1 /*
2 * Copyright (c) 2014, 2023, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2023 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 "classfile/javaClasses.hpp"
29 #include "compiler/disassembler.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interpreter/bytecodeHistogram.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interpreterRuntime.hpp"
34 #include "interpreter/interp_masm.hpp"
35 #include "interpreter/templateInterpreterGenerator.hpp"
36 #include "interpreter/templateTable.hpp"
37 #include "oops/arrayOop.hpp"
38 #include "oops/method.hpp"
39 #include "oops/methodCounters.hpp"
40 #include "oops/methodData.hpp"
41 #include "oops/oop.inline.hpp"
42 #include "oops/resolvedIndyEntry.hpp"
43 #include "oops/resolvedMethodEntry.hpp"
44 #include "prims/jvmtiExport.hpp"
45 #include "prims/jvmtiThreadState.hpp"
46 #include "runtime/arguments.hpp"
47 #include "runtime/deoptimization.hpp"
48 #include "runtime/frame.inline.hpp"
49 #include "runtime/jniHandles.hpp"
50 #include "runtime/sharedRuntime.hpp"
51 #include "runtime/stubRoutines.hpp"
52 #include "runtime/synchronizer.hpp"
53 #include "runtime/timer.hpp"
54 #include "runtime/vframeArray.hpp"
55 #include "runtime/vm_version.hpp"
56 #include "utilities/debug.hpp"
57 #include "utilities/macros.hpp"
58
59 #undef __
60 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
61
62 // Size of interpreter code. Increase if too small. Interpreter will
63 // fail with a guarantee ("not enough space for interpreter generation");
64 // if too small.
65 // Run with +PrintInterpreter to get the VM to print out the size.
66 // Max size with JVMTI
67 int TemplateInterpreter::InterpreterCodeSize = 256*K;
68
69 #ifdef PRODUCT
70 #define BLOCK_COMMENT(str) /* nothing */
71 #else
72 #define BLOCK_COMMENT(str) __ block_comment(str)
73 #endif
74
75 #define BIND(label) __ bind(label); BLOCK_COMMENT(#label ":")
76
77 //-----------------------------------------------------------------------------
78
79 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
80 // Slow_signature handler that respects the PPC C calling conventions.
81 //
82 // We get called by the native entry code with our output register
83 // area == 8. First we call InterpreterRuntime::get_result_handler
84 // to copy the pointer to the signature string temporarily to the
85 // first C-argument and to return the result_handler in
86 // R3_RET. Since native_entry will copy the jni-pointer to the
87 // first C-argument slot later on, it is OK to occupy this slot
88 // temporarily. Then we copy the argument list on the java
89 // expression stack into native varargs format on the native stack
90 // and load arguments into argument registers. Integer arguments in
91 // the varargs vector will be sign-extended to 8 bytes.
92 //
93 // On entry:
94 // R3_ARG1 - intptr_t* Address of java argument list in memory.
95 // R15_prev_state - BytecodeInterpreter* Address of interpreter state for
96 // this method
97 // R19_method
98 //
99 // On exit (just before return instruction):
100 // R3_RET - contains the address of the result_handler.
101 // R4_ARG2 - is not updated for static methods and contains "this" otherwise.
102 // R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double,
103 // ARGi contains this argument. Otherwise, ARGi is not updated.
104 // F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double.
105
106 const int LogSizeOfTwoInstructions = 3;
107
108 // FIXME: use Argument:: GL: Argument names different numbers!
109 const int max_fp_register_arguments = 13;
110 const int max_int_register_arguments = 6; // first 2 are reserved
111
112 const Register arg_java = R21_tmp1;
113 const Register arg_c = R22_tmp2;
114 const Register signature = R23_tmp3; // is string
115 const Register sig_byte = R24_tmp4;
116 const Register fpcnt = R25_tmp5;
117 const Register argcnt = R26_tmp6;
118 const Register intSlot = R27_tmp7;
119 const Register target_sp = R28_tmp8;
120 const FloatRegister floatSlot = F0;
121
122 address entry = __ function_entry();
123
124 __ save_LR_CR(R0);
125 __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
126 // We use target_sp for storing arguments in the C frame.
127 __ mr(target_sp, R1_SP);
128 __ push_frame_reg_args_nonvolatiles(0, R11_scratch1);
129
130 __ mr(arg_java, R3_ARG1);
131
132 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method);
133
134 // Signature is in R3_RET. Signature is callee saved.
135 __ mr(signature, R3_RET);
136
137 // Get the result handler.
138 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method);
139
140 {
141 Label L;
142 // test if static
143 // _access_flags._flags must be at offset 0.
144 // TODO PPC port: requires change in shared code.
145 //assert(in_bytes(AccessFlags::flags_offset()) == 0,
146 // "MethodDesc._access_flags == MethodDesc._access_flags._flags");
147 // _access_flags must be a 32 bit value.
148 assert(sizeof(AccessFlags) == 4, "wrong size");
149 __ lwa(R11_scratch1/*access_flags*/, method_(access_flags));
150 // testbit with condition register.
151 __ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT);
152 __ btrue(CCR0, L);
153 // For non-static functions, pass "this" in R4_ARG2 and copy it
154 // to 2nd C-arg slot.
155 // We need to box the Java object here, so we use arg_java
156 // (address of current Java stack slot) as argument and don't
157 // dereference it as in case of ints, floats, etc.
158 __ mr(R4_ARG2, arg_java);
159 __ addi(arg_java, arg_java, -BytesPerWord);
160 __ std(R4_ARG2, _abi0(carg_2), target_sp);
161 __ bind(L);
162 }
163
164 // Will be incremented directly after loop_start. argcnt=0
165 // corresponds to 3rd C argument.
166 __ li(argcnt, -1);
167 // arg_c points to 3rd C argument
168 __ addi(arg_c, target_sp, _abi0(carg_3));
169 // no floating-point args parsed so far
170 __ li(fpcnt, 0);
171
172 Label move_intSlot_to_ARG, move_floatSlot_to_FARG;
173 Label loop_start, loop_end;
174 Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed;
175
176 // signature points to '(' at entry
177 #ifdef ASSERT
178 __ lbz(sig_byte, 0, signature);
179 __ cmplwi(CCR0, sig_byte, '(');
180 __ bne(CCR0, do_dontreachhere);
181 #endif
182
183 __ bind(loop_start);
184
185 __ addi(argcnt, argcnt, 1);
186 __ lbzu(sig_byte, 1, signature);
187
188 __ cmplwi(CCR0, sig_byte, ')'); // end of signature
189 __ beq(CCR0, loop_end);
190
191 __ cmplwi(CCR0, sig_byte, 'B'); // byte
192 __ beq(CCR0, do_int);
193
194 __ cmplwi(CCR0, sig_byte, 'C'); // char
195 __ beq(CCR0, do_int);
196
197 __ cmplwi(CCR0, sig_byte, 'D'); // double
198 __ beq(CCR0, do_double);
199
200 __ cmplwi(CCR0, sig_byte, 'F'); // float
201 __ beq(CCR0, do_float);
202
203 __ cmplwi(CCR0, sig_byte, 'I'); // int
204 __ beq(CCR0, do_int);
205
206 __ cmplwi(CCR0, sig_byte, 'J'); // long
207 __ beq(CCR0, do_long);
208
209 __ cmplwi(CCR0, sig_byte, 'S'); // short
210 __ beq(CCR0, do_int);
211
212 __ cmplwi(CCR0, sig_byte, 'Z'); // boolean
213 __ beq(CCR0, do_int);
214
215 __ cmplwi(CCR0, sig_byte, 'L'); // object
216 __ beq(CCR0, do_object);
217
218 __ cmplwi(CCR0, sig_byte, '['); // array
219 __ beq(CCR0, do_array);
220
221 // __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type
222 // __ beq(CCR0, do_void);
223
224 __ bind(do_dontreachhere);
225
226 __ unimplemented("ShouldNotReachHere in slow_signature_handler");
227
228 __ bind(do_array);
229
230 {
231 Label start_skip, end_skip;
232
233 __ bind(start_skip);
234 __ lbzu(sig_byte, 1, signature);
235 __ cmplwi(CCR0, sig_byte, '[');
236 __ beq(CCR0, start_skip); // skip further brackets
237 __ cmplwi(CCR0, sig_byte, '9');
238 __ bgt(CCR0, end_skip); // no optional size
239 __ cmplwi(CCR0, sig_byte, '0');
240 __ bge(CCR0, start_skip); // skip optional size
241 __ bind(end_skip);
242
243 __ cmplwi(CCR0, sig_byte, 'L');
244 __ beq(CCR0, do_object); // for arrays of objects, the name of the object must be skipped
245 __ b(do_boxed); // otherwise, go directly to do_boxed
246 }
247
248 __ bind(do_object);
249 {
250 Label L;
251 __ bind(L);
252 __ lbzu(sig_byte, 1, signature);
253 __ cmplwi(CCR0, sig_byte, ';');
254 __ bne(CCR0, L);
255 }
256 // Need to box the Java object here, so we use arg_java (address of
257 // current Java stack slot) as argument and don't dereference it as
258 // in case of ints, floats, etc.
259 Label do_null;
260 __ bind(do_boxed);
261 __ ld(R0,0, arg_java);
262 __ cmpdi(CCR0, R0, 0);
263 __ li(intSlot,0);
264 __ beq(CCR0, do_null);
265 __ mr(intSlot, arg_java);
266 __ bind(do_null);
267 __ std(intSlot, 0, arg_c);
268 __ addi(arg_java, arg_java, -BytesPerWord);
269 __ addi(arg_c, arg_c, BytesPerWord);
270 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
271 __ blt(CCR0, move_intSlot_to_ARG);
272 __ b(loop_start);
273
274 __ bind(do_int);
275 __ lwa(intSlot, 0, arg_java);
276 __ std(intSlot, 0, arg_c);
277 __ addi(arg_java, arg_java, -BytesPerWord);
278 __ addi(arg_c, arg_c, BytesPerWord);
279 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
280 __ blt(CCR0, move_intSlot_to_ARG);
281 __ b(loop_start);
282
283 __ bind(do_long);
284 __ ld(intSlot, -BytesPerWord, arg_java);
285 __ std(intSlot, 0, arg_c);
286 __ addi(arg_java, arg_java, - 2 * BytesPerWord);
287 __ addi(arg_c, arg_c, BytesPerWord);
288 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
289 __ blt(CCR0, move_intSlot_to_ARG);
290 __ b(loop_start);
291
292 __ bind(do_float);
293 __ lfs(floatSlot, 0, arg_java);
294 __ stfs(floatSlot, Argument::float_on_stack_offset_in_bytes_c, arg_c);
295 __ addi(arg_java, arg_java, -BytesPerWord);
296 __ addi(arg_c, arg_c, BytesPerWord);
297 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
298 __ blt(CCR0, move_floatSlot_to_FARG);
299 __ b(loop_start);
300
301 __ bind(do_double);
302 __ lfd(floatSlot, - BytesPerWord, arg_java);
303 __ stfd(floatSlot, 0, arg_c);
304 __ addi(arg_java, arg_java, - 2 * BytesPerWord);
305 __ addi(arg_c, arg_c, BytesPerWord);
306 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
307 __ blt(CCR0, move_floatSlot_to_FARG);
308 __ b(loop_start);
309
310 __ bind(loop_end);
311
312 __ pop_frame();
313 __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
314 __ restore_LR_CR(R0);
315
316 __ blr();
317
318 Label move_int_arg, move_float_arg;
319 __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
320 __ mr(R5_ARG3, intSlot); __ b(loop_start);
321 __ mr(R6_ARG4, intSlot); __ b(loop_start);
322 __ mr(R7_ARG5, intSlot); __ b(loop_start);
323 __ mr(R8_ARG6, intSlot); __ b(loop_start);
324 __ mr(R9_ARG7, intSlot); __ b(loop_start);
325 __ mr(R10_ARG8, intSlot); __ b(loop_start);
326
327 __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
328 __ fmr(F1_ARG1, floatSlot); __ b(loop_start);
329 __ fmr(F2_ARG2, floatSlot); __ b(loop_start);
330 __ fmr(F3_ARG3, floatSlot); __ b(loop_start);
331 __ fmr(F4_ARG4, floatSlot); __ b(loop_start);
332 __ fmr(F5_ARG5, floatSlot); __ b(loop_start);
333 __ fmr(F6_ARG6, floatSlot); __ b(loop_start);
334 __ fmr(F7_ARG7, floatSlot); __ b(loop_start);
335 __ fmr(F8_ARG8, floatSlot); __ b(loop_start);
336 __ fmr(F9_ARG9, floatSlot); __ b(loop_start);
337 __ fmr(F10_ARG10, floatSlot); __ b(loop_start);
338 __ fmr(F11_ARG11, floatSlot); __ b(loop_start);
339 __ fmr(F12_ARG12, floatSlot); __ b(loop_start);
340 __ fmr(F13_ARG13, floatSlot); __ b(loop_start);
341
342 __ bind(move_intSlot_to_ARG);
343 __ sldi(R0, argcnt, LogSizeOfTwoInstructions);
344 __ load_const(R11_scratch1, move_int_arg); // Label must be bound here.
345 __ add(R11_scratch1, R0, R11_scratch1);
346 __ mtctr(R11_scratch1/*branch_target*/);
347 __ bctr();
348 __ bind(move_floatSlot_to_FARG);
349 __ sldi(R0, fpcnt, LogSizeOfTwoInstructions);
350 __ addi(fpcnt, fpcnt, 1);
351 __ load_const(R11_scratch1, move_float_arg); // Label must be bound here.
352 __ add(R11_scratch1, R0, R11_scratch1);
353 __ mtctr(R11_scratch1/*branch_target*/);
354 __ bctr();
355
356 return entry;
357 }
358
359 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
360 //
361 // Registers alive
362 // R3_RET
363 // LR
364 //
365 // Registers updated
366 // R3_RET
367 //
368
369 Label done;
370 address entry = __ pc();
371
372 switch (type) {
373 case T_BOOLEAN:
374 // convert !=0 to 1
375 __ neg(R0, R3_RET);
376 __ orr(R0, R3_RET, R0);
377 __ srwi(R3_RET, R0, 31);
378 break;
379 case T_BYTE:
380 // sign extend 8 bits
381 __ extsb(R3_RET, R3_RET);
382 break;
383 case T_CHAR:
384 // zero extend 16 bits
385 __ clrldi(R3_RET, R3_RET, 48);
386 break;
387 case T_SHORT:
388 // sign extend 16 bits
389 __ extsh(R3_RET, R3_RET);
390 break;
391 case T_INT:
392 // sign extend 32 bits
393 __ extsw(R3_RET, R3_RET);
394 break;
395 case T_LONG:
396 break;
397 case T_OBJECT:
398 // JNIHandles::resolve result.
399 __ resolve_jobject(R3_RET, R11_scratch1, R31, MacroAssembler::PRESERVATION_FRAME_LR); // kills R31
400 break;
401 case T_FLOAT:
402 break;
403 case T_DOUBLE:
404 break;
405 case T_VOID:
406 break;
407 default: ShouldNotReachHere();
408 }
409
410 BIND(done);
411 __ blr();
412
413 return entry;
414 }
415
416 // Abstract method entry.
417 //
418 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
419 address entry = __ pc();
420
421 //
422 // Registers alive
423 // R16_thread - JavaThread*
424 // R19_method - callee's method (method to be invoked)
425 // R1_SP - SP prepared such that caller's outgoing args are near top
426 // LR - return address to caller
427 //
428 // Stack layout at this point:
429 //
430 // 0 [TOP_IJAVA_FRAME_ABI] <-- R1_SP
431 // alignment (optional)
432 // [outgoing Java arguments]
433 // ...
434 // PARENT [PARENT_IJAVA_FRAME_ABI]
435 // ...
436 //
437
438 // Can't use call_VM here because we have not set up a new
439 // interpreter state. Make the call to the vm and make it look like
440 // our caller set up the JavaFrameAnchor.
441 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
442
443 // Push a new C frame and save LR.
444 __ save_LR_CR(R0);
445 __ push_frame_reg_args(0, R11_scratch1);
446
447 // This is not a leaf but we have a JavaFrameAnchor now and we will
448 // check (create) exceptions afterward so this is ok.
449 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
450 R16_thread, R19_method);
451
452 // Pop the C frame and restore LR.
453 __ pop_frame();
454 __ restore_LR_CR(R0);
455
456 // Reset JavaFrameAnchor from call_VM_leaf above.
457 __ reset_last_Java_frame();
458
459 // We don't know our caller, so jump to the general forward exception stub,
460 // which will also pop our full frame off. Satisfy the interface of
461 // SharedRuntime::generate_forward_exception()
462 __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0);
463 __ mtctr(R11_scratch1);
464 __ bctr();
465
466 return entry;
467 }
468
469 // Interpreter intrinsic for WeakReference.get().
470 // 1. Don't push a full blown frame and go on dispatching, but fetch the value
471 // into R8 and return quickly
472 // 2. If G1 is active we *must* execute this intrinsic for corrrectness:
473 // It contains a GC barrier which puts the reference into the satb buffer
474 // to indicate that someone holds a strong reference to the object the
475 // weak ref points to!
476 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
477 // Code: _aload_0, _getfield, _areturn
478 // parameter size = 1
479 //
480 // The code that gets generated by this routine is split into 2 parts:
481 // 1. the "intrinsified" code for G1 (or any SATB based GC),
482 // 2. the slow path - which is an expansion of the regular method entry.
483 //
484 // Notes:
485 // * In the G1 code we do not check whether we need to block for
486 // a safepoint. If G1 is enabled then we must execute the specialized
487 // code for Reference.get (except when the Reference object is null)
488 // so that we can log the value in the referent field with an SATB
489 // update buffer.
490 // If the code for the getfield template is modified so that the
491 // G1 pre-barrier code is executed when the current method is
492 // Reference.get() then going through the normal method entry
493 // will be fine.
494 // * The G1 code can, however, check the receiver object (the instance
495 // of java.lang.Reference) and jump to the slow path if null. If the
496 // Reference object is null then we obviously cannot fetch the referent
497 // and so we don't need to call the G1 pre-barrier. Thus we can use the
498 // regular method entry code to generate the NPE.
499 //
500
501 address entry = __ pc();
502
503 const int referent_offset = java_lang_ref_Reference::referent_offset();
504
505 Label slow_path;
506
507 // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH);
508
509 // In the G1 code we don't check if we need to reach a safepoint. We
510 // continue and the thread will safepoint at the next bytecode dispatch.
511
512 // If the receiver is null then it is OK to jump to the slow path.
513 __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver
514
515 // Check if receiver == nullptr and go the slow path.
516 __ cmpdi(CCR0, R3_RET, 0);
517 __ beq(CCR0, slow_path);
518
519 __ load_heap_oop(R3_RET, referent_offset, R3_RET,
520 /* non-volatile temp */ R31, R11_scratch1,
521 MacroAssembler::PRESERVATION_FRAME_LR,
522 ON_WEAK_OOP_REF);
523
524 // Generate the G1 pre-barrier code to log the value of
525 // the referent field in an SATB buffer. Note with
526 // these parameters the pre-barrier does not generate
527 // the load of the previous value.
528
529 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
530 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
531
532 __ blr();
533
534 __ bind(slow_path);
535 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
536 return entry;
537 }
538
539 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
540 address entry = __ pc();
541
542 // Expression stack must be empty before entering the VM if an
543 // exception happened.
544 __ empty_expression_stack();
545 // Throw exception.
546 __ call_VM(noreg,
547 CAST_FROM_FN_PTR(address,
548 InterpreterRuntime::throw_StackOverflowError));
549 return entry;
550 }
551
552 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
553 address entry = __ pc();
554 __ empty_expression_stack();
555 // R4_ARG2 already contains the array.
556 // Index is in R17_tos.
557 __ mr(R5_ARG3, R17_tos);
558 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R4_ARG2, R5_ARG3);
559 return entry;
560 }
561
562 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
563 address entry = __ pc();
564 // Expression stack must be empty before entering the VM if an
565 // exception happened.
566 __ empty_expression_stack();
567
568 // Load exception object.
569 // Thread will be loaded to R3_ARG1.
570 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
571 #ifdef ASSERT
572 // Above call must not return here since exception pending.
573 __ should_not_reach_here();
574 #endif
575 return entry;
576 }
577
578 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
579 address entry = __ pc();
580 //__ untested("generate_exception_handler_common");
581 Register Rexception = R17_tos;
582
583 // Expression stack must be empty before entering the VM if an exception happened.
584 __ empty_expression_stack();
585
586 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
587 if (pass_oop) {
588 __ mr(R5_ARG3, Rexception);
589 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception));
590 } else {
591 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
592 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception));
593 }
594
595 // Throw exception.
596 __ mr(R3_ARG1, Rexception);
597 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
598 __ mtctr(R11_scratch1);
599 __ bctr();
600
601 return entry;
602 }
603
604 // This entry is returned to when a call returns to the interpreter.
605 // When we arrive here, we expect that the callee stack frame is already popped.
606 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
607 address entry = __ pc();
608
609 // Move the value out of the return register back to the TOS cache of current frame.
610 switch (state) {
611 case ltos:
612 case btos:
613 case ztos:
614 case ctos:
615 case stos:
616 case atos:
617 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache
618 case ftos:
619 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
620 case vtos: break; // Nothing to do, this was a void return.
621 default : ShouldNotReachHere();
622 }
623
624 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
625
626 // Compiled code destroys templateTableBase, reload.
627 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
628
629 if (state == atos) {
630 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
631 }
632
633 const Register cache = R11_scratch1;
634 const Register size = R12_scratch2;
635 if (index_size == sizeof(u4)) {
636 __ load_resolved_indy_entry(cache, size /* tmp */);
637 __ lhz(size, in_bytes(ResolvedIndyEntry::num_parameters_offset()), cache);
638 } else {
639 assert(index_size == sizeof(u2), "Can only be u2");
640 __ load_method_entry(cache, size /* tmp */);
641 __ lhz(size, in_bytes(ResolvedMethodEntry::num_parameters_offset()), cache);
642 }
643 __ sldi(size, size, Interpreter::logStackElementSize);
644 __ add(R15_esp, R15_esp, size);
645
646 __ check_and_handle_popframe(R11_scratch1);
647 __ check_and_handle_earlyret(R11_scratch1);
648
649 __ dispatch_next(state, step);
650 return entry;
651 }
652
653 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
654 address entry = __ pc();
655 // If state != vtos, we're returning from a native method, which put it's result
656 // into the result register. So move the value out of the return register back
657 // to the TOS cache of current frame.
658
659 switch (state) {
660 case ltos:
661 case btos:
662 case ztos:
663 case ctos:
664 case stos:
665 case atos:
666 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache
667 case ftos:
668 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
669 case vtos: break; // Nothing to do, this was a void return.
670 default : ShouldNotReachHere();
671 }
672
673 // Load LcpoolCache @@@ should be already set!
674 __ get_constant_pool_cache(R27_constPoolCache);
675
676 // Handle a pending exception, fall through if none.
677 __ check_and_forward_exception(R11_scratch1, R12_scratch2);
678
679 // Start executing bytecodes.
680 if (continuation == nullptr) {
681 __ dispatch_next(state, step);
682 } else {
683 __ jump_to_entry(continuation, R11_scratch1);
684 }
685
686 return entry;
687 }
688
689 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
690 address entry = __ pc();
691
692 __ push(state);
693 __ push_cont_fastpath();
694 __ call_VM(noreg, runtime_entry);
695 __ pop_cont_fastpath();
696 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
697
698 return entry;
699 }
700
701 // Helpers for commoning out cases in the various type of method entries.
702
703 // Increment invocation count & check for overflow.
704 //
705 // Note: checking for negative value instead of overflow
706 // so we have a 'sticky' overflow test.
707 //
708 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
709 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
710 Register Rscratch1 = R11_scratch1;
711 Register Rscratch2 = R12_scratch2;
712 Register R3_counters = R3_ARG1;
713 Label done;
714
715 const int increment = InvocationCounter::count_increment;
716 Label no_mdo;
717 if (ProfileInterpreter) {
718 const Register Rmdo = R3_counters;
719 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
720 __ cmpdi(CCR0, Rmdo, 0);
721 __ beq(CCR0, no_mdo);
722
723 // Increment invocation counter in the MDO.
724 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
725 __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
726 __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo);
727 __ addi(Rscratch2, Rscratch2, increment);
728 __ stw(Rscratch2, mdo_ic_offs, Rmdo);
729 __ and_(Rscratch1, Rscratch2, Rscratch1);
730 __ bne(CCR0, done);
731 __ b(*overflow);
732 }
733
734 // Increment counter in MethodCounters*.
735 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
736 __ bind(no_mdo);
737 __ get_method_counters(R19_method, R3_counters, done);
738 __ lwz(Rscratch2, mo_ic_offs, R3_counters);
739 __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters);
740 __ addi(Rscratch2, Rscratch2, increment);
741 __ stw(Rscratch2, mo_ic_offs, R3_counters);
742 __ and_(Rscratch1, Rscratch2, Rscratch1);
743 __ beq(CCR0, *overflow);
744
745 __ bind(done);
746 }
747
748 // Generate code to initiate compilation on invocation counter overflow.
749 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
750 // Generate code to initiate compilation on the counter overflow.
751
752 // InterpreterRuntime::frequency_counter_overflow takes one arguments,
753 // which indicates if the counter overflow occurs at a backwards branch (null bcp)
754 // We pass zero in.
755 // The call returns the address of the verified entry point for the method or null
756 // if the compilation did not complete (either went background or bailed out).
757 //
758 // Unlike the C++ interpreter above: Check exceptions!
759 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
760 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
761
762 __ li(R4_ARG2, 0);
763 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
764
765 // Returns verified_entry_point or null.
766 // We ignore it in any case.
767 __ b(continue_entry);
768 }
769
770 // See if we've got enough room on the stack for locals plus overhead below
771 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
772 // without going through the signal handler, i.e., reserved and yellow zones
773 // will not be made usable. The shadow zone must suffice to handle the
774 // overflow.
775 //
776 // Kills Rmem_frame_size, Rscratch1.
777 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
778 Label done;
779 assert_different_registers(Rmem_frame_size, Rscratch1);
780
781 BLOCK_COMMENT("stack_overflow_check_with_compare {");
782 __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
783 __ ld(Rscratch1, thread_(stack_overflow_limit));
784 __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
785 __ bgt(CCR0/*is_stack_overflow*/, done);
786
787 // The stack overflows. Load target address of the runtime stub and call it.
788 assert(StubRoutines::throw_StackOverflowError_entry() != nullptr, "generated in wrong order");
789 __ load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0);
790 __ mtctr(Rscratch1);
791 // Restore caller_sp (c2i adapter may exist, but no shrinking of interpreted caller frame).
792 #ifdef ASSERT
793 Label frame_not_shrunk;
794 __ cmpld(CCR0, R1_SP, R21_sender_SP);
795 __ ble(CCR0, frame_not_shrunk);
796 __ stop("frame shrunk");
797 __ bind(frame_not_shrunk);
798 __ ld(Rscratch1, 0, R1_SP);
799 __ ld(R0, 0, R21_sender_SP);
800 __ cmpd(CCR0, R0, Rscratch1);
801 __ asm_assert_eq("backlink");
802 #endif // ASSERT
803 __ mr(R1_SP, R21_sender_SP);
804 __ bctr();
805
806 __ align(32, 12);
807 __ bind(done);
808 BLOCK_COMMENT("} stack_overflow_check_with_compare");
809 }
810
811 // Lock the current method, interpreter register window must be set up!
812 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
813 const Register Robj_to_lock = Rscratch2;
814
815 {
816 if (!flags_preloaded) {
817 __ lwz(Rflags, method_(access_flags));
818 }
819
820 #ifdef ASSERT
821 // Check if methods needs synchronization.
822 {
823 Label Lok;
824 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
825 __ btrue(CCR0,Lok);
826 __ stop("method doesn't need synchronization");
827 __ bind(Lok);
828 }
829 #endif // ASSERT
830 }
831
832 // Get synchronization object to Rscratch2.
833 {
834 Label Lstatic;
835 Label Ldone;
836
837 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
838 __ btrue(CCR0, Lstatic);
839
840 // Non-static case: load receiver obj from stack and we're done.
841 __ ld(Robj_to_lock, R18_locals);
842 __ b(Ldone);
843
844 __ bind(Lstatic); // Static case: Lock the java mirror
845 // Load mirror from interpreter frame.
846 __ ld(Robj_to_lock, _abi0(callers_sp), R1_SP);
847 __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock);
848
849 __ bind(Ldone);
850 __ verify_oop(Robj_to_lock);
851 }
852
853 // Got the oop to lock => execute!
854 __ add_monitor_to_stack(true, Rscratch1, R0);
855
856 __ std(Robj_to_lock, in_bytes(BasicObjectLock::obj_offset()), R26_monitor);
857 __ lock_object(R26_monitor, Robj_to_lock);
858 }
859
860 // Generate a fixed interpreter frame for pure interpreter
861 // and I2N native transition frames.
862 //
863 // Before (stack grows downwards):
864 //
865 // | ... |
866 // |------------- |
867 // | java arg0 |
868 // | ... |
869 // | java argn |
870 // | | <- R15_esp
871 // | |
872 // |--------------|
873 // | abi_112 |
874 // | | <- R1_SP
875 // |==============|
876 //
877 //
878 // After:
879 //
880 // | ... |
881 // | java arg0 |<- R18_locals
882 // | ... |
883 // | java argn |
884 // |--------------|
885 // | |
886 // | java locals |
887 // | |
888 // |--------------|
889 // | abi_48 |
890 // |==============|
891 // | |
892 // | istate |
893 // | |
894 // |--------------|
895 // | monitor |<- R26_monitor
896 // |--------------|
897 // | |<- R15_esp
898 // | expression |
899 // | stack |
900 // | |
901 // |--------------|
902 // | |
903 // | abi_112 |<- R1_SP
904 // |==============|
905 //
906 // The top most frame needs an abi space of 112 bytes. This space is needed,
907 // since we call to c. The c function may spill their arguments to the caller
908 // frame. When we call to java, we don't need these spill slots. In order to save
909 // space on the stack, we resize the caller. However, java locals reside in
910 // the caller frame and the frame has to be increased. The frame_size for the
911 // current frame was calculated based on max_stack as size for the expression
912 // stack. At the call, just a part of the expression stack might be used.
913 // We don't want to waste this space and cut the frame back accordingly.
914 // The resulting amount for resizing is calculated as follows:
915 // resize = (number_of_locals - number_of_arguments) * slot_size
916 // + (R1_SP - R15_esp) + 48
917 //
918 // The size for the callee frame is calculated:
919 // framesize = 112 + max_stack + monitor + state_size
920 //
921 // maxstack: Max number of slots on the expression stack, loaded from the method.
922 // monitor: We statically reserve room for one monitor object.
923 // state_size: We save the current state of the interpreter to this area.
924 //
925 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
926 Register Rparent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
927 Rtop_frame_size = R7_ARG5,
928 Rconst_method = R8_ARG6,
929 Rconst_pool = R9_ARG7,
930 Rmirror = R10_ARG8;
931
932 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, Rparent_frame_resize, Rtop_frame_size,
933 Rconst_method, Rconst_pool);
934
935 __ ld(Rconst_method, method_(const));
936 __ lhz(Rsize_of_parameters /* number of params */,
937 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
938 if (native_call) {
939 // If we're calling a native method, we reserve space for the worst-case signature
940 // handler varargs vector, which is max(Argument::n_int_register_parameters_c, parameter_count+2).
941 // We add two slots to the parameter_count, one for the jni
942 // environment and one for a possible native mirror.
943 Label skip_native_calculate_max_stack;
944 __ addi(Rtop_frame_size, Rsize_of_parameters, 2);
945 __ cmpwi(CCR0, Rtop_frame_size, Argument::n_int_register_parameters_c);
946 __ bge(CCR0, skip_native_calculate_max_stack);
947 __ li(Rtop_frame_size, Argument::n_int_register_parameters_c);
948 __ bind(skip_native_calculate_max_stack);
949 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
950 __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
951 __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
952 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
953 } else {
954 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
955 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
956 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
957 __ lhz(Rtop_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
958 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
959 __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
960 __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
961 __ add(Rparent_frame_resize, Rparent_frame_resize, R11_scratch1);
962 }
963
964 // Compute top frame size.
965 __ addi(Rtop_frame_size, Rtop_frame_size, frame::top_ijava_frame_abi_size + frame::ijava_state_size);
966
967 // Cut back area between esp and max_stack.
968 __ addi(Rparent_frame_resize, Rparent_frame_resize, frame::parent_ijava_frame_abi_size - Interpreter::stackElementSize);
969
970 __ round_to(Rtop_frame_size, frame::alignment_in_bytes);
971 __ round_to(Rparent_frame_resize, frame::alignment_in_bytes);
972 // Rparent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
973 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
974
975 if (!native_call) {
976 // Stack overflow check.
977 // Native calls don't need the stack size check since they have no
978 // expression stack and the arguments are already on the stack and
979 // we only add a handful of words to the stack.
980 __ add(R11_scratch1, Rparent_frame_resize, Rtop_frame_size);
981 generate_stack_overflow_check(R11_scratch1, R12_scratch2);
982 }
983
984 // Set up interpreter state registers.
985
986 __ add(R18_locals, R15_esp, Rsize_of_parameters);
987 __ ld(Rconst_pool, in_bytes(ConstMethod::constants_offset()), Rconst_method);
988 __ ld(R27_constPoolCache, ConstantPool::cache_offset(), Rconst_pool);
989
990 // Set method data pointer.
991 if (ProfileInterpreter) {
992 Label zero_continue;
993 __ ld(R28_mdx, method_(method_data));
994 __ cmpdi(CCR0, R28_mdx, 0);
995 __ beq(CCR0, zero_continue);
996 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
997 __ bind(zero_continue);
998 }
999
1000 if (native_call) {
1001 __ li(R14_bcp, 0); // Must initialize.
1002 } else {
1003 __ addi(R14_bcp, Rconst_method, in_bytes(ConstMethod::codes_offset()));
1004 }
1005
1006 // Resize parent frame.
1007 __ mflr(R12_scratch2);
1008 __ neg(Rparent_frame_resize, Rparent_frame_resize);
1009 __ resize_frame(Rparent_frame_resize, R11_scratch1);
1010 __ std(R12_scratch2, _abi0(lr), R1_SP);
1011
1012 // Get mirror and store it in the frame as GC root for this Method*.
1013 __ ld(Rmirror, ConstantPool::pool_holder_offset(), Rconst_pool);
1014 __ ld(Rmirror, in_bytes(Klass::java_mirror_offset()), Rmirror);
1015 __ resolve_oop_handle(Rmirror, R11_scratch1, R12_scratch2, MacroAssembler::PRESERVATION_FRAME_LR_GP_REGS);
1016
1017 __ addi(R26_monitor, R1_SP, -frame::ijava_state_size);
1018 __ addi(R15_esp, R26_monitor, -Interpreter::stackElementSize);
1019
1020 // Store values.
1021 __ std(R19_method, _ijava_state_neg(method), R1_SP);
1022 __ std(Rmirror, _ijava_state_neg(mirror), R1_SP);
1023 __ sub(R12_scratch2, R18_locals, R1_SP);
1024 __ srdi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1025 // Store relativized R18_locals, see frame::interpreter_frame_locals().
1026 __ std(R12_scratch2, _ijava_state_neg(locals), R1_SP);
1027 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
1028
1029 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
1030 // be found in the frame after save_interpreter_state is done. This is always true
1031 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
1032 // because e.g. frame::interpreter_frame_bcp() will not access the correct value
1033 // (Enhanced Stack Trace).
1034 // The signal handler does not save the interpreter state into the frame.
1035
1036 // We have to initialize some of these frame slots for native calls (accessed by GC).
1037 // Also initialize them for non-native calls for better tool support (even though
1038 // you may not get the most recent version as described above).
1039 __ li(R0, 0);
1040 __ li(R12_scratch2, -(frame::ijava_state_size / wordSize));
1041 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
1042 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
1043 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
1044 __ sub(R12_scratch2, R15_esp, R1_SP);
1045 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1046 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
1047 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); // only used for native_call
1048
1049 // Store sender's SP and this frame's top SP.
1050 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
1051 __ neg(R12_scratch2, Rtop_frame_size);
1052 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1053 // Store relativized top_frame_sp
1054 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
1055
1056 // Push top frame.
1057 __ push_frame(Rtop_frame_size, R11_scratch1);
1058 }
1059
1060 // End of helpers
1061
1062 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1063
1064 // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1065 bool use_instruction = false;
1066 address runtime_entry = nullptr;
1067 int num_args = 1;
1068 bool double_precision = true;
1069
1070 // PPC64 specific:
1071 switch (kind) {
1072 case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break;
1073 case Interpreter::java_lang_math_abs: use_instruction = true; break;
1074 case Interpreter::java_lang_math_fmaF:
1075 case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1076 default: break; // Fall back to runtime call.
1077 }
1078
1079 switch (kind) {
1080 case Interpreter::java_lang_math_sin : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); break;
1081 case Interpreter::java_lang_math_cos : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); break;
1082 case Interpreter::java_lang_math_tan : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); break;
1083 case Interpreter::java_lang_math_abs : /* run interpreted */ break;
1084 case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); break;
1085 case Interpreter::java_lang_math_log : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); break;
1086 case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1087 case Interpreter::java_lang_math_pow : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1088 case Interpreter::java_lang_math_exp : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); break;
1089 case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1090 case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1091 default: ShouldNotReachHere();
1092 }
1093
1094 // Use normal entry if neither instruction nor runtime call is used.
1095 if (!use_instruction && runtime_entry == nullptr) return nullptr;
1096
1097 address entry = __ pc();
1098
1099 // Load arguments
1100 assert(num_args <= 13, "passed in registers");
1101 if (double_precision) {
1102 int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1103 for (int i = 0; i < num_args; ++i) {
1104 __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1105 offset -= 2 * Interpreter::stackElementSize;
1106 }
1107 } else {
1108 int offset = num_args * Interpreter::stackElementSize;
1109 for (int i = 0; i < num_args; ++i) {
1110 __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1111 offset -= Interpreter::stackElementSize;
1112 }
1113 }
1114
1115 if (use_instruction) {
1116 switch (kind) {
1117 case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1); break;
1118 case Interpreter::java_lang_math_abs: __ fabs(F1_RET, F1); break;
1119 case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break;
1120 case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3); break;
1121 default: ShouldNotReachHere();
1122 }
1123 } else {
1124 // Comment: Can use tail call if the unextended frame is always C ABI compliant:
1125 //__ load_const_optimized(R12_scratch2, runtime_entry, R0);
1126 //__ call_c_and_return_to_caller(R12_scratch2);
1127
1128 // Push a new C frame and save LR.
1129 __ save_LR_CR(R0);
1130 __ push_frame_reg_args(0, R11_scratch1);
1131
1132 __ call_VM_leaf(runtime_entry);
1133
1134 // Pop the C frame and restore LR.
1135 __ pop_frame();
1136 __ restore_LR_CR(R0);
1137 }
1138
1139 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1140 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1141 __ blr();
1142
1143 __ flush();
1144
1145 return entry;
1146 }
1147
1148 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1149 // Quick & dirty stack overflow checking: bang the stack & handle trap.
1150 // Note that we do the banging after the frame is setup, since the exception
1151 // handling code expects to find a valid interpreter frame on the stack.
1152 // Doing the banging earlier fails if the caller frame is not an interpreter
1153 // frame.
1154 // (Also, the exception throwing code expects to unlock any synchronized
1155 // method receiever, so do the banging after locking the receiver.)
1156
1157 // Bang each page in the shadow zone. We can't assume it's been done for
1158 // an interpreter frame with greater than a page of locals, so each page
1159 // needs to be checked. Only true for non-native.
1160 const size_t page_size = os::vm_page_size();
1161 const int n_shadow_pages = StackOverflow::stack_shadow_zone_size() / page_size;
1162 const int start_page = native_call ? n_shadow_pages : 1;
1163 BLOCK_COMMENT("bang_stack_shadow_pages:");
1164 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1165 __ bang_stack_with_offset(pages*page_size);
1166 }
1167 }
1168
1169 // Interpreter stub for calling a native method. (asm interpreter)
1170 // This sets up a somewhat different looking stack for calling the
1171 // native method than the typical interpreter frame setup.
1172 //
1173 // On entry:
1174 // R19_method - method
1175 // R16_thread - JavaThread*
1176 // R15_esp - intptr_t* sender tos
1177 //
1178 // abstract stack (grows up)
1179 // [ IJava (caller of JNI callee) ] <-- ASP
1180 // ...
1181 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1182
1183 address entry = __ pc();
1184
1185 const bool inc_counter = UseCompiler || CountCompiledCalls;
1186
1187 // -----------------------------------------------------------------------------
1188 // Allocate a new frame that represents the native callee (i2n frame).
1189 // This is not a full-blown interpreter frame, but in particular, the
1190 // following registers are valid after this:
1191 // - R19_method
1192 // - R18_local (points to start of arguments to native function)
1193 //
1194 // abstract stack (grows up)
1195 // [ IJava (caller of JNI callee) ] <-- ASP
1196 // ...
1197
1198 const Register signature_handler_fd = R11_scratch1;
1199 const Register pending_exception = R0;
1200 const Register result_handler_addr = R31;
1201 const Register native_method_fd = R11_scratch1;
1202 const Register access_flags = R22_tmp2;
1203 const Register active_handles = R11_scratch1; // R26_monitor saved to state.
1204 const Register sync_state = R12_scratch2;
1205 const Register sync_state_addr = sync_state; // Address is dead after use.
1206 const Register suspend_flags = R11_scratch1;
1207
1208 //=============================================================================
1209 // Allocate new frame and initialize interpreter state.
1210
1211 Label exception_return;
1212 Label exception_return_sync_check;
1213 Label stack_overflow_return;
1214
1215 // Generate new interpreter state and jump to stack_overflow_return in case of
1216 // a stack overflow.
1217 //generate_compute_interpreter_state(stack_overflow_return);
1218
1219 Register size_of_parameters = R22_tmp2;
1220
1221 generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
1222
1223 //=============================================================================
1224 // Increment invocation counter. On overflow, entry to JNI method
1225 // will be compiled.
1226 Label invocation_counter_overflow, continue_after_compile;
1227 if (inc_counter) {
1228 if (synchronized) {
1229 // Since at this point in the method invocation the exception handler
1230 // would try to exit the monitor of synchronized methods which hasn't
1231 // been entered yet, we set the thread local variable
1232 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1233 // runtime, exception handling i.e. unlock_if_synchronized_method will
1234 // check this thread local flag.
1235 // This flag has two effects, one is to force an unwind in the topmost
1236 // interpreter frame and not perform an unlock while doing so.
1237 __ li(R0, 1);
1238 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1239 }
1240 generate_counter_incr(&invocation_counter_overflow);
1241
1242 BIND(continue_after_compile);
1243 }
1244
1245 bang_stack_shadow_pages(true);
1246
1247 if (inc_counter) {
1248 // Reset the _do_not_unlock_if_synchronized flag.
1249 if (synchronized) {
1250 __ li(R0, 0);
1251 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1252 }
1253 }
1254
1255 // access_flags = method->access_flags();
1256 // Load access flags.
1257 assert(access_flags->is_nonvolatile(),
1258 "access_flags must be in a non-volatile register");
1259 // Type check.
1260 assert(4 == sizeof(AccessFlags), "unexpected field size");
1261 __ lwz(access_flags, method_(access_flags));
1262
1263 // We don't want to reload R19_method and access_flags after calls
1264 // to some helper functions.
1265 assert(R19_method->is_nonvolatile(),
1266 "R19_method must be a non-volatile register");
1267
1268 // Check for synchronized methods. Must happen AFTER invocation counter
1269 // check, so method is not locked if counter overflows.
1270
1271 if (synchronized) {
1272 lock_method(access_flags, R11_scratch1, R12_scratch2, true);
1273
1274 // Update monitor in state.
1275 __ ld(R11_scratch1, 0, R1_SP);
1276 __ sub(R12_scratch2, R26_monitor, R11_scratch1);
1277 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1278 __ std(R12_scratch2, _ijava_state_neg(monitors), R11_scratch1);
1279 }
1280
1281 // jvmti/jvmpi support
1282 __ notify_method_entry();
1283
1284 //=============================================================================
1285 // Get and call the signature handler.
1286
1287 __ ld(signature_handler_fd, method_(signature_handler));
1288 Label call_signature_handler;
1289
1290 __ cmpdi(CCR0, signature_handler_fd, 0);
1291 __ bne(CCR0, call_signature_handler);
1292
1293 // Method has never been called. Either generate a specialized
1294 // handler or point to the slow one.
1295 //
1296 // Pass parameter 'false' to avoid exception check in call_VM.
1297 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
1298
1299 // Check for an exception while looking up the target method. If we
1300 // incurred one, bail.
1301 __ ld(pending_exception, thread_(pending_exception));
1302 __ cmpdi(CCR0, pending_exception, 0);
1303 __ bne(CCR0, exception_return_sync_check); // Has pending exception.
1304
1305 // Reload signature handler, it may have been created/assigned in the meanwhile.
1306 __ ld(signature_handler_fd, method_(signature_handler));
1307 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
1308
1309 BIND(call_signature_handler);
1310
1311 // Before we call the signature handler we push a new frame to
1312 // protect the interpreter frame volatile registers when we return
1313 // from jni but before we can get back to Java.
1314
1315 // First set the frame anchor while the SP/FP registers are
1316 // convenient and the slow signature handler can use this same frame
1317 // anchor.
1318
1319 // We have a TOP_IJAVA_FRAME here, which belongs to us.
1320 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
1321
1322 // Now the interpreter frame (and its call chain) have been
1323 // invalidated and flushed. We are now protected against eager
1324 // being enabled in native code. Even if it goes eager the
1325 // registers will be reloaded as clean and we will invalidate after
1326 // the call so no spurious flush should be possible.
1327
1328 // Call signature handler and pass locals address.
1329 //
1330 // Our signature handlers copy required arguments to the C stack
1331 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
1332 __ mr(R3_ARG1, R18_locals);
1333 #if !defined(ABI_ELFv2)
1334 __ ld(signature_handler_fd, 0, signature_handler_fd);
1335 #endif
1336
1337 __ call_stub(signature_handler_fd);
1338
1339 // Remove the register parameter varargs slots we allocated in
1340 // compute_interpreter_state. SP+16 ends up pointing to the ABI
1341 // outgoing argument area.
1342 //
1343 // Not needed on PPC64.
1344 //__ add(SP, SP, Argument::n_int_register_parameters_c*BytesPerWord);
1345
1346 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
1347 // Save across call to native method.
1348 __ mr(result_handler_addr, R3_RET);
1349
1350 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
1351
1352 // Set up fixed parameters and call the native method.
1353 // If the method is static, get mirror into R4_ARG2.
1354 {
1355 Label method_is_not_static;
1356 // Access_flags is non-volatile and still, no need to restore it.
1357
1358 // Restore access flags.
1359 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
1360 __ bfalse(CCR0, method_is_not_static);
1361
1362 __ ld(R11_scratch1, _abi0(callers_sp), R1_SP);
1363 // Load mirror from interpreter frame.
1364 __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1);
1365 // R4_ARG2 = &state->_oop_temp;
1366 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
1367 __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
1368 BIND(method_is_not_static);
1369 }
1370
1371 // At this point, arguments have been copied off the stack into
1372 // their JNI positions. Oops are boxed in-place on the stack, with
1373 // handles copied to arguments. The result handler address is in a
1374 // register.
1375
1376 // Pass JNIEnv address as first parameter.
1377 __ addir(R3_ARG1, thread_(jni_environment));
1378
1379 // Load the native_method entry before we change the thread state.
1380 __ ld(native_method_fd, method_(native_function));
1381
1382 //=============================================================================
1383 // Transition from _thread_in_Java to _thread_in_native. As soon as
1384 // we make this change the safepoint code needs to be certain that
1385 // the last Java frame we established is good. The pc in that frame
1386 // just needs to be near here not an actual return address.
1387
1388 // We use release_store_fence to update values like the thread state, where
1389 // we don't want the current thread to continue until all our prior memory
1390 // accesses (including the new thread state) are visible to other threads.
1391 __ li(R0, _thread_in_native);
1392 __ release();
1393
1394 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
1395 __ stw(R0, thread_(thread_state));
1396
1397 //=============================================================================
1398 // Call the native method. Argument registers must not have been
1399 // overwritten since "__ call_stub(signature_handler);" (except for
1400 // ARG1 and ARG2 for static methods).
1401 __ call_c(native_method_fd);
1402
1403 __ li(R0, 0);
1404 __ ld(R11_scratch1, 0, R1_SP);
1405 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1406 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1407 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
1408
1409 // Note: C++ interpreter needs the following here:
1410 // The frame_manager_lr field, which we use for setting the last
1411 // java frame, gets overwritten by the signature handler. Restore
1412 // it now.
1413 //__ get_PC_trash_LR(R11_scratch1);
1414 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
1415
1416 // Because of GC R19_method may no longer be valid.
1417
1418 // Block, if necessary, before resuming in _thread_in_Java state.
1419 // In order for GC to work, don't clear the last_Java_sp until after
1420 // blocking.
1421
1422 //=============================================================================
1423 // Switch thread to "native transition" state before reading the
1424 // synchronization state. This additional state is necessary
1425 // because reading and testing the synchronization state is not
1426 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1427 // in _thread_in_native state, loads _not_synchronized and is
1428 // preempted. VM thread changes sync state to synchronizing and
1429 // suspends threads for GC. Thread A is resumed to finish this
1430 // native method, but doesn't block here since it didn't see any
1431 // synchronization in progress, and escapes.
1432
1433 // We use release_store_fence to update values like the thread state, where
1434 // we don't want the current thread to continue until all our prior memory
1435 // accesses (including the new thread state) are visible to other threads.
1436 __ li(R0/*thread_state*/, _thread_in_native_trans);
1437 __ release();
1438 __ stw(R0/*thread_state*/, thread_(thread_state));
1439 if (!UseSystemMemoryBarrier) {
1440 __ fence();
1441 }
1442
1443 // Now before we return to java we must look for a current safepoint
1444 // (a new safepoint can not start since we entered native_trans).
1445 // We must check here because a current safepoint could be modifying
1446 // the callers registers right this moment.
1447
1448 // Acquire isn't strictly necessary here because of the fence, but
1449 // sync_state is declared to be volatile, so we do it anyway
1450 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1451
1452 Label do_safepoint, sync_check_done;
1453 // No synchronization in progress nor yet synchronized.
1454 __ safepoint_poll(do_safepoint, sync_state, true /* at_return */, false /* in_nmethod */);
1455
1456 // Not suspended.
1457 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1458 __ lwz(suspend_flags, thread_(suspend_flags));
1459 __ cmpwi(CCR1, suspend_flags, 0);
1460 __ beq(CCR1, sync_check_done);
1461
1462 __ bind(do_safepoint);
1463 __ isync();
1464 // Block. We do the call directly and leave the current
1465 // last_Java_frame setup undisturbed. We must save any possible
1466 // native result across the call. No oop is present.
1467
1468 __ mr(R3_ARG1, R16_thread);
1469 #if defined(ABI_ELFv2)
1470 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
1471 relocInfo::none);
1472 #else
1473 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
1474 relocInfo::none);
1475 #endif
1476
1477 __ bind(sync_check_done);
1478
1479 //=============================================================================
1480 // <<<<<< Back in Interpreter Frame >>>>>
1481
1482 // We are in thread_in_native_trans here and back in the normal
1483 // interpreter frame. We don't have to do anything special about
1484 // safepoints and we can switch to Java mode anytime we are ready.
1485
1486 // Note: frame::interpreter_frame_result has a dependency on how the
1487 // method result is saved across the call to post_method_exit. For
1488 // native methods it assumes that the non-FPU/non-void result is
1489 // saved in _native_lresult and a FPU result in _native_fresult. If
1490 // this changes then the interpreter_frame_result implementation
1491 // will need to be updated too.
1492
1493 // On PPC64, we have stored the result directly after the native call.
1494
1495 //=============================================================================
1496 // Back in Java
1497
1498 // We use release_store_fence to update values like the thread state, where
1499 // we don't want the current thread to continue until all our prior memory
1500 // accesses (including the new thread state) are visible to other threads.
1501 __ li(R0/*thread_state*/, _thread_in_Java);
1502 __ lwsync(); // Acquire safepoint and suspend state, release thread state.
1503 __ stw(R0/*thread_state*/, thread_(thread_state));
1504
1505 if (CheckJNICalls) {
1506 // clear_pending_jni_exception_check
1507 __ load_const_optimized(R0, 0L);
1508 __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread);
1509 }
1510
1511 __ reset_last_Java_frame();
1512
1513 // Jvmdi/jvmpi support. Whether we've got an exception pending or
1514 // not, and whether unlocking throws an exception or not, we notify
1515 // on native method exit. If we do have an exception, we'll end up
1516 // in the caller's context to handle it, so if we don't do the
1517 // notify here, we'll drop it on the floor.
1518 __ notify_method_exit(true/*native method*/,
1519 ilgl /*illegal state (not used for native methods)*/,
1520 InterpreterMacroAssembler::NotifyJVMTI,
1521 false /*check_exceptions*/);
1522
1523 //=============================================================================
1524 // Handle exceptions
1525
1526 if (synchronized) {
1527 __ unlock_object(R26_monitor); // Can also unlock methods.
1528 }
1529
1530 // Reset active handles after returning from native.
1531 // thread->active_handles()->clear();
1532 __ ld(active_handles, thread_(active_handles));
1533 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1534 __ li(R0, 0);
1535 __ stw(R0, in_bytes(JNIHandleBlock::top_offset()), active_handles);
1536
1537 Label exception_return_sync_check_already_unlocked;
1538 __ ld(R0/*pending_exception*/, thread_(pending_exception));
1539 __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1540 __ bne(CCR0, exception_return_sync_check_already_unlocked);
1541
1542 //-----------------------------------------------------------------------------
1543 // No exception pending.
1544
1545 // Move native method result back into proper registers and return.
1546 // Invoke result handler (may unbox/promote).
1547 __ ld(R11_scratch1, 0, R1_SP);
1548 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1549 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1550 __ call_stub(result_handler_addr);
1551
1552 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1553
1554 // Must use the return pc which was loaded from the caller's frame
1555 // as the VM uses return-pc-patching for deoptimization.
1556 __ mtlr(R0);
1557 __ blr();
1558
1559 //-----------------------------------------------------------------------------
1560 // An exception is pending. We call into the runtime only if the
1561 // caller was not interpreted. If it was interpreted the
1562 // interpreter will do the correct thing. If it isn't interpreted
1563 // (call stub/compiled code) we will change our return and continue.
1564
1565 BIND(exception_return_sync_check);
1566
1567 if (synchronized) {
1568 __ unlock_object(R26_monitor); // Can also unlock methods.
1569 }
1570 BIND(exception_return_sync_check_already_unlocked);
1571
1572 const Register return_pc = R31;
1573
1574 __ ld(return_pc, 0, R1_SP);
1575 __ ld(return_pc, _abi0(lr), return_pc);
1576
1577 // Get the address of the exception handler.
1578 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1579 R16_thread,
1580 return_pc /* return pc */);
1581 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1582
1583 // Load the PC of the exception handler into LR.
1584 __ mtlr(R3_RET);
1585
1586 // Load exception into R3_ARG1 and clear pending exception in thread.
1587 __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1588 __ li(R4_ARG2, 0);
1589 __ std(R4_ARG2, thread_(pending_exception));
1590
1591 // Load the original return pc into R4_ARG2.
1592 __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1593
1594 // Return to exception handler.
1595 __ blr();
1596
1597 //=============================================================================
1598 // Counter overflow.
1599
1600 if (inc_counter) {
1601 // Handle invocation counter overflow.
1602 __ bind(invocation_counter_overflow);
1603
1604 generate_counter_overflow(continue_after_compile);
1605 }
1606
1607 return entry;
1608 }
1609
1610 // Generic interpreted method entry to (asm) interpreter.
1611 //
1612 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1613 bool inc_counter = UseCompiler || CountCompiledCalls;
1614 address entry = __ pc();
1615 // Generate the code to allocate the interpreter stack frame.
1616 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1617 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame.
1618
1619 // Does also a stack check to assure this frame fits on the stack.
1620 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1621
1622 // --------------------------------------------------------------------------
1623 // Zero out non-parameter locals.
1624 // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1625 // worth to ask the flag, just do it.
1626 Register Rslot_addr = R6_ARG4,
1627 Rnum = R7_ARG5;
1628 Label Lno_locals, Lzero_loop;
1629
1630 // Set up the zeroing loop.
1631 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1632 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1633 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1634 __ beq(CCR0, Lno_locals);
1635 __ li(R0, 0);
1636 __ mtctr(Rnum);
1637
1638 // The zero locals loop.
1639 __ bind(Lzero_loop);
1640 __ std(R0, 0, Rslot_addr);
1641 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1642 __ bdnz(Lzero_loop);
1643
1644 __ bind(Lno_locals);
1645
1646 // --------------------------------------------------------------------------
1647 // Counter increment and overflow check.
1648 Label invocation_counter_overflow;
1649 Label continue_after_compile;
1650 if (inc_counter || ProfileInterpreter) {
1651
1652 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1653 if (synchronized) {
1654 // Since at this point in the method invocation the exception handler
1655 // would try to exit the monitor of synchronized methods which hasn't
1656 // been entered yet, we set the thread local variable
1657 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1658 // runtime, exception handling i.e. unlock_if_synchronized_method will
1659 // check this thread local flag.
1660 // This flag has two effects, one is to force an unwind in the topmost
1661 // interpreter frame and not perform an unlock while doing so.
1662 __ li(R0, 1);
1663 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1664 }
1665
1666 // Argument and return type profiling.
1667 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1668
1669 // Increment invocation counter and check for overflow.
1670 if (inc_counter) {
1671 generate_counter_incr(&invocation_counter_overflow);
1672 }
1673
1674 __ bind(continue_after_compile);
1675 }
1676
1677 bang_stack_shadow_pages(false);
1678
1679 if (inc_counter || ProfileInterpreter) {
1680 // Reset the _do_not_unlock_if_synchronized flag.
1681 if (synchronized) {
1682 __ li(R0, 0);
1683 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1684 }
1685 }
1686
1687 // --------------------------------------------------------------------------
1688 // Locking of synchronized methods. Must happen AFTER invocation_counter
1689 // check and stack overflow check, so method is not locked if overflows.
1690 if (synchronized) {
1691 lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1692 }
1693 #ifdef ASSERT
1694 else {
1695 Label Lok;
1696 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1697 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1698 __ asm_assert_eq("method needs synchronization");
1699 __ bind(Lok);
1700 }
1701 #endif // ASSERT
1702
1703 // --------------------------------------------------------------------------
1704 // JVMTI support
1705 __ notify_method_entry();
1706
1707 // --------------------------------------------------------------------------
1708 // Start executing instructions.
1709 __ dispatch_next(vtos);
1710
1711 // --------------------------------------------------------------------------
1712 if (inc_counter) {
1713 // Handle invocation counter overflow.
1714 __ bind(invocation_counter_overflow);
1715 generate_counter_overflow(continue_after_compile);
1716 }
1717 return entry;
1718 }
1719
1720 // CRC32 Intrinsics.
1721 //
1722 // Contract on scratch and work registers.
1723 // =======================================
1724 //
1725 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1726 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1727 // You can't rely on these registers across calls.
1728 //
1729 // The generators for CRC32_update and for CRC32_updateBytes use the
1730 // scratch/work register set internally, passing the work registers
1731 // as arguments to the MacroAssembler emitters as required.
1732 //
1733 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1734 // Their contents is not constant but may change according to the requirements
1735 // of the emitted code.
1736 //
1737 // All other registers from the scratch/work register set are used "internally"
1738 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1739 // Basically, only R3_RET contains a defined value which is the function result.
1740 //
1741 /**
1742 * Method entry for static native methods:
1743 * int java.util.zip.CRC32.update(int crc, int b)
1744 */
1745 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1746 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1747 address start = __ pc(); // Remember stub start address (is rtn value).
1748 Label slow_path;
1749
1750 // Safepoint check
1751 const Register sync_state = R11_scratch1;
1752 __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1753
1754 // We don't generate local frame and don't align stack because
1755 // we not even call stub code (we generate the code inline)
1756 // and there is no safepoint on this path.
1757
1758 // Load java parameters.
1759 // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1760 const Register argP = R15_esp;
1761 const Register crc = R3_ARG1; // crc value
1762 const Register data = R4_ARG2;
1763 const Register table = R5_ARG3; // address of crc32 table
1764
1765 BLOCK_COMMENT("CRC32_update {");
1766
1767 // Arguments are reversed on java expression stack
1768 #ifdef VM_LITTLE_ENDIAN
1769 int data_offs = 0+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1770 // Being passed as an int, the single byte is at offset +0.
1771 #else
1772 int data_offs = 3+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1773 // Being passed from java as an int, the single byte is at offset +3.
1774 #endif
1775 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1776 __ lbz(data, data_offs, argP); // Byte from buffer, zero-extended.
1777 __ load_const_optimized(table, StubRoutines::crc_table_addr(), R0);
1778 __ kernel_crc32_singleByteReg(crc, data, table, true);
1779
1780 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1781 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1782 __ blr();
1783
1784 // Generate a vanilla native entry as the slow path.
1785 BLOCK_COMMENT("} CRC32_update");
1786 BIND(slow_path);
1787 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1788 return start;
1789 }
1790
1791 /**
1792 * Method entry for static native methods:
1793 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len)
1794 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1795 */
1796 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1797 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1798 address start = __ pc(); // Remember stub start address (is rtn value).
1799 Label slow_path;
1800
1801 // Safepoint check
1802 const Register sync_state = R11_scratch1;
1803 __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1804
1805 // We don't generate local frame and don't align stack because
1806 // we not even call stub code (we generate the code inline)
1807 // and there is no safepoint on this path.
1808
1809 // Load parameters.
1810 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1811 const Register argP = R15_esp;
1812 const Register crc = R3_ARG1; // crc value
1813 const Register data = R4_ARG2; // address of java byte array
1814 const Register dataLen = R5_ARG3; // source data len
1815 const Register tmp = R11_scratch1;
1816
1817 // Arguments are reversed on java expression stack.
1818 // Calculate address of start element.
1819 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1820 BLOCK_COMMENT("CRC32_updateByteBuffer {");
1821 // crc @ (SP + 5W) (32bit)
1822 // buf @ (SP + 3W) (64bit ptr to long array)
1823 // off @ (SP + 2W) (32bit)
1824 // dataLen @ (SP + 1W) (32bit)
1825 // data = buf + off
1826 __ ld( data, 3*wordSize, argP); // start of byte buffer
1827 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1828 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1829 __ lwz( crc, 5*wordSize, argP); // current crc state
1830 __ add( data, data, tmp); // Add byte buffer offset.
1831 } else { // Used for "updateBytes update".
1832 BLOCK_COMMENT("CRC32_updateBytes {");
1833 // crc @ (SP + 4W) (32bit)
1834 // buf @ (SP + 3W) (64bit ptr to byte array)
1835 // off @ (SP + 2W) (32bit)
1836 // dataLen @ (SP + 1W) (32bit)
1837 // data = buf + off + base_offset
1838 __ ld( data, 3*wordSize, argP); // start of byte buffer
1839 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1840 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1841 __ add( data, data, tmp); // add byte buffer offset
1842 __ lwz( crc, 4*wordSize, argP); // current crc state
1843 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1844 }
1845
1846 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, false);
1847
1848 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1849 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1850 __ blr();
1851
1852 // Generate a vanilla native entry as the slow path.
1853 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1854 BIND(slow_path);
1855 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1856 return start;
1857 }
1858
1859
1860 /**
1861 * Method entry for intrinsic-candidate (non-native) methods:
1862 * int java.util.zip.CRC32C.updateBytes( int crc, byte[] b, int off, int end)
1863 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
1864 * Unlike CRC32, CRC32C does not have any methods marked as native
1865 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1866 **/
1867 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1868 assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1869 address start = __ pc(); // Remember stub start address (is rtn value).
1870
1871 // We don't generate local frame and don't align stack because
1872 // we not even call stub code (we generate the code inline)
1873 // and there is no safepoint on this path.
1874
1875 // Load parameters.
1876 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1877 const Register argP = R15_esp;
1878 const Register crc = R3_ARG1; // crc value
1879 const Register data = R4_ARG2; // address of java byte array
1880 const Register dataLen = R5_ARG3; // source data len
1881 const Register tmp = R11_scratch1;
1882
1883 // Arguments are reversed on java expression stack.
1884 // Calculate address of start element.
1885 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer".
1886 BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
1887 // crc @ (SP + 5W) (32bit)
1888 // buf @ (SP + 3W) (64bit ptr to long array)
1889 // off @ (SP + 2W) (32bit)
1890 // dataLen @ (SP + 1W) (32bit)
1891 // data = buf + off
1892 __ ld( data, 3*wordSize, argP); // start of byte buffer
1893 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1894 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1895 __ lwz( crc, 5*wordSize, argP); // current crc state
1896 __ add( data, data, tmp); // Add byte buffer offset.
1897 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1898 } else { // Used for "updateBytes update".
1899 BLOCK_COMMENT("CRC32C_updateBytes {");
1900 // crc @ (SP + 4W) (32bit)
1901 // buf @ (SP + 3W) (64bit ptr to byte array)
1902 // off @ (SP + 2W) (32bit)
1903 // dataLen @ (SP + 1W) (32bit)
1904 // data = buf + off + base_offset
1905 __ ld( data, 3*wordSize, argP); // start of byte buffer
1906 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1907 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1908 __ add( data, data, tmp); // add byte buffer offset
1909 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1910 __ lwz( crc, 4*wordSize, argP); // current crc state
1911 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1912 }
1913
1914 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, true);
1915
1916 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1917 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1918 __ blr();
1919
1920 BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
1921 return start;
1922 }
1923
1924 // Not supported
1925 address TemplateInterpreterGenerator::generate_currentThread() { return nullptr; }
1926 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
1927 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
1928 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
1929 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
1930 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() { return nullptr; }
1931 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() { return nullptr; }
1932
1933 // =============================================================================
1934 // Exceptions
1935
1936 void TemplateInterpreterGenerator::generate_throw_exception() {
1937 Register Rexception = R17_tos,
1938 Rcontinuation = R3_RET;
1939
1940 // --------------------------------------------------------------------------
1941 // Entry point if an method returns with a pending exception (rethrow).
1942 Interpreter::_rethrow_exception_entry = __ pc();
1943 {
1944 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
1945
1946 // Compiled code destroys templateTableBase, reload.
1947 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
1948 }
1949
1950 // Entry point if a interpreted method throws an exception (throw).
1951 Interpreter::_throw_exception_entry = __ pc();
1952 {
1953 __ mr(Rexception, R3_RET);
1954
1955 __ verify_oop(Rexception);
1956
1957 // Expression stack must be empty before entering the VM in case of an exception.
1958 __ empty_expression_stack();
1959 // Find exception handler address and preserve exception oop.
1960 // Call C routine to find handler and jump to it.
1961 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
1962 __ mtctr(Rcontinuation);
1963 // Push exception for exception handler bytecodes.
1964 __ push_ptr(Rexception);
1965
1966 // Jump to exception handler (may be remove activation entry!).
1967 __ bctr();
1968 }
1969
1970 // If the exception is not handled in the current frame the frame is
1971 // removed and the exception is rethrown (i.e. exception
1972 // continuation is _rethrow_exception).
1973 //
1974 // Note: At this point the bci is still the bxi for the instruction
1975 // which caused the exception and the expression stack is
1976 // empty. Thus, for any VM calls at this point, GC will find a legal
1977 // oop map (with empty expression stack).
1978
1979 // In current activation
1980 // tos: exception
1981 // bcp: exception bcp
1982
1983 // --------------------------------------------------------------------------
1984 // JVMTI PopFrame support
1985
1986 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1987 {
1988 // Set the popframe_processing bit in popframe_condition indicating that we are
1989 // currently handling popframe, so that call_VMs that may happen later do not
1990 // trigger new popframe handling cycles.
1991 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1992 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
1993 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1994
1995 // Empty the expression stack, as in normal exception handling.
1996 __ empty_expression_stack();
1997 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
1998
1999 // Check to see whether we are returning to a deoptimized frame.
2000 // (The PopFrame call ensures that the caller of the popped frame is
2001 // either interpreted or compiled and deoptimizes it if compiled.)
2002 // Note that we don't compare the return PC against the
2003 // deoptimization blob's unpack entry because of the presence of
2004 // adapter frames in C2.
2005 Label Lcaller_not_deoptimized;
2006 Register return_pc = R3_ARG1;
2007 __ ld(return_pc, 0, R1_SP);
2008 __ ld(return_pc, _abi0(lr), return_pc);
2009 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
2010 __ cmpdi(CCR0, R3_RET, 0);
2011 __ bne(CCR0, Lcaller_not_deoptimized);
2012
2013 // The deoptimized case.
2014 // In this case, we can't call dispatch_next() after the frame is
2015 // popped, but instead must save the incoming arguments and restore
2016 // them after deoptimization has occurred.
2017 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
2018 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
2019 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
2020 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
2021 __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
2022 // Save these arguments.
2023 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
2024
2025 // Inform deoptimization that it is responsible for restoring these arguments.
2026 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
2027 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2028
2029 // Return from the current method into the deoptimization blob. Will eventually
2030 // end up in the deopt interpreter entry, deoptimization prepared everything that
2031 // we will reexecute the call that called us.
2032 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
2033 __ mtlr(return_pc);
2034 __ pop_cont_fastpath();
2035 __ blr();
2036
2037 // The non-deoptimized case.
2038 __ bind(Lcaller_not_deoptimized);
2039
2040 // Clear the popframe condition flag.
2041 __ li(R0, 0);
2042 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2043
2044 // Get out of the current method and re-execute the call that called us.
2045 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2046 __ pop_cont_fastpath();
2047 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
2048 if (ProfileInterpreter) {
2049 __ set_method_data_pointer_for_bcp();
2050 __ ld(R11_scratch1, 0, R1_SP);
2051 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
2052 }
2053 #if INCLUDE_JVMTI
2054 Label L_done;
2055
2056 __ lbz(R11_scratch1, 0, R14_bcp);
2057 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
2058 __ bne(CCR0, L_done);
2059
2060 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2061 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
2062 __ ld(R4_ARG2, 0, R18_locals);
2063 __ call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp);
2064
2065 __ cmpdi(CCR0, R4_ARG2, 0);
2066 __ beq(CCR0, L_done);
2067 __ std(R4_ARG2, wordSize, R15_esp);
2068 __ bind(L_done);
2069 #endif // INCLUDE_JVMTI
2070 __ dispatch_next(vtos);
2071 }
2072 // end of JVMTI PopFrame support
2073
2074 // --------------------------------------------------------------------------
2075 // Remove activation exception entry.
2076 // This is jumped to if an interpreted method can't handle an exception itself
2077 // (we come from the throw/rethrow exception entry above). We're going to call
2078 // into the VM to find the exception handler in the caller, pop the current
2079 // frame and return the handler we calculated.
2080 Interpreter::_remove_activation_entry = __ pc();
2081 {
2082 __ pop_ptr(Rexception);
2083 __ verify_oop(Rexception);
2084 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
2085
2086 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
2087 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
2088
2089 __ get_vm_result(Rexception);
2090
2091 // We are done with this activation frame; find out where to go next.
2092 // The continuation point will be an exception handler, which expects
2093 // the following registers set up:
2094 //
2095 // RET: exception oop
2096 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
2097
2098 Register return_pc = R31; // Needs to survive the runtime call.
2099 __ ld(return_pc, 0, R1_SP);
2100 __ ld(return_pc, _abi0(lr), return_pc);
2101 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
2102
2103 // Remove the current activation.
2104 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2105 __ pop_cont_fastpath();
2106
2107 __ mr(R4_ARG2, return_pc);
2108 __ mtlr(R3_RET);
2109 __ mr(R3_RET, Rexception);
2110 __ blr();
2111 }
2112 }
2113
2114 // JVMTI ForceEarlyReturn support.
2115 // Returns "in the middle" of a method with a "fake" return value.
2116 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2117
2118 Register Rscratch1 = R11_scratch1,
2119 Rscratch2 = R12_scratch2;
2120
2121 address entry = __ pc();
2122 __ empty_expression_stack();
2123
2124 __ load_earlyret_value(state, Rscratch1);
2125
2126 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
2127 // Clear the earlyret state.
2128 __ li(R0, 0);
2129 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
2130
2131 __ remove_activation(state, false, false);
2132 // Copied from TemplateTable::_return.
2133 // Restoration of lr done by remove_activation.
2134 switch (state) {
2135 // Narrow result if state is itos but result type is smaller.
2136 case btos:
2137 case ztos:
2138 case ctos:
2139 case stos:
2140 case itos: __ narrow(R17_tos); /* fall through */
2141 case ltos:
2142 case atos: __ mr(R3_RET, R17_tos); break;
2143 case ftos:
2144 case dtos: __ fmr(F1_RET, F15_ftos); break;
2145 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2146 // to get visible before the reference to the object gets stored anywhere.
2147 __ membar(Assembler::StoreStore); break;
2148 default : ShouldNotReachHere();
2149 }
2150 __ blr();
2151
2152 return entry;
2153 } // end of ForceEarlyReturn support
2154
2155 //-----------------------------------------------------------------------------
2156 // Helper for vtos entry point generation
2157
2158 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2159 address& bep,
2160 address& cep,
2161 address& sep,
2162 address& aep,
2163 address& iep,
2164 address& lep,
2165 address& fep,
2166 address& dep,
2167 address& vep) {
2168 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2169 Label L;
2170
2171 aep = __ pc(); __ push_ptr(); __ b(L);
2172 fep = __ pc(); __ push_f(); __ b(L);
2173 dep = __ pc(); __ push_d(); __ b(L);
2174 lep = __ pc(); __ push_l(); __ b(L);
2175 __ align(32, 12, 24); // align L
2176 bep = cep = sep =
2177 iep = __ pc(); __ push_i();
2178 vep = __ pc();
2179 __ bind(L);
2180 generate_and_dispatch(t);
2181 }
2182
2183 //-----------------------------------------------------------------------------
2184
2185 // Non-product code
2186 #ifndef PRODUCT
2187 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2188 //__ flush_bundle();
2189 address entry = __ pc();
2190
2191 const char *bname = nullptr;
2192 uint tsize = 0;
2193 switch(state) {
2194 case ftos:
2195 bname = "trace_code_ftos {";
2196 tsize = 2;
2197 break;
2198 case btos:
2199 bname = "trace_code_btos {";
2200 tsize = 2;
2201 break;
2202 case ztos:
2203 bname = "trace_code_ztos {";
2204 tsize = 2;
2205 break;
2206 case ctos:
2207 bname = "trace_code_ctos {";
2208 tsize = 2;
2209 break;
2210 case stos:
2211 bname = "trace_code_stos {";
2212 tsize = 2;
2213 break;
2214 case itos:
2215 bname = "trace_code_itos {";
2216 tsize = 2;
2217 break;
2218 case ltos:
2219 bname = "trace_code_ltos {";
2220 tsize = 3;
2221 break;
2222 case atos:
2223 bname = "trace_code_atos {";
2224 tsize = 2;
2225 break;
2226 case vtos:
2227 // Note: In case of vtos, the topmost of stack value could be a int or doubl
2228 // In case of a double (2 slots) we won't see the 2nd stack value.
2229 // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
2230 bname = "trace_code_vtos {";
2231 tsize = 2;
2232
2233 break;
2234 case dtos:
2235 bname = "trace_code_dtos {";
2236 tsize = 3;
2237 break;
2238 default:
2239 ShouldNotReachHere();
2240 }
2241 BLOCK_COMMENT(bname);
2242
2243 // Support short-cut for TraceBytecodesAt.
2244 // Don't call into the VM if we don't want to trace to speed up things.
2245 Label Lskip_vm_call;
2246 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2247 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
2248 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2249 __ ld(R11_scratch1, offs1, R11_scratch1);
2250 __ lwa(R12_scratch2, offs2, R12_scratch2);
2251 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2252 __ blt(CCR0, Lskip_vm_call);
2253 }
2254
2255 __ push(state);
2256 // Load 2 topmost expression stack values.
2257 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
2258 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
2259 __ mflr(R31);
2260 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
2261 __ mtlr(R31);
2262 __ pop(state);
2263
2264 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2265 __ bind(Lskip_vm_call);
2266 }
2267 __ blr();
2268 BLOCK_COMMENT("} trace_code");
2269 return entry;
2270 }
2271
2272 void TemplateInterpreterGenerator::count_bytecode() {
2273 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
2274 __ lwz(R12_scratch2, offs, R11_scratch1);
2275 __ addi(R12_scratch2, R12_scratch2, 1);
2276 __ stw(R12_scratch2, offs, R11_scratch1);
2277 }
2278
2279 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2280 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
2281 __ lwz(R12_scratch2, offs, R11_scratch1);
2282 __ addi(R12_scratch2, R12_scratch2, 1);
2283 __ stw(R12_scratch2, offs, R11_scratch1);
2284 }
2285
2286 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2287 const Register addr = R11_scratch1,
2288 tmp = R12_scratch2;
2289 // Get index, shift out old bytecode, bring in new bytecode, and store it.
2290 // _index = (_index >> log2_number_of_codes) |
2291 // (bytecode << log2_number_of_codes);
2292 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
2293 __ lwz(tmp, offs1, addr);
2294 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
2295 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2296 __ stw(tmp, offs1, addr);
2297
2298 // Bump bucket contents.
2299 // _counters[_index] ++;
2300 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
2301 __ sldi(tmp, tmp, LogBytesPerInt);
2302 __ add(addr, tmp, addr);
2303 __ lwz(tmp, offs2, addr);
2304 __ addi(tmp, tmp, 1);
2305 __ stw(tmp, offs2, addr);
2306 }
2307
2308 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2309 // Call a little run-time stub to avoid blow-up for each bytecode.
2310 // The run-time runtime saves the right registers, depending on
2311 // the tosca in-state for the given template.
2312
2313 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2314 "entry must have been generated");
2315
2316 // Note: we destroy LR here.
2317 __ bl(Interpreter::trace_code(t->tos_in()));
2318 }
2319
2320 void TemplateInterpreterGenerator::stop_interpreter_at() {
2321 Label L;
2322 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
2323 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2324 __ ld(R11_scratch1, offs1, R11_scratch1);
2325 __ lwa(R12_scratch2, offs2, R12_scratch2);
2326 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2327 __ bne(CCR0, L);
2328 __ illtrap();
2329 __ bind(L);
2330 }
2331
2332 #endif // !PRODUCT