
|
846 return NULL; |
|
847 } |
|
848 |
|
849 /** |
|
850 * Method entry for static native methods: |
|
851 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len) |
|
852 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) |
|
853 */ |
|
854 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { |
|
855 |
|
856 if (UseCRC32Intrinsics) { |
|
857 address entry = __ pc(); |
|
858 |
|
859 Label L_slow_path; |
|
860 // If we need a safepoint check, generate full interpreter entry. |
|
861 ExternalAddress state(SafepointSynchronize::address_of_state()); |
|
862 __ set(ExternalAddress(SafepointSynchronize::address_of_state()), O2); |
|
863 __ set(SafepointSynchronize::_not_synchronized, O3); |
|
864 __ cmp_and_br_short(O2, O3, Assembler::notEqual, Assembler::pt, L_slow_path); |
|
865 |
|
866 // Load parameters from the stack |
|
867 const Register crc = O0; // initial crc |
|
868 const Register buf = O1; // source java byte array address |
|
869 const Register len = O2; // len |
|
870 const Register offset = O3; // offset |
|
871 |
|
872 // Arguments are reversed on java expression stack |
|
873 // Calculate address of start element |
|
874 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { |
|
875 __ lduw(Gargs, 0, len); |
|
876 __ lduw(Gargs, 8, offset); |
|
877 __ ldx( Gargs, 16, buf); |
|
878 __ lduw(Gargs, 32, crc); |
|
879 __ add(buf, offset, buf); |
|
880 } else { |
|
881 __ lduw(Gargs, 0, len); |
|
882 __ lduw(Gargs, 8, offset); |
|
883 __ ldx( Gargs, 16, buf); |
|
884 __ lduw(Gargs, 24, crc); |
|
885 __ add(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE), buf); // account for the header size |
|
886 __ add(buf ,offset, buf); |
|
887 } |
|
888 |
|
889 // Call the crc32 kernel |
|
890 __ MacroAssembler::save_thread(L7_thread_cache); |
|
891 __ kernel_crc32(crc, buf, len, O3); |
|
892 __ MacroAssembler::restore_thread(L7_thread_cache); |
|
893 |
|
894 // result in O0 |
|
895 __ retl(); |
|
896 __ delayed()->nop(); |
|
897 |
|
898 // generate a vanilla native entry as the slow path |
|
899 __ bind(L_slow_path); |
|
900 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native)); |
|
901 return entry; |
|
902 } |
|
903 return NULL; |
|
904 } |
|
905 |
|
906 // |
|
907 // Interpreter stub for calling a native method. (asm interpreter) |
|
908 // This sets up a somewhat different looking stack for calling the native method |
|
909 // than the typical interpreter frame setup. |
|
910 // |
|
911 |
|
912 address InterpreterGenerator::generate_native_entry(bool synchronized) { |
|
913 address entry = __ pc(); |
|
914 |
|
915 // the following temporary registers are used during frame creation |
|
916 const Register Gtmp1 = G3_scratch ; |
|
917 const Register Gtmp2 = G1_scratch; |
|
918 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; |
|
919 |
|
920 // make sure registers are different! |
|
921 assert_different_registers(G2_thread, G5_method, Gargs, Gtmp1, Gtmp2); |
|
922 |
|
923 const Address Laccess_flags(Lmethod, Method::access_flags_offset()); |
|
924 |
|
925 const Register Glocals_size = G3; |
|
926 assert_different_registers(Glocals_size, G4_scratch, Gframe_size); |
|
927 |
|
928 // make sure method is native & not abstract |
|
929 // rethink these assertions - they can be simplified and shared (gri 2/25/2000) |
|
930 #ifdef ASSERT |
|
931 __ ld(G5_method, Method::access_flags_offset(), Gtmp1); |
|
932 { |
|
933 Label L; |
|
934 __ btst(JVM_ACC_NATIVE, Gtmp1); |
|
935 __ br(Assembler::notZero, false, Assembler::pt, L); |
|
936 __ delayed()->nop(); |
|
937 __ stop("tried to execute non-native method as native"); |
|
938 __ bind(L); |
|
939 } |
|
940 { Label L; |
|
941 __ btst(JVM_ACC_ABSTRACT, Gtmp1); |
|
942 __ br(Assembler::zero, false, Assembler::pt, L); |
|
943 __ delayed()->nop(); |
|
944 __ stop("tried to execute abstract method as non-abstract"); |
|
945 __ bind(L); |
|
946 } |
|
947 #endif // ASSERT |
|
948 |
|
949 // generate the code to allocate the interpreter stack frame |
|
950 generate_fixed_frame(true); |
|
951 |
|
952 // |
|
953 // No locals to initialize for native method |
|
954 // |
|
955 |
|
956 // this slot will be set later, we initialize it to null here just in |
|
957 // case we get a GC before the actual value is stored later |
|
958 __ st_ptr(G0, FP, (frame::interpreter_frame_oop_temp_offset * wordSize) + STACK_BIAS); |
|
959 |
|
960 const Address do_not_unlock_if_synchronized(G2_thread, |
|
961 JavaThread::do_not_unlock_if_synchronized_offset()); |
|
962 // Since at this point in the method invocation the exception handler |
|
963 // would try to exit the monitor of synchronized methods which hasn't |
|
964 // been entered yet, we set the thread local variable |
|
965 // _do_not_unlock_if_synchronized to true. If any exception was thrown by |
|
966 // runtime, exception handling i.e. unlock_if_synchronized_method will |
|
967 // check this thread local flag. |
|
968 // This flag has two effects, one is to force an unwind in the topmost |
|
969 // interpreter frame and not perform an unlock while doing so. |
|
970 |
|
971 __ movbool(true, G3_scratch); |
|
972 __ stbool(G3_scratch, do_not_unlock_if_synchronized); |
|
973 |
|
974 // increment invocation counter and check for overflow |
|
975 // |
|
976 // Note: checking for negative value instead of overflow |
|
977 // so we have a 'sticky' overflow test (may be of |
|
978 // importance as soon as we have true MT/MP) |
|
979 Label invocation_counter_overflow; |
|
980 Label Lcontinue; |
|
981 if (inc_counter) { |
|
982 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); |
|
983 |
|
984 } |
|
985 __ bind(Lcontinue); |
|
986 |
|
987 bang_stack_shadow_pages(true); |
|
988 |
|
989 // reset the _do_not_unlock_if_synchronized flag |
|
990 __ stbool(G0, do_not_unlock_if_synchronized); |
|
991 |
|
992 // check for synchronized methods |
|
993 // Must happen AFTER invocation_counter check and stack overflow check, |
|
994 // so method is not locked if overflows. |
|
995 |
|
996 if (synchronized) { |
|
997 lock_method(); |
|
998 } else { |
|
999 #ifdef ASSERT |
|
1000 { Label ok; |
|
1001 __ ld(Laccess_flags, O0); |
|
1002 __ btst(JVM_ACC_SYNCHRONIZED, O0); |
|
1003 __ br( Assembler::zero, false, Assembler::pt, ok); |
|
1004 __ delayed()->nop(); |
|
1005 __ stop("method needs synchronization"); |
|
1006 __ bind(ok); |
|
1007 } |
|
1008 #endif // ASSERT |
|
1009 } |
|
1010 |
|
1011 |
|
1012 // start execution |
|
1013 __ verify_thread(); |
|
1014 |
|
1015 // JVMTI support |
|
1016 __ notify_method_entry(); |
|
1017 |
|
1018 // native call |
|
1019 |
|
1020 // (note that O0 is never an oop--at most it is a handle) |
|
1021 // It is important not to smash any handles created by this call, |
|
1022 // until any oop handle in O0 is dereferenced. |
|
1023 |
|
1024 // (note that the space for outgoing params is preallocated) |
|
1025 |
|
1026 // get signature handler |
|
1027 { Label L; |
|
1028 Address signature_handler(Lmethod, Method::signature_handler_offset()); |
|
1029 __ ld_ptr(signature_handler, G3_scratch); |
|
1030 __ br_notnull_short(G3_scratch, Assembler::pt, L); |
|
1031 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), Lmethod); |
|
1032 __ ld_ptr(signature_handler, G3_scratch); |
|
1033 __ bind(L); |
|
1034 } |
|
1035 |
|
1036 // Push a new frame so that the args will really be stored in |
|
1037 // Copy a few locals across so the new frame has the variables |
|
1038 // we need but these values will be dead at the jni call and |
|
1039 // therefore not gc volatile like the values in the current |
|
1040 // frame (Lmethod in particular) |
|
1041 |
|
1042 // Flush the method pointer to the register save area |
|
1043 __ st_ptr(Lmethod, SP, (Lmethod->sp_offset_in_saved_window() * wordSize) + STACK_BIAS); |
|
1044 __ mov(Llocals, O1); |
|
1045 |
|
1046 // calculate where the mirror handle body is allocated in the interpreter frame: |
|
1047 __ add(FP, (frame::interpreter_frame_oop_temp_offset * wordSize) + STACK_BIAS, O2); |
|
1048 |
|
1049 // Calculate current frame size |
|
1050 __ sub(SP, FP, O3); // Calculate negative of current frame size |
|
1051 __ save(SP, O3, SP); // Allocate an identical sized frame |
|
1052 |
|
1053 // Note I7 has leftover trash. Slow signature handler will fill it in |
|
1054 // should we get there. Normal jni call will set reasonable last_Java_pc |
|
1055 // below (and fix I7 so the stack trace doesn't have a meaningless frame |
|
1056 // in it). |
|
1057 |
|
1058 // Load interpreter frame's Lmethod into same register here |
|
1059 |
|
1060 __ ld_ptr(FP, (Lmethod->sp_offset_in_saved_window() * wordSize) + STACK_BIAS, Lmethod); |
|
1061 |
|
1062 __ mov(I1, Llocals); |
|
1063 __ mov(I2, Lscratch2); // save the address of the mirror |
|
1064 |
|
1065 |
|
1066 // ONLY Lmethod and Llocals are valid here! |
|
1067 |
|
1068 // call signature handler, It will move the arg properly since Llocals in current frame |
|
1069 // matches that in outer frame |
|
1070 |
|
1071 __ callr(G3_scratch, 0); |
|
1072 __ delayed()->nop(); |
|
1073 |
|
1074 // Result handler is in Lscratch |
|
1075 |
|
1076 // Reload interpreter frame's Lmethod since slow signature handler may block |
|
1077 __ ld_ptr(FP, (Lmethod->sp_offset_in_saved_window() * wordSize) + STACK_BIAS, Lmethod); |
|
1078 |
|
1079 { Label not_static; |
|
1080 |
|
1081 __ ld(Laccess_flags, O0); |
|
1082 __ btst(JVM_ACC_STATIC, O0); |
|
1083 __ br( Assembler::zero, false, Assembler::pt, not_static); |
|
1084 // get native function entry point(O0 is a good temp until the very end) |
|
1085 __ delayed()->ld_ptr(Lmethod, in_bytes(Method::native_function_offset()), O0); |
|
1086 // for static methods insert the mirror argument |
|
1087 const int mirror_offset = in_bytes(Klass::java_mirror_offset()); |
|
1088 |
|
1089 __ ld_ptr(Lmethod, Method:: const_offset(), O1); |
|
1090 __ ld_ptr(O1, ConstMethod::constants_offset(), O1); |
|
1091 __ ld_ptr(O1, ConstantPool::pool_holder_offset_in_bytes(), O1); |
|
1092 __ ld_ptr(O1, mirror_offset, O1); |
|
1093 #ifdef ASSERT |
|
1094 if (!PrintSignatureHandlers) // do not dirty the output with this |
|
1095 { Label L; |
|
1096 __ br_notnull_short(O1, Assembler::pt, L); |
|
1097 __ stop("mirror is missing"); |
|
1098 __ bind(L); |
|
1099 } |
|
1100 #endif // ASSERT |
|
1101 __ st_ptr(O1, Lscratch2, 0); |
|
1102 __ mov(Lscratch2, O1); |
|
1103 __ bind(not_static); |
|
1104 } |
|
1105 |
|
1106 // At this point, arguments have been copied off of stack into |
|
1107 // their JNI positions, which are O1..O5 and SP[68..]. |
|
1108 // Oops are boxed in-place on the stack, with handles copied to arguments. |
|
1109 // The result handler is in Lscratch. O0 will shortly hold the JNIEnv*. |
|
1110 |
|
1111 #ifdef ASSERT |
|
1112 { Label L; |
|
1113 __ br_notnull_short(O0, Assembler::pt, L); |
|
1114 __ stop("native entry point is missing"); |
|
1115 __ bind(L); |
|
1116 } |
|
1117 #endif // ASSERT |
|
1118 |
|
1119 // |
|
1120 // setup the frame anchor |
|
1121 // |
|
1122 // The scavenge function only needs to know that the PC of this frame is |
|
1123 // in the interpreter method entry code, it doesn't need to know the exact |
|
1124 // PC and hence we can use O7 which points to the return address from the |
|
1125 // previous call in the code stream (signature handler function) |
|
1126 // |
|
1127 // The other trick is we set last_Java_sp to FP instead of the usual SP because |
|
1128 // we have pushed the extra frame in order to protect the volatile register(s) |
|
1129 // in that frame when we return from the jni call |
|
1130 // |
|
1131 |
|
1132 __ set_last_Java_frame(FP, O7); |
|
1133 __ mov(O7, I7); // make dummy interpreter frame look like one above, |
|
1134 // not meaningless information that'll confuse me. |
|
1135 |
|
1136 // flush the windows now. We don't care about the current (protection) frame |
|
1137 // only the outer frames |
|
1138 |
|
1139 __ flushw(); |
|
1140 |
|
1141 // mark windows as flushed |
|
1142 Address flags(G2_thread, JavaThread::frame_anchor_offset() + JavaFrameAnchor::flags_offset()); |
|
1143 __ set(JavaFrameAnchor::flushed, G3_scratch); |
|
1144 __ st(G3_scratch, flags); |
|
1145 |
|
1146 // Transition from _thread_in_Java to _thread_in_native. We are already safepoint ready. |
|
1147 |
|
1148 Address thread_state(G2_thread, JavaThread::thread_state_offset()); |
|
1149 #ifdef ASSERT |
|
1150 { Label L; |
|
1151 __ ld(thread_state, G3_scratch); |
|
1152 __ cmp_and_br_short(G3_scratch, _thread_in_Java, Assembler::equal, Assembler::pt, L); |
|
1153 __ stop("Wrong thread state in native stub"); |
|
1154 __ bind(L); |
|
1155 } |
|
1156 #endif // ASSERT |
|
1157 __ set(_thread_in_native, G3_scratch); |
|
1158 __ st(G3_scratch, thread_state); |
|
1159 |
|
1160 // Call the jni method, using the delay slot to set the JNIEnv* argument. |
|
1161 __ save_thread(L7_thread_cache); // save Gthread |
|
1162 __ callr(O0, 0); |
|
1163 __ delayed()-> |
|
1164 add(L7_thread_cache, in_bytes(JavaThread::jni_environment_offset()), O0); |
|
1165 |
|
1166 // Back from jni method Lmethod in this frame is DEAD, DEAD, DEAD |
|
1167 |
|
1168 __ restore_thread(L7_thread_cache); // restore G2_thread |
|
1169 __ reinit_heapbase(); |
|
1170 |
|
1171 // must we block? |
|
1172 |
|
1173 // Block, if necessary, before resuming in _thread_in_Java state. |
|
1174 // In order for GC to work, don't clear the last_Java_sp until after blocking. |
|
1175 { Label no_block; |
|
1176 AddressLiteral sync_state(SafepointSynchronize::address_of_state()); |
|
1177 |
|
1178 // Switch thread to "native transition" state before reading the synchronization state. |
|
1179 // This additional state is necessary because reading and testing the synchronization |
|
1180 // state is not atomic w.r.t. GC, as this scenario demonstrates: |
|
1181 // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted. |
|
1182 // VM thread changes sync state to synchronizing and suspends threads for GC. |
|
1183 // Thread A is resumed to finish this native method, but doesn't block here since it |
|
1184 // didn't see any synchronization is progress, and escapes. |
|
1185 __ set(_thread_in_native_trans, G3_scratch); |
|
1186 __ st(G3_scratch, thread_state); |
|
1187 if(os::is_MP()) { |
|
1188 if (UseMembar) { |
|
1189 // Force this write out before the read below |
|
1190 __ membar(Assembler::StoreLoad); |
|
1191 } else { |
|
1192 // Write serialization page so VM thread can do a pseudo remote membar. |
|
1193 // We use the current thread pointer to calculate a thread specific |
|
1194 // offset to write to within the page. This minimizes bus traffic |
|
1195 // due to cache line collision. |
|
1196 __ serialize_memory(G2_thread, G1_scratch, G3_scratch); |
|
1197 } |
|
1198 } |
|
1199 __ load_contents(sync_state, G3_scratch); |
|
1200 __ cmp(G3_scratch, SafepointSynchronize::_not_synchronized); |
|
1201 |
|
1202 Label L; |
|
1203 __ br(Assembler::notEqual, false, Assembler::pn, L); |
|
1204 __ delayed()->ld(G2_thread, JavaThread::suspend_flags_offset(), G3_scratch); |
|
1205 __ cmp_and_br_short(G3_scratch, 0, Assembler::equal, Assembler::pt, no_block); |
|
1206 __ bind(L); |
|
1207 |
|
1208 // Block. Save any potential method result value before the operation and |
|
1209 // use a leaf call to leave the last_Java_frame setup undisturbed. |
|
1210 save_native_result(); |
|
1211 __ call_VM_leaf(L7_thread_cache, |
|
1212 CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), |
|
1213 G2_thread); |
|
1214 |
|
1215 // Restore any method result value |
|
1216 restore_native_result(); |
|
1217 __ bind(no_block); |
|
1218 } |
|
1219 |
|
1220 // Clear the frame anchor now |
|
1221 |
|
1222 __ reset_last_Java_frame(); |
|
1223 |
|
1224 // Move the result handler address |
|
1225 __ mov(Lscratch, G3_scratch); |
|
1226 // return possible result to the outer frame |
|
1227 #ifndef __LP64 |
|
1228 __ mov(O0, I0); |
|
1229 __ restore(O1, G0, O1); |
|
1230 #else |
|
1231 __ restore(O0, G0, O0); |
|
1232 #endif /* __LP64 */ |
|
1233 |
|
1234 // Move result handler to expected register |
|
1235 __ mov(G3_scratch, Lscratch); |
|
1236 |
|
1237 // Back in normal (native) interpreter frame. State is thread_in_native_trans |
|
1238 // switch to thread_in_Java. |
|
1239 |
|
1240 __ set(_thread_in_Java, G3_scratch); |
|
1241 __ st(G3_scratch, thread_state); |
|
1242 |
|
1243 // reset handle block |
|
1244 __ ld_ptr(G2_thread, JavaThread::active_handles_offset(), G3_scratch); |
|
1245 __ st(G0, G3_scratch, JNIHandleBlock::top_offset_in_bytes()); |
|
1246 |
|
1247 // If we have an oop result store it where it will be safe for any further gc |
|
1248 // until we return now that we've released the handle it might be protected by |
|
1249 |
|
1250 { |
|
1251 Label no_oop, store_result; |
|
1252 |
|
1253 __ set((intptr_t)AbstractInterpreter::result_handler(T_OBJECT), G3_scratch); |
|
1254 __ cmp_and_brx_short(G3_scratch, Lscratch, Assembler::notEqual, Assembler::pt, no_oop); |
|
1255 __ addcc(G0, O0, O0); |
|
1256 __ brx(Assembler::notZero, true, Assembler::pt, store_result); // if result is not NULL: |
|
1257 __ delayed()->ld_ptr(O0, 0, O0); // unbox it |
|
1258 __ mov(G0, O0); |
|
1259 |
|
1260 __ bind(store_result); |
|
1261 // Store it where gc will look for it and result handler expects it. |
|
1262 __ st_ptr(O0, FP, (frame::interpreter_frame_oop_temp_offset*wordSize) + STACK_BIAS); |
|
1263 |
|
1264 __ bind(no_oop); |
|
1265 |
|
1266 } |
|
1267 |
|
1268 |
|
1269 // handle exceptions (exception handling will handle unlocking!) |
|
1270 { Label L; |
|
1271 Address exception_addr(G2_thread, Thread::pending_exception_offset()); |
|
1272 __ ld_ptr(exception_addr, Gtemp); |
|
1273 __ br_null_short(Gtemp, Assembler::pt, L); |
|
1274 // Note: This could be handled more efficiently since we know that the native |
|
1275 // method doesn't have an exception handler. We could directly return |
|
1276 // to the exception handler for the caller. |
|
1277 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception)); |
|
1278 __ should_not_reach_here(); |
|
1279 __ bind(L); |
|
1280 } |
|
1281 |
|
1282 // JVMTI support (preserves thread register) |
|
1283 __ notify_method_exit(true, ilgl, InterpreterMacroAssembler::NotifyJVMTI); |
|
1284 |
|
1285 if (synchronized) { |
|
1286 // save and restore any potential method result value around the unlocking operation |
|
1287 save_native_result(); |
|
1288 |
|
1289 __ add( __ top_most_monitor(), O1); |
|
1290 __ unlock_object(O1); |
|
1291 |
|
1292 restore_native_result(); |
|
1293 } |
|
1294 |
|
1295 #if defined(COMPILER2) && !defined(_LP64) |
|
1296 |
|
1297 // C2 expects long results in G1 we can't tell if we're returning to interpreted |
|
1298 // or compiled so just be safe. |
|
1299 |
|
1300 __ sllx(O0, 32, G1); // Shift bits into high G1 |
|
1301 __ srl (O1, 0, O1); // Zero extend O1 |
|
1302 __ or3 (O1, G1, G1); // OR 64 bits into G1 |
|
1303 |
|
1304 #endif /* COMPILER2 && !_LP64 */ |
|
1305 |
|
1306 // dispose of return address and remove activation |
|
1307 #ifdef ASSERT |
|
1308 { |
|
1309 Label ok; |
|
1310 __ cmp_and_brx_short(I5_savedSP, FP, Assembler::greaterEqualUnsigned, Assembler::pt, ok); |
|
1311 __ stop("bad I5_savedSP value"); |
|
1312 __ should_not_reach_here(); |
|
1313 __ bind(ok); |
|
1314 } |
|
1315 #endif |
|
1316 if (TraceJumps) { |
|
1317 // Move target to register that is recordable |
|
1318 __ mov(Lscratch, G3_scratch); |
|
1319 __ JMP(G3_scratch, 0); |
|
1320 } else { |
|
1321 __ jmp(Lscratch, 0); |
|
1322 } |
|
1323 __ delayed()->nop(); |
|
1324 |
|
1325 |
|
1326 if (inc_counter) { |
|
1327 // handle invocation counter overflow |
|
1328 __ bind(invocation_counter_overflow); |
|
1329 generate_counter_overflow(Lcontinue); |
|
1330 } |
|
1331 |
|
1332 |
|
1333 |
|
1334 return entry; |
|
1335 } |
|
1336 |
|
1337 |
|
1338 // Generic method entry to (asm) interpreter |
|
1339 address InterpreterGenerator::generate_normal_entry(bool synchronized) { |
|
1340 address entry = __ pc(); |
|
1341 |
|
1342 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; |
|
1343 |
|
1344 // the following temporary registers are used during frame creation |
|
1345 const Register Gtmp1 = G3_scratch ; |
|
1346 const Register Gtmp2 = G1_scratch; |
|
1347 |
|
1348 // make sure registers are different! |
|
1349 assert_different_registers(G2_thread, G5_method, Gargs, Gtmp1, Gtmp2); |
|
1350 |
|
1351 const Address constMethod (G5_method, Method::const_offset()); |
|
1352 // Seems like G5_method is live at the point this is used. So we could make this look consistent |
|
1353 // and use in the asserts. |
|
1354 const Address access_flags (Lmethod, Method::access_flags_offset()); |
|
1355 |
|
1356 const Register Glocals_size = G3; |
|
1357 assert_different_registers(Glocals_size, G4_scratch, Gframe_size); |
|
1358 |
|
1359 // make sure method is not native & not abstract |
|
1360 // rethink these assertions - they can be simplified and shared (gri 2/25/2000) |
|
1361 #ifdef ASSERT |
|
1362 __ ld(G5_method, Method::access_flags_offset(), Gtmp1); |
|
1363 { |
|
1364 Label L; |
|
1365 __ btst(JVM_ACC_NATIVE, Gtmp1); |
|
1366 __ br(Assembler::zero, false, Assembler::pt, L); |
|
1367 __ delayed()->nop(); |
|
1368 __ stop("tried to execute native method as non-native"); |
|
1369 __ bind(L); |
|
1370 } |
|
1371 { Label L; |
|
1372 __ btst(JVM_ACC_ABSTRACT, Gtmp1); |
|
1373 __ br(Assembler::zero, false, Assembler::pt, L); |
|
1374 __ delayed()->nop(); |
|
1375 __ stop("tried to execute abstract method as non-abstract"); |
|
1376 __ bind(L); |
|
1377 } |
|
1378 #endif // ASSERT |
|
1379 |
|
1380 // generate the code to allocate the interpreter stack frame |
|
1381 |
|
1382 generate_fixed_frame(false); |
|
1383 |
|
1384 #ifdef FAST_DISPATCH |
|
1385 __ set((intptr_t)Interpreter::dispatch_table(), IdispatchTables); |
|
1386 // set bytecode dispatch table base |
|
1387 #endif |
|
1388 |
|
1389 // |
|
1390 // Code to initialize the extra (i.e. non-parm) locals |
|
1391 // |
|
1392 Register init_value = noreg; // will be G0 if we must clear locals |
|
1393 // The way the code was setup before zerolocals was always true for vanilla java entries. |
|
1394 // It could only be false for the specialized entries like accessor or empty which have |
|
1395 // no extra locals so the testing was a waste of time and the extra locals were always |
|
1396 // initialized. We removed this extra complication to already over complicated code. |
|
1397 |
|
1398 init_value = G0; |
|
1399 Label clear_loop; |
|
1400 |
|
1401 const Register RconstMethod = O1; |
|
1402 const Address size_of_parameters(RconstMethod, ConstMethod::size_of_parameters_offset()); |
|
1403 const Address size_of_locals (RconstMethod, ConstMethod::size_of_locals_offset()); |
|
1404 |
|
1405 // NOTE: If you change the frame layout, this code will need to |
|
1406 // be updated! |
|
1407 __ ld_ptr( constMethod, RconstMethod ); |
|
1408 __ lduh( size_of_locals, O2 ); |
|
1409 __ lduh( size_of_parameters, O1 ); |
|
1410 __ sll( O2, Interpreter::logStackElementSize, O2); |
|
1411 __ sll( O1, Interpreter::logStackElementSize, O1 ); |
|
1412 __ sub( Llocals, O2, O2 ); |
|
1413 __ sub( Llocals, O1, O1 ); |
|
1414 |
|
1415 __ bind( clear_loop ); |
|
1416 __ inc( O2, wordSize ); |
|
1417 |
|
1418 __ cmp( O2, O1 ); |
|
1419 __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, clear_loop ); |
|
1420 __ delayed()->st_ptr( init_value, O2, 0 ); |
|
1421 |
|
1422 const Address do_not_unlock_if_synchronized(G2_thread, |
|
1423 JavaThread::do_not_unlock_if_synchronized_offset()); |
|
1424 // Since at this point in the method invocation the exception handler |
|
1425 // would try to exit the monitor of synchronized methods which hasn't |
|
1426 // been entered yet, we set the thread local variable |
|
1427 // _do_not_unlock_if_synchronized to true. If any exception was thrown by |
|
1428 // runtime, exception handling i.e. unlock_if_synchronized_method will |
|
1429 // check this thread local flag. |
|
1430 __ movbool(true, G3_scratch); |
|
1431 __ stbool(G3_scratch, do_not_unlock_if_synchronized); |
|
1432 |
|
1433 __ profile_parameters_type(G1_scratch, G3_scratch, G4_scratch, Lscratch); |
|
1434 // increment invocation counter and check for overflow |
|
1435 // |
|
1436 // Note: checking for negative value instead of overflow |
|
1437 // so we have a 'sticky' overflow test (may be of |
|
1438 // importance as soon as we have true MT/MP) |
|
1439 Label invocation_counter_overflow; |
|
1440 Label profile_method; |
|
1441 Label profile_method_continue; |
|
1442 Label Lcontinue; |
|
1443 if (inc_counter) { |
|
1444 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue); |
|
1445 if (ProfileInterpreter) { |
|
1446 __ bind(profile_method_continue); |
|
1447 } |
|
1448 } |
|
1449 __ bind(Lcontinue); |
|
1450 |
|
1451 bang_stack_shadow_pages(false); |
|
1452 |
|
1453 // reset the _do_not_unlock_if_synchronized flag |
|
1454 __ stbool(G0, do_not_unlock_if_synchronized); |
|
1455 |
|
1456 // check for synchronized methods |
|
1457 // Must happen AFTER invocation_counter check and stack overflow check, |
|
1458 // so method is not locked if overflows. |
|
1459 |
|
1460 if (synchronized) { |
|
1461 lock_method(); |
|
1462 } else { |
|
1463 #ifdef ASSERT |
|
1464 { Label ok; |
|
1465 __ ld(access_flags, O0); |
|
1466 __ btst(JVM_ACC_SYNCHRONIZED, O0); |
|
1467 __ br( Assembler::zero, false, Assembler::pt, ok); |
|
1468 __ delayed()->nop(); |
|
1469 __ stop("method needs synchronization"); |
|
1470 __ bind(ok); |
|
1471 } |
|
1472 #endif // ASSERT |
|
1473 } |
|
1474 |
|
1475 // start execution |
|
1476 |
|
1477 __ verify_thread(); |
|
1478 |
|
1479 // jvmti support |
|
1480 __ notify_method_entry(); |
|
1481 |
|
1482 // start executing instructions |
|
1483 __ dispatch_next(vtos); |
|
1484 |
|
1485 |
|
1486 if (inc_counter) { |
|
1487 if (ProfileInterpreter) { |
|
1488 // We have decided to profile this method in the interpreter |
|
1489 __ bind(profile_method); |
|
1490 |
|
1491 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); |
|
1492 __ set_method_data_pointer_for_bcp(); |
|
1493 __ ba_short(profile_method_continue); |
|
1494 } |
|
1495 |
|
1496 // handle invocation counter overflow |
|
1497 __ bind(invocation_counter_overflow); |
|
1498 generate_counter_overflow(Lcontinue); |
|
1499 } |
|
1500 |
|
1501 |
|
1502 return entry; |
|
1503 } |
|
1504 |
|
1505 //---------------------------------------------------------------------------------------------------- |
|
1506 // Exceptions |
|
1507 void TemplateInterpreterGenerator::generate_throw_exception() { |
|
1508 |
|
1509 // Entry point in previous activation (i.e., if the caller was interpreted) |
|
1510 Interpreter::_rethrow_exception_entry = __ pc(); |
|
1511 // O0: exception |
|
1512 |
|
1513 // entry point for exceptions thrown within interpreter code |
|
1514 Interpreter::_throw_exception_entry = __ pc(); |
|
1515 __ verify_thread(); |
|
1516 // expression stack is undefined here |
|
1517 // O0: exception, i.e. Oexception |
|
1518 // Lbcp: exception bcp |
|
1519 __ verify_oop(Oexception); |
|
1520 |
|
1521 |
|
1522 // expression stack must be empty before entering the VM in case of an exception |
|
1523 __ empty_expression_stack(); |
|
1524 // find exception handler address and preserve exception oop |
|
1525 // call C routine to find handler and jump to it |
|
1526 __ call_VM(O1, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Oexception); |
|
1527 __ push_ptr(O1); // push exception for exception handler bytecodes |
|
1528 |
|
1529 __ JMP(O0, 0); // jump to exception handler (may be remove activation entry!) |
|
1530 __ delayed()->nop(); |
|
1531 |
|
1532 |
|
1533 // if the exception is not handled in the current frame |
|
1534 // the frame is removed and the exception is rethrown |
|
1535 // (i.e. exception continuation is _rethrow_exception) |
|
1536 // |
|
1537 // Note: At this point the bci is still the bxi for the instruction which caused |
|
1538 // the exception and the expression stack is empty. Thus, for any VM calls |
|
1539 // at this point, GC will find a legal oop map (with empty expression stack). |
|
1540 |
|
1541 // in current activation |
|
1542 // tos: exception |
|
1543 // Lbcp: exception bcp |
|
1544 |
|
1545 // |
|
1546 // JVMTI PopFrame support |
|
1547 // |
|
1548 |
|
1549 Interpreter::_remove_activation_preserving_args_entry = __ pc(); |
|
1550 Address popframe_condition_addr(G2_thread, JavaThread::popframe_condition_offset()); |
|
1551 // Set the popframe_processing bit in popframe_condition indicating that we are |
|
1552 // currently handling popframe, so that call_VMs that may happen later do not trigger new |
|
1553 // popframe handling cycles. |
|
1554 |
|
1555 __ ld(popframe_condition_addr, G3_scratch); |
|
1556 __ or3(G3_scratch, JavaThread::popframe_processing_bit, G3_scratch); |
|
1557 __ stw(G3_scratch, popframe_condition_addr); |
|
1558 |
|
1559 // Empty the expression stack, as in normal exception handling |
|
1560 __ empty_expression_stack(); |
|
1561 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false); |
|
1562 |
|
1563 { |
|
1564 // Check to see whether we are returning to a deoptimized frame. |
|
1565 // (The PopFrame call ensures that the caller of the popped frame is |
|
1566 // either interpreted or compiled and deoptimizes it if compiled.) |
|
1567 // In this case, we can't call dispatch_next() after the frame is |
|
1568 // popped, but instead must save the incoming arguments and restore |
|
1569 // them after deoptimization has occurred. |
|
1570 // |
|
1571 // Note that we don't compare the return PC against the |
|
1572 // deoptimization blob's unpack entry because of the presence of |
|
1573 // adapter frames in C2. |
|
1574 Label caller_not_deoptimized; |
|
1575 __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), I7); |
|
1576 __ br_notnull_short(O0, Assembler::pt, caller_not_deoptimized); |
|
1577 |
|
1578 const Register Gtmp1 = G3_scratch; |
|
1579 const Register Gtmp2 = G1_scratch; |
|
1580 const Register RconstMethod = Gtmp1; |
|
1581 const Address constMethod(Lmethod, Method::const_offset()); |
|
1582 const Address size_of_parameters(RconstMethod, ConstMethod::size_of_parameters_offset()); |
|
1583 |
|
1584 // Compute size of arguments for saving when returning to deoptimized caller |
|
1585 __ ld_ptr(constMethod, RconstMethod); |
|
1586 __ lduh(size_of_parameters, Gtmp1); |
|
1587 __ sll(Gtmp1, Interpreter::logStackElementSize, Gtmp1); |
|
1588 __ sub(Llocals, Gtmp1, Gtmp2); |
|
1589 __ add(Gtmp2, wordSize, Gtmp2); |
|
1590 // Save these arguments |
|
1591 __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), G2_thread, Gtmp1, Gtmp2); |
|
1592 // Inform deoptimization that it is responsible for restoring these arguments |
|
1593 __ set(JavaThread::popframe_force_deopt_reexecution_bit, Gtmp1); |
|
1594 Address popframe_condition_addr(G2_thread, JavaThread::popframe_condition_offset()); |
|
1595 __ st(Gtmp1, popframe_condition_addr); |
|
1596 |
|
1597 // Return from the current method |
|
1598 // The caller's SP was adjusted upon method entry to accomodate |
|
1599 // the callee's non-argument locals. Undo that adjustment. |
|
1600 __ ret(); |
|
1601 __ delayed()->restore(I5_savedSP, G0, SP); |
|
1602 |
|
1603 __ bind(caller_not_deoptimized); |
|
1604 } |
|
1605 |
|
1606 // Clear the popframe condition flag |
|
1607 __ stw(G0 /* popframe_inactive */, popframe_condition_addr); |
|
1608 |
|
1609 // Get out of the current method (how this is done depends on the particular compiler calling |
|
1610 // convention that the interpreter currently follows) |
|
1611 // The caller's SP was adjusted upon method entry to accomodate |
|
1612 // the callee's non-argument locals. Undo that adjustment. |
|
1613 __ restore(I5_savedSP, G0, SP); |
|
1614 // The method data pointer was incremented already during |
|
1615 // call profiling. We have to restore the mdp for the current bcp. |
|
1616 if (ProfileInterpreter) { |
|
1617 __ set_method_data_pointer_for_bcp(); |
|
1618 } |
|
1619 |
|
1620 #if INCLUDE_JVMTI |
|
1621 { |
|
1622 Label L_done; |
|
1623 |
|
1624 __ ldub(Address(Lbcp, 0), G1_scratch); // Load current bytecode |
|
1625 __ cmp_and_br_short(G1_scratch, Bytecodes::_invokestatic, Assembler::notEqual, Assembler::pn, L_done); |
|
1626 |
|
1627 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. |
|
1628 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL. |
|
1629 |
|
1630 __ call_VM(G1_scratch, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), I0, Lmethod, Lbcp); |
|
1631 |
|
1632 __ br_null(G1_scratch, false, Assembler::pn, L_done); |
|
1633 __ delayed()->nop(); |
|
1634 |
|
1635 __ st_ptr(G1_scratch, Lesp, wordSize); |
|
1636 __ bind(L_done); |
|
1637 } |
|
1638 #endif // INCLUDE_JVMTI |
|
1639 |
|
1640 // Resume bytecode interpretation at the current bcp |
|
1641 __ dispatch_next(vtos); |
|
1642 // end of JVMTI PopFrame support |
|
1643 |
|
1644 Interpreter::_remove_activation_entry = __ pc(); |
|
1645 |
|
1646 // preserve exception over this code sequence (remove activation calls the vm, but oopmaps are not correct here) |
|
1647 __ pop_ptr(Oexception); // get exception |
|
1648 |
|
1649 // Intel has the following comment: |
|
1650 //// remove the activation (without doing throws on illegalMonitorExceptions) |
|
1651 // They remove the activation without checking for bad monitor state. |
|
1652 // %%% We should make sure this is the right semantics before implementing. |
|
1653 |
|
1654 __ set_vm_result(Oexception); |
|
1655 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false); |
|
1656 |
|
1657 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI); |
|
1658 |
|
1659 __ get_vm_result(Oexception); |
|
1660 __ verify_oop(Oexception); |
|
1661 |
|
1662 const int return_reg_adjustment = frame::pc_return_offset; |
|
1663 Address issuing_pc_addr(I7, return_reg_adjustment); |
|
1664 |
|
1665 // We are done with this activation frame; find out where to go next. |
|
1666 // The continuation point will be an exception handler, which expects |
|
1667 // the following registers set up: |
|
1668 // |
|
1669 // Oexception: exception |
|
1670 // Oissuing_pc: the local call that threw exception |
|
1671 // Other On: garbage |
|
1672 // In/Ln: the contents of the caller's register window |
|
1673 // |
|
1674 // We do the required restore at the last possible moment, because we |
|
1675 // need to preserve some state across a runtime call. |
|
1676 // (Remember that the caller activation is unknown--it might not be |
|
1677 // interpreted, so things like Lscratch are useless in the caller.) |
|
1678 |
|
1679 // Although the Intel version uses call_C, we can use the more |
|
1680 // compact call_VM. (The only real difference on SPARC is a |
|
1681 // harmlessly ignored [re]set_last_Java_frame, compared with |
|
1682 // the Intel code which lacks this.) |
|
1683 __ mov(Oexception, Oexception ->after_save()); // get exception in I0 so it will be on O0 after restore |
|
1684 __ add(issuing_pc_addr, Oissuing_pc->after_save()); // likewise set I1 to a value local to the caller |
|
1685 __ super_call_VM_leaf(L7_thread_cache, |
|
1686 CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), |
|
1687 G2_thread, Oissuing_pc->after_save()); |
|
1688 |
|
1689 // The caller's SP was adjusted upon method entry to accomodate |
|
1690 // the callee's non-argument locals. Undo that adjustment. |
|
1691 __ JMP(O0, 0); // return exception handler in caller |
|
1692 __ delayed()->restore(I5_savedSP, G0, SP); |
|
1693 |
|
1694 // (same old exception object is already in Oexception; see above) |
|
1695 // Note that an "issuing PC" is actually the next PC after the call |
|
1696 } |
|
1697 |
|
1698 |
|
1699 // |
|
1700 // JVMTI ForceEarlyReturn support |
|
1701 // |
|
1702 |
|
1703 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { |
|
1704 address entry = __ pc(); |
|
1705 |
|
1706 __ empty_expression_stack(); |
|
1707 __ load_earlyret_value(state); |
|
1708 |
|
1709 __ ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), G3_scratch); |
|
1710 Address cond_addr(G3_scratch, JvmtiThreadState::earlyret_state_offset()); |
|
1711 |
|
1712 // Clear the earlyret state |
|
1713 __ stw(G0 /* JvmtiThreadState::earlyret_inactive */, cond_addr); |
|
1714 |
|
1715 __ remove_activation(state, |
|
1716 /* throw_monitor_exception */ false, |
|
1717 /* install_monitor_exception */ false); |
|
1718 |
|
1719 // The caller's SP was adjusted upon method entry to accomodate |
|
1720 // the callee's non-argument locals. Undo that adjustment. |
|
1721 __ ret(); // return to caller |
|
1722 __ delayed()->restore(I5_savedSP, G0, SP); |
|
1723 |
|
1724 return entry; |
|
1725 } // end of JVMTI ForceEarlyReturn support |
|
1726 |
|
1727 |
|
1728 //------------------------------------------------------------------------------------------------------------------------ |
|
1729 // Helper for vtos entry point generation |
|
1730 |
|
1731 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) { |
|
1732 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); |
|
1733 Label L; |
|
1734 aep = __ pc(); __ push_ptr(); __ ba_short(L); |
|
1735 fep = __ pc(); __ push_f(); __ ba_short(L); |
|
1736 dep = __ pc(); __ push_d(); __ ba_short(L); |
|
1737 lep = __ pc(); __ push_l(); __ ba_short(L); |
|
1738 iep = __ pc(); __ push_i(); |
|
1739 bep = cep = sep = iep; // there aren't any |
|
1740 vep = __ pc(); __ bind(L); // fall through |
|
1741 generate_and_dispatch(t); |
|
1742 } |
|
1743 |
|
1744 // -------------------------------------------------------------------------------- |
|
1745 |
|
1746 |
|
1747 InterpreterGenerator::InterpreterGenerator(StubQueue* code) |
|
1748 : TemplateInterpreterGenerator(code) { |
|
1749 generate_all(); // down here so it can be "virtual" |
|
1750 } |
|
1751 |
|
1752 // -------------------------------------------------------------------------------- |
|
1753 |
|
1754 // Non-product code |
|
1755 #ifndef PRODUCT |
|
1756 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { |
|
1757 address entry = __ pc(); |
|
1758 |
|
1759 __ push(state); |
|
1760 __ mov(O7, Lscratch); // protect return address within interpreter |
|
1761 |
|
1762 // Pass a 0 (not used in sparc) and the top of stack to the bytecode tracer |
|
1763 __ mov( Otos_l2, G3_scratch ); |
|
1764 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), G0, Otos_l1, G3_scratch); |
|
1765 __ mov(Lscratch, O7); // restore return address |
|
1766 __ pop(state); |
|
1767 __ retl(); |
|
1768 __ delayed()->nop(); |
|
1769 |
|
1770 return entry; |
|
1771 } |
|
1772 |
|
1773 |
|
1774 // helpers for generate_and_dispatch |
|
1775 |
|
1776 void TemplateInterpreterGenerator::count_bytecode() { |
|
1777 __ inc_counter(&BytecodeCounter::_counter_value, G3_scratch, G4_scratch); |
|
1778 } |
|
1779 |
|
1780 |
|
1781 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { |
|
1782 __ inc_counter(&BytecodeHistogram::_counters[t->bytecode()], G3_scratch, G4_scratch); |
|
1783 } |
|
1784 |
|
1785 |
|
1786 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { |
|
1787 AddressLiteral index (&BytecodePairHistogram::_index); |
|
1788 AddressLiteral counters((address) &BytecodePairHistogram::_counters); |
|
1789 |
|
1790 // get index, shift out old bytecode, bring in new bytecode, and store it |
|
1791 // _index = (_index >> log2_number_of_codes) | |
|
1792 // (bytecode << log2_number_of_codes); |
|
1793 |
|
1794 __ load_contents(index, G4_scratch); |
|
1795 __ srl( G4_scratch, BytecodePairHistogram::log2_number_of_codes, G4_scratch ); |
|
1796 __ set( ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes, G3_scratch ); |
|
1797 __ or3( G3_scratch, G4_scratch, G4_scratch ); |
|
1798 __ store_contents(G4_scratch, index, G3_scratch); |
|
1799 |
|
1800 // bump bucket contents |
|
1801 // _counters[_index] ++; |
|
1802 |
|
1803 __ set(counters, G3_scratch); // loads into G3_scratch |
|
1804 __ sll( G4_scratch, LogBytesPerWord, G4_scratch ); // Index is word address |
|
1805 __ add (G3_scratch, G4_scratch, G3_scratch); // Add in index |
|
1806 __ ld (G3_scratch, 0, G4_scratch); |
|
1807 __ inc (G4_scratch); |
|
1808 __ st (G4_scratch, 0, G3_scratch); |
|
1809 } |
|
1810 |
|
1811 |
|
1812 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { |
|
1813 // Call a little run-time stub to avoid blow-up for each bytecode. |
|
1814 // The run-time runtime saves the right registers, depending on |
|
1815 // the tosca in-state for the given template. |
|
1816 address entry = Interpreter::trace_code(t->tos_in()); |
|
1817 guarantee(entry != NULL, "entry must have been generated"); |
|
1818 __ call(entry, relocInfo::none); |
|
1819 __ delayed()->nop(); |
|
1820 } |
|
1821 |
|
1822 |
|
1823 void TemplateInterpreterGenerator::stop_interpreter_at() { |
|
1824 AddressLiteral counter(&BytecodeCounter::_counter_value); |
|
1825 __ load_contents(counter, G3_scratch); |
|
1826 AddressLiteral stop_at(&StopInterpreterAt); |
|
1827 __ load_ptr_contents(stop_at, G4_scratch); |
|
1828 __ cmp(G3_scratch, G4_scratch); |
|
1829 __ breakpoint_trap(Assembler::equal, Assembler::icc); |
|
1830 } |
|
1831 #endif // not PRODUCT |
|
1832 #endif // !CC_INTERP |