1 /*
2 * Copyright (c) 2014, 2024, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2024 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(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(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 __ normalize_bool(R3_RET);
376 break;
377 case T_BYTE:
378 // sign extend 8 bits
379 __ extsb(R3_RET, R3_RET);
380 break;
381 case T_CHAR:
382 // zero extend 16 bits
383 __ clrldi(R3_RET, R3_RET, 48);
384 break;
385 case T_SHORT:
386 // sign extend 16 bits
387 __ extsh(R3_RET, R3_RET);
388 break;
389 case T_INT:
390 // sign extend 32 bits
391 __ extsw(R3_RET, R3_RET);
392 break;
393 case T_LONG:
394 break;
395 case T_OBJECT:
396 // JNIHandles::resolve result.
397 __ resolve_jobject(R3_RET, R11_scratch1, R31, MacroAssembler::PRESERVATION_FRAME_LR); // kills R31
398 break;
399 case T_FLOAT:
400 break;
401 case T_DOUBLE:
402 break;
403 case T_VOID:
404 break;
405 default: ShouldNotReachHere();
406 }
407
408 BIND(done);
409 __ blr();
410
411 return entry;
412 }
413
414 // Abstract method entry.
415 //
416 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
417 address entry = __ pc();
418
419 //
420 // Registers alive
421 // R16_thread - JavaThread*
422 // R19_method - callee's method (method to be invoked)
423 // R1_SP - SP prepared such that caller's outgoing args are near top
424 // LR - return address to caller
425 //
426 // Stack layout at this point:
427 //
428 // 0 [TOP_IJAVA_FRAME_ABI] <-- R1_SP
429 // alignment (optional)
430 // [outgoing Java arguments]
431 // ...
432 // PARENT [PARENT_IJAVA_FRAME_ABI]
433 // ...
434 //
435
436 // Can't use call_VM here because we have not set up a new
437 // interpreter state. Make the call to the vm and make it look like
438 // our caller set up the JavaFrameAnchor.
439 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
440
441 // Push a new C frame and save LR.
442 __ save_LR(R0);
443 __ push_frame_reg_args(0, R11_scratch1);
444
445 // This is not a leaf but we have a JavaFrameAnchor now and we will
446 // check (create) exceptions afterward so this is ok.
447 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
448 R16_thread, R19_method);
449
450 // Pop the C frame and restore LR.
451 __ pop_frame();
452 __ restore_LR(R0);
453
454 // Reset JavaFrameAnchor from call_VM_leaf above.
455 __ reset_last_Java_frame();
456
457 // We don't know our caller, so jump to the general forward exception stub,
458 // which will also pop our full frame off. Satisfy the interface of
459 // SharedRuntime::generate_forward_exception()
460 __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0);
461 __ mtctr(R11_scratch1);
462 __ bctr();
463
464 return entry;
465 }
466
467 // Interpreter intrinsic for WeakReference.get().
468 // 1. Don't push a full blown frame and go on dispatching, but fetch the value
469 // into R8 and return quickly
470 // 2. If G1 is active we *must* execute this intrinsic for corrrectness:
471 // It contains a GC barrier which puts the reference into the satb buffer
472 // to indicate that someone holds a strong reference to the object the
473 // weak ref points to!
474 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
475 // Code: _aload_0, _getfield, _areturn
476 // parameter size = 1
477 //
478 // The code that gets generated by this routine is split into 2 parts:
479 // 1. the "intrinsified" code for G1 (or any SATB based GC),
480 // 2. the slow path - which is an expansion of the regular method entry.
481 //
482 // Notes:
483 // * In the G1 code we do not check whether we need to block for
484 // a safepoint. If G1 is enabled then we must execute the specialized
485 // code for Reference.get (except when the Reference object is null)
486 // so that we can log the value in the referent field with an SATB
487 // update buffer.
488 // If the code for the getfield template is modified so that the
489 // G1 pre-barrier code is executed when the current method is
490 // Reference.get() then going through the normal method entry
491 // will be fine.
492 // * The G1 code can, however, check the receiver object (the instance
493 // of java.lang.Reference) and jump to the slow path if null. If the
494 // Reference object is null then we obviously cannot fetch the referent
495 // and so we don't need to call the G1 pre-barrier. Thus we can use the
496 // regular method entry code to generate the NPE.
497 //
498
499 address entry = __ pc();
500
501 const int referent_offset = java_lang_ref_Reference::referent_offset();
502
503 Label slow_path;
504
505 // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH);
506
507 // In the G1 code we don't check if we need to reach a safepoint. We
508 // continue and the thread will safepoint at the next bytecode dispatch.
509
510 // If the receiver is null then it is OK to jump to the slow path.
511 __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver
512
513 // Check if receiver == nullptr and go the slow path.
514 __ cmpdi(CCR0, R3_RET, 0);
515 __ beq(CCR0, slow_path);
516
517 __ load_heap_oop(R3_RET, referent_offset, R3_RET,
518 /* non-volatile temp */ R31, R11_scratch1,
519 MacroAssembler::PRESERVATION_FRAME_LR,
520 ON_WEAK_OOP_REF);
521
522 // Generate the G1 pre-barrier code to log the value of
523 // the referent field in an SATB buffer. Note with
524 // these parameters the pre-barrier does not generate
525 // the load of the previous value.
526
527 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
528 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
529
530 __ blr();
531
532 __ bind(slow_path);
533 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
534 return entry;
535 }
536
537 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
538 address entry = __ pc();
539
540 // Expression stack must be empty before entering the VM if an
541 // exception happened.
542 __ empty_expression_stack();
543 // Throw exception.
544 __ call_VM(noreg,
545 CAST_FROM_FN_PTR(address,
546 InterpreterRuntime::throw_StackOverflowError));
547 return entry;
548 }
549
550 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
551 address entry = __ pc();
552 __ empty_expression_stack();
553 // R4_ARG2 already contains the array.
554 // Index is in R17_tos.
555 __ mr(R5_ARG3, R17_tos);
556 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R4_ARG2, R5_ARG3);
557 return entry;
558 }
559
560 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
561 address entry = __ pc();
562 // Expression stack must be empty before entering the VM if an
563 // exception happened.
564 __ empty_expression_stack();
565
566 // Load exception object.
567 // Thread will be loaded to R3_ARG1.
568 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
569 #ifdef ASSERT
570 // Above call must not return here since exception pending.
571 __ should_not_reach_here();
572 #endif
573 return entry;
574 }
575
576 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
577 address entry = __ pc();
578 //__ untested("generate_exception_handler_common");
579 Register Rexception = R17_tos;
580
581 // Expression stack must be empty before entering the VM if an exception happened.
582 __ empty_expression_stack();
583
584 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
585 if (pass_oop) {
586 __ mr(R5_ARG3, Rexception);
587 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception));
588 } else {
589 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
590 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception));
591 }
592
593 // Throw exception.
594 __ mr(R3_ARG1, Rexception);
595 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
596 __ mtctr(R11_scratch1);
597 __ bctr();
598
599 return entry;
600 }
601
602 // This entry is returned to when a call returns to the interpreter.
603 // When we arrive here, we expect that the callee stack frame is already popped.
604 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
605 address entry = __ pc();
606
607 // Move the value out of the return register back to the TOS cache of current frame.
608 switch (state) {
609 case ltos:
610 case btos:
611 case ztos:
612 case ctos:
613 case stos:
614 case atos:
615 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache
616 case ftos:
617 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
618 case vtos: break; // Nothing to do, this was a void return.
619 default : ShouldNotReachHere();
620 }
621
622 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
623
624 // Compiled code destroys templateTableBase, reload.
625 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
626
627 if (state == atos) {
628 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
629 }
630
631 const Register cache = R11_scratch1;
632 const Register size = R12_scratch2;
633 if (index_size == sizeof(u4)) {
634 __ load_resolved_indy_entry(cache, size /* tmp */);
635 __ lhz(size, in_bytes(ResolvedIndyEntry::num_parameters_offset()), cache);
636 } else {
637 assert(index_size == sizeof(u2), "Can only be u2");
638 __ load_method_entry(cache, size /* tmp */);
639 __ lhz(size, in_bytes(ResolvedMethodEntry::num_parameters_offset()), cache);
640 }
641 __ sldi(size, size, Interpreter::logStackElementSize);
642 __ add(R15_esp, R15_esp, size);
643
644 __ check_and_handle_popframe(R11_scratch1);
645 __ check_and_handle_earlyret(R11_scratch1);
646
647 __ dispatch_next(state, step);
648 return entry;
649 }
650
651 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
652 address entry = __ pc();
653 // If state != vtos, we're returning from a native method, which put it's result
654 // into the result register. So move the value out of the return register back
655 // to the TOS cache of current frame.
656
657 switch (state) {
658 case ltos:
659 case btos:
660 case ztos:
661 case ctos:
662 case stos:
663 case atos:
664 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache
665 case ftos:
666 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
667 case vtos: break; // Nothing to do, this was a void return.
668 default : ShouldNotReachHere();
669 }
670
671 // Load LcpoolCache @@@ should be already set!
672 __ get_constant_pool_cache(R27_constPoolCache);
673
674 // Handle a pending exception, fall through if none.
675 __ check_and_forward_exception(R11_scratch1, R12_scratch2);
676
677 // Start executing bytecodes.
678 if (continuation == nullptr) {
679 __ dispatch_next(state, step);
680 } else {
681 __ jump_to_entry(continuation, R11_scratch1);
682 }
683
684 return entry;
685 }
686
687 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
688 address entry = __ pc();
689
690 __ push(state);
691 __ push_cont_fastpath();
692 __ call_VM(noreg, runtime_entry);
693 __ pop_cont_fastpath();
694 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
695
696 return entry;
697 }
698
699 address TemplateInterpreterGenerator::generate_cont_resume_interpreter_adapter() {
700 if (!Continuations::enabled()) return nullptr;
701 address start = __ pc();
702
703 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
704 __ restore_interpreter_state(R11_scratch1, false, true /*restore_top_frame_sp*/);
705 __ blr();
706
707 return start;
708 }
709
710 // Helpers for commoning out cases in the various type of method entries.
711
712 // Increment invocation count & check for overflow.
713 //
714 // Note: checking for negative value instead of overflow
715 // so we have a 'sticky' overflow test.
716 //
717 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
718 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
719 Register Rscratch1 = R11_scratch1;
720 Register Rscratch2 = R12_scratch2;
721 Register R3_counters = R3_ARG1;
722 Label done;
723
724 const int increment = InvocationCounter::count_increment;
725 Label no_mdo;
726 if (ProfileInterpreter) {
727 const Register Rmdo = R3_counters;
728 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
729 __ cmpdi(CCR0, Rmdo, 0);
730 __ beq(CCR0, no_mdo);
731
732 // Increment invocation counter in the MDO.
733 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
734 __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
735 __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo);
736 __ addi(Rscratch2, Rscratch2, increment);
737 __ stw(Rscratch2, mdo_ic_offs, Rmdo);
738 __ and_(Rscratch1, Rscratch2, Rscratch1);
739 __ bne(CCR0, done);
740 __ b(*overflow);
741 }
742
743 // Increment counter in MethodCounters*.
744 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
745 __ bind(no_mdo);
746 __ get_method_counters(R19_method, R3_counters, done);
747 __ lwz(Rscratch2, mo_ic_offs, R3_counters);
748 __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters);
749 __ addi(Rscratch2, Rscratch2, increment);
750 __ stw(Rscratch2, mo_ic_offs, R3_counters);
751 __ and_(Rscratch1, Rscratch2, Rscratch1);
752 __ beq(CCR0, *overflow);
753
754 __ bind(done);
755 }
756
757 // Generate code to initiate compilation on invocation counter overflow.
758 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
759 // Generate code to initiate compilation on the counter overflow.
760
761 // InterpreterRuntime::frequency_counter_overflow takes one arguments,
762 // which indicates if the counter overflow occurs at a backwards branch (null bcp)
763 // We pass zero in.
764 // The call returns the address of the verified entry point for the method or null
765 // if the compilation did not complete (either went background or bailed out).
766 //
767 // Unlike the C++ interpreter above: Check exceptions!
768 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
769 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
770
771 __ li(R4_ARG2, 0);
772 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
773
774 // Returns verified_entry_point or null.
775 // We ignore it in any case.
776 __ b(continue_entry);
777 }
778
779 // See if we've got enough room on the stack for locals plus overhead below
780 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
781 // without going through the signal handler, i.e., reserved and yellow zones
782 // will not be made usable. The shadow zone must suffice to handle the
783 // overflow.
784 //
785 // Kills Rmem_frame_size, Rscratch1.
786 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
787 Label done;
788 assert_different_registers(Rmem_frame_size, Rscratch1);
789
790 BLOCK_COMMENT("stack_overflow_check_with_compare {");
791 __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
792 __ ld(Rscratch1, thread_(stack_overflow_limit));
793 __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
794 __ bgt(CCR0/*is_stack_overflow*/, done);
795
796 // The stack overflows. Load target address of the runtime stub and call it.
797 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "generated in wrong order");
798 __ load_const_optimized(Rscratch1, (SharedRuntime::throw_StackOverflowError_entry()), R0);
799 __ mtctr(Rscratch1);
800 // Restore caller_sp (c2i adapter may exist, but no shrinking of interpreted caller frame).
801 #ifdef ASSERT
802 Label frame_not_shrunk;
803 __ cmpld(CCR0, R1_SP, R21_sender_SP);
804 __ ble(CCR0, frame_not_shrunk);
805 __ stop("frame shrunk");
806 __ bind(frame_not_shrunk);
807 __ ld(Rscratch1, 0, R1_SP);
808 __ ld(R0, 0, R21_sender_SP);
809 __ cmpd(CCR0, R0, Rscratch1);
810 __ asm_assert_eq("backlink");
811 #endif // ASSERT
812 __ mr(R1_SP, R21_sender_SP);
813 __ bctr();
814
815 __ align(32, 12);
816 __ bind(done);
817 BLOCK_COMMENT("} stack_overflow_check_with_compare");
818 }
819
820 // Lock the current method, interpreter register window must be set up!
821 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
822 const Register Robj_to_lock = Rscratch2;
823
824 {
825 if (!flags_preloaded) {
826 __ lwz(Rflags, method_(access_flags));
827 }
828
829 #ifdef ASSERT
830 // Check if methods needs synchronization.
831 {
832 Label Lok;
833 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
834 __ btrue(CCR0,Lok);
835 __ stop("method doesn't need synchronization");
836 __ bind(Lok);
837 }
838 #endif // ASSERT
839 }
840
841 // Get synchronization object to Rscratch2.
842 {
843 Label Lstatic;
844 Label Ldone;
845
846 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
847 __ btrue(CCR0, Lstatic);
848
849 // Non-static case: load receiver obj from stack and we're done.
850 __ ld(Robj_to_lock, R18_locals);
851 __ b(Ldone);
852
853 __ bind(Lstatic); // Static case: Lock the java mirror
854 // Load mirror from interpreter frame.
855 __ ld(Robj_to_lock, _abi0(callers_sp), R1_SP);
856 __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock);
857
858 __ bind(Ldone);
859 __ verify_oop(Robj_to_lock);
860 }
861
862 // Got the oop to lock => execute!
863 __ add_monitor_to_stack(true, Rscratch1, R0);
864
865 __ std(Robj_to_lock, in_bytes(BasicObjectLock::obj_offset()), R26_monitor);
866 __ lock_object(R26_monitor, Robj_to_lock);
867 }
868
869 // Generate a fixed interpreter frame for pure interpreter
870 // and I2N native transition frames.
871 //
872 // Before (stack grows downwards):
873 //
874 // | ... |
875 // |------------- |
876 // | java arg0 |
877 // | ... |
878 // | java argn |
879 // | | <- R15_esp
880 // | |
881 // |--------------|
882 // | abi_112 |
883 // | | <- R1_SP
884 // |==============|
885 //
886 //
887 // After:
888 //
889 // | ... |
890 // | java arg0 |<- R18_locals
891 // | ... |
892 // | java argn |
893 // |--------------|
894 // | |
895 // | java locals |
896 // | |
897 // |--------------|
898 // | abi_48 |
899 // |==============|
900 // | |
901 // | istate |
902 // | |
903 // |--------------|
904 // | monitor |<- R26_monitor
905 // |--------------|
906 // | |<- R15_esp
907 // | expression |
908 // | stack |
909 // | |
910 // |--------------|
911 // | |
912 // | abi_112 |<- R1_SP
913 // |==============|
914 //
915 // The top most frame needs an abi space of 112 bytes. This space is needed,
916 // since we call to c. The c function may spill their arguments to the caller
917 // frame. When we call to java, we don't need these spill slots. In order to save
918 // space on the stack, we resize the caller. However, java locals reside in
919 // the caller frame and the frame has to be increased. The frame_size for the
920 // current frame was calculated based on max_stack as size for the expression
921 // stack. At the call, just a part of the expression stack might be used.
922 // We don't want to waste this space and cut the frame back accordingly.
923 // The resulting amount for resizing is calculated as follows:
924 // resize = (number_of_locals - number_of_arguments) * slot_size
925 // + (R1_SP - R15_esp) + 48
926 //
927 // The size for the callee frame is calculated:
928 // framesize = 112 + max_stack + monitor + state_size
929 //
930 // maxstack: Max number of slots on the expression stack, loaded from the method.
931 // monitor: We statically reserve room for one monitor object.
932 // state_size: We save the current state of the interpreter to this area.
933 //
934 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
935 Register Rparent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
936 Rtop_frame_size = R7_ARG5,
937 Rconst_method = R8_ARG6,
938 Rconst_pool = R9_ARG7,
939 Rmirror = R10_ARG8;
940
941 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, Rparent_frame_resize, Rtop_frame_size,
942 Rconst_method, Rconst_pool);
943
944 __ ld(Rconst_method, method_(const));
945 __ lhz(Rsize_of_parameters /* number of params */,
946 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
947 if (native_call) {
948 // If we're calling a native method, we reserve space for the worst-case signature
949 // handler varargs vector, which is max(Argument::n_int_register_parameters_c, parameter_count+2).
950 // We add two slots to the parameter_count, one for the jni
951 // environment and one for a possible native mirror.
952 Label skip_native_calculate_max_stack;
953 __ addi(Rtop_frame_size, Rsize_of_parameters, 2);
954 __ cmpwi(CCR0, Rtop_frame_size, Argument::n_int_register_parameters_c);
955 __ bge(CCR0, skip_native_calculate_max_stack);
956 __ li(Rtop_frame_size, Argument::n_int_register_parameters_c);
957 __ bind(skip_native_calculate_max_stack);
958 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
959 __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
960 __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
961 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
962 } else {
963 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
964 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
965 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
966 __ lhz(Rtop_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
967 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
968 __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
969 __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
970 __ add(Rparent_frame_resize, Rparent_frame_resize, R11_scratch1);
971 }
972
973 // Compute top frame size.
974 __ addi(Rtop_frame_size, Rtop_frame_size, frame::top_ijava_frame_abi_size + frame::ijava_state_size);
975
976 // Cut back area between esp and max_stack.
977 __ addi(Rparent_frame_resize, Rparent_frame_resize, frame::parent_ijava_frame_abi_size - Interpreter::stackElementSize);
978
979 __ round_to(Rtop_frame_size, frame::alignment_in_bytes);
980 __ round_to(Rparent_frame_resize, frame::alignment_in_bytes);
981 // Rparent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
982 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
983
984 if (!native_call) {
985 // Stack overflow check.
986 // Native calls don't need the stack size check since they have no
987 // expression stack and the arguments are already on the stack and
988 // we only add a handful of words to the stack.
989 __ add(R11_scratch1, Rparent_frame_resize, Rtop_frame_size);
990 generate_stack_overflow_check(R11_scratch1, R12_scratch2);
991 }
992
993 // Set up interpreter state registers.
994
995 __ add(R18_locals, R15_esp, Rsize_of_parameters);
996 __ ld(Rconst_pool, in_bytes(ConstMethod::constants_offset()), Rconst_method);
997 __ ld(R27_constPoolCache, ConstantPool::cache_offset(), Rconst_pool);
998
999 // Set method data pointer.
1000 if (ProfileInterpreter) {
1001 Label zero_continue;
1002 __ ld(R28_mdx, method_(method_data));
1003 __ cmpdi(CCR0, R28_mdx, 0);
1004 __ beq(CCR0, zero_continue);
1005 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1006 __ bind(zero_continue);
1007 }
1008
1009 if (native_call) {
1010 __ li(R14_bcp, 0); // Must initialize.
1011 } else {
1012 __ addi(R14_bcp, Rconst_method, in_bytes(ConstMethod::codes_offset()));
1013 }
1014
1015 // Resize parent frame.
1016 __ mflr(R12_scratch2);
1017 __ neg(Rparent_frame_resize, Rparent_frame_resize);
1018 __ resize_frame(Rparent_frame_resize, R11_scratch1);
1019 __ std(R12_scratch2, _abi0(lr), R1_SP);
1020
1021 // Get mirror and store it in the frame as GC root for this Method*.
1022 __ ld(Rmirror, ConstantPool::pool_holder_offset(), Rconst_pool);
1023 __ ld(Rmirror, in_bytes(Klass::java_mirror_offset()), Rmirror);
1024 __ resolve_oop_handle(Rmirror, R11_scratch1, R12_scratch2, MacroAssembler::PRESERVATION_FRAME_LR_GP_REGS);
1025
1026 __ addi(R26_monitor, R1_SP, -frame::ijava_state_size);
1027 __ addi(R15_esp, R26_monitor, -Interpreter::stackElementSize);
1028
1029 // Store values.
1030 __ std(R19_method, _ijava_state_neg(method), R1_SP);
1031 __ std(Rmirror, _ijava_state_neg(mirror), R1_SP);
1032 __ sub(R12_scratch2, R18_locals, R1_SP);
1033 __ srdi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1034 // Store relativized R18_locals, see frame::interpreter_frame_locals().
1035 __ std(R12_scratch2, _ijava_state_neg(locals), R1_SP);
1036 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
1037
1038 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
1039 // be found in the frame after save_interpreter_state is done. This is always true
1040 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
1041 // because e.g. frame::interpreter_frame_bcp() will not access the correct value
1042 // (Enhanced Stack Trace).
1043 // The signal handler does not save the interpreter state into the frame.
1044
1045 // We have to initialize some of these frame slots for native calls (accessed by GC).
1046 // Also initialize them for non-native calls for better tool support (even though
1047 // you may not get the most recent version as described above).
1048 __ li(R0, 0);
1049 __ li(R12_scratch2, -(frame::ijava_state_size / wordSize));
1050 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
1051 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
1052 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
1053 __ sub(R12_scratch2, R15_esp, R1_SP);
1054 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1055 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
1056 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); // only used for native_call
1057
1058 // Store sender's SP and this frame's top SP.
1059 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
1060 __ neg(R12_scratch2, Rtop_frame_size);
1061 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1062 // Store relativized top_frame_sp
1063 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
1064
1065 // Push top frame.
1066 __ push_frame(Rtop_frame_size, R11_scratch1);
1067 }
1068
1069 // End of helpers
1070
1071 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1072
1073 // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1074 bool use_instruction = false;
1075 address runtime_entry = nullptr;
1076 int num_args = 1;
1077 bool double_precision = true;
1078
1079 // PPC64 specific:
1080 switch (kind) {
1081 case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break;
1082 case Interpreter::java_lang_math_abs: use_instruction = true; break;
1083 case Interpreter::java_lang_math_fmaF:
1084 case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1085 default: break; // Fall back to runtime call.
1086 }
1087
1088 switch (kind) {
1089 case Interpreter::java_lang_math_sin : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); break;
1090 case Interpreter::java_lang_math_cos : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); break;
1091 case Interpreter::java_lang_math_tan : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); break;
1092 case Interpreter::java_lang_math_tanh : /* run interpreted */ break;
1093 case Interpreter::java_lang_math_abs : /* run interpreted */ break;
1094 case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); break;
1095 case Interpreter::java_lang_math_log : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); break;
1096 case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1097 case Interpreter::java_lang_math_pow : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1098 case Interpreter::java_lang_math_exp : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); break;
1099 case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1100 case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1101 default: ShouldNotReachHere();
1102 }
1103
1104 // Use normal entry if neither instruction nor runtime call is used.
1105 if (!use_instruction && runtime_entry == nullptr) return nullptr;
1106
1107 address entry = __ pc();
1108
1109 // Load arguments
1110 assert(num_args <= 13, "passed in registers");
1111 if (double_precision) {
1112 int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1113 for (int i = 0; i < num_args; ++i) {
1114 __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1115 offset -= 2 * Interpreter::stackElementSize;
1116 }
1117 } else {
1118 int offset = num_args * Interpreter::stackElementSize;
1119 for (int i = 0; i < num_args; ++i) {
1120 __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1121 offset -= Interpreter::stackElementSize;
1122 }
1123 }
1124
1125 if (use_instruction) {
1126 switch (kind) {
1127 case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1); break;
1128 case Interpreter::java_lang_math_abs: __ fabs(F1_RET, F1); break;
1129 case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break;
1130 case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3); break;
1131 default: ShouldNotReachHere();
1132 }
1133 } else {
1134 // Comment: Can use tail call if the unextended frame is always C ABI compliant:
1135 //__ load_const_optimized(R12_scratch2, runtime_entry, R0);
1136 //__ call_c_and_return_to_caller(R12_scratch2);
1137
1138 // Push a new C frame and save LR.
1139 __ save_LR(R0);
1140 __ push_frame_reg_args(0, R11_scratch1);
1141
1142 __ call_VM_leaf(runtime_entry);
1143
1144 // Pop the C frame and restore LR.
1145 __ pop_frame();
1146 __ restore_LR(R0);
1147 }
1148
1149 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1150 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1151 __ blr();
1152
1153 __ flush();
1154
1155 return entry;
1156 }
1157
1158 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1159 // Quick & dirty stack overflow checking: bang the stack & handle trap.
1160 // Note that we do the banging after the frame is setup, since the exception
1161 // handling code expects to find a valid interpreter frame on the stack.
1162 // Doing the banging earlier fails if the caller frame is not an interpreter
1163 // frame.
1164 // (Also, the exception throwing code expects to unlock any synchronized
1165 // method receiever, so do the banging after locking the receiver.)
1166
1167 // Bang each page in the shadow zone. We can't assume it's been done for
1168 // an interpreter frame with greater than a page of locals, so each page
1169 // needs to be checked. Only true for non-native.
1170 const size_t page_size = os::vm_page_size();
1171 const int n_shadow_pages = StackOverflow::stack_shadow_zone_size() / page_size;
1172 const int start_page = native_call ? n_shadow_pages : 1;
1173 BLOCK_COMMENT("bang_stack_shadow_pages:");
1174 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1175 __ bang_stack_with_offset(pages*page_size);
1176 }
1177 }
1178
1179 // Interpreter stub for calling a native method. (asm interpreter)
1180 // This sets up a somewhat different looking stack for calling the
1181 // native method than the typical interpreter frame setup.
1182 //
1183 // On entry:
1184 // R19_method - method
1185 // R16_thread - JavaThread*
1186 // R15_esp - intptr_t* sender tos
1187 //
1188 // abstract stack (grows up)
1189 // [ IJava (caller of JNI callee) ] <-- ASP
1190 // ...
1191 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized, bool runtime_upcalls) {
1192
1193 address entry = __ pc();
1194
1195 const bool inc_counter = UseCompiler || CountCompiledCalls;
1196
1197 // -----------------------------------------------------------------------------
1198 // Allocate a new frame that represents the native callee (i2n frame).
1199 // This is not a full-blown interpreter frame, but in particular, the
1200 // following registers are valid after this:
1201 // - R19_method
1202 // - R18_local (points to start of arguments to native function)
1203 //
1204 // abstract stack (grows up)
1205 // [ IJava (caller of JNI callee) ] <-- ASP
1206 // ...
1207
1208 const Register signature_handler_fd = R11_scratch1;
1209 const Register pending_exception = R0;
1210 const Register result_handler_addr = R31;
1211 const Register native_method_fd = R12_scratch2; // preferred in MacroAssembler::branch_to
1212 const Register access_flags = R22_tmp2;
1213 const Register active_handles = R11_scratch1; // R26_monitor saved to state.
1214 const Register sync_state = R12_scratch2;
1215 const Register sync_state_addr = sync_state; // Address is dead after use.
1216 const Register suspend_flags = R11_scratch1;
1217
1218 //=============================================================================
1219 // Allocate new frame and initialize interpreter state.
1220
1221 Label exception_return;
1222 Label exception_return_sync_check;
1223 Label stack_overflow_return;
1224
1225 Register size_of_parameters = R22_tmp2;
1226
1227 generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
1228
1229 //=============================================================================
1230 // Increment invocation counter. On overflow, entry to JNI method
1231 // will be compiled.
1232 Label invocation_counter_overflow, continue_after_compile;
1233 if (inc_counter) {
1234 if (synchronized) {
1235 // Since at this point in the method invocation the exception handler
1236 // would try to exit the monitor of synchronized methods which hasn't
1237 // been entered yet, we set the thread local variable
1238 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1239 // runtime, exception handling i.e. unlock_if_synchronized_method will
1240 // check this thread local flag.
1241 // This flag has two effects, one is to force an unwind in the topmost
1242 // interpreter frame and not perform an unlock while doing so.
1243 __ li(R0, 1);
1244 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1245 }
1246 generate_counter_incr(&invocation_counter_overflow);
1247
1248 BIND(continue_after_compile);
1249 }
1250
1251 bang_stack_shadow_pages(true);
1252
1253 if (inc_counter) {
1254 // Reset the _do_not_unlock_if_synchronized flag.
1255 if (synchronized) {
1256 __ li(R0, 0);
1257 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1258 }
1259 }
1260
1261 // access_flags = method->access_flags();
1262 // Load access flags.
1263 assert(__ nonvolatile_accross_vthread_preemtion(access_flags),
1264 "access_flags not preserved");
1265 // Type check.
1266 assert(4 == sizeof(AccessFlags), "unexpected field size");
1267 __ lwz(access_flags, method_(access_flags));
1268
1269 // We don't want to reload R19_method and access_flags after calls
1270 // to some helper functions.
1271 assert(R19_method->is_nonvolatile(),
1272 "R19_method must be a non-volatile register");
1273
1274 // Check for synchronized methods. Must happen AFTER invocation counter
1275 // check, so method is not locked if counter overflows.
1276
1277 if (synchronized) {
1278 lock_method(access_flags, R11_scratch1, R12_scratch2, true);
1279
1280 // Update monitor in state.
1281 __ ld(R11_scratch1, 0, R1_SP);
1282 __ sub(R12_scratch2, R26_monitor, R11_scratch1);
1283 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1284 __ std(R12_scratch2, _ijava_state_neg(monitors), R11_scratch1);
1285 }
1286
1287 // jvmti/jvmpi support
1288 __ notify_method_entry();
1289
1290 //=============================================================================
1291 // Get and call the signature handler.
1292
1293 __ ld(signature_handler_fd, method_(signature_handler));
1294 Label call_signature_handler;
1295
1296 __ cmpdi(CCR0, signature_handler_fd, 0);
1297 __ bne(CCR0, call_signature_handler);
1298
1299 // Method has never been called. Either generate a specialized
1300 // handler or point to the slow one.
1301 //
1302 // Pass parameter 'false' to avoid exception check in call_VM.
1303 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
1304
1305 // Check for an exception while looking up the target method. If we
1306 // incurred one, bail.
1307 __ ld(pending_exception, thread_(pending_exception));
1308 __ cmpdi(CCR0, pending_exception, 0);
1309 __ bne(CCR0, exception_return_sync_check); // Has pending exception.
1310
1311 // Reload signature handler, it may have been created/assigned in the meanwhile.
1312 __ ld(signature_handler_fd, method_(signature_handler));
1313 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
1314
1315 BIND(call_signature_handler);
1316
1317 // Before we call the signature handler we push a new frame to
1318 // protect the interpreter frame volatile registers when we return
1319 // from jni but before we can get back to Java.
1320
1321 // First set the frame anchor while the SP/FP registers are
1322 // convenient and the slow signature handler can use this same frame
1323 // anchor.
1324
1325 bool support_vthread_preemption = Continuations::enabled() && LockingMode != LM_LEGACY;
1326
1327 // We have a TOP_IJAVA_FRAME here, which belongs to us.
1328 Label last_java_pc;
1329 Label *resume_pc = support_vthread_preemption ? &last_java_pc : nullptr;
1330 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R3_ARG1/*tmp*/, resume_pc);
1331
1332 // Now the interpreter frame (and its call chain) have been
1333 // invalidated and flushed. We are now protected against eager
1334 // being enabled in native code. Even if it goes eager the
1335 // registers will be reloaded as clean and we will invalidate after
1336 // the call so no spurious flush should be possible.
1337
1338 // Call signature handler and pass locals address.
1339 //
1340 // Our signature handlers copy required arguments to the C stack
1341 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
1342 __ mr(R3_ARG1, R18_locals);
1343 #if !defined(ABI_ELFv2)
1344 __ ld(signature_handler_fd, 0, signature_handler_fd);
1345 #endif
1346
1347 __ call_stub(signature_handler_fd);
1348
1349 assert(__ nonvolatile_accross_vthread_preemtion(result_handler_addr),
1350 "result_handler_addr not preserved");
1351 // Save across call to native method.
1352 __ mr(result_handler_addr, R3_RET);
1353 __ ld(R11_scratch1, _abi0(callers_sp), R1_SP); // load FP
1354
1355 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
1356
1357 // Set up fixed parameters and call the native method.
1358 // If the method is static, get mirror into R4_ARG2.
1359 {
1360 Label method_is_not_static;
1361 // Access_flags is non-volatile and still, no need to restore it.
1362
1363 // Restore access flags.
1364 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
1365 __ bfalse(CCR0, method_is_not_static);
1366
1367 // Load mirror from interpreter frame (FP in R11_scratch1)
1368 __ ld(R21_tmp1, _ijava_state_neg(mirror), R11_scratch1);
1369 // R4_ARG2 = &state->_oop_temp;
1370 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
1371 __ std(R21_tmp1/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
1372 BIND(method_is_not_static);
1373 }
1374
1375 // At this point, arguments have been copied off the stack into
1376 // their JNI positions. Oops are boxed in-place on the stack, with
1377 // handles copied to arguments. The result handler address is in a
1378 // register.
1379
1380 // Pass JNIEnv address as first parameter.
1381 __ addir(R3_ARG1, thread_(jni_environment));
1382
1383 // Load the native_method entry before we change the thread state.
1384 __ ld(native_method_fd, method_(native_function));
1385
1386 //=============================================================================
1387 // Transition from _thread_in_Java to _thread_in_native. As soon as
1388 // we make this change the safepoint code needs to be certain that
1389 // the last Java frame we established is good. The pc in that frame
1390 // just needs to be near here not an actual return address.
1391
1392 // We use release_store_fence to update values like the thread state, where
1393 // we don't want the current thread to continue until all our prior memory
1394 // accesses (including the new thread state) are visible to other threads.
1395 __ li(R0, _thread_in_native);
1396 __ release();
1397
1398 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
1399 __ stw(R0, thread_(thread_state));
1400
1401 //=============================================================================
1402 // Call the native method. Argument registers must not have been
1403 // overwritten since "__ call_stub(signature_handler);" (except for
1404 // ARG1 and ARG2 for static methods).
1405
1406 if (support_vthread_preemption) {
1407 // result_handler_addr is a nonvolatile register. Its value will be preserved across
1408 // the native call but only if the call isn't preempted. To preserve its value even
1409 // in the case of preemption we save it in the lresult slot. It is restored at
1410 // resume_pc if, and only if the call was preempted. This works because only
1411 // j.l.Object::wait calls are preempted which don't return a result.
1412 __ std(result_handler_addr, _ijava_state_neg(lresult), R11_scratch1);
1413 }
1414 __ push_cont_fastpath();
1415 __ call_c(native_method_fd);
1416 __ pop_cont_fastpath();
1417
1418 __ li(R0, 0);
1419 __ ld(R11_scratch1, 0, R1_SP);
1420 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1421 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1422 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
1423
1424 // Note: C++ interpreter needs the following here:
1425 // The frame_manager_lr field, which we use for setting the last
1426 // java frame, gets overwritten by the signature handler. Restore
1427 // it now.
1428 //__ get_PC_trash_LR(R11_scratch1);
1429 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
1430
1431 // Because of GC R19_method may no longer be valid.
1432
1433 // Block, if necessary, before resuming in _thread_in_Java state.
1434 // In order for GC to work, don't clear the last_Java_sp until after
1435 // blocking.
1436
1437 //=============================================================================
1438 // Switch thread to "native transition" state before reading the
1439 // synchronization state. This additional state is necessary
1440 // because reading and testing the synchronization state is not
1441 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1442 // in _thread_in_native state, loads _not_synchronized and is
1443 // preempted. VM thread changes sync state to synchronizing and
1444 // suspends threads for GC. Thread A is resumed to finish this
1445 // native method, but doesn't block here since it didn't see any
1446 // synchronization in progress, and escapes.
1447
1448 // We use release_store_fence to update values like the thread state, where
1449 // we don't want the current thread to continue until all our prior memory
1450 // accesses (including the new thread state) are visible to other threads.
1451 __ li(R0/*thread_state*/, _thread_in_native_trans);
1452 __ release();
1453 __ stw(R0/*thread_state*/, thread_(thread_state));
1454 if (!UseSystemMemoryBarrier) {
1455 __ fence();
1456 }
1457
1458 // Now before we return to java we must look for a current safepoint
1459 // (a new safepoint can not start since we entered native_trans).
1460 // We must check here because a current safepoint could be modifying
1461 // the callers registers right this moment.
1462
1463 // Acquire isn't strictly necessary here because of the fence, but
1464 // sync_state is declared to be volatile, so we do it anyway
1465 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1466
1467 Label do_safepoint, sync_check_done;
1468 // No synchronization in progress nor yet synchronized.
1469 __ safepoint_poll(do_safepoint, sync_state, true /* at_return */, false /* in_nmethod */);
1470
1471 // Not suspended.
1472 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1473 __ lwz(suspend_flags, thread_(suspend_flags));
1474 __ cmpwi(CCR1, suspend_flags, 0);
1475 __ beq(CCR1, sync_check_done);
1476
1477 __ bind(do_safepoint);
1478 __ isync();
1479 // Block. We do the call directly and leave the current
1480 // last_Java_frame setup undisturbed. We must save any possible
1481 // native result across the call. No oop is present.
1482
1483 __ mr(R3_ARG1, R16_thread);
1484 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1485
1486 __ bind(sync_check_done);
1487
1488 //=============================================================================
1489 // <<<<<< Back in Interpreter Frame >>>>>
1490
1491 // We are in thread_in_native_trans here and back in the normal
1492 // interpreter frame. We don't have to do anything special about
1493 // safepoints and we can switch to Java mode anytime we are ready.
1494
1495 // Note: frame::interpreter_frame_result has a dependency on how the
1496 // method result is saved across the call to post_method_exit. For
1497 // native methods it assumes that the non-FPU/non-void result is
1498 // saved in _native_lresult and a FPU result in _native_fresult. If
1499 // this changes then the interpreter_frame_result implementation
1500 // will need to be updated too.
1501
1502 // On PPC64, we have stored the result directly after the native call.
1503
1504 //=============================================================================
1505 // Back in Java
1506
1507 // We use release_store_fence to update values like the thread state, where
1508 // we don't want the current thread to continue until all our prior memory
1509 // accesses (including the new thread state) are visible to other threads.
1510 __ li(R0/*thread_state*/, _thread_in_Java);
1511 __ lwsync(); // Acquire safepoint and suspend state, release thread state.
1512 __ stw(R0/*thread_state*/, thread_(thread_state));
1513
1514 if (support_vthread_preemption) {
1515 // Check preemption for Object.wait()
1516 Label not_preempted;
1517 __ ld(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
1518 __ cmpdi(CCR0, R0, 0);
1519 __ beq(CCR0, not_preempted);
1520 __ mtlr(R0);
1521 __ li(R0, 0);
1522 __ std(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
1523 __ blr();
1524
1525 // Execution will be resumed here when the vthread becomes runnable again.
1526 __ bind(*resume_pc);
1527 __ restore_after_resume(R11_scratch1 /* fp */);
1528 // We saved the result handler before the call
1529 __ ld(result_handler_addr, _ijava_state_neg(lresult), R11_scratch1);
1530 #ifdef ASSERT
1531 // Clobber result slots. Only native methods returning void can be preemted currently.
1532 __ load_const(R3_RET, UCONST64(0xbad01001));
1533 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1534 __ std(R3_RET, _ijava_state_neg(fresult), R11_scratch1);
1535 // reset_last_Java_frame() below asserts that a last java sp is set
1536 __ asm_assert_mem8_is_zero(in_bytes(JavaThread::last_Java_sp_offset()),
1537 R16_thread, FILE_AND_LINE ": Last java sp should not be set when resuming");
1538 __ std(R3_RET, in_bytes(JavaThread::last_Java_sp_offset()), R16_thread);
1539 #endif
1540 __ bind(not_preempted);
1541 }
1542
1543 if (CheckJNICalls) {
1544 // clear_pending_jni_exception_check
1545 __ load_const_optimized(R0, 0L);
1546 __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread);
1547 }
1548
1549 __ reset_last_Java_frame();
1550
1551 // Jvmdi/jvmpi support. Whether we've got an exception pending or
1552 // not, and whether unlocking throws an exception or not, we notify
1553 // on native method exit. If we do have an exception, we'll end up
1554 // in the caller's context to handle it, so if we don't do the
1555 // notify here, we'll drop it on the floor.
1556 __ notify_method_exit(true/*native method*/,
1557 ilgl /*illegal state (not used for native methods)*/,
1558 InterpreterMacroAssembler::NotifyJVMTI,
1559 false /*check_exceptions*/);
1560
1561 //=============================================================================
1562 // Handle exceptions
1563
1564 if (synchronized) {
1565 __ unlock_object(R26_monitor); // Can also unlock methods.
1566 }
1567
1568 // Reset active handles after returning from native.
1569 // thread->active_handles()->clear();
1570 __ ld(active_handles, thread_(active_handles));
1571 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1572 __ li(R0, 0);
1573 __ stw(R0, in_bytes(JNIHandleBlock::top_offset()), active_handles);
1574
1575 Label exception_return_sync_check_already_unlocked;
1576 __ ld(R0/*pending_exception*/, thread_(pending_exception));
1577 __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1578 __ bne(CCR0, exception_return_sync_check_already_unlocked);
1579
1580 //-----------------------------------------------------------------------------
1581 // No exception pending.
1582
1583 // Move native method result back into proper registers and return.
1584 // Invoke result handler (may unbox/promote).
1585 __ ld(R11_scratch1, 0, R1_SP);
1586 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1587 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1588 __ call_stub(result_handler_addr);
1589
1590 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1591
1592 // Must use the return pc which was loaded from the caller's frame
1593 // as the VM uses return-pc-patching for deoptimization.
1594 __ mtlr(R0);
1595 __ blr();
1596
1597 //-----------------------------------------------------------------------------
1598 // An exception is pending. We call into the runtime only if the
1599 // caller was not interpreted. If it was interpreted the
1600 // interpreter will do the correct thing. If it isn't interpreted
1601 // (call stub/compiled code) we will change our return and continue.
1602
1603 BIND(exception_return_sync_check);
1604
1605 if (synchronized) {
1606 __ unlock_object(R26_monitor); // Can also unlock methods.
1607 }
1608 BIND(exception_return_sync_check_already_unlocked);
1609
1610 const Register return_pc = R31;
1611
1612 __ ld(return_pc, 0, R1_SP);
1613 __ ld(return_pc, _abi0(lr), return_pc);
1614
1615 // Get the address of the exception handler.
1616 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1617 R16_thread,
1618 return_pc /* return pc */);
1619 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1620
1621 // Load the PC of the exception handler into LR.
1622 __ mtlr(R3_RET);
1623
1624 // Load exception into R3_ARG1 and clear pending exception in thread.
1625 __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1626 __ li(R4_ARG2, 0);
1627 __ std(R4_ARG2, thread_(pending_exception));
1628
1629 // Load the original return pc into R4_ARG2.
1630 __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1631
1632 // Return to exception handler.
1633 __ blr();
1634
1635 //=============================================================================
1636 // Counter overflow.
1637
1638 if (inc_counter) {
1639 // Handle invocation counter overflow.
1640 __ bind(invocation_counter_overflow);
1641
1642 generate_counter_overflow(continue_after_compile);
1643 }
1644
1645 return entry;
1646 }
1647
1648 // Generic interpreted method entry to (asm) interpreter.
1649 //
1650 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized, bool runtime_upcalls) {
1651 bool inc_counter = UseCompiler || CountCompiledCalls;
1652 address entry = __ pc();
1653 // Generate the code to allocate the interpreter stack frame.
1654 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1655 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame.
1656
1657 // Does also a stack check to assure this frame fits on the stack.
1658 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1659
1660 // --------------------------------------------------------------------------
1661 // Zero out non-parameter locals.
1662 // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1663 // worth to ask the flag, just do it.
1664 Register Rslot_addr = R6_ARG4,
1665 Rnum = R7_ARG5;
1666 Label Lno_locals, Lzero_loop;
1667
1668 // Set up the zeroing loop.
1669 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1670 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1671 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1672 __ beq(CCR0, Lno_locals);
1673 __ li(R0, 0);
1674 __ mtctr(Rnum);
1675
1676 // The zero locals loop.
1677 __ bind(Lzero_loop);
1678 __ std(R0, 0, Rslot_addr);
1679 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1680 __ bdnz(Lzero_loop);
1681
1682 __ bind(Lno_locals);
1683
1684 // --------------------------------------------------------------------------
1685 // Counter increment and overflow check.
1686 Label invocation_counter_overflow;
1687 Label continue_after_compile;
1688 if (inc_counter || ProfileInterpreter) {
1689
1690 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1691 if (synchronized) {
1692 // Since at this point in the method invocation the exception handler
1693 // would try to exit the monitor of synchronized methods which hasn't
1694 // been entered yet, we set the thread local variable
1695 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1696 // runtime, exception handling i.e. unlock_if_synchronized_method will
1697 // check this thread local flag.
1698 // This flag has two effects, one is to force an unwind in the topmost
1699 // interpreter frame and not perform an unlock while doing so.
1700 __ li(R0, 1);
1701 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1702 }
1703
1704 // Argument and return type profiling.
1705 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1706
1707 // Increment invocation counter and check for overflow.
1708 if (inc_counter) {
1709 generate_counter_incr(&invocation_counter_overflow);
1710 }
1711
1712 __ bind(continue_after_compile);
1713 }
1714
1715 bang_stack_shadow_pages(false);
1716
1717 if (inc_counter || ProfileInterpreter) {
1718 // Reset the _do_not_unlock_if_synchronized flag.
1719 if (synchronized) {
1720 __ li(R0, 0);
1721 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1722 }
1723 }
1724
1725 // --------------------------------------------------------------------------
1726 // Locking of synchronized methods. Must happen AFTER invocation_counter
1727 // check and stack overflow check, so method is not locked if overflows.
1728 if (synchronized) {
1729 lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1730 }
1731 #ifdef ASSERT
1732 else {
1733 Label Lok;
1734 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1735 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1736 __ asm_assert_eq("method needs synchronization");
1737 __ bind(Lok);
1738 }
1739 #endif // ASSERT
1740
1741 // --------------------------------------------------------------------------
1742 // JVMTI support
1743 __ notify_method_entry();
1744
1745 // --------------------------------------------------------------------------
1746 // Start executing instructions.
1747 __ dispatch_next(vtos);
1748
1749 // --------------------------------------------------------------------------
1750 if (inc_counter) {
1751 // Handle invocation counter overflow.
1752 __ bind(invocation_counter_overflow);
1753 generate_counter_overflow(continue_after_compile);
1754 }
1755 return entry;
1756 }
1757
1758 // CRC32 Intrinsics.
1759 //
1760 // Contract on scratch and work registers.
1761 // =======================================
1762 //
1763 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1764 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1765 // You can't rely on these registers across calls.
1766 //
1767 // The generators for CRC32_update and for CRC32_updateBytes use the
1768 // scratch/work register set internally, passing the work registers
1769 // as arguments to the MacroAssembler emitters as required.
1770 //
1771 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1772 // Their contents is not constant but may change according to the requirements
1773 // of the emitted code.
1774 //
1775 // All other registers from the scratch/work register set are used "internally"
1776 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1777 // Basically, only R3_RET contains a defined value which is the function result.
1778 //
1779 /**
1780 * Method entry for static native methods:
1781 * int java.util.zip.CRC32.update(int crc, int b)
1782 */
1783 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1784 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1785 address start = __ pc(); // Remember stub start address (is rtn value).
1786 Label slow_path;
1787
1788 // Safepoint check
1789 const Register sync_state = R11_scratch1;
1790 __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1791
1792 // We don't generate local frame and don't align stack because
1793 // we not even call stub code (we generate the code inline)
1794 // and there is no safepoint on this path.
1795
1796 // Load java parameters.
1797 // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1798 const Register argP = R15_esp;
1799 const Register crc = R3_ARG1; // crc value
1800 const Register data = R4_ARG2;
1801 const Register table = R5_ARG3; // address of crc32 table
1802
1803 BLOCK_COMMENT("CRC32_update {");
1804
1805 // Arguments are reversed on java expression stack
1806 #ifdef VM_LITTLE_ENDIAN
1807 int data_offs = 0+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1808 // Being passed as an int, the single byte is at offset +0.
1809 #else
1810 int data_offs = 3+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1811 // Being passed from java as an int, the single byte is at offset +3.
1812 #endif
1813 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1814 __ lbz(data, data_offs, argP); // Byte from buffer, zero-extended.
1815 __ load_const_optimized(table, StubRoutines::crc_table_addr(), R0);
1816 __ kernel_crc32_singleByteReg(crc, data, table, true);
1817
1818 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1819 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1820 __ blr();
1821
1822 // Generate a vanilla native entry as the slow path.
1823 BLOCK_COMMENT("} CRC32_update");
1824 BIND(slow_path);
1825 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1826 return start;
1827 }
1828
1829 /**
1830 * Method entry for static native methods:
1831 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len)
1832 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1833 */
1834 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1835 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1836 address start = __ pc(); // Remember stub start address (is rtn value).
1837 Label slow_path;
1838
1839 // Safepoint check
1840 const Register sync_state = R11_scratch1;
1841 __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1842
1843 // We don't generate local frame and don't align stack because
1844 // we not even call stub code (we generate the code inline)
1845 // and there is no safepoint on this path.
1846
1847 // Load parameters.
1848 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1849 const Register argP = R15_esp;
1850 const Register crc = R3_ARG1; // crc value
1851 const Register data = R4_ARG2; // address of java byte array
1852 const Register dataLen = R5_ARG3; // source data len
1853 const Register tmp = R11_scratch1;
1854
1855 // Arguments are reversed on java expression stack.
1856 // Calculate address of start element.
1857 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1858 BLOCK_COMMENT("CRC32_updateByteBuffer {");
1859 // crc @ (SP + 5W) (32bit)
1860 // buf @ (SP + 3W) (64bit ptr to long array)
1861 // off @ (SP + 2W) (32bit)
1862 // dataLen @ (SP + 1W) (32bit)
1863 // data = buf + off
1864 __ ld( data, 3*wordSize, argP); // start of byte buffer
1865 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1866 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1867 __ lwz( crc, 5*wordSize, argP); // current crc state
1868 __ add( data, data, tmp); // Add byte buffer offset.
1869 } else { // Used for "updateBytes update".
1870 BLOCK_COMMENT("CRC32_updateBytes {");
1871 // crc @ (SP + 4W) (32bit)
1872 // buf @ (SP + 3W) (64bit ptr to byte array)
1873 // off @ (SP + 2W) (32bit)
1874 // dataLen @ (SP + 1W) (32bit)
1875 // data = buf + off + base_offset
1876 __ ld( data, 3*wordSize, argP); // start of byte buffer
1877 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1878 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1879 __ add( data, data, tmp); // add byte buffer offset
1880 __ lwz( crc, 4*wordSize, argP); // current crc state
1881 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1882 }
1883
1884 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, false);
1885
1886 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1887 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1888 __ blr();
1889
1890 // Generate a vanilla native entry as the slow path.
1891 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1892 BIND(slow_path);
1893 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1894 return start;
1895 }
1896
1897
1898 /**
1899 * Method entry for intrinsic-candidate (non-native) methods:
1900 * int java.util.zip.CRC32C.updateBytes( int crc, byte[] b, int off, int end)
1901 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
1902 * Unlike CRC32, CRC32C does not have any methods marked as native
1903 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1904 **/
1905 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1906 assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1907 address start = __ pc(); // Remember stub start address (is rtn value).
1908
1909 // We don't generate local frame and don't align stack because
1910 // we not even call stub code (we generate the code inline)
1911 // and there is no safepoint on this path.
1912
1913 // Load parameters.
1914 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1915 const Register argP = R15_esp;
1916 const Register crc = R3_ARG1; // crc value
1917 const Register data = R4_ARG2; // address of java byte array
1918 const Register dataLen = R5_ARG3; // source data len
1919 const Register tmp = R11_scratch1;
1920
1921 // Arguments are reversed on java expression stack.
1922 // Calculate address of start element.
1923 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer".
1924 BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
1925 // crc @ (SP + 5W) (32bit)
1926 // buf @ (SP + 3W) (64bit ptr to long array)
1927 // off @ (SP + 2W) (32bit)
1928 // dataLen @ (SP + 1W) (32bit)
1929 // data = buf + off
1930 __ ld( data, 3*wordSize, argP); // start of byte buffer
1931 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1932 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1933 __ lwz( crc, 5*wordSize, argP); // current crc state
1934 __ add( data, data, tmp); // Add byte buffer offset.
1935 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1936 } else { // Used for "updateBytes update".
1937 BLOCK_COMMENT("CRC32C_updateBytes {");
1938 // crc @ (SP + 4W) (32bit)
1939 // buf @ (SP + 3W) (64bit ptr to byte array)
1940 // off @ (SP + 2W) (32bit)
1941 // dataLen @ (SP + 1W) (32bit)
1942 // data = buf + off + base_offset
1943 __ ld( data, 3*wordSize, argP); // start of byte buffer
1944 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1945 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1946 __ add( data, data, tmp); // add byte buffer offset
1947 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1948 __ lwz( crc, 4*wordSize, argP); // current crc state
1949 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1950 }
1951
1952 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, true);
1953
1954 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1955 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1956 __ blr();
1957
1958 BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
1959 return start;
1960 }
1961
1962 // Not supported
1963 address TemplateInterpreterGenerator::generate_currentThread() { return nullptr; }
1964 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
1965 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
1966 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
1967 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
1968 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() { return nullptr; }
1969 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() { return nullptr; }
1970
1971 // =============================================================================
1972 // Exceptions
1973
1974 void TemplateInterpreterGenerator::generate_throw_exception() {
1975 Register Rexception = R17_tos,
1976 Rcontinuation = R3_RET;
1977
1978 // --------------------------------------------------------------------------
1979 // Entry point if an method returns with a pending exception (rethrow).
1980 Interpreter::_rethrow_exception_entry = __ pc();
1981 {
1982 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
1983
1984 // Compiled code destroys templateTableBase, reload.
1985 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
1986 }
1987
1988 // Entry point if a interpreted method throws an exception (throw).
1989 Interpreter::_throw_exception_entry = __ pc();
1990 {
1991 __ mr(Rexception, R3_RET);
1992
1993 __ verify_oop(Rexception);
1994
1995 // Expression stack must be empty before entering the VM in case of an exception.
1996 __ empty_expression_stack();
1997 // Find exception handler address and preserve exception oop.
1998 // Call C routine to find handler and jump to it.
1999 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
2000 __ mtctr(Rcontinuation);
2001 // Push exception for exception handler bytecodes.
2002 __ push_ptr(Rexception);
2003
2004 // Jump to exception handler (may be remove activation entry!).
2005 __ bctr();
2006 }
2007
2008 // If the exception is not handled in the current frame the frame is
2009 // removed and the exception is rethrown (i.e. exception
2010 // continuation is _rethrow_exception).
2011 //
2012 // Note: At this point the bci is still the bxi for the instruction
2013 // which caused the exception and the expression stack is
2014 // empty. Thus, for any VM calls at this point, GC will find a legal
2015 // oop map (with empty expression stack).
2016
2017 // In current activation
2018 // tos: exception
2019 // bcp: exception bcp
2020
2021 // --------------------------------------------------------------------------
2022 // JVMTI PopFrame support
2023
2024 Interpreter::_remove_activation_preserving_args_entry = __ pc();
2025 {
2026 // Set the popframe_processing bit in popframe_condition indicating that we are
2027 // currently handling popframe, so that call_VMs that may happen later do not
2028 // trigger new popframe handling cycles.
2029 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2030 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
2031 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2032
2033 // Empty the expression stack, as in normal exception handling.
2034 __ empty_expression_stack();
2035 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
2036
2037 // Check to see whether we are returning to a deoptimized frame.
2038 // (The PopFrame call ensures that the caller of the popped frame is
2039 // either interpreted or compiled and deoptimizes it if compiled.)
2040 // Note that we don't compare the return PC against the
2041 // deoptimization blob's unpack entry because of the presence of
2042 // adapter frames in C2.
2043 Label Lcaller_not_deoptimized;
2044 Register return_pc = R3_ARG1;
2045 __ ld(return_pc, 0, R1_SP);
2046 __ ld(return_pc, _abi0(lr), return_pc);
2047 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
2048 __ cmpdi(CCR0, R3_RET, 0);
2049 __ bne(CCR0, Lcaller_not_deoptimized);
2050
2051 // The deoptimized case.
2052 // In this case, we can't call dispatch_next() after the frame is
2053 // popped, but instead must save the incoming arguments and restore
2054 // them after deoptimization has occurred.
2055 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
2056 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
2057 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
2058 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
2059 __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
2060 // Save these arguments.
2061 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
2062
2063 // Inform deoptimization that it is responsible for restoring these arguments.
2064 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
2065 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2066
2067 // Return from the current method into the deoptimization blob. Will eventually
2068 // end up in the deopt interpreter entry, deoptimization prepared everything that
2069 // we will reexecute the call that called us.
2070 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
2071 __ mtlr(return_pc);
2072 __ pop_cont_fastpath();
2073 __ blr();
2074
2075 // The non-deoptimized case.
2076 __ bind(Lcaller_not_deoptimized);
2077
2078 // Clear the popframe condition flag.
2079 __ li(R0, 0);
2080 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2081
2082 // Get out of the current method and re-execute the call that called us.
2083 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2084 __ pop_cont_fastpath();
2085 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
2086 if (ProfileInterpreter) {
2087 __ set_method_data_pointer_for_bcp();
2088 __ ld(R11_scratch1, 0, R1_SP);
2089 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
2090 }
2091 #if INCLUDE_JVMTI
2092 Label L_done;
2093
2094 __ lbz(R11_scratch1, 0, R14_bcp);
2095 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
2096 __ bne(CCR0, L_done);
2097
2098 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2099 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
2100 __ ld(R4_ARG2, 0, R18_locals);
2101 __ call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp);
2102
2103 __ cmpdi(CCR0, R4_ARG2, 0);
2104 __ beq(CCR0, L_done);
2105 __ std(R4_ARG2, wordSize, R15_esp);
2106 __ bind(L_done);
2107 #endif // INCLUDE_JVMTI
2108 __ dispatch_next(vtos);
2109 }
2110 // end of JVMTI PopFrame support
2111
2112 // --------------------------------------------------------------------------
2113 // Remove activation exception entry.
2114 // This is jumped to if an interpreted method can't handle an exception itself
2115 // (we come from the throw/rethrow exception entry above). We're going to call
2116 // into the VM to find the exception handler in the caller, pop the current
2117 // frame and return the handler we calculated.
2118 Interpreter::_remove_activation_entry = __ pc();
2119 {
2120 __ pop_ptr(Rexception);
2121 __ verify_oop(Rexception);
2122 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
2123
2124 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
2125 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
2126
2127 __ get_vm_result(Rexception);
2128
2129 // We are done with this activation frame; find out where to go next.
2130 // The continuation point will be an exception handler, which expects
2131 // the following registers set up:
2132 //
2133 // RET: exception oop
2134 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
2135
2136 Register return_pc = R31; // Needs to survive the runtime call.
2137 __ ld(return_pc, 0, R1_SP);
2138 __ ld(return_pc, _abi0(lr), return_pc);
2139 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
2140
2141 // Remove the current activation.
2142 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2143 __ pop_cont_fastpath();
2144
2145 __ mr(R4_ARG2, return_pc);
2146 __ mtlr(R3_RET);
2147 __ mr(R3_RET, Rexception);
2148 __ blr();
2149 }
2150 }
2151
2152 // JVMTI ForceEarlyReturn support.
2153 // Returns "in the middle" of a method with a "fake" return value.
2154 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2155
2156 Register Rscratch1 = R11_scratch1,
2157 Rscratch2 = R12_scratch2;
2158
2159 address entry = __ pc();
2160 __ empty_expression_stack();
2161
2162 __ load_earlyret_value(state, Rscratch1);
2163
2164 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
2165 // Clear the earlyret state.
2166 __ li(R0, 0);
2167 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
2168
2169 __ remove_activation(state, false, false);
2170 // Copied from TemplateTable::_return.
2171 // Restoration of lr done by remove_activation.
2172 switch (state) {
2173 // Narrow result if state is itos but result type is smaller.
2174 case btos:
2175 case ztos:
2176 case ctos:
2177 case stos:
2178 case itos: __ narrow(R17_tos); /* fall through */
2179 case ltos:
2180 case atos: __ mr(R3_RET, R17_tos); break;
2181 case ftos:
2182 case dtos: __ fmr(F1_RET, F15_ftos); break;
2183 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2184 // to get visible before the reference to the object gets stored anywhere.
2185 __ membar(Assembler::StoreStore); break;
2186 default : ShouldNotReachHere();
2187 }
2188 __ blr();
2189
2190 return entry;
2191 } // end of ForceEarlyReturn support
2192
2193 //-----------------------------------------------------------------------------
2194 // Helper for vtos entry point generation
2195
2196 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2197 address& bep,
2198 address& cep,
2199 address& sep,
2200 address& aep,
2201 address& iep,
2202 address& lep,
2203 address& fep,
2204 address& dep,
2205 address& vep) {
2206 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2207 Label L;
2208
2209 aep = __ pc(); __ push_ptr(); __ b(L);
2210 fep = __ pc(); __ push_f(); __ b(L);
2211 dep = __ pc(); __ push_d(); __ b(L);
2212 lep = __ pc(); __ push_l(); __ b(L);
2213 __ align(32, 12, 24); // align L
2214 bep = cep = sep =
2215 iep = __ pc(); __ push_i();
2216 vep = __ pc();
2217 __ bind(L);
2218 generate_and_dispatch(t);
2219 }
2220
2221 //-----------------------------------------------------------------------------
2222
2223 void TemplateInterpreterGenerator::count_bytecode() {
2224 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
2225 __ lwz(R12_scratch2, offs, R11_scratch1);
2226 __ addi(R12_scratch2, R12_scratch2, 1);
2227 __ stw(R12_scratch2, offs, R11_scratch1);
2228 }
2229
2230 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2231 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
2232 __ lwz(R12_scratch2, offs, R11_scratch1);
2233 __ addi(R12_scratch2, R12_scratch2, 1);
2234 __ stw(R12_scratch2, offs, R11_scratch1);
2235 }
2236
2237 // Non-product code
2238 #ifndef PRODUCT
2239 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2240 //__ flush_bundle();
2241 address entry = __ pc();
2242
2243 const char *bname = nullptr;
2244 uint tsize = 0;
2245 switch(state) {
2246 case ftos:
2247 bname = "trace_code_ftos {";
2248 tsize = 2;
2249 break;
2250 case btos:
2251 bname = "trace_code_btos {";
2252 tsize = 2;
2253 break;
2254 case ztos:
2255 bname = "trace_code_ztos {";
2256 tsize = 2;
2257 break;
2258 case ctos:
2259 bname = "trace_code_ctos {";
2260 tsize = 2;
2261 break;
2262 case stos:
2263 bname = "trace_code_stos {";
2264 tsize = 2;
2265 break;
2266 case itos:
2267 bname = "trace_code_itos {";
2268 tsize = 2;
2269 break;
2270 case ltos:
2271 bname = "trace_code_ltos {";
2272 tsize = 3;
2273 break;
2274 case atos:
2275 bname = "trace_code_atos {";
2276 tsize = 2;
2277 break;
2278 case vtos:
2279 // Note: In case of vtos, the topmost of stack value could be a int or doubl
2280 // In case of a double (2 slots) we won't see the 2nd stack value.
2281 // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
2282 bname = "trace_code_vtos {";
2283 tsize = 2;
2284
2285 break;
2286 case dtos:
2287 bname = "trace_code_dtos {";
2288 tsize = 3;
2289 break;
2290 default:
2291 ShouldNotReachHere();
2292 }
2293 BLOCK_COMMENT(bname);
2294
2295 // Support short-cut for TraceBytecodesAt.
2296 // Don't call into the VM if we don't want to trace to speed up things.
2297 Label Lskip_vm_call;
2298 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2299 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
2300 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2301 __ ld(R11_scratch1, offs1, R11_scratch1);
2302 __ lwa(R12_scratch2, offs2, R12_scratch2);
2303 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2304 __ blt(CCR0, Lskip_vm_call);
2305 }
2306
2307 __ push(state);
2308 // Load 2 topmost expression stack values.
2309 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
2310 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
2311 __ mflr(R31);
2312 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
2313 __ mtlr(R31);
2314 __ pop(state);
2315
2316 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2317 __ bind(Lskip_vm_call);
2318 }
2319 __ blr();
2320 BLOCK_COMMENT("} trace_code");
2321 return entry;
2322 }
2323
2324 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2325 const Register addr = R11_scratch1,
2326 tmp = R12_scratch2;
2327 // Get index, shift out old bytecode, bring in new bytecode, and store it.
2328 // _index = (_index >> log2_number_of_codes) |
2329 // (bytecode << log2_number_of_codes);
2330 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
2331 __ lwz(tmp, offs1, addr);
2332 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
2333 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2334 __ stw(tmp, offs1, addr);
2335
2336 // Bump bucket contents.
2337 // _counters[_index] ++;
2338 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
2339 __ sldi(tmp, tmp, LogBytesPerInt);
2340 __ add(addr, tmp, addr);
2341 __ lwz(tmp, offs2, addr);
2342 __ addi(tmp, tmp, 1);
2343 __ stw(tmp, offs2, addr);
2344 }
2345
2346 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2347 // Call a little run-time stub to avoid blow-up for each bytecode.
2348 // The run-time runtime saves the right registers, depending on
2349 // the tosca in-state for the given template.
2350
2351 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2352 "entry must have been generated");
2353
2354 // Note: we destroy LR here.
2355 __ bl(Interpreter::trace_code(t->tos_in()));
2356 }
2357
2358 void TemplateInterpreterGenerator::stop_interpreter_at() {
2359 Label L;
2360 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
2361 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2362 __ ld(R11_scratch1, offs1, R11_scratch1);
2363 __ lwa(R12_scratch2, offs2, R12_scratch2);
2364 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2365 __ bne(CCR0, L);
2366 __ illtrap();
2367 __ bind(L);
2368 }
2369
2370 #endif // !PRODUCT