rev 55858 : 8228649: [PPC64] SA reads wrong slots from interpreter frames
Summary: Make frame layout consistent between dbg and product build and implement offsets accordingly.
Reviewed-by: goetz, gromero
1 /*
2 * Copyright (c) 2014, 2019, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2019, 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 "gc/shared/barrierSetAssembler.hpp"
29 #include "interpreter/bytecodeHistogram.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "interpreter/interpreterRuntime.hpp"
32 #include "interpreter/interp_masm.hpp"
33 #include "interpreter/templateInterpreterGenerator.hpp"
34 #include "interpreter/templateTable.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/methodData.hpp"
37 #include "oops/method.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "prims/jvmtiExport.hpp"
40 #include "prims/jvmtiThreadState.hpp"
41 #include "runtime/arguments.hpp"
42 #include "runtime/deoptimization.hpp"
43 #include "runtime/frame.inline.hpp"
44 #include "runtime/sharedRuntime.hpp"
45 #include "runtime/stubRoutines.hpp"
46 #include "runtime/synchronizer.hpp"
47 #include "runtime/timer.hpp"
48 #include "runtime/vframeArray.hpp"
49 #include "utilities/debug.hpp"
50 #include "utilities/macros.hpp"
51
52 #undef __
53 #define __ _masm->
54
55 // Size of interpreter code. Increase if too small. Interpreter will
56 // fail with a guarantee ("not enough space for interpreter generation");
57 // if too small.
58 // Run with +PrintInterpreter to get the VM to print out the size.
59 // Max size with JVMTI
60 int TemplateInterpreter::InterpreterCodeSize = 256*K;
61
62 #ifdef PRODUCT
63 #define BLOCK_COMMENT(str) /* nothing */
64 #else
65 #define BLOCK_COMMENT(str) __ block_comment(str)
66 #endif
67
68 #define BIND(label) __ bind(label); BLOCK_COMMENT(#label ":")
69
70 //-----------------------------------------------------------------------------
71
72 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
73 // Slow_signature handler that respects the PPC C calling conventions.
74 //
75 // We get called by the native entry code with our output register
76 // area == 8. First we call InterpreterRuntime::get_result_handler
77 // to copy the pointer to the signature string temporarily to the
78 // first C-argument and to return the result_handler in
79 // R3_RET. Since native_entry will copy the jni-pointer to the
80 // first C-argument slot later on, it is OK to occupy this slot
81 // temporarilly. Then we copy the argument list on the java
82 // expression stack into native varargs format on the native stack
83 // and load arguments into argument registers. Integer arguments in
84 // the varargs vector will be sign-extended to 8 bytes.
85 //
86 // On entry:
87 // R3_ARG1 - intptr_t* Address of java argument list in memory.
88 // R15_prev_state - BytecodeInterpreter* Address of interpreter state for
89 // this method
90 // R19_method
91 //
92 // On exit (just before return instruction):
93 // R3_RET - contains the address of the result_handler.
94 // R4_ARG2 - is not updated for static methods and contains "this" otherwise.
95 // R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double,
96 // ARGi contains this argument. Otherwise, ARGi is not updated.
97 // F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double.
98
99 const int LogSizeOfTwoInstructions = 3;
100
101 // FIXME: use Argument:: GL: Argument names different numbers!
102 const int max_fp_register_arguments = 13;
103 const int max_int_register_arguments = 6; // first 2 are reserved
104
105 const Register arg_java = R21_tmp1;
106 const Register arg_c = R22_tmp2;
107 const Register signature = R23_tmp3; // is string
108 const Register sig_byte = R24_tmp4;
109 const Register fpcnt = R25_tmp5;
110 const Register argcnt = R26_tmp6;
111 const Register intSlot = R27_tmp7;
112 const Register target_sp = R28_tmp8;
113 const FloatRegister floatSlot = F0;
114
115 address entry = __ function_entry();
116
117 __ save_LR_CR(R0);
118 __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
119 // We use target_sp for storing arguments in the C frame.
120 __ mr(target_sp, R1_SP);
121 __ push_frame_reg_args_nonvolatiles(0, R11_scratch1);
122
123 __ mr(arg_java, R3_ARG1);
124
125 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method);
126
127 // Signature is in R3_RET. Signature is callee saved.
128 __ mr(signature, R3_RET);
129
130 // Get the result handler.
131 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method);
132
133 {
134 Label L;
135 // test if static
136 // _access_flags._flags must be at offset 0.
137 // TODO PPC port: requires change in shared code.
138 //assert(in_bytes(AccessFlags::flags_offset()) == 0,
139 // "MethodDesc._access_flags == MethodDesc._access_flags._flags");
140 // _access_flags must be a 32 bit value.
141 assert(sizeof(AccessFlags) == 4, "wrong size");
142 __ lwa(R11_scratch1/*access_flags*/, method_(access_flags));
143 // testbit with condition register.
144 __ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT);
145 __ btrue(CCR0, L);
146 // For non-static functions, pass "this" in R4_ARG2 and copy it
147 // to 2nd C-arg slot.
148 // We need to box the Java object here, so we use arg_java
149 // (address of current Java stack slot) as argument and don't
150 // dereference it as in case of ints, floats, etc.
151 __ mr(R4_ARG2, arg_java);
152 __ addi(arg_java, arg_java, -BytesPerWord);
153 __ std(R4_ARG2, _abi(carg_2), target_sp);
154 __ bind(L);
155 }
156
157 // Will be incremented directly after loop_start. argcnt=0
158 // corresponds to 3rd C argument.
159 __ li(argcnt, -1);
160 // arg_c points to 3rd C argument
161 __ addi(arg_c, target_sp, _abi(carg_3));
162 // no floating-point args parsed so far
163 __ li(fpcnt, 0);
164
165 Label move_intSlot_to_ARG, move_floatSlot_to_FARG;
166 Label loop_start, loop_end;
167 Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed;
168
169 // signature points to '(' at entry
170 #ifdef ASSERT
171 __ lbz(sig_byte, 0, signature);
172 __ cmplwi(CCR0, sig_byte, '(');
173 __ bne(CCR0, do_dontreachhere);
174 #endif
175
176 __ bind(loop_start);
177
178 __ addi(argcnt, argcnt, 1);
179 __ lbzu(sig_byte, 1, signature);
180
181 __ cmplwi(CCR0, sig_byte, ')'); // end of signature
182 __ beq(CCR0, loop_end);
183
184 __ cmplwi(CCR0, sig_byte, 'B'); // byte
185 __ beq(CCR0, do_int);
186
187 __ cmplwi(CCR0, sig_byte, 'C'); // char
188 __ beq(CCR0, do_int);
189
190 __ cmplwi(CCR0, sig_byte, 'D'); // double
191 __ beq(CCR0, do_double);
192
193 __ cmplwi(CCR0, sig_byte, 'F'); // float
194 __ beq(CCR0, do_float);
195
196 __ cmplwi(CCR0, sig_byte, 'I'); // int
197 __ beq(CCR0, do_int);
198
199 __ cmplwi(CCR0, sig_byte, 'J'); // long
200 __ beq(CCR0, do_long);
201
202 __ cmplwi(CCR0, sig_byte, 'S'); // short
203 __ beq(CCR0, do_int);
204
205 __ cmplwi(CCR0, sig_byte, 'Z'); // boolean
206 __ beq(CCR0, do_int);
207
208 __ cmplwi(CCR0, sig_byte, 'L'); // object
209 __ beq(CCR0, do_object);
210
211 __ cmplwi(CCR0, sig_byte, '['); // array
212 __ beq(CCR0, do_array);
213
214 // __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type
215 // __ beq(CCR0, do_void);
216
217 __ bind(do_dontreachhere);
218
219 __ unimplemented("ShouldNotReachHere in slow_signature_handler", 120);
220
221 __ bind(do_array);
222
223 {
224 Label start_skip, end_skip;
225
226 __ bind(start_skip);
227 __ lbzu(sig_byte, 1, signature);
228 __ cmplwi(CCR0, sig_byte, '[');
229 __ beq(CCR0, start_skip); // skip further brackets
230 __ cmplwi(CCR0, sig_byte, '9');
231 __ bgt(CCR0, end_skip); // no optional size
232 __ cmplwi(CCR0, sig_byte, '0');
233 __ bge(CCR0, start_skip); // skip optional size
234 __ bind(end_skip);
235
236 __ cmplwi(CCR0, sig_byte, 'L');
237 __ beq(CCR0, do_object); // for arrays of objects, the name of the object must be skipped
238 __ b(do_boxed); // otherwise, go directly to do_boxed
239 }
240
241 __ bind(do_object);
242 {
243 Label L;
244 __ bind(L);
245 __ lbzu(sig_byte, 1, signature);
246 __ cmplwi(CCR0, sig_byte, ';');
247 __ bne(CCR0, L);
248 }
249 // Need to box the Java object here, so we use arg_java (address of
250 // current Java stack slot) as argument and don't dereference it as
251 // in case of ints, floats, etc.
252 Label do_null;
253 __ bind(do_boxed);
254 __ ld(R0,0, arg_java);
255 __ cmpdi(CCR0, R0, 0);
256 __ li(intSlot,0);
257 __ beq(CCR0, do_null);
258 __ mr(intSlot, arg_java);
259 __ bind(do_null);
260 __ std(intSlot, 0, arg_c);
261 __ addi(arg_java, arg_java, -BytesPerWord);
262 __ addi(arg_c, arg_c, BytesPerWord);
263 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
264 __ blt(CCR0, move_intSlot_to_ARG);
265 __ b(loop_start);
266
267 __ bind(do_int);
268 __ lwa(intSlot, 0, arg_java);
269 __ std(intSlot, 0, arg_c);
270 __ addi(arg_java, arg_java, -BytesPerWord);
271 __ addi(arg_c, arg_c, BytesPerWord);
272 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
273 __ blt(CCR0, move_intSlot_to_ARG);
274 __ b(loop_start);
275
276 __ bind(do_long);
277 __ ld(intSlot, -BytesPerWord, arg_java);
278 __ std(intSlot, 0, arg_c);
279 __ addi(arg_java, arg_java, - 2 * BytesPerWord);
280 __ addi(arg_c, arg_c, BytesPerWord);
281 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
282 __ blt(CCR0, move_intSlot_to_ARG);
283 __ b(loop_start);
284
285 __ bind(do_float);
286 __ lfs(floatSlot, 0, arg_java);
287 #if defined(LINUX)
288 // Linux uses ELF ABI. Both original ELF and ELFv2 ABIs have float
289 // in the least significant word of an argument slot.
290 #if defined(VM_LITTLE_ENDIAN)
291 __ stfs(floatSlot, 0, arg_c);
292 #else
293 __ stfs(floatSlot, 4, arg_c);
294 #endif
295 #elif defined(AIX)
296 // Although AIX runs on big endian CPU, float is in most significant
297 // word of an argument slot.
298 __ stfs(floatSlot, 0, arg_c);
299 #else
300 #error "unknown OS"
301 #endif
302 __ addi(arg_java, arg_java, -BytesPerWord);
303 __ addi(arg_c, arg_c, BytesPerWord);
304 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
305 __ blt(CCR0, move_floatSlot_to_FARG);
306 __ b(loop_start);
307
308 __ bind(do_double);
309 __ lfd(floatSlot, - BytesPerWord, arg_java);
310 __ stfd(floatSlot, 0, arg_c);
311 __ addi(arg_java, arg_java, - 2 * BytesPerWord);
312 __ addi(arg_c, arg_c, BytesPerWord);
313 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
314 __ blt(CCR0, move_floatSlot_to_FARG);
315 __ b(loop_start);
316
317 __ bind(loop_end);
318
319 __ pop_frame();
320 __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
321 __ restore_LR_CR(R0);
322
323 __ blr();
324
325 Label move_int_arg, move_float_arg;
326 __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
327 __ mr(R5_ARG3, intSlot); __ b(loop_start);
328 __ mr(R6_ARG4, intSlot); __ b(loop_start);
329 __ mr(R7_ARG5, intSlot); __ b(loop_start);
330 __ mr(R8_ARG6, intSlot); __ b(loop_start);
331 __ mr(R9_ARG7, intSlot); __ b(loop_start);
332 __ mr(R10_ARG8, intSlot); __ b(loop_start);
333
334 __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
335 __ fmr(F1_ARG1, floatSlot); __ b(loop_start);
336 __ fmr(F2_ARG2, floatSlot); __ b(loop_start);
337 __ fmr(F3_ARG3, floatSlot); __ b(loop_start);
338 __ fmr(F4_ARG4, floatSlot); __ b(loop_start);
339 __ fmr(F5_ARG5, floatSlot); __ b(loop_start);
340 __ fmr(F6_ARG6, floatSlot); __ b(loop_start);
341 __ fmr(F7_ARG7, floatSlot); __ b(loop_start);
342 __ fmr(F8_ARG8, floatSlot); __ b(loop_start);
343 __ fmr(F9_ARG9, floatSlot); __ b(loop_start);
344 __ fmr(F10_ARG10, floatSlot); __ b(loop_start);
345 __ fmr(F11_ARG11, floatSlot); __ b(loop_start);
346 __ fmr(F12_ARG12, floatSlot); __ b(loop_start);
347 __ fmr(F13_ARG13, floatSlot); __ b(loop_start);
348
349 __ bind(move_intSlot_to_ARG);
350 __ sldi(R0, argcnt, LogSizeOfTwoInstructions);
351 __ load_const(R11_scratch1, move_int_arg); // Label must be bound here.
352 __ add(R11_scratch1, R0, R11_scratch1);
353 __ mtctr(R11_scratch1/*branch_target*/);
354 __ bctr();
355 __ bind(move_floatSlot_to_FARG);
356 __ sldi(R0, fpcnt, LogSizeOfTwoInstructions);
357 __ addi(fpcnt, fpcnt, 1);
358 __ load_const(R11_scratch1, move_float_arg); // Label must be bound here.
359 __ add(R11_scratch1, R0, R11_scratch1);
360 __ mtctr(R11_scratch1/*branch_target*/);
361 __ bctr();
362
363 return entry;
364 }
365
366 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
367 //
368 // Registers alive
369 // R3_RET
370 // LR
371 //
372 // Registers updated
373 // R3_RET
374 //
375
376 Label done;
377 address entry = __ pc();
378
379 switch (type) {
380 case T_BOOLEAN:
381 // convert !=0 to 1
382 __ neg(R0, R3_RET);
383 __ orr(R0, R3_RET, R0);
384 __ srwi(R3_RET, R0, 31);
385 break;
386 case T_BYTE:
387 // sign extend 8 bits
388 __ extsb(R3_RET, R3_RET);
389 break;
390 case T_CHAR:
391 // zero extend 16 bits
392 __ clrldi(R3_RET, R3_RET, 48);
393 break;
394 case T_SHORT:
395 // sign extend 16 bits
396 __ extsh(R3_RET, R3_RET);
397 break;
398 case T_INT:
399 // sign extend 32 bits
400 __ extsw(R3_RET, R3_RET);
401 break;
402 case T_LONG:
403 break;
404 case T_OBJECT:
405 // JNIHandles::resolve result.
406 __ resolve_jobject(R3_RET, R11_scratch1, R31, /* needs_frame */ true); // kills R31
407 break;
408 case T_FLOAT:
409 break;
410 case T_DOUBLE:
411 break;
412 case T_VOID:
413 break;
414 default: ShouldNotReachHere();
415 }
416
417 BIND(done);
418 __ blr();
419
420 return entry;
421 }
422
423 // Abstract method entry.
424 //
425 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
426 address entry = __ pc();
427
428 //
429 // Registers alive
430 // R16_thread - JavaThread*
431 // R19_method - callee's method (method to be invoked)
432 // R1_SP - SP prepared such that caller's outgoing args are near top
433 // LR - return address to caller
434 //
435 // Stack layout at this point:
436 //
437 // 0 [TOP_IJAVA_FRAME_ABI] <-- R1_SP
438 // alignment (optional)
439 // [outgoing Java arguments]
440 // ...
441 // PARENT [PARENT_IJAVA_FRAME_ABI]
442 // ...
443 //
444
445 // Can't use call_VM here because we have not set up a new
446 // interpreter state. Make the call to the vm and make it look like
447 // our caller set up the JavaFrameAnchor.
448 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
449
450 // Push a new C frame and save LR.
451 __ save_LR_CR(R0);
452 __ push_frame_reg_args(0, R11_scratch1);
453
454 // This is not a leaf but we have a JavaFrameAnchor now and we will
455 // check (create) exceptions afterward so this is ok.
456 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
457 R16_thread, R19_method);
458
459 // Pop the C frame and restore LR.
460 __ pop_frame();
461 __ restore_LR_CR(R0);
462
463 // Reset JavaFrameAnchor from call_VM_leaf above.
464 __ reset_last_Java_frame();
465
466 // We don't know our caller, so jump to the general forward exception stub,
467 // which will also pop our full frame off. Satisfy the interface of
468 // SharedRuntime::generate_forward_exception()
469 __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0);
470 __ mtctr(R11_scratch1);
471 __ bctr();
472
473 return entry;
474 }
475
476 // Interpreter intrinsic for WeakReference.get().
477 // 1. Don't push a full blown frame and go on dispatching, but fetch the value
478 // into R8 and return quickly
479 // 2. If G1 is active we *must* execute this intrinsic for corrrectness:
480 // It contains a GC barrier which puts the reference into the satb buffer
481 // to indicate that someone holds a strong reference to the object the
482 // weak ref points to!
483 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
484 // Code: _aload_0, _getfield, _areturn
485 // parameter size = 1
486 //
487 // The code that gets generated by this routine is split into 2 parts:
488 // 1. the "intrinsified" code for G1 (or any SATB based GC),
489 // 2. the slow path - which is an expansion of the regular method entry.
490 //
491 // Notes:
492 // * In the G1 code we do not check whether we need to block for
493 // a safepoint. If G1 is enabled then we must execute the specialized
494 // code for Reference.get (except when the Reference object is null)
495 // so that we can log the value in the referent field with an SATB
496 // update buffer.
497 // If the code for the getfield template is modified so that the
498 // G1 pre-barrier code is executed when the current method is
499 // Reference.get() then going through the normal method entry
500 // will be fine.
501 // * The G1 code can, however, check the receiver object (the instance
502 // of java.lang.Reference) and jump to the slow path if null. If the
503 // Reference object is null then we obviously cannot fetch the referent
504 // and so we don't need to call the G1 pre-barrier. Thus we can use the
505 // regular method entry code to generate the NPE.
506 //
507
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_heap_oop(R3_RET, referent_offset, R3_RET,
528 /* non-volatile temp */ R31, R11_scratch1, true, ON_WEAK_OOP_REF);
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 (from compiled) and for resized sender frame (from interpreted).
536 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
537
538 __ blr();
539
540 __ bind(slow_path);
541 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
542 return entry;
543 }
544
545 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
546 address entry = __ pc();
547
548 // Expression stack must be empty before entering the VM if an
549 // exception happened.
550 __ empty_expression_stack();
551 // Throw exception.
552 __ call_VM(noreg,
553 CAST_FROM_FN_PTR(address,
554 InterpreterRuntime::throw_StackOverflowError));
555 return entry;
556 }
557
558 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
559 address entry = __ pc();
560 __ empty_expression_stack();
561 // R4_ARG2 already contains the array.
562 // Index is in R17_tos.
563 __ mr(R5_ARG3, R17_tos);
564 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R4_ARG2, R5_ARG3);
565 return entry;
566 }
567
568 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
569 address entry = __ pc();
570 // Expression stack must be empty before entering the VM if an
571 // exception happened.
572 __ empty_expression_stack();
573
574 // Load exception object.
575 // Thread will be loaded to R3_ARG1.
576 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
577 #ifdef ASSERT
578 // Above call must not return here since exception pending.
579 __ should_not_reach_here();
580 #endif
581 return entry;
582 }
583
584 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
585 address entry = __ pc();
586 //__ untested("generate_exception_handler_common");
587 Register Rexception = R17_tos;
588
589 // Expression stack must be empty before entering the VM if an exception happened.
590 __ empty_expression_stack();
591
592 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
593 if (pass_oop) {
594 __ mr(R5_ARG3, Rexception);
595 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false);
596 } else {
597 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
598 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false);
599 }
600
601 // Throw exception.
602 __ mr(R3_ARG1, Rexception);
603 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
604 __ mtctr(R11_scratch1);
605 __ bctr();
606
607 return entry;
608 }
609
610 // This entry is returned to when a call returns to the interpreter.
611 // When we arrive here, we expect that the callee stack frame is already popped.
612 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
613 address entry = __ pc();
614
615 // Move the value out of the return register back to the TOS cache of current frame.
616 switch (state) {
617 case ltos:
618 case btos:
619 case ztos:
620 case ctos:
621 case stos:
622 case atos:
623 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache
624 case ftos:
625 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
626 case vtos: break; // Nothing to do, this was a void return.
627 default : ShouldNotReachHere();
628 }
629
630 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
631 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
632 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
633
634 // Compiled code destroys templateTableBase, reload.
635 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
636
637 if (state == atos) {
638 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
639 }
640
641 const Register cache = R11_scratch1;
642 const Register size = R12_scratch2;
643 __ get_cache_and_index_at_bcp(cache, 1, index_size);
644
645 // Get least significant byte of 64 bit value:
646 #if defined(VM_LITTLE_ENDIAN)
647 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache);
648 #else
649 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
650 #endif
651 __ sldi(size, size, Interpreter::logStackElementSize);
652 __ add(R15_esp, R15_esp, size);
653
654 __ check_and_handle_popframe(R11_scratch1);
655 __ check_and_handle_earlyret(R11_scratch1);
656
657 __ dispatch_next(state, step);
658 return entry;
659 }
660
661 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
662 address entry = __ pc();
663 // If state != vtos, we're returning from a native method, which put it's result
664 // into the result register. So move the value out of the return register back
665 // to the TOS cache of current frame.
666
667 switch (state) {
668 case ltos:
669 case btos:
670 case ztos:
671 case ctos:
672 case stos:
673 case atos:
674 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache
675 case ftos:
676 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
677 case vtos: break; // Nothing to do, this was a void return.
678 default : ShouldNotReachHere();
679 }
680
681 // Load LcpoolCache @@@ should be already set!
682 __ get_constant_pool_cache(R27_constPoolCache);
683
684 // Handle a pending exception, fall through if none.
685 __ check_and_forward_exception(R11_scratch1, R12_scratch2);
686
687 // Start executing bytecodes.
688 if (continuation == NULL) {
689 __ dispatch_next(state, step);
690 } else {
691 __ jump_to_entry(continuation, R11_scratch1);
692 }
693
694 return entry;
695 }
696
697 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
698 address entry = __ pc();
699
700 __ push(state);
701 __ call_VM(noreg, runtime_entry);
702 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
703
704 return entry;
705 }
706
707 // Helpers for commoning out cases in the various type of method entries.
708
709 // Increment invocation count & check for overflow.
710 //
711 // Note: checking for negative value instead of overflow
712 // so we have a 'sticky' overflow test.
713 //
714 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
715 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
716 Register Rscratch1 = R11_scratch1;
717 Register Rscratch2 = R12_scratch2;
718 Register R3_counters = R3_ARG1;
719 Label done;
720
721 if (TieredCompilation) {
722 const int increment = InvocationCounter::count_increment;
723 Label no_mdo;
724 if (ProfileInterpreter) {
725 const Register Rmdo = R3_counters;
726 // If no method data exists, go to profile_continue.
727 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
728 __ cmpdi(CCR0, Rmdo, 0);
729 __ beq(CCR0, no_mdo);
730
731 // Increment invocation counter in the MDO.
732 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
733 __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
734 __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo);
735 __ addi(Rscratch2, Rscratch2, increment);
736 __ stw(Rscratch2, mdo_ic_offs, Rmdo);
737 __ and_(Rscratch1, Rscratch2, Rscratch1);
738 __ bne(CCR0, done);
739 __ b(*overflow);
740 }
741
742 // Increment counter in MethodCounters*.
743 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
744 __ bind(no_mdo);
745 __ get_method_counters(R19_method, R3_counters, done);
746 __ lwz(Rscratch2, mo_ic_offs, R3_counters);
747 __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters);
748 __ addi(Rscratch2, Rscratch2, increment);
749 __ stw(Rscratch2, mo_ic_offs, R3_counters);
750 __ and_(Rscratch1, Rscratch2, Rscratch1);
751 __ beq(CCR0, *overflow);
752
753 __ bind(done);
754
755 } else {
756
757 // Update standard invocation counters.
758 Register Rsum_ivc_bec = R4_ARG2;
759 __ get_method_counters(R19_method, R3_counters, done);
760 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2);
761 // Increment interpreter invocation counter.
762 if (ProfileInterpreter) { // %%% Merge this into methodDataOop.
763 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
764 __ addi(R12_scratch2, R12_scratch2, 1);
765 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
766 }
767 // Check if we must create a method data obj.
768 if (ProfileInterpreter && profile_method != NULL) {
769 const Register profile_limit = Rscratch1;
770 __ lwz(profile_limit, in_bytes(MethodCounters::interpreter_profile_limit_offset()), R3_counters);
771 // Test to see if we should create a method data oop.
772 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit);
773 __ blt(CCR0, *profile_method_continue);
774 // If no method data exists, go to profile_method.
775 __ test_method_data_pointer(*profile_method);
776 }
777 // Finally check for counter overflow.
778 if (overflow) {
779 const Register invocation_limit = Rscratch1;
780 __ lwz(invocation_limit, in_bytes(MethodCounters::interpreter_invocation_limit_offset()), R3_counters);
781 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit);
782 __ bge(CCR0, *overflow);
783 }
784
785 __ bind(done);
786 }
787 }
788
789 // Generate code to initiate compilation on invocation counter overflow.
790 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
791 // Generate code to initiate compilation on the counter overflow.
792
793 // InterpreterRuntime::frequency_counter_overflow takes one arguments,
794 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp)
795 // We pass zero in.
796 // The call returns the address of the verified entry point for the method or NULL
797 // if the compilation did not complete (either went background or bailed out).
798 //
799 // Unlike the C++ interpreter above: Check exceptions!
800 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
801 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
802
803 __ li(R4_ARG2, 0);
804 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
805
806 // Returns verified_entry_point or NULL.
807 // We ignore it in any case.
808 __ b(continue_entry);
809 }
810
811 // See if we've got enough room on the stack for locals plus overhead below
812 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
813 // without going through the signal handler, i.e., reserved and yellow zones
814 // will not be made usable. The shadow zone must suffice to handle the
815 // overflow.
816 //
817 // Kills Rmem_frame_size, Rscratch1.
818 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
819 Label done;
820 assert_different_registers(Rmem_frame_size, Rscratch1);
821
822 BLOCK_COMMENT("stack_overflow_check_with_compare {");
823 __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
824 __ ld(Rscratch1, thread_(stack_overflow_limit));
825 __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
826 __ bgt(CCR0/*is_stack_overflow*/, done);
827
828 // The stack overflows. Load target address of the runtime stub and call it.
829 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order");
830 __ load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0);
831 __ mtctr(Rscratch1);
832 // Restore caller_sp (c2i adapter may exist, but no shrinking of interpreted caller frame).
833 #ifdef ASSERT
834 Label frame_not_shrunk;
835 __ cmpld(CCR0, R1_SP, R21_sender_SP);
836 __ ble(CCR0, frame_not_shrunk);
837 __ stop("frame shrunk", 0x546);
838 __ bind(frame_not_shrunk);
839 __ ld(Rscratch1, 0, R1_SP);
840 __ ld(R0, 0, R21_sender_SP);
841 __ cmpd(CCR0, R0, Rscratch1);
842 __ asm_assert_eq("backlink", 0x547);
843 #endif // ASSERT
844 __ mr(R1_SP, R21_sender_SP);
845 __ bctr();
846
847 __ align(32, 12);
848 __ bind(done);
849 BLOCK_COMMENT("} stack_overflow_check_with_compare");
850 }
851
852 // Lock the current method, interpreter register window must be set up!
853 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
854 const Register Robj_to_lock = Rscratch2;
855
856 {
857 if (!flags_preloaded) {
858 __ lwz(Rflags, method_(access_flags));
859 }
860
861 #ifdef ASSERT
862 // Check if methods needs synchronization.
863 {
864 Label Lok;
865 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
866 __ btrue(CCR0,Lok);
867 __ stop("method doesn't need synchronization");
868 __ bind(Lok);
869 }
870 #endif // ASSERT
871 }
872
873 // Get synchronization object to Rscratch2.
874 {
875 Label Lstatic;
876 Label Ldone;
877
878 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
879 __ btrue(CCR0, Lstatic);
880
881 // Non-static case: load receiver obj from stack and we're done.
882 __ ld(Robj_to_lock, R18_locals);
883 __ b(Ldone);
884
885 __ bind(Lstatic); // Static case: Lock the java mirror
886 // Load mirror from interpreter frame.
887 __ ld(Robj_to_lock, _abi(callers_sp), R1_SP);
888 __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock);
889
890 __ bind(Ldone);
891 __ verify_oop(Robj_to_lock);
892 }
893
894 // Got the oop to lock => execute!
895 __ add_monitor_to_stack(true, Rscratch1, R0);
896
897 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
898 __ lock_object(R26_monitor, Robj_to_lock);
899 }
900
901 // Generate a fixed interpreter frame for pure interpreter
902 // and I2N native transition frames.
903 //
904 // Before (stack grows downwards):
905 //
906 // | ... |
907 // |------------- |
908 // | java arg0 |
909 // | ... |
910 // | java argn |
911 // | | <- R15_esp
912 // | |
913 // |--------------|
914 // | abi_112 |
915 // | | <- R1_SP
916 // |==============|
917 //
918 //
919 // After:
920 //
921 // | ... |
922 // | java arg0 |<- R18_locals
923 // | ... |
924 // | java argn |
925 // |--------------|
926 // | |
927 // | java locals |
928 // | |
929 // |--------------|
930 // | abi_48 |
931 // |==============|
932 // | |
933 // | istate |
934 // | |
935 // |--------------|
936 // | monitor |<- R26_monitor
937 // |--------------|
938 // | |<- R15_esp
939 // | expression |
940 // | stack |
941 // | |
942 // |--------------|
943 // | |
944 // | abi_112 |<- R1_SP
945 // |==============|
946 //
947 // The top most frame needs an abi space of 112 bytes. This space is needed,
948 // since we call to c. The c function may spill their arguments to the caller
949 // frame. When we call to java, we don't need these spill slots. In order to save
950 // space on the stack, we resize the caller. However, java locals reside in
951 // the caller frame and the frame has to be increased. The frame_size for the
952 // current frame was calculated based on max_stack as size for the expression
953 // stack. At the call, just a part of the expression stack might be used.
954 // We don't want to waste this space and cut the frame back accordingly.
955 // The resulting amount for resizing is calculated as follows:
956 // resize = (number_of_locals - number_of_arguments) * slot_size
957 // + (R1_SP - R15_esp) + 48
958 //
959 // The size for the callee frame is calculated:
960 // framesize = 112 + max_stack + monitor + state_size
961 //
962 // maxstack: Max number of slots on the expression stack, loaded from the method.
963 // monitor: We statically reserve room for one monitor object.
964 // state_size: We save the current state of the interpreter to this area.
965 //
966 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
967 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
968 top_frame_size = R7_ARG5,
969 Rconst_method = R8_ARG6;
970
971 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size);
972
973 __ ld(Rconst_method, method_(const));
974 __ lhz(Rsize_of_parameters /* number of params */,
975 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
976 if (native_call) {
977 // If we're calling a native method, we reserve space for the worst-case signature
978 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2).
979 // We add two slots to the parameter_count, one for the jni
980 // environment and one for a possible native mirror.
981 Label skip_native_calculate_max_stack;
982 __ addi(top_frame_size, Rsize_of_parameters, 2);
983 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters);
984 __ bge(CCR0, skip_native_calculate_max_stack);
985 __ li(top_frame_size, Argument::n_register_parameters);
986 __ bind(skip_native_calculate_max_stack);
987 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
988 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
989 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
990 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
991 } else {
992 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
993 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
994 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
995 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
996 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
997 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
998 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
999 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1);
1000 }
1001
1002 // Compute top frame size.
1003 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size);
1004
1005 // Cut back area between esp and max_stack.
1006 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize);
1007
1008 __ round_to(top_frame_size, frame::alignment_in_bytes);
1009 __ round_to(parent_frame_resize, frame::alignment_in_bytes);
1010 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
1011 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
1012
1013 if (!native_call) {
1014 // Stack overflow check.
1015 // Native calls don't need the stack size check since they have no
1016 // expression stack and the arguments are already on the stack and
1017 // we only add a handful of words to the stack.
1018 __ add(R11_scratch1, parent_frame_resize, top_frame_size);
1019 generate_stack_overflow_check(R11_scratch1, R12_scratch2);
1020 }
1021
1022 // Set up interpreter state registers.
1023
1024 __ add(R18_locals, R15_esp, Rsize_of_parameters);
1025 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
1026 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache);
1027
1028 // Set method data pointer.
1029 if (ProfileInterpreter) {
1030 Label zero_continue;
1031 __ ld(R28_mdx, method_(method_data));
1032 __ cmpdi(CCR0, R28_mdx, 0);
1033 __ beq(CCR0, zero_continue);
1034 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1035 __ bind(zero_continue);
1036 }
1037
1038 if (native_call) {
1039 __ li(R14_bcp, 0); // Must initialize.
1040 } else {
1041 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method);
1042 }
1043
1044 // Resize parent frame.
1045 __ mflr(R12_scratch2);
1046 __ neg(parent_frame_resize, parent_frame_resize);
1047 __ resize_frame(parent_frame_resize, R11_scratch1);
1048 __ std(R12_scratch2, _abi(lr), R1_SP);
1049
1050 // Get mirror and store it in the frame as GC root for this Method*.
1051 __ load_mirror_from_const_method(R12_scratch2, Rconst_method);
1052
1053 __ addi(R26_monitor, R1_SP, - frame::ijava_state_size);
1054 __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
1055
1056 // Store values.
1057 // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls
1058 // in InterpreterMacroAssembler::call_from_interpreter.
1059 __ std(R19_method, _ijava_state_neg(method), R1_SP);
1060 __ std(R12_scratch2, _ijava_state_neg(mirror), R1_SP);
1061 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
1062 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
1063 __ std(R18_locals, _ijava_state_neg(locals), R1_SP);
1064
1065 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
1066 // be found in the frame after save_interpreter_state is done. This is always true
1067 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
1068 // because e.g. frame::interpreter_frame_bcp() will not access the correct value
1069 // (Enhanced Stack Trace).
1070 // The signal handler does not save the interpreter state into the frame.
1071 __ li(R0, 0);
1072 #ifdef ASSERT
1073 // Fill remaining slots with constants.
1074 __ load_const_optimized(R11_scratch1, 0x5afe);
1075 __ load_const_optimized(R12_scratch2, 0xdead);
1076 #endif
1077 // We have to initialize some frame slots for native calls (accessed by GC).
1078 if (native_call) {
1079 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP);
1080 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
1081 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
1082 }
1083 #ifdef ASSERT
1084 else {
1085 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
1086 __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP);
1087 __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP);
1088 }
1089 __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP);
1090 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
1091 __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP);
1092 __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP);
1093 #endif
1094 __ subf(R12_scratch2, top_frame_size, R1_SP);
1095 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP);
1096 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
1097
1098 // Push top frame.
1099 __ push_frame(top_frame_size, R11_scratch1);
1100 }
1101
1102 // End of helpers
1103
1104 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1105
1106 // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1107 bool use_instruction = false;
1108 address runtime_entry = NULL;
1109 int num_args = 1;
1110 bool double_precision = true;
1111
1112 // PPC64 specific:
1113 switch (kind) {
1114 case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break;
1115 case Interpreter::java_lang_math_abs: use_instruction = true; break;
1116 case Interpreter::java_lang_math_fmaF:
1117 case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1118 default: break; // Fall back to runtime call.
1119 }
1120
1121 switch (kind) {
1122 case Interpreter::java_lang_math_sin : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); break;
1123 case Interpreter::java_lang_math_cos : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); break;
1124 case Interpreter::java_lang_math_tan : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); break;
1125 case Interpreter::java_lang_math_abs : /* run interpreted */ break;
1126 case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); break;
1127 case Interpreter::java_lang_math_log : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); break;
1128 case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1129 case Interpreter::java_lang_math_pow : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1130 case Interpreter::java_lang_math_exp : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); break;
1131 case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1132 case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1133 default: ShouldNotReachHere();
1134 }
1135
1136 // Use normal entry if neither instruction nor runtime call is used.
1137 if (!use_instruction && runtime_entry == NULL) return NULL;
1138
1139 address entry = __ pc();
1140
1141 // Load arguments
1142 assert(num_args <= 13, "passed in registers");
1143 if (double_precision) {
1144 int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1145 for (int i = 0; i < num_args; ++i) {
1146 __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1147 offset -= 2 * Interpreter::stackElementSize;
1148 }
1149 } else {
1150 int offset = num_args * Interpreter::stackElementSize;
1151 for (int i = 0; i < num_args; ++i) {
1152 __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1153 offset -= Interpreter::stackElementSize;
1154 }
1155 }
1156
1157 if (use_instruction) {
1158 switch (kind) {
1159 case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1); break;
1160 case Interpreter::java_lang_math_abs: __ fabs(F1_RET, F1); break;
1161 case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break;
1162 case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3); break;
1163 default: ShouldNotReachHere();
1164 }
1165 } else {
1166 // Comment: Can use tail call if the unextended frame is always C ABI compliant:
1167 //__ load_const_optimized(R12_scratch2, runtime_entry, R0);
1168 //__ call_c_and_return_to_caller(R12_scratch2);
1169
1170 // Push a new C frame and save LR.
1171 __ save_LR_CR(R0);
1172 __ push_frame_reg_args(0, R11_scratch1);
1173
1174 __ call_VM_leaf(runtime_entry);
1175
1176 // Pop the C frame and restore LR.
1177 __ pop_frame();
1178 __ restore_LR_CR(R0);
1179 }
1180
1181 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1182 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1183 __ blr();
1184
1185 __ flush();
1186
1187 return entry;
1188 }
1189
1190 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1191 // Quick & dirty stack overflow checking: bang the stack & handle trap.
1192 // Note that we do the banging after the frame is setup, since the exception
1193 // handling code expects to find a valid interpreter frame on the stack.
1194 // Doing the banging earlier fails if the caller frame is not an interpreter
1195 // frame.
1196 // (Also, the exception throwing code expects to unlock any synchronized
1197 // method receiever, so do the banging after locking the receiver.)
1198
1199 // Bang each page in the shadow zone. We can't assume it's been done for
1200 // an interpreter frame with greater than a page of locals, so each page
1201 // needs to be checked. Only true for non-native.
1202 if (UseStackBanging) {
1203 const int page_size = os::vm_page_size();
1204 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
1205 const int start_page = native_call ? n_shadow_pages : 1;
1206 BLOCK_COMMENT("bang_stack_shadow_pages:");
1207 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1208 __ bang_stack_with_offset(pages*page_size);
1209 }
1210 }
1211 }
1212
1213 // Interpreter stub for calling a native method. (asm interpreter)
1214 // This sets up a somewhat different looking stack for calling the
1215 // native method than the typical interpreter frame setup.
1216 //
1217 // On entry:
1218 // R19_method - method
1219 // R16_thread - JavaThread*
1220 // R15_esp - intptr_t* sender tos
1221 //
1222 // abstract stack (grows up)
1223 // [ IJava (caller of JNI callee) ] <-- ASP
1224 // ...
1225 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1226
1227 address entry = __ pc();
1228
1229 const bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1230
1231 // -----------------------------------------------------------------------------
1232 // Allocate a new frame that represents the native callee (i2n frame).
1233 // This is not a full-blown interpreter frame, but in particular, the
1234 // following registers are valid after this:
1235 // - R19_method
1236 // - R18_local (points to start of arguments to native function)
1237 //
1238 // abstract stack (grows up)
1239 // [ IJava (caller of JNI callee) ] <-- ASP
1240 // ...
1241
1242 const Register signature_handler_fd = R11_scratch1;
1243 const Register pending_exception = R0;
1244 const Register result_handler_addr = R31;
1245 const Register native_method_fd = R11_scratch1;
1246 const Register access_flags = R22_tmp2;
1247 const Register active_handles = R11_scratch1; // R26_monitor saved to state.
1248 const Register sync_state = R12_scratch2;
1249 const Register sync_state_addr = sync_state; // Address is dead after use.
1250 const Register suspend_flags = R11_scratch1;
1251
1252 //=============================================================================
1253 // Allocate new frame and initialize interpreter state.
1254
1255 Label exception_return;
1256 Label exception_return_sync_check;
1257 Label stack_overflow_return;
1258
1259 // Generate new interpreter state and jump to stack_overflow_return in case of
1260 // a stack overflow.
1261 //generate_compute_interpreter_state(stack_overflow_return);
1262
1263 Register size_of_parameters = R22_tmp2;
1264
1265 generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
1266
1267 //=============================================================================
1268 // Increment invocation counter. On overflow, entry to JNI method
1269 // will be compiled.
1270 Label invocation_counter_overflow, continue_after_compile;
1271 if (inc_counter) {
1272 if (synchronized) {
1273 // Since at this point in the method invocation the exception handler
1274 // would try to exit the monitor of synchronized methods which hasn't
1275 // been entered yet, we set the thread local variable
1276 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1277 // runtime, exception handling i.e. unlock_if_synchronized_method will
1278 // check this thread local flag.
1279 // This flag has two effects, one is to force an unwind in the topmost
1280 // interpreter frame and not perform an unlock while doing so.
1281 __ li(R0, 1);
1282 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1283 }
1284 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1285
1286 BIND(continue_after_compile);
1287 }
1288
1289 bang_stack_shadow_pages(true);
1290
1291 if (inc_counter) {
1292 // Reset the _do_not_unlock_if_synchronized flag.
1293 if (synchronized) {
1294 __ li(R0, 0);
1295 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1296 }
1297 }
1298
1299 // access_flags = method->access_flags();
1300 // Load access flags.
1301 assert(access_flags->is_nonvolatile(),
1302 "access_flags must be in a non-volatile register");
1303 // Type check.
1304 assert(4 == sizeof(AccessFlags), "unexpected field size");
1305 __ lwz(access_flags, method_(access_flags));
1306
1307 // We don't want to reload R19_method and access_flags after calls
1308 // to some helper functions.
1309 assert(R19_method->is_nonvolatile(),
1310 "R19_method must be a non-volatile register");
1311
1312 // Check for synchronized methods. Must happen AFTER invocation counter
1313 // check, so method is not locked if counter overflows.
1314
1315 if (synchronized) {
1316 lock_method(access_flags, R11_scratch1, R12_scratch2, true);
1317
1318 // Update monitor in state.
1319 __ ld(R11_scratch1, 0, R1_SP);
1320 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1);
1321 }
1322
1323 // jvmti/jvmpi support
1324 __ notify_method_entry();
1325
1326 //=============================================================================
1327 // Get and call the signature handler.
1328
1329 __ ld(signature_handler_fd, method_(signature_handler));
1330 Label call_signature_handler;
1331
1332 __ cmpdi(CCR0, signature_handler_fd, 0);
1333 __ bne(CCR0, call_signature_handler);
1334
1335 // Method has never been called. Either generate a specialized
1336 // handler or point to the slow one.
1337 //
1338 // Pass parameter 'false' to avoid exception check in call_VM.
1339 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
1340
1341 // Check for an exception while looking up the target method. If we
1342 // incurred one, bail.
1343 __ ld(pending_exception, thread_(pending_exception));
1344 __ cmpdi(CCR0, pending_exception, 0);
1345 __ bne(CCR0, exception_return_sync_check); // Has pending exception.
1346
1347 // Reload signature handler, it may have been created/assigned in the meanwhile.
1348 __ ld(signature_handler_fd, method_(signature_handler));
1349 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
1350
1351 BIND(call_signature_handler);
1352
1353 // Before we call the signature handler we push a new frame to
1354 // protect the interpreter frame volatile registers when we return
1355 // from jni but before we can get back to Java.
1356
1357 // First set the frame anchor while the SP/FP registers are
1358 // convenient and the slow signature handler can use this same frame
1359 // anchor.
1360
1361 // We have a TOP_IJAVA_FRAME here, which belongs to us.
1362 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
1363
1364 // Now the interpreter frame (and its call chain) have been
1365 // invalidated and flushed. We are now protected against eager
1366 // being enabled in native code. Even if it goes eager the
1367 // registers will be reloaded as clean and we will invalidate after
1368 // the call so no spurious flush should be possible.
1369
1370 // Call signature handler and pass locals address.
1371 //
1372 // Our signature handlers copy required arguments to the C stack
1373 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
1374 __ mr(R3_ARG1, R18_locals);
1375 #if !defined(ABI_ELFv2)
1376 __ ld(signature_handler_fd, 0, signature_handler_fd);
1377 #endif
1378
1379 __ call_stub(signature_handler_fd);
1380
1381 // Remove the register parameter varargs slots we allocated in
1382 // compute_interpreter_state. SP+16 ends up pointing to the ABI
1383 // outgoing argument area.
1384 //
1385 // Not needed on PPC64.
1386 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord);
1387
1388 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
1389 // Save across call to native method.
1390 __ mr(result_handler_addr, R3_RET);
1391
1392 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
1393
1394 // Set up fixed parameters and call the native method.
1395 // If the method is static, get mirror into R4_ARG2.
1396 {
1397 Label method_is_not_static;
1398 // Access_flags is non-volatile and still, no need to restore it.
1399
1400 // Restore access flags.
1401 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
1402 __ bfalse(CCR0, method_is_not_static);
1403
1404 __ ld(R11_scratch1, _abi(callers_sp), R1_SP);
1405 // Load mirror from interpreter frame.
1406 __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1);
1407 // R4_ARG2 = &state->_oop_temp;
1408 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
1409 __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
1410 BIND(method_is_not_static);
1411 }
1412
1413 // At this point, arguments have been copied off the stack into
1414 // their JNI positions. Oops are boxed in-place on the stack, with
1415 // handles copied to arguments. The result handler address is in a
1416 // register.
1417
1418 // Pass JNIEnv address as first parameter.
1419 __ addir(R3_ARG1, thread_(jni_environment));
1420
1421 // Load the native_method entry before we change the thread state.
1422 __ ld(native_method_fd, method_(native_function));
1423
1424 //=============================================================================
1425 // Transition from _thread_in_Java to _thread_in_native. As soon as
1426 // we make this change the safepoint code needs to be certain that
1427 // the last Java frame we established is good. The pc in that frame
1428 // just needs to be near here not an actual return address.
1429
1430 // We use release_store_fence to update values like the thread state, where
1431 // we don't want the current thread to continue until all our prior memory
1432 // accesses (including the new thread state) are visible to other threads.
1433 __ li(R0, _thread_in_native);
1434 __ release();
1435
1436 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
1437 __ stw(R0, thread_(thread_state));
1438
1439 //=============================================================================
1440 // Call the native method. Argument registers must not have been
1441 // overwritten since "__ call_stub(signature_handler);" (except for
1442 // ARG1 and ARG2 for static methods).
1443 __ call_c(native_method_fd);
1444
1445 __ li(R0, 0);
1446 __ ld(R11_scratch1, 0, R1_SP);
1447 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1448 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1449 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
1450
1451 // Note: C++ interpreter needs the following here:
1452 // The frame_manager_lr field, which we use for setting the last
1453 // java frame, gets overwritten by the signature handler. Restore
1454 // it now.
1455 //__ get_PC_trash_LR(R11_scratch1);
1456 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
1457
1458 // Because of GC R19_method may no longer be valid.
1459
1460 // Block, if necessary, before resuming in _thread_in_Java state.
1461 // In order for GC to work, don't clear the last_Java_sp until after
1462 // blocking.
1463
1464 //=============================================================================
1465 // Switch thread to "native transition" state before reading the
1466 // synchronization state. This additional state is necessary
1467 // because reading and testing the synchronization state is not
1468 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1469 // in _thread_in_native state, loads _not_synchronized and is
1470 // preempted. VM thread changes sync state to synchronizing and
1471 // suspends threads for GC. Thread A is resumed to finish this
1472 // native method, but doesn't block here since it didn't see any
1473 // synchronization in progress, and escapes.
1474
1475 // We use release_store_fence to update values like the thread state, where
1476 // we don't want the current thread to continue until all our prior memory
1477 // accesses (including the new thread state) are visible to other threads.
1478 __ li(R0/*thread_state*/, _thread_in_native_trans);
1479 __ release();
1480 __ stw(R0/*thread_state*/, thread_(thread_state));
1481 __ fence();
1482
1483 // Now before we return to java we must look for a current safepoint
1484 // (a new safepoint can not start since we entered native_trans).
1485 // We must check here because a current safepoint could be modifying
1486 // the callers registers right this moment.
1487
1488 // Acquire isn't strictly necessary here because of the fence, but
1489 // sync_state is declared to be volatile, so we do it anyway
1490 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1491
1492 Label do_safepoint, sync_check_done;
1493 // No synchronization in progress nor yet synchronized.
1494 __ safepoint_poll(do_safepoint, sync_state);
1495
1496 // Not suspended.
1497 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1498 __ lwz(suspend_flags, thread_(suspend_flags));
1499 __ cmpwi(CCR1, suspend_flags, 0);
1500 __ beq(CCR1, sync_check_done);
1501
1502 __ bind(do_safepoint);
1503 __ isync();
1504 // Block. We do the call directly and leave the current
1505 // last_Java_frame setup undisturbed. We must save any possible
1506 // native result across the call. No oop is present.
1507
1508 __ mr(R3_ARG1, R16_thread);
1509 #if defined(ABI_ELFv2)
1510 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
1511 relocInfo::none);
1512 #else
1513 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
1514 relocInfo::none);
1515 #endif
1516
1517 __ bind(sync_check_done);
1518
1519 //=============================================================================
1520 // <<<<<< Back in Interpreter Frame >>>>>
1521
1522 // We are in thread_in_native_trans here and back in the normal
1523 // interpreter frame. We don't have to do anything special about
1524 // safepoints and we can switch to Java mode anytime we are ready.
1525
1526 // Note: frame::interpreter_frame_result has a dependency on how the
1527 // method result is saved across the call to post_method_exit. For
1528 // native methods it assumes that the non-FPU/non-void result is
1529 // saved in _native_lresult and a FPU result in _native_fresult. If
1530 // this changes then the interpreter_frame_result implementation
1531 // will need to be updated too.
1532
1533 // On PPC64, we have stored the result directly after the native call.
1534
1535 //=============================================================================
1536 // Back in Java
1537
1538 // We use release_store_fence to update values like the thread state, where
1539 // we don't want the current thread to continue until all our prior memory
1540 // accesses (including the new thread state) are visible to other threads.
1541 __ li(R0/*thread_state*/, _thread_in_Java);
1542 __ lwsync(); // Acquire safepoint and suspend state, release thread state.
1543 __ stw(R0/*thread_state*/, thread_(thread_state));
1544
1545 if (CheckJNICalls) {
1546 // clear_pending_jni_exception_check
1547 __ load_const_optimized(R0, 0L);
1548 __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread);
1549 }
1550
1551 __ reset_last_Java_frame();
1552
1553 // Jvmdi/jvmpi support. Whether we've got an exception pending or
1554 // not, and whether unlocking throws an exception or not, we notify
1555 // on native method exit. If we do have an exception, we'll end up
1556 // in the caller's context to handle it, so if we don't do the
1557 // notify here, we'll drop it on the floor.
1558 __ notify_method_exit(true/*native method*/,
1559 ilgl /*illegal state (not used for native methods)*/,
1560 InterpreterMacroAssembler::NotifyJVMTI,
1561 false /*check_exceptions*/);
1562
1563 //=============================================================================
1564 // Handle exceptions
1565
1566 if (synchronized) {
1567 // Don't check for exceptions since we're still in the i2n frame. Do that
1568 // manually afterwards.
1569 __ unlock_object(R26_monitor, false); // Can also unlock methods.
1570 }
1571
1572 // Reset active handles after returning from native.
1573 // thread->active_handles()->clear();
1574 __ ld(active_handles, thread_(active_handles));
1575 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1576 __ li(R0, 0);
1577 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles);
1578
1579 Label exception_return_sync_check_already_unlocked;
1580 __ ld(R0/*pending_exception*/, thread_(pending_exception));
1581 __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1582 __ bne(CCR0, exception_return_sync_check_already_unlocked);
1583
1584 //-----------------------------------------------------------------------------
1585 // No exception pending.
1586
1587 // Move native method result back into proper registers and return.
1588 // Invoke result handler (may unbox/promote).
1589 __ ld(R11_scratch1, 0, R1_SP);
1590 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1591 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1592 __ call_stub(result_handler_addr);
1593
1594 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1595
1596 // Must use the return pc which was loaded from the caller's frame
1597 // as the VM uses return-pc-patching for deoptimization.
1598 __ mtlr(R0);
1599 __ blr();
1600
1601 //-----------------------------------------------------------------------------
1602 // An exception is pending. We call into the runtime only if the
1603 // caller was not interpreted. If it was interpreted the
1604 // interpreter will do the correct thing. If it isn't interpreted
1605 // (call stub/compiled code) we will change our return and continue.
1606
1607 BIND(exception_return_sync_check);
1608
1609 if (synchronized) {
1610 // Don't check for exceptions since we're still in the i2n frame. Do that
1611 // manually afterwards.
1612 __ unlock_object(R26_monitor, false); // Can also unlock methods.
1613 }
1614 BIND(exception_return_sync_check_already_unlocked);
1615
1616 const Register return_pc = R31;
1617
1618 __ ld(return_pc, 0, R1_SP);
1619 __ ld(return_pc, _abi(lr), return_pc);
1620
1621 // Get the address of the exception handler.
1622 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1623 R16_thread,
1624 return_pc /* return pc */);
1625 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1626
1627 // Load the PC of the the exception handler into LR.
1628 __ mtlr(R3_RET);
1629
1630 // Load exception into R3_ARG1 and clear pending exception in thread.
1631 __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1632 __ li(R4_ARG2, 0);
1633 __ std(R4_ARG2, thread_(pending_exception));
1634
1635 // Load the original return pc into R4_ARG2.
1636 __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1637
1638 // Return to exception handler.
1639 __ blr();
1640
1641 //=============================================================================
1642 // Counter overflow.
1643
1644 if (inc_counter) {
1645 // Handle invocation counter overflow.
1646 __ bind(invocation_counter_overflow);
1647
1648 generate_counter_overflow(continue_after_compile);
1649 }
1650
1651 return entry;
1652 }
1653
1654 // Generic interpreted method entry to (asm) interpreter.
1655 //
1656 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1657 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1658 address entry = __ pc();
1659 // Generate the code to allocate the interpreter stack frame.
1660 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1661 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame.
1662
1663 // Does also a stack check to assure this frame fits on the stack.
1664 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1665
1666 // --------------------------------------------------------------------------
1667 // Zero out non-parameter locals.
1668 // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1669 // worth to ask the flag, just do it.
1670 Register Rslot_addr = R6_ARG4,
1671 Rnum = R7_ARG5;
1672 Label Lno_locals, Lzero_loop;
1673
1674 // Set up the zeroing loop.
1675 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1676 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1677 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1678 __ beq(CCR0, Lno_locals);
1679 __ li(R0, 0);
1680 __ mtctr(Rnum);
1681
1682 // The zero locals loop.
1683 __ bind(Lzero_loop);
1684 __ std(R0, 0, Rslot_addr);
1685 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1686 __ bdnz(Lzero_loop);
1687
1688 __ bind(Lno_locals);
1689
1690 // --------------------------------------------------------------------------
1691 // Counter increment and overflow check.
1692 Label invocation_counter_overflow,
1693 profile_method,
1694 profile_method_continue;
1695 if (inc_counter || ProfileInterpreter) {
1696
1697 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1698 if (synchronized) {
1699 // Since at this point in the method invocation the exception handler
1700 // would try to exit the monitor of synchronized methods which hasn't
1701 // been entered yet, we set the thread local variable
1702 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1703 // runtime, exception handling i.e. unlock_if_synchronized_method will
1704 // check this thread local flag.
1705 // This flag has two effects, one is to force an unwind in the topmost
1706 // interpreter frame and not perform an unlock while doing so.
1707 __ li(R0, 1);
1708 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1709 }
1710
1711 // Argument and return type profiling.
1712 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1713
1714 // Increment invocation counter and check for overflow.
1715 if (inc_counter) {
1716 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1717 }
1718
1719 __ bind(profile_method_continue);
1720 }
1721
1722 bang_stack_shadow_pages(false);
1723
1724 if (inc_counter || ProfileInterpreter) {
1725 // Reset the _do_not_unlock_if_synchronized flag.
1726 if (synchronized) {
1727 __ li(R0, 0);
1728 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1729 }
1730 }
1731
1732 // --------------------------------------------------------------------------
1733 // Locking of synchronized methods. Must happen AFTER invocation_counter
1734 // check and stack overflow check, so method is not locked if overflows.
1735 if (synchronized) {
1736 lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1737 }
1738 #ifdef ASSERT
1739 else {
1740 Label Lok;
1741 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1742 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1743 __ asm_assert_eq("method needs synchronization", 0x8521);
1744 __ bind(Lok);
1745 }
1746 #endif // ASSERT
1747
1748 __ verify_thread();
1749
1750 // --------------------------------------------------------------------------
1751 // JVMTI support
1752 __ notify_method_entry();
1753
1754 // --------------------------------------------------------------------------
1755 // Start executing instructions.
1756 __ dispatch_next(vtos);
1757
1758 // --------------------------------------------------------------------------
1759 // Out of line counter overflow and MDO creation code.
1760 if (ProfileInterpreter) {
1761 // We have decided to profile this method in the interpreter.
1762 __ bind(profile_method);
1763 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1764 __ set_method_data_pointer_for_bcp();
1765 __ b(profile_method_continue);
1766 }
1767
1768 if (inc_counter) {
1769 // Handle invocation counter overflow.
1770 __ bind(invocation_counter_overflow);
1771 generate_counter_overflow(profile_method_continue);
1772 }
1773 return entry;
1774 }
1775
1776 // CRC32 Intrinsics.
1777 //
1778 // Contract on scratch and work registers.
1779 // =======================================
1780 //
1781 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1782 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1783 // You can't rely on these registers across calls.
1784 //
1785 // The generators for CRC32_update and for CRC32_updateBytes use the
1786 // scratch/work register set internally, passing the work registers
1787 // as arguments to the MacroAssembler emitters as required.
1788 //
1789 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1790 // Their contents is not constant but may change according to the requirements
1791 // of the emitted code.
1792 //
1793 // All other registers from the scratch/work register set are used "internally"
1794 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1795 // Basically, only R3_RET contains a defined value which is the function result.
1796 //
1797 /**
1798 * Method entry for static native methods:
1799 * int java.util.zip.CRC32.update(int crc, int b)
1800 */
1801 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1802 if (UseCRC32Intrinsics) {
1803 address start = __ pc(); // Remember stub start address (is rtn value).
1804 Label slow_path;
1805
1806 // Safepoint check
1807 const Register sync_state = R11_scratch1;
1808 __ safepoint_poll(slow_path, sync_state);
1809
1810 // We don't generate local frame and don't align stack because
1811 // we not even call stub code (we generate the code inline)
1812 // and there is no safepoint on this path.
1813
1814 // Load java parameters.
1815 // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1816 const Register argP = R15_esp;
1817 const Register crc = R3_ARG1; // crc value
1818 const Register data = R4_ARG2;
1819 const Register table = R5_ARG3; // address of crc32 table
1820
1821 BLOCK_COMMENT("CRC32_update {");
1822
1823 // Arguments are reversed on java expression stack
1824 #ifdef VM_LITTLE_ENDIAN
1825 int data_offs = 0+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1826 // Being passed as an int, the single byte is at offset +0.
1827 #else
1828 int data_offs = 3+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1829 // Being passed from java as an int, the single byte is at offset +3.
1830 #endif
1831 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1832 __ lbz(data, data_offs, argP); // Byte from buffer, zero-extended.
1833 __ load_const_optimized(table, StubRoutines::crc_table_addr(), R0);
1834 __ kernel_crc32_singleByteReg(crc, data, table, true);
1835
1836 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1837 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1838 __ blr();
1839
1840 // Generate a vanilla native entry as the slow path.
1841 BLOCK_COMMENT("} CRC32_update");
1842 BIND(slow_path);
1843 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1844 return start;
1845 }
1846
1847 return NULL;
1848 }
1849
1850 /**
1851 * Method entry for static native methods:
1852 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len)
1853 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1854 */
1855 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1856 if (UseCRC32Intrinsics) {
1857 address start = __ pc(); // Remember stub start address (is rtn value).
1858 Label slow_path;
1859
1860 // Safepoint check
1861 const Register sync_state = R11_scratch1;
1862 __ safepoint_poll(slow_path, sync_state);
1863
1864 // We don't generate local frame and don't align stack because
1865 // we not even call stub code (we generate the code inline)
1866 // and there is no safepoint on this path.
1867
1868 // Load parameters.
1869 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1870 const Register argP = R15_esp;
1871 const Register crc = R3_ARG1; // crc value
1872 const Register data = R4_ARG2; // address of java byte array
1873 const Register dataLen = R5_ARG3; // source data len
1874 const Register tmp = R11_scratch1;
1875
1876 // Arguments are reversed on java expression stack.
1877 // Calculate address of start element.
1878 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1879 BLOCK_COMMENT("CRC32_updateByteBuffer {");
1880 // crc @ (SP + 5W) (32bit)
1881 // buf @ (SP + 3W) (64bit ptr to long array)
1882 // off @ (SP + 2W) (32bit)
1883 // dataLen @ (SP + 1W) (32bit)
1884 // data = buf + off
1885 __ ld( data, 3*wordSize, argP); // start of byte buffer
1886 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1887 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1888 __ lwz( crc, 5*wordSize, argP); // current crc state
1889 __ add( data, data, tmp); // Add byte buffer offset.
1890 } else { // Used for "updateBytes update".
1891 BLOCK_COMMENT("CRC32_updateBytes {");
1892 // crc @ (SP + 4W) (32bit)
1893 // buf @ (SP + 3W) (64bit ptr to byte array)
1894 // off @ (SP + 2W) (32bit)
1895 // dataLen @ (SP + 1W) (32bit)
1896 // data = buf + off + base_offset
1897 __ ld( data, 3*wordSize, argP); // start of byte buffer
1898 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1899 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1900 __ add( data, data, tmp); // add byte buffer offset
1901 __ lwz( crc, 4*wordSize, argP); // current crc state
1902 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1903 }
1904
1905 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, false);
1906
1907 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1908 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1909 __ blr();
1910
1911 // Generate a vanilla native entry as the slow path.
1912 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1913 BIND(slow_path);
1914 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1915 return start;
1916 }
1917
1918 return NULL;
1919 }
1920
1921
1922 /**
1923 * Method entry for intrinsic-candidate (non-native) methods:
1924 * int java.util.zip.CRC32C.updateBytes( int crc, byte[] b, int off, int end)
1925 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
1926 * Unlike CRC32, CRC32C does not have any methods marked as native
1927 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1928 **/
1929 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1930 if (UseCRC32CIntrinsics) {
1931 address start = __ pc(); // Remember stub start address (is rtn value).
1932
1933 // We don't generate local frame and don't align stack because
1934 // we not even call stub code (we generate the code inline)
1935 // and there is no safepoint on this path.
1936
1937 // Load parameters.
1938 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1939 const Register argP = R15_esp;
1940 const Register crc = R3_ARG1; // crc value
1941 const Register data = R4_ARG2; // address of java byte array
1942 const Register dataLen = R5_ARG3; // source data len
1943 const Register tmp = R11_scratch1;
1944
1945 // Arguments are reversed on java expression stack.
1946 // Calculate address of start element.
1947 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer".
1948 BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
1949 // crc @ (SP + 5W) (32bit)
1950 // buf @ (SP + 3W) (64bit ptr to long array)
1951 // off @ (SP + 2W) (32bit)
1952 // dataLen @ (SP + 1W) (32bit)
1953 // data = buf + off
1954 __ ld( data, 3*wordSize, argP); // start of byte buffer
1955 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1956 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1957 __ lwz( crc, 5*wordSize, argP); // current crc state
1958 __ add( data, data, tmp); // Add byte buffer offset.
1959 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1960 } else { // Used for "updateBytes update".
1961 BLOCK_COMMENT("CRC32C_updateBytes {");
1962 // crc @ (SP + 4W) (32bit)
1963 // buf @ (SP + 3W) (64bit ptr to byte array)
1964 // off @ (SP + 2W) (32bit)
1965 // dataLen @ (SP + 1W) (32bit)
1966 // data = buf + off + base_offset
1967 __ ld( data, 3*wordSize, argP); // start of byte buffer
1968 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1969 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1970 __ add( data, data, tmp); // add byte buffer offset
1971 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1972 __ lwz( crc, 4*wordSize, argP); // current crc state
1973 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1974 }
1975
1976 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, true);
1977
1978 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1979 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1980 __ blr();
1981
1982 BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
1983 return start;
1984 }
1985
1986 return NULL;
1987 }
1988
1989 // =============================================================================
1990 // Exceptions
1991
1992 void TemplateInterpreterGenerator::generate_throw_exception() {
1993 Register Rexception = R17_tos,
1994 Rcontinuation = R3_RET;
1995
1996 // --------------------------------------------------------------------------
1997 // Entry point if an method returns with a pending exception (rethrow).
1998 Interpreter::_rethrow_exception_entry = __ pc();
1999 {
2000 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
2001 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
2002 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
2003
2004 // Compiled code destroys templateTableBase, reload.
2005 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
2006 }
2007
2008 // Entry point if a interpreted method throws an exception (throw).
2009 Interpreter::_throw_exception_entry = __ pc();
2010 {
2011 __ mr(Rexception, R3_RET);
2012
2013 __ verify_thread();
2014 __ verify_oop(Rexception);
2015
2016 // Expression stack must be empty before entering the VM in case of an exception.
2017 __ empty_expression_stack();
2018 // Find exception handler address and preserve exception oop.
2019 // Call C routine to find handler and jump to it.
2020 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
2021 __ mtctr(Rcontinuation);
2022 // Push exception for exception handler bytecodes.
2023 __ push_ptr(Rexception);
2024
2025 // Jump to exception handler (may be remove activation entry!).
2026 __ bctr();
2027 }
2028
2029 // If the exception is not handled in the current frame the frame is
2030 // removed and the exception is rethrown (i.e. exception
2031 // continuation is _rethrow_exception).
2032 //
2033 // Note: At this point the bci is still the bxi for the instruction
2034 // which caused the exception and the expression stack is
2035 // empty. Thus, for any VM calls at this point, GC will find a legal
2036 // oop map (with empty expression stack).
2037
2038 // In current activation
2039 // tos: exception
2040 // bcp: exception bcp
2041
2042 // --------------------------------------------------------------------------
2043 // JVMTI PopFrame support
2044
2045 Interpreter::_remove_activation_preserving_args_entry = __ pc();
2046 {
2047 // Set the popframe_processing bit in popframe_condition indicating that we are
2048 // currently handling popframe, so that call_VMs that may happen later do not
2049 // trigger new popframe handling cycles.
2050 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2051 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
2052 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2053
2054 // Empty the expression stack, as in normal exception handling.
2055 __ empty_expression_stack();
2056 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
2057
2058 // Check to see whether we are returning to a deoptimized frame.
2059 // (The PopFrame call ensures that the caller of the popped frame is
2060 // either interpreted or compiled and deoptimizes it if compiled.)
2061 // Note that we don't compare the return PC against the
2062 // deoptimization blob's unpack entry because of the presence of
2063 // adapter frames in C2.
2064 Label Lcaller_not_deoptimized;
2065 Register return_pc = R3_ARG1;
2066 __ ld(return_pc, 0, R1_SP);
2067 __ ld(return_pc, _abi(lr), return_pc);
2068 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
2069 __ cmpdi(CCR0, R3_RET, 0);
2070 __ bne(CCR0, Lcaller_not_deoptimized);
2071
2072 // The deoptimized case.
2073 // In this case, we can't call dispatch_next() after the frame is
2074 // popped, but instead must save the incoming arguments and restore
2075 // them after deoptimization has occurred.
2076 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
2077 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
2078 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
2079 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
2080 __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
2081 // Save these arguments.
2082 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
2083
2084 // Inform deoptimization that it is responsible for restoring these arguments.
2085 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
2086 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2087
2088 // Return from the current method into the deoptimization blob. Will eventually
2089 // end up in the deopt interpeter entry, deoptimization prepared everything that
2090 // we will reexecute the call that called us.
2091 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
2092 __ mtlr(return_pc);
2093 __ blr();
2094
2095 // The non-deoptimized case.
2096 __ bind(Lcaller_not_deoptimized);
2097
2098 // Clear the popframe condition flag.
2099 __ li(R0, 0);
2100 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2101
2102 // Get out of the current method and re-execute the call that called us.
2103 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2104 __ restore_interpreter_state(R11_scratch1);
2105 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
2106 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
2107 if (ProfileInterpreter) {
2108 __ set_method_data_pointer_for_bcp();
2109 __ ld(R11_scratch1, 0, R1_SP);
2110 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
2111 }
2112 #if INCLUDE_JVMTI
2113 Label L_done;
2114
2115 __ lbz(R11_scratch1, 0, R14_bcp);
2116 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
2117 __ bne(CCR0, L_done);
2118
2119 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2120 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
2121 __ ld(R4_ARG2, 0, R18_locals);
2122 __ MacroAssembler::call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp, false);
2123 __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2124 __ cmpdi(CCR0, R4_ARG2, 0);
2125 __ beq(CCR0, L_done);
2126 __ std(R4_ARG2, wordSize, R15_esp);
2127 __ bind(L_done);
2128 #endif // INCLUDE_JVMTI
2129 __ dispatch_next(vtos);
2130 }
2131 // end of JVMTI PopFrame support
2132
2133 // --------------------------------------------------------------------------
2134 // Remove activation exception entry.
2135 // This is jumped to if an interpreted method can't handle an exception itself
2136 // (we come from the throw/rethrow exception entry above). We're going to call
2137 // into the VM to find the exception handler in the caller, pop the current
2138 // frame and return the handler we calculated.
2139 Interpreter::_remove_activation_entry = __ pc();
2140 {
2141 __ pop_ptr(Rexception);
2142 __ verify_thread();
2143 __ verify_oop(Rexception);
2144 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
2145
2146 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
2147 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
2148
2149 __ get_vm_result(Rexception);
2150
2151 // We are done with this activation frame; find out where to go next.
2152 // The continuation point will be an exception handler, which expects
2153 // the following registers set up:
2154 //
2155 // RET: exception oop
2156 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
2157
2158 Register return_pc = R31; // Needs to survive the runtime call.
2159 __ ld(return_pc, 0, R1_SP);
2160 __ ld(return_pc, _abi(lr), return_pc);
2161 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
2162
2163 // Remove the current activation.
2164 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2165
2166 __ mr(R4_ARG2, return_pc);
2167 __ mtlr(R3_RET);
2168 __ mr(R3_RET, Rexception);
2169 __ blr();
2170 }
2171 }
2172
2173 // JVMTI ForceEarlyReturn support.
2174 // Returns "in the middle" of a method with a "fake" return value.
2175 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2176
2177 Register Rscratch1 = R11_scratch1,
2178 Rscratch2 = R12_scratch2;
2179
2180 address entry = __ pc();
2181 __ empty_expression_stack();
2182
2183 __ load_earlyret_value(state, Rscratch1);
2184
2185 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
2186 // Clear the earlyret state.
2187 __ li(R0, 0);
2188 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
2189
2190 __ remove_activation(state, false, false);
2191 // Copied from TemplateTable::_return.
2192 // Restoration of lr done by remove_activation.
2193 switch (state) {
2194 // Narrow result if state is itos but result type is smaller.
2195 case btos:
2196 case ztos:
2197 case ctos:
2198 case stos:
2199 case itos: __ narrow(R17_tos); /* fall through */
2200 case ltos:
2201 case atos: __ mr(R3_RET, R17_tos); break;
2202 case ftos:
2203 case dtos: __ fmr(F1_RET, F15_ftos); break;
2204 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2205 // to get visible before the reference to the object gets stored anywhere.
2206 __ membar(Assembler::StoreStore); break;
2207 default : ShouldNotReachHere();
2208 }
2209 __ blr();
2210
2211 return entry;
2212 } // end of ForceEarlyReturn support
2213
2214 //-----------------------------------------------------------------------------
2215 // Helper for vtos entry point generation
2216
2217 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2218 address& bep,
2219 address& cep,
2220 address& sep,
2221 address& aep,
2222 address& iep,
2223 address& lep,
2224 address& fep,
2225 address& dep,
2226 address& vep) {
2227 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2228 Label L;
2229
2230 aep = __ pc(); __ push_ptr(); __ b(L);
2231 fep = __ pc(); __ push_f(); __ b(L);
2232 dep = __ pc(); __ push_d(); __ b(L);
2233 lep = __ pc(); __ push_l(); __ b(L);
2234 __ align(32, 12, 24); // align L
2235 bep = cep = sep =
2236 iep = __ pc(); __ push_i();
2237 vep = __ pc();
2238 __ bind(L);
2239 generate_and_dispatch(t);
2240 }
2241
2242 //-----------------------------------------------------------------------------
2243
2244 // Non-product code
2245 #ifndef PRODUCT
2246 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2247 //__ flush_bundle();
2248 address entry = __ pc();
2249
2250 const char *bname = NULL;
2251 uint tsize = 0;
2252 switch(state) {
2253 case ftos:
2254 bname = "trace_code_ftos {";
2255 tsize = 2;
2256 break;
2257 case btos:
2258 bname = "trace_code_btos {";
2259 tsize = 2;
2260 break;
2261 case ztos:
2262 bname = "trace_code_ztos {";
2263 tsize = 2;
2264 break;
2265 case ctos:
2266 bname = "trace_code_ctos {";
2267 tsize = 2;
2268 break;
2269 case stos:
2270 bname = "trace_code_stos {";
2271 tsize = 2;
2272 break;
2273 case itos:
2274 bname = "trace_code_itos {";
2275 tsize = 2;
2276 break;
2277 case ltos:
2278 bname = "trace_code_ltos {";
2279 tsize = 3;
2280 break;
2281 case atos:
2282 bname = "trace_code_atos {";
2283 tsize = 2;
2284 break;
2285 case vtos:
2286 // Note: In case of vtos, the topmost of stack value could be a int or doubl
2287 // In case of a double (2 slots) we won't see the 2nd stack value.
2288 // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
2289 bname = "trace_code_vtos {";
2290 tsize = 2;
2291
2292 break;
2293 case dtos:
2294 bname = "trace_code_dtos {";
2295 tsize = 3;
2296 break;
2297 default:
2298 ShouldNotReachHere();
2299 }
2300 BLOCK_COMMENT(bname);
2301
2302 // Support short-cut for TraceBytecodesAt.
2303 // Don't call into the VM if we don't want to trace to speed up things.
2304 Label Lskip_vm_call;
2305 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2306 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
2307 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2308 __ ld(R11_scratch1, offs1, R11_scratch1);
2309 __ lwa(R12_scratch2, offs2, R12_scratch2);
2310 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2311 __ blt(CCR0, Lskip_vm_call);
2312 }
2313
2314 __ push(state);
2315 // Load 2 topmost expression stack values.
2316 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
2317 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
2318 __ mflr(R31);
2319 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
2320 __ mtlr(R31);
2321 __ pop(state);
2322
2323 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2324 __ bind(Lskip_vm_call);
2325 }
2326 __ blr();
2327 BLOCK_COMMENT("} trace_code");
2328 return entry;
2329 }
2330
2331 void TemplateInterpreterGenerator::count_bytecode() {
2332 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
2333 __ lwz(R12_scratch2, offs, R11_scratch1);
2334 __ addi(R12_scratch2, R12_scratch2, 1);
2335 __ stw(R12_scratch2, offs, R11_scratch1);
2336 }
2337
2338 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2339 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
2340 __ lwz(R12_scratch2, offs, R11_scratch1);
2341 __ addi(R12_scratch2, R12_scratch2, 1);
2342 __ stw(R12_scratch2, offs, R11_scratch1);
2343 }
2344
2345 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2346 const Register addr = R11_scratch1,
2347 tmp = R12_scratch2;
2348 // Get index, shift out old bytecode, bring in new bytecode, and store it.
2349 // _index = (_index >> log2_number_of_codes) |
2350 // (bytecode << log2_number_of_codes);
2351 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
2352 __ lwz(tmp, offs1, addr);
2353 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
2354 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2355 __ stw(tmp, offs1, addr);
2356
2357 // Bump bucket contents.
2358 // _counters[_index] ++;
2359 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
2360 __ sldi(tmp, tmp, LogBytesPerInt);
2361 __ add(addr, tmp, addr);
2362 __ lwz(tmp, offs2, addr);
2363 __ addi(tmp, tmp, 1);
2364 __ stw(tmp, offs2, addr);
2365 }
2366
2367 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2368 // Call a little run-time stub to avoid blow-up for each bytecode.
2369 // The run-time runtime saves the right registers, depending on
2370 // the tosca in-state for the given template.
2371
2372 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2373 "entry must have been generated");
2374
2375 // Note: we destroy LR here.
2376 __ bl(Interpreter::trace_code(t->tos_in()));
2377 }
2378
2379 void TemplateInterpreterGenerator::stop_interpreter_at() {
2380 Label L;
2381 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
2382 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2383 __ ld(R11_scratch1, offs1, R11_scratch1);
2384 __ lwa(R12_scratch2, offs2, R12_scratch2);
2385 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2386 __ bne(CCR0, L);
2387 __ illtrap();
2388 __ bind(L);
2389 }
2390
2391 #endif // !PRODUCT
--- EOF ---