1 /*
2 * Copyright (c) 2014, 2020, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2019, SAP SE. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "gc/shared/barrierSetAssembler.hpp"
29 #include "interpreter/bytecodeHistogram.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "interpreter/interpreterRuntime.hpp"
32 #include "interpreter/interp_masm.hpp"
33 #include "interpreter/templateInterpreterGenerator.hpp"
34 #include "interpreter/templateTable.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/methodData.hpp"
37 #include "oops/method.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "prims/jvmtiExport.hpp"
40 #include "prims/jvmtiThreadState.hpp"
41 #include "runtime/arguments.hpp"
42 #include "runtime/deoptimization.hpp"
43 #include "runtime/frame.inline.hpp"
44 #include "runtime/sharedRuntime.hpp"
45 #include "runtime/stubRoutines.hpp"
46 #include "runtime/synchronizer.hpp"
47 #include "runtime/timer.hpp"
48 #include "runtime/vframeArray.hpp"
49 #include "utilities/debug.hpp"
50 #include "utilities/macros.hpp"
51
52 #undef __
53 #define __ _masm->
54
55 // Size of interpreter code. Increase if too small. Interpreter will
56 // fail with a guarantee ("not enough space for interpreter generation");
57 // if too small.
58 // Run with +PrintInterpreter to get the VM to print out the size.
59 // Max size with JVMTI
60 int TemplateInterpreter::InterpreterCodeSize = 256*K;
61
62 #ifdef PRODUCT
63 #define BLOCK_COMMENT(str) /* nothing */
64 #else
65 #define BLOCK_COMMENT(str) __ block_comment(str)
66 #endif
67
68 #define BIND(label) __ bind(label); BLOCK_COMMENT(#label ":")
69
70 //-----------------------------------------------------------------------------
71
generate_slow_signature_handler()72 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
73 // Slow_signature handler that respects the PPC C calling conventions.
74 //
75 // We get called by the native entry code with our output register
76 // area == 8. First we call InterpreterRuntime::get_result_handler
77 // to copy the pointer to the signature string temporarily to the
78 // first C-argument and to return the result_handler in
79 // R3_RET. Since native_entry will copy the jni-pointer to the
80 // first C-argument slot later on, it is OK to occupy this slot
81 // temporarilly. Then we copy the argument list on the java
82 // expression stack into native varargs format on the native stack
83 // and load arguments into argument registers. Integer arguments in
84 // the varargs vector will be sign-extended to 8 bytes.
85 //
86 // On entry:
87 // R3_ARG1 - intptr_t* Address of java argument list in memory.
88 // R15_prev_state - BytecodeInterpreter* Address of interpreter state for
89 // this method
90 // R19_method
91 //
92 // On exit (just before return instruction):
93 // R3_RET - contains the address of the result_handler.
94 // R4_ARG2 - is not updated for static methods and contains "this" otherwise.
95 // R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double,
96 // ARGi contains this argument. Otherwise, ARGi is not updated.
97 // F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double.
98
99 const int LogSizeOfTwoInstructions = 3;
100
101 // FIXME: use Argument:: GL: Argument names different numbers!
102 const int max_fp_register_arguments = 13;
103 const int max_int_register_arguments = 6; // first 2 are reserved
104
105 const Register arg_java = R21_tmp1;
106 const Register arg_c = R22_tmp2;
107 const Register signature = R23_tmp3; // is string
108 const Register sig_byte = R24_tmp4;
109 const Register fpcnt = R25_tmp5;
110 const Register argcnt = R26_tmp6;
111 const Register intSlot = R27_tmp7;
112 const Register target_sp = R28_tmp8;
113 const FloatRegister floatSlot = F0;
114
115 address entry = __ function_entry();
116
117 __ save_LR_CR(R0);
118 __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
119 // We use target_sp for storing arguments in the C frame.
120 __ mr(target_sp, R1_SP);
121 __ push_frame_reg_args_nonvolatiles(0, R11_scratch1);
122
123 __ mr(arg_java, R3_ARG1);
124
125 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method);
126
127 // Signature is in R3_RET. Signature is callee saved.
128 __ mr(signature, R3_RET);
129
130 // Get the result handler.
131 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method);
132
133 {
134 Label L;
135 // test if static
136 // _access_flags._flags must be at offset 0.
137 // TODO PPC port: requires change in shared code.
138 //assert(in_bytes(AccessFlags::flags_offset()) == 0,
139 // "MethodDesc._access_flags == MethodDesc._access_flags._flags");
140 // _access_flags must be a 32 bit value.
141 assert(sizeof(AccessFlags) == 4, "wrong size");
142 __ lwa(R11_scratch1/*access_flags*/, method_(access_flags));
143 // testbit with condition register.
144 __ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT);
145 __ btrue(CCR0, L);
146 // For non-static functions, pass "this" in R4_ARG2 and copy it
147 // to 2nd C-arg slot.
148 // We need to box the Java object here, so we use arg_java
149 // (address of current Java stack slot) as argument and don't
150 // dereference it as in case of ints, floats, etc.
151 __ mr(R4_ARG2, arg_java);
152 __ addi(arg_java, arg_java, -BytesPerWord);
153 __ std(R4_ARG2, _abi(carg_2), target_sp);
154 __ bind(L);
155 }
156
157 // Will be incremented directly after loop_start. argcnt=0
158 // corresponds to 3rd C argument.
159 __ li(argcnt, -1);
160 // arg_c points to 3rd C argument
161 __ addi(arg_c, target_sp, _abi(carg_3));
162 // no floating-point args parsed so far
163 __ li(fpcnt, 0);
164
165 Label move_intSlot_to_ARG, move_floatSlot_to_FARG;
166 Label loop_start, loop_end;
167 Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed;
168
169 // signature points to '(' at entry
170 #ifdef ASSERT
171 __ lbz(sig_byte, 0, signature);
172 __ cmplwi(CCR0, sig_byte, '(');
173 __ bne(CCR0, do_dontreachhere);
174 #endif
175
176 __ bind(loop_start);
177
178 __ addi(argcnt, argcnt, 1);
179 __ lbzu(sig_byte, 1, signature);
180
181 __ cmplwi(CCR0, sig_byte, ')'); // end of signature
182 __ beq(CCR0, loop_end);
183
184 __ cmplwi(CCR0, sig_byte, 'B'); // byte
185 __ beq(CCR0, do_int);
186
187 __ cmplwi(CCR0, sig_byte, 'C'); // char
188 __ beq(CCR0, do_int);
189
190 __ cmplwi(CCR0, sig_byte, 'D'); // double
191 __ beq(CCR0, do_double);
192
193 __ cmplwi(CCR0, sig_byte, 'F'); // float
194 __ beq(CCR0, do_float);
195
196 __ cmplwi(CCR0, sig_byte, 'I'); // int
197 __ beq(CCR0, do_int);
198
199 __ cmplwi(CCR0, sig_byte, 'J'); // long
200 __ beq(CCR0, do_long);
201
202 __ cmplwi(CCR0, sig_byte, 'S'); // short
203 __ beq(CCR0, do_int);
204
205 __ cmplwi(CCR0, sig_byte, 'Z'); // boolean
206 __ beq(CCR0, do_int);
207
208 __ cmplwi(CCR0, sig_byte, 'L'); // object
209 __ beq(CCR0, do_object);
210
211 __ cmplwi(CCR0, sig_byte, '['); // array
212 __ beq(CCR0, do_array);
213
214 // __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type
215 // __ beq(CCR0, do_void);
216
217 __ bind(do_dontreachhere);
218
219 __ unimplemented("ShouldNotReachHere in slow_signature_handler");
220
221 __ bind(do_array);
222
223 {
224 Label start_skip, end_skip;
225
226 __ bind(start_skip);
227 __ lbzu(sig_byte, 1, signature);
228 __ cmplwi(CCR0, sig_byte, '[');
229 __ beq(CCR0, start_skip); // skip further brackets
230 __ cmplwi(CCR0, sig_byte, '9');
231 __ bgt(CCR0, end_skip); // no optional size
232 __ cmplwi(CCR0, sig_byte, '0');
233 __ bge(CCR0, start_skip); // skip optional size
234 __ bind(end_skip);
235
236 __ cmplwi(CCR0, sig_byte, 'L');
237 __ beq(CCR0, do_object); // for arrays of objects, the name of the object must be skipped
238 __ b(do_boxed); // otherwise, go directly to do_boxed
239 }
240
241 __ bind(do_object);
242 {
243 Label L;
244 __ bind(L);
245 __ lbzu(sig_byte, 1, signature);
246 __ cmplwi(CCR0, sig_byte, ';');
247 __ bne(CCR0, L);
248 }
249 // Need to box the Java object here, so we use arg_java (address of
250 // current Java stack slot) as argument and don't dereference it as
251 // in case of ints, floats, etc.
252 Label do_null;
253 __ bind(do_boxed);
254 __ ld(R0,0, arg_java);
255 __ cmpdi(CCR0, R0, 0);
256 __ li(intSlot,0);
257 __ beq(CCR0, do_null);
258 __ mr(intSlot, arg_java);
259 __ bind(do_null);
260 __ std(intSlot, 0, arg_c);
261 __ addi(arg_java, arg_java, -BytesPerWord);
262 __ addi(arg_c, arg_c, BytesPerWord);
263 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
264 __ blt(CCR0, move_intSlot_to_ARG);
265 __ b(loop_start);
266
267 __ bind(do_int);
268 __ lwa(intSlot, 0, arg_java);
269 __ std(intSlot, 0, arg_c);
270 __ addi(arg_java, arg_java, -BytesPerWord);
271 __ addi(arg_c, arg_c, BytesPerWord);
272 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
273 __ blt(CCR0, move_intSlot_to_ARG);
274 __ b(loop_start);
275
276 __ bind(do_long);
277 __ ld(intSlot, -BytesPerWord, arg_java);
278 __ std(intSlot, 0, arg_c);
279 __ addi(arg_java, arg_java, - 2 * BytesPerWord);
280 __ addi(arg_c, arg_c, BytesPerWord);
281 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
282 __ blt(CCR0, move_intSlot_to_ARG);
283 __ b(loop_start);
284
285 __ bind(do_float);
286 __ lfs(floatSlot, 0, arg_java);
287 #if defined(LINUX)
288 // Linux uses ELF ABI. Both original ELF and ELFv2 ABIs have float
289 // in the least significant word of an argument slot.
290 #if defined(VM_LITTLE_ENDIAN)
291 __ stfs(floatSlot, 0, arg_c);
292 #else
293 __ stfs(floatSlot, 4, arg_c);
294 #endif
295 #elif defined(AIX)
296 // Although AIX runs on big endian CPU, float is in most significant
297 // word of an argument slot.
298 __ stfs(floatSlot, 0, arg_c);
299 #elif defined(_ALLBSD_SOURCE)
300 __ stfs(floatSlot, 4, arg_c);
301 #else
302 #error "unknown OS"
303 #endif
304 __ addi(arg_java, arg_java, -BytesPerWord);
305 __ addi(arg_c, arg_c, BytesPerWord);
306 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
307 __ blt(CCR0, move_floatSlot_to_FARG);
308 __ b(loop_start);
309
310 __ bind(do_double);
311 __ lfd(floatSlot, - BytesPerWord, arg_java);
312 __ stfd(floatSlot, 0, arg_c);
313 __ addi(arg_java, arg_java, - 2 * BytesPerWord);
314 __ addi(arg_c, arg_c, BytesPerWord);
315 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
316 __ blt(CCR0, move_floatSlot_to_FARG);
317 __ b(loop_start);
318
319 __ bind(loop_end);
320
321 __ pop_frame();
322 __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
323 __ restore_LR_CR(R0);
324
325 __ blr();
326
327 Label move_int_arg, move_float_arg;
328 __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
329 __ mr(R5_ARG3, intSlot); __ b(loop_start);
330 __ mr(R6_ARG4, intSlot); __ b(loop_start);
331 __ mr(R7_ARG5, intSlot); __ b(loop_start);
332 __ mr(R8_ARG6, intSlot); __ b(loop_start);
333 __ mr(R9_ARG7, intSlot); __ b(loop_start);
334 __ mr(R10_ARG8, intSlot); __ b(loop_start);
335
336 __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
337 __ fmr(F1_ARG1, floatSlot); __ b(loop_start);
338 __ fmr(F2_ARG2, floatSlot); __ b(loop_start);
339 __ fmr(F3_ARG3, floatSlot); __ b(loop_start);
340 __ fmr(F4_ARG4, floatSlot); __ b(loop_start);
341 __ fmr(F5_ARG5, floatSlot); __ b(loop_start);
342 __ fmr(F6_ARG6, floatSlot); __ b(loop_start);
343 __ fmr(F7_ARG7, floatSlot); __ b(loop_start);
344 __ fmr(F8_ARG8, floatSlot); __ b(loop_start);
345 __ fmr(F9_ARG9, floatSlot); __ b(loop_start);
346 __ fmr(F10_ARG10, floatSlot); __ b(loop_start);
347 __ fmr(F11_ARG11, floatSlot); __ b(loop_start);
348 __ fmr(F12_ARG12, floatSlot); __ b(loop_start);
349 __ fmr(F13_ARG13, floatSlot); __ b(loop_start);
350
351 __ bind(move_intSlot_to_ARG);
352 __ sldi(R0, argcnt, LogSizeOfTwoInstructions);
353 __ load_const(R11_scratch1, move_int_arg); // Label must be bound here.
354 __ add(R11_scratch1, R0, R11_scratch1);
355 __ mtctr(R11_scratch1/*branch_target*/);
356 __ bctr();
357 __ bind(move_floatSlot_to_FARG);
358 __ sldi(R0, fpcnt, LogSizeOfTwoInstructions);
359 __ addi(fpcnt, fpcnt, 1);
360 __ load_const(R11_scratch1, move_float_arg); // Label must be bound here.
361 __ add(R11_scratch1, R0, R11_scratch1);
362 __ mtctr(R11_scratch1/*branch_target*/);
363 __ bctr();
364
365 return entry;
366 }
367
generate_result_handler_for(BasicType type)368 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
369 //
370 // Registers alive
371 // R3_RET
372 // LR
373 //
374 // Registers updated
375 // R3_RET
376 //
377
378 Label done;
379 address entry = __ pc();
380
381 switch (type) {
382 case T_BOOLEAN:
383 // convert !=0 to 1
384 __ neg(R0, R3_RET);
385 __ orr(R0, R3_RET, R0);
386 __ srwi(R3_RET, R0, 31);
387 break;
388 case T_BYTE:
389 // sign extend 8 bits
390 __ extsb(R3_RET, R3_RET);
391 break;
392 case T_CHAR:
393 // zero extend 16 bits
394 __ clrldi(R3_RET, R3_RET, 48);
395 break;
396 case T_SHORT:
397 // sign extend 16 bits
398 __ extsh(R3_RET, R3_RET);
399 break;
400 case T_INT:
401 // sign extend 32 bits
402 __ extsw(R3_RET, R3_RET);
403 break;
404 case T_LONG:
405 break;
406 case T_OBJECT:
407 // JNIHandles::resolve result.
408 __ resolve_jobject(R3_RET, R11_scratch1, R31, /* needs_frame */ true); // kills R31
409 break;
410 case T_FLOAT:
411 break;
412 case T_DOUBLE:
413 break;
414 case T_VOID:
415 break;
416 default: ShouldNotReachHere();
417 }
418
419 BIND(done);
420 __ blr();
421
422 return entry;
423 }
424
425 // Abstract method entry.
426 //
generate_abstract_entry(void)427 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
428 address entry = __ pc();
429
430 //
431 // Registers alive
432 // R16_thread - JavaThread*
433 // R19_method - callee's method (method to be invoked)
434 // R1_SP - SP prepared such that caller's outgoing args are near top
435 // LR - return address to caller
436 //
437 // Stack layout at this point:
438 //
439 // 0 [TOP_IJAVA_FRAME_ABI] <-- R1_SP
440 // alignment (optional)
441 // [outgoing Java arguments]
442 // ...
443 // PARENT [PARENT_IJAVA_FRAME_ABI]
444 // ...
445 //
446
447 // Can't use call_VM here because we have not set up a new
448 // interpreter state. Make the call to the vm and make it look like
449 // our caller set up the JavaFrameAnchor.
450 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
451
452 // Push a new C frame and save LR.
453 __ save_LR_CR(R0);
454 __ push_frame_reg_args(0, R11_scratch1);
455
456 // This is not a leaf but we have a JavaFrameAnchor now and we will
457 // check (create) exceptions afterward so this is ok.
458 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
459 R16_thread, R19_method);
460
461 // Pop the C frame and restore LR.
462 __ pop_frame();
463 __ restore_LR_CR(R0);
464
465 // Reset JavaFrameAnchor from call_VM_leaf above.
466 __ reset_last_Java_frame();
467
468 // We don't know our caller, so jump to the general forward exception stub,
469 // which will also pop our full frame off. Satisfy the interface of
470 // SharedRuntime::generate_forward_exception()
471 __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0);
472 __ mtctr(R11_scratch1);
473 __ bctr();
474
475 return entry;
476 }
477
478 // Interpreter intrinsic for WeakReference.get().
479 // 1. Don't push a full blown frame and go on dispatching, but fetch the value
480 // into R8 and return quickly
481 // 2. If G1 is active we *must* execute this intrinsic for corrrectness:
482 // It contains a GC barrier which puts the reference into the satb buffer
483 // to indicate that someone holds a strong reference to the object the
484 // weak ref points to!
generate_Reference_get_entry(void)485 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
486 // Code: _aload_0, _getfield, _areturn
487 // parameter size = 1
488 //
489 // The code that gets generated by this routine is split into 2 parts:
490 // 1. the "intrinsified" code for G1 (or any SATB based GC),
491 // 2. the slow path - which is an expansion of the regular method entry.
492 //
493 // Notes:
494 // * In the G1 code we do not check whether we need to block for
495 // a safepoint. If G1 is enabled then we must execute the specialized
496 // code for Reference.get (except when the Reference object is null)
497 // so that we can log the value in the referent field with an SATB
498 // update buffer.
499 // If the code for the getfield template is modified so that the
500 // G1 pre-barrier code is executed when the current method is
501 // Reference.get() then going through the normal method entry
502 // will be fine.
503 // * The G1 code can, however, check the receiver object (the instance
504 // of java.lang.Reference) and jump to the slow path if null. If the
505 // Reference object is null then we obviously cannot fetch the referent
506 // and so we don't need to call the G1 pre-barrier. Thus we can use the
507 // regular method entry code to generate the NPE.
508 //
509
510 address entry = __ pc();
511
512 const int referent_offset = java_lang_ref_Reference::referent_offset();
513
514 Label slow_path;
515
516 // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH);
517
518 // In the G1 code we don't check if we need to reach a safepoint. We
519 // continue and the thread will safepoint at the next bytecode dispatch.
520
521 // If the receiver is null then it is OK to jump to the slow path.
522 __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver
523
524 // Check if receiver == NULL and go the slow path.
525 __ cmpdi(CCR0, R3_RET, 0);
526 __ beq(CCR0, slow_path);
527
528 __ load_heap_oop(R3_RET, referent_offset, R3_RET,
529 /* non-volatile temp */ R31, R11_scratch1, true, ON_WEAK_OOP_REF);
530
531 // Generate the G1 pre-barrier code to log the value of
532 // the referent field in an SATB buffer. Note with
533 // these parameters the pre-barrier does not generate
534 // the load of the previous value.
535
536 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
537 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
538
539 __ blr();
540
541 __ bind(slow_path);
542 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
543 return entry;
544 }
545
generate_StackOverflowError_handler()546 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
547 address entry = __ pc();
548
549 // Expression stack must be empty before entering the VM if an
550 // exception happened.
551 __ empty_expression_stack();
552 // Throw exception.
553 __ call_VM(noreg,
554 CAST_FROM_FN_PTR(address,
555 InterpreterRuntime::throw_StackOverflowError));
556 return entry;
557 }
558
generate_ArrayIndexOutOfBounds_handler()559 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
560 address entry = __ pc();
561 __ empty_expression_stack();
562 // R4_ARG2 already contains the array.
563 // Index is in R17_tos.
564 __ mr(R5_ARG3, R17_tos);
565 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R4_ARG2, R5_ARG3);
566 return entry;
567 }
568
generate_ClassCastException_handler()569 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
570 address entry = __ pc();
571 // Expression stack must be empty before entering the VM if an
572 // exception happened.
573 __ empty_expression_stack();
574
575 // Load exception object.
576 // Thread will be loaded to R3_ARG1.
577 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
578 #ifdef ASSERT
579 // Above call must not return here since exception pending.
580 __ should_not_reach_here();
581 #endif
582 return entry;
583 }
584
generate_exception_handler_common(const char * name,const char * message,bool pass_oop)585 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
586 address entry = __ pc();
587 //__ untested("generate_exception_handler_common");
588 Register Rexception = R17_tos;
589
590 // Expression stack must be empty before entering the VM if an exception happened.
591 __ empty_expression_stack();
592
593 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
594 if (pass_oop) {
595 __ mr(R5_ARG3, Rexception);
596 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception));
597 } else {
598 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
599 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception));
600 }
601
602 // Throw exception.
603 __ mr(R3_ARG1, Rexception);
604 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
605 __ mtctr(R11_scratch1);
606 __ bctr();
607
608 return entry;
609 }
610
611 // This entry is returned to when a call returns to the interpreter.
612 // When we arrive here, we expect that the callee stack frame is already popped.
generate_return_entry_for(TosState state,int step,size_t index_size)613 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
614 address entry = __ pc();
615
616 // Move the value out of the return register back to the TOS cache of current frame.
617 switch (state) {
618 case ltos:
619 case btos:
620 case ztos:
621 case ctos:
622 case stos:
623 case atos:
624 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache
625 case ftos:
626 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
627 case vtos: break; // Nothing to do, this was a void return.
628 default : ShouldNotReachHere();
629 }
630
631 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
632 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
633 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
634
635 // Compiled code destroys templateTableBase, reload.
636 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
637
638 if (state == atos) {
639 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
640 }
641
642 const Register cache = R11_scratch1;
643 const Register size = R12_scratch2;
644 __ get_cache_and_index_at_bcp(cache, 1, index_size);
645
646 // Get least significant byte of 64 bit value:
647 #if defined(VM_LITTLE_ENDIAN)
648 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache);
649 #else
650 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
651 #endif
652 __ sldi(size, size, Interpreter::logStackElementSize);
653 __ add(R15_esp, R15_esp, size);
654
655 __ check_and_handle_popframe(R11_scratch1);
656 __ check_and_handle_earlyret(R11_scratch1);
657
658 __ dispatch_next(state, step);
659 return entry;
660 }
661
generate_deopt_entry_for(TosState state,int step,address continuation)662 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
663 address entry = __ pc();
664 // If state != vtos, we're returning from a native method, which put it's result
665 // into the result register. So move the value out of the return register back
666 // to the TOS cache of current frame.
667
668 switch (state) {
669 case ltos:
670 case btos:
671 case ztos:
672 case ctos:
673 case stos:
674 case atos:
675 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache
676 case ftos:
677 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
678 case vtos: break; // Nothing to do, this was a void return.
679 default : ShouldNotReachHere();
680 }
681
682 // Load LcpoolCache @@@ should be already set!
683 __ get_constant_pool_cache(R27_constPoolCache);
684
685 // Handle a pending exception, fall through if none.
686 __ check_and_forward_exception(R11_scratch1, R12_scratch2);
687
688 // Start executing bytecodes.
689 if (continuation == NULL) {
690 __ dispatch_next(state, step);
691 } else {
692 __ jump_to_entry(continuation, R11_scratch1);
693 }
694
695 return entry;
696 }
697
generate_safept_entry_for(TosState state,address runtime_entry)698 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
699 address entry = __ pc();
700
701 __ push(state);
702 __ call_VM(noreg, runtime_entry);
703 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
704
705 return entry;
706 }
707
708 // Helpers for commoning out cases in the various type of method entries.
709
710 // Increment invocation count & check for overflow.
711 //
712 // Note: checking for negative value instead of overflow
713 // so we have a 'sticky' overflow test.
714 //
generate_counter_incr(Label * overflow,Label * profile_method,Label * profile_method_continue)715 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
716 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
717 Register Rscratch1 = R11_scratch1;
718 Register Rscratch2 = R12_scratch2;
719 Register R3_counters = R3_ARG1;
720 Label done;
721
722 if (TieredCompilation) {
723 const int increment = InvocationCounter::count_increment;
724 Label no_mdo;
725 if (ProfileInterpreter) {
726 const Register Rmdo = R3_counters;
727 // If no method data exists, go to profile_continue.
728 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
729 __ cmpdi(CCR0, Rmdo, 0);
730 __ beq(CCR0, no_mdo);
731
732 // Increment invocation counter in the MDO.
733 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
734 __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
735 __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo);
736 __ addi(Rscratch2, Rscratch2, increment);
737 __ stw(Rscratch2, mdo_ic_offs, Rmdo);
738 __ and_(Rscratch1, Rscratch2, Rscratch1);
739 __ bne(CCR0, done);
740 __ b(*overflow);
741 }
742
743 // Increment counter in MethodCounters*.
744 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
745 __ bind(no_mdo);
746 __ get_method_counters(R19_method, R3_counters, done);
747 __ lwz(Rscratch2, mo_ic_offs, R3_counters);
748 __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters);
749 __ addi(Rscratch2, Rscratch2, increment);
750 __ stw(Rscratch2, mo_ic_offs, R3_counters);
751 __ and_(Rscratch1, Rscratch2, Rscratch1);
752 __ beq(CCR0, *overflow);
753
754 __ bind(done);
755
756 } else {
757
758 // Update standard invocation counters.
759 Register Rsum_ivc_bec = R4_ARG2;
760 __ get_method_counters(R19_method, R3_counters, done);
761 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2);
762 // Increment interpreter invocation counter.
763 if (ProfileInterpreter) { // %%% Merge this into methodDataOop.
764 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
765 __ addi(R12_scratch2, R12_scratch2, 1);
766 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
767 }
768 // Check if we must create a method data obj.
769 if (ProfileInterpreter && profile_method != NULL) {
770 const Register profile_limit = Rscratch1;
771 __ lwz(profile_limit, in_bytes(MethodCounters::interpreter_profile_limit_offset()), R3_counters);
772 // Test to see if we should create a method data oop.
773 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit);
774 __ blt(CCR0, *profile_method_continue);
775 // If no method data exists, go to profile_method.
776 __ test_method_data_pointer(*profile_method);
777 }
778 // Finally check for counter overflow.
779 if (overflow) {
780 const Register invocation_limit = Rscratch1;
781 __ lwz(invocation_limit, in_bytes(MethodCounters::interpreter_invocation_limit_offset()), R3_counters);
782 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit);
783 __ bge(CCR0, *overflow);
784 }
785
786 __ bind(done);
787 }
788 }
789
790 // Generate code to initiate compilation on invocation counter overflow.
generate_counter_overflow(Label & continue_entry)791 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
792 // Generate code to initiate compilation on the counter overflow.
793
794 // InterpreterRuntime::frequency_counter_overflow takes one arguments,
795 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp)
796 // We pass zero in.
797 // The call returns the address of the verified entry point for the method or NULL
798 // if the compilation did not complete (either went background or bailed out).
799 //
800 // Unlike the C++ interpreter above: Check exceptions!
801 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
802 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
803
804 __ li(R4_ARG2, 0);
805 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
806
807 // Returns verified_entry_point or NULL.
808 // We ignore it in any case.
809 __ b(continue_entry);
810 }
811
812 // See if we've got enough room on the stack for locals plus overhead below
813 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
814 // without going through the signal handler, i.e., reserved and yellow zones
815 // will not be made usable. The shadow zone must suffice to handle the
816 // overflow.
817 //
818 // Kills Rmem_frame_size, Rscratch1.
generate_stack_overflow_check(Register Rmem_frame_size,Register Rscratch1)819 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
820 Label done;
821 assert_different_registers(Rmem_frame_size, Rscratch1);
822
823 BLOCK_COMMENT("stack_overflow_check_with_compare {");
824 __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
825 __ ld(Rscratch1, thread_(stack_overflow_limit));
826 __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
827 __ bgt(CCR0/*is_stack_overflow*/, done);
828
829 // The stack overflows. Load target address of the runtime stub and call it.
830 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order");
831 __ load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0);
832 __ mtctr(Rscratch1);
833 // Restore caller_sp (c2i adapter may exist, but no shrinking of interpreted caller frame).
834 #ifdef ASSERT
835 Label frame_not_shrunk;
836 __ cmpld(CCR0, R1_SP, R21_sender_SP);
837 __ ble(CCR0, frame_not_shrunk);
838 __ stop("frame shrunk");
839 __ bind(frame_not_shrunk);
840 __ ld(Rscratch1, 0, R1_SP);
841 __ ld(R0, 0, R21_sender_SP);
842 __ cmpd(CCR0, R0, Rscratch1);
843 __ asm_assert_eq("backlink");
844 #endif // ASSERT
845 __ mr(R1_SP, R21_sender_SP);
846 __ bctr();
847
848 __ align(32, 12);
849 __ bind(done);
850 BLOCK_COMMENT("} stack_overflow_check_with_compare");
851 }
852
853 // Lock the current method, interpreter register window must be set up!
lock_method(Register Rflags,Register Rscratch1,Register Rscratch2,bool flags_preloaded)854 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
855 const Register Robj_to_lock = Rscratch2;
856
857 {
858 if (!flags_preloaded) {
859 __ lwz(Rflags, method_(access_flags));
860 }
861
862 #ifdef ASSERT
863 // Check if methods needs synchronization.
864 {
865 Label Lok;
866 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
867 __ btrue(CCR0,Lok);
868 __ stop("method doesn't need synchronization");
869 __ bind(Lok);
870 }
871 #endif // ASSERT
872 }
873
874 // Get synchronization object to Rscratch2.
875 {
876 Label Lstatic;
877 Label Ldone;
878
879 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
880 __ btrue(CCR0, Lstatic);
881
882 // Non-static case: load receiver obj from stack and we're done.
883 __ ld(Robj_to_lock, R18_locals);
884 __ b(Ldone);
885
886 __ bind(Lstatic); // Static case: Lock the java mirror
887 // Load mirror from interpreter frame.
888 __ ld(Robj_to_lock, _abi(callers_sp), R1_SP);
889 __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock);
890
891 __ bind(Ldone);
892 __ verify_oop(Robj_to_lock);
893 }
894
895 // Got the oop to lock => execute!
896 __ add_monitor_to_stack(true, Rscratch1, R0);
897
898 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
899 __ lock_object(R26_monitor, Robj_to_lock);
900 }
901
902 // Generate a fixed interpreter frame for pure interpreter
903 // and I2N native transition frames.
904 //
905 // Before (stack grows downwards):
906 //
907 // | ... |
908 // |------------- |
909 // | java arg0 |
910 // | ... |
911 // | java argn |
912 // | | <- R15_esp
913 // | |
914 // |--------------|
915 // | abi_112 |
916 // | | <- R1_SP
917 // |==============|
918 //
919 //
920 // After:
921 //
922 // | ... |
923 // | java arg0 |<- R18_locals
924 // | ... |
925 // | java argn |
926 // |--------------|
927 // | |
928 // | java locals |
929 // | |
930 // |--------------|
931 // | abi_48 |
932 // |==============|
933 // | |
934 // | istate |
935 // | |
936 // |--------------|
937 // | monitor |<- R26_monitor
938 // |--------------|
939 // | |<- R15_esp
940 // | expression |
941 // | stack |
942 // | |
943 // |--------------|
944 // | |
945 // | abi_112 |<- R1_SP
946 // |==============|
947 //
948 // The top most frame needs an abi space of 112 bytes. This space is needed,
949 // since we call to c. The c function may spill their arguments to the caller
950 // frame. When we call to java, we don't need these spill slots. In order to save
951 // space on the stack, we resize the caller. However, java locals reside in
952 // the caller frame and the frame has to be increased. The frame_size for the
953 // current frame was calculated based on max_stack as size for the expression
954 // stack. At the call, just a part of the expression stack might be used.
955 // We don't want to waste this space and cut the frame back accordingly.
956 // The resulting amount for resizing is calculated as follows:
957 // resize = (number_of_locals - number_of_arguments) * slot_size
958 // + (R1_SP - R15_esp) + 48
959 //
960 // The size for the callee frame is calculated:
961 // framesize = 112 + max_stack + monitor + state_size
962 //
963 // maxstack: Max number of slots on the expression stack, loaded from the method.
964 // monitor: We statically reserve room for one monitor object.
965 // state_size: We save the current state of the interpreter to this area.
966 //
generate_fixed_frame(bool native_call,Register Rsize_of_parameters,Register Rsize_of_locals)967 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
968 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
969 top_frame_size = R7_ARG5,
970 Rconst_method = R8_ARG6;
971
972 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size);
973
974 __ ld(Rconst_method, method_(const));
975 __ lhz(Rsize_of_parameters /* number of params */,
976 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
977 if (native_call) {
978 // If we're calling a native method, we reserve space for the worst-case signature
979 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2).
980 // We add two slots to the parameter_count, one for the jni
981 // environment and one for a possible native mirror.
982 Label skip_native_calculate_max_stack;
983 __ addi(top_frame_size, Rsize_of_parameters, 2);
984 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters);
985 __ bge(CCR0, skip_native_calculate_max_stack);
986 __ li(top_frame_size, Argument::n_register_parameters);
987 __ bind(skip_native_calculate_max_stack);
988 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
989 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
990 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
991 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
992 } else {
993 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
994 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
995 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
996 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
997 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
998 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
999 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
1000 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1);
1001 }
1002
1003 // Compute top frame size.
1004 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size);
1005
1006 // Cut back area between esp and max_stack.
1007 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize);
1008
1009 __ round_to(top_frame_size, frame::alignment_in_bytes);
1010 __ round_to(parent_frame_resize, frame::alignment_in_bytes);
1011 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
1012 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
1013
1014 if (!native_call) {
1015 // Stack overflow check.
1016 // Native calls don't need the stack size check since they have no
1017 // expression stack and the arguments are already on the stack and
1018 // we only add a handful of words to the stack.
1019 __ add(R11_scratch1, parent_frame_resize, top_frame_size);
1020 generate_stack_overflow_check(R11_scratch1, R12_scratch2);
1021 }
1022
1023 // Set up interpreter state registers.
1024
1025 __ add(R18_locals, R15_esp, Rsize_of_parameters);
1026 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
1027 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache);
1028
1029 // Set method data pointer.
1030 if (ProfileInterpreter) {
1031 Label zero_continue;
1032 __ ld(R28_mdx, method_(method_data));
1033 __ cmpdi(CCR0, R28_mdx, 0);
1034 __ beq(CCR0, zero_continue);
1035 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1036 __ bind(zero_continue);
1037 }
1038
1039 if (native_call) {
1040 __ li(R14_bcp, 0); // Must initialize.
1041 } else {
1042 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method);
1043 }
1044
1045 // Resize parent frame.
1046 __ mflr(R12_scratch2);
1047 __ neg(parent_frame_resize, parent_frame_resize);
1048 __ resize_frame(parent_frame_resize, R11_scratch1);
1049 __ std(R12_scratch2, _abi(lr), R1_SP);
1050
1051 // Get mirror and store it in the frame as GC root for this Method*.
1052 __ load_mirror_from_const_method(R12_scratch2, Rconst_method);
1053
1054 __ addi(R26_monitor, R1_SP, -frame::ijava_state_size);
1055 __ addi(R15_esp, R26_monitor, -Interpreter::stackElementSize);
1056
1057 // Store values.
1058 __ std(R19_method, _ijava_state_neg(method), R1_SP);
1059 __ std(R12_scratch2, _ijava_state_neg(mirror), R1_SP);
1060 __ std(R18_locals, _ijava_state_neg(locals), R1_SP);
1061 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
1062
1063 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
1064 // be found in the frame after save_interpreter_state is done. This is always true
1065 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
1066 // because e.g. frame::interpreter_frame_bcp() will not access the correct value
1067 // (Enhanced Stack Trace).
1068 // The signal handler does not save the interpreter state into the frame.
1069
1070 // We have to initialize some of these frame slots for native calls (accessed by GC).
1071 // Also initialize them for non-native calls for better tool support (even though
1072 // you may not get the most recent version as described above).
1073 __ li(R0, 0);
1074 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP);
1075 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
1076 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
1077 __ std(R15_esp, _ijava_state_neg(esp), R1_SP);
1078 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); // only used for native_call
1079
1080 // Store sender's SP and this frame's top SP.
1081 __ subf(R12_scratch2, top_frame_size, R1_SP);
1082 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
1083 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
1084
1085 // Push top frame.
1086 __ push_frame(top_frame_size, R11_scratch1);
1087 }
1088
1089 // End of helpers
1090
generate_math_entry(AbstractInterpreter::MethodKind kind)1091 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1092
1093 // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1094 bool use_instruction = false;
1095 address runtime_entry = NULL;
1096 int num_args = 1;
1097 bool double_precision = true;
1098
1099 // PPC64 specific:
1100 switch (kind) {
1101 case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break;
1102 case Interpreter::java_lang_math_abs: use_instruction = true; break;
1103 case Interpreter::java_lang_math_fmaF:
1104 case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1105 default: break; // Fall back to runtime call.
1106 }
1107
1108 switch (kind) {
1109 case Interpreter::java_lang_math_sin : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); break;
1110 case Interpreter::java_lang_math_cos : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); break;
1111 case Interpreter::java_lang_math_tan : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); break;
1112 case Interpreter::java_lang_math_abs : /* run interpreted */ break;
1113 case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); break;
1114 case Interpreter::java_lang_math_log : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); break;
1115 case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1116 case Interpreter::java_lang_math_pow : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1117 case Interpreter::java_lang_math_exp : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); break;
1118 case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1119 case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1120 default: ShouldNotReachHere();
1121 }
1122
1123 // Use normal entry if neither instruction nor runtime call is used.
1124 if (!use_instruction && runtime_entry == NULL) return NULL;
1125
1126 address entry = __ pc();
1127
1128 // Load arguments
1129 assert(num_args <= 13, "passed in registers");
1130 if (double_precision) {
1131 int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1132 for (int i = 0; i < num_args; ++i) {
1133 __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1134 offset -= 2 * Interpreter::stackElementSize;
1135 }
1136 } else {
1137 int offset = num_args * Interpreter::stackElementSize;
1138 for (int i = 0; i < num_args; ++i) {
1139 __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1140 offset -= Interpreter::stackElementSize;
1141 }
1142 }
1143
1144 if (use_instruction) {
1145 switch (kind) {
1146 case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1); break;
1147 case Interpreter::java_lang_math_abs: __ fabs(F1_RET, F1); break;
1148 case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break;
1149 case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3); break;
1150 default: ShouldNotReachHere();
1151 }
1152 } else {
1153 // Comment: Can use tail call if the unextended frame is always C ABI compliant:
1154 //__ load_const_optimized(R12_scratch2, runtime_entry, R0);
1155 //__ call_c_and_return_to_caller(R12_scratch2);
1156
1157 // Push a new C frame and save LR.
1158 __ save_LR_CR(R0);
1159 __ push_frame_reg_args(0, R11_scratch1);
1160
1161 __ call_VM_leaf(runtime_entry);
1162
1163 // Pop the C frame and restore LR.
1164 __ pop_frame();
1165 __ restore_LR_CR(R0);
1166 }
1167
1168 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1169 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1170 __ blr();
1171
1172 __ flush();
1173
1174 return entry;
1175 }
1176
bang_stack_shadow_pages(bool native_call)1177 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1178 // Quick & dirty stack overflow checking: bang the stack & handle trap.
1179 // Note that we do the banging after the frame is setup, since the exception
1180 // handling code expects to find a valid interpreter frame on the stack.
1181 // Doing the banging earlier fails if the caller frame is not an interpreter
1182 // frame.
1183 // (Also, the exception throwing code expects to unlock any synchronized
1184 // method receiever, so do the banging after locking the receiver.)
1185
1186 // Bang each page in the shadow zone. We can't assume it's been done for
1187 // an interpreter frame with greater than a page of locals, so each page
1188 // needs to be checked. Only true for non-native.
1189 if (UseStackBanging) {
1190 const int page_size = os::vm_page_size();
1191 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
1192 const int start_page = native_call ? n_shadow_pages : 1;
1193 BLOCK_COMMENT("bang_stack_shadow_pages:");
1194 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1195 __ bang_stack_with_offset(pages*page_size);
1196 }
1197 }
1198 }
1199
1200 // Interpreter stub for calling a native method. (asm interpreter)
1201 // This sets up a somewhat different looking stack for calling the
1202 // native method than the typical interpreter frame setup.
1203 //
1204 // On entry:
1205 // R19_method - method
1206 // R16_thread - JavaThread*
1207 // R15_esp - intptr_t* sender tos
1208 //
1209 // abstract stack (grows up)
1210 // [ IJava (caller of JNI callee) ] <-- ASP
1211 // ...
generate_native_entry(bool synchronized)1212 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1213
1214 address entry = __ pc();
1215
1216 const bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1217
1218 // -----------------------------------------------------------------------------
1219 // Allocate a new frame that represents the native callee (i2n frame).
1220 // This is not a full-blown interpreter frame, but in particular, the
1221 // following registers are valid after this:
1222 // - R19_method
1223 // - R18_local (points to start of arguments to native function)
1224 //
1225 // abstract stack (grows up)
1226 // [ IJava (caller of JNI callee) ] <-- ASP
1227 // ...
1228
1229 const Register signature_handler_fd = R11_scratch1;
1230 const Register pending_exception = R0;
1231 const Register result_handler_addr = R31;
1232 const Register native_method_fd = R11_scratch1;
1233 const Register access_flags = R22_tmp2;
1234 const Register active_handles = R11_scratch1; // R26_monitor saved to state.
1235 const Register sync_state = R12_scratch2;
1236 const Register sync_state_addr = sync_state; // Address is dead after use.
1237 const Register suspend_flags = R11_scratch1;
1238
1239 //=============================================================================
1240 // Allocate new frame and initialize interpreter state.
1241
1242 Label exception_return;
1243 Label exception_return_sync_check;
1244 Label stack_overflow_return;
1245
1246 // Generate new interpreter state and jump to stack_overflow_return in case of
1247 // a stack overflow.
1248 //generate_compute_interpreter_state(stack_overflow_return);
1249
1250 Register size_of_parameters = R22_tmp2;
1251
1252 generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
1253
1254 //=============================================================================
1255 // Increment invocation counter. On overflow, entry to JNI method
1256 // will be compiled.
1257 Label invocation_counter_overflow, continue_after_compile;
1258 if (inc_counter) {
1259 if (synchronized) {
1260 // Since at this point in the method invocation the exception handler
1261 // would try to exit the monitor of synchronized methods which hasn't
1262 // been entered yet, we set the thread local variable
1263 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1264 // runtime, exception handling i.e. unlock_if_synchronized_method will
1265 // check this thread local flag.
1266 // This flag has two effects, one is to force an unwind in the topmost
1267 // interpreter frame and not perform an unlock while doing so.
1268 __ li(R0, 1);
1269 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1270 }
1271 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1272
1273 BIND(continue_after_compile);
1274 }
1275
1276 bang_stack_shadow_pages(true);
1277
1278 if (inc_counter) {
1279 // Reset the _do_not_unlock_if_synchronized flag.
1280 if (synchronized) {
1281 __ li(R0, 0);
1282 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1283 }
1284 }
1285
1286 // access_flags = method->access_flags();
1287 // Load access flags.
1288 assert(access_flags->is_nonvolatile(),
1289 "access_flags must be in a non-volatile register");
1290 // Type check.
1291 assert(4 == sizeof(AccessFlags), "unexpected field size");
1292 __ lwz(access_flags, method_(access_flags));
1293
1294 // We don't want to reload R19_method and access_flags after calls
1295 // to some helper functions.
1296 assert(R19_method->is_nonvolatile(),
1297 "R19_method must be a non-volatile register");
1298
1299 // Check for synchronized methods. Must happen AFTER invocation counter
1300 // check, so method is not locked if counter overflows.
1301
1302 if (synchronized) {
1303 lock_method(access_flags, R11_scratch1, R12_scratch2, true);
1304
1305 // Update monitor in state.
1306 __ ld(R11_scratch1, 0, R1_SP);
1307 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1);
1308 }
1309
1310 // jvmti/jvmpi support
1311 __ notify_method_entry();
1312
1313 //=============================================================================
1314 // Get and call the signature handler.
1315
1316 __ ld(signature_handler_fd, method_(signature_handler));
1317 Label call_signature_handler;
1318
1319 __ cmpdi(CCR0, signature_handler_fd, 0);
1320 __ bne(CCR0, call_signature_handler);
1321
1322 // Method has never been called. Either generate a specialized
1323 // handler or point to the slow one.
1324 //
1325 // Pass parameter 'false' to avoid exception check in call_VM.
1326 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
1327
1328 // Check for an exception while looking up the target method. If we
1329 // incurred one, bail.
1330 __ ld(pending_exception, thread_(pending_exception));
1331 __ cmpdi(CCR0, pending_exception, 0);
1332 __ bne(CCR0, exception_return_sync_check); // Has pending exception.
1333
1334 // Reload signature handler, it may have been created/assigned in the meanwhile.
1335 __ ld(signature_handler_fd, method_(signature_handler));
1336 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
1337
1338 BIND(call_signature_handler);
1339
1340 // Before we call the signature handler we push a new frame to
1341 // protect the interpreter frame volatile registers when we return
1342 // from jni but before we can get back to Java.
1343
1344 // First set the frame anchor while the SP/FP registers are
1345 // convenient and the slow signature handler can use this same frame
1346 // anchor.
1347
1348 // We have a TOP_IJAVA_FRAME here, which belongs to us.
1349 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
1350
1351 // Now the interpreter frame (and its call chain) have been
1352 // invalidated and flushed. We are now protected against eager
1353 // being enabled in native code. Even if it goes eager the
1354 // registers will be reloaded as clean and we will invalidate after
1355 // the call so no spurious flush should be possible.
1356
1357 // Call signature handler and pass locals address.
1358 //
1359 // Our signature handlers copy required arguments to the C stack
1360 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
1361 __ mr(R3_ARG1, R18_locals);
1362 #if !defined(ABI_ELFv2)
1363 __ ld(signature_handler_fd, 0, signature_handler_fd);
1364 #endif
1365
1366 __ call_stub(signature_handler_fd);
1367
1368 // Remove the register parameter varargs slots we allocated in
1369 // compute_interpreter_state. SP+16 ends up pointing to the ABI
1370 // outgoing argument area.
1371 //
1372 // Not needed on PPC64.
1373 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord);
1374
1375 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
1376 // Save across call to native method.
1377 __ mr(result_handler_addr, R3_RET);
1378
1379 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
1380
1381 // Set up fixed parameters and call the native method.
1382 // If the method is static, get mirror into R4_ARG2.
1383 {
1384 Label method_is_not_static;
1385 // Access_flags is non-volatile and still, no need to restore it.
1386
1387 // Restore access flags.
1388 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
1389 __ bfalse(CCR0, method_is_not_static);
1390
1391 __ ld(R11_scratch1, _abi(callers_sp), R1_SP);
1392 // Load mirror from interpreter frame.
1393 __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1);
1394 // R4_ARG2 = &state->_oop_temp;
1395 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
1396 __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
1397 BIND(method_is_not_static);
1398 }
1399
1400 // At this point, arguments have been copied off the stack into
1401 // their JNI positions. Oops are boxed in-place on the stack, with
1402 // handles copied to arguments. The result handler address is in a
1403 // register.
1404
1405 // Pass JNIEnv address as first parameter.
1406 __ addir(R3_ARG1, thread_(jni_environment));
1407
1408 // Load the native_method entry before we change the thread state.
1409 __ ld(native_method_fd, method_(native_function));
1410
1411 //=============================================================================
1412 // Transition from _thread_in_Java to _thread_in_native. As soon as
1413 // we make this change the safepoint code needs to be certain that
1414 // the last Java frame we established is good. The pc in that frame
1415 // just needs to be near here not an actual return address.
1416
1417 // We use release_store_fence to update values like the thread state, where
1418 // we don't want the current thread to continue until all our prior memory
1419 // accesses (including the new thread state) are visible to other threads.
1420 __ li(R0, _thread_in_native);
1421 __ release();
1422
1423 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
1424 __ stw(R0, thread_(thread_state));
1425
1426 //=============================================================================
1427 // Call the native method. Argument registers must not have been
1428 // overwritten since "__ call_stub(signature_handler);" (except for
1429 // ARG1 and ARG2 for static methods).
1430 __ call_c(native_method_fd);
1431
1432 __ li(R0, 0);
1433 __ ld(R11_scratch1, 0, R1_SP);
1434 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1435 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1436 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
1437
1438 // Note: C++ interpreter needs the following here:
1439 // The frame_manager_lr field, which we use for setting the last
1440 // java frame, gets overwritten by the signature handler. Restore
1441 // it now.
1442 //__ get_PC_trash_LR(R11_scratch1);
1443 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
1444
1445 // Because of GC R19_method may no longer be valid.
1446
1447 // Block, if necessary, before resuming in _thread_in_Java state.
1448 // In order for GC to work, don't clear the last_Java_sp until after
1449 // blocking.
1450
1451 //=============================================================================
1452 // Switch thread to "native transition" state before reading the
1453 // synchronization state. This additional state is necessary
1454 // because reading and testing the synchronization state is not
1455 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1456 // in _thread_in_native state, loads _not_synchronized and is
1457 // preempted. VM thread changes sync state to synchronizing and
1458 // suspends threads for GC. Thread A is resumed to finish this
1459 // native method, but doesn't block here since it didn't see any
1460 // synchronization in progress, and escapes.
1461
1462 // We use release_store_fence to update values like the thread state, where
1463 // we don't want the current thread to continue until all our prior memory
1464 // accesses (including the new thread state) are visible to other threads.
1465 __ li(R0/*thread_state*/, _thread_in_native_trans);
1466 __ release();
1467 __ stw(R0/*thread_state*/, thread_(thread_state));
1468 __ fence();
1469
1470 // Now before we return to java we must look for a current safepoint
1471 // (a new safepoint can not start since we entered native_trans).
1472 // We must check here because a current safepoint could be modifying
1473 // the callers registers right this moment.
1474
1475 // Acquire isn't strictly necessary here because of the fence, but
1476 // sync_state is declared to be volatile, so we do it anyway
1477 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1478
1479 Label do_safepoint, sync_check_done;
1480 // No synchronization in progress nor yet synchronized.
1481 __ safepoint_poll(do_safepoint, sync_state);
1482
1483 // Not suspended.
1484 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1485 __ lwz(suspend_flags, thread_(suspend_flags));
1486 __ cmpwi(CCR1, suspend_flags, 0);
1487 __ beq(CCR1, sync_check_done);
1488
1489 __ bind(do_safepoint);
1490 __ isync();
1491 // Block. We do the call directly and leave the current
1492 // last_Java_frame setup undisturbed. We must save any possible
1493 // native result across the call. No oop is present.
1494
1495 __ mr(R3_ARG1, R16_thread);
1496 #if defined(ABI_ELFv2)
1497 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
1498 relocInfo::none);
1499 #else
1500 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
1501 relocInfo::none);
1502 #endif
1503
1504 __ bind(sync_check_done);
1505
1506 //=============================================================================
1507 // <<<<<< Back in Interpreter Frame >>>>>
1508
1509 // We are in thread_in_native_trans here and back in the normal
1510 // interpreter frame. We don't have to do anything special about
1511 // safepoints and we can switch to Java mode anytime we are ready.
1512
1513 // Note: frame::interpreter_frame_result has a dependency on how the
1514 // method result is saved across the call to post_method_exit. For
1515 // native methods it assumes that the non-FPU/non-void result is
1516 // saved in _native_lresult and a FPU result in _native_fresult. If
1517 // this changes then the interpreter_frame_result implementation
1518 // will need to be updated too.
1519
1520 // On PPC64, we have stored the result directly after the native call.
1521
1522 //=============================================================================
1523 // Back in Java
1524
1525 // We use release_store_fence to update values like the thread state, where
1526 // we don't want the current thread to continue until all our prior memory
1527 // accesses (including the new thread state) are visible to other threads.
1528 __ li(R0/*thread_state*/, _thread_in_Java);
1529 __ lwsync(); // Acquire safepoint and suspend state, release thread state.
1530 __ stw(R0/*thread_state*/, thread_(thread_state));
1531
1532 if (CheckJNICalls) {
1533 // clear_pending_jni_exception_check
1534 __ load_const_optimized(R0, 0L);
1535 __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread);
1536 }
1537
1538 __ reset_last_Java_frame();
1539
1540 // Jvmdi/jvmpi support. Whether we've got an exception pending or
1541 // not, and whether unlocking throws an exception or not, we notify
1542 // on native method exit. If we do have an exception, we'll end up
1543 // in the caller's context to handle it, so if we don't do the
1544 // notify here, we'll drop it on the floor.
1545 __ notify_method_exit(true/*native method*/,
1546 ilgl /*illegal state (not used for native methods)*/,
1547 InterpreterMacroAssembler::NotifyJVMTI,
1548 false /*check_exceptions*/);
1549
1550 //=============================================================================
1551 // Handle exceptions
1552
1553 if (synchronized) {
1554 // Don't check for exceptions since we're still in the i2n frame. Do that
1555 // manually afterwards.
1556 __ unlock_object(R26_monitor, false); // Can also unlock methods.
1557 }
1558
1559 // Reset active handles after returning from native.
1560 // thread->active_handles()->clear();
1561 __ ld(active_handles, thread_(active_handles));
1562 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1563 __ li(R0, 0);
1564 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles);
1565
1566 Label exception_return_sync_check_already_unlocked;
1567 __ ld(R0/*pending_exception*/, thread_(pending_exception));
1568 __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1569 __ bne(CCR0, exception_return_sync_check_already_unlocked);
1570
1571 //-----------------------------------------------------------------------------
1572 // No exception pending.
1573
1574 // Move native method result back into proper registers and return.
1575 // Invoke result handler (may unbox/promote).
1576 __ ld(R11_scratch1, 0, R1_SP);
1577 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1578 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1579 __ call_stub(result_handler_addr);
1580
1581 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1582
1583 // Must use the return pc which was loaded from the caller's frame
1584 // as the VM uses return-pc-patching for deoptimization.
1585 __ mtlr(R0);
1586 __ blr();
1587
1588 //-----------------------------------------------------------------------------
1589 // An exception is pending. We call into the runtime only if the
1590 // caller was not interpreted. If it was interpreted the
1591 // interpreter will do the correct thing. If it isn't interpreted
1592 // (call stub/compiled code) we will change our return and continue.
1593
1594 BIND(exception_return_sync_check);
1595
1596 if (synchronized) {
1597 // Don't check for exceptions since we're still in the i2n frame. Do that
1598 // manually afterwards.
1599 __ unlock_object(R26_monitor, false); // Can also unlock methods.
1600 }
1601 BIND(exception_return_sync_check_already_unlocked);
1602
1603 const Register return_pc = R31;
1604
1605 __ ld(return_pc, 0, R1_SP);
1606 __ ld(return_pc, _abi(lr), return_pc);
1607
1608 // Get the address of the exception handler.
1609 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1610 R16_thread,
1611 return_pc /* return pc */);
1612 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1613
1614 // Load the PC of the the exception handler into LR.
1615 __ mtlr(R3_RET);
1616
1617 // Load exception into R3_ARG1 and clear pending exception in thread.
1618 __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1619 __ li(R4_ARG2, 0);
1620 __ std(R4_ARG2, thread_(pending_exception));
1621
1622 // Load the original return pc into R4_ARG2.
1623 __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1624
1625 // Return to exception handler.
1626 __ blr();
1627
1628 //=============================================================================
1629 // Counter overflow.
1630
1631 if (inc_counter) {
1632 // Handle invocation counter overflow.
1633 __ bind(invocation_counter_overflow);
1634
1635 generate_counter_overflow(continue_after_compile);
1636 }
1637
1638 return entry;
1639 }
1640
1641 // Generic interpreted method entry to (asm) interpreter.
1642 //
generate_normal_entry(bool synchronized)1643 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1644 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1645 address entry = __ pc();
1646 // Generate the code to allocate the interpreter stack frame.
1647 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1648 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame.
1649
1650 // Does also a stack check to assure this frame fits on the stack.
1651 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1652
1653 // --------------------------------------------------------------------------
1654 // Zero out non-parameter locals.
1655 // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1656 // worth to ask the flag, just do it.
1657 Register Rslot_addr = R6_ARG4,
1658 Rnum = R7_ARG5;
1659 Label Lno_locals, Lzero_loop;
1660
1661 // Set up the zeroing loop.
1662 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1663 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1664 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1665 __ beq(CCR0, Lno_locals);
1666 __ li(R0, 0);
1667 __ mtctr(Rnum);
1668
1669 // The zero locals loop.
1670 __ bind(Lzero_loop);
1671 __ std(R0, 0, Rslot_addr);
1672 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1673 __ bdnz(Lzero_loop);
1674
1675 __ bind(Lno_locals);
1676
1677 // --------------------------------------------------------------------------
1678 // Counter increment and overflow check.
1679 Label invocation_counter_overflow,
1680 profile_method,
1681 profile_method_continue;
1682 if (inc_counter || ProfileInterpreter) {
1683
1684 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1685 if (synchronized) {
1686 // Since at this point in the method invocation the exception handler
1687 // would try to exit the monitor of synchronized methods which hasn't
1688 // been entered yet, we set the thread local variable
1689 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1690 // runtime, exception handling i.e. unlock_if_synchronized_method will
1691 // check this thread local flag.
1692 // This flag has two effects, one is to force an unwind in the topmost
1693 // interpreter frame and not perform an unlock while doing so.
1694 __ li(R0, 1);
1695 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1696 }
1697
1698 // Argument and return type profiling.
1699 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1700
1701 // Increment invocation counter and check for overflow.
1702 if (inc_counter) {
1703 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1704 }
1705
1706 __ bind(profile_method_continue);
1707 }
1708
1709 bang_stack_shadow_pages(false);
1710
1711 if (inc_counter || ProfileInterpreter) {
1712 // Reset the _do_not_unlock_if_synchronized flag.
1713 if (synchronized) {
1714 __ li(R0, 0);
1715 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1716 }
1717 }
1718
1719 // --------------------------------------------------------------------------
1720 // Locking of synchronized methods. Must happen AFTER invocation_counter
1721 // check and stack overflow check, so method is not locked if overflows.
1722 if (synchronized) {
1723 lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1724 }
1725 #ifdef ASSERT
1726 else {
1727 Label Lok;
1728 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1729 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1730 __ asm_assert_eq("method needs synchronization");
1731 __ bind(Lok);
1732 }
1733 #endif // ASSERT
1734
1735 __ verify_thread();
1736
1737 // --------------------------------------------------------------------------
1738 // JVMTI support
1739 __ notify_method_entry();
1740
1741 // --------------------------------------------------------------------------
1742 // Start executing instructions.
1743 __ dispatch_next(vtos);
1744
1745 // --------------------------------------------------------------------------
1746 // Out of line counter overflow and MDO creation code.
1747 if (ProfileInterpreter) {
1748 // We have decided to profile this method in the interpreter.
1749 __ bind(profile_method);
1750 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1751 __ set_method_data_pointer_for_bcp();
1752 __ b(profile_method_continue);
1753 }
1754
1755 if (inc_counter) {
1756 // Handle invocation counter overflow.
1757 __ bind(invocation_counter_overflow);
1758 generate_counter_overflow(profile_method_continue);
1759 }
1760 return entry;
1761 }
1762
1763 // CRC32 Intrinsics.
1764 //
1765 // Contract on scratch and work registers.
1766 // =======================================
1767 //
1768 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1769 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1770 // You can't rely on these registers across calls.
1771 //
1772 // The generators for CRC32_update and for CRC32_updateBytes use the
1773 // scratch/work register set internally, passing the work registers
1774 // as arguments to the MacroAssembler emitters as required.
1775 //
1776 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1777 // Their contents is not constant but may change according to the requirements
1778 // of the emitted code.
1779 //
1780 // All other registers from the scratch/work register set are used "internally"
1781 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1782 // Basically, only R3_RET contains a defined value which is the function result.
1783 //
1784 /**
1785 * Method entry for static native methods:
1786 * int java.util.zip.CRC32.update(int crc, int b)
1787 */
generate_CRC32_update_entry()1788 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1789 if (UseCRC32Intrinsics) {
1790 address start = __ pc(); // Remember stub start address (is rtn value).
1791 Label slow_path;
1792
1793 // Safepoint check
1794 const Register sync_state = R11_scratch1;
1795 __ safepoint_poll(slow_path, sync_state);
1796
1797 // We don't generate local frame and don't align stack because
1798 // we not even call stub code (we generate the code inline)
1799 // and there is no safepoint on this path.
1800
1801 // Load java parameters.
1802 // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1803 const Register argP = R15_esp;
1804 const Register crc = R3_ARG1; // crc value
1805 const Register data = R4_ARG2;
1806 const Register table = R5_ARG3; // address of crc32 table
1807
1808 BLOCK_COMMENT("CRC32_update {");
1809
1810 // Arguments are reversed on java expression stack
1811 #ifdef VM_LITTLE_ENDIAN
1812 int data_offs = 0+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1813 // Being passed as an int, the single byte is at offset +0.
1814 #else
1815 int data_offs = 3+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1816 // Being passed from java as an int, the single byte is at offset +3.
1817 #endif
1818 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1819 __ lbz(data, data_offs, argP); // Byte from buffer, zero-extended.
1820 __ load_const_optimized(table, StubRoutines::crc_table_addr(), R0);
1821 __ kernel_crc32_singleByteReg(crc, data, table, true);
1822
1823 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1824 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1825 __ blr();
1826
1827 // Generate a vanilla native entry as the slow path.
1828 BLOCK_COMMENT("} CRC32_update");
1829 BIND(slow_path);
1830 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1831 return start;
1832 }
1833
1834 return NULL;
1835 }
1836
1837 /**
1838 * Method entry for static native methods:
1839 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len)
1840 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1841 */
generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind)1842 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1843 if (UseCRC32Intrinsics) {
1844 address start = __ pc(); // Remember stub start address (is rtn value).
1845 Label slow_path;
1846
1847 // Safepoint check
1848 const Register sync_state = R11_scratch1;
1849 __ safepoint_poll(slow_path, sync_state);
1850
1851 // We don't generate local frame and don't align stack because
1852 // we not even call stub code (we generate the code inline)
1853 // and there is no safepoint on this path.
1854
1855 // Load parameters.
1856 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1857 const Register argP = R15_esp;
1858 const Register crc = R3_ARG1; // crc value
1859 const Register data = R4_ARG2; // address of java byte array
1860 const Register dataLen = R5_ARG3; // source data len
1861 const Register tmp = R11_scratch1;
1862
1863 // Arguments are reversed on java expression stack.
1864 // Calculate address of start element.
1865 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1866 BLOCK_COMMENT("CRC32_updateByteBuffer {");
1867 // crc @ (SP + 5W) (32bit)
1868 // buf @ (SP + 3W) (64bit ptr to long array)
1869 // off @ (SP + 2W) (32bit)
1870 // dataLen @ (SP + 1W) (32bit)
1871 // data = buf + off
1872 __ ld( data, 3*wordSize, argP); // start of byte buffer
1873 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1874 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1875 __ lwz( crc, 5*wordSize, argP); // current crc state
1876 __ add( data, data, tmp); // Add byte buffer offset.
1877 } else { // Used for "updateBytes update".
1878 BLOCK_COMMENT("CRC32_updateBytes {");
1879 // crc @ (SP + 4W) (32bit)
1880 // buf @ (SP + 3W) (64bit ptr to byte array)
1881 // off @ (SP + 2W) (32bit)
1882 // dataLen @ (SP + 1W) (32bit)
1883 // data = buf + off + base_offset
1884 __ ld( data, 3*wordSize, argP); // start of byte buffer
1885 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1886 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1887 __ add( data, data, tmp); // add byte buffer offset
1888 __ lwz( crc, 4*wordSize, argP); // current crc state
1889 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1890 }
1891
1892 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, false);
1893
1894 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1895 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1896 __ blr();
1897
1898 // Generate a vanilla native entry as the slow path.
1899 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1900 BIND(slow_path);
1901 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1902 return start;
1903 }
1904
1905 return NULL;
1906 }
1907
1908
1909 /**
1910 * Method entry for intrinsic-candidate (non-native) methods:
1911 * int java.util.zip.CRC32C.updateBytes( int crc, byte[] b, int off, int end)
1912 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
1913 * Unlike CRC32, CRC32C does not have any methods marked as native
1914 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1915 **/
generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind)1916 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1917 if (UseCRC32CIntrinsics) {
1918 address start = __ pc(); // Remember stub start address (is rtn value).
1919
1920 // We don't generate local frame and don't align stack because
1921 // we not even call stub code (we generate the code inline)
1922 // and there is no safepoint on this path.
1923
1924 // Load parameters.
1925 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1926 const Register argP = R15_esp;
1927 const Register crc = R3_ARG1; // crc value
1928 const Register data = R4_ARG2; // address of java byte array
1929 const Register dataLen = R5_ARG3; // source data len
1930 const Register tmp = R11_scratch1;
1931
1932 // Arguments are reversed on java expression stack.
1933 // Calculate address of start element.
1934 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer".
1935 BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
1936 // crc @ (SP + 5W) (32bit)
1937 // buf @ (SP + 3W) (64bit ptr to long array)
1938 // off @ (SP + 2W) (32bit)
1939 // dataLen @ (SP + 1W) (32bit)
1940 // data = buf + off
1941 __ ld( data, 3*wordSize, argP); // start of byte buffer
1942 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1943 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1944 __ lwz( crc, 5*wordSize, argP); // current crc state
1945 __ add( data, data, tmp); // Add byte buffer offset.
1946 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1947 } else { // Used for "updateBytes update".
1948 BLOCK_COMMENT("CRC32C_updateBytes {");
1949 // crc @ (SP + 4W) (32bit)
1950 // buf @ (SP + 3W) (64bit ptr to byte array)
1951 // off @ (SP + 2W) (32bit)
1952 // dataLen @ (SP + 1W) (32bit)
1953 // data = buf + off + base_offset
1954 __ ld( data, 3*wordSize, argP); // start of byte buffer
1955 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1956 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1957 __ add( data, data, tmp); // add byte buffer offset
1958 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1959 __ lwz( crc, 4*wordSize, argP); // current crc state
1960 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1961 }
1962
1963 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, true);
1964
1965 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1966 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1967 __ blr();
1968
1969 BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
1970 return start;
1971 }
1972
1973 return NULL;
1974 }
1975
1976 // =============================================================================
1977 // Exceptions
1978
generate_throw_exception()1979 void TemplateInterpreterGenerator::generate_throw_exception() {
1980 Register Rexception = R17_tos,
1981 Rcontinuation = R3_RET;
1982
1983 // --------------------------------------------------------------------------
1984 // Entry point if an method returns with a pending exception (rethrow).
1985 Interpreter::_rethrow_exception_entry = __ pc();
1986 {
1987 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
1988 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
1989 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
1990
1991 // Compiled code destroys templateTableBase, reload.
1992 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
1993 }
1994
1995 // Entry point if a interpreted method throws an exception (throw).
1996 Interpreter::_throw_exception_entry = __ pc();
1997 {
1998 __ mr(Rexception, R3_RET);
1999
2000 __ verify_thread();
2001 __ verify_oop(Rexception);
2002
2003 // Expression stack must be empty before entering the VM in case of an exception.
2004 __ empty_expression_stack();
2005 // Find exception handler address and preserve exception oop.
2006 // Call C routine to find handler and jump to it.
2007 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
2008 __ mtctr(Rcontinuation);
2009 // Push exception for exception handler bytecodes.
2010 __ push_ptr(Rexception);
2011
2012 // Jump to exception handler (may be remove activation entry!).
2013 __ bctr();
2014 }
2015
2016 // If the exception is not handled in the current frame the frame is
2017 // removed and the exception is rethrown (i.e. exception
2018 // continuation is _rethrow_exception).
2019 //
2020 // Note: At this point the bci is still the bxi for the instruction
2021 // which caused the exception and the expression stack is
2022 // empty. Thus, for any VM calls at this point, GC will find a legal
2023 // oop map (with empty expression stack).
2024
2025 // In current activation
2026 // tos: exception
2027 // bcp: exception bcp
2028
2029 // --------------------------------------------------------------------------
2030 // JVMTI PopFrame support
2031
2032 Interpreter::_remove_activation_preserving_args_entry = __ pc();
2033 {
2034 // Set the popframe_processing bit in popframe_condition indicating that we are
2035 // currently handling popframe, so that call_VMs that may happen later do not
2036 // trigger new popframe handling cycles.
2037 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2038 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
2039 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2040
2041 // Empty the expression stack, as in normal exception handling.
2042 __ empty_expression_stack();
2043 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
2044
2045 // Check to see whether we are returning to a deoptimized frame.
2046 // (The PopFrame call ensures that the caller of the popped frame is
2047 // either interpreted or compiled and deoptimizes it if compiled.)
2048 // Note that we don't compare the return PC against the
2049 // deoptimization blob's unpack entry because of the presence of
2050 // adapter frames in C2.
2051 Label Lcaller_not_deoptimized;
2052 Register return_pc = R3_ARG1;
2053 __ ld(return_pc, 0, R1_SP);
2054 __ ld(return_pc, _abi(lr), return_pc);
2055 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
2056 __ cmpdi(CCR0, R3_RET, 0);
2057 __ bne(CCR0, Lcaller_not_deoptimized);
2058
2059 // The deoptimized case.
2060 // In this case, we can't call dispatch_next() after the frame is
2061 // popped, but instead must save the incoming arguments and restore
2062 // them after deoptimization has occurred.
2063 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
2064 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
2065 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
2066 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
2067 __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
2068 // Save these arguments.
2069 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
2070
2071 // Inform deoptimization that it is responsible for restoring these arguments.
2072 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
2073 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2074
2075 // Return from the current method into the deoptimization blob. Will eventually
2076 // end up in the deopt interpeter entry, deoptimization prepared everything that
2077 // we will reexecute the call that called us.
2078 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
2079 __ mtlr(return_pc);
2080 __ blr();
2081
2082 // The non-deoptimized case.
2083 __ bind(Lcaller_not_deoptimized);
2084
2085 // Clear the popframe condition flag.
2086 __ li(R0, 0);
2087 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2088
2089 // Get out of the current method and re-execute the call that called us.
2090 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2091 __ restore_interpreter_state(R11_scratch1);
2092 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
2093 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
2094 if (ProfileInterpreter) {
2095 __ set_method_data_pointer_for_bcp();
2096 __ ld(R11_scratch1, 0, R1_SP);
2097 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
2098 }
2099 #if INCLUDE_JVMTI
2100 Label L_done;
2101
2102 __ lbz(R11_scratch1, 0, R14_bcp);
2103 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
2104 __ bne(CCR0, L_done);
2105
2106 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2107 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
2108 __ ld(R4_ARG2, 0, R18_locals);
2109 __ call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp);
2110 __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2111 __ cmpdi(CCR0, R4_ARG2, 0);
2112 __ beq(CCR0, L_done);
2113 __ std(R4_ARG2, wordSize, R15_esp);
2114 __ bind(L_done);
2115 #endif // INCLUDE_JVMTI
2116 __ dispatch_next(vtos);
2117 }
2118 // end of JVMTI PopFrame support
2119
2120 // --------------------------------------------------------------------------
2121 // Remove activation exception entry.
2122 // This is jumped to if an interpreted method can't handle an exception itself
2123 // (we come from the throw/rethrow exception entry above). We're going to call
2124 // into the VM to find the exception handler in the caller, pop the current
2125 // frame and return the handler we calculated.
2126 Interpreter::_remove_activation_entry = __ pc();
2127 {
2128 __ pop_ptr(Rexception);
2129 __ verify_thread();
2130 __ verify_oop(Rexception);
2131 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
2132
2133 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
2134 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
2135
2136 __ get_vm_result(Rexception);
2137
2138 // We are done with this activation frame; find out where to go next.
2139 // The continuation point will be an exception handler, which expects
2140 // the following registers set up:
2141 //
2142 // RET: exception oop
2143 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
2144
2145 Register return_pc = R31; // Needs to survive the runtime call.
2146 __ ld(return_pc, 0, R1_SP);
2147 __ ld(return_pc, _abi(lr), return_pc);
2148 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
2149
2150 // Remove the current activation.
2151 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2152
2153 __ mr(R4_ARG2, return_pc);
2154 __ mtlr(R3_RET);
2155 __ mr(R3_RET, Rexception);
2156 __ blr();
2157 }
2158 }
2159
2160 // JVMTI ForceEarlyReturn support.
2161 // Returns "in the middle" of a method with a "fake" return value.
generate_earlyret_entry_for(TosState state)2162 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2163
2164 Register Rscratch1 = R11_scratch1,
2165 Rscratch2 = R12_scratch2;
2166
2167 address entry = __ pc();
2168 __ empty_expression_stack();
2169
2170 __ load_earlyret_value(state, Rscratch1);
2171
2172 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
2173 // Clear the earlyret state.
2174 __ li(R0, 0);
2175 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
2176
2177 __ remove_activation(state, false, false);
2178 // Copied from TemplateTable::_return.
2179 // Restoration of lr done by remove_activation.
2180 switch (state) {
2181 // Narrow result if state is itos but result type is smaller.
2182 case btos:
2183 case ztos:
2184 case ctos:
2185 case stos:
2186 case itos: __ narrow(R17_tos); /* fall through */
2187 case ltos:
2188 case atos: __ mr(R3_RET, R17_tos); break;
2189 case ftos:
2190 case dtos: __ fmr(F1_RET, F15_ftos); break;
2191 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2192 // to get visible before the reference to the object gets stored anywhere.
2193 __ membar(Assembler::StoreStore); break;
2194 default : ShouldNotReachHere();
2195 }
2196 __ blr();
2197
2198 return entry;
2199 } // end of ForceEarlyReturn support
2200
2201 //-----------------------------------------------------------------------------
2202 // Helper for vtos entry point generation
2203
set_vtos_entry_points(Template * t,address & bep,address & cep,address & sep,address & aep,address & iep,address & lep,address & fep,address & dep,address & vep)2204 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2205 address& bep,
2206 address& cep,
2207 address& sep,
2208 address& aep,
2209 address& iep,
2210 address& lep,
2211 address& fep,
2212 address& dep,
2213 address& vep) {
2214 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2215 Label L;
2216
2217 aep = __ pc(); __ push_ptr(); __ b(L);
2218 fep = __ pc(); __ push_f(); __ b(L);
2219 dep = __ pc(); __ push_d(); __ b(L);
2220 lep = __ pc(); __ push_l(); __ b(L);
2221 __ align(32, 12, 24); // align L
2222 bep = cep = sep =
2223 iep = __ pc(); __ push_i();
2224 vep = __ pc();
2225 __ bind(L);
2226 generate_and_dispatch(t);
2227 }
2228
2229 //-----------------------------------------------------------------------------
2230
2231 // Non-product code
2232 #ifndef PRODUCT
generate_trace_code(TosState state)2233 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2234 //__ flush_bundle();
2235 address entry = __ pc();
2236
2237 const char *bname = NULL;
2238 uint tsize = 0;
2239 switch(state) {
2240 case ftos:
2241 bname = "trace_code_ftos {";
2242 tsize = 2;
2243 break;
2244 case btos:
2245 bname = "trace_code_btos {";
2246 tsize = 2;
2247 break;
2248 case ztos:
2249 bname = "trace_code_ztos {";
2250 tsize = 2;
2251 break;
2252 case ctos:
2253 bname = "trace_code_ctos {";
2254 tsize = 2;
2255 break;
2256 case stos:
2257 bname = "trace_code_stos {";
2258 tsize = 2;
2259 break;
2260 case itos:
2261 bname = "trace_code_itos {";
2262 tsize = 2;
2263 break;
2264 case ltos:
2265 bname = "trace_code_ltos {";
2266 tsize = 3;
2267 break;
2268 case atos:
2269 bname = "trace_code_atos {";
2270 tsize = 2;
2271 break;
2272 case vtos:
2273 // Note: In case of vtos, the topmost of stack value could be a int or doubl
2274 // In case of a double (2 slots) we won't see the 2nd stack value.
2275 // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
2276 bname = "trace_code_vtos {";
2277 tsize = 2;
2278
2279 break;
2280 case dtos:
2281 bname = "trace_code_dtos {";
2282 tsize = 3;
2283 break;
2284 default:
2285 ShouldNotReachHere();
2286 }
2287 BLOCK_COMMENT(bname);
2288
2289 // Support short-cut for TraceBytecodesAt.
2290 // Don't call into the VM if we don't want to trace to speed up things.
2291 Label Lskip_vm_call;
2292 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2293 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
2294 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2295 __ ld(R11_scratch1, offs1, R11_scratch1);
2296 __ lwa(R12_scratch2, offs2, R12_scratch2);
2297 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2298 __ blt(CCR0, Lskip_vm_call);
2299 }
2300
2301 __ push(state);
2302 // Load 2 topmost expression stack values.
2303 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
2304 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
2305 __ mflr(R31);
2306 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
2307 __ mtlr(R31);
2308 __ pop(state);
2309
2310 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2311 __ bind(Lskip_vm_call);
2312 }
2313 __ blr();
2314 BLOCK_COMMENT("} trace_code");
2315 return entry;
2316 }
2317
count_bytecode()2318 void TemplateInterpreterGenerator::count_bytecode() {
2319 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
2320 __ lwz(R12_scratch2, offs, R11_scratch1);
2321 __ addi(R12_scratch2, R12_scratch2, 1);
2322 __ stw(R12_scratch2, offs, R11_scratch1);
2323 }
2324
histogram_bytecode(Template * t)2325 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2326 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
2327 __ lwz(R12_scratch2, offs, R11_scratch1);
2328 __ addi(R12_scratch2, R12_scratch2, 1);
2329 __ stw(R12_scratch2, offs, R11_scratch1);
2330 }
2331
histogram_bytecode_pair(Template * t)2332 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2333 const Register addr = R11_scratch1,
2334 tmp = R12_scratch2;
2335 // Get index, shift out old bytecode, bring in new bytecode, and store it.
2336 // _index = (_index >> log2_number_of_codes) |
2337 // (bytecode << log2_number_of_codes);
2338 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
2339 __ lwz(tmp, offs1, addr);
2340 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
2341 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2342 __ stw(tmp, offs1, addr);
2343
2344 // Bump bucket contents.
2345 // _counters[_index] ++;
2346 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
2347 __ sldi(tmp, tmp, LogBytesPerInt);
2348 __ add(addr, tmp, addr);
2349 __ lwz(tmp, offs2, addr);
2350 __ addi(tmp, tmp, 1);
2351 __ stw(tmp, offs2, addr);
2352 }
2353
trace_bytecode(Template * t)2354 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2355 // Call a little run-time stub to avoid blow-up for each bytecode.
2356 // The run-time runtime saves the right registers, depending on
2357 // the tosca in-state for the given template.
2358
2359 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2360 "entry must have been generated");
2361
2362 // Note: we destroy LR here.
2363 __ bl(Interpreter::trace_code(t->tos_in()));
2364 }
2365
stop_interpreter_at()2366 void TemplateInterpreterGenerator::stop_interpreter_at() {
2367 Label L;
2368 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
2369 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2370 __ ld(R11_scratch1, offs1, R11_scratch1);
2371 __ lwa(R12_scratch2, offs2, R12_scratch2);
2372 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2373 __ bne(CCR0, L);
2374 __ illtrap();
2375 __ bind(L);
2376 }
2377
2378 #endif // !PRODUCT
2379