1 /*
2 * Copyright (c) 2014, 2025, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2025 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 "asm/macroAssembler.inline.hpp"
27 #include "classfile/javaClasses.hpp"
28 #include "compiler/disassembler.hpp"
29 #include "gc/shared/barrierSetAssembler.hpp"
30 #include "interpreter/bytecodeHistogram.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "interpreter/interp_masm.hpp"
34 #include "interpreter/templateInterpreterGenerator.hpp"
35 #include "interpreter/templateTable.hpp"
36 #include "oops/arrayOop.hpp"
37 #include "oops/method.hpp"
38 #include "oops/methodCounters.hpp"
39 #include "oops/methodData.hpp"
40 #include "oops/oop.inline.hpp"
41 #include "oops/resolvedIndyEntry.hpp"
42 #include "oops/resolvedMethodEntry.hpp"
43 #include "prims/jvmtiExport.hpp"
44 #include "prims/jvmtiThreadState.hpp"
45 #include "runtime/arguments.hpp"
46 #include "runtime/deoptimization.hpp"
47 #include "runtime/frame.inline.hpp"
48 #include "runtime/jniHandles.hpp"
49 #include "runtime/sharedRuntime.hpp"
50 #include "runtime/stubRoutines.hpp"
51 #include "runtime/synchronizer.hpp"
52 #include "runtime/timer.hpp"
53 #include "runtime/vframeArray.hpp"
54 #include "runtime/vm_version.hpp"
55 #include "utilities/debug.hpp"
56 #include "utilities/macros.hpp"
57
58 #undef __
59 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
60
61 // Size of interpreter code. Increase if too small. Interpreter will
62 // fail with a guarantee ("not enough space for interpreter generation");
63 // if too small.
64 // Run with +PrintInterpreter to get the VM to print out the size.
65 // Max size with JVMTI
66 int TemplateInterpreter::InterpreterCodeSize = 256*K;
67
68 #ifdef PRODUCT
69 #define BLOCK_COMMENT(str) /* nothing */
70 #else
71 #define BLOCK_COMMENT(str) __ block_comment(str)
72 #endif
73
74 #define BIND(label) __ bind(label); BLOCK_COMMENT(#label ":")
75
76 //-----------------------------------------------------------------------------
77
78 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
79 // Slow_signature handler that respects the PPC C calling conventions.
80 //
81 // We get called by the native entry code with our output register
82 // area == 8. First we call InterpreterRuntime::get_result_handler
83 // to copy the pointer to the signature string temporarily to the
84 // first C-argument and to return the result_handler in
85 // R3_RET. Since native_entry will copy the jni-pointer to the
86 // first C-argument slot later on, it is OK to occupy this slot
87 // temporarily. Then we copy the argument list on the java
88 // expression stack into native varargs format on the native stack
89 // and load arguments into argument registers. Integer arguments in
90 // the varargs vector will be sign-extended to 8 bytes.
91 //
92 // On entry:
93 // R3_ARG1 - intptr_t* Address of java argument list in memory.
94 // R15_prev_state - BytecodeInterpreter* Address of interpreter state for
95 // this method
96 // R19_method
97 //
98 // On exit (just before return instruction):
99 // R3_RET - contains the address of the result_handler.
100 // R4_ARG2 - is not updated for static methods and contains "this" otherwise.
101 // R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double,
102 // ARGi contains this argument. Otherwise, ARGi is not updated.
103 // F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double.
104
105 const int LogSizeOfTwoInstructions = 3;
106
107 // FIXME: use Argument:: GL: Argument names different numbers!
108 const int max_fp_register_arguments = 13;
109 const int max_int_register_arguments = 6; // first 2 are reserved
110
111 const Register arg_java = R21_tmp1;
112 const Register arg_c = R22_tmp2;
113 const Register signature = R23_tmp3; // is string
114 const Register sig_byte = R24_tmp4;
115 const Register fpcnt = R25_tmp5;
116 const Register argcnt = R26_tmp6;
117 const Register intSlot = R27_tmp7;
118 const Register target_sp = R28_tmp8;
119 const FloatRegister floatSlot = F0;
120
121 address entry = __ function_entry();
122
123 __ save_LR(R0);
124 __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
125 // We use target_sp for storing arguments in the C frame.
126 __ mr(target_sp, R1_SP);
127 __ push_frame_reg_args_nonvolatiles(0, R11_scratch1);
128
129 __ mr(arg_java, R3_ARG1);
130
131 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method);
132
133 // Signature is in R3_RET. Signature is callee saved.
134 __ mr(signature, R3_RET);
135
136 // Get the result handler.
137 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method);
138
139 {
140 Label L;
141 // test if static
142 // _access_flags._flags must be at offset 0.
143 // TODO PPC port: requires change in shared code.
144 //assert(in_bytes(AccessFlags::flags_offset()) == 0,
145 // "MethodDesc._access_flags == MethodDesc._access_flags._flags");
146 // _access_flags must be a 16 bit value.
147 assert(sizeof(AccessFlags) == 2, "wrong size");
148 __ lhz(R11_scratch1/*access_flags*/, method_(access_flags));
149 // testbit with condition register.
150 __ testbitdi(CR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT);
151 __ btrue(CR0, L);
152 // For non-static functions, pass "this" in R4_ARG2 and copy it
153 // to 2nd C-arg slot.
154 // We need to box the Java object here, so we use arg_java
155 // (address of current Java stack slot) as argument and don't
156 // dereference it as in case of ints, floats, etc.
157 __ mr(R4_ARG2, arg_java);
158 __ addi(arg_java, arg_java, -BytesPerWord);
159 __ std(R4_ARG2, _abi0(carg_2), target_sp);
160 __ bind(L);
161 }
162
163 // Will be incremented directly after loop_start. argcnt=0
164 // corresponds to 3rd C argument.
165 __ li(argcnt, -1);
166 // arg_c points to 3rd C argument
167 __ addi(arg_c, target_sp, _abi0(carg_3));
168 // no floating-point args parsed so far
169 __ li(fpcnt, 0);
170
171 Label move_intSlot_to_ARG, move_floatSlot_to_FARG;
172 Label loop_start, loop_end;
173 Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed;
174
175 // signature points to '(' at entry
176 #ifdef ASSERT
177 __ lbz(sig_byte, 0, signature);
178 __ cmplwi(CR0, sig_byte, '(');
179 __ bne(CR0, do_dontreachhere);
180 #endif
181
182 __ bind(loop_start);
183
184 __ addi(argcnt, argcnt, 1);
185 __ lbzu(sig_byte, 1, signature);
186
187 __ cmplwi(CR0, sig_byte, ')'); // end of signature
188 __ beq(CR0, loop_end);
189
190 __ cmplwi(CR0, sig_byte, 'B'); // byte
191 __ beq(CR0, do_int);
192
193 __ cmplwi(CR0, sig_byte, 'C'); // char
194 __ beq(CR0, do_int);
195
196 __ cmplwi(CR0, sig_byte, 'D'); // double
197 __ beq(CR0, do_double);
198
199 __ cmplwi(CR0, sig_byte, 'F'); // float
200 __ beq(CR0, do_float);
201
202 __ cmplwi(CR0, sig_byte, 'I'); // int
203 __ beq(CR0, do_int);
204
205 __ cmplwi(CR0, sig_byte, 'J'); // long
206 __ beq(CR0, do_long);
207
208 __ cmplwi(CR0, sig_byte, 'S'); // short
209 __ beq(CR0, do_int);
210
211 __ cmplwi(CR0, sig_byte, 'Z'); // boolean
212 __ beq(CR0, do_int);
213
214 __ cmplwi(CR0, sig_byte, 'L'); // object
215 __ beq(CR0, do_object);
216
217 __ cmplwi(CR0, sig_byte, '['); // array
218 __ beq(CR0, do_array);
219
220 // __ cmplwi(CR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type
221 // __ beq(CR0, do_void);
222
223 __ bind(do_dontreachhere);
224
225 __ unimplemented("ShouldNotReachHere in slow_signature_handler");
226
227 __ bind(do_array);
228
229 {
230 Label start_skip, end_skip;
231
232 __ bind(start_skip);
233 __ lbzu(sig_byte, 1, signature);
234 __ cmplwi(CR0, sig_byte, '[');
235 __ beq(CR0, start_skip); // skip further brackets
236 __ cmplwi(CR0, sig_byte, '9');
237 __ bgt(CR0, end_skip); // no optional size
238 __ cmplwi(CR0, sig_byte, '0');
239 __ bge(CR0, start_skip); // skip optional size
240 __ bind(end_skip);
241
242 __ cmplwi(CR0, sig_byte, 'L');
243 __ beq(CR0, do_object); // for arrays of objects, the name of the object must be skipped
244 __ b(do_boxed); // otherwise, go directly to do_boxed
245 }
246
247 __ bind(do_object);
248 {
249 Label L;
250 __ bind(L);
251 __ lbzu(sig_byte, 1, signature);
252 __ cmplwi(CR0, sig_byte, ';');
253 __ bne(CR0, L);
254 }
255 // Need to box the Java object here, so we use arg_java (address of
256 // current Java stack slot) as argument and don't dereference it as
257 // in case of ints, floats, etc.
258 Label do_null;
259 __ bind(do_boxed);
260 __ ld(R0,0, arg_java);
261 __ cmpdi(CR0, R0, 0);
262 __ li(intSlot,0);
263 __ beq(CR0, do_null);
264 __ mr(intSlot, arg_java);
265 __ bind(do_null);
266 __ std(intSlot, 0, arg_c);
267 __ addi(arg_java, arg_java, -BytesPerWord);
268 __ addi(arg_c, arg_c, BytesPerWord);
269 __ cmplwi(CR0, argcnt, max_int_register_arguments);
270 __ blt(CR0, move_intSlot_to_ARG);
271 __ b(loop_start);
272
273 __ bind(do_int);
274 __ lwa(intSlot, 0, arg_java);
275 __ std(intSlot, 0, arg_c);
276 __ addi(arg_java, arg_java, -BytesPerWord);
277 __ addi(arg_c, arg_c, BytesPerWord);
278 __ cmplwi(CR0, argcnt, max_int_register_arguments);
279 __ blt(CR0, move_intSlot_to_ARG);
280 __ b(loop_start);
281
282 __ bind(do_long);
283 __ ld(intSlot, -BytesPerWord, arg_java);
284 __ std(intSlot, 0, arg_c);
285 __ addi(arg_java, arg_java, - 2 * BytesPerWord);
286 __ addi(arg_c, arg_c, BytesPerWord);
287 __ cmplwi(CR0, argcnt, max_int_register_arguments);
288 __ blt(CR0, move_intSlot_to_ARG);
289 __ b(loop_start);
290
291 __ bind(do_float);
292 __ lfs(floatSlot, 0, arg_java);
293 __ stfs(floatSlot, Argument::float_on_stack_offset_in_bytes_c, arg_c);
294 __ addi(arg_java, arg_java, -BytesPerWord);
295 __ addi(arg_c, arg_c, BytesPerWord);
296 __ cmplwi(CR0, fpcnt, max_fp_register_arguments);
297 __ blt(CR0, move_floatSlot_to_FARG);
298 __ b(loop_start);
299
300 __ bind(do_double);
301 __ lfd(floatSlot, - BytesPerWord, arg_java);
302 __ stfd(floatSlot, 0, arg_c);
303 __ addi(arg_java, arg_java, - 2 * BytesPerWord);
304 __ addi(arg_c, arg_c, BytesPerWord);
305 __ cmplwi(CR0, fpcnt, max_fp_register_arguments);
306 __ blt(CR0, move_floatSlot_to_FARG);
307 __ b(loop_start);
308
309 __ bind(loop_end);
310
311 __ pop_frame();
312 __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
313 __ restore_LR(R0);
314
315 __ blr();
316
317 Label move_int_arg, move_float_arg;
318 __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
319 __ mr(R5_ARG3, intSlot); __ b(loop_start);
320 __ mr(R6_ARG4, intSlot); __ b(loop_start);
321 __ mr(R7_ARG5, intSlot); __ b(loop_start);
322 __ mr(R8_ARG6, intSlot); __ b(loop_start);
323 __ mr(R9_ARG7, intSlot); __ b(loop_start);
324 __ mr(R10_ARG8, intSlot); __ b(loop_start);
325
326 __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
327 __ fmr(F1_ARG1, floatSlot); __ b(loop_start);
328 __ fmr(F2_ARG2, floatSlot); __ b(loop_start);
329 __ fmr(F3_ARG3, floatSlot); __ b(loop_start);
330 __ fmr(F4_ARG4, floatSlot); __ b(loop_start);
331 __ fmr(F5_ARG5, floatSlot); __ b(loop_start);
332 __ fmr(F6_ARG6, floatSlot); __ b(loop_start);
333 __ fmr(F7_ARG7, floatSlot); __ b(loop_start);
334 __ fmr(F8_ARG8, floatSlot); __ b(loop_start);
335 __ fmr(F9_ARG9, floatSlot); __ b(loop_start);
336 __ fmr(F10_ARG10, floatSlot); __ b(loop_start);
337 __ fmr(F11_ARG11, floatSlot); __ b(loop_start);
338 __ fmr(F12_ARG12, floatSlot); __ b(loop_start);
339 __ fmr(F13_ARG13, floatSlot); __ b(loop_start);
340
341 __ bind(move_intSlot_to_ARG);
342 __ sldi(R0, argcnt, LogSizeOfTwoInstructions);
343 __ load_const(R11_scratch1, move_int_arg); // Label must be bound here.
344 __ add(R11_scratch1, R0, R11_scratch1);
345 __ mtctr(R11_scratch1/*branch_target*/);
346 __ bctr();
347 __ bind(move_floatSlot_to_FARG);
348 __ sldi(R0, fpcnt, LogSizeOfTwoInstructions);
349 __ addi(fpcnt, fpcnt, 1);
350 __ load_const(R11_scratch1, move_float_arg); // Label must be bound here.
351 __ add(R11_scratch1, R0, R11_scratch1);
352 __ mtctr(R11_scratch1/*branch_target*/);
353 __ bctr();
354
355 return entry;
356 }
357
358 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
359 //
360 // Registers alive
361 // R3_RET
362 // LR
363 //
364 // Registers updated
365 // R3_RET
366 //
367
368 Label done;
369 address entry = __ pc();
370
371 switch (type) {
372 case T_BOOLEAN:
373 // convert !=0 to 1
374 __ normalize_bool(R3_RET);
375 break;
376 case T_BYTE:
377 // sign extend 8 bits
378 __ extsb(R3_RET, R3_RET);
379 break;
380 case T_CHAR:
381 // zero extend 16 bits
382 __ clrldi(R3_RET, R3_RET, 48);
383 break;
384 case T_SHORT:
385 // sign extend 16 bits
386 __ extsh(R3_RET, R3_RET);
387 break;
388 case T_INT:
389 // sign extend 32 bits
390 __ extsw(R3_RET, R3_RET);
391 break;
392 case T_LONG:
393 break;
394 case T_OBJECT:
395 // JNIHandles::resolve result.
396 __ resolve_jobject(R3_RET, R11_scratch1, R31, MacroAssembler::PRESERVATION_FRAME_LR); // kills R31
397 break;
398 case T_FLOAT:
399 break;
400 case T_DOUBLE:
401 break;
402 case T_VOID:
403 break;
404 default: ShouldNotReachHere();
405 }
406
407 BIND(done);
408 __ blr();
409
410 return entry;
411 }
412
413 // Abstract method entry.
414 //
415 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
416 address entry = __ pc();
417
418 //
419 // Registers alive
420 // R16_thread - JavaThread*
421 // R19_method - callee's method (method to be invoked)
422 // R1_SP - SP prepared such that caller's outgoing args are near top
423 // LR - return address to caller
424 //
425 // Stack layout at this point:
426 //
427 // 0 [TOP_IJAVA_FRAME_ABI] <-- R1_SP
428 // alignment (optional)
429 // [outgoing Java arguments]
430 // ...
431 // PARENT [PARENT_IJAVA_FRAME_ABI]
432 // ...
433 //
434
435 // Can't use call_VM here because we have not set up a new
436 // interpreter state. Make the call to the vm and make it look like
437 // our caller set up the JavaFrameAnchor.
438 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
439
440 // Push a new C frame and save LR.
441 __ save_LR(R0);
442 __ push_frame_reg_args(0, R11_scratch1);
443
444 // This is not a leaf but we have a JavaFrameAnchor now and we will
445 // check (create) exceptions afterward so this is ok.
446 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
447 R16_thread, R19_method);
448
449 // Pop the C frame and restore LR.
450 __ pop_frame();
451 __ restore_LR(R0);
452
453 // Reset JavaFrameAnchor from call_VM_leaf above.
454 __ reset_last_Java_frame();
455
456 // We don't know our caller, so jump to the general forward exception stub,
457 // which will also pop our full frame off. Satisfy the interface of
458 // SharedRuntime::generate_forward_exception()
459 __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0);
460 __ mtctr(R11_scratch1);
461 __ bctr();
462
463 return entry;
464 }
465
466 // Interpreter intrinsic for WeakReference.get().
467 // 1. Don't push a full blown frame and go on dispatching, but fetch the value
468 // into R8 and return quickly
469 // 2. If G1 is active we *must* execute this intrinsic for corrrectness:
470 // It contains a GC barrier which puts the reference into the satb buffer
471 // to indicate that someone holds a strong reference to the object the
472 // weak ref points to!
473 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
474 // Code: _aload_0, _getfield, _areturn
475 // parameter size = 1
476 //
477 // The code that gets generated by this routine is split into 2 parts:
478 // 1. the "intrinsified" code for G1 (or any SATB based GC),
479 // 2. the slow path - which is an expansion of the regular method entry.
480 //
481 // Notes:
482 // * In the G1 code we do not check whether we need to block for
483 // a safepoint. If G1 is enabled then we must execute the specialized
484 // code for Reference.get (except when the Reference object is null)
485 // so that we can log the value in the referent field with an SATB
486 // update buffer.
487 // If the code for the getfield template is modified so that the
488 // G1 pre-barrier code is executed when the current method is
489 // Reference.get() then going through the normal method entry
490 // will be fine.
491 // * The G1 code can, however, check the receiver object (the instance
492 // of java.lang.Reference) and jump to the slow path if null. If the
493 // Reference object is null then we obviously cannot fetch the referent
494 // and so we don't need to call the G1 pre-barrier. Thus we can use the
495 // regular method entry code to generate the NPE.
496 //
497
498 address entry = __ pc();
499
500 const int referent_offset = java_lang_ref_Reference::referent_offset();
501
502 Label slow_path;
503
504 // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH);
505
506 // In the G1 code we don't check if we need to reach a safepoint. We
507 // continue and the thread will safepoint at the next bytecode dispatch.
508
509 // If the receiver is null then it is OK to jump to the slow path.
510 __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver
511
512 // Check if receiver == nullptr and go the slow path.
513 __ cmpdi(CR0, R3_RET, 0);
514 __ beq(CR0, slow_path);
515
516 __ load_heap_oop(R3_RET, referent_offset, R3_RET,
517 /* non-volatile temp */ R31, R11_scratch1,
518 MacroAssembler::PRESERVATION_FRAME_LR,
519 ON_WEAK_OOP_REF);
520
521 // Generate the G1 pre-barrier code to log the value of
522 // the referent field in an SATB buffer. Note with
523 // these parameters the pre-barrier does not generate
524 // the load of the previous value.
525
526 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
527 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
528
529 __ blr();
530
531 __ bind(slow_path);
532 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
533 return entry;
534 }
535
536 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
537 address entry = __ pc();
538
539 // Expression stack must be empty before entering the VM if an
540 // exception happened.
541 __ empty_expression_stack();
542 // Throw exception.
543 __ call_VM(noreg,
544 CAST_FROM_FN_PTR(address,
545 InterpreterRuntime::throw_StackOverflowError));
546 return entry;
547 }
548
549 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
550 address entry = __ pc();
551 __ empty_expression_stack();
552 // R4_ARG2 already contains the array.
553 // Index is in R17_tos.
554 __ mr(R5_ARG3, R17_tos);
555 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R4_ARG2, R5_ARG3);
556 return entry;
557 }
558
559 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
560 address entry = __ pc();
561 // Expression stack must be empty before entering the VM if an
562 // exception happened.
563 __ empty_expression_stack();
564
565 // Load exception object.
566 // Thread will be loaded to R3_ARG1.
567 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
568 #ifdef ASSERT
569 // Above call must not return here since exception pending.
570 __ should_not_reach_here();
571 #endif
572 return entry;
573 }
574
575 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
576 address entry = __ pc();
577 //__ untested("generate_exception_handler_common");
578 Register Rexception = R17_tos;
579
580 // Expression stack must be empty before entering the VM if an exception happened.
581 __ empty_expression_stack();
582
583 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
584 if (pass_oop) {
585 __ mr(R5_ARG3, Rexception);
586 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception));
587 } else {
588 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
589 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception));
590 }
591
592 // Throw exception.
593 __ mr(R3_ARG1, Rexception);
594 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
595 __ mtctr(R11_scratch1);
596 __ bctr();
597
598 return entry;
599 }
600
601 // This entry is returned to when a call returns to the interpreter.
602 // When we arrive here, we expect that the callee stack frame is already popped.
603 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
604 address entry = __ pc();
605
606 // Move the value out of the return register back to the TOS cache of current frame.
607 switch (state) {
608 case ltos:
609 case btos:
610 case ztos:
611 case ctos:
612 case stos:
613 case atos:
614 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache
615 case ftos:
616 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
617 case vtos: break; // Nothing to do, this was a void return.
618 default : ShouldNotReachHere();
619 }
620
621 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
622
623 // Compiled code destroys templateTableBase, reload.
624 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
625
626 if (state == atos) {
627 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
628 }
629
630 const Register cache = R11_scratch1;
631 const Register size = R12_scratch2;
632 if (index_size == sizeof(u4)) {
633 __ load_resolved_indy_entry(cache, size /* tmp */);
634 __ lhz(size, in_bytes(ResolvedIndyEntry::num_parameters_offset()), cache);
635 } else {
636 assert(index_size == sizeof(u2), "Can only be u2");
637 __ load_method_entry(cache, size /* tmp */);
638 __ lhz(size, in_bytes(ResolvedMethodEntry::num_parameters_offset()), cache);
639 }
640 __ sldi(size, size, Interpreter::logStackElementSize);
641 __ add(R15_esp, R15_esp, size);
642
643 __ check_and_handle_popframe(R11_scratch1);
644 __ check_and_handle_earlyret(R11_scratch1);
645
646 __ dispatch_next(state, step);
647 return entry;
648 }
649
650 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
651 address entry = __ pc();
652 // If state != vtos, we're returning from a native method, which put it's result
653 // into the result register. So move the value out of the return register back
654 // to the TOS cache of current frame.
655
656 switch (state) {
657 case ltos:
658 case btos:
659 case ztos:
660 case ctos:
661 case stos:
662 case atos:
663 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache
664 case ftos:
665 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
666 case vtos: break; // Nothing to do, this was a void return.
667 default : ShouldNotReachHere();
668 }
669
670 // Load LcpoolCache @@@ should be already set!
671 __ get_constant_pool_cache(R27_constPoolCache);
672
673 // Handle a pending exception, fall through if none.
674 __ check_and_forward_exception(R11_scratch1, R12_scratch2);
675
676 // Start executing bytecodes.
677 if (continuation == nullptr) {
678 __ dispatch_next(state, step);
679 } else {
680 __ jump_to_entry(continuation, R11_scratch1);
681 }
682
683 return entry;
684 }
685
686 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
687 address entry = __ pc();
688
689 __ push(state);
690 __ push_cont_fastpath();
691 __ call_VM(noreg, runtime_entry);
692 __ pop_cont_fastpath();
693 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
694
695 return entry;
696 }
697
698 address TemplateInterpreterGenerator::generate_cont_resume_interpreter_adapter() {
699 if (!Continuations::enabled()) return nullptr;
700 address start = __ pc();
701
702 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
703 __ restore_interpreter_state(R11_scratch1, false, true /*restore_top_frame_sp*/);
704 __ blr();
705
706 return start;
707 }
708
709 // Helpers for commoning out cases in the various type of method entries.
710
711 // Increment invocation count & check for overflow.
712 //
713 // Note: checking for negative value instead of overflow
714 // so we have a 'sticky' overflow test.
715 //
716 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
717 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
718 Register Rscratch1 = R11_scratch1;
719 Register Rscratch2 = R12_scratch2;
720 Register R3_counters = R3_ARG1;
721 Label done;
722
723 const int increment = InvocationCounter::count_increment;
724 Label no_mdo;
725 if (ProfileInterpreter) {
726 const Register Rmdo = R3_counters;
727 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
728 __ cmpdi(CR0, Rmdo, 0);
729 __ beq(CR0, no_mdo);
730
731 // Increment invocation counter in the MDO.
732 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
733 __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
734 __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo);
735 __ addi(Rscratch2, Rscratch2, increment);
736 __ stw(Rscratch2, mdo_ic_offs, Rmdo);
737 __ and_(Rscratch1, Rscratch2, Rscratch1);
738 __ bne(CR0, done);
739 __ b(*overflow);
740 }
741
742 // Increment counter in MethodCounters*.
743 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
744 __ bind(no_mdo);
745 __ get_method_counters(R19_method, R3_counters, done);
746 __ lwz(Rscratch2, mo_ic_offs, R3_counters);
747 __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters);
748 __ addi(Rscratch2, Rscratch2, increment);
749 __ stw(Rscratch2, mo_ic_offs, R3_counters);
750 __ and_(Rscratch1, Rscratch2, Rscratch1);
751 __ beq(CR0, *overflow);
752
753 __ bind(done);
754 }
755
756 // Generate code to initiate compilation on invocation counter overflow.
757 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
758 // Generate code to initiate compilation on the counter overflow.
759
760 // InterpreterRuntime::frequency_counter_overflow takes one arguments,
761 // which indicates if the counter overflow occurs at a backwards branch (null bcp)
762 // We pass zero in.
763 // The call returns the address of the verified entry point for the method or null
764 // if the compilation did not complete (either went background or bailed out).
765 //
766 // Unlike the C++ interpreter above: Check exceptions!
767 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
768 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
769
770 __ li(R4_ARG2, 0);
771 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
772
773 // Returns verified_entry_point or null.
774 // We ignore it in any case.
775 __ b(continue_entry);
776 }
777
778 // See if we've got enough room on the stack for locals plus overhead below
779 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
780 // without going through the signal handler, i.e., reserved and yellow zones
781 // will not be made usable. The shadow zone must suffice to handle the
782 // overflow.
783 //
784 // Kills Rmem_frame_size, Rscratch1.
785 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
786 Label done;
787 assert_different_registers(Rmem_frame_size, Rscratch1);
788
789 BLOCK_COMMENT("stack_overflow_check_with_compare {");
790 __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
791 __ ld(Rscratch1, thread_(stack_overflow_limit));
792 __ cmpld(CR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
793 __ bgt(CR0/*is_stack_overflow*/, done);
794
795 // The stack overflows. Load target address of the runtime stub and call it.
796 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "generated in wrong order");
797 __ load_const_optimized(Rscratch1, (SharedRuntime::throw_StackOverflowError_entry()), R0);
798 __ mtctr(Rscratch1);
799 // Restore caller_sp (c2i adapter may exist, but no shrinking of interpreted caller frame).
800 #ifdef ASSERT
801 Label frame_not_shrunk;
802 __ cmpld(CR0, R1_SP, R21_sender_SP);
803 __ ble(CR0, frame_not_shrunk);
804 __ stop("frame shrunk");
805 __ bind(frame_not_shrunk);
806 __ ld(Rscratch1, 0, R1_SP);
807 __ ld(R0, 0, R21_sender_SP);
808 __ cmpd(CR0, R0, Rscratch1);
809 __ asm_assert_eq("backlink");
810 #endif // ASSERT
811 __ mr(R1_SP, R21_sender_SP);
812 __ bctr();
813
814 __ align(32, 12);
815 __ bind(done);
816 BLOCK_COMMENT("} stack_overflow_check_with_compare");
817 }
818
819 // Lock the current method, interpreter register window must be set up!
820 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
821 const Register Robj_to_lock = Rscratch2;
822
823 {
824 if (!flags_preloaded) {
825 __ lhz(Rflags, method_(access_flags));
826 }
827
828 #ifdef ASSERT
829 // Check if methods needs synchronization.
830 {
831 Label Lok;
832 __ testbitdi(CR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
833 __ btrue(CR0,Lok);
834 __ stop("method doesn't need synchronization");
835 __ bind(Lok);
836 }
837 #endif // ASSERT
838 }
839
840 // Get synchronization object to Rscratch2.
841 {
842 Label Lstatic;
843 Label Ldone;
844
845 __ testbitdi(CR0, R0, Rflags, JVM_ACC_STATIC_BIT);
846 __ btrue(CR0, Lstatic);
847
848 // Non-static case: load receiver obj from stack and we're done.
849 __ ld(Robj_to_lock, R18_locals);
850 __ b(Ldone);
851
852 __ bind(Lstatic); // Static case: Lock the java mirror
853 // Load mirror from interpreter frame.
854 __ ld(Robj_to_lock, _abi0(callers_sp), R1_SP);
855 __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock);
856
857 __ bind(Ldone);
858 __ verify_oop(Robj_to_lock);
859 }
860
861 // Got the oop to lock => execute!
862 __ add_monitor_to_stack(true, Rscratch1, R0);
863
864 __ std(Robj_to_lock, in_bytes(BasicObjectLock::obj_offset()), R26_monitor);
865 __ lock_object(R26_monitor, Robj_to_lock);
866 }
867
868 // Generate a fixed interpreter frame for pure interpreter
869 // and I2N native transition frames.
870 //
871 // Before (stack grows downwards):
872 //
873 // | ... |
874 // |------------- |
875 // | java arg0 |
876 // | ... |
877 // | java argn |
878 // | | <- R15_esp
879 // | |
880 // |--------------|
881 // | abi_112 |
882 // | | <- R1_SP
883 // |==============|
884 //
885 //
886 // After:
887 //
888 // | ... |
889 // | java arg0 |<- R18_locals
890 // | ... |
891 // | java argn |
892 // |--------------|
893 // | |
894 // | java locals |
895 // | |
896 // |--------------|
897 // | abi_48 |
898 // |==============|
899 // | |
900 // | istate |
901 // | |
902 // |--------------|
903 // | monitor |<- R26_monitor
904 // |--------------|
905 // | |<- R15_esp
906 // | expression |
907 // | stack |
908 // | |
909 // |--------------|
910 // | |
911 // | abi_112 |<- R1_SP
912 // |==============|
913 //
914 // The top most frame needs an abi space of 112 bytes. This space is needed,
915 // since we call to c. The c function may spill their arguments to the caller
916 // frame. When we call to java, we don't need these spill slots. In order to save
917 // space on the stack, we resize the caller. However, java locals reside in
918 // the caller frame and the frame has to be increased. The frame_size for the
919 // current frame was calculated based on max_stack as size for the expression
920 // stack. At the call, just a part of the expression stack might be used.
921 // We don't want to waste this space and cut the frame back accordingly.
922 // The resulting amount for resizing is calculated as follows:
923 // resize = (number_of_locals - number_of_arguments) * slot_size
924 // + (R1_SP - R15_esp) + 48
925 //
926 // The size for the callee frame is calculated:
927 // framesize = 112 + max_stack + monitor + state_size
928 //
929 // maxstack: Max number of slots on the expression stack, loaded from the method.
930 // monitor: We statically reserve room for one monitor object.
931 // state_size: We save the current state of the interpreter to this area.
932 //
933 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
934 Register Rparent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
935 Rtop_frame_size = R7_ARG5,
936 Rconst_method = R8_ARG6,
937 Rconst_pool = R9_ARG7,
938 Rmirror = R10_ARG8;
939
940 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, Rparent_frame_resize, Rtop_frame_size,
941 Rconst_method, Rconst_pool);
942
943 __ ld(Rconst_method, method_(const));
944 __ lhz(Rsize_of_parameters /* number of params */,
945 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
946 if (native_call) {
947 // If we're calling a native method, we reserve space for the worst-case signature
948 // handler varargs vector, which is max(Argument::n_int_register_parameters_c, parameter_count+2).
949 // We add two slots to the parameter_count, one for the jni
950 // environment and one for a possible native mirror.
951 Label skip_native_calculate_max_stack;
952 __ addi(Rtop_frame_size, Rsize_of_parameters, 2);
953 __ cmpwi(CR0, Rtop_frame_size, Argument::n_int_register_parameters_c);
954 __ bge(CR0, skip_native_calculate_max_stack);
955 __ li(Rtop_frame_size, Argument::n_int_register_parameters_c);
956 __ bind(skip_native_calculate_max_stack);
957 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
958 __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
959 __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
960 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
961 } else {
962 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
963 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
964 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
965 __ lhz(Rtop_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
966 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
967 __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
968 __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
969 __ add(Rparent_frame_resize, Rparent_frame_resize, R11_scratch1);
970 }
971
972 // Compute top frame size.
973 __ addi(Rtop_frame_size, Rtop_frame_size, frame::top_ijava_frame_abi_size + frame::ijava_state_size);
974
975 // Cut back area between esp and max_stack.
976 __ addi(Rparent_frame_resize, Rparent_frame_resize, frame::parent_ijava_frame_abi_size - Interpreter::stackElementSize);
977
978 __ round_to(Rtop_frame_size, frame::alignment_in_bytes);
979 __ round_to(Rparent_frame_resize, frame::alignment_in_bytes);
980 // Rparent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
981 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
982
983 if (!native_call) {
984 // Stack overflow check.
985 // Native calls don't need the stack size check since they have no
986 // expression stack and the arguments are already on the stack and
987 // we only add a handful of words to the stack.
988 __ add(R11_scratch1, Rparent_frame_resize, Rtop_frame_size);
989 generate_stack_overflow_check(R11_scratch1, R12_scratch2);
990 }
991
992 // Set up interpreter state registers.
993
994 __ add(R18_locals, R15_esp, Rsize_of_parameters);
995 __ ld(Rconst_pool, in_bytes(ConstMethod::constants_offset()), Rconst_method);
996 __ ld(R27_constPoolCache, ConstantPool::cache_offset(), Rconst_pool);
997
998 // Set method data pointer.
999 if (ProfileInterpreter) {
1000 Label zero_continue;
1001 __ ld(R28_mdx, method_(method_data));
1002 __ cmpdi(CR0, R28_mdx, 0);
1003 __ beq(CR0, zero_continue);
1004 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1005 __ bind(zero_continue);
1006 }
1007
1008 if (native_call) {
1009 __ li(R14_bcp, 0); // Must initialize.
1010 } else {
1011 __ addi(R14_bcp, Rconst_method, in_bytes(ConstMethod::codes_offset()));
1012 }
1013
1014 // Resize parent frame.
1015 __ mflr(R12_scratch2);
1016 __ neg(Rparent_frame_resize, Rparent_frame_resize);
1017 __ resize_frame(Rparent_frame_resize, R11_scratch1);
1018 __ std(R12_scratch2, _abi0(lr), R1_SP);
1019
1020 // Get mirror and store it in the frame as GC root for this Method*.
1021 __ ld(Rmirror, ConstantPool::pool_holder_offset(), Rconst_pool);
1022 __ ld(Rmirror, in_bytes(Klass::java_mirror_offset()), Rmirror);
1023 __ resolve_oop_handle(Rmirror, R11_scratch1, R12_scratch2, MacroAssembler::PRESERVATION_FRAME_LR_GP_REGS);
1024
1025 __ addi(R26_monitor, R1_SP, -frame::ijava_state_size);
1026 __ addi(R15_esp, R26_monitor, -Interpreter::stackElementSize);
1027
1028 // Store values.
1029 __ std(R19_method, _ijava_state_neg(method), R1_SP);
1030 __ std(Rmirror, _ijava_state_neg(mirror), R1_SP);
1031 __ sub(R12_scratch2, R18_locals, R1_SP);
1032 __ srdi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1033 // Store relativized R18_locals, see frame::interpreter_frame_locals().
1034 __ std(R12_scratch2, _ijava_state_neg(locals), R1_SP);
1035 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
1036
1037 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
1038 // be found in the frame after save_interpreter_state is done. This is always true
1039 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
1040 // because e.g. frame::interpreter_frame_bcp() will not access the correct value
1041 // (Enhanced Stack Trace).
1042 // The signal handler does not save the interpreter state into the frame.
1043
1044 // We have to initialize some of these frame slots for native calls (accessed by GC).
1045 // Also initialize them for non-native calls for better tool support (even though
1046 // you may not get the most recent version as described above).
1047 __ li(R0, 0);
1048 __ li(R12_scratch2, -(frame::ijava_state_size / wordSize));
1049 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
1050 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
1051 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
1052 __ sub(R12_scratch2, R15_esp, R1_SP);
1053 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1054 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
1055 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); // only used for native_call
1056
1057 // Store sender's SP and this frame's top SP.
1058 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
1059 __ neg(R12_scratch2, Rtop_frame_size);
1060 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1061 // Store relativized top_frame_sp
1062 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
1063
1064 // Push top frame.
1065 __ push_frame(Rtop_frame_size, R11_scratch1);
1066 }
1067
1068 // End of helpers
1069
1070 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1071
1072 // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1073 bool use_instruction = false;
1074 address runtime_entry = nullptr;
1075 int num_args = 1;
1076 bool double_precision = true;
1077
1078 // PPC64 specific:
1079 switch (kind) {
1080 case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break;
1081 case Interpreter::java_lang_math_abs: use_instruction = true; break;
1082 case Interpreter::java_lang_math_fmaF:
1083 case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1084 default: break; // Fall back to runtime call.
1085 }
1086
1087 switch (kind) {
1088 case Interpreter::java_lang_math_sin : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); break;
1089 case Interpreter::java_lang_math_cos : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); break;
1090 case Interpreter::java_lang_math_tan : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); break;
1091 case Interpreter::java_lang_math_tanh : /* run interpreted */ break;
1092 case Interpreter::java_lang_math_abs : /* run interpreted */ break;
1093 case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); break;
1094 case Interpreter::java_lang_math_log : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); break;
1095 case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1096 case Interpreter::java_lang_math_pow : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1097 case Interpreter::java_lang_math_exp : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); break;
1098 case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1099 case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1100 default: ShouldNotReachHere();
1101 }
1102
1103 // Use normal entry if neither instruction nor runtime call is used.
1104 if (!use_instruction && runtime_entry == nullptr) return nullptr;
1105
1106 address entry = __ pc();
1107
1108 // Load arguments
1109 assert(num_args <= 13, "passed in registers");
1110 if (double_precision) {
1111 int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1112 for (int i = 0; i < num_args; ++i) {
1113 __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1114 offset -= 2 * Interpreter::stackElementSize;
1115 }
1116 } else {
1117 int offset = num_args * Interpreter::stackElementSize;
1118 for (int i = 0; i < num_args; ++i) {
1119 __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1120 offset -= Interpreter::stackElementSize;
1121 }
1122 }
1123
1124 if (use_instruction) {
1125 switch (kind) {
1126 case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1); break;
1127 case Interpreter::java_lang_math_abs: __ fabs(F1_RET, F1); break;
1128 case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break;
1129 case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3); break;
1130 default: ShouldNotReachHere();
1131 }
1132 } else {
1133 // Comment: Can use tail call if the unextended frame is always C ABI compliant:
1134 //__ load_const_optimized(R12_scratch2, runtime_entry, R0);
1135 //__ call_c_and_return_to_caller(R12_scratch2);
1136
1137 // Push a new C frame and save LR.
1138 __ save_LR(R0);
1139 __ push_frame_reg_args(0, R11_scratch1);
1140
1141 __ call_VM_leaf(runtime_entry);
1142
1143 // Pop the C frame and restore LR.
1144 __ pop_frame();
1145 __ restore_LR(R0);
1146 }
1147
1148 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1149 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1150 __ blr();
1151
1152 __ flush();
1153
1154 return entry;
1155 }
1156
1157 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() {
1158 if (!VM_Version::supports_float16()) return nullptr;
1159
1160 address entry = __ pc();
1161
1162 __ lfs(F1, Interpreter::stackElementSize, R15_esp);
1163 __ f2hf(R3_RET, F1, F0);
1164
1165 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1166 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1167 __ blr();
1168
1169 __ flush();
1170
1171 return entry;
1172 }
1173
1174 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() {
1175 if (!VM_Version::supports_float16()) return nullptr;
1176
1177 address entry = __ pc();
1178
1179 // Note: Could also use:
1180 //__ li(R3, Interpreter::stackElementSize);
1181 //__ lfiwax(F1_RET, R15_esp, R3); // short stored as 32 bit integer
1182 //__ xscvhpdp(F1_RET->to_vsr(), F1_RET->to_vsr());
1183 __ lwa(R3, Interpreter::stackElementSize, R15_esp);
1184 __ hf2f(F1_RET, R3);
1185
1186 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1187 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1188 __ blr();
1189
1190 __ flush();
1191
1192 return entry;
1193 }
1194
1195 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1196 // Quick & dirty stack overflow checking: bang the stack & handle trap.
1197 // Note that we do the banging after the frame is setup, since the exception
1198 // handling code expects to find a valid interpreter frame on the stack.
1199 // Doing the banging earlier fails if the caller frame is not an interpreter
1200 // frame.
1201 // (Also, the exception throwing code expects to unlock any synchronized
1202 // method receiever, so do the banging after locking the receiver.)
1203
1204 // Bang each page in the shadow zone. We can't assume it's been done for
1205 // an interpreter frame with greater than a page of locals, so each page
1206 // needs to be checked. Only true for non-native.
1207 const size_t page_size = os::vm_page_size();
1208 const int n_shadow_pages = StackOverflow::stack_shadow_zone_size() / page_size;
1209 const int start_page = native_call ? n_shadow_pages : 1;
1210 BLOCK_COMMENT("bang_stack_shadow_pages:");
1211 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1212 __ bang_stack_with_offset(pages*page_size);
1213 }
1214 }
1215
1216 // Interpreter stub for calling a native method. (asm interpreter)
1217 // This sets up a somewhat different looking stack for calling the
1218 // native method than the typical interpreter frame setup.
1219 //
1220 // On entry:
1221 // R19_method - method
1222 // R16_thread - JavaThread*
1223 // R15_esp - intptr_t* sender tos
1224 //
1225 // abstract stack (grows up)
1226 // [ IJava (caller of JNI callee) ] <-- ASP
1227 // ...
1228 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1229
1230 address entry = __ pc();
1231
1232 const bool inc_counter = UseCompiler || CountCompiledCalls;
1233
1234 // -----------------------------------------------------------------------------
1235 // Allocate a new frame that represents the native callee (i2n frame).
1236 // This is not a full-blown interpreter frame, but in particular, the
1237 // following registers are valid after this:
1238 // - R19_method
1239 // - R18_local (points to start of arguments to native function)
1240 //
1241 // abstract stack (grows up)
1242 // [ IJava (caller of JNI callee) ] <-- ASP
1243 // ...
1244
1245 const Register signature_handler_fd = R11_scratch1;
1246 const Register pending_exception = R0;
1247 const Register result_handler_addr = R31;
1248 const Register native_method_fd = R12_scratch2; // preferred in MacroAssembler::branch_to
1249 const Register access_flags = R22_tmp2;
1250 const Register active_handles = R11_scratch1; // R26_monitor saved to state.
1251 const Register sync_state = R12_scratch2;
1252 const Register sync_state_addr = sync_state; // Address is dead after use.
1253 const Register suspend_flags = R11_scratch1;
1254
1255 //=============================================================================
1256 // Allocate new frame and initialize interpreter state.
1257
1258 Label exception_return;
1259 Label exception_return_sync_check;
1260 Label stack_overflow_return;
1261
1262 Register size_of_parameters = R22_tmp2;
1263
1264 generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
1265
1266 //=============================================================================
1267 // Increment invocation counter. On overflow, entry to JNI method
1268 // will be compiled.
1269 Label invocation_counter_overflow, continue_after_compile;
1270 if (inc_counter) {
1271 if (synchronized) {
1272 // Since at this point in the method invocation the exception handler
1273 // would try to exit the monitor of synchronized methods which hasn't
1274 // been entered yet, we set the thread local variable
1275 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1276 // runtime, exception handling i.e. unlock_if_synchronized_method will
1277 // check this thread local flag.
1278 // This flag has two effects, one is to force an unwind in the topmost
1279 // interpreter frame and not perform an unlock while doing so.
1280 __ li(R0, 1);
1281 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1282 }
1283 generate_counter_incr(&invocation_counter_overflow);
1284
1285 BIND(continue_after_compile);
1286 }
1287
1288 bang_stack_shadow_pages(true);
1289
1290 if (inc_counter) {
1291 // Reset the _do_not_unlock_if_synchronized flag.
1292 if (synchronized) {
1293 __ li(R0, 0);
1294 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1295 }
1296 }
1297
1298 // access_flags = method->access_flags();
1299 // Load access flags.
1300 assert(__ nonvolatile_accross_vthread_preemtion(access_flags),
1301 "access_flags not preserved");
1302 // Type check.
1303 assert(2 == sizeof(AccessFlags), "unexpected field size");
1304 __ lhz(access_flags, method_(access_flags));
1305
1306 // We don't want to reload R19_method and access_flags after calls
1307 // to some helper functions.
1308 assert(R19_method->is_nonvolatile(),
1309 "R19_method must be a non-volatile register");
1310
1311 // Check for synchronized methods. Must happen AFTER invocation counter
1312 // check, so method is not locked if counter overflows.
1313
1314 if (synchronized) {
1315 lock_method(access_flags, R11_scratch1, R12_scratch2, true);
1316
1317 // Update monitor in state.
1318 __ ld(R11_scratch1, 0, R1_SP);
1319 __ sub(R12_scratch2, R26_monitor, R11_scratch1);
1320 __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1321 __ std(R12_scratch2, _ijava_state_neg(monitors), R11_scratch1);
1322 }
1323
1324 // jvmti/jvmpi support
1325 __ notify_method_entry();
1326
1327 //=============================================================================
1328 // Get and call the signature handler.
1329
1330 __ ld(signature_handler_fd, method_(signature_handler));
1331 Label call_signature_handler;
1332
1333 __ cmpdi(CR0, signature_handler_fd, 0);
1334 __ bne(CR0, call_signature_handler);
1335
1336 // Method has never been called. Either generate a specialized
1337 // handler or point to the slow one.
1338 //
1339 // Pass parameter 'false' to avoid exception check in call_VM.
1340 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
1341
1342 // Check for an exception while looking up the target method. If we
1343 // incurred one, bail.
1344 __ ld(pending_exception, thread_(pending_exception));
1345 __ cmpdi(CR0, pending_exception, 0);
1346 __ bne(CR0, exception_return_sync_check); // Has pending exception.
1347
1348 // Reload signature handler, it may have been created/assigned in the meanwhile.
1349 __ ld(signature_handler_fd, method_(signature_handler));
1350 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
1351
1352 BIND(call_signature_handler);
1353
1354 // Before we call the signature handler we push a new frame to
1355 // protect the interpreter frame volatile registers when we return
1356 // from jni but before we can get back to Java.
1357
1358 // First set the frame anchor while the SP/FP registers are
1359 // convenient and the slow signature handler can use this same frame
1360 // anchor.
1361
1362 bool support_vthread_preemption = Continuations::enabled() && LockingMode != LM_LEGACY;
1363
1364 // We have a TOP_IJAVA_FRAME here, which belongs to us.
1365 Label last_java_pc;
1366 Label *resume_pc = support_vthread_preemption ? &last_java_pc : nullptr;
1367 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R3_ARG1/*tmp*/, resume_pc);
1368
1369 // Now the interpreter frame (and its call chain) have been
1370 // invalidated and flushed. We are now protected against eager
1371 // being enabled in native code. Even if it goes eager the
1372 // registers will be reloaded as clean and we will invalidate after
1373 // the call so no spurious flush should be possible.
1374
1375 // Call signature handler and pass locals address.
1376 //
1377 // Our signature handlers copy required arguments to the C stack
1378 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
1379 __ mr(R3_ARG1, R18_locals);
1380 #if !defined(ABI_ELFv2)
1381 __ ld(signature_handler_fd, 0, signature_handler_fd);
1382 #endif
1383
1384 __ call_stub(signature_handler_fd);
1385
1386 assert(__ nonvolatile_accross_vthread_preemtion(result_handler_addr),
1387 "result_handler_addr not preserved");
1388 // Save across call to native method.
1389 __ mr(result_handler_addr, R3_RET);
1390 __ ld(R11_scratch1, _abi0(callers_sp), R1_SP); // load FP
1391
1392 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
1393
1394 // Set up fixed parameters and call the native method.
1395 // If the method is static, get mirror into R4_ARG2.
1396 {
1397 Label method_is_not_static;
1398 // Access_flags is non-volatile and still, no need to restore it.
1399
1400 // Restore access flags.
1401 __ testbitdi(CR0, R0, access_flags, JVM_ACC_STATIC_BIT);
1402 __ bfalse(CR0, method_is_not_static);
1403
1404 // Load mirror from interpreter frame (FP in R11_scratch1)
1405 __ ld(R21_tmp1, _ijava_state_neg(mirror), R11_scratch1);
1406 // R4_ARG2 = &state->_oop_temp;
1407 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
1408 __ std(R21_tmp1/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
1409 BIND(method_is_not_static);
1410 }
1411
1412 // At this point, arguments have been copied off the stack into
1413 // their JNI positions. Oops are boxed in-place on the stack, with
1414 // handles copied to arguments. The result handler address is in a
1415 // register.
1416
1417 // Pass JNIEnv address as first parameter.
1418 __ addir(R3_ARG1, thread_(jni_environment));
1419
1420 // Load the native_method entry before we change the thread state.
1421 __ ld(native_method_fd, method_(native_function));
1422
1423 //=============================================================================
1424 // Transition from _thread_in_Java to _thread_in_native. As soon as
1425 // we make this change the safepoint code needs to be certain that
1426 // the last Java frame we established is good. The pc in that frame
1427 // just needs to be near here not an actual return address.
1428
1429 // We use release_store_fence to update values like the thread state, where
1430 // we don't want the current thread to continue until all our prior memory
1431 // accesses (including the new thread state) are visible to other threads.
1432 __ li(R0, _thread_in_native);
1433 __ release();
1434
1435 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
1436 __ stw(R0, thread_(thread_state));
1437
1438 //=============================================================================
1439 // Call the native method. Argument registers must not have been
1440 // overwritten since "__ call_stub(signature_handler);" (except for
1441 // ARG1 and ARG2 for static methods).
1442
1443 if (support_vthread_preemption) {
1444 // result_handler_addr is a nonvolatile register. Its value will be preserved across
1445 // the native call but only if the call isn't preempted. To preserve its value even
1446 // in the case of preemption we save it in the lresult slot. It is restored at
1447 // resume_pc if, and only if the call was preempted. This works because only
1448 // j.l.Object::wait calls are preempted which don't return a result.
1449 __ std(result_handler_addr, _ijava_state_neg(lresult), R11_scratch1);
1450 }
1451 __ push_cont_fastpath();
1452 __ call_c(native_method_fd);
1453 __ pop_cont_fastpath();
1454
1455 __ li(R0, 0);
1456 __ ld(R11_scratch1, 0, R1_SP);
1457 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1458 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1459 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
1460
1461 // Note: C++ interpreter needs the following here:
1462 // The frame_manager_lr field, which we use for setting the last
1463 // java frame, gets overwritten by the signature handler. Restore
1464 // it now.
1465 //__ get_PC_trash_LR(R11_scratch1);
1466 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
1467
1468 // Because of GC R19_method may no longer be valid.
1469
1470 // Block, if necessary, before resuming in _thread_in_Java state.
1471 // In order for GC to work, don't clear the last_Java_sp until after
1472 // blocking.
1473
1474 //=============================================================================
1475 // Switch thread to "native transition" state before reading the
1476 // synchronization state. This additional state is necessary
1477 // because reading and testing the synchronization state is not
1478 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1479 // in _thread_in_native state, loads _not_synchronized and is
1480 // preempted. VM thread changes sync state to synchronizing and
1481 // suspends threads for GC. Thread A is resumed to finish this
1482 // native method, but doesn't block here since it didn't see any
1483 // synchronization in progress, and escapes.
1484
1485 // We use release_store_fence to update values like the thread state, where
1486 // we don't want the current thread to continue until all our prior memory
1487 // accesses (including the new thread state) are visible to other threads.
1488 __ li(R0/*thread_state*/, _thread_in_native_trans);
1489 __ release();
1490 __ stw(R0/*thread_state*/, thread_(thread_state));
1491 if (!UseSystemMemoryBarrier) {
1492 __ fence();
1493 }
1494
1495 // Now before we return to java we must look for a current safepoint
1496 // (a new safepoint can not start since we entered native_trans).
1497 // We must check here because a current safepoint could be modifying
1498 // the callers registers right this moment.
1499
1500 // Acquire isn't strictly necessary here because of the fence, but
1501 // sync_state is declared to be volatile, so we do it anyway
1502 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1503
1504 Label do_safepoint, sync_check_done;
1505 // No synchronization in progress nor yet synchronized.
1506 __ safepoint_poll(do_safepoint, sync_state, true /* at_return */, false /* in_nmethod */);
1507
1508 // Not suspended.
1509 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1510 __ lwz(suspend_flags, thread_(suspend_flags));
1511 __ cmpwi(CR1, suspend_flags, 0);
1512 __ beq(CR1, sync_check_done);
1513
1514 __ bind(do_safepoint);
1515 __ isync();
1516 // Block. We do the call directly and leave the current
1517 // last_Java_frame setup undisturbed. We must save any possible
1518 // native result across the call. No oop is present.
1519
1520 __ mr(R3_ARG1, R16_thread);
1521 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1522
1523 __ bind(sync_check_done);
1524
1525 //=============================================================================
1526 // <<<<<< Back in Interpreter Frame >>>>>
1527
1528 // We are in thread_in_native_trans here and back in the normal
1529 // interpreter frame. We don't have to do anything special about
1530 // safepoints and we can switch to Java mode anytime we are ready.
1531
1532 // Note: frame::interpreter_frame_result has a dependency on how the
1533 // method result is saved across the call to post_method_exit. For
1534 // native methods it assumes that the non-FPU/non-void result is
1535 // saved in _native_lresult and a FPU result in _native_fresult. If
1536 // this changes then the interpreter_frame_result implementation
1537 // will need to be updated too.
1538
1539 // On PPC64, we have stored the result directly after the native call.
1540
1541 //=============================================================================
1542 // Back in Java
1543
1544 // We use release_store_fence to update values like the thread state, where
1545 // we don't want the current thread to continue until all our prior memory
1546 // accesses (including the new thread state) are visible to other threads.
1547 __ li(R0/*thread_state*/, _thread_in_Java);
1548 __ lwsync(); // Acquire safepoint and suspend state, release thread state.
1549 __ stw(R0/*thread_state*/, thread_(thread_state));
1550
1551 if (support_vthread_preemption) {
1552 // Check preemption for Object.wait()
1553 Label not_preempted;
1554 __ ld(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
1555 __ cmpdi(CR0, R0, 0);
1556 __ beq(CR0, not_preempted);
1557 __ mtlr(R0);
1558 __ li(R0, 0);
1559 __ std(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
1560 __ blr();
1561
1562 // Execution will be resumed here when the vthread becomes runnable again.
1563 __ bind(*resume_pc);
1564 __ restore_after_resume(R11_scratch1 /* fp */);
1565 // We saved the result handler before the call
1566 __ ld(result_handler_addr, _ijava_state_neg(lresult), R11_scratch1);
1567 #ifdef ASSERT
1568 // Clobber result slots. Only native methods returning void can be preemted currently.
1569 __ load_const(R3_RET, UCONST64(0xbad01001));
1570 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1571 __ std(R3_RET, _ijava_state_neg(fresult), R11_scratch1);
1572 // reset_last_Java_frame() below asserts that a last java sp is set
1573 __ asm_assert_mem8_is_zero(in_bytes(JavaThread::last_Java_sp_offset()),
1574 R16_thread, FILE_AND_LINE ": Last java sp should not be set when resuming");
1575 __ std(R3_RET, in_bytes(JavaThread::last_Java_sp_offset()), R16_thread);
1576 #endif
1577 __ bind(not_preempted);
1578 }
1579
1580 if (CheckJNICalls) {
1581 // clear_pending_jni_exception_check
1582 __ load_const_optimized(R0, 0L);
1583 __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread);
1584 }
1585
1586 __ reset_last_Java_frame();
1587
1588 // Jvmdi/jvmpi support. Whether we've got an exception pending or
1589 // not, and whether unlocking throws an exception or not, we notify
1590 // on native method exit. If we do have an exception, we'll end up
1591 // in the caller's context to handle it, so if we don't do the
1592 // notify here, we'll drop it on the floor.
1593 __ notify_method_exit(true/*native method*/,
1594 ilgl /*illegal state (not used for native methods)*/,
1595 InterpreterMacroAssembler::NotifyJVMTI,
1596 false /*check_exceptions*/);
1597
1598 //=============================================================================
1599 // Handle exceptions
1600
1601 if (synchronized) {
1602 __ unlock_object(R26_monitor); // Can also unlock methods.
1603 }
1604
1605 // Reset active handles after returning from native.
1606 // thread->active_handles()->clear();
1607 __ ld(active_handles, thread_(active_handles));
1608 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1609 __ li(R0, 0);
1610 __ stw(R0, in_bytes(JNIHandleBlock::top_offset()), active_handles);
1611
1612 Label exception_return_sync_check_already_unlocked;
1613 __ ld(R0/*pending_exception*/, thread_(pending_exception));
1614 __ cmpdi(CR0, R0/*pending_exception*/, 0);
1615 __ bne(CR0, exception_return_sync_check_already_unlocked);
1616
1617 //-----------------------------------------------------------------------------
1618 // No exception pending.
1619
1620 // Move native method result back into proper registers and return.
1621 // Invoke result handler (may unbox/promote).
1622 __ ld(R11_scratch1, 0, R1_SP);
1623 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1624 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1625 __ call_stub(result_handler_addr);
1626
1627 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1628
1629 // Must use the return pc which was loaded from the caller's frame
1630 // as the VM uses return-pc-patching for deoptimization.
1631 __ mtlr(R0);
1632 __ blr();
1633
1634 //-----------------------------------------------------------------------------
1635 // An exception is pending. We call into the runtime only if the
1636 // caller was not interpreted. If it was interpreted the
1637 // interpreter will do the correct thing. If it isn't interpreted
1638 // (call stub/compiled code) we will change our return and continue.
1639
1640 BIND(exception_return_sync_check);
1641
1642 if (synchronized) {
1643 __ unlock_object(R26_monitor); // Can also unlock methods.
1644 }
1645 BIND(exception_return_sync_check_already_unlocked);
1646
1647 const Register return_pc = R31;
1648
1649 __ ld(return_pc, 0, R1_SP);
1650 __ ld(return_pc, _abi0(lr), return_pc);
1651
1652 // Get the address of the exception handler.
1653 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1654 R16_thread,
1655 return_pc /* return pc */);
1656 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1657
1658 // Load the PC of the exception handler into LR.
1659 __ mtlr(R3_RET);
1660
1661 // Load exception into R3_ARG1 and clear pending exception in thread.
1662 __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1663 __ li(R4_ARG2, 0);
1664 __ std(R4_ARG2, thread_(pending_exception));
1665
1666 // Load the original return pc into R4_ARG2.
1667 __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1668
1669 // Return to exception handler.
1670 __ blr();
1671
1672 //=============================================================================
1673 // Counter overflow.
1674
1675 if (inc_counter) {
1676 // Handle invocation counter overflow.
1677 __ bind(invocation_counter_overflow);
1678
1679 generate_counter_overflow(continue_after_compile);
1680 }
1681
1682 return entry;
1683 }
1684
1685 // Generic interpreted method entry to (asm) interpreter.
1686 //
1687 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1688 bool inc_counter = UseCompiler || CountCompiledCalls;
1689 address entry = __ pc();
1690 // Generate the code to allocate the interpreter stack frame.
1691 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1692 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame.
1693
1694 // Does also a stack check to assure this frame fits on the stack.
1695 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1696
1697 // --------------------------------------------------------------------------
1698 // Zero out non-parameter locals.
1699 // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1700 // worth to ask the flag, just do it.
1701 Register Rslot_addr = R6_ARG4,
1702 Rnum = R7_ARG5;
1703 Label Lno_locals, Lzero_loop;
1704
1705 // Set up the zeroing loop.
1706 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1707 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1708 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1709 __ beq(CR0, Lno_locals);
1710 __ li(R0, 0);
1711 __ mtctr(Rnum);
1712
1713 // The zero locals loop.
1714 __ bind(Lzero_loop);
1715 __ std(R0, 0, Rslot_addr);
1716 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1717 __ bdnz(Lzero_loop);
1718
1719 __ bind(Lno_locals);
1720
1721 // --------------------------------------------------------------------------
1722 // Counter increment and overflow check.
1723 Label invocation_counter_overflow;
1724 Label continue_after_compile;
1725 if (inc_counter || ProfileInterpreter) {
1726
1727 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1728 if (synchronized) {
1729 // Since at this point in the method invocation the exception handler
1730 // would try to exit the monitor of synchronized methods which hasn't
1731 // been entered yet, we set the thread local variable
1732 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1733 // runtime, exception handling i.e. unlock_if_synchronized_method will
1734 // check this thread local flag.
1735 // This flag has two effects, one is to force an unwind in the topmost
1736 // interpreter frame and not perform an unlock while doing so.
1737 __ li(R0, 1);
1738 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1739 }
1740
1741 // Argument and return type profiling.
1742 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1743
1744 // Increment invocation counter and check for overflow.
1745 if (inc_counter) {
1746 generate_counter_incr(&invocation_counter_overflow);
1747 }
1748
1749 __ bind(continue_after_compile);
1750 }
1751
1752 bang_stack_shadow_pages(false);
1753
1754 if (inc_counter || ProfileInterpreter) {
1755 // Reset the _do_not_unlock_if_synchronized flag.
1756 if (synchronized) {
1757 __ li(R0, 0);
1758 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1759 }
1760 }
1761
1762 // --------------------------------------------------------------------------
1763 // Locking of synchronized methods. Must happen AFTER invocation_counter
1764 // check and stack overflow check, so method is not locked if overflows.
1765 if (synchronized) {
1766 lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1767 }
1768 #ifdef ASSERT
1769 else {
1770 Label Lok;
1771 __ lhz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1772 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1773 __ asm_assert_eq("method needs synchronization");
1774 __ bind(Lok);
1775 }
1776 #endif // ASSERT
1777
1778 // --------------------------------------------------------------------------
1779 // JVMTI support
1780 __ notify_method_entry();
1781
1782 // --------------------------------------------------------------------------
1783 // Start executing instructions.
1784 __ dispatch_next(vtos);
1785
1786 // --------------------------------------------------------------------------
1787 if (inc_counter) {
1788 // Handle invocation counter overflow.
1789 __ bind(invocation_counter_overflow);
1790 generate_counter_overflow(continue_after_compile);
1791 }
1792 return entry;
1793 }
1794
1795 // CRC32 Intrinsics.
1796 //
1797 // Contract on scratch and work registers.
1798 // =======================================
1799 //
1800 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1801 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1802 // You can't rely on these registers across calls.
1803 //
1804 // The generators for CRC32_update and for CRC32_updateBytes use the
1805 // scratch/work register set internally, passing the work registers
1806 // as arguments to the MacroAssembler emitters as required.
1807 //
1808 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1809 // Their contents is not constant but may change according to the requirements
1810 // of the emitted code.
1811 //
1812 // All other registers from the scratch/work register set are used "internally"
1813 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1814 // Basically, only R3_RET contains a defined value which is the function result.
1815 //
1816 /**
1817 * Method entry for static native methods:
1818 * int java.util.zip.CRC32.update(int crc, int b)
1819 */
1820 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1821 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1822 address start = __ pc(); // Remember stub start address (is rtn value).
1823 Label slow_path;
1824
1825 // Safepoint check
1826 const Register sync_state = R11_scratch1;
1827 __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1828
1829 // We don't generate local frame and don't align stack because
1830 // we not even call stub code (we generate the code inline)
1831 // and there is no safepoint on this path.
1832
1833 // Load java parameters.
1834 // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1835 const Register argP = R15_esp;
1836 const Register crc = R3_ARG1; // crc value
1837 const Register data = R4_ARG2;
1838 const Register table = R5_ARG3; // address of crc32 table
1839
1840 BLOCK_COMMENT("CRC32_update {");
1841
1842 // Arguments are reversed on java expression stack
1843 #ifdef VM_LITTLE_ENDIAN
1844 int data_offs = 0+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1845 // Being passed as an int, the single byte is at offset +0.
1846 #else
1847 int data_offs = 3+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1848 // Being passed from java as an int, the single byte is at offset +3.
1849 #endif
1850 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1851 __ lbz(data, data_offs, argP); // Byte from buffer, zero-extended.
1852 __ load_const_optimized(table, StubRoutines::crc_table_addr(), R0);
1853 __ kernel_crc32_singleByteReg(crc, data, table, true);
1854
1855 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1856 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1857 __ blr();
1858
1859 // Generate a vanilla native entry as the slow path.
1860 BLOCK_COMMENT("} CRC32_update");
1861 BIND(slow_path);
1862 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1863 return start;
1864 }
1865
1866 /**
1867 * Method entry for static native methods:
1868 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len)
1869 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1870 */
1871 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1872 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1873 address start = __ pc(); // Remember stub start address (is rtn value).
1874 Label slow_path;
1875
1876 // Safepoint check
1877 const Register sync_state = R11_scratch1;
1878 __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1879
1880 // We don't generate local frame and don't align stack because
1881 // we not even call stub code (we generate the code inline)
1882 // and there is no safepoint on this path.
1883
1884 // Load parameters.
1885 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1886 const Register argP = R15_esp;
1887 const Register crc = R3_ARG1; // crc value
1888 const Register data = R4_ARG2; // address of java byte array
1889 const Register dataLen = R5_ARG3; // source data len
1890 const Register tmp = R11_scratch1;
1891
1892 // Arguments are reversed on java expression stack.
1893 // Calculate address of start element.
1894 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1895 BLOCK_COMMENT("CRC32_updateByteBuffer {");
1896 // crc @ (SP + 5W) (32bit)
1897 // buf @ (SP + 3W) (64bit ptr to long array)
1898 // off @ (SP + 2W) (32bit)
1899 // dataLen @ (SP + 1W) (32bit)
1900 // data = buf + off
1901 __ ld( data, 3*wordSize, argP); // start of byte buffer
1902 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1903 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1904 __ lwz( crc, 5*wordSize, argP); // current crc state
1905 __ add( data, data, tmp); // Add byte buffer offset.
1906 } else { // Used for "updateBytes update".
1907 BLOCK_COMMENT("CRC32_updateBytes {");
1908 // crc @ (SP + 4W) (32bit)
1909 // buf @ (SP + 3W) (64bit ptr to byte array)
1910 // off @ (SP + 2W) (32bit)
1911 // dataLen @ (SP + 1W) (32bit)
1912 // data = buf + off + base_offset
1913 __ ld( data, 3*wordSize, argP); // start of byte buffer
1914 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1915 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1916 __ add( data, data, tmp); // add byte buffer offset
1917 __ lwz( crc, 4*wordSize, argP); // current crc state
1918 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1919 }
1920
1921 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, false);
1922
1923 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1924 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1925 __ blr();
1926
1927 // Generate a vanilla native entry as the slow path.
1928 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1929 BIND(slow_path);
1930 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1931 return start;
1932 }
1933
1934
1935 /**
1936 * Method entry for intrinsic-candidate (non-native) methods:
1937 * int java.util.zip.CRC32C.updateBytes( int crc, byte[] b, int off, int end)
1938 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
1939 * Unlike CRC32, CRC32C does not have any methods marked as native
1940 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1941 **/
1942 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1943 assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1944 address start = __ pc(); // Remember stub start address (is rtn value).
1945
1946 // We don't generate local frame and don't align stack because
1947 // we not even call stub code (we generate the code inline)
1948 // and there is no safepoint on this path.
1949
1950 // Load parameters.
1951 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1952 const Register argP = R15_esp;
1953 const Register crc = R3_ARG1; // crc value
1954 const Register data = R4_ARG2; // address of java byte array
1955 const Register dataLen = R5_ARG3; // source data len
1956 const Register tmp = R11_scratch1;
1957
1958 // Arguments are reversed on java expression stack.
1959 // Calculate address of start element.
1960 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer".
1961 BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
1962 // crc @ (SP + 5W) (32bit)
1963 // buf @ (SP + 3W) (64bit ptr to long array)
1964 // off @ (SP + 2W) (32bit)
1965 // dataLen @ (SP + 1W) (32bit)
1966 // data = buf + off
1967 __ ld( data, 3*wordSize, argP); // start of byte buffer
1968 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1969 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1970 __ lwz( crc, 5*wordSize, argP); // current crc state
1971 __ add( data, data, tmp); // Add byte buffer offset.
1972 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1973 } else { // Used for "updateBytes update".
1974 BLOCK_COMMENT("CRC32C_updateBytes {");
1975 // crc @ (SP + 4W) (32bit)
1976 // buf @ (SP + 3W) (64bit ptr to byte array)
1977 // off @ (SP + 2W) (32bit)
1978 // dataLen @ (SP + 1W) (32bit)
1979 // data = buf + off + base_offset
1980 __ ld( data, 3*wordSize, argP); // start of byte buffer
1981 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1982 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1983 __ add( data, data, tmp); // add byte buffer offset
1984 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1985 __ lwz( crc, 4*wordSize, argP); // current crc state
1986 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1987 }
1988
1989 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, true);
1990
1991 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1992 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1993 __ blr();
1994
1995 BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
1996 return start;
1997 }
1998
1999 // Not supported
2000 address TemplateInterpreterGenerator::generate_currentThread() { return nullptr; }
2001 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
2002 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
2003 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
2004 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
2005
2006 // =============================================================================
2007 // Exceptions
2008
2009 void TemplateInterpreterGenerator::generate_throw_exception() {
2010 Register Rexception = R17_tos,
2011 Rcontinuation = R3_RET;
2012
2013 // --------------------------------------------------------------------------
2014 // Entry point if an method returns with a pending exception (rethrow).
2015 Interpreter::_rethrow_exception_entry = __ pc();
2016 {
2017 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
2018
2019 // Compiled code destroys templateTableBase, reload.
2020 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
2021 }
2022
2023 // Entry point if a interpreted method throws an exception (throw).
2024 Interpreter::_throw_exception_entry = __ pc();
2025 {
2026 __ mr(Rexception, R3_RET);
2027
2028 __ verify_oop(Rexception);
2029
2030 // Expression stack must be empty before entering the VM in case of an exception.
2031 __ empty_expression_stack();
2032 // Find exception handler address and preserve exception oop.
2033 // Call C routine to find handler and jump to it.
2034 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
2035 __ mtctr(Rcontinuation);
2036 // Push exception for exception handler bytecodes.
2037 __ push_ptr(Rexception);
2038
2039 // Jump to exception handler (may be remove activation entry!).
2040 __ bctr();
2041 }
2042
2043 // If the exception is not handled in the current frame the frame is
2044 // removed and the exception is rethrown (i.e. exception
2045 // continuation is _rethrow_exception).
2046 //
2047 // Note: At this point the bci is still the bxi for the instruction
2048 // which caused the exception and the expression stack is
2049 // empty. Thus, for any VM calls at this point, GC will find a legal
2050 // oop map (with empty expression stack).
2051
2052 // In current activation
2053 // tos: exception
2054 // bcp: exception bcp
2055
2056 // --------------------------------------------------------------------------
2057 // JVMTI PopFrame support
2058
2059 Interpreter::_remove_activation_preserving_args_entry = __ pc();
2060 {
2061 // Set the popframe_processing bit in popframe_condition indicating that we are
2062 // currently handling popframe, so that call_VMs that may happen later do not
2063 // trigger new popframe handling cycles.
2064 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2065 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
2066 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2067
2068 // Empty the expression stack, as in normal exception handling.
2069 __ empty_expression_stack();
2070 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
2071
2072 // Check to see whether we are returning to a deoptimized frame.
2073 // (The PopFrame call ensures that the caller of the popped frame is
2074 // either interpreted or compiled and deoptimizes it if compiled.)
2075 // Note that we don't compare the return PC against the
2076 // deoptimization blob's unpack entry because of the presence of
2077 // adapter frames in C2.
2078 Label Lcaller_not_deoptimized;
2079 Register return_pc = R3_ARG1;
2080 __ ld(return_pc, 0, R1_SP);
2081 __ ld(return_pc, _abi0(lr), return_pc);
2082 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
2083 __ cmpdi(CR0, R3_RET, 0);
2084 __ bne(CR0, Lcaller_not_deoptimized);
2085
2086 // The deoptimized case.
2087 // In this case, we can't call dispatch_next() after the frame is
2088 // popped, but instead must save the incoming arguments and restore
2089 // them after deoptimization has occurred.
2090 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
2091 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
2092 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
2093 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
2094 __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
2095 // Save these arguments.
2096 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
2097
2098 // Inform deoptimization that it is responsible for restoring these arguments.
2099 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
2100 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2101
2102 // Return from the current method into the deoptimization blob. Will eventually
2103 // end up in the deopt interpreter entry, deoptimization prepared everything that
2104 // we will reexecute the call that called us.
2105 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
2106 __ mtlr(return_pc);
2107 __ pop_cont_fastpath();
2108 __ blr();
2109
2110 // The non-deoptimized case.
2111 __ bind(Lcaller_not_deoptimized);
2112
2113 // Clear the popframe condition flag.
2114 __ li(R0, 0);
2115 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2116
2117 // Get out of the current method and re-execute the call that called us.
2118 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2119 __ pop_cont_fastpath();
2120 __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
2121 if (ProfileInterpreter) {
2122 __ set_method_data_pointer_for_bcp();
2123 __ ld(R11_scratch1, 0, R1_SP);
2124 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
2125 }
2126 #if INCLUDE_JVMTI
2127 Label L_done;
2128
2129 __ lbz(R11_scratch1, 0, R14_bcp);
2130 __ cmpwi(CR0, R11_scratch1, Bytecodes::_invokestatic);
2131 __ bne(CR0, L_done);
2132
2133 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2134 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
2135 __ ld(R4_ARG2, 0, R18_locals);
2136 __ call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp);
2137
2138 __ cmpdi(CR0, R4_ARG2, 0);
2139 __ beq(CR0, L_done);
2140 __ std(R4_ARG2, wordSize, R15_esp);
2141 __ bind(L_done);
2142 #endif // INCLUDE_JVMTI
2143 __ dispatch_next(vtos);
2144 }
2145 // end of JVMTI PopFrame support
2146
2147 // --------------------------------------------------------------------------
2148 // Remove activation exception entry.
2149 // This is jumped to if an interpreted method can't handle an exception itself
2150 // (we come from the throw/rethrow exception entry above). We're going to call
2151 // into the VM to find the exception handler in the caller, pop the current
2152 // frame and return the handler we calculated.
2153 Interpreter::_remove_activation_entry = __ pc();
2154 {
2155 __ pop_ptr(Rexception);
2156 __ verify_oop(Rexception);
2157 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
2158
2159 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
2160 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
2161
2162 __ get_vm_result(Rexception);
2163
2164 // We are done with this activation frame; find out where to go next.
2165 // The continuation point will be an exception handler, which expects
2166 // the following registers set up:
2167 //
2168 // RET: exception oop
2169 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
2170
2171 Register return_pc = R31; // Needs to survive the runtime call.
2172 __ ld(return_pc, 0, R1_SP);
2173 __ ld(return_pc, _abi0(lr), return_pc);
2174 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
2175
2176 // Remove the current activation.
2177 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2178 __ pop_cont_fastpath();
2179
2180 __ mr(R4_ARG2, return_pc);
2181 __ mtlr(R3_RET);
2182 __ mr(R3_RET, Rexception);
2183 __ blr();
2184 }
2185 }
2186
2187 // JVMTI ForceEarlyReturn support.
2188 // Returns "in the middle" of a method with a "fake" return value.
2189 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2190
2191 Register Rscratch1 = R11_scratch1,
2192 Rscratch2 = R12_scratch2;
2193
2194 address entry = __ pc();
2195 __ empty_expression_stack();
2196
2197 __ load_earlyret_value(state, Rscratch1);
2198
2199 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
2200 // Clear the earlyret state.
2201 __ li(R0, 0);
2202 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
2203
2204 __ remove_activation(state, false, false);
2205 // Copied from TemplateTable::_return.
2206 // Restoration of lr done by remove_activation.
2207 switch (state) {
2208 // Narrow result if state is itos but result type is smaller.
2209 case btos:
2210 case ztos:
2211 case ctos:
2212 case stos:
2213 case itos: __ narrow(R17_tos); /* fall through */
2214 case ltos:
2215 case atos: __ mr(R3_RET, R17_tos); break;
2216 case ftos:
2217 case dtos: __ fmr(F1_RET, F15_ftos); break;
2218 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2219 // to get visible before the reference to the object gets stored anywhere.
2220 __ membar(Assembler::StoreStore); break;
2221 default : ShouldNotReachHere();
2222 }
2223 __ blr();
2224
2225 return entry;
2226 } // end of ForceEarlyReturn support
2227
2228 //-----------------------------------------------------------------------------
2229 // Helper for vtos entry point generation
2230
2231 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2232 address& bep,
2233 address& cep,
2234 address& sep,
2235 address& aep,
2236 address& iep,
2237 address& lep,
2238 address& fep,
2239 address& dep,
2240 address& vep) {
2241 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2242 Label L;
2243
2244 aep = __ pc(); __ push_ptr(); __ b(L);
2245 fep = __ pc(); __ push_f(); __ b(L);
2246 dep = __ pc(); __ push_d(); __ b(L);
2247 lep = __ pc(); __ push_l(); __ b(L);
2248 __ align(32, 12, 24); // align L
2249 bep = cep = sep =
2250 iep = __ pc(); __ push_i();
2251 vep = __ pc();
2252 __ bind(L);
2253 generate_and_dispatch(t);
2254 }
2255
2256 //-----------------------------------------------------------------------------
2257
2258 // Non-product code
2259 #ifndef PRODUCT
2260 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2261 //__ flush_bundle();
2262 address entry = __ pc();
2263
2264 const char *bname = nullptr;
2265 uint tsize = 0;
2266 switch(state) {
2267 case ftos:
2268 bname = "trace_code_ftos {";
2269 tsize = 2;
2270 break;
2271 case btos:
2272 bname = "trace_code_btos {";
2273 tsize = 2;
2274 break;
2275 case ztos:
2276 bname = "trace_code_ztos {";
2277 tsize = 2;
2278 break;
2279 case ctos:
2280 bname = "trace_code_ctos {";
2281 tsize = 2;
2282 break;
2283 case stos:
2284 bname = "trace_code_stos {";
2285 tsize = 2;
2286 break;
2287 case itos:
2288 bname = "trace_code_itos {";
2289 tsize = 2;
2290 break;
2291 case ltos:
2292 bname = "trace_code_ltos {";
2293 tsize = 3;
2294 break;
2295 case atos:
2296 bname = "trace_code_atos {";
2297 tsize = 2;
2298 break;
2299 case vtos:
2300 // Note: In case of vtos, the topmost of stack value could be a int or doubl
2301 // In case of a double (2 slots) we won't see the 2nd stack value.
2302 // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
2303 bname = "trace_code_vtos {";
2304 tsize = 2;
2305
2306 break;
2307 case dtos:
2308 bname = "trace_code_dtos {";
2309 tsize = 3;
2310 break;
2311 default:
2312 ShouldNotReachHere();
2313 }
2314 BLOCK_COMMENT(bname);
2315
2316 // Support short-cut for TraceBytecodesAt.
2317 // Don't call into the VM if we don't want to trace to speed up things.
2318 Label Lskip_vm_call;
2319 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2320 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
2321 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2322 __ ld(R11_scratch1, offs1, R11_scratch1);
2323 __ lwa(R12_scratch2, offs2, R12_scratch2);
2324 __ cmpd(CR0, R12_scratch2, R11_scratch1);
2325 __ blt(CR0, Lskip_vm_call);
2326 }
2327
2328 __ push(state);
2329 // Load 2 topmost expression stack values.
2330 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
2331 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
2332 __ mflr(R31);
2333 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
2334 __ mtlr(R31);
2335 __ pop(state);
2336
2337 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2338 __ bind(Lskip_vm_call);
2339 }
2340 __ blr();
2341 BLOCK_COMMENT("} trace_code");
2342 return entry;
2343 }
2344
2345 void TemplateInterpreterGenerator::count_bytecode() {
2346 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
2347 __ lwz(R12_scratch2, offs, R11_scratch1);
2348 __ addi(R12_scratch2, R12_scratch2, 1);
2349 __ stw(R12_scratch2, offs, R11_scratch1);
2350 }
2351
2352 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2353 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
2354 __ lwz(R12_scratch2, offs, R11_scratch1);
2355 __ addi(R12_scratch2, R12_scratch2, 1);
2356 __ stw(R12_scratch2, offs, R11_scratch1);
2357 }
2358
2359 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2360 const Register addr = R11_scratch1,
2361 tmp = R12_scratch2;
2362 // Get index, shift out old bytecode, bring in new bytecode, and store it.
2363 // _index = (_index >> log2_number_of_codes) |
2364 // (bytecode << log2_number_of_codes);
2365 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
2366 __ lwz(tmp, offs1, addr);
2367 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
2368 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2369 __ stw(tmp, offs1, addr);
2370
2371 // Bump bucket contents.
2372 // _counters[_index] ++;
2373 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
2374 __ sldi(tmp, tmp, LogBytesPerInt);
2375 __ add(addr, tmp, addr);
2376 __ lwz(tmp, offs2, addr);
2377 __ addi(tmp, tmp, 1);
2378 __ stw(tmp, offs2, addr);
2379 }
2380
2381 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2382 // Call a little run-time stub to avoid blow-up for each bytecode.
2383 // The run-time runtime saves the right registers, depending on
2384 // the tosca in-state for the given template.
2385
2386 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2387 "entry must have been generated");
2388
2389 // Note: we destroy LR here.
2390 __ bl(Interpreter::trace_code(t->tos_in()));
2391 }
2392
2393 void TemplateInterpreterGenerator::stop_interpreter_at() {
2394 Label L;
2395 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
2396 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2397 __ ld(R11_scratch1, offs1, R11_scratch1);
2398 __ lwa(R12_scratch2, offs2, R12_scratch2);
2399 __ cmpd(CR0, R12_scratch2, R11_scratch1);
2400 __ bne(CR0, L);
2401 __ illtrap();
2402 __ bind(L);
2403 }
2404
2405 #endif // !PRODUCT