1 /*
2 * Copyright (c) 2003, 2020, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2012, 2018 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
27 #include "precompiled.hpp"
28 #include "asm/macroAssembler.inline.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interp_masm_ppc.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "prims/jvmtiThreadState.hpp"
34 #include "runtime/frame.inline.hpp"
35 #include "runtime/safepointMechanism.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "utilities/powerOfTwo.hpp"
38
39 // Implementation of InterpreterMacroAssembler.
40
41 // This file specializes the assembler with interpreter-specific macros.
42
43 #ifdef PRODUCT
44 #define BLOCK_COMMENT(str) // nothing
45 #else
46 #define BLOCK_COMMENT(str) block_comment(str)
47 #endif
48
49 void InterpreterMacroAssembler::null_check_throw(Register a, int offset, Register temp_reg) {
50 address exception_entry = Interpreter::throw_NullPointerException_entry();
51 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry);
52 }
53
54 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) {
55 assert(entry, "Entry must have been generated by now");
56 if (is_within_range_of_b(entry, pc())) {
57 b(entry);
58 } else {
59 load_const_optimized(Rscratch, entry, R0);
60 mtctr(Rscratch);
61 bctr();
62 }
63 }
64
65 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) {
66 Register bytecode = R12_scratch2;
67 if (bcp_incr != 0) {
68 lbzu(bytecode, bcp_incr, R14_bcp);
69 } else {
70 lbz(bytecode, 0, R14_bcp);
71 }
72
73 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state), generate_poll);
74 }
75
76 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
77 // Load current bytecode.
78 Register bytecode = R12_scratch2;
79 lbz(bytecode, 0, R14_bcp);
80 dispatch_Lbyte_code(state, bytecode, table);
81 }
82
83 // Dispatch code executed in the prolog of a bytecode which does not do it's
84 // own dispatch. The dispatch address is computed and placed in R24_dispatch_addr.
85 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
86 Register bytecode = R12_scratch2;
87 lbz(bytecode, bcp_incr, R14_bcp);
88
89 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state));
90
91 sldi(bytecode, bytecode, LogBytesPerWord);
92 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode);
93 }
94
95 // Dispatch code executed in the epilog of a bytecode which does not do it's
96 // own dispatch. The dispatch address in R24_dispatch_addr is used for the
97 // dispatch.
98 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
99 if (bcp_incr) { addi(R14_bcp, R14_bcp, bcp_incr); }
100 mtctr(R24_dispatch_addr);
101 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
102 }
103
104 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
105 assert(scratch_reg != R0, "can't use R0 as scratch_reg here");
106 if (JvmtiExport::can_pop_frame()) {
107 Label L;
108
109 // Check the "pending popframe condition" flag in the current thread.
110 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
111
112 // Initiate popframe handling only if it is not already being
113 // processed. If the flag has the popframe_processing bit set, it
114 // means that this code is called *during* popframe handling - we
115 // don't want to reenter.
116 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit);
117 beq(CCR0, L);
118
119 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit);
120 bne(CCR0, L);
121
122 // Call the Interpreter::remove_activation_preserving_args_entry()
123 // func to get the address of the same-named entrypoint in the
124 // generated interpreter code.
125 #if defined(ABI_ELFv2)
126 call_c(CAST_FROM_FN_PTR(address,
127 Interpreter::remove_activation_preserving_args_entry),
128 relocInfo::none);
129 #else
130 call_c(CAST_FROM_FN_PTR(FunctionDescriptor*,
131 Interpreter::remove_activation_preserving_args_entry),
132 relocInfo::none);
133 #endif
134
135 // Jump to Interpreter::_remove_activation_preserving_args_entry.
136 mtctr(R3_RET);
137 bctr();
138
139 align(32, 12);
140 bind(L);
141 }
142 }
143
144 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
145 const Register Rthr_state_addr = scratch_reg;
146 if (JvmtiExport::can_force_early_return()) {
147 Label Lno_early_ret;
148 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
149 cmpdi(CCR0, Rthr_state_addr, 0);
150 beq(CCR0, Lno_early_ret);
151
152 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr);
153 cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending);
154 bne(CCR0, Lno_early_ret);
155
156 // Jump to Interpreter::_earlyret_entry.
157 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr);
158 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry));
159 mtlr(R3_RET);
160 blr();
161
162 align(32, 12);
163 bind(Lno_early_ret);
164 }
165 }
166
167 void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) {
168 const Register RjvmtiState = Rscratch1;
169 const Register Rscratch2 = R0;
170
171 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
172 li(Rscratch2, 0);
173
174 switch (state) {
175 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
176 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
177 break;
178 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
179 break;
180 case btos: // fall through
181 case ztos: // fall through
182 case ctos: // fall through
183 case stos: // fall through
184 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
185 break;
186 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
187 break;
188 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
189 break;
190 case vtos: break;
191 default : ShouldNotReachHere();
192 }
193
194 // Clean up tos value in the jvmti thread state.
195 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
196 // Set tos state field to illegal value.
197 li(Rscratch2, ilgl);
198 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState);
199 }
200
201 // Common code to dispatch and dispatch_only.
202 // Dispatch value in Lbyte_code and increment Lbcp.
203
204 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) {
205 address table_base = (address)Interpreter::dispatch_table((TosState)0);
206 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base;
207 if (is_simm16(table_offs)) {
208 addi(dst, R25_templateTableBase, (int)table_offs);
209 } else {
210 load_const_optimized(dst, table, R0);
211 }
212 }
213
214 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode,
215 address* table, bool generate_poll) {
216 assert_different_registers(bytecode, R11_scratch1);
217
218 // Calc dispatch table address.
219 load_dispatch_table(R11_scratch1, table);
220
221 if (generate_poll) {
222 address *sfpt_tbl = Interpreter::safept_table(state);
223 if (table != sfpt_tbl) {
224 Label dispatch;
225 ld(R0, in_bytes(Thread::polling_page_offset()), R16_thread);
226 // Armed page has poll_bit set, if poll bit is cleared just continue.
227 andi_(R0, R0, SafepointMechanism::poll_bit());
228 beq(CCR0, dispatch);
229 load_dispatch_table(R11_scratch1, sfpt_tbl);
230 align(32, 16);
231 bind(dispatch);
232 }
233 }
234
235 sldi(R12_scratch2, bytecode, LogBytesPerWord);
236 ldx(R11_scratch1, R11_scratch1, R12_scratch2);
237
238 // Jump off!
239 mtctr(R11_scratch1);
240 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
241 }
242
243 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) {
244 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize);
245 ldx(Rrecv_dst, Rrecv_dst, R15_esp);
246 }
247
248 // helpers for expression stack
249
250 void InterpreterMacroAssembler::pop_i(Register r) {
251 lwzu(r, Interpreter::stackElementSize, R15_esp);
252 }
253
254 void InterpreterMacroAssembler::pop_ptr(Register r) {
255 ldu(r, Interpreter::stackElementSize, R15_esp);
256 }
257
258 void InterpreterMacroAssembler::pop_l(Register r) {
259 ld(r, Interpreter::stackElementSize, R15_esp);
260 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
261 }
262
263 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
264 lfsu(f, Interpreter::stackElementSize, R15_esp);
265 }
266
267 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
268 lfd(f, Interpreter::stackElementSize, R15_esp);
269 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
270 }
271
272 void InterpreterMacroAssembler::push_i(Register r) {
273 stw(r, 0, R15_esp);
274 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
275 }
276
277 void InterpreterMacroAssembler::push_ptr(Register r) {
278 std(r, 0, R15_esp);
279 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
280 }
281
282 void InterpreterMacroAssembler::push_l(Register r) {
283 // Clear unused slot.
284 load_const_optimized(R0, 0L);
285 std(R0, 0, R15_esp);
286 std(r, - Interpreter::stackElementSize, R15_esp);
287 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
288 }
289
290 void InterpreterMacroAssembler::push_f(FloatRegister f) {
291 stfs(f, 0, R15_esp);
292 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
293 }
294
295 void InterpreterMacroAssembler::push_d(FloatRegister f) {
296 stfd(f, - Interpreter::stackElementSize, R15_esp);
297 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
298 }
299
300 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) {
301 std(first, 0, R15_esp);
302 std(second, -Interpreter::stackElementSize, R15_esp);
303 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
304 }
305
306 void InterpreterMacroAssembler::move_l_to_d(Register l, FloatRegister d) {
307 if (VM_Version::has_mtfprd()) {
308 mtfprd(d, l);
309 } else {
310 std(l, 0, R15_esp);
311 lfd(d, 0, R15_esp);
312 }
313 }
314
315 void InterpreterMacroAssembler::move_d_to_l(FloatRegister d, Register l) {
316 if (VM_Version::has_mtfprd()) {
317 mffprd(l, d);
318 } else {
319 stfd(d, 0, R15_esp);
320 ld(l, 0, R15_esp);
321 }
322 }
323
324 void InterpreterMacroAssembler::push(TosState state) {
325 switch (state) {
326 case atos: push_ptr(); break;
327 case btos:
328 case ztos:
329 case ctos:
330 case stos:
331 case itos: push_i(); break;
332 case ltos: push_l(); break;
333 case ftos: push_f(); break;
334 case dtos: push_d(); break;
335 case vtos: /* nothing to do */ break;
336 default : ShouldNotReachHere();
337 }
338 }
339
340 void InterpreterMacroAssembler::pop(TosState state) {
341 switch (state) {
342 case atos: pop_ptr(); break;
343 case btos:
344 case ztos:
345 case ctos:
346 case stos:
347 case itos: pop_i(); break;
348 case ltos: pop_l(); break;
349 case ftos: pop_f(); break;
350 case dtos: pop_d(); break;
351 case vtos: /* nothing to do */ break;
352 default : ShouldNotReachHere();
353 }
354 verify_oop(R17_tos, state);
355 }
356
357 void InterpreterMacroAssembler::empty_expression_stack() {
358 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
359 }
360
361 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset,
362 Register Rdst,
363 signedOrNot is_signed) {
364 #if defined(VM_LITTLE_ENDIAN)
365 if (bcp_offset) {
366 load_const_optimized(Rdst, bcp_offset);
367 lhbrx(Rdst, R14_bcp, Rdst);
368 } else {
369 lhbrx(Rdst, R14_bcp);
370 }
371 if (is_signed == Signed) {
372 extsh(Rdst, Rdst);
373 }
374 #else
375 // Read Java big endian format.
376 if (is_signed == Signed) {
377 lha(Rdst, bcp_offset, R14_bcp);
378 } else {
379 lhz(Rdst, bcp_offset, R14_bcp);
380 }
381 #endif
382 }
383
384 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset,
385 Register Rdst,
386 signedOrNot is_signed) {
387 #if defined(VM_LITTLE_ENDIAN)
388 if (bcp_offset) {
389 load_const_optimized(Rdst, bcp_offset);
390 lwbrx(Rdst, R14_bcp, Rdst);
391 } else {
392 lwbrx(Rdst, R14_bcp);
393 }
394 if (is_signed == Signed) {
395 extsw(Rdst, Rdst);
396 }
397 #else
398 // Read Java big endian format.
399 if (bcp_offset & 3) { // Offset unaligned?
400 load_const_optimized(Rdst, bcp_offset);
401 if (is_signed == Signed) {
402 lwax(Rdst, R14_bcp, Rdst);
403 } else {
404 lwzx(Rdst, R14_bcp, Rdst);
405 }
406 } else {
407 if (is_signed == Signed) {
408 lwa(Rdst, bcp_offset, R14_bcp);
409 } else {
410 lwz(Rdst, bcp_offset, R14_bcp);
411 }
412 }
413 #endif
414 }
415
416
417 // Load the constant pool cache index from the bytecode stream.
418 //
419 // Kills / writes:
420 // - Rdst, Rscratch
421 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset,
422 size_t index_size) {
423 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
424 // Cache index is always in the native format, courtesy of Rewriter.
425 if (index_size == sizeof(u2)) {
426 lhz(Rdst, bcp_offset, R14_bcp);
427 } else if (index_size == sizeof(u4)) {
428 if (bcp_offset & 3) {
429 load_const_optimized(Rdst, bcp_offset);
430 lwax(Rdst, R14_bcp, Rdst);
431 } else {
432 lwa(Rdst, bcp_offset, R14_bcp);
433 }
434 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
435 nand(Rdst, Rdst, Rdst); // convert to plain index
436 } else if (index_size == sizeof(u1)) {
437 lbz(Rdst, bcp_offset, R14_bcp);
438 } else {
439 ShouldNotReachHere();
440 }
441 // Rdst now contains cp cache index.
442 }
443
444 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset,
445 size_t index_size) {
446 get_cache_index_at_bcp(cache, bcp_offset, index_size);
447 sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord));
448 add(cache, R27_constPoolCache, cache);
449 }
450
451 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format)
452 // from (Rsrc)+offset.
453 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset,
454 signedOrNot is_signed) {
455 #if defined(VM_LITTLE_ENDIAN)
456 if (offset) {
457 load_const_optimized(Rdst, offset);
458 lwbrx(Rdst, Rdst, Rsrc);
459 } else {
460 lwbrx(Rdst, Rsrc);
461 }
462 if (is_signed == Signed) {
463 extsw(Rdst, Rdst);
464 }
465 #else
466 if (is_signed == Signed) {
467 lwa(Rdst, offset, Rsrc);
468 } else {
469 lwz(Rdst, offset, Rsrc);
470 }
471 #endif
472 }
473
474 // Load object from cpool->resolved_references(index).
475 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index, Label *L_handle_null) {
476 assert_different_registers(result, index);
477 get_constant_pool(result);
478
479 // Convert from field index to resolved_references() index and from
480 // word index to byte offset. Since this is a java object, it can be compressed.
481 Register tmp = index; // reuse
482 sldi(tmp, index, LogBytesPerHeapOop);
483 // Load pointer for resolved_references[] objArray.
484 ld(result, ConstantPool::cache_offset_in_bytes(), result);
485 ld(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result);
486 resolve_oop_handle(result);
487 #ifdef ASSERT
488 Label index_ok;
489 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result);
490 sldi(R0, R0, LogBytesPerHeapOop);
491 cmpd(CCR0, tmp, R0);
492 blt(CCR0, index_ok);
493 stop("resolved reference index out of bounds");
494 bind(index_ok);
495 #endif
496 // Add in the index.
497 add(result, tmp, result);
498 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result, tmp, R0, false, 0, L_handle_null);
499 }
500
501 // load cpool->resolved_klass_at(index)
502 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool, Register Roffset, Register Rklass) {
503 // int value = *(Rcpool->int_at_addr(which));
504 // int resolved_klass_index = extract_low_short_from_int(value);
505 add(Roffset, Rcpool, Roffset);
506 #if defined(VM_LITTLE_ENDIAN)
507 lhz(Roffset, sizeof(ConstantPool), Roffset); // Roffset = resolved_klass_index
508 #else
509 lhz(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index
510 #endif
511
512 ld(Rklass, ConstantPool::resolved_klasses_offset_in_bytes(), Rcpool); // Rklass = Rcpool->_resolved_klasses
513
514 sldi(Roffset, Roffset, LogBytesPerWord);
515 addi(Roffset, Roffset, Array<Klass*>::base_offset_in_bytes());
516 isync(); // Order load of instance Klass wrt. tags.
517 ldx(Rklass, Rklass, Roffset);
518 }
519
520 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
521 Register cache,
522 Register method) {
523 const int method_offset = in_bytes(
524 ConstantPoolCache::base_offset() +
525 ((byte_no == TemplateTable::f2_byte)
526 ? ConstantPoolCacheEntry::f2_offset()
527 : ConstantPoolCacheEntry::f1_offset()));
528
529 ld(method, method_offset, cache); // get f1 Method*
530 }
531
532 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
533 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2.
534 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1,
535 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) {
536 // Profile the not-null value's klass.
537 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2);
538 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
539 profile_typecheck_failed(Rtmp1, Rtmp2);
540 }
541
542 // Separate these two to allow for delay slot in middle.
543 // These are used to do a test and full jump to exception-throwing code.
544
545 // Check that index is in range for array, then shift index by index_shift,
546 // and put arrayOop + shifted_index into res.
547 // Note: res is still shy of address by array offset into object.
548
549 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex,
550 int index_shift, Register Rtmp, Register Rres) {
551 // Check that index is in range for array, then shift index by index_shift,
552 // and put arrayOop + shifted_index into res.
553 // Note: res is still shy of address by array offset into object.
554 // Kills:
555 // - Rindex
556 // Writes:
557 // - Rres: Address that corresponds to the array index if check was successful.
558 verify_oop(Rarray);
559 const Register Rlength = R0;
560 const Register RsxtIndex = Rtmp;
561 Label LisNull, LnotOOR;
562
563 // Array nullcheck
564 if (!ImplicitNullChecks) {
565 cmpdi(CCR0, Rarray, 0);
566 beq(CCR0, LisNull);
567 } else {
568 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex);
569 }
570
571 // Rindex might contain garbage in upper bits (remember that we don't sign extend
572 // during integer arithmetic operations). So kill them and put value into same register
573 // where ArrayIndexOutOfBounds would expect the index in.
574 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit
575
576 // Index check
577 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray);
578 cmplw(CCR0, Rindex, Rlength);
579 sldi(RsxtIndex, RsxtIndex, index_shift);
580 blt(CCR0, LnotOOR);
581 // Index should be in R17_tos, array should be in R4_ARG2.
582 mr_if_needed(R17_tos, Rindex);
583 mr_if_needed(R4_ARG2, Rarray);
584 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
585 mtctr(Rtmp);
586 bctr();
587
588 if (!ImplicitNullChecks) {
589 bind(LisNull);
590 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry);
591 mtctr(Rtmp);
592 bctr();
593 }
594
595 align(32, 16);
596 bind(LnotOOR);
597
598 // Calc address
599 add(Rres, RsxtIndex, Rarray);
600 }
601
602 void InterpreterMacroAssembler::index_check(Register array, Register index,
603 int index_shift, Register tmp, Register res) {
604 // pop array
605 pop_ptr(array);
606
607 // check array
608 index_check_without_pop(array, index, index_shift, tmp, res);
609 }
610
611 void InterpreterMacroAssembler::get_const(Register Rdst) {
612 ld(Rdst, in_bytes(Method::const_offset()), R19_method);
613 }
614
615 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
616 get_const(Rdst);
617 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
618 }
619
620 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
621 get_constant_pool(Rdst);
622 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst);
623 }
624
625 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
626 get_constant_pool(Rcpool);
627 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool);
628 }
629
630 // Unlock if synchronized method.
631 //
632 // Unlock the receiver if this is a synchronized method.
633 // Unlock any Java monitors from synchronized blocks.
634 //
635 // If there are locked Java monitors
636 // If throw_monitor_exception
637 // throws IllegalMonitorStateException
638 // Else if install_monitor_exception
639 // installs IllegalMonitorStateException
640 // Else
641 // no error processing
642 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
643 bool throw_monitor_exception,
644 bool install_monitor_exception) {
645 Label Lunlocked, Lno_unlock;
646 {
647 Register Rdo_not_unlock_flag = R11_scratch1;
648 Register Raccess_flags = R12_scratch2;
649
650 // Check if synchronized method or unlocking prevented by
651 // JavaThread::do_not_unlock_if_synchronized flag.
652 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
653 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method);
654 li(R0, 0);
655 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag
656
657 push(state);
658
659 // Skip if we don't have to unlock.
660 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0.
661 beq(CCR0, Lunlocked);
662
663 cmpwi(CCR0, Rdo_not_unlock_flag, 0);
664 bne(CCR0, Lno_unlock);
665 }
666
667 // Unlock
668 {
669 Register Rmonitor_base = R11_scratch1;
670
671 Label Lunlock;
672 // If it's still locked, everything is ok, unlock it.
673 ld(Rmonitor_base, 0, R1_SP);
674 addi(Rmonitor_base, Rmonitor_base,
675 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
676
677 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base);
678 cmpdi(CCR0, R0, 0);
679 bne(CCR0, Lunlock);
680
681 // If it's already unlocked, throw exception.
682 if (throw_monitor_exception) {
683 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
684 should_not_reach_here();
685 } else {
686 if (install_monitor_exception) {
687 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
688 b(Lunlocked);
689 }
690 }
691
692 bind(Lunlock);
693 unlock_object(Rmonitor_base);
694 }
695
696 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not.
697 bind(Lunlocked);
698 {
699 Label Lexception, Lrestart;
700 Register Rcurrent_obj_addr = R11_scratch1;
701 const int delta = frame::interpreter_frame_monitor_size_in_bytes();
702 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords");
703
704 bind(Lrestart);
705 // Set up search loop: Calc num of iterations.
706 {
707 Register Riterations = R12_scratch2;
708 Register Rmonitor_base = Rcurrent_obj_addr;
709 ld(Rmonitor_base, 0, R1_SP);
710 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base
711
712 subf_(Riterations, R26_monitor, Rmonitor_base);
713 ble(CCR0, Lno_unlock);
714
715 addi(Rcurrent_obj_addr, Rmonitor_base,
716 BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes());
717 // Check if any monitor is on stack, bail out if not
718 srdi(Riterations, Riterations, exact_log2(delta));
719 mtctr(Riterations);
720 }
721
722 // The search loop: Look for locked monitors.
723 {
724 const Register Rcurrent_obj = R0;
725 Label Lloop;
726
727 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
728 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
729 bind(Lloop);
730
731 // Check if current entry is used.
732 cmpdi(CCR0, Rcurrent_obj, 0);
733 bne(CCR0, Lexception);
734 // Preload next iteration's compare value.
735 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
736 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
737 bdnz(Lloop);
738 }
739 // Fell through: Everything's unlocked => finish.
740 b(Lno_unlock);
741
742 // An object is still locked => need to throw exception.
743 bind(Lexception);
744 if (throw_monitor_exception) {
745 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
746 should_not_reach_here();
747 } else {
748 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
749 // Unlock does not block, so don't have to worry about the frame.
750 Register Rmonitor_addr = R11_scratch1;
751 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta);
752 unlock_object(Rmonitor_addr);
753 if (install_monitor_exception) {
754 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
755 }
756 b(Lrestart);
757 }
758 }
759
760 align(32, 12);
761 bind(Lno_unlock);
762 pop(state);
763 }
764
765 // Support function for remove_activation & Co.
766 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc,
767 Register Rscratch1, Register Rscratch2) {
768 // Pop interpreter frame.
769 ld(Rscratch1, 0, R1_SP); // *SP
770 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp
771 ld(Rscratch2, 0, Rscratch1); // **SP
772 if (return_pc!=noreg) {
773 ld(return_pc, _abi(lr), Rscratch1); // LR
774 }
775
776 // Merge top frames.
777 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP
778 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP
779 }
780
781 void InterpreterMacroAssembler::narrow(Register result) {
782 Register ret_type = R11_scratch1;
783 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
784 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1);
785
786 Label notBool, notByte, notChar, done;
787
788 // common case first
789 cmpwi(CCR0, ret_type, T_INT);
790 beq(CCR0, done);
791
792 cmpwi(CCR0, ret_type, T_BOOLEAN);
793 bne(CCR0, notBool);
794 andi(result, result, 0x1);
795 b(done);
796
797 bind(notBool);
798 cmpwi(CCR0, ret_type, T_BYTE);
799 bne(CCR0, notByte);
800 extsb(result, result);
801 b(done);
802
803 bind(notByte);
804 cmpwi(CCR0, ret_type, T_CHAR);
805 bne(CCR0, notChar);
806 andi(result, result, 0xffff);
807 b(done);
808
809 bind(notChar);
810 // cmpwi(CCR0, ret_type, T_SHORT); // all that's left
811 // bne(CCR0, done);
812 extsh(result, result);
813
814 // Nothing to do for T_INT
815 bind(done);
816 }
817
818 // Remove activation.
819 //
820 // Unlock the receiver if this is a synchronized method.
821 // Unlock any Java monitors from synchronized blocks.
822 // Remove the activation from the stack.
823 //
824 // If there are locked Java monitors
825 // If throw_monitor_exception
826 // throws IllegalMonitorStateException
827 // Else if install_monitor_exception
828 // installs IllegalMonitorStateException
829 // Else
830 // no error processing
831 void InterpreterMacroAssembler::remove_activation(TosState state,
832 bool throw_monitor_exception,
833 bool install_monitor_exception) {
834 BLOCK_COMMENT("remove_activation {");
835 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
836
837 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
838 notify_method_exit(false, state, NotifyJVMTI, true);
839
840 BLOCK_COMMENT("reserved_stack_check:");
841 if (StackReservedPages > 0) {
842 // Test if reserved zone needs to be enabled.
843 Label no_reserved_zone_enabling;
844
845 // Compare frame pointers. There is no good stack pointer, as with stack
846 // frame compression we can get different SPs when we do calls. A subsequent
847 // call could have a smaller SP, so that this compare succeeds for an
848 // inner call of the method annotated with ReservedStack.
849 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread);
850 ld_ptr(R11_scratch1, _abi(callers_sp), R1_SP); // Load frame pointer.
851 cmpld(CCR0, R11_scratch1, R0);
852 blt_predict_taken(CCR0, no_reserved_zone_enabling);
853
854 // Enable reserved zone again, throw stack overflow exception.
855 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread);
856 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
857
858 should_not_reach_here();
859
860 bind(no_reserved_zone_enabling);
861 }
862
863 verify_oop(R17_tos, state);
864 verify_thread();
865
866 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
867 mtlr(R0);
868 BLOCK_COMMENT("} remove_activation");
869 }
870
871 // Lock object
872 //
873 // Registers alive
874 // monitor - Address of the BasicObjectLock to be used for locking,
875 // which must be initialized with the object to lock.
876 // object - Address of the object to be locked.
877 //
878 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
879 if (UseHeavyMonitors) {
880 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
881 monitor, /*check_for_exceptions=*/true);
882 } else {
883 // template code:
884 //
885 // markWord displaced_header = obj->mark().set_unlocked();
886 // monitor->lock()->set_displaced_header(displaced_header);
887 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
888 // // We stored the monitor address into the object's mark word.
889 // } else if (THREAD->is_lock_owned((address)displaced_header))
890 // // Simple recursive case.
891 // monitor->lock()->set_displaced_header(NULL);
892 // } else {
893 // // Slow path.
894 // InterpreterRuntime::monitorenter(THREAD, monitor);
895 // }
896
897 const Register displaced_header = R7_ARG5;
898 const Register object_mark_addr = R8_ARG6;
899 const Register current_header = R9_ARG7;
900 const Register tmp = R10_ARG8;
901
902 Label done;
903 Label cas_failed, slow_case;
904
905 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp);
906
907 // markWord displaced_header = obj->mark().set_unlocked();
908
909 // Load markWord from object into displaced_header.
910 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object);
911
912 if (UseBiasedLocking) {
913 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case);
914 }
915
916 // Set displaced_header to be (markWord of object | UNLOCK_VALUE).
917 ori(displaced_header, displaced_header, markWord::unlocked_value);
918
919 // monitor->lock()->set_displaced_header(displaced_header);
920
921 // Initialize the box (Must happen before we update the object mark!).
922 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() +
923 BasicLock::displaced_header_offset_in_bytes(), monitor);
924
925 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
926
927 // Store stack address of the BasicObjectLock (this is monitor) into object.
928 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
929
930 // Must fence, otherwise, preceding store(s) may float below cmpxchg.
931 // CmpxchgX sets CCR0 to cmpX(current, displaced).
932 cmpxchgd(/*flag=*/CCR0,
933 /*current_value=*/current_header,
934 /*compare_value=*/displaced_header, /*exchange_value=*/monitor,
935 /*where=*/object_mark_addr,
936 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq,
937 MacroAssembler::cmpxchgx_hint_acquire_lock(),
938 noreg,
939 &cas_failed,
940 /*check without membar and ldarx first*/true);
941
942 // If the compare-and-exchange succeeded, then we found an unlocked
943 // object and we have now locked it.
944 b(done);
945 bind(cas_failed);
946
947 // } else if (THREAD->is_lock_owned((address)displaced_header))
948 // // Simple recursive case.
949 // monitor->lock()->set_displaced_header(NULL);
950
951 // We did not see an unlocked object so try the fast recursive case.
952
953 // Check if owner is self by comparing the value in the markWord of object
954 // (current_header) with the stack pointer.
955 sub(current_header, current_header, R1_SP);
956
957 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
958 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markWord::lock_mask_in_place);
959
960 and_(R0/*==0?*/, current_header, tmp);
961 // If condition is true we are done and hence we can store 0 in the displaced
962 // header indicating it is a recursive lock.
963 bne(CCR0, slow_case);
964 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() +
965 BasicLock::displaced_header_offset_in_bytes(), monitor);
966 b(done);
967
968 // } else {
969 // // Slow path.
970 // InterpreterRuntime::monitorenter(THREAD, monitor);
971
972 // None of the above fast optimizations worked so we have to get into the
973 // slow case of monitor enter.
974 bind(slow_case);
975 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
976 monitor, /*check_for_exceptions=*/true);
977 // }
978 align(32, 12);
979 bind(done);
980 }
981 }
982
983 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
984 //
985 // Registers alive
986 // monitor - Address of the BasicObjectLock to be used for locking,
987 // which must be initialized with the object to lock.
988 //
989 // Throw IllegalMonitorException if object is not locked by current thread.
990 void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) {
991 if (UseHeavyMonitors) {
992 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
993 monitor, check_for_exceptions);
994 } else {
995
996 // template code:
997 //
998 // if ((displaced_header = monitor->displaced_header()) == NULL) {
999 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL.
1000 // monitor->set_obj(NULL);
1001 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1002 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1003 // monitor->set_obj(NULL);
1004 // } else {
1005 // // Slow path.
1006 // InterpreterRuntime::monitorexit(THREAD, monitor);
1007 // }
1008
1009 const Register object = R7_ARG5;
1010 const Register displaced_header = R8_ARG6;
1011 const Register object_mark_addr = R9_ARG7;
1012 const Register current_header = R10_ARG8;
1013
1014 Label free_slot;
1015 Label slow_case;
1016
1017 assert_different_registers(object, displaced_header, object_mark_addr, current_header);
1018
1019 if (UseBiasedLocking) {
1020 // The object address from the monitor is in object.
1021 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor);
1022 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0");
1023 biased_locking_exit(CCR0, object, displaced_header, free_slot);
1024 }
1025
1026 // Test first if we are in the fast recursive case.
1027 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() +
1028 BasicLock::displaced_header_offset_in_bytes(), monitor);
1029
1030 // If the displaced header is zero, we have a recursive unlock.
1031 cmpdi(CCR0, displaced_header, 0);
1032 beq(CCR0, free_slot); // recursive unlock
1033
1034 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1035 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1036 // monitor->set_obj(NULL);
1037
1038 // If we still have a lightweight lock, unlock the object and be done.
1039
1040 // The object address from the monitor is in object.
1041 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); }
1042 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
1043
1044 // We have the displaced header in displaced_header. If the lock is still
1045 // lightweight, it will contain the monitor address and we'll store the
1046 // displaced header back into the object's mark word.
1047 // CmpxchgX sets CCR0 to cmpX(current, monitor).
1048 cmpxchgd(/*flag=*/CCR0,
1049 /*current_value=*/current_header,
1050 /*compare_value=*/monitor, /*exchange_value=*/displaced_header,
1051 /*where=*/object_mark_addr,
1052 MacroAssembler::MemBarRel,
1053 MacroAssembler::cmpxchgx_hint_release_lock(),
1054 noreg,
1055 &slow_case);
1056 b(free_slot);
1057
1058 // } else {
1059 // // Slow path.
1060 // InterpreterRuntime::monitorexit(THREAD, monitor);
1061
1062 // The lock has been converted into a heavy lock and hence
1063 // we need to get into the slow case.
1064 bind(slow_case);
1065 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1066 monitor, check_for_exceptions);
1067 // }
1068
1069 Label done;
1070 b(done); // Monitor register may be overwritten! Runtime has already freed the slot.
1071
1072 // Exchange worked, do monitor->set_obj(NULL);
1073 align(32, 12);
1074 bind(free_slot);
1075 li(R0, 0);
1076 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor);
1077 bind(done);
1078 }
1079 }
1080
1081 // Load compiled (i2c) or interpreter entry when calling from interpreted and
1082 // do the call. Centralized so that all interpreter calls will do the same actions.
1083 // If jvmti single stepping is on for a thread we must not call compiled code.
1084 //
1085 // Input:
1086 // - Rtarget_method: method to call
1087 // - Rret_addr: return address
1088 // - 2 scratch regs
1089 //
1090 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr,
1091 Register Rscratch1, Register Rscratch2) {
1092 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr);
1093 // Assume we want to go compiled if available.
1094 const Register Rtarget_addr = Rscratch1;
1095 const Register Rinterp_only = Rscratch2;
1096
1097 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method);
1098
1099 if (JvmtiExport::can_post_interpreter_events()) {
1100 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
1101
1102 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
1103 // compiled code in threads for which the event is enabled. Check here for
1104 // interp_only_mode if these events CAN be enabled.
1105 Label done;
1106 verify_thread();
1107 cmpwi(CCR0, Rinterp_only, 0);
1108 beq(CCR0, done);
1109 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method);
1110 align(32, 12);
1111 bind(done);
1112 }
1113
1114 #ifdef ASSERT
1115 {
1116 Label Lok;
1117 cmpdi(CCR0, Rtarget_addr, 0);
1118 bne(CCR0, Lok);
1119 stop("null entry point");
1120 bind(Lok);
1121 }
1122 #endif // ASSERT
1123
1124 mr(R21_sender_SP, R1_SP);
1125
1126 // Calc a precise SP for the call. The SP value we calculated in
1127 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space
1128 // if esp is not max. Also, the i2c adapter extends the stack space without restoring
1129 // our pre-calced value, so repeating calls via i2c would result in stack overflow.
1130 // Since esp already points to an empty slot, we just have to sub 1 additional slot
1131 // to meet the abi scratch requirements.
1132 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in
1133 // the return entry of the interpreter.
1134 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size);
1135 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address
1136 resize_frame_absolute(Rscratch2, Rscratch2, R0);
1137
1138 mr_if_needed(R19_method, Rtarget_method);
1139 mtctr(Rtarget_addr);
1140 mtlr(Rret_addr);
1141
1142 save_interpreter_state(Rscratch2);
1143 #ifdef ASSERT
1144 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp
1145 cmpd(CCR0, R21_sender_SP, Rscratch1);
1146 asm_assert_eq("top_frame_sp incorrect");
1147 #endif
1148
1149 bctr();
1150 }
1151
1152 // Set the method data pointer for the current bcp.
1153 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1154 assert(ProfileInterpreter, "must be profiling interpreter");
1155 Label get_continue;
1156 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method);
1157 test_method_data_pointer(get_continue);
1158 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp);
1159
1160 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1161 add(R28_mdx, R28_mdx, R3_RET);
1162 bind(get_continue);
1163 }
1164
1165 // Test ImethodDataPtr. If it is null, continue at the specified label.
1166 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1167 assert(ProfileInterpreter, "must be profiling interpreter");
1168 cmpdi(CCR0, R28_mdx, 0);
1169 beq(CCR0, zero_continue);
1170 }
1171
1172 void InterpreterMacroAssembler::verify_method_data_pointer() {
1173 assert(ProfileInterpreter, "must be profiling interpreter");
1174 #ifdef ASSERT
1175 Label verify_continue;
1176 test_method_data_pointer(verify_continue);
1177
1178 // If the mdp is valid, it will point to a DataLayout header which is
1179 // consistent with the bcp. The converse is highly probable also.
1180 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx);
1181 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method);
1182 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1183 add(R11_scratch1, R12_scratch2, R12_scratch2);
1184 cmpd(CCR0, R11_scratch1, R14_bcp);
1185 beq(CCR0, verify_continue);
1186
1187 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx);
1188
1189 bind(verify_continue);
1190 #endif
1191 }
1192
1193 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1194 Register method_counters,
1195 Register Rscratch,
1196 Label &profile_continue) {
1197 assert(ProfileInterpreter, "must be profiling interpreter");
1198 // Control will flow to "profile_continue" if the counter is less than the
1199 // limit or if we call profile_method().
1200 Label done;
1201
1202 // If no method data exists, and the counter is high enough, make one.
1203 lwz(Rscratch, in_bytes(MethodCounters::interpreter_profile_limit_offset()), method_counters);
1204
1205 cmpdi(CCR0, R28_mdx, 0);
1206 // Test to see if we should create a method data oop.
1207 cmpd(CCR1, Rscratch, invocation_count);
1208 bne(CCR0, done);
1209 bge(CCR1, profile_continue);
1210
1211 // Build it now.
1212 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1213 set_method_data_pointer_for_bcp();
1214 b(profile_continue);
1215
1216 align(32, 12);
1217 bind(done);
1218 }
1219
1220 void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register method_counters,
1221 Register target_bcp, Register disp, Register Rtmp) {
1222 assert_different_registers(backedge_count, target_bcp, disp, Rtmp, R4_ARG2);
1223 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
1224
1225 Label did_not_overflow;
1226 Label overflow_with_error;
1227
1228 lwz(Rtmp, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()), method_counters);
1229 cmpw(CCR0, backedge_count, Rtmp);
1230
1231 blt(CCR0, did_not_overflow);
1232
1233 // When ProfileInterpreter is on, the backedge_count comes from the
1234 // methodDataOop, which value does not get reset on the call to
1235 // frequency_counter_overflow(). To avoid excessive calls to the overflow
1236 // routine while the method is being compiled, add a second test to make sure
1237 // the overflow function is called only once every overflow_frequency.
1238 if (ProfileInterpreter) {
1239 const int overflow_frequency = 1024;
1240 andi_(Rtmp, backedge_count, overflow_frequency-1);
1241 bne(CCR0, did_not_overflow);
1242 }
1243
1244 // Overflow in loop, pass branch bytecode.
1245 subf(R4_ARG2, disp, target_bcp); // Compute branch bytecode (previous bcp).
1246 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
1247
1248 // Was an OSR adapter generated?
1249 cmpdi(CCR0, R3_RET, 0);
1250 beq(CCR0, overflow_with_error);
1251
1252 // Has the nmethod been invalidated already?
1253 lbz(Rtmp, nmethod::state_offset(), R3_RET);
1254 cmpwi(CCR0, Rtmp, nmethod::in_use);
1255 bne(CCR0, overflow_with_error);
1256
1257 // Migrate the interpreter frame off of the stack.
1258 // We can use all registers because we will not return to interpreter from this point.
1259
1260 // Save nmethod.
1261 const Register osr_nmethod = R31;
1262 mr(osr_nmethod, R3_RET);
1263 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1264 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1265 reset_last_Java_frame();
1266 // OSR buffer is in ARG1
1267
1268 // Remove the interpreter frame.
1269 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1270
1271 // Jump to the osr code.
1272 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1273 mtlr(R0);
1274 mtctr(R11_scratch1);
1275 bctr();
1276
1277 align(32, 12);
1278 bind(overflow_with_error);
1279 bind(did_not_overflow);
1280 }
1281
1282 // Store a value at some constant offset from the method data pointer.
1283 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1284 assert(ProfileInterpreter, "must be profiling interpreter");
1285
1286 std(value, constant, R28_mdx);
1287 }
1288
1289 // Increment the value at some constant offset from the method data pointer.
1290 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1291 Register counter_addr,
1292 Register Rbumped_count,
1293 bool decrement) {
1294 // Locate the counter at a fixed offset from the mdp:
1295 addi(counter_addr, R28_mdx, constant);
1296 increment_mdp_data_at(counter_addr, Rbumped_count, decrement);
1297 }
1298
1299 // Increment the value at some non-fixed (reg + constant) offset from
1300 // the method data pointer.
1301 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1302 int constant,
1303 Register scratch,
1304 Register Rbumped_count,
1305 bool decrement) {
1306 // Add the constant to reg to get the offset.
1307 add(scratch, R28_mdx, reg);
1308 // Then calculate the counter address.
1309 addi(scratch, scratch, constant);
1310 increment_mdp_data_at(scratch, Rbumped_count, decrement);
1311 }
1312
1313 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr,
1314 Register Rbumped_count,
1315 bool decrement) {
1316 assert(ProfileInterpreter, "must be profiling interpreter");
1317
1318 // Load the counter.
1319 ld(Rbumped_count, 0, counter_addr);
1320
1321 if (decrement) {
1322 // Decrement the register. Set condition codes.
1323 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment);
1324 // Store the decremented counter, if it is still negative.
1325 std(Rbumped_count, 0, counter_addr);
1326 // Note: add/sub overflow check are not ported, since 64 bit
1327 // calculation should never overflow.
1328 } else {
1329 // Increment the register. Set carry flag.
1330 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment);
1331 // Store the incremented counter.
1332 std(Rbumped_count, 0, counter_addr);
1333 }
1334 }
1335
1336 // Set a flag value at the current method data pointer position.
1337 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1338 Register scratch) {
1339 assert(ProfileInterpreter, "must be profiling interpreter");
1340 // Load the data header.
1341 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1342 // Set the flag.
1343 ori(scratch, scratch, flag_constant);
1344 // Store the modified header.
1345 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1346 }
1347
1348 // Test the location at some offset from the method data pointer.
1349 // If it is not equal to value, branch to the not_equal_continue Label.
1350 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1351 Register value,
1352 Label& not_equal_continue,
1353 Register test_out) {
1354 assert(ProfileInterpreter, "must be profiling interpreter");
1355
1356 ld(test_out, offset, R28_mdx);
1357 cmpd(CCR0, value, test_out);
1358 bne(CCR0, not_equal_continue);
1359 }
1360
1361 // Update the method data pointer by the displacement located at some fixed
1362 // offset from the method data pointer.
1363 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1364 Register scratch) {
1365 assert(ProfileInterpreter, "must be profiling interpreter");
1366
1367 ld(scratch, offset_of_disp, R28_mdx);
1368 add(R28_mdx, scratch, R28_mdx);
1369 }
1370
1371 // Update the method data pointer by the displacement located at the
1372 // offset (reg + offset_of_disp).
1373 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1374 int offset_of_disp,
1375 Register scratch) {
1376 assert(ProfileInterpreter, "must be profiling interpreter");
1377
1378 add(scratch, reg, R28_mdx);
1379 ld(scratch, offset_of_disp, scratch);
1380 add(R28_mdx, scratch, R28_mdx);
1381 }
1382
1383 // Update the method data pointer by a simple constant displacement.
1384 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1385 assert(ProfileInterpreter, "must be profiling interpreter");
1386 addi(R28_mdx, R28_mdx, constant);
1387 }
1388
1389 // Update the method data pointer for a _ret bytecode whose target
1390 // was not among our cached targets.
1391 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1392 Register return_bci) {
1393 assert(ProfileInterpreter, "must be profiling interpreter");
1394
1395 push(state);
1396 assert(return_bci->is_nonvolatile(), "need to protect return_bci");
1397 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1398 pop(state);
1399 }
1400
1401 // Increments the backedge counter.
1402 // Returns backedge counter + invocation counter in Rdst.
1403 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst,
1404 const Register Rtmp1, Register Rscratch) {
1405 assert(UseCompiler, "incrementing must be useful");
1406 assert_different_registers(Rdst, Rtmp1);
1407 const Register invocation_counter = Rtmp1;
1408 const Register counter = Rdst;
1409 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size.");
1410
1411 // Load backedge counter.
1412 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1413 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1414 // Load invocation counter.
1415 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) +
1416 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1417
1418 // Add the delta to the backedge counter.
1419 addi(counter, counter, InvocationCounter::count_increment);
1420
1421 // Mask the invocation counter.
1422 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value);
1423
1424 // Store new counter value.
1425 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1426 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1427 // Return invocation counter + backedge counter.
1428 add(counter, counter, invocation_counter);
1429 }
1430
1431 // Count a taken branch in the bytecodes.
1432 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1433 if (ProfileInterpreter) {
1434 Label profile_continue;
1435
1436 // If no method data exists, go to profile_continue.
1437 test_method_data_pointer(profile_continue);
1438
1439 // We are taking a branch. Increment the taken count.
1440 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count);
1441
1442 // The method data pointer needs to be updated to reflect the new target.
1443 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1444 bind (profile_continue);
1445 }
1446 }
1447
1448 // Count a not-taken branch in the bytecodes.
1449 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) {
1450 if (ProfileInterpreter) {
1451 Label profile_continue;
1452
1453 // If no method data exists, go to profile_continue.
1454 test_method_data_pointer(profile_continue);
1455
1456 // We are taking a branch. Increment the not taken count.
1457 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2);
1458
1459 // The method data pointer needs to be updated to correspond to the
1460 // next bytecode.
1461 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1462 bind (profile_continue);
1463 }
1464 }
1465
1466 // Count a non-virtual call in the bytecodes.
1467 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) {
1468 if (ProfileInterpreter) {
1469 Label profile_continue;
1470
1471 // If no method data exists, go to profile_continue.
1472 test_method_data_pointer(profile_continue);
1473
1474 // We are making a call. Increment the count.
1475 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1476
1477 // The method data pointer needs to be updated to reflect the new target.
1478 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1479 bind (profile_continue);
1480 }
1481 }
1482
1483 // Count a final call in the bytecodes.
1484 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) {
1485 if (ProfileInterpreter) {
1486 Label profile_continue;
1487
1488 // If no method data exists, go to profile_continue.
1489 test_method_data_pointer(profile_continue);
1490
1491 // We are making a call. Increment the count.
1492 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1493
1494 // The method data pointer needs to be updated to reflect the new target.
1495 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1496 bind (profile_continue);
1497 }
1498 }
1499
1500 // Count a virtual call in the bytecodes.
1501 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver,
1502 Register Rscratch1,
1503 Register Rscratch2,
1504 bool receiver_can_be_null) {
1505 if (!ProfileInterpreter) { return; }
1506 Label profile_continue;
1507
1508 // If no method data exists, go to profile_continue.
1509 test_method_data_pointer(profile_continue);
1510
1511 Label skip_receiver_profile;
1512 if (receiver_can_be_null) {
1513 Label not_null;
1514 cmpdi(CCR0, Rreceiver, 0);
1515 bne(CCR0, not_null);
1516 // We are making a call. Increment the count for null receiver.
1517 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2);
1518 b(skip_receiver_profile);
1519 bind(not_null);
1520 }
1521
1522 // Record the receiver type.
1523 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true);
1524 bind(skip_receiver_profile);
1525
1526 // The method data pointer needs to be updated to reflect the new target.
1527 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1528 bind (profile_continue);
1529 }
1530
1531 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) {
1532 if (ProfileInterpreter) {
1533 Label profile_continue;
1534
1535 // If no method data exists, go to profile_continue.
1536 test_method_data_pointer(profile_continue);
1537
1538 int mdp_delta = in_bytes(BitData::bit_data_size());
1539 if (TypeProfileCasts) {
1540 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1541
1542 // Record the object type.
1543 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false);
1544 }
1545
1546 // The method data pointer needs to be updated.
1547 update_mdp_by_constant(mdp_delta);
1548
1549 bind (profile_continue);
1550 }
1551 }
1552
1553 void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) {
1554 if (ProfileInterpreter && TypeProfileCasts) {
1555 Label profile_continue;
1556
1557 // If no method data exists, go to profile_continue.
1558 test_method_data_pointer(profile_continue);
1559
1560 int count_offset = in_bytes(CounterData::count_offset());
1561 // Back up the address, since we have already bumped the mdp.
1562 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1563
1564 // *Decrement* the counter. We expect to see zero or small negatives.
1565 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true);
1566
1567 bind (profile_continue);
1568 }
1569 }
1570
1571 // Count a ret in the bytecodes.
1572 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci,
1573 Register scratch1, Register scratch2) {
1574 if (ProfileInterpreter) {
1575 Label profile_continue;
1576 uint row;
1577
1578 // If no method data exists, go to profile_continue.
1579 test_method_data_pointer(profile_continue);
1580
1581 // Update the total ret count.
1582 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 );
1583
1584 for (row = 0; row < RetData::row_limit(); row++) {
1585 Label next_test;
1586
1587 // See if return_bci is equal to bci[n]:
1588 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1);
1589
1590 // return_bci is equal to bci[n]. Increment the count.
1591 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2);
1592
1593 // The method data pointer needs to be updated to reflect the new target.
1594 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1);
1595 b(profile_continue);
1596 bind(next_test);
1597 }
1598
1599 update_mdp_for_ret(state, return_bci);
1600
1601 bind (profile_continue);
1602 }
1603 }
1604
1605 // Count the default case of a switch construct.
1606 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) {
1607 if (ProfileInterpreter) {
1608 Label profile_continue;
1609
1610 // If no method data exists, go to profile_continue.
1611 test_method_data_pointer(profile_continue);
1612
1613 // Update the default case count
1614 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1615 scratch1, scratch2);
1616
1617 // The method data pointer needs to be updated.
1618 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()),
1619 scratch1);
1620
1621 bind (profile_continue);
1622 }
1623 }
1624
1625 // Count the index'th case of a switch construct.
1626 void InterpreterMacroAssembler::profile_switch_case(Register index,
1627 Register scratch1,
1628 Register scratch2,
1629 Register scratch3) {
1630 if (ProfileInterpreter) {
1631 assert_different_registers(index, scratch1, scratch2, scratch3);
1632 Label profile_continue;
1633
1634 // If no method data exists, go to profile_continue.
1635 test_method_data_pointer(profile_continue);
1636
1637 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes().
1638 li(scratch3, in_bytes(MultiBranchData::case_array_offset()));
1639
1640 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works");
1641 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1642 add(scratch1, scratch1, scratch3);
1643
1644 // Update the case count.
1645 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3);
1646
1647 // The method data pointer needs to be updated.
1648 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2);
1649
1650 bind (profile_continue);
1651 }
1652 }
1653
1654 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) {
1655 if (ProfileInterpreter) {
1656 assert_different_registers(Rscratch1, Rscratch2);
1657 Label profile_continue;
1658
1659 // If no method data exists, go to profile_continue.
1660 test_method_data_pointer(profile_continue);
1661
1662 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1);
1663
1664 // The method data pointer needs to be updated.
1665 int mdp_delta = in_bytes(BitData::bit_data_size());
1666 if (TypeProfileCasts) {
1667 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1668 }
1669 update_mdp_by_constant(mdp_delta);
1670
1671 bind (profile_continue);
1672 }
1673 }
1674
1675 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver,
1676 Register Rscratch1, Register Rscratch2,
1677 bool is_virtual_call) {
1678 assert(ProfileInterpreter, "must be profiling");
1679 assert_different_registers(Rreceiver, Rscratch1, Rscratch2);
1680
1681 Label done;
1682 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call);
1683 bind (done);
1684 }
1685
1686 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1687 Register receiver, Register scratch1, Register scratch2,
1688 int start_row, Label& done, bool is_virtual_call) {
1689 if (TypeProfileWidth == 0) {
1690 if (is_virtual_call) {
1691 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1692 }
1693 return;
1694 }
1695
1696 int last_row = VirtualCallData::row_limit() - 1;
1697 assert(start_row <= last_row, "must be work left to do");
1698 // Test this row for both the receiver and for null.
1699 // Take any of three different outcomes:
1700 // 1. found receiver => increment count and goto done
1701 // 2. found null => keep looking for case 1, maybe allocate this cell
1702 // 3. found something else => keep looking for cases 1 and 2
1703 // Case 3 is handled by a recursive call.
1704 for (int row = start_row; row <= last_row; row++) {
1705 Label next_test;
1706 bool test_for_null_also = (row == start_row);
1707
1708 // See if the receiver is receiver[n].
1709 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1710 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1);
1711 // delayed()->tst(scratch);
1712
1713 // The receiver is receiver[n]. Increment count[n].
1714 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1715 increment_mdp_data_at(count_offset, scratch1, scratch2);
1716 b(done);
1717 bind(next_test);
1718
1719 if (test_for_null_also) {
1720 Label found_null;
1721 // Failed the equality check on receiver[n]... Test for null.
1722 if (start_row == last_row) {
1723 // The only thing left to do is handle the null case.
1724 if (is_virtual_call) {
1725 // Scratch1 contains test_out from test_mdp_data_at.
1726 cmpdi(CCR0, scratch1, 0);
1727 beq(CCR0, found_null);
1728 // Receiver did not match any saved receiver and there is no empty row for it.
1729 // Increment total counter to indicate polymorphic case.
1730 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1731 b(done);
1732 bind(found_null);
1733 } else {
1734 cmpdi(CCR0, scratch1, 0);
1735 bne(CCR0, done);
1736 }
1737 break;
1738 }
1739 // Since null is rare, make it be the branch-taken case.
1740 cmpdi(CCR0, scratch1, 0);
1741 beq(CCR0, found_null);
1742
1743 // Put all the "Case 3" tests here.
1744 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call);
1745
1746 // Found a null. Keep searching for a matching receiver,
1747 // but remember that this is an empty (unused) slot.
1748 bind(found_null);
1749 }
1750 }
1751
1752 // In the fall-through case, we found no matching receiver, but we
1753 // observed the receiver[start_row] is NULL.
1754
1755 // Fill in the receiver field and increment the count.
1756 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1757 set_mdp_data_at(recvr_offset, receiver);
1758 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1759 li(scratch1, DataLayout::counter_increment);
1760 set_mdp_data_at(count_offset, scratch1);
1761 if (start_row > 0) {
1762 b(done);
1763 }
1764 }
1765
1766 // Argument and return type profilig.
1767 // kills: tmp, tmp2, R0, CR0, CR1
1768 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base,
1769 RegisterOrConstant mdo_addr_offs,
1770 Register tmp, Register tmp2) {
1771 Label do_nothing, do_update;
1772
1773 // tmp2 = obj is allowed
1774 assert_different_registers(obj, mdo_addr_base, tmp, R0);
1775 assert_different_registers(tmp2, mdo_addr_base, tmp, R0);
1776 const Register klass = tmp2;
1777
1778 verify_oop(obj);
1779
1780 ld(tmp, mdo_addr_offs, mdo_addr_base);
1781
1782 // Set null_seen if obj is 0.
1783 cmpdi(CCR0, obj, 0);
1784 ori(R0, tmp, TypeEntries::null_seen);
1785 beq(CCR0, do_update);
1786
1787 load_klass(klass, obj);
1788
1789 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
1790 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
1791 cmpd(CCR1, R0, klass);
1792 // Klass seen before, nothing to do (regardless of unknown bit).
1793 //beq(CCR1, do_nothing);
1794
1795 andi_(R0, klass, TypeEntries::type_unknown);
1796 // Already unknown. Nothing to do anymore.
1797 //bne(CCR0, do_nothing);
1798 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne
1799 beq(CCR0, do_nothing);
1800
1801 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
1802 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1803 beq(CCR0, do_update); // First time here. Set profile type.
1804
1805 // Different than before. Cannot keep accurate profile.
1806 ori(R0, tmp, TypeEntries::type_unknown);
1807
1808 bind(do_update);
1809 // update profile
1810 std(R0, mdo_addr_offs, mdo_addr_base);
1811
1812 align(32, 12);
1813 bind(do_nothing);
1814 }
1815
1816 void InterpreterMacroAssembler::profile_arguments_type(Register callee,
1817 Register tmp1, Register tmp2,
1818 bool is_virtual) {
1819 if (!ProfileInterpreter) {
1820 return;
1821 }
1822
1823 assert_different_registers(callee, tmp1, tmp2, R28_mdx);
1824
1825 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1826 Label profile_continue;
1827
1828 test_method_data_pointer(profile_continue);
1829
1830 int off_to_start = is_virtual ?
1831 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1832
1833 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx);
1834 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1835 bne(CCR0, profile_continue);
1836
1837 if (MethodData::profile_arguments()) {
1838 Label done;
1839 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1840 add(R28_mdx, off_to_args, R28_mdx);
1841
1842 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1843 if (i > 0 || MethodData::profile_return()) {
1844 // If return value type is profiled we may have no argument to profile.
1845 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1846 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count());
1847 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count());
1848 blt(CCR0, done);
1849 }
1850 ld(tmp1, in_bytes(Method::const_offset()), callee);
1851 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1);
1852 // Stack offset o (zero based) from the start of the argument
1853 // list, for n arguments translates into offset n - o - 1 from
1854 // the end of the argument list. But there's an extra slot at
1855 // the top of the stack. So the offset is n - o from Lesp.
1856 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx);
1857 subf(tmp1, tmp2, tmp1);
1858
1859 sldi(tmp1, tmp1, Interpreter::logStackElementSize);
1860 ldx(tmp1, tmp1, R15_esp);
1861
1862 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1);
1863
1864 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1865 addi(R28_mdx, R28_mdx, to_add);
1866 off_to_args += to_add;
1867 }
1868
1869 if (MethodData::profile_return()) {
1870 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1871 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1872 }
1873
1874 bind(done);
1875
1876 if (MethodData::profile_return()) {
1877 // We're right after the type profile for the last
1878 // argument. tmp1 is the number of cells left in the
1879 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1880 // if there's a return to profile.
1881 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(),
1882 "can't move past ret type");
1883 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size));
1884 add(R28_mdx, tmp1, R28_mdx);
1885 }
1886 } else {
1887 assert(MethodData::profile_return(), "either profile call args or call ret");
1888 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
1889 }
1890
1891 // Mdp points right after the end of the
1892 // CallTypeData/VirtualCallTypeData, right after the cells for the
1893 // return value type if there's one.
1894 align(32, 12);
1895 bind(profile_continue);
1896 }
1897 }
1898
1899 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
1900 assert_different_registers(ret, tmp1, tmp2);
1901 if (ProfileInterpreter && MethodData::profile_return()) {
1902 Label profile_continue;
1903
1904 test_method_data_pointer(profile_continue);
1905
1906 if (MethodData::profile_return_jsr292_only()) {
1907 // If we don't profile all invoke bytecodes we must make sure
1908 // it's a bytecode we indeed profile. We can't go back to the
1909 // begining of the ProfileData we intend to update to check its
1910 // type because we're right after it and we don't known its
1911 // length.
1912 lbz(tmp1, 0, R14_bcp);
1913 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method);
1914 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic);
1915 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle);
1916 cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
1917 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm);
1918 cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
1919 bne(CCR0, profile_continue);
1920 }
1921
1922 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2);
1923
1924 align(32, 12);
1925 bind(profile_continue);
1926 }
1927 }
1928
1929 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2,
1930 Register tmp3, Register tmp4) {
1931 if (ProfileInterpreter && MethodData::profile_parameters()) {
1932 Label profile_continue, done;
1933
1934 test_method_data_pointer(profile_continue);
1935
1936 // Load the offset of the area within the MDO used for
1937 // parameters. If it's negative we're not profiling any parameters.
1938 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx);
1939 cmpwi(CCR0, tmp1, 0);
1940 blt(CCR0, profile_continue);
1941
1942 // Compute a pointer to the area for parameters from the offset
1943 // and move the pointer to the slot for the last
1944 // parameters. Collect profiling from last parameter down.
1945 // mdo start + parameters offset + array length - 1
1946
1947 // Pointer to the parameter area in the MDO.
1948 const Register mdp = tmp1;
1949 add(mdp, tmp1, R28_mdx);
1950
1951 // Offset of the current profile entry to update.
1952 const Register entry_offset = tmp2;
1953 // entry_offset = array len in number of cells
1954 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp);
1955
1956 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1957 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
1958
1959 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
1960 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size));
1961 // entry_offset in bytes
1962 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1963
1964 Label loop;
1965 align(32, 12);
1966 bind(loop);
1967
1968 // Load offset on the stack from the slot for this parameter.
1969 ld(tmp3, entry_offset, mdp);
1970 sldi(tmp3, tmp3, Interpreter::logStackElementSize);
1971 neg(tmp3, tmp3);
1972 // Read the parameter from the local area.
1973 ldx(tmp3, tmp3, R18_locals);
1974
1975 // Make entry_offset now point to the type field for this parameter.
1976 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1977 assert(type_base > off_base, "unexpected");
1978 addi(entry_offset, entry_offset, type_base - off_base);
1979
1980 // Profile the parameter.
1981 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3);
1982
1983 // Go to next parameter.
1984 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base);
1985 cmpdi(CCR0, entry_offset, off_base + delta);
1986 addi(entry_offset, entry_offset, -delta);
1987 bge(CCR0, loop);
1988
1989 align(32, 12);
1990 bind(profile_continue);
1991 }
1992 }
1993
1994 // Add a InterpMonitorElem to stack (see frame_sparc.hpp).
1995 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) {
1996
1997 // Very-local scratch registers.
1998 const Register esp = Rtemp1;
1999 const Register slot = Rtemp2;
2000
2001 // Extracted monitor_size.
2002 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes();
2003 assert(Assembler::is_aligned((unsigned int)monitor_size,
2004 (unsigned int)frame::alignment_in_bytes),
2005 "size of a monitor must respect alignment of SP");
2006
2007 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor
2008 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp
2009
2010 // Shuffle expression stack down. Recall that stack_base points
2011 // just above the new expression stack bottom. Old_tos and new_tos
2012 // are used to scan thru the old and new expression stacks.
2013 if (!stack_is_empty) {
2014 Label copy_slot, copy_slot_finished;
2015 const Register n_slots = slot;
2016
2017 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack).
2018 subf(n_slots, esp, R26_monitor);
2019 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy.
2020 assert(LogBytesPerWord == 3, "conflicts assembler instructions");
2021 beq(CCR0, copy_slot_finished); // Nothing to copy.
2022
2023 mtctr(n_slots);
2024
2025 // loop
2026 bind(copy_slot);
2027 ld(slot, 0, esp); // Move expression stack down.
2028 std(slot, -monitor_size, esp); // distance = monitor_size
2029 addi(esp, esp, BytesPerWord);
2030 bdnz(copy_slot);
2031
2032 bind(copy_slot_finished);
2033 }
2034
2035 addi(R15_esp, R15_esp, -monitor_size);
2036 addi(R26_monitor, R26_monitor, -monitor_size);
2037
2038 // Restart interpreter
2039 }
2040
2041 // ============================================================================
2042 // Java locals access
2043
2044 // Load a local variable at index in Rindex into register Rdst_value.
2045 // Also puts address of local into Rdst_address as a service.
2046 // Kills:
2047 // - Rdst_value
2048 // - Rdst_address
2049 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) {
2050 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2051 subf(Rdst_address, Rdst_address, R18_locals);
2052 lwz(Rdst_value, 0, Rdst_address);
2053 }
2054
2055 // Load a local variable at index in Rindex into register Rdst_value.
2056 // Also puts address of local into Rdst_address as a service.
2057 // Kills:
2058 // - Rdst_value
2059 // - Rdst_address
2060 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) {
2061 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2062 subf(Rdst_address, Rdst_address, R18_locals);
2063 ld(Rdst_value, -8, Rdst_address);
2064 }
2065
2066 // Load a local variable at index in Rindex into register Rdst_value.
2067 // Also puts address of local into Rdst_address as a service.
2068 // Input:
2069 // - Rindex: slot nr of local variable
2070 // Kills:
2071 // - Rdst_value
2072 // - Rdst_address
2073 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value,
2074 Register Rdst_address,
2075 Register Rindex) {
2076 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2077 subf(Rdst_address, Rdst_address, R18_locals);
2078 ld(Rdst_value, 0, Rdst_address);
2079 }
2080
2081 // Load a local variable at index in Rindex into register Rdst_value.
2082 // Also puts address of local into Rdst_address as a service.
2083 // Kills:
2084 // - Rdst_value
2085 // - Rdst_address
2086 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value,
2087 Register Rdst_address,
2088 Register Rindex) {
2089 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2090 subf(Rdst_address, Rdst_address, R18_locals);
2091 lfs(Rdst_value, 0, Rdst_address);
2092 }
2093
2094 // Load a local variable at index in Rindex into register Rdst_value.
2095 // Also puts address of local into Rdst_address as a service.
2096 // Kills:
2097 // - Rdst_value
2098 // - Rdst_address
2099 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value,
2100 Register Rdst_address,
2101 Register Rindex) {
2102 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2103 subf(Rdst_address, Rdst_address, R18_locals);
2104 lfd(Rdst_value, -8, Rdst_address);
2105 }
2106
2107 // Store an int value at local variable slot Rindex.
2108 // Kills:
2109 // - Rindex
2110 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) {
2111 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2112 subf(Rindex, Rindex, R18_locals);
2113 stw(Rvalue, 0, Rindex);
2114 }
2115
2116 // Store a long value at local variable slot Rindex.
2117 // Kills:
2118 // - Rindex
2119 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) {
2120 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2121 subf(Rindex, Rindex, R18_locals);
2122 std(Rvalue, -8, Rindex);
2123 }
2124
2125 // Store an oop value at local variable slot Rindex.
2126 // Kills:
2127 // - Rindex
2128 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) {
2129 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2130 subf(Rindex, Rindex, R18_locals);
2131 std(Rvalue, 0, Rindex);
2132 }
2133
2134 // Store an int value at local variable slot Rindex.
2135 // Kills:
2136 // - Rindex
2137 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) {
2138 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2139 subf(Rindex, Rindex, R18_locals);
2140 stfs(Rvalue, 0, Rindex);
2141 }
2142
2143 // Store an int value at local variable slot Rindex.
2144 // Kills:
2145 // - Rindex
2146 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) {
2147 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2148 subf(Rindex, Rindex, R18_locals);
2149 stfd(Rvalue, -8, Rindex);
2150 }
2151
2152 // Read pending exception from thread and jump to interpreter.
2153 // Throw exception entry if one if pending. Fall through otherwise.
2154 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) {
2155 assert_different_registers(Rscratch1, Rscratch2, R3);
2156 Register Rexception = Rscratch1;
2157 Register Rtmp = Rscratch2;
2158 Label Ldone;
2159 // Get pending exception oop.
2160 ld(Rexception, thread_(pending_exception));
2161 cmpdi(CCR0, Rexception, 0);
2162 beq(CCR0, Ldone);
2163 li(Rtmp, 0);
2164 mr_if_needed(R3, Rexception);
2165 std(Rtmp, thread_(pending_exception)); // Clear exception in thread
2166 if (Interpreter::rethrow_exception_entry() != NULL) {
2167 // Already got entry address.
2168 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry());
2169 } else {
2170 // Dynamically load entry address.
2171 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true);
2172 ld(Rtmp, simm16_rest, Rtmp);
2173 }
2174 mtctr(Rtmp);
2175 save_interpreter_state(Rtmp);
2176 bctr();
2177
2178 align(32, 12);
2179 bind(Ldone);
2180 }
2181
2182 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) {
2183 save_interpreter_state(R11_scratch1);
2184
2185 MacroAssembler::call_VM(oop_result, entry_point, false);
2186
2187 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2188
2189 check_and_handle_popframe(R11_scratch1);
2190 check_and_handle_earlyret(R11_scratch1);
2191 // Now check exceptions manually.
2192 if (check_exceptions) {
2193 check_and_forward_exception(R11_scratch1, R12_scratch2);
2194 }
2195 }
2196
2197 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2198 Register arg_1, bool check_exceptions) {
2199 // ARG1 is reserved for the thread.
2200 mr_if_needed(R4_ARG2, arg_1);
2201 call_VM(oop_result, entry_point, check_exceptions);
2202 }
2203
2204 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2205 Register arg_1, Register arg_2,
2206 bool check_exceptions) {
2207 // ARG1 is reserved for the thread.
2208 mr_if_needed(R4_ARG2, arg_1);
2209 assert(arg_2 != R4_ARG2, "smashed argument");
2210 mr_if_needed(R5_ARG3, arg_2);
2211 call_VM(oop_result, entry_point, check_exceptions);
2212 }
2213
2214 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2215 Register arg_1, Register arg_2, Register arg_3,
2216 bool check_exceptions) {
2217 // ARG1 is reserved for the thread.
2218 mr_if_needed(R4_ARG2, arg_1);
2219 assert(arg_2 != R4_ARG2, "smashed argument");
2220 mr_if_needed(R5_ARG3, arg_2);
2221 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument");
2222 mr_if_needed(R6_ARG4, arg_3);
2223 call_VM(oop_result, entry_point, check_exceptions);
2224 }
2225
2226 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) {
2227 ld(scratch, 0, R1_SP);
2228 std(R15_esp, _ijava_state_neg(esp), scratch);
2229 std(R14_bcp, _ijava_state_neg(bcp), scratch);
2230 std(R26_monitor, _ijava_state_neg(monitors), scratch);
2231 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); }
2232 // Other entries should be unchanged.
2233 }
2234
2235 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) {
2236 ld(scratch, 0, R1_SP);
2237 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception).
2238 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code.
2239 if (!bcp_and_mdx_only) {
2240 // Following ones are Metadata.
2241 ld(R19_method, _ijava_state_neg(method), scratch);
2242 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch);
2243 // Following ones are stack addresses and don't require reload.
2244 ld(R15_esp, _ijava_state_neg(esp), scratch);
2245 ld(R18_locals, _ijava_state_neg(locals), scratch);
2246 ld(R26_monitor, _ijava_state_neg(monitors), scratch);
2247 }
2248 #ifdef ASSERT
2249 {
2250 Label Lok;
2251 subf(R0, R1_SP, scratch);
2252 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size);
2253 bge(CCR0, Lok);
2254 stop("frame too small (restore istate)");
2255 bind(Lok);
2256 }
2257 #endif
2258 }
2259
2260 void InterpreterMacroAssembler::get_method_counters(Register method,
2261 Register Rcounters,
2262 Label& skip) {
2263 BLOCK_COMMENT("Load and ev. allocate counter object {");
2264 Label has_counters;
2265 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2266 cmpdi(CCR0, Rcounters, 0);
2267 bne(CCR0, has_counters);
2268 call_VM(noreg, CAST_FROM_FN_PTR(address,
2269 InterpreterRuntime::build_method_counters), method);
2270 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2271 cmpdi(CCR0, Rcounters, 0);
2272 beq(CCR0, skip); // No MethodCounters, OutOfMemory.
2273 BLOCK_COMMENT("} Load and ev. allocate counter object");
2274
2275 bind(has_counters);
2276 }
2277
2278 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters,
2279 Register iv_be_count,
2280 Register Rtmp_r0) {
2281 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful");
2282 Register invocation_count = iv_be_count;
2283 Register backedge_count = Rtmp_r0;
2284 int delta = InvocationCounter::count_increment;
2285
2286 // Load each counter in a register.
2287 // ld(inv_counter, Rtmp);
2288 // ld(be_counter, Rtmp2);
2289 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() +
2290 InvocationCounter::counter_offset());
2291 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() +
2292 InvocationCounter::counter_offset());
2293
2294 BLOCK_COMMENT("Increment profiling counters {");
2295
2296 // Load the backedge counter.
2297 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int
2298 // Mask the backedge counter.
2299 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value);
2300
2301 // Load the invocation counter.
2302 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int
2303 // Add the delta to the invocation counter and store the result.
2304 addi(invocation_count, invocation_count, delta);
2305 // Store value.
2306 stw(invocation_count, inv_counter_offset, Rcounters);
2307
2308 // Add invocation counter + backedge counter.
2309 add(iv_be_count, backedge_count, invocation_count);
2310
2311 // Note that this macro must leave the backedge_count + invocation_count in
2312 // register iv_be_count!
2313 BLOCK_COMMENT("} Increment profiling counters");
2314 }
2315
2316 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2317 if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); }
2318 }
2319
2320 // Local helper function for the verify_oop_or_return_address macro.
2321 static bool verify_return_address(Method* m, int bci) {
2322 #ifndef PRODUCT
2323 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci;
2324 // Assume it is a valid return address if it is inside m and is preceded by a jsr.
2325 if (!m->contains(pc)) return false;
2326 address jsr_pc;
2327 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2328 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2329 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2330 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2331 #endif // PRODUCT
2332 return false;
2333 }
2334
2335 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2336 if (VerifyFPU) {
2337 unimplemented("verfiyFPU");
2338 }
2339 }
2340
2341 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2342 if (!VerifyOops) return;
2343
2344 // The VM documentation for the astore[_wide] bytecode allows
2345 // the TOS to be not only an oop but also a return address.
2346 Label test;
2347 Label skip;
2348 // See if it is an address (in the current method):
2349
2350 const int log2_bytecode_size_limit = 16;
2351 srdi_(Rtmp, reg, log2_bytecode_size_limit);
2352 bne(CCR0, test);
2353
2354 address fd = CAST_FROM_FN_PTR(address, verify_return_address);
2355 const int nbytes_save = MacroAssembler::num_volatile_regs * 8;
2356 save_volatile_gprs(R1_SP, -nbytes_save); // except R0
2357 save_LR_CR(Rtmp); // Save in old frame.
2358 push_frame_reg_args(nbytes_save, Rtmp);
2359
2360 load_const_optimized(Rtmp, fd, R0);
2361 mr_if_needed(R4_ARG2, reg);
2362 mr(R3_ARG1, R19_method);
2363 call_c(Rtmp); // call C
2364
2365 pop_frame();
2366 restore_LR_CR(Rtmp);
2367 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
2368 b(skip);
2369
2370 // Perform a more elaborate out-of-line call.
2371 // Not an address; verify it:
2372 bind(test);
2373 verify_oop(reg);
2374 bind(skip);
2375 }
2376
2377 // Inline assembly for:
2378 //
2379 // if (thread is in interp_only_mode) {
2380 // InterpreterRuntime::post_method_entry();
2381 // }
2382 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) ||
2383 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) {
2384 // SharedRuntime::jvmpi_method_entry(method, receiver);
2385 // }
2386 void InterpreterMacroAssembler::notify_method_entry() {
2387 // JVMTI
2388 // Whenever JVMTI puts a thread in interp_only_mode, method
2389 // entry/exit events are sent for that thread to track stack
2390 // depth. If it is possible to enter interp_only_mode we add
2391 // the code to check if the event should be sent.
2392 if (JvmtiExport::can_post_interpreter_events()) {
2393 Label jvmti_post_done;
2394
2395 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2396 cmpwi(CCR0, R0, 0);
2397 beq(CCR0, jvmti_post_done);
2398 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry),
2399 /*check_exceptions=*/true);
2400
2401 bind(jvmti_post_done);
2402 }
2403 }
2404
2405 // Inline assembly for:
2406 //
2407 // if (thread is in interp_only_mode) {
2408 // // save result
2409 // InterpreterRuntime::post_method_exit();
2410 // // restore result
2411 // }
2412 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) {
2413 // // save result
2414 // SharedRuntime::jvmpi_method_exit();
2415 // // restore result
2416 // }
2417 //
2418 // Native methods have their result stored in d_tmp and l_tmp.
2419 // Java methods have their result stored in the expression stack.
2420 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state,
2421 NotifyMethodExitMode mode, bool check_exceptions) {
2422 // JVMTI
2423 // Whenever JVMTI puts a thread in interp_only_mode, method
2424 // entry/exit events are sent for that thread to track stack
2425 // depth. If it is possible to enter interp_only_mode we add
2426 // the code to check if the event should be sent.
2427 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2428 Label jvmti_post_done;
2429
2430 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2431 cmpwi(CCR0, R0, 0);
2432 beq(CCR0, jvmti_post_done);
2433 if (!is_native_method) { push(state); } // Expose tos to GC.
2434 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit),
2435 /*check_exceptions=*/check_exceptions);
2436 if (!is_native_method) { pop(state); }
2437
2438 align(32, 12);
2439 bind(jvmti_post_done);
2440 }
2441
2442 // Dtrace support not implemented.
2443 }