1 /*
2 * Copyright (c) 2014, 2019, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2019, SAP SE. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "gc/shared/barrierSetAssembler.hpp"
29 #include "interpreter/bytecodeHistogram.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "interpreter/interpreterRuntime.hpp"
32 #include "interpreter/interp_masm.hpp"
33 #include "interpreter/templateInterpreterGenerator.hpp"
34 #include "interpreter/templateTable.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/methodData.hpp"
37 #include "oops/method.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "prims/jvmtiExport.hpp"
40 #include "prims/jvmtiThreadState.hpp"
41 #include "runtime/arguments.hpp"
42 #include "runtime/deoptimization.hpp"
43 #include "runtime/frame.inline.hpp"
44 #include "runtime/sharedRuntime.hpp"
45 #include "runtime/stubRoutines.hpp"
46 #include "runtime/synchronizer.hpp"
47 #include "runtime/timer.hpp"
48 #include "runtime/vframeArray.hpp"
49 #include "utilities/debug.hpp"
50 #include "utilities/macros.hpp"
51
52 #undef __
53 #define __ _masm->
54
55 // Size of interpreter code. Increase if too small. Interpreter will
56 // fail with a guarantee ("not enough space for interpreter generation");
57 // if too small.
58 // Run with +PrintInterpreter to get the VM to print out the size.
59 // Max size with JVMTI
60 int TemplateInterpreter::InterpreterCodeSize = 256*K;
61
62 #ifdef PRODUCT
63 #define BLOCK_COMMENT(str) /* nothing */
64 #else
65 #define BLOCK_COMMENT(str) __ block_comment(str)
66 #endif
67
68 #define BIND(label) __ bind(label); BLOCK_COMMENT(#label ":")
69
70 //-----------------------------------------------------------------------------
71
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", 120);
220
221 __ bind(do_array);
222
223 {
224 Label start_skip, end_skip;
225
226 __ bind(start_skip);
227 __ lbzu(sig_byte, 1, signature);
228 __ cmplwi(CCR0, sig_byte, '[');
229 __ beq(CCR0, start_skip); // skip further brackets
230 __ cmplwi(CCR0, sig_byte, '9');
231 __ bgt(CCR0, end_skip); // no optional size
232 __ cmplwi(CCR0, sig_byte, '0');
233 __ bge(CCR0, start_skip); // skip optional size
234 __ bind(end_skip);
235
236 __ cmplwi(CCR0, sig_byte, 'L');
237 __ beq(CCR0, do_object); // for arrays of objects, the name of the object must be skipped
238 __ b(do_boxed); // otherwise, go directly to do_boxed
239 }
240
241 __ bind(do_object);
242 {
243 Label L;
244 __ bind(L);
245 __ lbzu(sig_byte, 1, signature);
246 __ cmplwi(CCR0, sig_byte, ';');
247 __ bne(CCR0, L);
248 }
249 // Need to box the Java object here, so we use arg_java (address of
250 // current Java stack slot) as argument and don't dereference it as
251 // in case of ints, floats, etc.
252 Label do_null;
253 __ bind(do_boxed);
254 __ ld(R0,0, arg_java);
255 __ cmpdi(CCR0, R0, 0);
256 __ li(intSlot,0);
257 __ beq(CCR0, do_null);
258 __ mr(intSlot, arg_java);
259 __ bind(do_null);
260 __ std(intSlot, 0, arg_c);
261 __ addi(arg_java, arg_java, -BytesPerWord);
262 __ addi(arg_c, arg_c, BytesPerWord);
263 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
264 __ blt(CCR0, move_intSlot_to_ARG);
265 __ b(loop_start);
266
267 __ bind(do_int);
268 __ lwa(intSlot, 0, arg_java);
269 __ std(intSlot, 0, arg_c);
270 __ addi(arg_java, arg_java, -BytesPerWord);
271 __ addi(arg_c, arg_c, BytesPerWord);
272 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
273 __ blt(CCR0, move_intSlot_to_ARG);
274 __ b(loop_start);
275
276 __ bind(do_long);
277 __ ld(intSlot, -BytesPerWord, arg_java);
278 __ std(intSlot, 0, arg_c);
279 __ addi(arg_java, arg_java, - 2 * BytesPerWord);
280 __ addi(arg_c, arg_c, BytesPerWord);
281 __ cmplwi(CCR0, argcnt, max_int_register_arguments);
282 __ blt(CCR0, move_intSlot_to_ARG);
283 __ b(loop_start);
284
285 __ bind(do_float);
286 __ lfs(floatSlot, 0, arg_java);
287 #if defined(LINUX)
288 // Linux uses ELF ABI. Both original ELF and ELFv2 ABIs have float
289 // in the least significant word of an argument slot.
290 #if defined(VM_LITTLE_ENDIAN)
291 __ stfs(floatSlot, 0, arg_c);
292 #else
293 __ stfs(floatSlot, 4, arg_c);
294 #endif
295 #elif defined(AIX)
296 // Although AIX runs on big endian CPU, float is in most significant
297 // word of an argument slot.
298 __ stfs(floatSlot, 0, arg_c);
299 #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 guarantee(referent_offset > 0, "referent offset not initialized");
514
515 Label slow_path;
516
517 // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH);
518
519 // In the G1 code we don't check if we need to reach a safepoint. We
520 // continue and the thread will safepoint at the next bytecode dispatch.
521
522 // If the receiver is null then it is OK to jump to the slow path.
523 __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver
524
525 // Check if receiver == NULL and go the slow path.
526 __ cmpdi(CCR0, R3_RET, 0);
527 __ beq(CCR0, slow_path);
528
529 __ load_heap_oop(R3_RET, referent_offset, R3_RET,
530 /* non-volatile temp */ R31, R11_scratch1, true, ON_WEAK_OOP_REF);
531
532 // Generate the G1 pre-barrier code to log the value of
533 // the referent field in an SATB buffer. Note with
534 // these parameters the pre-barrier does not generate
535 // the load of the previous value.
536
537 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
538 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
539
540 __ blr();
541
542 __ bind(slow_path);
543 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
544 return entry;
545 }
546
generate_StackOverflowError_handler()547 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
548 address entry = __ pc();
549
550 // Expression stack must be empty before entering the VM if an
551 // exception happened.
552 __ empty_expression_stack();
553 // Throw exception.
554 __ call_VM(noreg,
555 CAST_FROM_FN_PTR(address,
556 InterpreterRuntime::throw_StackOverflowError));
557 return entry;
558 }
559
generate_ArrayIndexOutOfBounds_handler()560 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
561 address entry = __ pc();
562 __ empty_expression_stack();
563 // R4_ARG2 already contains the array.
564 // Index is in R17_tos.
565 __ mr(R5_ARG3, R17_tos);
566 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R4_ARG2, R5_ARG3);
567 return entry;
568 }
569
generate_ClassCastException_handler()570 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
571 address entry = __ pc();
572 // Expression stack must be empty before entering the VM if an
573 // exception happened.
574 __ empty_expression_stack();
575
576 // Load exception object.
577 // Thread will be loaded to R3_ARG1.
578 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
579 #ifdef ASSERT
580 // Above call must not return here since exception pending.
581 __ should_not_reach_here();
582 #endif
583 return entry;
584 }
585
generate_exception_handler_common(const char * name,const char * message,bool pass_oop)586 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
587 address entry = __ pc();
588 //__ untested("generate_exception_handler_common");
589 Register Rexception = R17_tos;
590
591 // Expression stack must be empty before entering the VM if an exception happened.
592 __ empty_expression_stack();
593
594 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
595 if (pass_oop) {
596 __ mr(R5_ARG3, Rexception);
597 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception));
598 } else {
599 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
600 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception));
601 }
602
603 // Throw exception.
604 __ mr(R3_ARG1, Rexception);
605 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
606 __ mtctr(R11_scratch1);
607 __ bctr();
608
609 return entry;
610 }
611
612 // This entry is returned to when a call returns to the interpreter.
613 // 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)614 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
615 address entry = __ pc();
616
617 // Move the value out of the return register back to the TOS cache of current frame.
618 switch (state) {
619 case ltos:
620 case btos:
621 case ztos:
622 case ctos:
623 case stos:
624 case atos:
625 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache
626 case ftos:
627 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
628 case vtos: break; // Nothing to do, this was a void return.
629 default : ShouldNotReachHere();
630 }
631
632 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
633 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
634 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
635
636 // Compiled code destroys templateTableBase, reload.
637 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
638
639 if (state == atos) {
640 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
641 }
642
643 const Register cache = R11_scratch1;
644 const Register size = R12_scratch2;
645 __ get_cache_and_index_at_bcp(cache, 1, index_size);
646
647 // Get least significant byte of 64 bit value:
648 #if defined(VM_LITTLE_ENDIAN)
649 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache);
650 #else
651 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
652 #endif
653 __ sldi(size, size, Interpreter::logStackElementSize);
654 __ add(R15_esp, R15_esp, size);
655
656 __ check_and_handle_popframe(R11_scratch1);
657 __ check_and_handle_earlyret(R11_scratch1);
658
659 __ dispatch_next(state, step);
660 return entry;
661 }
662
generate_deopt_entry_for(TosState state,int step,address continuation)663 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
664 address entry = __ pc();
665 // If state != vtos, we're returning from a native method, which put it's result
666 // into the result register. So move the value out of the return register back
667 // to the TOS cache of current frame.
668
669 switch (state) {
670 case ltos:
671 case btos:
672 case ztos:
673 case ctos:
674 case stos:
675 case atos:
676 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache
677 case ftos:
678 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
679 case vtos: break; // Nothing to do, this was a void return.
680 default : ShouldNotReachHere();
681 }
682
683 // Load LcpoolCache @@@ should be already set!
684 __ get_constant_pool_cache(R27_constPoolCache);
685
686 // Handle a pending exception, fall through if none.
687 __ check_and_forward_exception(R11_scratch1, R12_scratch2);
688
689 // Start executing bytecodes.
690 if (continuation == NULL) {
691 __ dispatch_next(state, step);
692 } else {
693 __ jump_to_entry(continuation, R11_scratch1);
694 }
695
696 return entry;
697 }
698
generate_safept_entry_for(TosState state,address runtime_entry)699 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
700 address entry = __ pc();
701
702 __ push(state);
703 __ call_VM(noreg, runtime_entry);
704 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
705
706 return entry;
707 }
708
709 // Helpers for commoning out cases in the various type of method entries.
710
711 // Increment invocation count & check for overflow.
712 //
713 // Note: checking for negative value instead of overflow
714 // so we have a 'sticky' overflow test.
715 //
generate_counter_incr(Label * overflow,Label * profile_method,Label * profile_method_continue)716 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
717 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
718 Register Rscratch1 = R11_scratch1;
719 Register Rscratch2 = R12_scratch2;
720 Register R3_counters = R3_ARG1;
721 Label done;
722
723 if (TieredCompilation) {
724 const int increment = InvocationCounter::count_increment;
725 Label no_mdo;
726 if (ProfileInterpreter) {
727 const Register Rmdo = R3_counters;
728 // If no method data exists, go to profile_continue.
729 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
730 __ cmpdi(CCR0, Rmdo, 0);
731 __ beq(CCR0, no_mdo);
732
733 // Increment invocation counter in the MDO.
734 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
735 __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
736 __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo);
737 __ addi(Rscratch2, Rscratch2, increment);
738 __ stw(Rscratch2, mdo_ic_offs, Rmdo);
739 __ and_(Rscratch1, Rscratch2, Rscratch1);
740 __ bne(CCR0, done);
741 __ b(*overflow);
742 }
743
744 // Increment counter in MethodCounters*.
745 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
746 __ bind(no_mdo);
747 __ get_method_counters(R19_method, R3_counters, done);
748 __ lwz(Rscratch2, mo_ic_offs, R3_counters);
749 __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters);
750 __ addi(Rscratch2, Rscratch2, increment);
751 __ stw(Rscratch2, mo_ic_offs, R3_counters);
752 __ and_(Rscratch1, Rscratch2, Rscratch1);
753 __ beq(CCR0, *overflow);
754
755 __ bind(done);
756
757 } else {
758
759 // Update standard invocation counters.
760 Register Rsum_ivc_bec = R4_ARG2;
761 __ get_method_counters(R19_method, R3_counters, done);
762 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2);
763 // Increment interpreter invocation counter.
764 if (ProfileInterpreter) { // %%% Merge this into methodDataOop.
765 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
766 __ addi(R12_scratch2, R12_scratch2, 1);
767 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
768 }
769 // Check if we must create a method data obj.
770 if (ProfileInterpreter && profile_method != NULL) {
771 const Register profile_limit = Rscratch1;
772 __ lwz(profile_limit, in_bytes(MethodCounters::interpreter_profile_limit_offset()), R3_counters);
773 // Test to see if we should create a method data oop.
774 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit);
775 __ blt(CCR0, *profile_method_continue);
776 // If no method data exists, go to profile_method.
777 __ test_method_data_pointer(*profile_method);
778 }
779 // Finally check for counter overflow.
780 if (overflow) {
781 const Register invocation_limit = Rscratch1;
782 __ lwz(invocation_limit, in_bytes(MethodCounters::interpreter_invocation_limit_offset()), R3_counters);
783 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit);
784 __ bge(CCR0, *overflow);
785 }
786
787 __ bind(done);
788 }
789 }
790
791 // Generate code to initiate compilation on invocation counter overflow.
generate_counter_overflow(Label & continue_entry)792 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
793 // Generate code to initiate compilation on the counter overflow.
794
795 // InterpreterRuntime::frequency_counter_overflow takes one arguments,
796 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp)
797 // We pass zero in.
798 // The call returns the address of the verified entry point for the method or NULL
799 // if the compilation did not complete (either went background or bailed out).
800 //
801 // Unlike the C++ interpreter above: Check exceptions!
802 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
803 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
804
805 __ li(R4_ARG2, 0);
806 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
807
808 // Returns verified_entry_point or NULL.
809 // We ignore it in any case.
810 __ b(continue_entry);
811 }
812
813 // See if we've got enough room on the stack for locals plus overhead below
814 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
815 // without going through the signal handler, i.e., reserved and yellow zones
816 // will not be made usable. The shadow zone must suffice to handle the
817 // overflow.
818 //
819 // Kills Rmem_frame_size, Rscratch1.
generate_stack_overflow_check(Register Rmem_frame_size,Register Rscratch1)820 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
821 Label done;
822 assert_different_registers(Rmem_frame_size, Rscratch1);
823
824 BLOCK_COMMENT("stack_overflow_check_with_compare {");
825 __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
826 __ ld(Rscratch1, thread_(stack_overflow_limit));
827 __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
828 __ bgt(CCR0/*is_stack_overflow*/, done);
829
830 // The stack overflows. Load target address of the runtime stub and call it.
831 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order");
832 __ load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0);
833 __ mtctr(Rscratch1);
834 // Restore caller_sp (c2i adapter may exist, but no shrinking of interpreted caller frame).
835 #ifdef ASSERT
836 Label frame_not_shrunk;
837 __ cmpld(CCR0, R1_SP, R21_sender_SP);
838 __ ble(CCR0, frame_not_shrunk);
839 __ stop("frame shrunk", 0x546);
840 __ bind(frame_not_shrunk);
841 __ ld(Rscratch1, 0, R1_SP);
842 __ ld(R0, 0, R21_sender_SP);
843 __ cmpd(CCR0, R0, Rscratch1);
844 __ asm_assert_eq("backlink", 0x547);
845 #endif // ASSERT
846 __ mr(R1_SP, R21_sender_SP);
847 __ bctr();
848
849 __ align(32, 12);
850 __ bind(done);
851 BLOCK_COMMENT("} stack_overflow_check_with_compare");
852 }
853
854 // Lock the current method, interpreter register window must be set up!
lock_method(Register Rflags,Register Rscratch1,Register Rscratch2,bool flags_preloaded)855 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
856 const Register Robj_to_lock = Rscratch2;
857
858 {
859 if (!flags_preloaded) {
860 __ lwz(Rflags, method_(access_flags));
861 }
862
863 #ifdef ASSERT
864 // Check if methods needs synchronization.
865 {
866 Label Lok;
867 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
868 __ btrue(CCR0,Lok);
869 __ stop("method doesn't need synchronization");
870 __ bind(Lok);
871 }
872 #endif // ASSERT
873 }
874
875 // Get synchronization object to Rscratch2.
876 {
877 Label Lstatic;
878 Label Ldone;
879
880 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
881 __ btrue(CCR0, Lstatic);
882
883 // Non-static case: load receiver obj from stack and we're done.
884 __ ld(Robj_to_lock, R18_locals);
885 __ b(Ldone);
886
887 __ bind(Lstatic); // Static case: Lock the java mirror
888 // Load mirror from interpreter frame.
889 __ ld(Robj_to_lock, _abi(callers_sp), R1_SP);
890 __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock);
891
892 __ bind(Ldone);
893 __ verify_oop(Robj_to_lock);
894 }
895
896 // Got the oop to lock => execute!
897 __ add_monitor_to_stack(true, Rscratch1, R0);
898
899 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
900 __ lock_object(R26_monitor, Robj_to_lock);
901 }
902
903 // Generate a fixed interpreter frame for pure interpreter
904 // and I2N native transition frames.
905 //
906 // Before (stack grows downwards):
907 //
908 // | ... |
909 // |------------- |
910 // | java arg0 |
911 // | ... |
912 // | java argn |
913 // | | <- R15_esp
914 // | |
915 // |--------------|
916 // | abi_112 |
917 // | | <- R1_SP
918 // |==============|
919 //
920 //
921 // After:
922 //
923 // | ... |
924 // | java arg0 |<- R18_locals
925 // | ... |
926 // | java argn |
927 // |--------------|
928 // | |
929 // | java locals |
930 // | |
931 // |--------------|
932 // | abi_48 |
933 // |==============|
934 // | |
935 // | istate |
936 // | |
937 // |--------------|
938 // | monitor |<- R26_monitor
939 // |--------------|
940 // | |<- R15_esp
941 // | expression |
942 // | stack |
943 // | |
944 // |--------------|
945 // | |
946 // | abi_112 |<- R1_SP
947 // |==============|
948 //
949 // The top most frame needs an abi space of 112 bytes. This space is needed,
950 // since we call to c. The c function may spill their arguments to the caller
951 // frame. When we call to java, we don't need these spill slots. In order to save
952 // space on the stack, we resize the caller. However, java locals reside in
953 // the caller frame and the frame has to be increased. The frame_size for the
954 // current frame was calculated based on max_stack as size for the expression
955 // stack. At the call, just a part of the expression stack might be used.
956 // We don't want to waste this space and cut the frame back accordingly.
957 // The resulting amount for resizing is calculated as follows:
958 // resize = (number_of_locals - number_of_arguments) * slot_size
959 // + (R1_SP - R15_esp) + 48
960 //
961 // The size for the callee frame is calculated:
962 // framesize = 112 + max_stack + monitor + state_size
963 //
964 // maxstack: Max number of slots on the expression stack, loaded from the method.
965 // monitor: We statically reserve room for one monitor object.
966 // state_size: We save the current state of the interpreter to this area.
967 //
generate_fixed_frame(bool native_call,Register Rsize_of_parameters,Register Rsize_of_locals)968 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
969 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
970 top_frame_size = R7_ARG5,
971 Rconst_method = R8_ARG6;
972
973 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size);
974
975 __ ld(Rconst_method, method_(const));
976 __ lhz(Rsize_of_parameters /* number of params */,
977 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
978 if (native_call) {
979 // If we're calling a native method, we reserve space for the worst-case signature
980 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2).
981 // We add two slots to the parameter_count, one for the jni
982 // environment and one for a possible native mirror.
983 Label skip_native_calculate_max_stack;
984 __ addi(top_frame_size, Rsize_of_parameters, 2);
985 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters);
986 __ bge(CCR0, skip_native_calculate_max_stack);
987 __ li(top_frame_size, Argument::n_register_parameters);
988 __ bind(skip_native_calculate_max_stack);
989 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
990 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
991 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
992 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
993 } else {
994 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
995 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
996 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
997 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
998 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
999 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
1000 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
1001 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1);
1002 }
1003
1004 // Compute top frame size.
1005 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size);
1006
1007 // Cut back area between esp and max_stack.
1008 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize);
1009
1010 __ round_to(top_frame_size, frame::alignment_in_bytes);
1011 __ round_to(parent_frame_resize, frame::alignment_in_bytes);
1012 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
1013 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
1014
1015 if (!native_call) {
1016 // Stack overflow check.
1017 // Native calls don't need the stack size check since they have no
1018 // expression stack and the arguments are already on the stack and
1019 // we only add a handful of words to the stack.
1020 __ add(R11_scratch1, parent_frame_resize, top_frame_size);
1021 generate_stack_overflow_check(R11_scratch1, R12_scratch2);
1022 }
1023
1024 // Set up interpreter state registers.
1025
1026 __ add(R18_locals, R15_esp, Rsize_of_parameters);
1027 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
1028 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache);
1029
1030 // Set method data pointer.
1031 if (ProfileInterpreter) {
1032 Label zero_continue;
1033 __ ld(R28_mdx, method_(method_data));
1034 __ cmpdi(CCR0, R28_mdx, 0);
1035 __ beq(CCR0, zero_continue);
1036 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1037 __ bind(zero_continue);
1038 }
1039
1040 if (native_call) {
1041 __ li(R14_bcp, 0); // Must initialize.
1042 } else {
1043 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method);
1044 }
1045
1046 // Resize parent frame.
1047 __ mflr(R12_scratch2);
1048 __ neg(parent_frame_resize, parent_frame_resize);
1049 __ resize_frame(parent_frame_resize, R11_scratch1);
1050 __ std(R12_scratch2, _abi(lr), R1_SP);
1051
1052 // Get mirror and store it in the frame as GC root for this Method*.
1053 __ load_mirror_from_const_method(R12_scratch2, Rconst_method);
1054
1055 __ addi(R26_monitor, R1_SP, -frame::ijava_state_size);
1056 __ addi(R15_esp, R26_monitor, -Interpreter::stackElementSize);
1057
1058 // Store values.
1059 __ std(R19_method, _ijava_state_neg(method), R1_SP);
1060 __ std(R12_scratch2, _ijava_state_neg(mirror), R1_SP);
1061 __ std(R18_locals, _ijava_state_neg(locals), R1_SP);
1062 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
1063
1064 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
1065 // be found in the frame after save_interpreter_state is done. This is always true
1066 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
1067 // because e.g. frame::interpreter_frame_bcp() will not access the correct value
1068 // (Enhanced Stack Trace).
1069 // The signal handler does not save the interpreter state into the frame.
1070
1071 // We have to initialize some of these frame slots for native calls (accessed by GC).
1072 // Also initialize them for non-native calls for better tool support (even though
1073 // you may not get the most recent version as described above).
1074 __ li(R0, 0);
1075 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP);
1076 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
1077 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
1078 __ std(R15_esp, _ijava_state_neg(esp), R1_SP);
1079 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); // only used for native_call
1080
1081 // Store sender's SP and this frame's top SP.
1082 __ subf(R12_scratch2, top_frame_size, R1_SP);
1083 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
1084 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
1085
1086 // Push top frame.
1087 __ push_frame(top_frame_size, R11_scratch1);
1088 }
1089
1090 // End of helpers
1091
generate_math_entry(AbstractInterpreter::MethodKind kind)1092 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1093
1094 // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1095 bool use_instruction = false;
1096 address runtime_entry = NULL;
1097 int num_args = 1;
1098 bool double_precision = true;
1099
1100 // PPC64 specific:
1101 switch (kind) {
1102 case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break;
1103 case Interpreter::java_lang_math_abs: use_instruction = true; break;
1104 case Interpreter::java_lang_math_fmaF:
1105 case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1106 default: break; // Fall back to runtime call.
1107 }
1108
1109 switch (kind) {
1110 case Interpreter::java_lang_math_sin : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); break;
1111 case Interpreter::java_lang_math_cos : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); break;
1112 case Interpreter::java_lang_math_tan : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); break;
1113 case Interpreter::java_lang_math_abs : /* run interpreted */ break;
1114 case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); break;
1115 case Interpreter::java_lang_math_log : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); break;
1116 case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1117 case Interpreter::java_lang_math_pow : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1118 case Interpreter::java_lang_math_exp : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); break;
1119 case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1120 case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1121 default: ShouldNotReachHere();
1122 }
1123
1124 // Use normal entry if neither instruction nor runtime call is used.
1125 if (!use_instruction && runtime_entry == NULL) return NULL;
1126
1127 address entry = __ pc();
1128
1129 // Load arguments
1130 assert(num_args <= 13, "passed in registers");
1131 if (double_precision) {
1132 int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1133 for (int i = 0; i < num_args; ++i) {
1134 __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1135 offset -= 2 * Interpreter::stackElementSize;
1136 }
1137 } else {
1138 int offset = num_args * Interpreter::stackElementSize;
1139 for (int i = 0; i < num_args; ++i) {
1140 __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1141 offset -= Interpreter::stackElementSize;
1142 }
1143 }
1144
1145 if (use_instruction) {
1146 switch (kind) {
1147 case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1); break;
1148 case Interpreter::java_lang_math_abs: __ fabs(F1_RET, F1); break;
1149 case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break;
1150 case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3); break;
1151 default: ShouldNotReachHere();
1152 }
1153 } else {
1154 // Comment: Can use tail call if the unextended frame is always C ABI compliant:
1155 //__ load_const_optimized(R12_scratch2, runtime_entry, R0);
1156 //__ call_c_and_return_to_caller(R12_scratch2);
1157
1158 // Push a new C frame and save LR.
1159 __ save_LR_CR(R0);
1160 __ push_frame_reg_args(0, R11_scratch1);
1161
1162 __ call_VM_leaf(runtime_entry);
1163
1164 // Pop the C frame and restore LR.
1165 __ pop_frame();
1166 __ restore_LR_CR(R0);
1167 }
1168
1169 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1170 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1171 __ blr();
1172
1173 __ flush();
1174
1175 return entry;
1176 }
1177
bang_stack_shadow_pages(bool native_call)1178 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1179 // Quick & dirty stack overflow checking: bang the stack & handle trap.
1180 // Note that we do the banging after the frame is setup, since the exception
1181 // handling code expects to find a valid interpreter frame on the stack.
1182 // Doing the banging earlier fails if the caller frame is not an interpreter
1183 // frame.
1184 // (Also, the exception throwing code expects to unlock any synchronized
1185 // method receiever, so do the banging after locking the receiver.)
1186
1187 // Bang each page in the shadow zone. We can't assume it's been done for
1188 // an interpreter frame with greater than a page of locals, so each page
1189 // needs to be checked. Only true for non-native.
1190 if (UseStackBanging) {
1191 const int page_size = os::vm_page_size();
1192 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
1193 const int start_page = native_call ? n_shadow_pages : 1;
1194 BLOCK_COMMENT("bang_stack_shadow_pages:");
1195 for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1196 __ bang_stack_with_offset(pages*page_size);
1197 }
1198 }
1199 }
1200
1201 // Interpreter stub for calling a native method. (asm interpreter)
1202 // This sets up a somewhat different looking stack for calling the
1203 // native method than the typical interpreter frame setup.
1204 //
1205 // On entry:
1206 // R19_method - method
1207 // R16_thread - JavaThread*
1208 // R15_esp - intptr_t* sender tos
1209 //
1210 // abstract stack (grows up)
1211 // [ IJava (caller of JNI callee) ] <-- ASP
1212 // ...
generate_native_entry(bool synchronized)1213 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1214
1215 address entry = __ pc();
1216
1217 const bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1218
1219 // -----------------------------------------------------------------------------
1220 // Allocate a new frame that represents the native callee (i2n frame).
1221 // This is not a full-blown interpreter frame, but in particular, the
1222 // following registers are valid after this:
1223 // - R19_method
1224 // - R18_local (points to start of arguments to native function)
1225 //
1226 // abstract stack (grows up)
1227 // [ IJava (caller of JNI callee) ] <-- ASP
1228 // ...
1229
1230 const Register signature_handler_fd = R11_scratch1;
1231 const Register pending_exception = R0;
1232 const Register result_handler_addr = R31;
1233 const Register native_method_fd = R11_scratch1;
1234 const Register access_flags = R22_tmp2;
1235 const Register active_handles = R11_scratch1; // R26_monitor saved to state.
1236 const Register sync_state = R12_scratch2;
1237 const Register sync_state_addr = sync_state; // Address is dead after use.
1238 const Register suspend_flags = R11_scratch1;
1239
1240 //=============================================================================
1241 // Allocate new frame and initialize interpreter state.
1242
1243 Label exception_return;
1244 Label exception_return_sync_check;
1245 Label stack_overflow_return;
1246
1247 // Generate new interpreter state and jump to stack_overflow_return in case of
1248 // a stack overflow.
1249 //generate_compute_interpreter_state(stack_overflow_return);
1250
1251 Register size_of_parameters = R22_tmp2;
1252
1253 generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
1254
1255 //=============================================================================
1256 // Increment invocation counter. On overflow, entry to JNI method
1257 // will be compiled.
1258 Label invocation_counter_overflow, continue_after_compile;
1259 if (inc_counter) {
1260 if (synchronized) {
1261 // Since at this point in the method invocation the exception handler
1262 // would try to exit the monitor of synchronized methods which hasn't
1263 // been entered yet, we set the thread local variable
1264 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1265 // runtime, exception handling i.e. unlock_if_synchronized_method will
1266 // check this thread local flag.
1267 // This flag has two effects, one is to force an unwind in the topmost
1268 // interpreter frame and not perform an unlock while doing so.
1269 __ li(R0, 1);
1270 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1271 }
1272 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1273
1274 BIND(continue_after_compile);
1275 }
1276
1277 bang_stack_shadow_pages(true);
1278
1279 if (inc_counter) {
1280 // Reset the _do_not_unlock_if_synchronized flag.
1281 if (synchronized) {
1282 __ li(R0, 0);
1283 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1284 }
1285 }
1286
1287 // access_flags = method->access_flags();
1288 // Load access flags.
1289 assert(access_flags->is_nonvolatile(),
1290 "access_flags must be in a non-volatile register");
1291 // Type check.
1292 assert(4 == sizeof(AccessFlags), "unexpected field size");
1293 __ lwz(access_flags, method_(access_flags));
1294
1295 // We don't want to reload R19_method and access_flags after calls
1296 // to some helper functions.
1297 assert(R19_method->is_nonvolatile(),
1298 "R19_method must be a non-volatile register");
1299
1300 // Check for synchronized methods. Must happen AFTER invocation counter
1301 // check, so method is not locked if counter overflows.
1302
1303 if (synchronized) {
1304 lock_method(access_flags, R11_scratch1, R12_scratch2, true);
1305
1306 // Update monitor in state.
1307 __ ld(R11_scratch1, 0, R1_SP);
1308 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1);
1309 }
1310
1311 // jvmti/jvmpi support
1312 __ notify_method_entry();
1313
1314 //=============================================================================
1315 // Get and call the signature handler.
1316
1317 __ ld(signature_handler_fd, method_(signature_handler));
1318 Label call_signature_handler;
1319
1320 __ cmpdi(CCR0, signature_handler_fd, 0);
1321 __ bne(CCR0, call_signature_handler);
1322
1323 // Method has never been called. Either generate a specialized
1324 // handler or point to the slow one.
1325 //
1326 // Pass parameter 'false' to avoid exception check in call_VM.
1327 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
1328
1329 // Check for an exception while looking up the target method. If we
1330 // incurred one, bail.
1331 __ ld(pending_exception, thread_(pending_exception));
1332 __ cmpdi(CCR0, pending_exception, 0);
1333 __ bne(CCR0, exception_return_sync_check); // Has pending exception.
1334
1335 // Reload signature handler, it may have been created/assigned in the meanwhile.
1336 __ ld(signature_handler_fd, method_(signature_handler));
1337 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
1338
1339 BIND(call_signature_handler);
1340
1341 // Before we call the signature handler we push a new frame to
1342 // protect the interpreter frame volatile registers when we return
1343 // from jni but before we can get back to Java.
1344
1345 // First set the frame anchor while the SP/FP registers are
1346 // convenient and the slow signature handler can use this same frame
1347 // anchor.
1348
1349 // We have a TOP_IJAVA_FRAME here, which belongs to us.
1350 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
1351
1352 // Now the interpreter frame (and its call chain) have been
1353 // invalidated and flushed. We are now protected against eager
1354 // being enabled in native code. Even if it goes eager the
1355 // registers will be reloaded as clean and we will invalidate after
1356 // the call so no spurious flush should be possible.
1357
1358 // Call signature handler and pass locals address.
1359 //
1360 // Our signature handlers copy required arguments to the C stack
1361 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
1362 __ mr(R3_ARG1, R18_locals);
1363 #if !defined(ABI_ELFv2)
1364 __ ld(signature_handler_fd, 0, signature_handler_fd);
1365 #endif
1366
1367 __ call_stub(signature_handler_fd);
1368
1369 // Remove the register parameter varargs slots we allocated in
1370 // compute_interpreter_state. SP+16 ends up pointing to the ABI
1371 // outgoing argument area.
1372 //
1373 // Not needed on PPC64.
1374 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord);
1375
1376 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
1377 // Save across call to native method.
1378 __ mr(result_handler_addr, R3_RET);
1379
1380 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
1381
1382 // Set up fixed parameters and call the native method.
1383 // If the method is static, get mirror into R4_ARG2.
1384 {
1385 Label method_is_not_static;
1386 // Access_flags is non-volatile and still, no need to restore it.
1387
1388 // Restore access flags.
1389 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
1390 __ bfalse(CCR0, method_is_not_static);
1391
1392 __ ld(R11_scratch1, _abi(callers_sp), R1_SP);
1393 // Load mirror from interpreter frame.
1394 __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1);
1395 // R4_ARG2 = &state->_oop_temp;
1396 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
1397 __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
1398 BIND(method_is_not_static);
1399 }
1400
1401 // At this point, arguments have been copied off the stack into
1402 // their JNI positions. Oops are boxed in-place on the stack, with
1403 // handles copied to arguments. The result handler address is in a
1404 // register.
1405
1406 // Pass JNIEnv address as first parameter.
1407 __ addir(R3_ARG1, thread_(jni_environment));
1408
1409 // Load the native_method entry before we change the thread state.
1410 __ ld(native_method_fd, method_(native_function));
1411
1412 //=============================================================================
1413 // Transition from _thread_in_Java to _thread_in_native. As soon as
1414 // we make this change the safepoint code needs to be certain that
1415 // the last Java frame we established is good. The pc in that frame
1416 // just needs to be near here not an actual return address.
1417
1418 // We use release_store_fence to update values like the thread state, where
1419 // we don't want the current thread to continue until all our prior memory
1420 // accesses (including the new thread state) are visible to other threads.
1421 __ li(R0, _thread_in_native);
1422 __ release();
1423
1424 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
1425 __ stw(R0, thread_(thread_state));
1426
1427 //=============================================================================
1428 // Call the native method. Argument registers must not have been
1429 // overwritten since "__ call_stub(signature_handler);" (except for
1430 // ARG1 and ARG2 for static methods).
1431 __ call_c(native_method_fd);
1432
1433 __ li(R0, 0);
1434 __ ld(R11_scratch1, 0, R1_SP);
1435 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1436 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1437 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
1438
1439 // Note: C++ interpreter needs the following here:
1440 // The frame_manager_lr field, which we use for setting the last
1441 // java frame, gets overwritten by the signature handler. Restore
1442 // it now.
1443 //__ get_PC_trash_LR(R11_scratch1);
1444 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
1445
1446 // Because of GC R19_method may no longer be valid.
1447
1448 // Block, if necessary, before resuming in _thread_in_Java state.
1449 // In order for GC to work, don't clear the last_Java_sp until after
1450 // blocking.
1451
1452 //=============================================================================
1453 // Switch thread to "native transition" state before reading the
1454 // synchronization state. This additional state is necessary
1455 // because reading and testing the synchronization state is not
1456 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1457 // in _thread_in_native state, loads _not_synchronized and is
1458 // preempted. VM thread changes sync state to synchronizing and
1459 // suspends threads for GC. Thread A is resumed to finish this
1460 // native method, but doesn't block here since it didn't see any
1461 // synchronization in progress, and escapes.
1462
1463 // We use release_store_fence to update values like the thread state, where
1464 // we don't want the current thread to continue until all our prior memory
1465 // accesses (including the new thread state) are visible to other threads.
1466 __ li(R0/*thread_state*/, _thread_in_native_trans);
1467 __ release();
1468 __ stw(R0/*thread_state*/, thread_(thread_state));
1469 if (UseMembar) {
1470 __ fence();
1471 }
1472 // Write serialization page so that the VM thread can do a pseudo remote
1473 // membar. We use the current thread pointer to calculate a thread
1474 // specific offset to write to within the page. This minimizes bus
1475 // traffic due to cache line collision.
1476 else {
1477 __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2);
1478 }
1479
1480 // Now before we return to java we must look for a current safepoint
1481 // (a new safepoint can not start since we entered native_trans).
1482 // We must check here because a current safepoint could be modifying
1483 // the callers registers right this moment.
1484
1485 // Acquire isn't strictly necessary here because of the fence, but
1486 // sync_state is declared to be volatile, so we do it anyway
1487 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1488
1489 Label do_safepoint, sync_check_done;
1490 // No synchronization in progress nor yet synchronized.
1491 __ safepoint_poll(do_safepoint, sync_state);
1492
1493 // Not suspended.
1494 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1495 __ lwz(suspend_flags, thread_(suspend_flags));
1496 __ cmpwi(CCR1, suspend_flags, 0);
1497 __ beq(CCR1, sync_check_done);
1498
1499 __ bind(do_safepoint);
1500 __ isync();
1501 // Block. We do the call directly and leave the current
1502 // last_Java_frame setup undisturbed. We must save any possible
1503 // native result across the call. No oop is present.
1504
1505 __ mr(R3_ARG1, R16_thread);
1506 #if defined(ABI_ELFv2)
1507 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
1508 relocInfo::none);
1509 #else
1510 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
1511 relocInfo::none);
1512 #endif
1513
1514 __ bind(sync_check_done);
1515
1516 //=============================================================================
1517 // <<<<<< Back in Interpreter Frame >>>>>
1518
1519 // We are in thread_in_native_trans here and back in the normal
1520 // interpreter frame. We don't have to do anything special about
1521 // safepoints and we can switch to Java mode anytime we are ready.
1522
1523 // Note: frame::interpreter_frame_result has a dependency on how the
1524 // method result is saved across the call to post_method_exit. For
1525 // native methods it assumes that the non-FPU/non-void result is
1526 // saved in _native_lresult and a FPU result in _native_fresult. If
1527 // this changes then the interpreter_frame_result implementation
1528 // will need to be updated too.
1529
1530 // On PPC64, we have stored the result directly after the native call.
1531
1532 //=============================================================================
1533 // Back in Java
1534
1535 // We use release_store_fence to update values like the thread state, where
1536 // we don't want the current thread to continue until all our prior memory
1537 // accesses (including the new thread state) are visible to other threads.
1538 __ li(R0/*thread_state*/, _thread_in_Java);
1539 __ lwsync(); // Acquire safepoint and suspend state, release thread state.
1540 __ stw(R0/*thread_state*/, thread_(thread_state));
1541
1542 if (CheckJNICalls) {
1543 // clear_pending_jni_exception_check
1544 __ load_const_optimized(R0, 0L);
1545 __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread);
1546 }
1547
1548 __ reset_last_Java_frame();
1549
1550 // Jvmdi/jvmpi support. Whether we've got an exception pending or
1551 // not, and whether unlocking throws an exception or not, we notify
1552 // on native method exit. If we do have an exception, we'll end up
1553 // in the caller's context to handle it, so if we don't do the
1554 // notify here, we'll drop it on the floor.
1555 __ notify_method_exit(true/*native method*/,
1556 ilgl /*illegal state (not used for native methods)*/,
1557 InterpreterMacroAssembler::NotifyJVMTI,
1558 false /*check_exceptions*/);
1559
1560 //=============================================================================
1561 // Handle exceptions
1562
1563 if (synchronized) {
1564 // Don't check for exceptions since we're still in the i2n frame. Do that
1565 // manually afterwards.
1566 __ unlock_object(R26_monitor, false); // Can also unlock methods.
1567 }
1568
1569 // Reset active handles after returning from native.
1570 // thread->active_handles()->clear();
1571 __ ld(active_handles, thread_(active_handles));
1572 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1573 __ li(R0, 0);
1574 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles);
1575
1576 Label exception_return_sync_check_already_unlocked;
1577 __ ld(R0/*pending_exception*/, thread_(pending_exception));
1578 __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1579 __ bne(CCR0, exception_return_sync_check_already_unlocked);
1580
1581 //-----------------------------------------------------------------------------
1582 // No exception pending.
1583
1584 // Move native method result back into proper registers and return.
1585 // Invoke result handler (may unbox/promote).
1586 __ ld(R11_scratch1, 0, R1_SP);
1587 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1588 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1589 __ call_stub(result_handler_addr);
1590
1591 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1592
1593 // Must use the return pc which was loaded from the caller's frame
1594 // as the VM uses return-pc-patching for deoptimization.
1595 __ mtlr(R0);
1596 __ blr();
1597
1598 //-----------------------------------------------------------------------------
1599 // An exception is pending. We call into the runtime only if the
1600 // caller was not interpreted. If it was interpreted the
1601 // interpreter will do the correct thing. If it isn't interpreted
1602 // (call stub/compiled code) we will change our return and continue.
1603
1604 BIND(exception_return_sync_check);
1605
1606 if (synchronized) {
1607 // Don't check for exceptions since we're still in the i2n frame. Do that
1608 // manually afterwards.
1609 __ unlock_object(R26_monitor, false); // Can also unlock methods.
1610 }
1611 BIND(exception_return_sync_check_already_unlocked);
1612
1613 const Register return_pc = R31;
1614
1615 __ ld(return_pc, 0, R1_SP);
1616 __ ld(return_pc, _abi(lr), return_pc);
1617
1618 // Get the address of the exception handler.
1619 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1620 R16_thread,
1621 return_pc /* return pc */);
1622 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1623
1624 // Load the PC of the the exception handler into LR.
1625 __ mtlr(R3_RET);
1626
1627 // Load exception into R3_ARG1 and clear pending exception in thread.
1628 __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1629 __ li(R4_ARG2, 0);
1630 __ std(R4_ARG2, thread_(pending_exception));
1631
1632 // Load the original return pc into R4_ARG2.
1633 __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1634
1635 // Return to exception handler.
1636 __ blr();
1637
1638 //=============================================================================
1639 // Counter overflow.
1640
1641 if (inc_counter) {
1642 // Handle invocation counter overflow.
1643 __ bind(invocation_counter_overflow);
1644
1645 generate_counter_overflow(continue_after_compile);
1646 }
1647
1648 return entry;
1649 }
1650
1651 // Generic interpreted method entry to (asm) interpreter.
1652 //
generate_normal_entry(bool synchronized)1653 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1654 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1655 address entry = __ pc();
1656 // Generate the code to allocate the interpreter stack frame.
1657 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1658 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame.
1659
1660 // Does also a stack check to assure this frame fits on the stack.
1661 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1662
1663 // --------------------------------------------------------------------------
1664 // Zero out non-parameter locals.
1665 // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1666 // worth to ask the flag, just do it.
1667 Register Rslot_addr = R6_ARG4,
1668 Rnum = R7_ARG5;
1669 Label Lno_locals, Lzero_loop;
1670
1671 // Set up the zeroing loop.
1672 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1673 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1674 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1675 __ beq(CCR0, Lno_locals);
1676 __ li(R0, 0);
1677 __ mtctr(Rnum);
1678
1679 // The zero locals loop.
1680 __ bind(Lzero_loop);
1681 __ std(R0, 0, Rslot_addr);
1682 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1683 __ bdnz(Lzero_loop);
1684
1685 __ bind(Lno_locals);
1686
1687 // --------------------------------------------------------------------------
1688 // Counter increment and overflow check.
1689 Label invocation_counter_overflow,
1690 profile_method,
1691 profile_method_continue;
1692 if (inc_counter || ProfileInterpreter) {
1693
1694 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1695 if (synchronized) {
1696 // Since at this point in the method invocation the exception handler
1697 // would try to exit the monitor of synchronized methods which hasn't
1698 // been entered yet, we set the thread local variable
1699 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1700 // runtime, exception handling i.e. unlock_if_synchronized_method will
1701 // check this thread local flag.
1702 // This flag has two effects, one is to force an unwind in the topmost
1703 // interpreter frame and not perform an unlock while doing so.
1704 __ li(R0, 1);
1705 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1706 }
1707
1708 // Argument and return type profiling.
1709 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1710
1711 // Increment invocation counter and check for overflow.
1712 if (inc_counter) {
1713 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1714 }
1715
1716 __ bind(profile_method_continue);
1717 }
1718
1719 bang_stack_shadow_pages(false);
1720
1721 if (inc_counter || ProfileInterpreter) {
1722 // Reset the _do_not_unlock_if_synchronized flag.
1723 if (synchronized) {
1724 __ li(R0, 0);
1725 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1726 }
1727 }
1728
1729 // --------------------------------------------------------------------------
1730 // Locking of synchronized methods. Must happen AFTER invocation_counter
1731 // check and stack overflow check, so method is not locked if overflows.
1732 if (synchronized) {
1733 lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1734 }
1735 #ifdef ASSERT
1736 else {
1737 Label Lok;
1738 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1739 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1740 __ asm_assert_eq("method needs synchronization", 0x8521);
1741 __ bind(Lok);
1742 }
1743 #endif // ASSERT
1744
1745 __ verify_thread();
1746
1747 // --------------------------------------------------------------------------
1748 // JVMTI support
1749 __ notify_method_entry();
1750
1751 // --------------------------------------------------------------------------
1752 // Start executing instructions.
1753 __ dispatch_next(vtos);
1754
1755 // --------------------------------------------------------------------------
1756 // Out of line counter overflow and MDO creation code.
1757 if (ProfileInterpreter) {
1758 // We have decided to profile this method in the interpreter.
1759 __ bind(profile_method);
1760 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1761 __ set_method_data_pointer_for_bcp();
1762 __ b(profile_method_continue);
1763 }
1764
1765 if (inc_counter) {
1766 // Handle invocation counter overflow.
1767 __ bind(invocation_counter_overflow);
1768 generate_counter_overflow(profile_method_continue);
1769 }
1770 return entry;
1771 }
1772
1773 // CRC32 Intrinsics.
1774 //
1775 // Contract on scratch and work registers.
1776 // =======================================
1777 //
1778 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1779 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1780 // You can't rely on these registers across calls.
1781 //
1782 // The generators for CRC32_update and for CRC32_updateBytes use the
1783 // scratch/work register set internally, passing the work registers
1784 // as arguments to the MacroAssembler emitters as required.
1785 //
1786 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1787 // Their contents is not constant but may change according to the requirements
1788 // of the emitted code.
1789 //
1790 // All other registers from the scratch/work register set are used "internally"
1791 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1792 // Basically, only R3_RET contains a defined value which is the function result.
1793 //
1794 /**
1795 * Method entry for static native methods:
1796 * int java.util.zip.CRC32.update(int crc, int b)
1797 */
generate_CRC32_update_entry()1798 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1799 if (UseCRC32Intrinsics) {
1800 address start = __ pc(); // Remember stub start address (is rtn value).
1801 Label slow_path;
1802
1803 // Safepoint check
1804 const Register sync_state = R11_scratch1;
1805 __ safepoint_poll(slow_path, sync_state);
1806
1807 // We don't generate local frame and don't align stack because
1808 // we not even call stub code (we generate the code inline)
1809 // and there is no safepoint on this path.
1810
1811 // Load java parameters.
1812 // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1813 const Register argP = R15_esp;
1814 const Register crc = R3_ARG1; // crc value
1815 const Register data = R4_ARG2;
1816 const Register table = R5_ARG3; // address of crc32 table
1817
1818 BLOCK_COMMENT("CRC32_update {");
1819
1820 // Arguments are reversed on java expression stack
1821 #ifdef VM_LITTLE_ENDIAN
1822 int data_offs = 0+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1823 // Being passed as an int, the single byte is at offset +0.
1824 #else
1825 int data_offs = 3+1*wordSize; // (stack) address of byte value. Emitter expects address, not value.
1826 // Being passed from java as an int, the single byte is at offset +3.
1827 #endif
1828 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1829 __ lbz(data, data_offs, argP); // Byte from buffer, zero-extended.
1830 __ load_const_optimized(table, StubRoutines::crc_table_addr(), R0);
1831 __ kernel_crc32_singleByteReg(crc, data, table, true);
1832
1833 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1834 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1835 __ blr();
1836
1837 // Generate a vanilla native entry as the slow path.
1838 BLOCK_COMMENT("} CRC32_update");
1839 BIND(slow_path);
1840 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1841 return start;
1842 }
1843
1844 return NULL;
1845 }
1846
1847 /**
1848 * Method entry for static native methods:
1849 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len)
1850 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1851 */
generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind)1852 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1853 if (UseCRC32Intrinsics) {
1854 address start = __ pc(); // Remember stub start address (is rtn value).
1855 Label slow_path;
1856
1857 // Safepoint check
1858 const Register sync_state = R11_scratch1;
1859 __ safepoint_poll(slow_path, sync_state);
1860
1861 // We don't generate local frame and don't align stack because
1862 // we not even call stub code (we generate the code inline)
1863 // and there is no safepoint on this path.
1864
1865 // Load parameters.
1866 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1867 const Register argP = R15_esp;
1868 const Register crc = R3_ARG1; // crc value
1869 const Register data = R4_ARG2; // address of java byte array
1870 const Register dataLen = R5_ARG3; // source data len
1871 const Register tmp = R11_scratch1;
1872
1873 // Arguments are reversed on java expression stack.
1874 // Calculate address of start element.
1875 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1876 BLOCK_COMMENT("CRC32_updateByteBuffer {");
1877 // crc @ (SP + 5W) (32bit)
1878 // buf @ (SP + 3W) (64bit ptr to long array)
1879 // off @ (SP + 2W) (32bit)
1880 // dataLen @ (SP + 1W) (32bit)
1881 // data = buf + off
1882 __ ld( data, 3*wordSize, argP); // start of byte buffer
1883 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1884 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1885 __ lwz( crc, 5*wordSize, argP); // current crc state
1886 __ add( data, data, tmp); // Add byte buffer offset.
1887 } else { // Used for "updateBytes update".
1888 BLOCK_COMMENT("CRC32_updateBytes {");
1889 // crc @ (SP + 4W) (32bit)
1890 // buf @ (SP + 3W) (64bit ptr to byte array)
1891 // off @ (SP + 2W) (32bit)
1892 // dataLen @ (SP + 1W) (32bit)
1893 // data = buf + off + base_offset
1894 __ ld( data, 3*wordSize, argP); // start of byte buffer
1895 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1896 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1897 __ add( data, data, tmp); // add byte buffer offset
1898 __ lwz( crc, 4*wordSize, argP); // current crc state
1899 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1900 }
1901
1902 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, false);
1903
1904 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1905 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1906 __ blr();
1907
1908 // Generate a vanilla native entry as the slow path.
1909 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1910 BIND(slow_path);
1911 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1912 return start;
1913 }
1914
1915 return NULL;
1916 }
1917
1918
1919 /**
1920 * Method entry for intrinsic-candidate (non-native) methods:
1921 * int java.util.zip.CRC32C.updateBytes( int crc, byte[] b, int off, int end)
1922 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
1923 * Unlike CRC32, CRC32C does not have any methods marked as native
1924 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1925 **/
generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind)1926 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1927 if (UseCRC32CIntrinsics) {
1928 address start = __ pc(); // Remember stub start address (is rtn value).
1929
1930 // We don't generate local frame and don't align stack because
1931 // we not even call stub code (we generate the code inline)
1932 // and there is no safepoint on this path.
1933
1934 // Load parameters.
1935 // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1936 const Register argP = R15_esp;
1937 const Register crc = R3_ARG1; // crc value
1938 const Register data = R4_ARG2; // address of java byte array
1939 const Register dataLen = R5_ARG3; // source data len
1940 const Register tmp = R11_scratch1;
1941
1942 // Arguments are reversed on java expression stack.
1943 // Calculate address of start element.
1944 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer".
1945 BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
1946 // crc @ (SP + 5W) (32bit)
1947 // buf @ (SP + 3W) (64bit ptr to long array)
1948 // off @ (SP + 2W) (32bit)
1949 // dataLen @ (SP + 1W) (32bit)
1950 // data = buf + off
1951 __ ld( data, 3*wordSize, argP); // start of byte buffer
1952 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1953 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1954 __ lwz( crc, 5*wordSize, argP); // current crc state
1955 __ add( data, data, tmp); // Add byte buffer offset.
1956 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1957 } else { // Used for "updateBytes update".
1958 BLOCK_COMMENT("CRC32C_updateBytes {");
1959 // crc @ (SP + 4W) (32bit)
1960 // buf @ (SP + 3W) (64bit ptr to byte array)
1961 // off @ (SP + 2W) (32bit)
1962 // dataLen @ (SP + 1W) (32bit)
1963 // data = buf + off + base_offset
1964 __ ld( data, 3*wordSize, argP); // start of byte buffer
1965 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset
1966 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process
1967 __ add( data, data, tmp); // add byte buffer offset
1968 __ sub( dataLen, dataLen, tmp); // (end_index - offset)
1969 __ lwz( crc, 4*wordSize, argP); // current crc state
1970 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1971 }
1972
1973 __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, true);
1974
1975 // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1976 __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1977 __ blr();
1978
1979 BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
1980 return start;
1981 }
1982
1983 return NULL;
1984 }
1985
1986 // =============================================================================
1987 // Exceptions
1988
generate_throw_exception()1989 void TemplateInterpreterGenerator::generate_throw_exception() {
1990 Register Rexception = R17_tos,
1991 Rcontinuation = R3_RET;
1992
1993 // --------------------------------------------------------------------------
1994 // Entry point if an method returns with a pending exception (rethrow).
1995 Interpreter::_rethrow_exception_entry = __ pc();
1996 {
1997 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
1998 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
1999 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
2000
2001 // Compiled code destroys templateTableBase, reload.
2002 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
2003 }
2004
2005 // Entry point if a interpreted method throws an exception (throw).
2006 Interpreter::_throw_exception_entry = __ pc();
2007 {
2008 __ mr(Rexception, R3_RET);
2009
2010 __ verify_thread();
2011 __ verify_oop(Rexception);
2012
2013 // Expression stack must be empty before entering the VM in case of an exception.
2014 __ empty_expression_stack();
2015 // Find exception handler address and preserve exception oop.
2016 // Call C routine to find handler and jump to it.
2017 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
2018 __ mtctr(Rcontinuation);
2019 // Push exception for exception handler bytecodes.
2020 __ push_ptr(Rexception);
2021
2022 // Jump to exception handler (may be remove activation entry!).
2023 __ bctr();
2024 }
2025
2026 // If the exception is not handled in the current frame the frame is
2027 // removed and the exception is rethrown (i.e. exception
2028 // continuation is _rethrow_exception).
2029 //
2030 // Note: At this point the bci is still the bxi for the instruction
2031 // which caused the exception and the expression stack is
2032 // empty. Thus, for any VM calls at this point, GC will find a legal
2033 // oop map (with empty expression stack).
2034
2035 // In current activation
2036 // tos: exception
2037 // bcp: exception bcp
2038
2039 // --------------------------------------------------------------------------
2040 // JVMTI PopFrame support
2041
2042 Interpreter::_remove_activation_preserving_args_entry = __ pc();
2043 {
2044 // Set the popframe_processing bit in popframe_condition indicating that we are
2045 // currently handling popframe, so that call_VMs that may happen later do not
2046 // trigger new popframe handling cycles.
2047 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2048 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
2049 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2050
2051 // Empty the expression stack, as in normal exception handling.
2052 __ empty_expression_stack();
2053 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
2054
2055 // Check to see whether we are returning to a deoptimized frame.
2056 // (The PopFrame call ensures that the caller of the popped frame is
2057 // either interpreted or compiled and deoptimizes it if compiled.)
2058 // Note that we don't compare the return PC against the
2059 // deoptimization blob's unpack entry because of the presence of
2060 // adapter frames in C2.
2061 Label Lcaller_not_deoptimized;
2062 Register return_pc = R3_ARG1;
2063 __ ld(return_pc, 0, R1_SP);
2064 __ ld(return_pc, _abi(lr), return_pc);
2065 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
2066 __ cmpdi(CCR0, R3_RET, 0);
2067 __ bne(CCR0, Lcaller_not_deoptimized);
2068
2069 // The deoptimized case.
2070 // In this case, we can't call dispatch_next() after the frame is
2071 // popped, but instead must save the incoming arguments and restore
2072 // them after deoptimization has occurred.
2073 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
2074 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
2075 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
2076 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
2077 __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
2078 // Save these arguments.
2079 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
2080
2081 // Inform deoptimization that it is responsible for restoring these arguments.
2082 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
2083 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2084
2085 // Return from the current method into the deoptimization blob. Will eventually
2086 // end up in the deopt interpeter entry, deoptimization prepared everything that
2087 // we will reexecute the call that called us.
2088 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
2089 __ mtlr(return_pc);
2090 __ blr();
2091
2092 // The non-deoptimized case.
2093 __ bind(Lcaller_not_deoptimized);
2094
2095 // Clear the popframe condition flag.
2096 __ li(R0, 0);
2097 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2098
2099 // Get out of the current method and re-execute the call that called us.
2100 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2101 __ restore_interpreter_state(R11_scratch1);
2102 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
2103 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
2104 if (ProfileInterpreter) {
2105 __ set_method_data_pointer_for_bcp();
2106 __ ld(R11_scratch1, 0, R1_SP);
2107 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
2108 }
2109 #if INCLUDE_JVMTI
2110 Label L_done;
2111
2112 __ lbz(R11_scratch1, 0, R14_bcp);
2113 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
2114 __ bne(CCR0, L_done);
2115
2116 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2117 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
2118 __ ld(R4_ARG2, 0, R18_locals);
2119 __ call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp);
2120 __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2121 __ cmpdi(CCR0, R4_ARG2, 0);
2122 __ beq(CCR0, L_done);
2123 __ std(R4_ARG2, wordSize, R15_esp);
2124 __ bind(L_done);
2125 #endif // INCLUDE_JVMTI
2126 __ dispatch_next(vtos);
2127 }
2128 // end of JVMTI PopFrame support
2129
2130 // --------------------------------------------------------------------------
2131 // Remove activation exception entry.
2132 // This is jumped to if an interpreted method can't handle an exception itself
2133 // (we come from the throw/rethrow exception entry above). We're going to call
2134 // into the VM to find the exception handler in the caller, pop the current
2135 // frame and return the handler we calculated.
2136 Interpreter::_remove_activation_entry = __ pc();
2137 {
2138 __ pop_ptr(Rexception);
2139 __ verify_thread();
2140 __ verify_oop(Rexception);
2141 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
2142
2143 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
2144 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
2145
2146 __ get_vm_result(Rexception);
2147
2148 // We are done with this activation frame; find out where to go next.
2149 // The continuation point will be an exception handler, which expects
2150 // the following registers set up:
2151 //
2152 // RET: exception oop
2153 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
2154
2155 Register return_pc = R31; // Needs to survive the runtime call.
2156 __ ld(return_pc, 0, R1_SP);
2157 __ ld(return_pc, _abi(lr), return_pc);
2158 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
2159
2160 // Remove the current activation.
2161 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2162
2163 __ mr(R4_ARG2, return_pc);
2164 __ mtlr(R3_RET);
2165 __ mr(R3_RET, Rexception);
2166 __ blr();
2167 }
2168 }
2169
2170 // JVMTI ForceEarlyReturn support.
2171 // Returns "in the middle" of a method with a "fake" return value.
generate_earlyret_entry_for(TosState state)2172 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2173
2174 Register Rscratch1 = R11_scratch1,
2175 Rscratch2 = R12_scratch2;
2176
2177 address entry = __ pc();
2178 __ empty_expression_stack();
2179
2180 __ load_earlyret_value(state, Rscratch1);
2181
2182 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
2183 // Clear the earlyret state.
2184 __ li(R0, 0);
2185 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
2186
2187 __ remove_activation(state, false, false);
2188 // Copied from TemplateTable::_return.
2189 // Restoration of lr done by remove_activation.
2190 switch (state) {
2191 // Narrow result if state is itos but result type is smaller.
2192 case btos:
2193 case ztos:
2194 case ctos:
2195 case stos:
2196 case itos: __ narrow(R17_tos); /* fall through */
2197 case ltos:
2198 case atos: __ mr(R3_RET, R17_tos); break;
2199 case ftos:
2200 case dtos: __ fmr(F1_RET, F15_ftos); break;
2201 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2202 // to get visible before the reference to the object gets stored anywhere.
2203 __ membar(Assembler::StoreStore); break;
2204 default : ShouldNotReachHere();
2205 }
2206 __ blr();
2207
2208 return entry;
2209 } // end of ForceEarlyReturn support
2210
2211 //-----------------------------------------------------------------------------
2212 // Helper for vtos entry point generation
2213
set_vtos_entry_points(Template * t,address & bep,address & cep,address & sep,address & aep,address & iep,address & lep,address & fep,address & dep,address & vep)2214 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2215 address& bep,
2216 address& cep,
2217 address& sep,
2218 address& aep,
2219 address& iep,
2220 address& lep,
2221 address& fep,
2222 address& dep,
2223 address& vep) {
2224 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2225 Label L;
2226
2227 aep = __ pc(); __ push_ptr(); __ b(L);
2228 fep = __ pc(); __ push_f(); __ b(L);
2229 dep = __ pc(); __ push_d(); __ b(L);
2230 lep = __ pc(); __ push_l(); __ b(L);
2231 __ align(32, 12, 24); // align L
2232 bep = cep = sep =
2233 iep = __ pc(); __ push_i();
2234 vep = __ pc();
2235 __ bind(L);
2236 generate_and_dispatch(t);
2237 }
2238
2239 //-----------------------------------------------------------------------------
2240
2241 // Non-product code
2242 #ifndef PRODUCT
generate_trace_code(TosState state)2243 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2244 //__ flush_bundle();
2245 address entry = __ pc();
2246
2247 const char *bname = NULL;
2248 uint tsize = 0;
2249 switch(state) {
2250 case ftos:
2251 bname = "trace_code_ftos {";
2252 tsize = 2;
2253 break;
2254 case btos:
2255 bname = "trace_code_btos {";
2256 tsize = 2;
2257 break;
2258 case ztos:
2259 bname = "trace_code_ztos {";
2260 tsize = 2;
2261 break;
2262 case ctos:
2263 bname = "trace_code_ctos {";
2264 tsize = 2;
2265 break;
2266 case stos:
2267 bname = "trace_code_stos {";
2268 tsize = 2;
2269 break;
2270 case itos:
2271 bname = "trace_code_itos {";
2272 tsize = 2;
2273 break;
2274 case ltos:
2275 bname = "trace_code_ltos {";
2276 tsize = 3;
2277 break;
2278 case atos:
2279 bname = "trace_code_atos {";
2280 tsize = 2;
2281 break;
2282 case vtos:
2283 // Note: In case of vtos, the topmost of stack value could be a int or doubl
2284 // In case of a double (2 slots) we won't see the 2nd stack value.
2285 // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
2286 bname = "trace_code_vtos {";
2287 tsize = 2;
2288
2289 break;
2290 case dtos:
2291 bname = "trace_code_dtos {";
2292 tsize = 3;
2293 break;
2294 default:
2295 ShouldNotReachHere();
2296 }
2297 BLOCK_COMMENT(bname);
2298
2299 // Support short-cut for TraceBytecodesAt.
2300 // Don't call into the VM if we don't want to trace to speed up things.
2301 Label Lskip_vm_call;
2302 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2303 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
2304 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2305 __ ld(R11_scratch1, offs1, R11_scratch1);
2306 __ lwa(R12_scratch2, offs2, R12_scratch2);
2307 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2308 __ blt(CCR0, Lskip_vm_call);
2309 }
2310
2311 __ push(state);
2312 // Load 2 topmost expression stack values.
2313 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
2314 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
2315 __ mflr(R31);
2316 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
2317 __ mtlr(R31);
2318 __ pop(state);
2319
2320 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2321 __ bind(Lskip_vm_call);
2322 }
2323 __ blr();
2324 BLOCK_COMMENT("} trace_code");
2325 return entry;
2326 }
2327
count_bytecode()2328 void TemplateInterpreterGenerator::count_bytecode() {
2329 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
2330 __ lwz(R12_scratch2, offs, R11_scratch1);
2331 __ addi(R12_scratch2, R12_scratch2, 1);
2332 __ stw(R12_scratch2, offs, R11_scratch1);
2333 }
2334
histogram_bytecode(Template * t)2335 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2336 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
2337 __ lwz(R12_scratch2, offs, R11_scratch1);
2338 __ addi(R12_scratch2, R12_scratch2, 1);
2339 __ stw(R12_scratch2, offs, R11_scratch1);
2340 }
2341
histogram_bytecode_pair(Template * t)2342 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2343 const Register addr = R11_scratch1,
2344 tmp = R12_scratch2;
2345 // Get index, shift out old bytecode, bring in new bytecode, and store it.
2346 // _index = (_index >> log2_number_of_codes) |
2347 // (bytecode << log2_number_of_codes);
2348 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
2349 __ lwz(tmp, offs1, addr);
2350 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
2351 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2352 __ stw(tmp, offs1, addr);
2353
2354 // Bump bucket contents.
2355 // _counters[_index] ++;
2356 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
2357 __ sldi(tmp, tmp, LogBytesPerInt);
2358 __ add(addr, tmp, addr);
2359 __ lwz(tmp, offs2, addr);
2360 __ addi(tmp, tmp, 1);
2361 __ stw(tmp, offs2, addr);
2362 }
2363
trace_bytecode(Template * t)2364 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2365 // Call a little run-time stub to avoid blow-up for each bytecode.
2366 // The run-time runtime saves the right registers, depending on
2367 // the tosca in-state for the given template.
2368
2369 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2370 "entry must have been generated");
2371
2372 // Note: we destroy LR here.
2373 __ bl(Interpreter::trace_code(t->tos_in()));
2374 }
2375
stop_interpreter_at()2376 void TemplateInterpreterGenerator::stop_interpreter_at() {
2377 Label L;
2378 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
2379 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2380 __ ld(R11_scratch1, offs1, R11_scratch1);
2381 __ lwa(R12_scratch2, offs2, R12_scratch2);
2382 __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2383 __ bne(CCR0, L);
2384 __ illtrap();
2385 __ bind(L);
2386 }
2387
2388 #endif // !PRODUCT
2389