1 /*
2 * Copyright (c) 2014, 2017, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2017 SAP SE. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "interpreter/bytecodeHistogram.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/interp_masm.hpp"
32 #include "interpreter/templateInterpreterGenerator.hpp"
33 #include "interpreter/templateTable.hpp"
34 #include "oops/arrayOop.hpp"
35 #include "oops/methodData.hpp"
36 #include "oops/method.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "prims/jvmtiExport.hpp"
39 #include "prims/jvmtiThreadState.hpp"
40 #include "runtime/arguments.hpp"
41 #include "runtime/deoptimization.hpp"
42 #include "runtime/frame.inline.hpp"
43 #include "runtime/sharedRuntime.hpp"
44 #include "runtime/stubRoutines.hpp"
45 #include "runtime/synchronizer.hpp"
46 #include "runtime/timer.hpp"
47 #include "runtime/vframeArray.hpp"
48 #include "utilities/debug.hpp"
49 #include "utilities/macros.hpp"
50
51 #undef __
52 #define __ _masm->
53
54 // Size of interpreter code. Increase if too small. Interpreter will
55 // fail with a guarantee ("not enough space for interpreter generation");
56 // if too small.
57 // Run with +PrintInterpreter to get the VM to print out the size.
58 // Max size with JVMTI
59 int TemplateInterpreter::InterpreterCodeSize = 230*K;
60
61 #ifdef PRODUCT
62 #define BLOCK_COMMENT(str) /* nothing */
63 #else
64 #define BLOCK_COMMENT(str) __ block_comment(str)
65 #endif
66
67 #define BIND(label) __ bind(label); BLOCK_COMMENT(#label ":")
68
69 //-----------------------------------------------------------------------------
70
71 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
72 // Slow_signature handler that respects the PPC C calling conventions.
73 //
74 // We get called by the native entry code with our output register
75 // area == 8. First we call InterpreterRuntime::get_result_handler
76 // to copy the pointer to the signature string temporarily to the
77 // first C-argument and to return the result_handler in
78 // R3_RET. Since native_entry will copy the jni-pointer to the
79 // first C-argument slot later on, it is OK to occupy this slot
80 // temporarilly. Then we copy the argument list on the java
81 // expression stack into native varargs format on the native stack
82 // and load arguments into argument registers. Integer arguments in
83 // the varargs vector will be sign-extended to 8 bytes.
84 //
85 // On entry:
86 // R3_ARG1 - intptr_t* Address of java argument list in memory.
87 // R15_prev_state - BytecodeInterpreter* Address of interpreter state for
88 // this method
89 // R19_method
90 //
91 // On exit (just before return instruction):
92 // R3_RET - contains the address of the result_handler.
93 // R4_ARG2 - is not updated for static methods and contains "this" otherwise.
94 // R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double,
95 // ARGi contains this argument. Otherwise, ARGi is not updated.
96 // F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double.
97
98 const int LogSizeOfTwoInstructions = 3;
99
100 // FIXME: use Argument:: GL: Argument names different numbers!
101 const int max_fp_register_arguments = 13;
102 const int max_int_register_arguments = 6; // first 2 are reserved
103
104 const Register arg_java = R21_tmp1;
105 const Register arg_c = R22_tmp2;
106 const Register signature = R23_tmp3; // is string
107 const Register sig_byte = R24_tmp4;
108 const Register fpcnt = R25_tmp5;
109 const Register argcnt = R26_tmp6;
110 const Register intSlot = R27_tmp7;
111 const Register target_sp = R28_tmp8;
112 const FloatRegister floatSlot = F0;
113
114 address entry = __ function_entry();
115
116 __ save_LR_CR(R0);
117 __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
118 // We use target_sp for storing arguments in the C frame.
119 __ mr(target_sp, R1_SP);
120 __ push_frame_reg_args_nonvolatiles(0, R11_scratch1);
121
122 __ mr(arg_java, R3_ARG1);
123
124 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method);
125
126 // Signature is in R3_RET. Signature is callee saved.
127 __ mr(signature, R3_RET);
128
129 // Get the result handler.
130 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method);
131
132 {
133 Label L;
134 // test if static
135 // _access_flags._flags must be at offset 0.
136 // TODO PPC port: requires change in shared code.
137 //assert(in_bytes(AccessFlags::flags_offset()) == 0,
138 // "MethodDesc._access_flags == MethodDesc._access_flags._flags");
139 // _access_flags must be a 32 bit value.
140 assert(sizeof(AccessFlags) == 4, "wrong size");
141 __ lwa(R11_scratch1/*access_flags*/, method_(access_flags));
142 // testbit with condition register.
143 __ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT);
144 __ btrue(CCR0, L);
145 // For non-static functions, pass "this" in R4_ARG2 and copy it
146 // to 2nd C-arg slot.
147 // We need to box the Java object here, so we use arg_java
148 // (address of current Java stack slot) as argument and don't
149 // dereference it as in case of ints, floats, etc.
150 __ mr(R4_ARG2, arg_java);
151 __ addi(arg_java, arg_java, -BytesPerWord);
152 __ std(R4_ARG2, _abi(carg_2), target_sp);
153 __ bind(L);
154 }
155
156 // Will be incremented directly after loop_start. argcnt=0
157 // corresponds to 3rd C argument.
158 __ li(argcnt, -1);
159 // arg_c points to 3rd C argument
160 __ addi(arg_c, target_sp, _abi(carg_3));
161 // no floating-point args parsed so far
162 __ li(fpcnt, 0);
163
164 Label move_intSlot_to_ARG, move_floatSlot_to_FARG;
165 Label loop_start, loop_end;
166 Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed;
167
168 // signature points to '(' at entry
169 #ifdef ASSERT
170 __ lbz(sig_byte, 0, signature);
171 __ cmplwi(CCR0, sig_byte, '(');
172 __ bne(CCR0, do_dontreachhere);
173 #endif
174
175 __ bind(loop_start);
176
177 __ addi(argcnt, argcnt, 1);
178 __ lbzu(sig_byte, 1, signature);
179
180 __ cmplwi(CCR0, sig_byte, ')'); // end of signature
181 __ beq(CCR0, loop_end);
182
183 __ cmplwi(CCR0, sig_byte, 'B'); // byte
184 __ beq(CCR0, do_int);
185
186 __ cmplwi(CCR0, sig_byte, 'C'); // char
187 __ beq(CCR0, do_int);
188
189 __ cmplwi(CCR0, sig_byte, 'D'); // double
190 __ beq(CCR0, do_double);
191
192 __ cmplwi(CCR0, sig_byte, 'F'); // float
193 __ beq(CCR0, do_float);
194
195 __ cmplwi(CCR0, sig_byte, 'I'); // int
196 __ beq(CCR0, do_int);
197
198 __ cmplwi(CCR0, sig_byte, 'J'); // long
199 __ beq(CCR0, do_long);
200
201 __ cmplwi(CCR0, sig_byte, 'S'); // short
202 __ beq(CCR0, do_int);
203
204 __ cmplwi(CCR0, sig_byte, 'Z'); // boolean
205 __ beq(CCR0, do_int);
206
207 __ cmplwi(CCR0, sig_byte, 'L'); // object
208 __ beq(CCR0, do_object);
209
210 __ cmplwi(CCR0, sig_byte, '['); // array
211 __ beq(CCR0, do_array);
212
213 // __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type
214 // __ beq(CCR0, do_void);
215
216 __ bind(do_dontreachhere);
217
218 __ unimplemented("ShouldNotReachHere in slow_signature_handler", 120);
219
220 __ bind(do_array);
221
222 {
223 Label start_skip, end_skip;
224
225 __ bind(start_skip);
226 __ lbzu(sig_byte, 1, signature);
227 __ cmplwi(CCR0, sig_byte, '[');
228 __ beq(CCR0, start_skip); // skip further brackets
229 __ cmplwi(CCR0, sig_byte, '9');
230 __ bgt(CCR0, end_skip); // no optional size
231 __ cmplwi(CCR0, sig_byte, '0');
232 __ bge(CCR0, start_skip); // skip optional size
233 __ bind(end_skip);
234
235 __ cmplwi(CCR0, sig_byte, 'L');
236 __ beq(CCR0, do_object); // for arrays of objects, the name of the object must be skipped
237 __ b(do_boxed); // otherwise, go directly to do_boxed
238 }
239
240 __ bind(do_object);
241 {
242 Label L;
243 __ bind(L);
244 __ lbzu(sig_byte, 1, signature);
245 __ cmplwi(CCR0, sig_byte, ';');
246 __ bne(CCR0, L);
247 }
248 // Need to box the Java object here, so we use arg_java (address of
249 // current Java stack slot) as argument and don't dereference it as
250 // in case of ints, floats, etc.
251 Label do_null;
252 __ bind(do_boxed);
253 __ ld(R0,0, arg_java);
254 __ cmpdi(CCR0, R0, 0);
255 __ li(intSlot,0);
256 __ beq(CCR0, do_null);
257 __ mr(intSlot, arg_java);
258 __ bind(do_null);
259 __ std(intSlot, 0, arg_c);
260 __ addi(arg_java, arg_java, -BytesPerWord);
261 __ addi(arg_c, arg_c, BytesPerWord);
262 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
263 __ blt(CCR0, move_intSlot_to_ARG);
264 __ b(loop_start);
265
266 __ bind(do_int);
267 __ lwa(intSlot, 0, arg_java);
268 __ std(intSlot, 0, arg_c);
269 __ addi(arg_java, arg_java, -BytesPerWord);
270 __ addi(arg_c, arg_c, BytesPerWord);
271 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
272 __ blt(CCR0, move_intSlot_to_ARG);
273 __ b(loop_start);
274
275 __ bind(do_long);
276 __ ld(intSlot, -BytesPerWord, arg_java);
277 __ std(intSlot, 0, arg_c);
278 __ addi(arg_java, arg_java, - 2 * BytesPerWord);
279 __ addi(arg_c, arg_c, BytesPerWord);
280 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
281 __ blt(CCR0, move_intSlot_to_ARG);
282 __ b(loop_start);
283
284 __ bind(do_float);
285 __ lfs(floatSlot, 0, arg_java);
286 #if defined(LINUX)
287 // Linux uses ELF ABI. Both original ELF and ELFv2 ABIs have float
288 // in the least significant word of an argument slot.
289 #if defined(VM_LITTLE_ENDIAN)
290 __ stfs(floatSlot, 0, arg_c);
291 #else
292 __ stfs(floatSlot, 4, arg_c);
293 #endif
294 #elif defined(AIX)
295 // Although AIX runs on big endian CPU, float is in most significant
296 // word of an argument slot.
297 __ stfs(floatSlot, 0, arg_c);
298 #else
299 #error "unknown OS"
300 #endif
301 __ addi(arg_java, arg_java, -BytesPerWord);
302 __ addi(arg_c, arg_c, BytesPerWord);
303 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
304 __ blt(CCR0, move_floatSlot_to_FARG);
305 __ b(loop_start);
306
307 __ bind(do_double);
308 __ lfd(floatSlot, - BytesPerWord, arg_java);
309 __ stfd(floatSlot, 0, arg_c);
310 __ addi(arg_java, arg_java, - 2 * BytesPerWord);
311 __ addi(arg_c, arg_c, BytesPerWord);
312 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
313 __ blt(CCR0, move_floatSlot_to_FARG);
314 __ b(loop_start);
315
316 __ bind(loop_end);
317
318 __ pop_frame();
319 __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
320 __ restore_LR_CR(R0);
321
322 __ blr();
323
324 Label move_int_arg, move_float_arg;
325 __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
326 __ mr(R5_ARG3, intSlot); __ b(loop_start);
327 __ mr(R6_ARG4, intSlot); __ b(loop_start);
328 __ mr(R7_ARG5, intSlot); __ b(loop_start);
329 __ mr(R8_ARG6, intSlot); __ b(loop_start);
330 __ mr(R9_ARG7, intSlot); __ b(loop_start);
331 __ mr(R10_ARG8, intSlot); __ b(loop_start);
332
333 __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
334 __ fmr(F1_ARG1, floatSlot); __ b(loop_start);
335 __ fmr(F2_ARG2, floatSlot); __ b(loop_start);
336 __ fmr(F3_ARG3, floatSlot); __ b(loop_start);
337 __ fmr(F4_ARG4, floatSlot); __ b(loop_start);
338 __ fmr(F5_ARG5, floatSlot); __ b(loop_start);
339 __ fmr(F6_ARG6, floatSlot); __ b(loop_start);
340 __ fmr(F7_ARG7, floatSlot); __ b(loop_start);
341 __ fmr(F8_ARG8, floatSlot); __ b(loop_start);
342 __ fmr(F9_ARG9, floatSlot); __ b(loop_start);
343 __ fmr(F10_ARG10, floatSlot); __ b(loop_start);
344 __ fmr(F11_ARG11, floatSlot); __ b(loop_start);
345 __ fmr(F12_ARG12, floatSlot); __ b(loop_start);
346 __ fmr(F13_ARG13, floatSlot); __ b(loop_start);
347
348 __ bind(move_intSlot_to_ARG);
349 __ sldi(R0, argcnt, LogSizeOfTwoInstructions);
350 __ load_const(R11_scratch1, move_int_arg); // Label must be bound here.
351 __ add(R11_scratch1, R0, R11_scratch1);
352 __ mtctr(R11_scratch1/*branch_target*/);
353 __ bctr();
354 __ bind(move_floatSlot_to_FARG);
355 __ sldi(R0, fpcnt, LogSizeOfTwoInstructions);
356 __ addi(fpcnt, fpcnt, 1);
357 __ load_const(R11_scratch1, move_float_arg); // Label must be bound here.
358 __ add(R11_scratch1, R0, R11_scratch1);
359 __ mtctr(R11_scratch1/*branch_target*/);
360 __ bctr();
361
362 return entry;
363 }
364
365 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
366 //
367 // Registers alive
368 // R3_RET
369 // LR
370 //
371 // Registers updated
372 // R3_RET
373 //
374
375 Label done;
376 address entry = __ pc();
377
378 switch (type) {
379 case T_BOOLEAN:
380 // convert !=0 to 1
381 __ neg(R0, R3_RET);
382 __ orr(R0, R3_RET, R0);
383 __ srwi(R3_RET, R0, 31);
384 break;
385 case T_BYTE:
386 // sign extend 8 bits
387 __ extsb(R3_RET, R3_RET);
388 break;
389 case T_CHAR:
390 // zero extend 16 bits
391 __ clrldi(R3_RET, R3_RET, 48);
392 break;
393 case T_SHORT:
394 // sign extend 16 bits
395 __ extsh(R3_RET, R3_RET);
396 break;
397 case T_INT:
398 // sign extend 32 bits
399 __ extsw(R3_RET, R3_RET);
400 break;
401 case T_LONG:
402 break;
403 case T_OBJECT:
404 // JNIHandles::resolve result.
405 __ resolve_jobject(R3_RET, R11_scratch1, R12_scratch2, /* needs_frame */ true); // kills R31
406 break;
407 case T_FLOAT:
408 break;
409 case T_DOUBLE:
410 break;
411 case T_VOID:
412 break;
413 default: ShouldNotReachHere();
414 }
415
416 BIND(done);
417 __ blr();
418
419 return entry;
420 }
421
422 // Abstract method entry.
423 //
424 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
425 address entry = __ pc();
426
427 //
428 // Registers alive
429 // R16_thread - JavaThread*
430 // R19_method - callee's method (method to be invoked)
431 // R1_SP - SP prepared such that caller's outgoing args are near top
432 // LR - return address to caller
433 //
434 // Stack layout at this point:
435 //
436 // 0 [TOP_IJAVA_FRAME_ABI] <-- R1_SP
437 // alignment (optional)
438 // [outgoing Java arguments]
439 // ...
440 // PARENT [PARENT_IJAVA_FRAME_ABI]
441 // ...
442 //
443
444 // Can't use call_VM here because we have not set up a new
445 // interpreter state. Make the call to the vm and make it look like
446 // our caller set up the JavaFrameAnchor.
447 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
448
449 // Push a new C frame and save LR.
450 __ save_LR_CR(R0);
451 __ push_frame_reg_args(0, R11_scratch1);
452
453 // This is not a leaf but we have a JavaFrameAnchor now and we will
454 // check (create) exceptions afterward so this is ok.
455 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError),
456 R16_thread);
457
458 // Pop the C frame and restore LR.
459 __ pop_frame();
460 __ restore_LR_CR(R0);
461
462 // Reset JavaFrameAnchor from call_VM_leaf above.
463 __ reset_last_Java_frame();
464
465 // We don't know our caller, so jump to the general forward exception stub,
466 // which will also pop our full frame off. Satisfy the interface of
467 // SharedRuntime::generate_forward_exception()
468 __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0);
469 __ mtctr(R11_scratch1);
470 __ bctr();
471
472 return entry;
473 }
474
475 // Interpreter intrinsic for WeakReference.get().
476 // 1. Don't push a full blown frame and go on dispatching, but fetch the value
477 // into R8 and return quickly
478 // 2. If G1 is active we *must* execute this intrinsic for corrrectness:
479 // It contains a GC barrier which puts the reference into the satb buffer
480 // to indicate that someone holds a strong reference to the object the
481 // weak ref points to!
482 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
483 // Code: _aload_0, _getfield, _areturn
484 // parameter size = 1
485 //
486 // The code that gets generated by this routine is split into 2 parts:
487 // 1. the "intrinsified" code for G1 (or any SATB based GC),
488 // 2. the slow path - which is an expansion of the regular method entry.
489 //
490 // Notes:
491 // * In the G1 code we do not check whether we need to block for
492 // a safepoint. If G1 is enabled then we must execute the specialized
493 // code for Reference.get (except when the Reference object is null)
494 // so that we can log the value in the referent field with an SATB
495 // update buffer.
496 // If the code for the getfield template is modified so that the
497 // G1 pre-barrier code is executed when the current method is
498 // Reference.get() then going through the normal method entry
499 // will be fine.
500 // * The G1 code can, however, check the receiver object (the instance
501 // of java.lang.Reference) and jump to the slow path if null. If the
502 // Reference object is null then we obviously cannot fetch the referent
503 // and so we don't need to call the G1 pre-barrier. Thus we can use the
504 // regular method entry code to generate the NPE.
505 //
506
507 if (UseG1GC) {
508 address entry = __ pc();
509
510 const int referent_offset = java_lang_ref_Reference::referent_offset;
511 guarantee(referent_offset > 0, "referent offset not initialized");
512
513 Label slow_path;
514
515 // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH);
516
517 // In the G1 code we don't check if we need to reach a safepoint. We
518 // continue and the thread will safepoint at the next bytecode dispatch.
519
520 // If the receiver is null then it is OK to jump to the slow path.
521 __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver
522
523 // Check if receiver == NULL and go the slow path.
524 __ cmpdi(CCR0, R3_RET, 0);
525 __ beq(CCR0, slow_path);
526
527 // Load the value of the referent field.
528 __ load_heap_oop(R3_RET, referent_offset, R3_RET);
529
530 // Generate the G1 pre-barrier code to log the value of
531 // the referent field in an SATB buffer. Note with
532 // these parameters the pre-barrier does not generate
533 // the load of the previous value.
534
535 // Restore caller sp for c2i case.
536 #ifdef ASSERT
537 __ ld(R9_ARG7, 0, R1_SP);
538 __ ld(R10_ARG8, 0, R21_sender_SP);
539 __ cmpd(CCR0, R9_ARG7, R10_ARG8);
540 __ asm_assert_eq("backlink", 0x544);
541 #endif // ASSERT
542 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
543
544 __ g1_write_barrier_pre(noreg, // obj
545 noreg, // offset
546 R3_RET, // pre_val
547 R11_scratch1, // tmp
548 R12_scratch2, // tmp
549 true); // needs_frame
550
551 __ blr();
552
553 // Generate regular method entry.
554 __ bind(slow_path);
555 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
556 return entry;
557 }
558
559 return NULL;
560 }
561
562 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
563 address entry = __ pc();
564
565 // Expression stack must be empty before entering the VM if an
566 // exception happened.
567 __ empty_expression_stack();
568 // Throw exception.
569 __ call_VM(noreg,
570 CAST_FROM_FN_PTR(address,
571 InterpreterRuntime::throw_StackOverflowError));
572 return entry;
573 }
574
575 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
576 address entry = __ pc();
577 __ empty_expression_stack();
578 __ load_const_optimized(R4_ARG2, (address) name);
579 // Index is in R17_tos.
580 __ mr(R5_ARG3, R17_tos);
581 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException));
582 return entry;
583 }
584
585 #if 0
586 // Call special ClassCastException constructor taking object to cast
587 // and target class as arguments.
588 address TemplateInterpreterGenerator::generate_ClassCastException_verbose_handler() {
589 address entry = __ pc();
590
591 // Expression stack must be empty before entering the VM if an
592 // exception happened.
593 __ empty_expression_stack();
594
595 // Thread will be loaded to R3_ARG1.
596 // Target class oop is in register R5_ARG3 by convention!
597 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose), R17_tos, R5_ARG3);
598 // Above call must not return here since exception pending.
599 DEBUG_ONLY(__ should_not_reach_here();)
600 return entry;
601 }
602 #endif
603
604 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
605 address entry = __ pc();
606 // Expression stack must be empty before entering the VM if an
607 // exception happened.
608 __ empty_expression_stack();
609
610 // Load exception object.
611 // Thread will be loaded to R3_ARG1.
612 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
613 #ifdef ASSERT
614 // Above call must not return here since exception pending.
615 __ should_not_reach_here();
616 #endif
617 return entry;
618 }
619
620 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
621 address entry = __ pc();
622 //__ untested("generate_exception_handler_common");
623 Register Rexception = R17_tos;
624
625 // Expression stack must be empty before entering the VM if an exception happened.
626 __ empty_expression_stack();
627
628 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
629 if (pass_oop) {
630 __ mr(R5_ARG3, Rexception);
631 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false);
632 } else {
633 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
634 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false);
635 }
636
637 // Throw exception.
638 __ mr(R3_ARG1, Rexception);
639 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
640 __ mtctr(R11_scratch1);
641 __ bctr();
642
643 return entry;
644 }
645
646 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
647 address entry = __ pc();
648 __ unimplemented("generate_continuation_for");
649 return entry;
650 }
651
652 // This entry is returned to when a call returns to the interpreter.
653 // When we arrive here, we expect that the callee stack frame is already popped.
654 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
655 address entry = __ pc();
656
657 // Move the value out of the return register back to the TOS cache of current frame.
658 switch (state) {
659 case ltos:
660 case btos:
661 case ztos:
662 case ctos:
663 case stos:
664 case atos:
665 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache
666 case ftos:
667 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
668 case vtos: break; // Nothing to do, this was a void return.
669 default : ShouldNotReachHere();
670 }
671
672 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
673 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
674 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
675
676 // Compiled code destroys templateTableBase, reload.
677 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
678
679 if (state == atos) {
680 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
681 }
682
683 const Register cache = R11_scratch1;
684 const Register size = R12_scratch2;
685 __ get_cache_and_index_at_bcp(cache, 1, index_size);
686
687 // Get least significant byte of 64 bit value:
688 #if defined(VM_LITTLE_ENDIAN)
689 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache);
690 #else
691 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
692 #endif
693 __ sldi(size, size, Interpreter::logStackElementSize);
694 __ add(R15_esp, R15_esp, size);
695 __ dispatch_next(state, step);
696 return entry;
697 }
698
699 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
700 address entry = __ pc();
701 // If state != vtos, we're returning from a native method, which put it's result
702 // into the result register. So move the value out of the return register back
703 // to the TOS cache of current frame.
704
705 switch (state) {
706 case ltos:
707 case btos:
708 case ztos:
709 case ctos:
710 case stos:
711 case atos:
712 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache
713 case ftos:
714 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
715 case vtos: break; // Nothing to do, this was a void return.
716 default : ShouldNotReachHere();
717 }
718
719 // Load LcpoolCache @@@ should be already set!
720 __ get_constant_pool_cache(R27_constPoolCache);
721
722 // Handle a pending exception, fall through if none.
723 __ check_and_forward_exception(R11_scratch1, R12_scratch2);
724
725 // Start executing bytecodes.
726 __ dispatch_next(state, step);
727
728 return entry;
729 }
730
731 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
732 address entry = __ pc();
733
734 __ push(state);
735 __ call_VM(noreg, runtime_entry);
736 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
737
738 return entry;
739 }
740
741 // Helpers for commoning out cases in the various type of method entries.
742
743 // Increment invocation count & check for overflow.
744 //
745 // Note: checking for negative value instead of overflow
746 // so we have a 'sticky' overflow test.
747 //
748 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
749 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
750 Register Rscratch1 = R11_scratch1;
751 Register Rscratch2 = R12_scratch2;
752 Register R3_counters = R3_ARG1;
753 Label done;
754
755 if (TieredCompilation) {
756 const int increment = InvocationCounter::count_increment;
757 Label no_mdo;
758 if (ProfileInterpreter) {
759 const Register Rmdo = R3_counters;
760 // If no method data exists, go to profile_continue.
761 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
762 __ cmpdi(CCR0, Rmdo, 0);
763 __ beq(CCR0, no_mdo);
764
765 // Increment invocation counter in the MDO.
766 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
767 __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
768 __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo);
769 __ addi(Rscratch2, Rscratch2, increment);
770 __ stw(Rscratch2, mdo_ic_offs, Rmdo);
771 __ and_(Rscratch1, Rscratch2, Rscratch1);
772 __ bne(CCR0, done);
773 __ b(*overflow);
774 }
775
776 // Increment counter in MethodCounters*.
777 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
778 __ bind(no_mdo);
779 __ get_method_counters(R19_method, R3_counters, done);
780 __ lwz(Rscratch2, mo_ic_offs, R3_counters);
781 __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters);
782 __ addi(Rscratch2, Rscratch2, increment);
783 __ stw(Rscratch2, mo_ic_offs, R3_counters);
784 __ and_(Rscratch1, Rscratch2, Rscratch1);
785 __ beq(CCR0, *overflow);
786
787 __ bind(done);
788
789 } else {
790
791 // Update standard invocation counters.
792 Register Rsum_ivc_bec = R4_ARG2;
793 __ get_method_counters(R19_method, R3_counters, done);
794 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2);
795 // Increment interpreter invocation counter.
796 if (ProfileInterpreter) { // %%% Merge this into methodDataOop.
797 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
798 __ addi(R12_scratch2, R12_scratch2, 1);
799 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
800 }
801 // Check if we must create a method data obj.
802 if (ProfileInterpreter && profile_method != NULL) {
803 const Register profile_limit = Rscratch1;
804 __ lwz(profile_limit, in_bytes(MethodCounters::interpreter_profile_limit_offset()), R3_counters);
805 // Test to see if we should create a method data oop.
806 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit);
807 __ blt(CCR0, *profile_method_continue);
808 // If no method data exists, go to profile_method.
809 __ test_method_data_pointer(*profile_method);
810 }
811 // Finally check for counter overflow.
812 if (overflow) {
813 const Register invocation_limit = Rscratch1;
814 __ lwz(invocation_limit, in_bytes(MethodCounters::interpreter_invocation_limit_offset()), R3_counters);
815 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit);
816 __ bge(CCR0, *overflow);
817 }
818
819 __ bind(done);
820 }
821 }
822
823 // Generate code to initiate compilation on invocation counter overflow.
824 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
825 // Generate code to initiate compilation on the counter overflow.
826
827 // InterpreterRuntime::frequency_counter_overflow takes one arguments,
828 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp)
829 // We pass zero in.
830 // The call returns the address of the verified entry point for the method or NULL
831 // if the compilation did not complete (either went background or bailed out).
832 //
833 // Unlike the C++ interpreter above: Check exceptions!
834 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
835 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
836
837 __ li(R4_ARG2, 0);
838 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
839
840 // Returns verified_entry_point or NULL.
841 // We ignore it in any case.
842 __ b(continue_entry);
843 }
844
845 // See if we've got enough room on the stack for locals plus overhead below
846 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
847 // without going through the signal handler, i.e., reserved and yellow zones
848 // will not be made usable. The shadow zone must suffice to handle the
849 // overflow.
850 //
851 // Kills Rmem_frame_size, Rscratch1.
852 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
853 Label done;
854 assert_different_registers(Rmem_frame_size, Rscratch1);
855
856 BLOCK_COMMENT("stack_overflow_check_with_compare {");
857 __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
858 __ ld(Rscratch1, thread_(stack_overflow_limit));
859 __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
860 __ bgt(CCR0/*is_stack_overflow*/, done);
861
862 // The stack overflows. Load target address of the runtime stub and call it.
863 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order");
864 __ load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0);
865 __ mtctr(Rscratch1);
866 // Restore caller_sp.
867 #ifdef ASSERT
868 __ ld(Rscratch1, 0, R1_SP);
869 __ ld(R0, 0, R21_sender_SP);
870 __ cmpd(CCR0, R0, Rscratch1);
871 __ asm_assert_eq("backlink", 0x547);
872 #endif // ASSERT
873 __ mr(R1_SP, R21_sender_SP);
874 __ bctr();
875
876 __ align(32, 12);
877 __ bind(done);
878 BLOCK_COMMENT("} stack_overflow_check_with_compare");
879 }
880
881 // Lock the current method, interpreter register window must be set up!
882 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
883 const Register Robj_to_lock = Rscratch2;
884
885 {
886 if (!flags_preloaded) {
887 __ lwz(Rflags, method_(access_flags));
888 }
889
890 #ifdef ASSERT
891 // Check if methods needs synchronization.
892 {
893 Label Lok;
894 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
895 __ btrue(CCR0,Lok);
896 __ stop("method doesn't need synchronization");
897 __ bind(Lok);
898 }
899 #endif // ASSERT
900 }
901
902 // Get synchronization object to Rscratch2.
903 {
904 Label Lstatic;
905 Label Ldone;
906
907 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
908 __ btrue(CCR0, Lstatic);
909
910 // Non-static case: load receiver obj from stack and we're done.
911 __ ld(Robj_to_lock, R18_locals);
912 __ b(Ldone);
913
914 __ bind(Lstatic); // Static case: Lock the java mirror
915 // Load mirror from interpreter frame.
916 __ ld(Robj_to_lock, _abi(callers_sp), R1_SP);
917 __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock);
918
919 __ bind(Ldone);
920 __ verify_oop(Robj_to_lock);
921 }
922
923 // Got the oop to lock => execute!
924 __ add_monitor_to_stack(true, Rscratch1, R0);
925
926 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
927 __ lock_object(R26_monitor, Robj_to_lock);
928 }
929
930 // Generate a fixed interpreter frame for pure interpreter
931 // and I2N native transition frames.
932 //
933 // Before (stack grows downwards):
934 //
935 // | ... |
936 // |------------- |
937 // | java arg0 |
938 // | ... |
939 // | java argn |
940 // | | <- R15_esp
941 // | |
942 // |--------------|
943 // | abi_112 |
944 // | | <- R1_SP
945 // |==============|
946 //
947 //
948 // After:
949 //
950 // | ... |
951 // | java arg0 |<- R18_locals
952 // | ... |
953 // | java argn |
954 // |--------------|
955 // | |
956 // | java locals |
957 // | |
958 // |--------------|
959 // | abi_48 |
960 // |==============|
961 // | |
962 // | istate |
963 // | |
964 // |--------------|
965 // | monitor |<- R26_monitor
966 // |--------------|
967 // | |<- R15_esp
968 // | expression |
969 // | stack |
970 // | |
971 // |--------------|
972 // | |
973 // | abi_112 |<- R1_SP
974 // |==============|
975 //
976 // The top most frame needs an abi space of 112 bytes. This space is needed,
977 // since we call to c. The c function may spill their arguments to the caller
978 // frame. When we call to java, we don't need these spill slots. In order to save
979 // space on the stack, we resize the caller. However, java locals reside in
980 // the caller frame and the frame has to be increased. The frame_size for the
981 // current frame was calculated based on max_stack as size for the expression
982 // stack. At the call, just a part of the expression stack might be used.
983 // We don't want to waste this space and cut the frame back accordingly.
984 // The resulting amount for resizing is calculated as follows:
985 // resize = (number_of_locals - number_of_arguments) * slot_size
986 // + (R1_SP - R15_esp) + 48
987 //
988 // The size for the callee frame is calculated:
989 // framesize = 112 + max_stack + monitor + state_size
990 //
991 // maxstack: Max number of slots on the expression stack, loaded from the method.
992 // monitor: We statically reserve room for one monitor object.
993 // state_size: We save the current state of the interpreter to this area.
994 //
995 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
996 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
997 top_frame_size = R7_ARG5,
998 Rconst_method = R8_ARG6;
999
1000 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size);
1001
1002 __ ld(Rconst_method, method_(const));
1003 __ lhz(Rsize_of_parameters /* number of params */,
1004 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
1005 if (native_call) {
1006 // If we're calling a native method, we reserve space for the worst-case signature
1007 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2).
1008 // We add two slots to the parameter_count, one for the jni
1009 // environment and one for a possible native mirror.
1010 Label skip_native_calculate_max_stack;
1011 __ addi(top_frame_size, Rsize_of_parameters, 2);
1012 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters);
1013 __ bge(CCR0, skip_native_calculate_max_stack);
1014 __ li(top_frame_size, Argument::n_register_parameters);
1015 __ bind(skip_native_calculate_max_stack);
1016 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
1017 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
1018 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
1019 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
1020 } else {
1021 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
1022 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
1023 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
1024 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
1025 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
1026 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
1027 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
1028 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1);
1029 }
1030
1031 // Compute top frame size.
1032 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size);
1033
1034 // Cut back area between esp and max_stack.
1035 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize);
1036
1037 __ round_to(top_frame_size, frame::alignment_in_bytes);
1038 __ round_to(parent_frame_resize, frame::alignment_in_bytes);
1039 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
1040 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
1041
1042 if (!native_call) {
1043 // Stack overflow check.
1044 // Native calls don't need the stack size check since they have no
1045 // expression stack and the arguments are already on the stack and
1046 // we only add a handful of words to the stack.
1047 __ add(R11_scratch1, parent_frame_resize, top_frame_size);
1048 generate_stack_overflow_check(R11_scratch1, R12_scratch2);
1049 }
1050
1051 // Set up interpreter state registers.
1052
1053 __ add(R18_locals, R15_esp, Rsize_of_parameters);
1054 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
1055 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache);
1056
1057 // Set method data pointer.
1058 if (ProfileInterpreter) {
1059 Label zero_continue;
1060 __ ld(R28_mdx, method_(method_data));
1061 __ cmpdi(CCR0, R28_mdx, 0);
1062 __ beq(CCR0, zero_continue);
1063 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1064 __ bind(zero_continue);
1065 }
1066
1067 if (native_call) {
1068 __ li(R14_bcp, 0); // Must initialize.
1069 } else {
1070 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method);
1071 }
1072
1073 // Resize parent frame.
1074 __ mflr(R12_scratch2);
1075 __ neg(parent_frame_resize, parent_frame_resize);
1076 __ resize_frame(parent_frame_resize, R11_scratch1);
1077 __ std(R12_scratch2, _abi(lr), R1_SP);
1078
1079 // Get mirror and store it in the frame as GC root for this Method*.
1080 __ load_mirror_from_const_method(R12_scratch2, Rconst_method);
1081
1082 __ addi(R26_monitor, R1_SP, - frame::ijava_state_size);
1083 __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
1084
1085 // Store values.
1086 // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls
1087 // in InterpreterMacroAssembler::call_from_interpreter.
1088 __ std(R19_method, _ijava_state_neg(method), R1_SP);
1089 __ std(R12_scratch2, _ijava_state_neg(mirror), R1_SP);
1090 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
1091 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
1092 __ std(R18_locals, _ijava_state_neg(locals), R1_SP);
1093
1094 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
1095 // be found in the frame after save_interpreter_state is done. This is always true
1096 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
1097 // because e.g. frame::interpreter_frame_bcp() will not access the correct value
1098 // (Enhanced Stack Trace).
1099 // The signal handler does not save the interpreter state into the frame.
1100 __ li(R0, 0);
1101 #ifdef ASSERT
1102 // Fill remaining slots with constants.
1103 __ load_const_optimized(R11_scratch1, 0x5afe);
1104 __ load_const_optimized(R12_scratch2, 0xdead);
1105 #endif
1106 // We have to initialize some frame slots for native calls (accessed by GC).
1107 if (native_call) {
1108 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP);
1109 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
1110 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
1111 }
1112 #ifdef ASSERT
1113 else {
1114 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
1115 __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP);
1116 __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP);
1117 }
1118 __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP);
1119 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
1120 __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP);
1121 __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP);
1122 #endif
1123 __ subf(R12_scratch2, top_frame_size, R1_SP);
1124 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP);
1125 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
1126
1127 // Push top frame.
1128 __ push_frame(top_frame_size, R11_scratch1);
1129 }
1130
1131 // End of helpers
1132
1133 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1134
1135 // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1136 bool use_instruction = false;
1137 address runtime_entry = NULL;
1138 int num_args = 1;
1139 bool double_precision = true;
1140
1141 // PPC64 specific:
1142 switch (kind) {
1143 case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break;
1144 case Interpreter::java_lang_math_abs: use_instruction = true; break;
1145 case Interpreter::java_lang_math_fmaF:
1146 case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1147 default: break; // Fall back to runtime call.
1148 }
1149
1150 switch (kind) {
1151 case Interpreter::java_lang_math_sin : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); break;
1152 case Interpreter::java_lang_math_cos : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); break;
1153 case Interpreter::java_lang_math_tan : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); break;
1154 case Interpreter::java_lang_math_abs : /* run interpreted */ break;
1155 case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); break;
1156 case Interpreter::java_lang_math_log : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); break;
1157 case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1158 case Interpreter::java_lang_math_pow : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1159 case Interpreter::java_lang_math_exp : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); break;
1160 case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1161 case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1162 default: ShouldNotReachHere();
1163 }
1164
1165 // Use normal entry if neither instruction nor runtime call is used.
1166 if (!use_instruction && runtime_entry == NULL) return NULL;
1167
1168 address entry = __ pc();
1169
1170 // Load arguments
1171 assert(num_args <= 13, "passed in registers");
1172 if (double_precision) {
1173 int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1174 for (int i = 0; i < num_args; ++i) {
1175 __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1176 offset -= 2 * Interpreter::stackElementSize;
1177 }
1178 } else {
1179 int offset = num_args * Interpreter::stackElementSize;
1180 for (int i = 0; i < num_args; ++i) {
1181 __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1182 offset -= Interpreter::stackElementSize;
1183 }
1184 }
1185
1186 // Pop c2i arguments (if any) off when we return.
1187 #ifdef ASSERT
1188 __ ld(R9_ARG7, 0, R1_SP);
1189 __ ld(R10_ARG8, 0, R21_sender_SP);
1190 __ cmpd(CCR0, R9_ARG7, R10_ARG8);
1191 __ asm_assert_eq("backlink", 0x545);
1192 #endif // ASSERT
1193 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
1194
1195 if (use_instruction) {
1196 switch (kind) {
1197 case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1); break;
1198 case Interpreter::java_lang_math_abs: __ fabs(F1_RET, F1); break;
1199 case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break;
1200 case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3); break;
1201 default: ShouldNotReachHere();
1202 }
1203 } else {
1204 // Comment: Can use tail call if the unextended frame is always C ABI compliant:
1205 //__ load_const_optimized(R12_scratch2, runtime_entry, R0);
1206 //__ call_c_and_return_to_caller(R12_scratch2);
1207
1208 // Push a new C frame and save LR.
1209 __ save_LR_CR(R0);
1210 __ push_frame_reg_args(0, R11_scratch1);
1211
1212 __ call_VM_leaf(runtime_entry);
1213
1214 // Pop the C frame and restore LR.
1215 __ pop_frame();
1216 __ restore_LR_CR(R0);
1217 }
1218
1219 __ blr();
1220
1221 __ flush();
1222
1223 return entry;
1224 }
1225
1226 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1227 // Quick & dirty stack overflow checking: bang the stack & handle trap.
1228 // Note that we do the banging after the frame is setup, since the exception
1229 // handling code expects to find a valid interpreter frame on the stack.
1230 // Doing the banging earlier fails if the caller frame is not an interpreter
1231 // frame.
1232 // (Also, the exception throwing code expects to unlock any synchronized
1233 // method receiever, so do the banging after locking the receiver.)
1234
1235 // Bang each page in the shadow zone. We can't assume it's been done for
1236 // an interpreter frame with greater than a page of locals, so each page
1237 // needs to be checked. Only true for non-native.
1238 if (UseStackBanging) {
1239 const int page_size = os::vm_page_size();
1240 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
1241 const int start_page = native_call ? n_shadow_pages : 1;
1242 BLOCK_COMMENT("bang_stack_shadow_pages:");
1243 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1244 __ bang_stack_with_offset(pages*page_size);
1245 }
1246 }
1247 }
1248
1249 // Interpreter stub for calling a native method. (asm interpreter)
1250 // This sets up a somewhat different looking stack for calling the
1251 // native method than the typical interpreter frame setup.
1252 //
1253 // On entry:
1254 // R19_method - method
1255 // R16_thread - JavaThread*
1256 // R15_esp - intptr_t* sender tos
1257 //
1258 // abstract stack (grows up)
1259 // [ IJava (caller of JNI callee) ] <-- ASP
1260 // ...
1261 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1262
1263 address entry = __ pc();
1264
1265 const bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1266
1267 // -----------------------------------------------------------------------------
1268 // Allocate a new frame that represents the native callee (i2n frame).
1269 // This is not a full-blown interpreter frame, but in particular, the
1270 // following registers are valid after this:
1271 // - R19_method
1272 // - R18_local (points to start of arguments to native function)
1273 //
1274 // abstract stack (grows up)
1275 // [ IJava (caller of JNI callee) ] <-- ASP
1276 // ...
1277
1278 const Register signature_handler_fd = R11_scratch1;
1279 const Register pending_exception = R0;
1280 const Register result_handler_addr = R31;
1281 const Register native_method_fd = R11_scratch1;
1282 const Register access_flags = R22_tmp2;
1283 const Register active_handles = R11_scratch1; // R26_monitor saved to state.
1284 const Register sync_state = R12_scratch2;
1285 const Register sync_state_addr = sync_state; // Address is dead after use.
1286 const Register suspend_flags = R11_scratch1;
1287
1288 //=============================================================================
1289 // Allocate new frame and initialize interpreter state.
1290
1291 Label exception_return;
1292 Label exception_return_sync_check;
1293 Label stack_overflow_return;
1294
1295 // Generate new interpreter state and jump to stack_overflow_return in case of
1296 // a stack overflow.
1297 //generate_compute_interpreter_state(stack_overflow_return);
1298
1299 Register size_of_parameters = R22_tmp2;
1300
1301 generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
1302
1303 //=============================================================================
1304 // Increment invocation counter. On overflow, entry to JNI method
1305 // will be compiled.
1306 Label invocation_counter_overflow, continue_after_compile;
1307 if (inc_counter) {
1308 if (synchronized) {
1309 // Since at this point in the method invocation the exception handler
1310 // would try to exit the monitor of synchronized methods which hasn't
1311 // been entered yet, we set the thread local variable
1312 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1313 // runtime, exception handling i.e. unlock_if_synchronized_method will
1314 // check this thread local flag.
1315 // This flag has two effects, one is to force an unwind in the topmost
1316 // interpreter frame and not perform an unlock while doing so.
1317 __ li(R0, 1);
1318 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1319 }
1320 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1321
1322 BIND(continue_after_compile);
1323 }
1324
1325 bang_stack_shadow_pages(true);
1326
1327 if (inc_counter) {
1328 // Reset the _do_not_unlock_if_synchronized flag.
1329 if (synchronized) {
1330 __ li(R0, 0);
1331 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1332 }
1333 }
1334
1335 // access_flags = method->access_flags();
1336 // Load access flags.
1337 assert(access_flags->is_nonvolatile(),
1338 "access_flags must be in a non-volatile register");
1339 // Type check.
1340 assert(4 == sizeof(AccessFlags), "unexpected field size");
1341 __ lwz(access_flags, method_(access_flags));
1342
1343 // We don't want to reload R19_method and access_flags after calls
1344 // to some helper functions.
1345 assert(R19_method->is_nonvolatile(),
1346 "R19_method must be a non-volatile register");
1347
1348 // Check for synchronized methods. Must happen AFTER invocation counter
1349 // check, so method is not locked if counter overflows.
1350
1351 if (synchronized) {
1352 lock_method(access_flags, R11_scratch1, R12_scratch2, true);
1353
1354 // Update monitor in state.
1355 __ ld(R11_scratch1, 0, R1_SP);
1356 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1);
1357 }
1358
1359 // jvmti/jvmpi support
1360 __ notify_method_entry();
1361
1362 //=============================================================================
1363 // Get and call the signature handler.
1364
1365 __ ld(signature_handler_fd, method_(signature_handler));
1366 Label call_signature_handler;
1367
1368 __ cmpdi(CCR0, signature_handler_fd, 0);
1369 __ bne(CCR0, call_signature_handler);
1370
1371 // Method has never been called. Either generate a specialized
1372 // handler or point to the slow one.
1373 //
1374 // Pass parameter 'false' to avoid exception check in call_VM.
1375 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
1376
1377 // Check for an exception while looking up the target method. If we
1378 // incurred one, bail.
1379 __ ld(pending_exception, thread_(pending_exception));
1380 __ cmpdi(CCR0, pending_exception, 0);
1381 __ bne(CCR0, exception_return_sync_check); // Has pending exception.
1382
1383 // Reload signature handler, it may have been created/assigned in the meanwhile.
1384 __ ld(signature_handler_fd, method_(signature_handler));
1385 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
1386
1387 BIND(call_signature_handler);
1388
1389 // Before we call the signature handler we push a new frame to
1390 // protect the interpreter frame volatile registers when we return
1391 // from jni but before we can get back to Java.
1392
1393 // First set the frame anchor while the SP/FP registers are
1394 // convenient and the slow signature handler can use this same frame
1395 // anchor.
1396
1397 // We have a TOP_IJAVA_FRAME here, which belongs to us.
1398 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
1399
1400 // Now the interpreter frame (and its call chain) have been
1401 // invalidated and flushed. We are now protected against eager
1402 // being enabled in native code. Even if it goes eager the
1403 // registers will be reloaded as clean and we will invalidate after
1404 // the call so no spurious flush should be possible.
1405
1406 // Call signature handler and pass locals address.
1407 //
1408 // Our signature handlers copy required arguments to the C stack
1409 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
1410 __ mr(R3_ARG1, R18_locals);
1411 #if !defined(ABI_ELFv2)
1412 __ ld(signature_handler_fd, 0, signature_handler_fd);
1413 #endif
1414
1415 __ call_stub(signature_handler_fd);
1416
1417 // Remove the register parameter varargs slots we allocated in
1418 // compute_interpreter_state. SP+16 ends up pointing to the ABI
1419 // outgoing argument area.
1420 //
1421 // Not needed on PPC64.
1422 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord);
1423
1424 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
1425 // Save across call to native method.
1426 __ mr(result_handler_addr, R3_RET);
1427
1428 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
1429
1430 // Set up fixed parameters and call the native method.
1431 // If the method is static, get mirror into R4_ARG2.
1432 {
1433 Label method_is_not_static;
1434 // Access_flags is non-volatile and still, no need to restore it.
1435
1436 // Restore access flags.
1437 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
1438 __ bfalse(CCR0, method_is_not_static);
1439
1440 __ ld(R11_scratch1, _abi(callers_sp), R1_SP);
1441 // Load mirror from interpreter frame.
1442 __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1);
1443 // R4_ARG2 = &state->_oop_temp;
1444 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
1445 __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
1446 BIND(method_is_not_static);
1447 }
1448
1449 // At this point, arguments have been copied off the stack into
1450 // their JNI positions. Oops are boxed in-place on the stack, with
1451 // handles copied to arguments. The result handler address is in a
1452 // register.
1453
1454 // Pass JNIEnv address as first parameter.
1455 __ addir(R3_ARG1, thread_(jni_environment));
1456
1457 // Load the native_method entry before we change the thread state.
1458 __ ld(native_method_fd, method_(native_function));
1459
1460 //=============================================================================
1461 // Transition from _thread_in_Java to _thread_in_native. As soon as
1462 // we make this change the safepoint code needs to be certain that
1463 // the last Java frame we established is good. The pc in that frame
1464 // just needs to be near here not an actual return address.
1465
1466 // We use release_store_fence to update values like the thread state, where
1467 // we don't want the current thread to continue until all our prior memory
1468 // accesses (including the new thread state) are visible to other threads.
1469 __ li(R0, _thread_in_native);
1470 __ release();
1471
1472 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
1473 __ stw(R0, thread_(thread_state));
1474
1475 if (UseMembar) {
1476 __ fence();
1477 }
1478
1479 //=============================================================================
1480 // Call the native method. Argument registers must not have been
1481 // overwritten since "__ call_stub(signature_handler);" (except for
1482 // ARG1 and ARG2 for static methods).
1483 __ call_c(native_method_fd);
1484
1485 __ li(R0, 0);
1486 __ ld(R11_scratch1, 0, R1_SP);
1487 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1488 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1489 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
1490
1491 // Note: C++ interpreter needs the following here:
1492 // The frame_manager_lr field, which we use for setting the last
1493 // java frame, gets overwritten by the signature handler. Restore
1494 // it now.
1495 //__ get_PC_trash_LR(R11_scratch1);
1496 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
1497
1498 // Because of GC R19_method may no longer be valid.
1499
1500 // Block, if necessary, before resuming in _thread_in_Java state.
1501 // In order for GC to work, don't clear the last_Java_sp until after
1502 // blocking.
1503
1504 //=============================================================================
1505 // Switch thread to "native transition" state before reading the
1506 // synchronization state. This additional state is necessary
1507 // because reading and testing the synchronization state is not
1508 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1509 // in _thread_in_native state, loads _not_synchronized and is
1510 // preempted. VM thread changes sync state to synchronizing and
1511 // suspends threads for GC. Thread A is resumed to finish this
1512 // native method, but doesn't block here since it didn't see any
1513 // synchronization in progress, and escapes.
1514
1515 // We use release_store_fence to update values like the thread state, where
1516 // we don't want the current thread to continue until all our prior memory
1517 // accesses (including the new thread state) are visible to other threads.
1518 __ li(R0/*thread_state*/, _thread_in_native_trans);
1519 __ release();
1520 __ stw(R0/*thread_state*/, thread_(thread_state));
1521 if (UseMembar) {
1522 __ fence();
1523 }
1524 // Write serialization page so that the VM thread can do a pseudo remote
1525 // membar. We use the current thread pointer to calculate a thread
1526 // specific offset to write to within the page. This minimizes bus
1527 // traffic due to cache line collision.
1528 else {
1529 __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2);
1530 }
1531
1532 // Now before we return to java we must look for a current safepoint
1533 // (a new safepoint can not start since we entered native_trans).
1534 // We must check here because a current safepoint could be modifying
1535 // the callers registers right this moment.
1536
1537 // Acquire isn't strictly necessary here because of the fence, but
1538 // sync_state is declared to be volatile, so we do it anyway
1539 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1540 int sync_state_offs = __ load_const_optimized(sync_state_addr, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
1541
1542 // TODO PPC port assert(4 == SafepointSynchronize::sz_state(), "unexpected field size");
1543 __ lwz(sync_state, sync_state_offs, sync_state_addr);
1544
1545 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1546 __ lwz(suspend_flags, thread_(suspend_flags));
1547
1548 Label sync_check_done;
1549 Label do_safepoint;
1550 // No synchronization in progress nor yet synchronized.
1551 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
1552 // Not suspended.
1553 __ cmpwi(CCR1, suspend_flags, 0);
1554
1555 __ bne(CCR0, do_safepoint);
1556 __ beq(CCR1, sync_check_done);
1557 __ bind(do_safepoint);
1558 __ isync();
1559 // Block. We do the call directly and leave the current
1560 // last_Java_frame setup undisturbed. We must save any possible
1561 // native result across the call. No oop is present.
1562
1563 __ mr(R3_ARG1, R16_thread);
1564 #if defined(ABI_ELFv2)
1565 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
1566 relocInfo::none);
1567 #else
1568 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
1569 relocInfo::none);
1570 #endif
1571
1572 __ bind(sync_check_done);
1573
1574 //=============================================================================
1575 // <<<<<< Back in Interpreter Frame >>>>>
1576
1577 // We are in thread_in_native_trans here and back in the normal
1578 // interpreter frame. We don't have to do anything special about
1579 // safepoints and we can switch to Java mode anytime we are ready.
1580
1581 // Note: frame::interpreter_frame_result has a dependency on how the
1582 // method result is saved across the call to post_method_exit. For
1583 // native methods it assumes that the non-FPU/non-void result is
1584 // saved in _native_lresult and a FPU result in _native_fresult. If
1585 // this changes then the interpreter_frame_result implementation
1586 // will need to be updated too.
1587
1588 // On PPC64, we have stored the result directly after the native call.
1589
1590 //=============================================================================
1591 // Back in Java
1592
1593 // We use release_store_fence to update values like the thread state, where
1594 // we don't want the current thread to continue until all our prior memory
1595 // accesses (including the new thread state) are visible to other threads.
1596 __ li(R0/*thread_state*/, _thread_in_Java);
1597 __ release();
1598 __ stw(R0/*thread_state*/, thread_(thread_state));
1599 if (UseMembar) {
1600 __ fence();
1601 }
1602
1603 if (CheckJNICalls) {
1604 // clear_pending_jni_exception_check
1605 __ load_const_optimized(R0, 0L);
1606 __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread);
1607 }
1608
1609 __ reset_last_Java_frame();
1610
1611 // Jvmdi/jvmpi support. Whether we've got an exception pending or
1612 // not, and whether unlocking throws an exception or not, we notify
1613 // on native method exit. If we do have an exception, we'll end up
1614 // in the caller's context to handle it, so if we don't do the
1615 // notify here, we'll drop it on the floor.
1616 __ notify_method_exit(true/*native method*/,
1617 ilgl /*illegal state (not used for native methods)*/,
1618 InterpreterMacroAssembler::NotifyJVMTI,
1619 false /*check_exceptions*/);
1620
1621 //=============================================================================
1622 // Handle exceptions
1623
1624 if (synchronized) {
1625 // Don't check for exceptions since we're still in the i2n frame. Do that
1626 // manually afterwards.
1627 __ unlock_object(R26_monitor, false); // Can also unlock methods.
1628 }
1629
1630 // Reset active handles after returning from native.
1631 // thread->active_handles()->clear();
1632 __ ld(active_handles, thread_(active_handles));
1633 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1634 __ li(R0, 0);
1635 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles);
1636
1637 Label exception_return_sync_check_already_unlocked;
1638 __ ld(R0/*pending_exception*/, thread_(pending_exception));
1639 __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1640 __ bne(CCR0, exception_return_sync_check_already_unlocked);
1641
1642 //-----------------------------------------------------------------------------
1643 // No exception pending.
1644
1645 // Move native method result back into proper registers and return.
1646 // Invoke result handler (may unbox/promote).
1647 __ ld(R11_scratch1, 0, R1_SP);
1648 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1649 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1650 __ call_stub(result_handler_addr);
1651
1652 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1653
1654 // Must use the return pc which was loaded from the caller's frame
1655 // as the VM uses return-pc-patching for deoptimization.
1656 __ mtlr(R0);
1657 __ blr();
1658
1659 //-----------------------------------------------------------------------------
1660 // An exception is pending. We call into the runtime only if the
1661 // caller was not interpreted. If it was interpreted the
1662 // interpreter will do the correct thing. If it isn't interpreted
1663 // (call stub/compiled code) we will change our return and continue.
1664
1665 BIND(exception_return_sync_check);
1666
1667 if (synchronized) {
1668 // Don't check for exceptions since we're still in the i2n frame. Do that
1669 // manually afterwards.
1670 __ unlock_object(R26_monitor, false); // Can also unlock methods.
1671 }
1672 BIND(exception_return_sync_check_already_unlocked);
1673
1674 const Register return_pc = R31;
1675
1676 __ ld(return_pc, 0, R1_SP);
1677 __ ld(return_pc, _abi(lr), return_pc);
1678
1679 // Get the address of the exception handler.
1680 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1681 R16_thread,
1682 return_pc /* return pc */);
1683 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1684
1685 // Load the PC of the the exception handler into LR.
1686 __ mtlr(R3_RET);
1687
1688 // Load exception into R3_ARG1 and clear pending exception in thread.
1689 __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1690 __ li(R4_ARG2, 0);
1691 __ std(R4_ARG2, thread_(pending_exception));
1692
1693 // Load the original return pc into R4_ARG2.
1694 __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1695
1696 // Return to exception handler.
1697 __ blr();
1698
1699 //=============================================================================
1700 // Counter overflow.
1701
1702 if (inc_counter) {
1703 // Handle invocation counter overflow.
1704 __ bind(invocation_counter_overflow);
1705
1706 generate_counter_overflow(continue_after_compile);
1707 }
1708
1709 return entry;
1710 }
1711
1712 // Generic interpreted method entry to (asm) interpreter.
1713 //
1714 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1715 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1716 address entry = __ pc();
1717 // Generate the code to allocate the interpreter stack frame.
1718 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1719 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame.
1720
1721 // Does also a stack check to assure this frame fits on the stack.
1722 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1723
1724 // --------------------------------------------------------------------------
1725 // Zero out non-parameter locals.
1726 // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1727 // worth to ask the flag, just do it.
1728 Register Rslot_addr = R6_ARG4,
1729 Rnum = R7_ARG5;
1730 Label Lno_locals, Lzero_loop;
1731
1732 // Set up the zeroing loop.
1733 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1734 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1735 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1736 __ beq(CCR0, Lno_locals);
1737 __ li(R0, 0);
1738 __ mtctr(Rnum);
1739
1740 // The zero locals loop.
1741 __ bind(Lzero_loop);
1742 __ std(R0, 0, Rslot_addr);
1743 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1744 __ bdnz(Lzero_loop);
1745
1746 __ bind(Lno_locals);
1747
1748 // --------------------------------------------------------------------------
1749 // Counter increment and overflow check.
1750 Label invocation_counter_overflow,
1751 profile_method,
1752 profile_method_continue;
1753 if (inc_counter || ProfileInterpreter) {
1754
1755 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1756 if (synchronized) {
1757 // Since at this point in the method invocation the exception handler
1758 // would try to exit the monitor of synchronized methods which hasn't
1759 // been entered yet, we set the thread local variable
1760 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1761 // runtime, exception handling i.e. unlock_if_synchronized_method will
1762 // check this thread local flag.
1763 // This flag has two effects, one is to force an unwind in the topmost
1764 // interpreter frame and not perform an unlock while doing so.
1765 __ li(R0, 1);
1766 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1767 }
1768
1769 // Argument and return type profiling.
1770 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1771
1772 // Increment invocation counter and check for overflow.
1773 if (inc_counter) {
1774 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1775 }
1776
1777 __ bind(profile_method_continue);
1778 }
1779
1780 bang_stack_shadow_pages(false);
1781
1782 if (inc_counter || ProfileInterpreter) {
1783 // Reset the _do_not_unlock_if_synchronized flag.
1784 if (synchronized) {
1785 __ li(R0, 0);
1786 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1787 }
1788 }
1789
1790 // --------------------------------------------------------------------------
1791 // Locking of synchronized methods. Must happen AFTER invocation_counter
1792 // check and stack overflow check, so method is not locked if overflows.
1793 if (synchronized) {
1794 lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1795 }
1796 #ifdef ASSERT
1797 else {
1798 Label Lok;
1799 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1800 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1801 __ asm_assert_eq("method needs synchronization", 0x8521);
1802 __ bind(Lok);
1803 }
1804 #endif // ASSERT
1805
1806 __ verify_thread();
1807
1808 // --------------------------------------------------------------------------
1809 // JVMTI support
1810 __ notify_method_entry();
1811
1812 // --------------------------------------------------------------------------
1813 // Start executing instructions.
1814 __ dispatch_next(vtos);
1815
1816 // --------------------------------------------------------------------------
1817 // Out of line counter overflow and MDO creation code.
1818 if (ProfileInterpreter) {
1819 // We have decided to profile this method in the interpreter.
1820 __ bind(profile_method);
1821 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1822 __ set_method_data_pointer_for_bcp();
1823 __ b(profile_method_continue);
1824 }
1825
1826 if (inc_counter) {
1827 // Handle invocation counter overflow.
1828 __ bind(invocation_counter_overflow);
1829 generate_counter_overflow(profile_method_continue);
1830 }
1831 return entry;
1832 }
1833
1834 // CRC32 Intrinsics.
1835 //
1836 // Contract on scratch and work registers.
1837 // =======================================
1838 //
1839 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1840 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1841 // You can't rely on these registers across calls.
1842 //
1843 // The generators for CRC32_update and for CRC32_updateBytes use the
1844 // scratch/work register set internally, passing the work registers
1845 // as arguments to the MacroAssembler emitters as required.
1846 //
1847 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1848 // Their contents is not constant but may change according to the requirements
1849 // of the emitted code.
1850 //
1851 // All other registers from the scratch/work register set are used "internally"
1852 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1853 // Basically, only R3_RET contains a defined value which is the function result.
1854 //
1855 /**
1856 * Method entry for static native methods:
1857 * int java.util.zip.CRC32.update(int crc, int b)
1858 */
1859 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1860 if (UseCRC32Intrinsics) {
1861 address start = __ pc(); // Remember stub start address (is rtn value).
1862 Label slow_path;
1863
1864 // Safepoint check
1865 const Register sync_state = R11_scratch1;
1866 int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
1867 __ lwz(sync_state, sync_state_offs, sync_state);
1868 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
1869 __ bne(CCR0, slow_path);
1870
1871 // We don't generate local frame and don't align stack because
1872 // we not even call stub code (we generate the code inline)
1873 // and there is no safepoint on this path.
1874
1875 // Load java parameters.
1876 // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1877 const Register argP = R15_esp;
1878 const Register crc = R3_ARG1; // crc value
1879 const Register data = R4_ARG2; // address of java byte value (kernel_crc32 needs address)
1880 const Register dataLen = R5_ARG3; // source data len (1 byte). Not used because calling the single-byte emitter.
1881 const Register table = R6_ARG4; // address of crc32 table
1882 const Register tmp = dataLen; // Reuse unused len register to show we don't actually need a separate tmp here.
1883
1884 BLOCK_COMMENT("CRC32_update {");
1885
1886 // Arguments are reversed on java expression stack
1887 #ifdef VM_LITTLE_ENDIAN
1888 __ addi(data, argP, 0+1*wordSize); // (stack) address of byte value. Emitter expects address, not value.
1889 // Being passed as an int, the single byte is at offset +0.
1890 #else
1891 __ addi(data, argP, 3+1*wordSize); // (stack) address of byte value. Emitter expects address, not value.
1892 // Being passed from java as an int, the single byte is at offset +3.
1893 #endif
1894 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1895
1896 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
1897 __ kernel_crc32_singleByte(crc, data, dataLen, table, tmp);
1898
1899 // Restore caller sp for c2i case and return.
1900 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
1901 __ blr();
1902
1903 // Generate a vanilla native entry as the slow path.
1904 BLOCK_COMMENT("} CRC32_update");
1905 BIND(slow_path);
1906 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1907 return start;
1908 }
1909
1910 return NULL;
1911 }
1912
1913 // CRC32 Intrinsics.
1914 /**
1915 * Method entry for static native methods:
1916 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len)
1917 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1918 */
1919 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1920 if (UseCRC32Intrinsics) {
1921 address start = __ pc(); // Remember stub start address (is rtn value).
1922 Label slow_path;
1923
1924 // Safepoint check
1925 const Register sync_state = R11_scratch1;
1926 int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
1927 __ lwz(sync_state, sync_state_offs, sync_state);
1928 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
1929 __ bne(CCR0, slow_path);
1930
1931 // We don't generate local frame and don't align stack because
1932 // we not even call stub code (we generate the code inline)
1933 // and there is no safepoint on this path.
1934
1935 // Load parameters.
1936 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1937 const Register argP = R15_esp;
1938 const Register crc = R3_ARG1; // crc value
1939 const Register data = R4_ARG2; // address of java byte array
1940 const Register dataLen = R5_ARG3; // source data len
1941 const Register table = R6_ARG4; // address of crc32 table
1942
1943 const Register t0 = R9; // scratch registers for crc calculation
1944 const Register t1 = R10;
1945 const Register t2 = R11;
1946 const Register t3 = R12;
1947
1948 const Register tc0 = R2; // registers to hold pre-calculated column addresses
1949 const Register tc1 = R7;
1950 const Register tc2 = R8;
1951 const Register tc3 = table; // table address is reconstructed at the end of kernel_crc32_* emitters
1952
1953 const Register tmp = t0; // Only used very locally to calculate byte buffer address.
1954
1955 // Arguments are reversed on java expression stack.
1956 // Calculate address of start element.
1957 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1958 BLOCK_COMMENT("CRC32_updateByteBuffer {");
1959 // crc @ (SP + 5W) (32bit)
1960 // buf @ (SP + 3W) (64bit ptr to long array)
1961 // off @ (SP + 2W) (32bit)
1962 // dataLen @ (SP + 1W) (32bit)
1963 // data = buf + off
1964 __ ld( data, 3*wordSize, argP); // start of byte buffer
1965 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1966 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1967 __ lwz( crc, 5*wordSize, argP); // current crc state
1968 __ add( data, data, tmp); // Add byte buffer offset.
1969 } else { // Used for "updateBytes update".
1970 BLOCK_COMMENT("CRC32_updateBytes {");
1971 // crc @ (SP + 4W) (32bit)
1972 // buf @ (SP + 3W) (64bit ptr to byte array)
1973 // off @ (SP + 2W) (32bit)
1974 // dataLen @ (SP + 1W) (32bit)
1975 // data = buf + off + base_offset
1976 __ ld( data, 3*wordSize, argP); // start of byte buffer
1977 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1978 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1979 __ add( data, data, tmp); // add byte buffer offset
1980 __ lwz( crc, 4*wordSize, argP); // current crc state
1981 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1982 }
1983
1984 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
1985
1986 // Performance measurements show the 1word and 2word variants to be almost equivalent,
1987 // with very light advantages for the 1word variant. We chose the 1word variant for
1988 // code compactness.
1989 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3);
1990
1991 // Restore caller sp for c2i case and return.
1992 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
1993 __ blr();
1994
1995 // Generate a vanilla native entry as the slow path.
1996 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1997 BIND(slow_path);
1998 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1999 return start;
2000 }
2001
2002 return NULL;
2003 }
2004
2005 // Not supported
2006 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
2007 return NULL;
2008 }
2009
2010 // =============================================================================
2011 // Exceptions
2012
2013 void TemplateInterpreterGenerator::generate_throw_exception() {
2014 Register Rexception = R17_tos,
2015 Rcontinuation = R3_RET;
2016
2017 // --------------------------------------------------------------------------
2018 // Entry point if an method returns with a pending exception (rethrow).
2019 Interpreter::_rethrow_exception_entry = __ pc();
2020 {
2021 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
2022 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
2023 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
2024
2025 // Compiled code destroys templateTableBase, reload.
2026 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
2027 }
2028
2029 // Entry point if a interpreted method throws an exception (throw).
2030 Interpreter::_throw_exception_entry = __ pc();
2031 {
2032 __ mr(Rexception, R3_RET);
2033
2034 __ verify_thread();
2035 __ verify_oop(Rexception);
2036
2037 // Expression stack must be empty before entering the VM in case of an exception.
2038 __ empty_expression_stack();
2039 // Find exception handler address and preserve exception oop.
2040 // Call C routine to find handler and jump to it.
2041 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
2042 __ mtctr(Rcontinuation);
2043 // Push exception for exception handler bytecodes.
2044 __ push_ptr(Rexception);
2045
2046 // Jump to exception handler (may be remove activation entry!).
2047 __ bctr();
2048 }
2049
2050 // If the exception is not handled in the current frame the frame is
2051 // removed and the exception is rethrown (i.e. exception
2052 // continuation is _rethrow_exception).
2053 //
2054 // Note: At this point the bci is still the bxi for the instruction
2055 // which caused the exception and the expression stack is
2056 // empty. Thus, for any VM calls at this point, GC will find a legal
2057 // oop map (with empty expression stack).
2058
2059 // In current activation
2060 // tos: exception
2061 // bcp: exception bcp
2062
2063 // --------------------------------------------------------------------------
2064 // JVMTI PopFrame support
2065
2066 Interpreter::_remove_activation_preserving_args_entry = __ pc();
2067 {
2068 // Set the popframe_processing bit in popframe_condition indicating that we are
2069 // currently handling popframe, so that call_VMs that may happen later do not
2070 // trigger new popframe handling cycles.
2071 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2072 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
2073 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2074
2075 // Empty the expression stack, as in normal exception handling.
2076 __ empty_expression_stack();
2077 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
2078
2079 // Check to see whether we are returning to a deoptimized frame.
2080 // (The PopFrame call ensures that the caller of the popped frame is
2081 // either interpreted or compiled and deoptimizes it if compiled.)
2082 // Note that we don't compare the return PC against the
2083 // deoptimization blob's unpack entry because of the presence of
2084 // adapter frames in C2.
2085 Label Lcaller_not_deoptimized;
2086 Register return_pc = R3_ARG1;
2087 __ ld(return_pc, 0, R1_SP);
2088 __ ld(return_pc, _abi(lr), return_pc);
2089 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
2090 __ cmpdi(CCR0, R3_RET, 0);
2091 __ bne(CCR0, Lcaller_not_deoptimized);
2092
2093 // The deoptimized case.
2094 // In this case, we can't call dispatch_next() after the frame is
2095 // popped, but instead must save the incoming arguments and restore
2096 // them after deoptimization has occurred.
2097 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
2098 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
2099 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
2100 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
2101 __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
2102 // Save these arguments.
2103 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
2104
2105 // Inform deoptimization that it is responsible for restoring these arguments.
2106 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
2107 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2108
2109 // Return from the current method into the deoptimization blob. Will eventually
2110 // end up in the deopt interpeter entry, deoptimization prepared everything that
2111 // we will reexecute the call that called us.
2112 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
2113 __ mtlr(return_pc);
2114 __ blr();
2115
2116 // The non-deoptimized case.
2117 __ bind(Lcaller_not_deoptimized);
2118
2119 // Clear the popframe condition flag.
2120 __ li(R0, 0);
2121 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2122
2123 // Get out of the current method and re-execute the call that called us.
2124 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2125 __ restore_interpreter_state(R11_scratch1);
2126 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
2127 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
2128 if (ProfileInterpreter) {
2129 __ set_method_data_pointer_for_bcp();
2130 __ ld(R11_scratch1, 0, R1_SP);
2131 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
2132 }
2133 #if INCLUDE_JVMTI
2134 Label L_done;
2135
2136 __ lbz(R11_scratch1, 0, R14_bcp);
2137 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
2138 __ bne(CCR0, L_done);
2139
2140 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2141 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
2142 __ ld(R4_ARG2, 0, R18_locals);
2143 __ MacroAssembler::call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp, false);
2144 __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2145 __ cmpdi(CCR0, R4_ARG2, 0);
2146 __ beq(CCR0, L_done);
2147 __ std(R4_ARG2, wordSize, R15_esp);
2148 __ bind(L_done);
2149 #endif // INCLUDE_JVMTI
2150 __ dispatch_next(vtos);
2151 }
2152 // end of JVMTI PopFrame support
2153
2154 // --------------------------------------------------------------------------
2155 // Remove activation exception entry.
2156 // This is jumped to if an interpreted method can't handle an exception itself
2157 // (we come from the throw/rethrow exception entry above). We're going to call
2158 // into the VM to find the exception handler in the caller, pop the current
2159 // frame and return the handler we calculated.
2160 Interpreter::_remove_activation_entry = __ pc();
2161 {
2162 __ pop_ptr(Rexception);
2163 __ verify_thread();
2164 __ verify_oop(Rexception);
2165 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
2166
2167 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
2168 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
2169
2170 __ get_vm_result(Rexception);
2171
2172 // We are done with this activation frame; find out where to go next.
2173 // The continuation point will be an exception handler, which expects
2174 // the following registers set up:
2175 //
2176 // RET: exception oop
2177 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
2178
2179 Register return_pc = R31; // Needs to survive the runtime call.
2180 __ ld(return_pc, 0, R1_SP);
2181 __ ld(return_pc, _abi(lr), return_pc);
2182 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
2183
2184 // Remove the current activation.
2185 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2186
2187 __ mr(R4_ARG2, return_pc);
2188 __ mtlr(R3_RET);
2189 __ mr(R3_RET, Rexception);
2190 __ blr();
2191 }
2192 }
2193
2194 // JVMTI ForceEarlyReturn support.
2195 // Returns "in the middle" of a method with a "fake" return value.
2196 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2197
2198 Register Rscratch1 = R11_scratch1,
2199 Rscratch2 = R12_scratch2;
2200
2201 address entry = __ pc();
2202 __ empty_expression_stack();
2203
2204 __ load_earlyret_value(state, Rscratch1);
2205
2206 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
2207 // Clear the earlyret state.
2208 __ li(R0, 0);
2209 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
2210
2211 __ remove_activation(state, false, false);
2212 // Copied from TemplateTable::_return.
2213 // Restoration of lr done by remove_activation.
2214 switch (state) {
2215 // Narrow result if state is itos but result type is smaller.
2216 case btos:
2217 case ztos:
2218 case ctos:
2219 case stos:
2220 case itos: __ narrow(R17_tos); /* fall through */
2221 case ltos:
2222 case atos: __ mr(R3_RET, R17_tos); break;
2223 case ftos:
2224 case dtos: __ fmr(F1_RET, F15_ftos); break;
2225 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2226 // to get visible before the reference to the object gets stored anywhere.
2227 __ membar(Assembler::StoreStore); break;
2228 default : ShouldNotReachHere();
2229 }
2230 __ blr();
2231
2232 return entry;
2233 } // end of ForceEarlyReturn support
2234
2235 //-----------------------------------------------------------------------------
2236 // Helper for vtos entry point generation
2237
2238 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2239 address& bep,
2240 address& cep,
2241 address& sep,
2242 address& aep,
2243 address& iep,
2244 address& lep,
2245 address& fep,
2246 address& dep,
2247 address& vep) {
2248 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2249 Label L;
2250
2251 aep = __ pc(); __ push_ptr(); __ b(L);
2252 fep = __ pc(); __ push_f(); __ b(L);
2253 dep = __ pc(); __ push_d(); __ b(L);
2254 lep = __ pc(); __ push_l(); __ b(L);
2255 __ align(32, 12, 24); // align L
2256 bep = cep = sep =
2257 iep = __ pc(); __ push_i();
2258 vep = __ pc();
2259 __ bind(L);
2260 generate_and_dispatch(t);
2261 }
2262
2263 //-----------------------------------------------------------------------------
2264
2265 // Non-product code
2266 #ifndef PRODUCT
2267 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2268 //__ flush_bundle();
2269 address entry = __ pc();
2270
2271 const char *bname = NULL;
2272 uint tsize = 0;
2273 switch(state) {
2274 case ftos:
2275 bname = "trace_code_ftos {";
2276 tsize = 2;
2277 break;
2278 case btos:
2279 bname = "trace_code_btos {";
2280 tsize = 2;
2281 break;
2282 case ztos:
2283 bname = "trace_code_ztos {";
2284 tsize = 2;
2285 break;
2286 case ctos:
2287 bname = "trace_code_ctos {";
2288 tsize = 2;
2289 break;
2290 case stos:
2291 bname = "trace_code_stos {";
2292 tsize = 2;
2293 break;
2294 case itos:
2295 bname = "trace_code_itos {";
2296 tsize = 2;
2297 break;
2298 case ltos:
2299 bname = "trace_code_ltos {";
2300 tsize = 3;
2301 break;
2302 case atos:
2303 bname = "trace_code_atos {";
2304 tsize = 2;
2305 break;
2306 case vtos:
2307 // Note: In case of vtos, the topmost of stack value could be a int or doubl
2308 // In case of a double (2 slots) we won't see the 2nd stack value.
2309 // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
2310 bname = "trace_code_vtos {";
2311 tsize = 2;
2312
2313 break;
2314 case dtos:
2315 bname = "trace_code_dtos {";
2316 tsize = 3;
2317 break;
2318 default:
2319 ShouldNotReachHere();
2320 }
2321 BLOCK_COMMENT(bname);
2322
2323 // Support short-cut for TraceBytecodesAt.
2324 // Don't call into the VM if we don't want to trace to speed up things.
2325 Label Lskip_vm_call;
2326 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2327 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
2328 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2329 __ ld(R11_scratch1, offs1, R11_scratch1);
2330 __ lwa(R12_scratch2, offs2, R12_scratch2);
2331 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2332 __ blt(CCR0, Lskip_vm_call);
2333 }
2334
2335 __ push(state);
2336 // Load 2 topmost expression stack values.
2337 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
2338 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
2339 __ mflr(R31);
2340 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
2341 __ mtlr(R31);
2342 __ pop(state);
2343
2344 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2345 __ bind(Lskip_vm_call);
2346 }
2347 __ blr();
2348 BLOCK_COMMENT("} trace_code");
2349 return entry;
2350 }
2351
2352 void TemplateInterpreterGenerator::count_bytecode() {
2353 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
2354 __ lwz(R12_scratch2, offs, R11_scratch1);
2355 __ addi(R12_scratch2, R12_scratch2, 1);
2356 __ stw(R12_scratch2, offs, R11_scratch1);
2357 }
2358
2359 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2360 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
2361 __ lwz(R12_scratch2, offs, R11_scratch1);
2362 __ addi(R12_scratch2, R12_scratch2, 1);
2363 __ stw(R12_scratch2, offs, R11_scratch1);
2364 }
2365
2366 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2367 const Register addr = R11_scratch1,
2368 tmp = R12_scratch2;
2369 // Get index, shift out old bytecode, bring in new bytecode, and store it.
2370 // _index = (_index >> log2_number_of_codes) |
2371 // (bytecode << log2_number_of_codes);
2372 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
2373 __ lwz(tmp, offs1, addr);
2374 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
2375 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2376 __ stw(tmp, offs1, addr);
2377
2378 // Bump bucket contents.
2379 // _counters[_index] ++;
2380 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
2381 __ sldi(tmp, tmp, LogBytesPerInt);
2382 __ add(addr, tmp, addr);
2383 __ lwz(tmp, offs2, addr);
2384 __ addi(tmp, tmp, 1);
2385 __ stw(tmp, offs2, addr);
2386 }
2387
2388 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2389 // Call a little run-time stub to avoid blow-up for each bytecode.
2390 // The run-time runtime saves the right registers, depending on
2391 // the tosca in-state for the given template.
2392
2393 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2394 "entry must have been generated");
2395
2396 // Note: we destroy LR here.
2397 __ bl(Interpreter::trace_code(t->tos_in()));
2398 }
2399
2400 void TemplateInterpreterGenerator::stop_interpreter_at() {
2401 Label L;
2402 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
2403 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2404 __ ld(R11_scratch1, offs1, R11_scratch1);
2405 __ lwa(R12_scratch2, offs2, R12_scratch2);
2406 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2407 __ bne(CCR0, L);
2408 __ illtrap();
2409 __ bind(L);
2410 }
2411
2412 #endif // !PRODUCT