1 /*
2 * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2012, 2023 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 "oops/methodData.hpp"
34 #include "prims/jvmtiExport.hpp"
35 #include "prims/jvmtiThreadState.hpp"
36 #include "runtime/frame.inline.hpp"
37 #include "runtime/safepointMechanism.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/vm_version.hpp"
40 #include "utilities/macros.hpp"
41 #include "utilities/powerOfTwo.hpp"
42
43 // Implementation of InterpreterMacroAssembler.
44
45 // This file specializes the assembler with interpreter-specific macros.
46
47 #ifdef PRODUCT
48 #define BLOCK_COMMENT(str) // nothing
49 #else
50 #define BLOCK_COMMENT(str) block_comment(str)
51 #endif
52
53 void InterpreterMacroAssembler::null_check_throw(Register a, int offset, Register temp_reg) {
54 address exception_entry = Interpreter::throw_NullPointerException_entry();
55 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry);
56 }
57
58 void InterpreterMacroAssembler::load_klass_check_null_throw(Register dst, Register src, Register temp_reg) {
59 null_check_throw(src, oopDesc::klass_offset_in_bytes(), temp_reg);
60 load_klass(dst, src);
61 }
62
63 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) {
64 assert(entry, "Entry must have been generated by now");
65 if (is_within_range_of_b(entry, pc())) {
66 b(entry);
67 } else {
68 load_const_optimized(Rscratch, entry, R0);
69 mtctr(Rscratch);
70 bctr();
71 }
72 }
73
74 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) {
75 Register bytecode = R12_scratch2;
76 if (bcp_incr != 0) {
77 lbzu(bytecode, bcp_incr, R14_bcp);
78 } else {
79 lbz(bytecode, 0, R14_bcp);
80 }
81
82 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state), generate_poll);
83 }
84
85 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
86 // Load current bytecode.
87 Register bytecode = R12_scratch2;
88 lbz(bytecode, 0, R14_bcp);
89 dispatch_Lbyte_code(state, bytecode, table);
90 }
91
92 // Dispatch code executed in the prolog of a bytecode which does not do it's
93 // own dispatch. The dispatch address is computed and placed in R24_dispatch_addr.
94 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
95 Register bytecode = R12_scratch2;
96 lbz(bytecode, bcp_incr, R14_bcp);
97
98 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state));
99
100 sldi(bytecode, bytecode, LogBytesPerWord);
101 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode);
102 }
103
104 // Dispatch code executed in the epilog of a bytecode which does not do it's
105 // own dispatch. The dispatch address in R24_dispatch_addr is used for the
106 // dispatch.
107 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
108 if (bcp_incr) { addi(R14_bcp, R14_bcp, bcp_incr); }
109 mtctr(R24_dispatch_addr);
110 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
111 }
112
113 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
114 assert(scratch_reg != R0, "can't use R0 as scratch_reg here");
115 if (JvmtiExport::can_pop_frame()) {
116 Label L;
117
118 // Check the "pending popframe condition" flag in the current thread.
119 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
120
121 // Initiate popframe handling only if it is not already being
122 // processed. If the flag has the popframe_processing bit set, it
123 // means that this code is called *during* popframe handling - we
124 // don't want to reenter.
125 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit);
126 beq(CCR0, L);
127
128 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit);
129 bne(CCR0, L);
130
131 // Call the Interpreter::remove_activation_preserving_args_entry()
132 // func to get the address of the same-named entrypoint in the
133 // generated interpreter code.
134 #if defined(ABI_ELFv2)
135 call_c(CAST_FROM_FN_PTR(address,
136 Interpreter::remove_activation_preserving_args_entry),
137 relocInfo::none);
138 #else
139 call_c(CAST_FROM_FN_PTR(FunctionDescriptor*,
140 Interpreter::remove_activation_preserving_args_entry),
141 relocInfo::none);
142 #endif
143
144 // Jump to Interpreter::_remove_activation_preserving_args_entry.
145 mtctr(R3_RET);
146 bctr();
147
148 align(32, 12);
149 bind(L);
150 }
151 }
152
153 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
154 const Register Rthr_state_addr = scratch_reg;
155 if (JvmtiExport::can_force_early_return()) {
156 Label Lno_early_ret;
157 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
158 cmpdi(CCR0, Rthr_state_addr, 0);
159 beq(CCR0, Lno_early_ret);
160
161 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr);
162 cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending);
163 bne(CCR0, Lno_early_ret);
164
165 // Jump to Interpreter::_earlyret_entry.
166 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr);
167 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry));
168 mtlr(R3_RET);
169 blr();
170
171 align(32, 12);
172 bind(Lno_early_ret);
173 }
174 }
175
176 void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) {
177 const Register RjvmtiState = Rscratch1;
178 const Register Rscratch2 = R0;
179
180 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
181 li(Rscratch2, 0);
182
183 switch (state) {
184 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
185 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
186 break;
187 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
188 break;
189 case btos: // fall through
190 case ztos: // fall through
191 case ctos: // fall through
192 case stos: // fall through
193 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
194 break;
195 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
196 break;
197 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
198 break;
199 case vtos: break;
200 default : ShouldNotReachHere();
201 }
202
203 // Clean up tos value in the jvmti thread state.
204 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
205 // Set tos state field to illegal value.
206 li(Rscratch2, ilgl);
207 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState);
208 }
209
210 // Common code to dispatch and dispatch_only.
211 // Dispatch value in Lbyte_code and increment Lbcp.
212
213 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) {
214 address table_base = (address)Interpreter::dispatch_table((TosState)0);
215 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base;
216 if (is_simm16(table_offs)) {
217 addi(dst, R25_templateTableBase, (int)table_offs);
218 } else {
219 load_const_optimized(dst, table, R0);
220 }
221 }
222
223 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode,
224 address* table, bool generate_poll) {
225 assert_different_registers(bytecode, R11_scratch1);
226
227 // Calc dispatch table address.
228 load_dispatch_table(R11_scratch1, table);
229
230 if (generate_poll) {
231 address *sfpt_tbl = Interpreter::safept_table(state);
232 if (table != sfpt_tbl) {
233 Label dispatch;
234 ld(R0, in_bytes(JavaThread::polling_word_offset()), R16_thread);
235 // Armed page has poll_bit set, if poll bit is cleared just continue.
236 andi_(R0, R0, SafepointMechanism::poll_bit());
237 beq(CCR0, dispatch);
238 load_dispatch_table(R11_scratch1, sfpt_tbl);
239 align(32, 16);
240 bind(dispatch);
241 }
242 }
243
244 sldi(R12_scratch2, bytecode, LogBytesPerWord);
245 ldx(R11_scratch1, R11_scratch1, R12_scratch2);
246
247 // Jump off!
248 mtctr(R11_scratch1);
249 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
250 }
251
252 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) {
253 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize);
254 ldx(Rrecv_dst, Rrecv_dst, R15_esp);
255 }
256
257 // helpers for expression stack
258
259 void InterpreterMacroAssembler::pop_i(Register r) {
260 lwzu(r, Interpreter::stackElementSize, R15_esp);
261 }
262
263 void InterpreterMacroAssembler::pop_ptr(Register r) {
264 ldu(r, Interpreter::stackElementSize, R15_esp);
265 }
266
267 void InterpreterMacroAssembler::pop_l(Register r) {
268 ld(r, Interpreter::stackElementSize, R15_esp);
269 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
270 }
271
272 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
273 lfsu(f, Interpreter::stackElementSize, R15_esp);
274 }
275
276 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
277 lfd(f, Interpreter::stackElementSize, R15_esp);
278 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
279 }
280
281 void InterpreterMacroAssembler::push_i(Register r) {
282 stw(r, 0, R15_esp);
283 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
284 }
285
286 void InterpreterMacroAssembler::push_ptr(Register r) {
287 std(r, 0, R15_esp);
288 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
289 }
290
291 void InterpreterMacroAssembler::push_l(Register r) {
292 // Clear unused slot.
293 load_const_optimized(R0, 0L);
294 std(R0, 0, R15_esp);
295 std(r, - Interpreter::stackElementSize, R15_esp);
296 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
297 }
298
299 void InterpreterMacroAssembler::push_f(FloatRegister f) {
300 stfs(f, 0, R15_esp);
301 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
302 }
303
304 void InterpreterMacroAssembler::push_d(FloatRegister f) {
305 stfd(f, - Interpreter::stackElementSize, R15_esp);
306 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
307 }
308
309 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) {
310 std(first, 0, R15_esp);
311 std(second, -Interpreter::stackElementSize, R15_esp);
312 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
313 }
314
315 void InterpreterMacroAssembler::move_l_to_d(Register l, FloatRegister d) {
316 if (VM_Version::has_mtfprd()) {
317 mtfprd(d, l);
318 } else {
319 std(l, 0, R15_esp);
320 lfd(d, 0, R15_esp);
321 }
322 }
323
324 void InterpreterMacroAssembler::move_d_to_l(FloatRegister d, Register l) {
325 if (VM_Version::has_mtfprd()) {
326 mffprd(l, d);
327 } else {
328 stfd(d, 0, R15_esp);
329 ld(l, 0, R15_esp);
330 }
331 }
332
333 void InterpreterMacroAssembler::push(TosState state) {
334 switch (state) {
335 case atos: push_ptr(); break;
336 case btos:
337 case ztos:
338 case ctos:
339 case stos:
340 case itos: push_i(); break;
341 case ltos: push_l(); break;
342 case ftos: push_f(); break;
343 case dtos: push_d(); break;
344 case vtos: /* nothing to do */ break;
345 default : ShouldNotReachHere();
346 }
347 }
348
349 void InterpreterMacroAssembler::pop(TosState state) {
350 switch (state) {
351 case atos: pop_ptr(); break;
352 case btos:
353 case ztos:
354 case ctos:
355 case stos:
356 case itos: pop_i(); break;
357 case ltos: pop_l(); break;
358 case ftos: pop_f(); break;
359 case dtos: pop_d(); break;
360 case vtos: /* nothing to do */ break;
361 default : ShouldNotReachHere();
362 }
363 verify_oop(R17_tos, state);
364 }
365
366 void InterpreterMacroAssembler::empty_expression_stack() {
367 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
368 }
369
370 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset,
371 Register Rdst,
372 signedOrNot is_signed) {
373 #if defined(VM_LITTLE_ENDIAN)
374 if (bcp_offset) {
375 load_const_optimized(Rdst, bcp_offset);
376 lhbrx(Rdst, R14_bcp, Rdst);
377 } else {
378 lhbrx(Rdst, R14_bcp);
379 }
380 if (is_signed == Signed) {
381 extsh(Rdst, Rdst);
382 }
383 #else
384 // Read Java big endian format.
385 if (is_signed == Signed) {
386 lha(Rdst, bcp_offset, R14_bcp);
387 } else {
388 lhz(Rdst, bcp_offset, R14_bcp);
389 }
390 #endif
391 }
392
393 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset,
394 Register Rdst,
395 signedOrNot is_signed) {
396 #if defined(VM_LITTLE_ENDIAN)
397 if (bcp_offset) {
398 load_const_optimized(Rdst, bcp_offset);
399 lwbrx(Rdst, R14_bcp, Rdst);
400 } else {
401 lwbrx(Rdst, R14_bcp);
402 }
403 if (is_signed == Signed) {
404 extsw(Rdst, Rdst);
405 }
406 #else
407 // Read Java big endian format.
408 if (bcp_offset & 3) { // Offset unaligned?
409 load_const_optimized(Rdst, bcp_offset);
410 if (is_signed == Signed) {
411 lwax(Rdst, R14_bcp, Rdst);
412 } else {
413 lwzx(Rdst, R14_bcp, Rdst);
414 }
415 } else {
416 if (is_signed == Signed) {
417 lwa(Rdst, bcp_offset, R14_bcp);
418 } else {
419 lwz(Rdst, bcp_offset, R14_bcp);
420 }
421 }
422 #endif
423 }
424
425
426 // Load the constant pool cache index from the bytecode stream.
427 //
428 // Kills / writes:
429 // - Rdst, Rscratch
430 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset,
431 size_t index_size) {
432 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
433 // Cache index is always in the native format, courtesy of Rewriter.
434 if (index_size == sizeof(u2)) {
435 lhz(Rdst, bcp_offset, R14_bcp);
436 } else if (index_size == sizeof(u4)) {
437 if (bcp_offset & 3) {
438 load_const_optimized(Rdst, bcp_offset);
439 lwax(Rdst, R14_bcp, Rdst);
440 } else {
441 lwa(Rdst, bcp_offset, R14_bcp);
442 }
443 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
444 nand(Rdst, Rdst, Rdst); // convert to plain index
445 } else if (index_size == sizeof(u1)) {
446 lbz(Rdst, bcp_offset, R14_bcp);
447 } else {
448 ShouldNotReachHere();
449 }
450 // Rdst now contains cp cache index.
451 }
452
453 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset,
454 size_t index_size) {
455 get_cache_index_at_bcp(cache, bcp_offset, index_size);
456 sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord));
457 add(cache, R27_constPoolCache, cache);
458 }
459
460 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format)
461 // from (Rsrc)+offset.
462 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset,
463 signedOrNot is_signed) {
464 #if defined(VM_LITTLE_ENDIAN)
465 if (offset) {
466 load_const_optimized(Rdst, offset);
467 lwbrx(Rdst, Rdst, Rsrc);
468 } else {
469 lwbrx(Rdst, Rsrc);
470 }
471 if (is_signed == Signed) {
472 extsw(Rdst, Rdst);
473 }
474 #else
475 if (is_signed == Signed) {
476 lwa(Rdst, offset, Rsrc);
477 } else {
478 lwz(Rdst, offset, Rsrc);
479 }
480 #endif
481 }
482
483 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
484 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
485 get_cache_index_at_bcp(index, 1, sizeof(u4));
486
487 // Get address of invokedynamic array
488 ld_ptr(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset()), R27_constPoolCache);
489 // Scale the index to be the entry index * sizeof(ResolvedInvokeDynamicInfo)
490 sldi(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
491 add(cache, cache, index);
492 }
493
494 // Load object from cpool->resolved_references(index).
495 // Kills:
496 // - index
497 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index,
498 Register tmp1, Register tmp2,
499 Label *L_handle_null) {
500 assert_different_registers(result, index, tmp1, tmp2);
501 assert(index->is_nonvolatile(), "needs to survive C-call in resolve_oop_handle");
502 get_constant_pool(result);
503
504 // Convert from field index to resolved_references() index and from
505 // word index to byte offset. Since this is a java object, it can be compressed.
506 sldi(index, index, LogBytesPerHeapOop);
507 // Load pointer for resolved_references[] objArray.
508 ld(result, ConstantPool::cache_offset(), result);
509 ld(result, ConstantPoolCache::resolved_references_offset(), result);
510 resolve_oop_handle(result, tmp1, tmp2, MacroAssembler::PRESERVATION_NONE);
511 #ifdef ASSERT
512 Label index_ok;
513 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result);
514 sldi(R0, R0, LogBytesPerHeapOop);
515 cmpd(CCR0, index, R0);
516 blt(CCR0, index_ok);
517 stop("resolved reference index out of bounds");
518 bind(index_ok);
519 #endif
520 // Add in the index.
521 add(result, index, result);
522 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result,
523 tmp1, tmp2,
524 MacroAssembler::PRESERVATION_NONE,
525 0, L_handle_null);
526 }
527
528 // load cpool->resolved_klass_at(index)
529 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool, Register Roffset, Register Rklass) {
530 // int value = *(Rcpool->int_at_addr(which));
531 // int resolved_klass_index = extract_low_short_from_int(value);
532 add(Roffset, Rcpool, Roffset);
533 #if defined(VM_LITTLE_ENDIAN)
534 lhz(Roffset, sizeof(ConstantPool), Roffset); // Roffset = resolved_klass_index
535 #else
536 lhz(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index
537 #endif
538
539 ld(Rklass, ConstantPool::resolved_klasses_offset(), Rcpool); // Rklass = Rcpool->_resolved_klasses
540
541 sldi(Roffset, Roffset, LogBytesPerWord);
542 addi(Roffset, Roffset, Array<Klass*>::base_offset_in_bytes());
543 isync(); // Order load of instance Klass wrt. tags.
544 ldx(Rklass, Rklass, Roffset);
545 }
546
547 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
548 Register cache,
549 Register method) {
550 const int method_offset = in_bytes(
551 ConstantPoolCache::base_offset() +
552 ((byte_no == TemplateTable::f2_byte)
553 ? ConstantPoolCacheEntry::f2_offset()
554 : ConstantPoolCacheEntry::f1_offset()));
555
556 ld(method, method_offset, cache); // get f1 Method*
557 }
558
559 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
560 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2.
561 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1,
562 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) {
563 // Profile the not-null value's klass.
564 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2);
565 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
566 profile_typecheck_failed(Rtmp1, Rtmp2);
567 }
568
569 // Separate these two to allow for delay slot in middle.
570 // These are used to do a test and full jump to exception-throwing code.
571
572 // Check that index is in range for array, then shift index by index_shift,
573 // and put arrayOop + shifted_index into res.
574 // Note: res is still shy of address by array offset into object.
575
576 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex,
577 int index_shift, Register Rtmp, Register Rres) {
578 // Check that index is in range for array, then shift index by index_shift,
579 // and put arrayOop + shifted_index into res.
580 // Note: res is still shy of address by array offset into object.
581 // Kills:
582 // - Rindex
583 // Writes:
584 // - Rres: Address that corresponds to the array index if check was successful.
585 verify_oop(Rarray);
586 const Register Rlength = R0;
587 const Register RsxtIndex = Rtmp;
588 Label LisNull, LnotOOR;
589
590 // Array nullcheck
591 if (!ImplicitNullChecks) {
592 cmpdi(CCR0, Rarray, 0);
593 beq(CCR0, LisNull);
594 } else {
595 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex);
596 }
597
598 // Rindex might contain garbage in upper bits (remember that we don't sign extend
599 // during integer arithmetic operations). So kill them and put value into same register
600 // where ArrayIndexOutOfBounds would expect the index in.
601 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit
602
603 // Index check
604 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray);
605 cmplw(CCR0, Rindex, Rlength);
606 sldi(RsxtIndex, RsxtIndex, index_shift);
607 blt(CCR0, LnotOOR);
608 // Index should be in R17_tos, array should be in R4_ARG2.
609 mr_if_needed(R17_tos, Rindex);
610 mr_if_needed(R4_ARG2, Rarray);
611 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
612 mtctr(Rtmp);
613 bctr();
614
615 if (!ImplicitNullChecks) {
616 bind(LisNull);
617 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry);
618 mtctr(Rtmp);
619 bctr();
620 }
621
622 align(32, 16);
623 bind(LnotOOR);
624
625 // Calc address
626 add(Rres, RsxtIndex, Rarray);
627 }
628
629 void InterpreterMacroAssembler::index_check(Register array, Register index,
630 int index_shift, Register tmp, Register res) {
631 // pop array
632 pop_ptr(array);
633
634 // check array
635 index_check_without_pop(array, index, index_shift, tmp, res);
636 }
637
638 void InterpreterMacroAssembler::get_const(Register Rdst) {
639 ld(Rdst, in_bytes(Method::const_offset()), R19_method);
640 }
641
642 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
643 get_const(Rdst);
644 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
645 }
646
647 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
648 get_constant_pool(Rdst);
649 ld(Rdst, ConstantPool::cache_offset(), Rdst);
650 }
651
652 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
653 get_constant_pool(Rcpool);
654 ld(Rtags, ConstantPool::tags_offset(), Rcpool);
655 }
656
657 // Unlock if synchronized method.
658 //
659 // Unlock the receiver if this is a synchronized method.
660 // Unlock any Java monitors from synchronized blocks.
661 //
662 // If there are locked Java monitors
663 // If throw_monitor_exception
664 // throws IllegalMonitorStateException
665 // Else if install_monitor_exception
666 // installs IllegalMonitorStateException
667 // Else
668 // no error processing
669 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
670 bool throw_monitor_exception,
671 bool install_monitor_exception) {
672 Label Lunlocked, Lno_unlock;
673 {
674 Register Rdo_not_unlock_flag = R11_scratch1;
675 Register Raccess_flags = R12_scratch2;
676
677 // Check if synchronized method or unlocking prevented by
678 // JavaThread::do_not_unlock_if_synchronized flag.
679 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
680 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method);
681 li(R0, 0);
682 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag
683
684 push(state);
685
686 // Skip if we don't have to unlock.
687 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0.
688 beq(CCR0, Lunlocked);
689
690 cmpwi(CCR0, Rdo_not_unlock_flag, 0);
691 bne(CCR0, Lno_unlock);
692 }
693
694 // Unlock
695 {
696 Register Rmonitor_base = R11_scratch1;
697
698 Label Lunlock;
699 // If it's still locked, everything is ok, unlock it.
700 ld(Rmonitor_base, 0, R1_SP);
701 addi(Rmonitor_base, Rmonitor_base,
702 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
703
704 ld(R0, BasicObjectLock::obj_offset(), Rmonitor_base);
705 cmpdi(CCR0, R0, 0);
706 bne(CCR0, Lunlock);
707
708 // If it's already unlocked, throw exception.
709 if (throw_monitor_exception) {
710 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
711 should_not_reach_here();
712 } else {
713 if (install_monitor_exception) {
714 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
715 b(Lunlocked);
716 }
717 }
718
719 bind(Lunlock);
720 unlock_object(Rmonitor_base);
721 }
722
723 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not.
724 bind(Lunlocked);
725 {
726 Label Lexception, Lrestart;
727 Register Rcurrent_obj_addr = R11_scratch1;
728 const int delta = frame::interpreter_frame_monitor_size_in_bytes();
729 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords");
730
731 bind(Lrestart);
732 // Set up search loop: Calc num of iterations.
733 {
734 Register Riterations = R12_scratch2;
735 Register Rmonitor_base = Rcurrent_obj_addr;
736 ld(Rmonitor_base, 0, R1_SP);
737 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base
738
739 subf_(Riterations, R26_monitor, Rmonitor_base);
740 ble(CCR0, Lno_unlock);
741
742 addi(Rcurrent_obj_addr, Rmonitor_base,
743 in_bytes(BasicObjectLock::obj_offset()) - frame::interpreter_frame_monitor_size_in_bytes());
744 // Check if any monitor is on stack, bail out if not
745 srdi(Riterations, Riterations, exact_log2(delta));
746 mtctr(Riterations);
747 }
748
749 // The search loop: Look for locked monitors.
750 {
751 const Register Rcurrent_obj = R0;
752 Label Lloop;
753
754 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
755 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
756 bind(Lloop);
757
758 // Check if current entry is used.
759 cmpdi(CCR0, Rcurrent_obj, 0);
760 bne(CCR0, Lexception);
761 // Preload next iteration's compare value.
762 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
763 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
764 bdnz(Lloop);
765 }
766 // Fell through: Everything's unlocked => finish.
767 b(Lno_unlock);
768
769 // An object is still locked => need to throw exception.
770 bind(Lexception);
771 if (throw_monitor_exception) {
772 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
773 should_not_reach_here();
774 } else {
775 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
776 // Unlock does not block, so don't have to worry about the frame.
777 Register Rmonitor_addr = R11_scratch1;
778 addi(Rmonitor_addr, Rcurrent_obj_addr, -in_bytes(BasicObjectLock::obj_offset()) + delta);
779 unlock_object(Rmonitor_addr);
780 if (install_monitor_exception) {
781 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
782 }
783 b(Lrestart);
784 }
785 }
786
787 align(32, 12);
788 bind(Lno_unlock);
789 pop(state);
790 }
791
792 // Support function for remove_activation & Co.
793 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc,
794 Register Rscratch1, Register Rscratch2) {
795 // Pop interpreter frame.
796 ld(Rscratch1, 0, R1_SP); // *SP
797 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp
798 ld(Rscratch2, 0, Rscratch1); // **SP
799 if (return_pc!=noreg) {
800 ld(return_pc, _abi0(lr), Rscratch1); // LR
801 }
802
803 // Merge top frames.
804 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP
805 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP
806 }
807
808 void InterpreterMacroAssembler::narrow(Register result) {
809 Register ret_type = R11_scratch1;
810 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
811 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1);
812
813 Label notBool, notByte, notChar, done;
814
815 // common case first
816 cmpwi(CCR0, ret_type, T_INT);
817 beq(CCR0, done);
818
819 cmpwi(CCR0, ret_type, T_BOOLEAN);
820 bne(CCR0, notBool);
821 andi(result, result, 0x1);
822 b(done);
823
824 bind(notBool);
825 cmpwi(CCR0, ret_type, T_BYTE);
826 bne(CCR0, notByte);
827 extsb(result, result);
828 b(done);
829
830 bind(notByte);
831 cmpwi(CCR0, ret_type, T_CHAR);
832 bne(CCR0, notChar);
833 andi(result, result, 0xffff);
834 b(done);
835
836 bind(notChar);
837 // cmpwi(CCR0, ret_type, T_SHORT); // all that's left
838 // bne(CCR0, done);
839 extsh(result, result);
840
841 // Nothing to do for T_INT
842 bind(done);
843 }
844
845 // Remove activation.
846 //
847 // Apply stack watermark barrier.
848 // Unlock the receiver if this is a synchronized method.
849 // Unlock any Java monitors from synchronized blocks.
850 // Remove the activation from the stack.
851 //
852 // If there are locked Java monitors
853 // If throw_monitor_exception
854 // throws IllegalMonitorStateException
855 // Else if install_monitor_exception
856 // installs IllegalMonitorStateException
857 // Else
858 // no error processing
859 void InterpreterMacroAssembler::remove_activation(TosState state,
860 bool throw_monitor_exception,
861 bool install_monitor_exception) {
862 BLOCK_COMMENT("remove_activation {");
863
864 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
865 // that would normally not be safe to use. Such bad returns into unsafe territory of
866 // the stack, will call InterpreterRuntime::at_unwind.
867 Label slow_path;
868 Label fast_path;
869 safepoint_poll(slow_path, R11_scratch1, true /* at_return */, false /* in_nmethod */);
870 b(fast_path);
871 bind(slow_path);
872 push(state);
873 set_last_Java_frame(R1_SP, noreg);
874 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), R16_thread);
875 reset_last_Java_frame();
876 pop(state);
877 align(32);
878 bind(fast_path);
879
880 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
881
882 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
883 notify_method_exit(false, state, NotifyJVMTI, true);
884
885 BLOCK_COMMENT("reserved_stack_check:");
886 if (StackReservedPages > 0) {
887 // Test if reserved zone needs to be enabled.
888 Label no_reserved_zone_enabling;
889
890 // check if already enabled - if so no re-enabling needed
891 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
892 lwz(R0, in_bytes(JavaThread::stack_guard_state_offset()), R16_thread);
893 cmpwi(CCR0, R0, StackOverflow::stack_guard_enabled);
894 beq_predict_taken(CCR0, no_reserved_zone_enabling);
895
896 // Compare frame pointers. There is no good stack pointer, as with stack
897 // frame compression we can get different SPs when we do calls. A subsequent
898 // call could have a smaller SP, so that this compare succeeds for an
899 // inner call of the method annotated with ReservedStack.
900 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread);
901 ld_ptr(R11_scratch1, _abi0(callers_sp), R1_SP); // Load frame pointer.
902 cmpld(CCR0, R11_scratch1, R0);
903 blt_predict_taken(CCR0, no_reserved_zone_enabling);
904
905 // Enable reserved zone again, throw stack overflow exception.
906 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread);
907 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
908
909 should_not_reach_here();
910
911 bind(no_reserved_zone_enabling);
912 }
913
914 verify_oop(R17_tos, state);
915
916 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
917 mtlr(R0);
918 pop_cont_fastpath();
919 BLOCK_COMMENT("} remove_activation");
920 }
921
922 // Lock object
923 //
924 // Registers alive
925 // monitor - Address of the BasicObjectLock to be used for locking,
926 // which must be initialized with the object to lock.
927 // object - Address of the object to be locked.
928 //
929 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
930 if (LockingMode == LM_MONITOR) {
931 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
932 } else {
933 // template code (for LM_LEGACY):
934 //
935 // markWord displaced_header = obj->mark().set_unlocked();
936 // monitor->lock()->set_displaced_header(displaced_header);
937 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
938 // // We stored the monitor address into the object's mark word.
939 // } else if (THREAD->is_lock_owned((address)displaced_header))
940 // // Simple recursive case.
941 // monitor->lock()->set_displaced_header(nullptr);
942 // } else {
943 // // Slow path.
944 // InterpreterRuntime::monitorenter(THREAD, monitor);
945 // }
946
947 const Register header = R7_ARG5;
948 const Register object_mark_addr = R8_ARG6;
949 const Register current_header = R9_ARG7;
950 const Register tmp = R10_ARG8;
951
952 Label count_locking, done;
953 Label cas_failed, slow_case;
954
955 assert_different_registers(header, object_mark_addr, current_header, tmp);
956
957 // markWord displaced_header = obj->mark().set_unlocked();
958
959 // Load markWord from object into header.
960 ld(header, oopDesc::mark_offset_in_bytes(), object);
961
962 if (DiagnoseSyncOnValueBasedClasses != 0) {
963 load_klass(tmp, object);
964 lwz(tmp, in_bytes(Klass::access_flags_offset()), tmp);
965 testbitdi(CCR0, R0, tmp, exact_log2(JVM_ACC_IS_VALUE_BASED_CLASS));
966 bne(CCR0, slow_case);
967 }
968
969 if (LockingMode == LM_LIGHTWEIGHT) {
970 lightweight_lock(object, /* mark word */ header, tmp, slow_case);
971 b(count_locking);
972 } else if (LockingMode == LM_LEGACY) {
973
974 // Set displaced_header to be (markWord of object | UNLOCK_VALUE).
975 ori(header, header, markWord::unlocked_value);
976
977 // monitor->lock()->set_displaced_header(displaced_header);
978 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
979 const int mark_offset = lock_offset +
980 BasicLock::displaced_header_offset_in_bytes();
981
982 // Initialize the box (Must happen before we update the object mark!).
983 std(header, mark_offset, monitor);
984
985 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
986
987 // Store stack address of the BasicObjectLock (this is monitor) into object.
988 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
989
990 // Must fence, otherwise, preceding store(s) may float below cmpxchg.
991 // CmpxchgX sets CCR0 to cmpX(current, displaced).
992 cmpxchgd(/*flag=*/CCR0,
993 /*current_value=*/current_header,
994 /*compare_value=*/header, /*exchange_value=*/monitor,
995 /*where=*/object_mark_addr,
996 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq,
997 MacroAssembler::cmpxchgx_hint_acquire_lock(),
998 noreg,
999 &cas_failed,
1000 /*check without membar and ldarx first*/true);
1001
1002 // If the compare-and-exchange succeeded, then we found an unlocked
1003 // object and we have now locked it.
1004 b(count_locking);
1005 bind(cas_failed);
1006
1007 // } else if (THREAD->is_lock_owned((address)displaced_header))
1008 // // Simple recursive case.
1009 // monitor->lock()->set_displaced_header(nullptr);
1010
1011 // We did not see an unlocked object so try the fast recursive case.
1012
1013 // Check if owner is self by comparing the value in the markWord of object
1014 // (current_header) with the stack pointer.
1015 sub(current_header, current_header, R1_SP);
1016
1017 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1018 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markWord::lock_mask_in_place);
1019
1020 and_(R0/*==0?*/, current_header, tmp);
1021 // If condition is true we are done and hence we can store 0 in the displaced
1022 // header indicating it is a recursive lock.
1023 bne(CCR0, slow_case);
1024 std(R0/*==0!*/, mark_offset, monitor);
1025 b(count_locking);
1026 }
1027
1028 // } else {
1029 // // Slow path.
1030 // InterpreterRuntime::monitorenter(THREAD, monitor);
1031
1032 // None of the above fast optimizations worked so we have to get into the
1033 // slow case of monitor enter.
1034 bind(slow_case);
1035 if (LockingMode == LM_LIGHTWEIGHT) {
1036 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj), object);
1037 } else {
1038 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
1039 }
1040 b(done);
1041 // }
1042 align(32, 12);
1043 bind(count_locking);
1044 inc_held_monitor_count(current_header /*tmp*/);
1045 bind(done);
1046 }
1047 }
1048
1049 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1050 //
1051 // Registers alive
1052 // monitor - Address of the BasicObjectLock to be used for locking,
1053 // which must be initialized with the object to lock.
1054 //
1055 // Throw IllegalMonitorException if object is not locked by current thread.
1056 void InterpreterMacroAssembler::unlock_object(Register monitor) {
1057 if (LockingMode == LM_MONITOR) {
1058 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1059 } else {
1060
1061 // template code (for LM_LEGACY):
1062 //
1063 // if ((displaced_header = monitor->displaced_header()) == nullptr) {
1064 // // Recursive unlock. Mark the monitor unlocked by setting the object field to null.
1065 // monitor->set_obj(nullptr);
1066 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1067 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1068 // monitor->set_obj(nullptr);
1069 // } else {
1070 // // Slow path.
1071 // InterpreterRuntime::monitorexit(monitor);
1072 // }
1073
1074 const Register object = R7_ARG5;
1075 const Register header = R8_ARG6;
1076 const Register object_mark_addr = R9_ARG7;
1077 const Register current_header = R10_ARG8;
1078
1079 Label free_slot;
1080 Label slow_case;
1081
1082 assert_different_registers(object, header, object_mark_addr, current_header);
1083
1084 if (LockingMode != LM_LIGHTWEIGHT) {
1085 // Test first if we are in the fast recursive case.
1086 ld(header, in_bytes(BasicObjectLock::lock_offset()) +
1087 BasicLock::displaced_header_offset_in_bytes(), monitor);
1088
1089 // If the displaced header is zero, we have a recursive unlock.
1090 cmpdi(CCR0, header, 0);
1091 beq(CCR0, free_slot); // recursive unlock
1092 }
1093
1094 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1095 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1096 // monitor->set_obj(nullptr);
1097
1098 // If we still have a lightweight lock, unlock the object and be done.
1099
1100 // The object address from the monitor is in object.
1101 ld(object, in_bytes(BasicObjectLock::obj_offset()), monitor);
1102
1103 if (LockingMode == LM_LIGHTWEIGHT) {
1104 // Check for non-symmetric locking. This is allowed by the spec and the interpreter
1105 // must handle it.
1106 Register tmp = current_header;
1107 // First check for lock-stack underflow.
1108 lwz(tmp, in_bytes(JavaThread::lock_stack_top_offset()), R16_thread);
1109 cmplwi(CCR0, tmp, (unsigned)LockStack::start_offset());
1110 ble(CCR0, slow_case);
1111 // Then check if the top of the lock-stack matches the unlocked object.
1112 addi(tmp, tmp, -oopSize);
1113 ldx(tmp, tmp, R16_thread);
1114 cmpd(CCR0, tmp, object);
1115 bne(CCR0, slow_case);
1116
1117 ld(header, oopDesc::mark_offset_in_bytes(), object);
1118 andi_(R0, header, markWord::monitor_value);
1119 bne(CCR0, slow_case);
1120 lightweight_unlock(object, header, slow_case);
1121 } else {
1122 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
1123
1124 // We have the displaced header in displaced_header. If the lock is still
1125 // lightweight, it will contain the monitor address and we'll store the
1126 // displaced header back into the object's mark word.
1127 // CmpxchgX sets CCR0 to cmpX(current, monitor).
1128 cmpxchgd(/*flag=*/CCR0,
1129 /*current_value=*/current_header,
1130 /*compare_value=*/monitor, /*exchange_value=*/header,
1131 /*where=*/object_mark_addr,
1132 MacroAssembler::MemBarRel,
1133 MacroAssembler::cmpxchgx_hint_release_lock(),
1134 noreg,
1135 &slow_case);
1136 }
1137 b(free_slot);
1138
1139 // } else {
1140 // // Slow path.
1141 // InterpreterRuntime::monitorexit(monitor);
1142
1143 // The lock has been converted into a heavy lock and hence
1144 // we need to get into the slow case.
1145 bind(slow_case);
1146 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1147 // }
1148
1149 Label done;
1150 b(done); // Monitor register may be overwritten! Runtime has already freed the slot.
1151
1152 // Exchange worked, do monitor->set_obj(nullptr);
1153 align(32, 12);
1154 bind(free_slot);
1155 li(R0, 0);
1156 std(R0, in_bytes(BasicObjectLock::obj_offset()), monitor);
1157 dec_held_monitor_count(current_header /*tmp*/);
1158 bind(done);
1159 }
1160 }
1161
1162 // Load compiled (i2c) or interpreter entry when calling from interpreted and
1163 // do the call. Centralized so that all interpreter calls will do the same actions.
1164 // If jvmti single stepping is on for a thread we must not call compiled code.
1165 //
1166 // Input:
1167 // - Rtarget_method: method to call
1168 // - Rret_addr: return address
1169 // - 2 scratch regs
1170 //
1171 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr,
1172 Register Rscratch1, Register Rscratch2) {
1173 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr);
1174 // Assume we want to go compiled if available.
1175 const Register Rtarget_addr = Rscratch1;
1176 const Register Rinterp_only = Rscratch2;
1177
1178 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method);
1179
1180 if (JvmtiExport::can_post_interpreter_events()) {
1181 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
1182
1183 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
1184 // compiled code in threads for which the event is enabled. Check here for
1185 // interp_only_mode if these events CAN be enabled.
1186 Label done;
1187 cmpwi(CCR0, Rinterp_only, 0);
1188 beq(CCR0, done);
1189 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method);
1190 align(32, 12);
1191 bind(done);
1192 }
1193
1194 #ifdef ASSERT
1195 {
1196 Label Lok;
1197 cmpdi(CCR0, Rtarget_addr, 0);
1198 bne(CCR0, Lok);
1199 stop("null entry point");
1200 bind(Lok);
1201 }
1202 #endif // ASSERT
1203
1204 mr(R21_sender_SP, R1_SP);
1205
1206 // Calc a precise SP for the call. The SP value we calculated in
1207 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space
1208 // if esp is not max. Also, the i2c adapter extends the stack space without restoring
1209 // our pre-calced value, so repeating calls via i2c would result in stack overflow.
1210 // Since esp already points to an empty slot, we just have to sub 1 additional slot
1211 // to meet the abi scratch requirements.
1212 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in
1213 // the return entry of the interpreter.
1214 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::top_ijava_frame_abi_size);
1215 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address
1216 resize_frame_absolute(Rscratch2, Rscratch2, R0);
1217
1218 mr_if_needed(R19_method, Rtarget_method);
1219 mtctr(Rtarget_addr);
1220 mtlr(Rret_addr);
1221
1222 save_interpreter_state(Rscratch2);
1223 #ifdef ASSERT
1224 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp
1225 cmpd(CCR0, R21_sender_SP, Rscratch1);
1226 asm_assert_eq("top_frame_sp incorrect");
1227 #endif
1228
1229 bctr();
1230 }
1231
1232 // Set the method data pointer for the current bcp.
1233 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1234 assert(ProfileInterpreter, "must be profiling interpreter");
1235 Label get_continue;
1236 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method);
1237 test_method_data_pointer(get_continue);
1238 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp);
1239
1240 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1241 add(R28_mdx, R28_mdx, R3_RET);
1242 bind(get_continue);
1243 }
1244
1245 // Test ImethodDataPtr. If it is null, continue at the specified label.
1246 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1247 assert(ProfileInterpreter, "must be profiling interpreter");
1248 cmpdi(CCR0, R28_mdx, 0);
1249 beq(CCR0, zero_continue);
1250 }
1251
1252 void InterpreterMacroAssembler::verify_method_data_pointer() {
1253 assert(ProfileInterpreter, "must be profiling interpreter");
1254 #ifdef ASSERT
1255 Label verify_continue;
1256 test_method_data_pointer(verify_continue);
1257
1258 // If the mdp is valid, it will point to a DataLayout header which is
1259 // consistent with the bcp. The converse is highly probable also.
1260 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx);
1261 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method);
1262 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1263 add(R11_scratch1, R12_scratch2, R12_scratch2);
1264 cmpd(CCR0, R11_scratch1, R14_bcp);
1265 beq(CCR0, verify_continue);
1266
1267 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx);
1268
1269 bind(verify_continue);
1270 #endif
1271 }
1272
1273 // Store a value at some constant offset from the method data pointer.
1274 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1275 assert(ProfileInterpreter, "must be profiling interpreter");
1276
1277 std(value, constant, R28_mdx);
1278 }
1279
1280 // Increment the value at some constant offset from the method data pointer.
1281 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1282 Register counter_addr,
1283 Register Rbumped_count,
1284 bool decrement) {
1285 // Locate the counter at a fixed offset from the mdp:
1286 addi(counter_addr, R28_mdx, constant);
1287 increment_mdp_data_at(counter_addr, Rbumped_count, decrement);
1288 }
1289
1290 // Increment the value at some non-fixed (reg + constant) offset from
1291 // the method data pointer.
1292 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1293 int constant,
1294 Register scratch,
1295 Register Rbumped_count,
1296 bool decrement) {
1297 // Add the constant to reg to get the offset.
1298 add(scratch, R28_mdx, reg);
1299 // Then calculate the counter address.
1300 addi(scratch, scratch, constant);
1301 increment_mdp_data_at(scratch, Rbumped_count, decrement);
1302 }
1303
1304 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr,
1305 Register Rbumped_count,
1306 bool decrement) {
1307 assert(ProfileInterpreter, "must be profiling interpreter");
1308
1309 // Load the counter.
1310 ld(Rbumped_count, 0, counter_addr);
1311
1312 if (decrement) {
1313 // Decrement the register. Set condition codes.
1314 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment);
1315 // Store the decremented counter, if it is still negative.
1316 std(Rbumped_count, 0, counter_addr);
1317 // Note: add/sub overflow check are not ported, since 64 bit
1318 // calculation should never overflow.
1319 } else {
1320 // Increment the register. Set carry flag.
1321 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment);
1322 // Store the incremented counter.
1323 std(Rbumped_count, 0, counter_addr);
1324 }
1325 }
1326
1327 // Set a flag value at the current method data pointer position.
1328 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1329 Register scratch) {
1330 assert(ProfileInterpreter, "must be profiling interpreter");
1331 // Load the data header.
1332 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1333 // Set the flag.
1334 ori(scratch, scratch, flag_constant);
1335 // Store the modified header.
1336 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1337 }
1338
1339 // Test the location at some offset from the method data pointer.
1340 // If it is not equal to value, branch to the not_equal_continue Label.
1341 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1342 Register value,
1343 Label& not_equal_continue,
1344 Register test_out) {
1345 assert(ProfileInterpreter, "must be profiling interpreter");
1346
1347 ld(test_out, offset, R28_mdx);
1348 cmpd(CCR0, value, test_out);
1349 bne(CCR0, not_equal_continue);
1350 }
1351
1352 // Update the method data pointer by the displacement located at some fixed
1353 // offset from the method data pointer.
1354 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1355 Register scratch) {
1356 assert(ProfileInterpreter, "must be profiling interpreter");
1357
1358 ld(scratch, offset_of_disp, R28_mdx);
1359 add(R28_mdx, scratch, R28_mdx);
1360 }
1361
1362 // Update the method data pointer by the displacement located at the
1363 // offset (reg + offset_of_disp).
1364 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1365 int offset_of_disp,
1366 Register scratch) {
1367 assert(ProfileInterpreter, "must be profiling interpreter");
1368
1369 add(scratch, reg, R28_mdx);
1370 ld(scratch, offset_of_disp, scratch);
1371 add(R28_mdx, scratch, R28_mdx);
1372 }
1373
1374 // Update the method data pointer by a simple constant displacement.
1375 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1376 assert(ProfileInterpreter, "must be profiling interpreter");
1377 addi(R28_mdx, R28_mdx, constant);
1378 }
1379
1380 // Update the method data pointer for a _ret bytecode whose target
1381 // was not among our cached targets.
1382 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1383 Register return_bci) {
1384 assert(ProfileInterpreter, "must be profiling interpreter");
1385
1386 push(state);
1387 assert(return_bci->is_nonvolatile(), "need to protect return_bci");
1388 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1389 pop(state);
1390 }
1391
1392 // Increments the backedge counter.
1393 // Returns backedge counter + invocation counter in Rdst.
1394 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst,
1395 const Register Rtmp1, Register Rscratch) {
1396 assert(UseCompiler, "incrementing must be useful");
1397 assert_different_registers(Rdst, Rtmp1);
1398 const Register invocation_counter = Rtmp1;
1399 const Register counter = Rdst;
1400 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size.");
1401
1402 // Load backedge counter.
1403 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1404 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1405 // Load invocation counter.
1406 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) +
1407 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1408
1409 // Add the delta to the backedge counter.
1410 addi(counter, counter, InvocationCounter::count_increment);
1411
1412 // Mask the invocation counter.
1413 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value);
1414
1415 // Store new counter value.
1416 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1417 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1418 // Return invocation counter + backedge counter.
1419 add(counter, counter, invocation_counter);
1420 }
1421
1422 // Count a taken branch in the bytecodes.
1423 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1424 if (ProfileInterpreter) {
1425 Label profile_continue;
1426
1427 // If no method data exists, go to profile_continue.
1428 test_method_data_pointer(profile_continue);
1429
1430 // We are taking a branch. Increment the taken count.
1431 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count);
1432
1433 // The method data pointer needs to be updated to reflect the new target.
1434 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1435 bind (profile_continue);
1436 }
1437 }
1438
1439 // Count a not-taken branch in the bytecodes.
1440 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) {
1441 if (ProfileInterpreter) {
1442 Label profile_continue;
1443
1444 // If no method data exists, go to profile_continue.
1445 test_method_data_pointer(profile_continue);
1446
1447 // We are taking a branch. Increment the not taken count.
1448 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2);
1449
1450 // The method data pointer needs to be updated to correspond to the
1451 // next bytecode.
1452 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1453 bind (profile_continue);
1454 }
1455 }
1456
1457 // Count a non-virtual call in the bytecodes.
1458 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) {
1459 if (ProfileInterpreter) {
1460 Label profile_continue;
1461
1462 // If no method data exists, go to profile_continue.
1463 test_method_data_pointer(profile_continue);
1464
1465 // We are making a call. Increment the count.
1466 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1467
1468 // The method data pointer needs to be updated to reflect the new target.
1469 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1470 bind (profile_continue);
1471 }
1472 }
1473
1474 // Count a final call in the bytecodes.
1475 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) {
1476 if (ProfileInterpreter) {
1477 Label profile_continue;
1478
1479 // If no method data exists, go to profile_continue.
1480 test_method_data_pointer(profile_continue);
1481
1482 // We are making a call. Increment the count.
1483 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1484
1485 // The method data pointer needs to be updated to reflect the new target.
1486 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1487 bind (profile_continue);
1488 }
1489 }
1490
1491 // Count a virtual call in the bytecodes.
1492 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver,
1493 Register Rscratch1,
1494 Register Rscratch2,
1495 bool receiver_can_be_null) {
1496 if (!ProfileInterpreter) { return; }
1497 Label profile_continue;
1498
1499 // If no method data exists, go to profile_continue.
1500 test_method_data_pointer(profile_continue);
1501
1502 Label skip_receiver_profile;
1503 if (receiver_can_be_null) {
1504 Label not_null;
1505 cmpdi(CCR0, Rreceiver, 0);
1506 bne(CCR0, not_null);
1507 // We are making a call. Increment the count for null receiver.
1508 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2);
1509 b(skip_receiver_profile);
1510 bind(not_null);
1511 }
1512
1513 // Record the receiver type.
1514 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true);
1515 bind(skip_receiver_profile);
1516
1517 // The method data pointer needs to be updated to reflect the new target.
1518 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1519 bind (profile_continue);
1520 }
1521
1522 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) {
1523 if (ProfileInterpreter) {
1524 Label profile_continue;
1525
1526 // If no method data exists, go to profile_continue.
1527 test_method_data_pointer(profile_continue);
1528
1529 int mdp_delta = in_bytes(BitData::bit_data_size());
1530 if (TypeProfileCasts) {
1531 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1532
1533 // Record the object type.
1534 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false);
1535 }
1536
1537 // The method data pointer needs to be updated.
1538 update_mdp_by_constant(mdp_delta);
1539
1540 bind (profile_continue);
1541 }
1542 }
1543
1544 void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) {
1545 if (ProfileInterpreter && TypeProfileCasts) {
1546 Label profile_continue;
1547
1548 // If no method data exists, go to profile_continue.
1549 test_method_data_pointer(profile_continue);
1550
1551 int count_offset = in_bytes(CounterData::count_offset());
1552 // Back up the address, since we have already bumped the mdp.
1553 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1554
1555 // *Decrement* the counter. We expect to see zero or small negatives.
1556 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true);
1557
1558 bind (profile_continue);
1559 }
1560 }
1561
1562 // Count a ret in the bytecodes.
1563 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci,
1564 Register scratch1, Register scratch2) {
1565 if (ProfileInterpreter) {
1566 Label profile_continue;
1567 uint row;
1568
1569 // If no method data exists, go to profile_continue.
1570 test_method_data_pointer(profile_continue);
1571
1572 // Update the total ret count.
1573 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 );
1574
1575 for (row = 0; row < RetData::row_limit(); row++) {
1576 Label next_test;
1577
1578 // See if return_bci is equal to bci[n]:
1579 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1);
1580
1581 // return_bci is equal to bci[n]. Increment the count.
1582 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2);
1583
1584 // The method data pointer needs to be updated to reflect the new target.
1585 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1);
1586 b(profile_continue);
1587 bind(next_test);
1588 }
1589
1590 update_mdp_for_ret(state, return_bci);
1591
1592 bind (profile_continue);
1593 }
1594 }
1595
1596 // Count the default case of a switch construct.
1597 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) {
1598 if (ProfileInterpreter) {
1599 Label profile_continue;
1600
1601 // If no method data exists, go to profile_continue.
1602 test_method_data_pointer(profile_continue);
1603
1604 // Update the default case count
1605 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1606 scratch1, scratch2);
1607
1608 // The method data pointer needs to be updated.
1609 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()),
1610 scratch1);
1611
1612 bind (profile_continue);
1613 }
1614 }
1615
1616 // Count the index'th case of a switch construct.
1617 void InterpreterMacroAssembler::profile_switch_case(Register index,
1618 Register scratch1,
1619 Register scratch2,
1620 Register scratch3) {
1621 if (ProfileInterpreter) {
1622 assert_different_registers(index, scratch1, scratch2, scratch3);
1623 Label profile_continue;
1624
1625 // If no method data exists, go to profile_continue.
1626 test_method_data_pointer(profile_continue);
1627
1628 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes().
1629 li(scratch3, in_bytes(MultiBranchData::case_array_offset()));
1630
1631 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works");
1632 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1633 add(scratch1, scratch1, scratch3);
1634
1635 // Update the case count.
1636 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3);
1637
1638 // The method data pointer needs to be updated.
1639 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2);
1640
1641 bind (profile_continue);
1642 }
1643 }
1644
1645 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) {
1646 if (ProfileInterpreter) {
1647 assert_different_registers(Rscratch1, Rscratch2);
1648 Label profile_continue;
1649
1650 // If no method data exists, go to profile_continue.
1651 test_method_data_pointer(profile_continue);
1652
1653 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1);
1654
1655 // The method data pointer needs to be updated.
1656 int mdp_delta = in_bytes(BitData::bit_data_size());
1657 if (TypeProfileCasts) {
1658 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1659 }
1660 update_mdp_by_constant(mdp_delta);
1661
1662 bind (profile_continue);
1663 }
1664 }
1665
1666 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver,
1667 Register Rscratch1, Register Rscratch2,
1668 bool is_virtual_call) {
1669 assert(ProfileInterpreter, "must be profiling");
1670 assert_different_registers(Rreceiver, Rscratch1, Rscratch2);
1671
1672 Label done;
1673 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call);
1674 bind (done);
1675 }
1676
1677 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1678 Register receiver, Register scratch1, Register scratch2,
1679 int start_row, Label& done, bool is_virtual_call) {
1680 if (TypeProfileWidth == 0) {
1681 if (is_virtual_call) {
1682 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1683 }
1684 return;
1685 }
1686
1687 int last_row = VirtualCallData::row_limit() - 1;
1688 assert(start_row <= last_row, "must be work left to do");
1689 // Test this row for both the receiver and for null.
1690 // Take any of three different outcomes:
1691 // 1. found receiver => increment count and goto done
1692 // 2. found null => keep looking for case 1, maybe allocate this cell
1693 // 3. found something else => keep looking for cases 1 and 2
1694 // Case 3 is handled by a recursive call.
1695 for (int row = start_row; row <= last_row; row++) {
1696 Label next_test;
1697 bool test_for_null_also = (row == start_row);
1698
1699 // See if the receiver is receiver[n].
1700 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1701 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1);
1702 // delayed()->tst(scratch);
1703
1704 // The receiver is receiver[n]. Increment count[n].
1705 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1706 increment_mdp_data_at(count_offset, scratch1, scratch2);
1707 b(done);
1708 bind(next_test);
1709
1710 if (test_for_null_also) {
1711 Label found_null;
1712 // Failed the equality check on receiver[n]... Test for null.
1713 if (start_row == last_row) {
1714 // The only thing left to do is handle the null case.
1715 if (is_virtual_call) {
1716 // Scratch1 contains test_out from test_mdp_data_at.
1717 cmpdi(CCR0, scratch1, 0);
1718 beq(CCR0, found_null);
1719 // Receiver did not match any saved receiver and there is no empty row for it.
1720 // Increment total counter to indicate polymorphic case.
1721 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1722 b(done);
1723 bind(found_null);
1724 } else {
1725 cmpdi(CCR0, scratch1, 0);
1726 bne(CCR0, done);
1727 }
1728 break;
1729 }
1730 // Since null is rare, make it be the branch-taken case.
1731 cmpdi(CCR0, scratch1, 0);
1732 beq(CCR0, found_null);
1733
1734 // Put all the "Case 3" tests here.
1735 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call);
1736
1737 // Found a null. Keep searching for a matching receiver,
1738 // but remember that this is an empty (unused) slot.
1739 bind(found_null);
1740 }
1741 }
1742
1743 // In the fall-through case, we found no matching receiver, but we
1744 // observed the receiver[start_row] is null.
1745
1746 // Fill in the receiver field and increment the count.
1747 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1748 set_mdp_data_at(recvr_offset, receiver);
1749 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1750 li(scratch1, DataLayout::counter_increment);
1751 set_mdp_data_at(count_offset, scratch1);
1752 if (start_row > 0) {
1753 b(done);
1754 }
1755 }
1756
1757 // Argument and return type profilig.
1758 // kills: tmp, tmp2, R0, CR0, CR1
1759 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base,
1760 RegisterOrConstant mdo_addr_offs,
1761 Register tmp, Register tmp2) {
1762 Label do_nothing, do_update;
1763
1764 // tmp2 = obj is allowed
1765 assert_different_registers(obj, mdo_addr_base, tmp, R0);
1766 assert_different_registers(tmp2, mdo_addr_base, tmp, R0);
1767 const Register klass = tmp2;
1768
1769 verify_oop(obj);
1770
1771 ld(tmp, mdo_addr_offs, mdo_addr_base);
1772
1773 // Set null_seen if obj is 0.
1774 cmpdi(CCR0, obj, 0);
1775 ori(R0, tmp, TypeEntries::null_seen);
1776 beq(CCR0, do_update);
1777
1778 load_klass(klass, obj);
1779
1780 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
1781 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
1782 cmpd(CCR1, R0, klass);
1783 // Klass seen before, nothing to do (regardless of unknown bit).
1784 //beq(CCR1, do_nothing);
1785
1786 andi_(R0, tmp, TypeEntries::type_unknown);
1787 // Already unknown. Nothing to do anymore.
1788 //bne(CCR0, do_nothing);
1789 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne
1790 beq(CCR0, do_nothing);
1791
1792 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
1793 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1794 beq(CCR0, do_update); // First time here. Set profile type.
1795
1796 // Different than before. Cannot keep accurate profile.
1797 ori(R0, tmp, TypeEntries::type_unknown);
1798
1799 bind(do_update);
1800 // update profile
1801 std(R0, mdo_addr_offs, mdo_addr_base);
1802
1803 align(32, 12);
1804 bind(do_nothing);
1805 }
1806
1807 void InterpreterMacroAssembler::profile_arguments_type(Register callee,
1808 Register tmp1, Register tmp2,
1809 bool is_virtual) {
1810 if (!ProfileInterpreter) {
1811 return;
1812 }
1813
1814 assert_different_registers(callee, tmp1, tmp2, R28_mdx);
1815
1816 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1817 Label profile_continue;
1818
1819 test_method_data_pointer(profile_continue);
1820
1821 int off_to_start = is_virtual ?
1822 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1823
1824 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx);
1825 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1826 bne(CCR0, profile_continue);
1827
1828 if (MethodData::profile_arguments()) {
1829 Label done;
1830 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1831 addi(R28_mdx, R28_mdx, off_to_args);
1832
1833 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1834 if (i > 0 || MethodData::profile_return()) {
1835 // If return value type is profiled we may have no argument to profile.
1836 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1837 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count());
1838 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count());
1839 blt(CCR0, done);
1840 }
1841 ld(tmp1, in_bytes(Method::const_offset()), callee);
1842 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1);
1843 // Stack offset o (zero based) from the start of the argument
1844 // list, for n arguments translates into offset n - o - 1 from
1845 // the end of the argument list. But there's an extra slot at
1846 // the top of the stack. So the offset is n - o from Lesp.
1847 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx);
1848 subf(tmp1, tmp2, tmp1);
1849
1850 sldi(tmp1, tmp1, Interpreter::logStackElementSize);
1851 ldx(tmp1, tmp1, R15_esp);
1852
1853 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1);
1854
1855 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1856 addi(R28_mdx, R28_mdx, to_add);
1857 off_to_args += to_add;
1858 }
1859
1860 if (MethodData::profile_return()) {
1861 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1862 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1863 }
1864
1865 bind(done);
1866
1867 if (MethodData::profile_return()) {
1868 // We're right after the type profile for the last
1869 // argument. tmp1 is the number of cells left in the
1870 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1871 // if there's a return to profile.
1872 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(),
1873 "can't move past ret type");
1874 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size));
1875 add(R28_mdx, tmp1, R28_mdx);
1876 }
1877 } else {
1878 assert(MethodData::profile_return(), "either profile call args or call ret");
1879 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
1880 }
1881
1882 // Mdp points right after the end of the
1883 // CallTypeData/VirtualCallTypeData, right after the cells for the
1884 // return value type if there's one.
1885 align(32, 12);
1886 bind(profile_continue);
1887 }
1888 }
1889
1890 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
1891 assert_different_registers(ret, tmp1, tmp2);
1892 if (ProfileInterpreter && MethodData::profile_return()) {
1893 Label profile_continue;
1894
1895 test_method_data_pointer(profile_continue);
1896
1897 if (MethodData::profile_return_jsr292_only()) {
1898 // If we don't profile all invoke bytecodes we must make sure
1899 // it's a bytecode we indeed profile. We can't go back to the
1900 // beginning of the ProfileData we intend to update to check its
1901 // type because we're right after it and we don't known its
1902 // length.
1903 lbz(tmp1, 0, R14_bcp);
1904 lbz(tmp2, in_bytes(Method::intrinsic_id_offset()), R19_method);
1905 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic);
1906 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle);
1907 cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
1908 cmpwi(CCR1, tmp2, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1909 cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
1910 bne(CCR0, profile_continue);
1911 }
1912
1913 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2);
1914
1915 align(32, 12);
1916 bind(profile_continue);
1917 }
1918 }
1919
1920 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2,
1921 Register tmp3, Register tmp4) {
1922 if (ProfileInterpreter && MethodData::profile_parameters()) {
1923 Label profile_continue, done;
1924
1925 test_method_data_pointer(profile_continue);
1926
1927 // Load the offset of the area within the MDO used for
1928 // parameters. If it's negative we're not profiling any parameters.
1929 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx);
1930 cmpwi(CCR0, tmp1, 0);
1931 blt(CCR0, profile_continue);
1932
1933 // Compute a pointer to the area for parameters from the offset
1934 // and move the pointer to the slot for the last
1935 // parameters. Collect profiling from last parameter down.
1936 // mdo start + parameters offset + array length - 1
1937
1938 // Pointer to the parameter area in the MDO.
1939 const Register mdp = tmp1;
1940 add(mdp, tmp1, R28_mdx);
1941
1942 // Offset of the current profile entry to update.
1943 const Register entry_offset = tmp2;
1944 // entry_offset = array len in number of cells
1945 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp);
1946
1947 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1948 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
1949
1950 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
1951 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size));
1952 // entry_offset in bytes
1953 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1954
1955 Label loop;
1956 align(32, 12);
1957 bind(loop);
1958
1959 // Load offset on the stack from the slot for this parameter.
1960 ld(tmp3, entry_offset, mdp);
1961 sldi(tmp3, tmp3, Interpreter::logStackElementSize);
1962 neg(tmp3, tmp3);
1963 // Read the parameter from the local area.
1964 ldx(tmp3, tmp3, R18_locals);
1965
1966 // Make entry_offset now point to the type field for this parameter.
1967 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1968 assert(type_base > off_base, "unexpected");
1969 addi(entry_offset, entry_offset, type_base - off_base);
1970
1971 // Profile the parameter.
1972 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3);
1973
1974 // Go to next parameter.
1975 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base);
1976 cmpdi(CCR0, entry_offset, off_base + delta);
1977 addi(entry_offset, entry_offset, -delta);
1978 bge(CCR0, loop);
1979
1980 align(32, 12);
1981 bind(profile_continue);
1982 }
1983 }
1984
1985 // Add a monitor (see frame_ppc.hpp).
1986 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) {
1987
1988 // Very-local scratch registers.
1989 const Register esp = Rtemp1;
1990 const Register slot = Rtemp2;
1991
1992 // Extracted monitor_size.
1993 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes();
1994 assert(Assembler::is_aligned((unsigned int)monitor_size,
1995 (unsigned int)frame::alignment_in_bytes),
1996 "size of a monitor must respect alignment of SP");
1997
1998 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor
1999 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp
2000
2001 // Shuffle expression stack down. Recall that stack_base points
2002 // just above the new expression stack bottom. Old_tos and new_tos
2003 // are used to scan thru the old and new expression stacks.
2004 if (!stack_is_empty) {
2005 Label copy_slot, copy_slot_finished;
2006 const Register n_slots = slot;
2007
2008 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack).
2009 subf(n_slots, esp, R26_monitor);
2010 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy.
2011 assert(LogBytesPerWord == 3, "conflicts assembler instructions");
2012 beq(CCR0, copy_slot_finished); // Nothing to copy.
2013
2014 mtctr(n_slots);
2015
2016 // loop
2017 bind(copy_slot);
2018 ld(slot, 0, esp); // Move expression stack down.
2019 std(slot, -monitor_size, esp); // distance = monitor_size
2020 addi(esp, esp, BytesPerWord);
2021 bdnz(copy_slot);
2022
2023 bind(copy_slot_finished);
2024 }
2025
2026 addi(R15_esp, R15_esp, -monitor_size);
2027 addi(R26_monitor, R26_monitor, -monitor_size);
2028
2029 // Restart interpreter
2030 }
2031
2032 // ============================================================================
2033 // Java locals access
2034
2035 // Load a local variable at index in Rindex into register Rdst_value.
2036 // Also puts address of local into Rdst_address as a service.
2037 // Kills:
2038 // - Rdst_value
2039 // - Rdst_address
2040 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) {
2041 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2042 subf(Rdst_address, Rdst_address, R18_locals);
2043 lwz(Rdst_value, 0, Rdst_address);
2044 }
2045
2046 // Load a local variable at index in Rindex into register Rdst_value.
2047 // Also puts address of local into Rdst_address as a service.
2048 // Kills:
2049 // - Rdst_value
2050 // - Rdst_address
2051 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) {
2052 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2053 subf(Rdst_address, Rdst_address, R18_locals);
2054 ld(Rdst_value, -8, Rdst_address);
2055 }
2056
2057 // Load a local variable at index in Rindex into register Rdst_value.
2058 // Also puts address of local into Rdst_address as a service.
2059 // Input:
2060 // - Rindex: slot nr of local variable
2061 // Kills:
2062 // - Rdst_value
2063 // - Rdst_address
2064 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value,
2065 Register Rdst_address,
2066 Register Rindex) {
2067 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2068 subf(Rdst_address, Rdst_address, R18_locals);
2069 ld(Rdst_value, 0, Rdst_address);
2070 }
2071
2072 // Load a local variable at index in Rindex into register Rdst_value.
2073 // Also puts address of local into Rdst_address as a service.
2074 // Kills:
2075 // - Rdst_value
2076 // - Rdst_address
2077 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value,
2078 Register Rdst_address,
2079 Register Rindex) {
2080 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2081 subf(Rdst_address, Rdst_address, R18_locals);
2082 lfs(Rdst_value, 0, Rdst_address);
2083 }
2084
2085 // Load a local variable at index in Rindex into register Rdst_value.
2086 // Also puts address of local into Rdst_address as a service.
2087 // Kills:
2088 // - Rdst_value
2089 // - Rdst_address
2090 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value,
2091 Register Rdst_address,
2092 Register Rindex) {
2093 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2094 subf(Rdst_address, Rdst_address, R18_locals);
2095 lfd(Rdst_value, -8, Rdst_address);
2096 }
2097
2098 // Store an int value at local variable slot Rindex.
2099 // Kills:
2100 // - Rindex
2101 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) {
2102 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2103 subf(Rindex, Rindex, R18_locals);
2104 stw(Rvalue, 0, Rindex);
2105 }
2106
2107 // Store a long value at local variable slot Rindex.
2108 // Kills:
2109 // - Rindex
2110 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) {
2111 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2112 subf(Rindex, Rindex, R18_locals);
2113 std(Rvalue, -8, Rindex);
2114 }
2115
2116 // Store an oop value at local variable slot Rindex.
2117 // Kills:
2118 // - Rindex
2119 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) {
2120 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2121 subf(Rindex, Rindex, R18_locals);
2122 std(Rvalue, 0, Rindex);
2123 }
2124
2125 // Store an int value at local variable slot Rindex.
2126 // Kills:
2127 // - Rindex
2128 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) {
2129 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2130 subf(Rindex, Rindex, R18_locals);
2131 stfs(Rvalue, 0, Rindex);
2132 }
2133
2134 // Store an int value at local variable slot Rindex.
2135 // Kills:
2136 // - Rindex
2137 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) {
2138 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2139 subf(Rindex, Rindex, R18_locals);
2140 stfd(Rvalue, -8, Rindex);
2141 }
2142
2143 // Read pending exception from thread and jump to interpreter.
2144 // Throw exception entry if one if pending. Fall through otherwise.
2145 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) {
2146 assert_different_registers(Rscratch1, Rscratch2, R3);
2147 Register Rexception = Rscratch1;
2148 Register Rtmp = Rscratch2;
2149 Label Ldone;
2150 // Get pending exception oop.
2151 ld(Rexception, thread_(pending_exception));
2152 cmpdi(CCR0, Rexception, 0);
2153 beq(CCR0, Ldone);
2154 li(Rtmp, 0);
2155 mr_if_needed(R3, Rexception);
2156 std(Rtmp, thread_(pending_exception)); // Clear exception in thread
2157 if (Interpreter::rethrow_exception_entry() != nullptr) {
2158 // Already got entry address.
2159 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry());
2160 } else {
2161 // Dynamically load entry address.
2162 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true);
2163 ld(Rtmp, simm16_rest, Rtmp);
2164 }
2165 mtctr(Rtmp);
2166 save_interpreter_state(Rtmp);
2167 bctr();
2168
2169 align(32, 12);
2170 bind(Ldone);
2171 }
2172
2173 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) {
2174 save_interpreter_state(R11_scratch1);
2175
2176 MacroAssembler::call_VM(oop_result, entry_point, false);
2177
2178 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2179
2180 check_and_handle_popframe(R11_scratch1);
2181 check_and_handle_earlyret(R11_scratch1);
2182 // Now check exceptions manually.
2183 if (check_exceptions) {
2184 check_and_forward_exception(R11_scratch1, R12_scratch2);
2185 }
2186 }
2187
2188 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2189 Register arg_1, bool check_exceptions) {
2190 // ARG1 is reserved for the thread.
2191 mr_if_needed(R4_ARG2, arg_1);
2192 call_VM(oop_result, entry_point, check_exceptions);
2193 }
2194
2195 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2196 Register arg_1, Register arg_2,
2197 bool check_exceptions) {
2198 // ARG1 is reserved for the thread.
2199 mr_if_needed(R4_ARG2, arg_1);
2200 assert(arg_2 != R4_ARG2, "smashed argument");
2201 mr_if_needed(R5_ARG3, arg_2);
2202 call_VM(oop_result, entry_point, check_exceptions);
2203 }
2204
2205 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2206 Register arg_1, Register arg_2, Register arg_3,
2207 bool check_exceptions) {
2208 // ARG1 is reserved for the thread.
2209 mr_if_needed(R4_ARG2, arg_1);
2210 assert(arg_2 != R4_ARG2, "smashed argument");
2211 mr_if_needed(R5_ARG3, arg_2);
2212 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument");
2213 mr_if_needed(R6_ARG4, arg_3);
2214 call_VM(oop_result, entry_point, check_exceptions);
2215 }
2216
2217 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) {
2218 ld(scratch, 0, R1_SP);
2219 std(R15_esp, _ijava_state_neg(esp), scratch);
2220 std(R14_bcp, _ijava_state_neg(bcp), scratch);
2221 std(R26_monitor, _ijava_state_neg(monitors), scratch);
2222 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); }
2223 // Other entries should be unchanged.
2224 }
2225
2226 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only, bool restore_top_frame_sp) {
2227 ld_ptr(scratch, _abi0(callers_sp), R1_SP); // Load frame pointer.
2228 if (restore_top_frame_sp) {
2229 // After thawing the top frame of a continuation we reach here with frame::java_abi.
2230 // therefore we have to restore top_frame_sp before the assertion below.
2231 assert(!bcp_and_mdx_only, "chose other registers");
2232 Register tfsp = R18_locals;
2233 Register scratch2 = R26_monitor;
2234 ld(tfsp, _ijava_state_neg(top_frame_sp), scratch);
2235 resize_frame_absolute(tfsp, scratch2, R0);
2236 }
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 sldi(R18_locals, R18_locals, Interpreter::logStackElementSize);
2247 add(R18_locals, R18_locals, scratch);
2248 ld(R26_monitor, _ijava_state_neg(monitors), scratch);
2249 }
2250 #ifdef ASSERT
2251 {
2252 Label Lok;
2253 subf(R0, R1_SP, scratch);
2254 cmpdi(CCR0, R0, frame::top_ijava_frame_abi_size + frame::ijava_state_size);
2255 bge(CCR0, Lok);
2256 stop("frame too small (restore istate)");
2257 bind(Lok);
2258 }
2259 #endif
2260 }
2261
2262 void InterpreterMacroAssembler::get_method_counters(Register method,
2263 Register Rcounters,
2264 Label& skip) {
2265 BLOCK_COMMENT("Load and ev. allocate counter object {");
2266 Label has_counters;
2267 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2268 cmpdi(CCR0, Rcounters, 0);
2269 bne(CCR0, has_counters);
2270 call_VM(noreg, CAST_FROM_FN_PTR(address,
2271 InterpreterRuntime::build_method_counters), method);
2272 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2273 cmpdi(CCR0, Rcounters, 0);
2274 beq(CCR0, skip); // No MethodCounters, OutOfMemory.
2275 BLOCK_COMMENT("} Load and ev. allocate counter object");
2276
2277 bind(has_counters);
2278 }
2279
2280 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters,
2281 Register iv_be_count,
2282 Register Rtmp_r0) {
2283 assert(UseCompiler, "incrementing must be useful");
2284 Register invocation_count = iv_be_count;
2285 Register backedge_count = Rtmp_r0;
2286 int delta = InvocationCounter::count_increment;
2287
2288 // Load each counter in a register.
2289 // ld(inv_counter, Rtmp);
2290 // ld(be_counter, Rtmp2);
2291 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() +
2292 InvocationCounter::counter_offset());
2293 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() +
2294 InvocationCounter::counter_offset());
2295
2296 BLOCK_COMMENT("Increment profiling counters {");
2297
2298 // Load the backedge counter.
2299 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int
2300 // Mask the backedge counter.
2301 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value);
2302
2303 // Load the invocation counter.
2304 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int
2305 // Add the delta to the invocation counter and store the result.
2306 addi(invocation_count, invocation_count, delta);
2307 // Store value.
2308 stw(invocation_count, inv_counter_offset, Rcounters);
2309
2310 // Add invocation counter + backedge counter.
2311 add(iv_be_count, backedge_count, invocation_count);
2312
2313 // Note that this macro must leave the backedge_count + invocation_count in
2314 // register iv_be_count!
2315 BLOCK_COMMENT("} Increment profiling counters");
2316 }
2317
2318 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2319 if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); }
2320 }
2321
2322 // Local helper function for the verify_oop_or_return_address macro.
2323 static bool verify_return_address(Method* m, int bci) {
2324 #ifndef PRODUCT
2325 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci;
2326 // Assume it is a valid return address if it is inside m and is preceded by a jsr.
2327 if (!m->contains(pc)) return false;
2328 address jsr_pc;
2329 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2330 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2331 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2332 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2333 #endif // PRODUCT
2334 return false;
2335 }
2336
2337 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2338 if (VerifyFPU) {
2339 unimplemented("verfiyFPU");
2340 }
2341 }
2342
2343 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2344 if (!VerifyOops) return;
2345
2346 // The VM documentation for the astore[_wide] bytecode allows
2347 // the TOS to be not only an oop but also a return address.
2348 Label test;
2349 Label skip;
2350 // See if it is an address (in the current method):
2351
2352 const int log2_bytecode_size_limit = 16;
2353 srdi_(Rtmp, reg, log2_bytecode_size_limit);
2354 bne(CCR0, test);
2355
2356 address fd = CAST_FROM_FN_PTR(address, verify_return_address);
2357 const int nbytes_save = MacroAssembler::num_volatile_regs * 8;
2358 save_volatile_gprs(R1_SP, -nbytes_save); // except R0
2359 save_LR_CR(Rtmp); // Save in old frame.
2360 push_frame_reg_args(nbytes_save, Rtmp);
2361
2362 load_const_optimized(Rtmp, fd, R0);
2363 mr_if_needed(R4_ARG2, reg);
2364 mr(R3_ARG1, R19_method);
2365 call_c(Rtmp); // call C
2366
2367 pop_frame();
2368 restore_LR_CR(Rtmp);
2369 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
2370 b(skip);
2371
2372 // Perform a more elaborate out-of-line call.
2373 // Not an address; verify it:
2374 bind(test);
2375 verify_oop(reg);
2376 bind(skip);
2377 }
2378
2379 // Inline assembly for:
2380 //
2381 // if (thread is in interp_only_mode) {
2382 // InterpreterRuntime::post_method_entry();
2383 // }
2384 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) ||
2385 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) {
2386 // SharedRuntime::jvmpi_method_entry(method, receiver);
2387 // }
2388 void InterpreterMacroAssembler::notify_method_entry() {
2389 // JVMTI
2390 // Whenever JVMTI puts a thread in interp_only_mode, method
2391 // entry/exit events are sent for that thread to track stack
2392 // depth. If it is possible to enter interp_only_mode we add
2393 // the code to check if the event should be sent.
2394 if (JvmtiExport::can_post_interpreter_events()) {
2395 Label jvmti_post_done;
2396
2397 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2398 cmpwi(CCR0, R0, 0);
2399 beq(CCR0, jvmti_post_done);
2400 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2401
2402 bind(jvmti_post_done);
2403 }
2404 }
2405
2406 // Inline assembly for:
2407 //
2408 // if (thread is in interp_only_mode) {
2409 // // save result
2410 // InterpreterRuntime::post_method_exit();
2411 // // restore result
2412 // }
2413 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) {
2414 // // save result
2415 // SharedRuntime::jvmpi_method_exit();
2416 // // restore result
2417 // }
2418 //
2419 // Native methods have their result stored in d_tmp and l_tmp.
2420 // Java methods have their result stored in the expression stack.
2421 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state,
2422 NotifyMethodExitMode mode, bool check_exceptions) {
2423 // JVMTI
2424 // Whenever JVMTI puts a thread in interp_only_mode, method
2425 // entry/exit events are sent for that thread to track stack
2426 // depth. If it is possible to enter interp_only_mode we add
2427 // the code to check if the event should be sent.
2428 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2429 Label jvmti_post_done;
2430
2431 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2432 cmpwi(CCR0, R0, 0);
2433 beq(CCR0, jvmti_post_done);
2434 if (!is_native_method) { push(state); } // Expose tos to GC.
2435 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), 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 }