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/methodCounters.hpp"
34 #include "oops/methodData.hpp"
35 #include "oops/resolvedFieldEntry.hpp"
36 #include "oops/resolvedIndyEntry.hpp"
37 #include "oops/resolvedMethodEntry.hpp"
38 #include "prims/jvmtiExport.hpp"
39 #include "prims/jvmtiThreadState.hpp"
40 #include "runtime/frame.inline.hpp"
41 #include "runtime/safepointMechanism.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/vm_version.hpp"
44 #include "utilities/macros.hpp"
45 #include "utilities/powerOfTwo.hpp"
46
47 // Implementation of InterpreterMacroAssembler.
48
49 // This file specializes the assembler with interpreter-specific macros.
50
51 #ifdef PRODUCT
52 #define BLOCK_COMMENT(str) // nothing
53 #else
54 #define BLOCK_COMMENT(str) block_comment(str)
55 #endif
56
57 void InterpreterMacroAssembler::null_check_throw(Register a, int offset, Register temp_reg) {
58 address exception_entry = Interpreter::throw_NullPointerException_entry();
59 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry);
60 }
61
62 void InterpreterMacroAssembler::load_klass_check_null_throw(Register dst, Register src, Register temp_reg) {
63 null_check_throw(src, oopDesc::klass_offset_in_bytes(), temp_reg);
64 load_klass(dst, src);
65 }
66
67 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) {
68 assert(entry, "Entry must have been generated by now");
69 if (is_within_range_of_b(entry, pc())) {
70 b(entry);
71 } else {
72 load_const_optimized(Rscratch, entry, R0);
73 mtctr(Rscratch);
74 bctr();
75 }
76 }
77
78 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) {
79 Register bytecode = R12_scratch2;
80 if (bcp_incr != 0) {
81 lbzu(bytecode, bcp_incr, R14_bcp);
82 } else {
83 lbz(bytecode, 0, R14_bcp);
84 }
85
86 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state), generate_poll);
87 }
88
89 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
90 // Load current bytecode.
91 Register bytecode = R12_scratch2;
92 lbz(bytecode, 0, R14_bcp);
93 dispatch_Lbyte_code(state, bytecode, table);
94 }
95
96 // Dispatch code executed in the prolog of a bytecode which does not do it's
97 // own dispatch. The dispatch address is computed and placed in R24_dispatch_addr.
98 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
99 Register bytecode = R12_scratch2;
100 lbz(bytecode, bcp_incr, R14_bcp);
101
102 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state));
103
104 sldi(bytecode, bytecode, LogBytesPerWord);
105 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode);
106 }
107
108 // Dispatch code executed in the epilog of a bytecode which does not do it's
109 // own dispatch. The dispatch address in R24_dispatch_addr is used for the
110 // dispatch.
111 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
112 if (bcp_incr) { addi(R14_bcp, R14_bcp, bcp_incr); }
113 mtctr(R24_dispatch_addr);
114 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
115 }
116
117 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
118 assert(scratch_reg != R0, "can't use R0 as scratch_reg here");
119 if (JvmtiExport::can_pop_frame()) {
120 Label L;
121
122 // Check the "pending popframe condition" flag in the current thread.
123 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
124
125 // Initiate popframe handling only if it is not already being
126 // processed. If the flag has the popframe_processing bit set, it
127 // means that this code is called *during* popframe handling - we
128 // don't want to reenter.
129 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit);
130 beq(CCR0, L);
131
132 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit);
133 bne(CCR0, L);
134
135 // Call the Interpreter::remove_activation_preserving_args_entry()
136 // func to get the address of the same-named entrypoint in the
137 // generated interpreter code.
138 #if defined(ABI_ELFv2)
139 call_c(CAST_FROM_FN_PTR(address,
140 Interpreter::remove_activation_preserving_args_entry),
141 relocInfo::none);
142 #else
143 call_c(CAST_FROM_FN_PTR(FunctionDescriptor*,
144 Interpreter::remove_activation_preserving_args_entry),
145 relocInfo::none);
146 #endif
147
148 // Jump to Interpreter::_remove_activation_preserving_args_entry.
149 mtctr(R3_RET);
150 bctr();
151
152 align(32, 12);
153 bind(L);
154 }
155 }
156
157 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
158 const Register Rthr_state_addr = scratch_reg;
159 if (JvmtiExport::can_force_early_return()) {
160 Label Lno_early_ret;
161 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
162 cmpdi(CCR0, Rthr_state_addr, 0);
163 beq(CCR0, Lno_early_ret);
164
165 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr);
166 cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending);
167 bne(CCR0, Lno_early_ret);
168
169 // Jump to Interpreter::_earlyret_entry.
170 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr);
171 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry));
172 mtlr(R3_RET);
173 blr();
174
175 align(32, 12);
176 bind(Lno_early_ret);
177 }
178 }
179
180 void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) {
181 const Register RjvmtiState = Rscratch1;
182 const Register Rscratch2 = R0;
183
184 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
185 li(Rscratch2, 0);
186
187 switch (state) {
188 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
189 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
190 break;
191 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
192 break;
193 case btos: // fall through
194 case ztos: // fall through
195 case ctos: // fall through
196 case stos: // fall through
197 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
198 break;
199 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
200 break;
201 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
202 break;
203 case vtos: break;
204 default : ShouldNotReachHere();
205 }
206
207 // Clean up tos value in the jvmti thread state.
208 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
209 // Set tos state field to illegal value.
210 li(Rscratch2, ilgl);
211 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState);
212 }
213
214 // Common code to dispatch and dispatch_only.
215 // Dispatch value in Lbyte_code and increment Lbcp.
216
217 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) {
218 address table_base = (address)Interpreter::dispatch_table((TosState)0);
219 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base;
220 if (is_simm16(table_offs)) {
221 addi(dst, R25_templateTableBase, (int)table_offs);
222 } else {
223 load_const_optimized(dst, table, R0);
224 }
225 }
226
227 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode,
228 address* table, bool generate_poll) {
229 assert_different_registers(bytecode, R11_scratch1);
230
231 // Calc dispatch table address.
232 load_dispatch_table(R11_scratch1, table);
233
234 if (generate_poll) {
235 address *sfpt_tbl = Interpreter::safept_table(state);
236 if (table != sfpt_tbl) {
237 Label dispatch;
238 ld(R0, in_bytes(JavaThread::polling_word_offset()), R16_thread);
239 // Armed page has poll_bit set, if poll bit is cleared just continue.
240 andi_(R0, R0, SafepointMechanism::poll_bit());
241 beq(CCR0, dispatch);
242 load_dispatch_table(R11_scratch1, sfpt_tbl);
243 align(32, 16);
244 bind(dispatch);
245 }
246 }
247
248 sldi(R12_scratch2, bytecode, LogBytesPerWord);
249 ldx(R11_scratch1, R11_scratch1, R12_scratch2);
250
251 // Jump off!
252 mtctr(R11_scratch1);
253 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
254 }
255
256 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) {
257 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize);
258 ldx(Rrecv_dst, Rrecv_dst, R15_esp);
259 }
260
261 // helpers for expression stack
262
263 void InterpreterMacroAssembler::pop_i(Register r) {
264 lwzu(r, Interpreter::stackElementSize, R15_esp);
265 }
266
267 void InterpreterMacroAssembler::pop_ptr(Register r) {
268 ldu(r, Interpreter::stackElementSize, R15_esp);
269 }
270
271 void InterpreterMacroAssembler::pop_l(Register r) {
272 ld(r, Interpreter::stackElementSize, R15_esp);
273 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
274 }
275
276 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
277 lfsu(f, Interpreter::stackElementSize, R15_esp);
278 }
279
280 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
281 lfd(f, Interpreter::stackElementSize, R15_esp);
282 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
283 }
284
285 void InterpreterMacroAssembler::push_i(Register r) {
286 stw(r, 0, R15_esp);
287 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
288 }
289
290 void InterpreterMacroAssembler::push_ptr(Register r) {
291 std(r, 0, R15_esp);
292 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
293 }
294
295 void InterpreterMacroAssembler::push_l(Register r) {
296 // Clear unused slot.
297 load_const_optimized(R0, 0L);
298 std(R0, 0, R15_esp);
299 std(r, - Interpreter::stackElementSize, R15_esp);
300 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
301 }
302
303 void InterpreterMacroAssembler::push_f(FloatRegister f) {
304 stfs(f, 0, R15_esp);
305 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
306 }
307
308 void InterpreterMacroAssembler::push_d(FloatRegister f) {
309 stfd(f, - Interpreter::stackElementSize, R15_esp);
310 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
311 }
312
313 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) {
314 std(first, 0, R15_esp);
315 std(second, -Interpreter::stackElementSize, R15_esp);
316 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
317 }
318
319 void InterpreterMacroAssembler::move_l_to_d(Register l, FloatRegister d) {
320 if (VM_Version::has_mtfprd()) {
321 mtfprd(d, l);
322 } else {
323 std(l, 0, R15_esp);
324 lfd(d, 0, R15_esp);
325 }
326 }
327
328 void InterpreterMacroAssembler::move_d_to_l(FloatRegister d, Register l) {
329 if (VM_Version::has_mtfprd()) {
330 mffprd(l, d);
331 } else {
332 stfd(d, 0, R15_esp);
333 ld(l, 0, R15_esp);
334 }
335 }
336
337 void InterpreterMacroAssembler::push(TosState state) {
338 switch (state) {
339 case atos: push_ptr(); break;
340 case btos:
341 case ztos:
342 case ctos:
343 case stos:
344 case itos: push_i(); break;
345 case ltos: push_l(); break;
346 case ftos: push_f(); break;
347 case dtos: push_d(); break;
348 case vtos: /* nothing to do */ break;
349 default : ShouldNotReachHere();
350 }
351 }
352
353 void InterpreterMacroAssembler::pop(TosState state) {
354 switch (state) {
355 case atos: pop_ptr(); break;
356 case btos:
357 case ztos:
358 case ctos:
359 case stos:
360 case itos: pop_i(); break;
361 case ltos: pop_l(); break;
362 case ftos: pop_f(); break;
363 case dtos: pop_d(); break;
364 case vtos: /* nothing to do */ break;
365 default : ShouldNotReachHere();
366 }
367 verify_oop(R17_tos, state);
368 }
369
370 void InterpreterMacroAssembler::empty_expression_stack() {
371 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
372 }
373
374 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset,
375 Register Rdst,
376 signedOrNot is_signed) {
377 #if defined(VM_LITTLE_ENDIAN)
378 if (bcp_offset) {
379 load_const_optimized(Rdst, bcp_offset);
380 lhbrx(Rdst, R14_bcp, Rdst);
381 } else {
382 lhbrx(Rdst, R14_bcp);
383 }
384 if (is_signed == Signed) {
385 extsh(Rdst, Rdst);
386 }
387 #else
388 // Read Java big endian format.
389 if (is_signed == Signed) {
390 lha(Rdst, bcp_offset, R14_bcp);
391 } else {
392 lhz(Rdst, bcp_offset, R14_bcp);
393 }
394 #endif
395 }
396
397 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset,
398 Register Rdst,
399 signedOrNot is_signed) {
400 #if defined(VM_LITTLE_ENDIAN)
401 if (bcp_offset) {
402 load_const_optimized(Rdst, bcp_offset);
403 lwbrx(Rdst, R14_bcp, Rdst);
404 } else {
405 lwbrx(Rdst, R14_bcp);
406 }
407 if (is_signed == Signed) {
408 extsw(Rdst, Rdst);
409 }
410 #else
411 // Read Java big endian format.
412 if (bcp_offset & 3) { // Offset unaligned?
413 load_const_optimized(Rdst, bcp_offset);
414 if (is_signed == Signed) {
415 lwax(Rdst, R14_bcp, Rdst);
416 } else {
417 lwzx(Rdst, R14_bcp, Rdst);
418 }
419 } else {
420 if (is_signed == Signed) {
421 lwa(Rdst, bcp_offset, R14_bcp);
422 } else {
423 lwz(Rdst, bcp_offset, R14_bcp);
424 }
425 }
426 #endif
427 }
428
429
430 // Load the constant pool cache index from the bytecode stream.
431 //
432 // Kills / writes:
433 // - Rdst, Rscratch
434 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset,
435 size_t index_size) {
436 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
437 // Cache index is always in the native format, courtesy of Rewriter.
438 if (index_size == sizeof(u2)) {
439 lhz(Rdst, bcp_offset, R14_bcp);
440 } else if (index_size == sizeof(u4)) {
441 if (bcp_offset & 3) {
442 load_const_optimized(Rdst, bcp_offset);
443 lwax(Rdst, R14_bcp, Rdst);
444 } else {
445 lwa(Rdst, bcp_offset, R14_bcp);
446 }
447 } else if (index_size == sizeof(u1)) {
448 lbz(Rdst, bcp_offset, R14_bcp);
449 } else {
450 ShouldNotReachHere();
451 }
452 // Rdst now contains cp cache index.
453 }
454
455 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format)
456 // from (Rsrc)+offset.
457 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset,
458 signedOrNot is_signed) {
459 #if defined(VM_LITTLE_ENDIAN)
460 if (offset) {
461 load_const_optimized(Rdst, offset);
462 lwbrx(Rdst, Rdst, Rsrc);
463 } else {
464 lwbrx(Rdst, Rsrc);
465 }
466 if (is_signed == Signed) {
467 extsw(Rdst, Rdst);
468 }
469 #else
470 if (is_signed == Signed) {
471 lwa(Rdst, offset, Rsrc);
472 } else {
473 lwz(Rdst, offset, Rsrc);
474 }
475 #endif
476 }
477
478 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
479 // Get index out of bytecode pointer
480 get_cache_index_at_bcp(index, 1, sizeof(u4));
481
482 // Get address of invokedynamic array
483 ld_ptr(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset()), R27_constPoolCache);
484 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
485 sldi(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
486 addi(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
487 add(cache, cache, index);
488 }
489
490 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
491 // Get index out of bytecode pointer
492 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
493 // Take shortcut if the size is a power of 2
494 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
495 // Scale index by power of 2
496 sldi(index, index, log2i_exact(sizeof(ResolvedFieldEntry)));
497 } else {
498 // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
499 mulli(index, index, sizeof(ResolvedFieldEntry));
500 }
501 // Get address of field entries array
502 ld_ptr(cache, in_bytes(ConstantPoolCache::field_entries_offset()), R27_constPoolCache);
503 addi(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
504 add(cache, cache, index);
505 }
506
507 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
508 // Get index out of bytecode pointer
509 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
510 // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
511 mulli(index, index, sizeof(ResolvedMethodEntry));
512
513 // Get address of field entries array
514 ld_ptr(cache, ConstantPoolCache::method_entries_offset(), R27_constPoolCache);
515 addi(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
516 add(cache, cache, index); // method_entries + base_offset + scaled index
517 }
518
519 // Load object from cpool->resolved_references(index).
520 // Kills:
521 // - index
522 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index,
523 Register tmp1, Register tmp2,
524 Label *L_handle_null) {
525 assert_different_registers(result, index, tmp1, tmp2);
526 assert(index->is_nonvolatile(), "needs to survive C-call in resolve_oop_handle");
527 get_constant_pool(result);
528
529 // Convert from field index to resolved_references() index and from
530 // word index to byte offset. Since this is a java object, it can be compressed.
531 sldi(index, index, LogBytesPerHeapOop);
532 // Load pointer for resolved_references[] objArray.
533 ld(result, ConstantPool::cache_offset(), result);
534 ld(result, ConstantPoolCache::resolved_references_offset(), result);
535 resolve_oop_handle(result, tmp1, tmp2, MacroAssembler::PRESERVATION_NONE);
536 #ifdef ASSERT
537 Label index_ok;
538 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result);
539 sldi(R0, R0, LogBytesPerHeapOop);
540 cmpd(CCR0, index, R0);
541 blt(CCR0, index_ok);
542 stop("resolved reference index out of bounds");
543 bind(index_ok);
544 #endif
545 // Add in the index.
546 add(result, index, result);
547 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result,
548 tmp1, tmp2,
549 MacroAssembler::PRESERVATION_NONE,
550 0, L_handle_null);
551 }
552
553 // load cpool->resolved_klass_at(index)
554 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool, Register Roffset, Register Rklass) {
555 // int value = *(Rcpool->int_at_addr(which));
556 // int resolved_klass_index = extract_low_short_from_int(value);
557 add(Roffset, Rcpool, Roffset);
558 #if defined(VM_LITTLE_ENDIAN)
559 lhz(Roffset, sizeof(ConstantPool), Roffset); // Roffset = resolved_klass_index
560 #else
561 lhz(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index
562 #endif
563
564 ld(Rklass, ConstantPool::resolved_klasses_offset(), Rcpool); // Rklass = Rcpool->_resolved_klasses
565
566 sldi(Roffset, Roffset, LogBytesPerWord);
567 addi(Roffset, Roffset, Array<Klass*>::base_offset_in_bytes());
568 isync(); // Order load of instance Klass wrt. tags.
569 ldx(Rklass, Rklass, Roffset);
570 }
571
572 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
573 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2.
574 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1,
575 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) {
576 // Profile the not-null value's klass.
577 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2);
578 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
579 }
580
581 // Separate these two to allow for delay slot in middle.
582 // These are used to do a test and full jump to exception-throwing code.
583
584 // Check that index is in range for array, then shift index by index_shift,
585 // and put arrayOop + shifted_index into res.
586 // Note: res is still shy of address by array offset into object.
587
588 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex,
589 int index_shift, Register Rtmp, Register Rres) {
590 // Check that index is in range for array, then shift index by index_shift,
591 // and put arrayOop + shifted_index into res.
592 // Note: res is still shy of address by array offset into object.
593 // Kills:
594 // - Rindex
595 // Writes:
596 // - Rres: Address that corresponds to the array index if check was successful.
597 verify_oop(Rarray);
598 const Register Rlength = R0;
599 const Register RsxtIndex = Rtmp;
600 Label LisNull, LnotOOR;
601
602 // Array nullcheck
603 if (!ImplicitNullChecks) {
604 cmpdi(CCR0, Rarray, 0);
605 beq(CCR0, LisNull);
606 } else {
607 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex);
608 }
609
610 // Rindex might contain garbage in upper bits (remember that we don't sign extend
611 // during integer arithmetic operations). So kill them and put value into same register
612 // where ArrayIndexOutOfBounds would expect the index in.
613 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit
614
615 // Index check
616 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray);
617 cmplw(CCR0, Rindex, Rlength);
618 sldi(RsxtIndex, RsxtIndex, index_shift);
619 blt(CCR0, LnotOOR);
620 // Index should be in R17_tos, array should be in R4_ARG2.
621 mr_if_needed(R17_tos, Rindex);
622 mr_if_needed(R4_ARG2, Rarray);
623 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
624 mtctr(Rtmp);
625 bctr();
626
627 if (!ImplicitNullChecks) {
628 bind(LisNull);
629 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry);
630 mtctr(Rtmp);
631 bctr();
632 }
633
634 align(32, 16);
635 bind(LnotOOR);
636
637 // Calc address
638 add(Rres, RsxtIndex, Rarray);
639 }
640
641 void InterpreterMacroAssembler::index_check(Register array, Register index,
642 int index_shift, Register tmp, Register res) {
643 // pop array
644 pop_ptr(array);
645
646 // check array
647 index_check_without_pop(array, index, index_shift, tmp, res);
648 }
649
650 void InterpreterMacroAssembler::get_const(Register Rdst) {
651 ld(Rdst, in_bytes(Method::const_offset()), R19_method);
652 }
653
654 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
655 get_const(Rdst);
656 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
657 }
658
659 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
660 get_constant_pool(Rdst);
661 ld(Rdst, ConstantPool::cache_offset(), Rdst);
662 }
663
664 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
665 get_constant_pool(Rcpool);
666 ld(Rtags, ConstantPool::tags_offset(), Rcpool);
667 }
668
669 // Unlock if synchronized method.
670 //
671 // Unlock the receiver if this is a synchronized method.
672 // Unlock any Java monitors from synchronized blocks.
673 //
674 // If there are locked Java monitors
675 // If throw_monitor_exception
676 // throws IllegalMonitorStateException
677 // Else if install_monitor_exception
678 // installs IllegalMonitorStateException
679 // Else
680 // no error processing
681 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
682 bool throw_monitor_exception,
683 bool install_monitor_exception) {
684 Label Lunlocked, Lno_unlock;
685 {
686 Register Rdo_not_unlock_flag = R11_scratch1;
687 Register Raccess_flags = R12_scratch2;
688
689 // Check if synchronized method or unlocking prevented by
690 // JavaThread::do_not_unlock_if_synchronized flag.
691 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
692 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method);
693 li(R0, 0);
694 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag
695
696 push(state);
697
698 // Skip if we don't have to unlock.
699 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0.
700 beq(CCR0, Lunlocked);
701
702 cmpwi(CCR0, Rdo_not_unlock_flag, 0);
703 bne(CCR0, Lno_unlock);
704 }
705
706 // Unlock
707 {
708 Register Rmonitor_base = R11_scratch1;
709
710 Label Lunlock;
711 // If it's still locked, everything is ok, unlock it.
712 ld(Rmonitor_base, 0, R1_SP);
713 addi(Rmonitor_base, Rmonitor_base,
714 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
715
716 ld(R0, BasicObjectLock::obj_offset(), Rmonitor_base);
717 cmpdi(CCR0, R0, 0);
718 bne(CCR0, Lunlock);
719
720 // If it's already unlocked, throw exception.
721 if (throw_monitor_exception) {
722 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
723 should_not_reach_here();
724 } else {
725 if (install_monitor_exception) {
726 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
727 b(Lunlocked);
728 }
729 }
730
731 bind(Lunlock);
732 unlock_object(Rmonitor_base);
733 }
734
735 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not.
736 bind(Lunlocked);
737 {
738 Label Lexception, Lrestart;
739 Register Rcurrent_obj_addr = R11_scratch1;
740 const int delta = frame::interpreter_frame_monitor_size_in_bytes();
741 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords");
742
743 bind(Lrestart);
744 // Set up search loop: Calc num of iterations.
745 {
746 Register Riterations = R12_scratch2;
747 Register Rmonitor_base = Rcurrent_obj_addr;
748 ld(Rmonitor_base, 0, R1_SP);
749 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base
750
751 subf_(Riterations, R26_monitor, Rmonitor_base);
752 ble(CCR0, Lno_unlock);
753
754 addi(Rcurrent_obj_addr, Rmonitor_base,
755 in_bytes(BasicObjectLock::obj_offset()) - frame::interpreter_frame_monitor_size_in_bytes());
756 // Check if any monitor is on stack, bail out if not
757 srdi(Riterations, Riterations, exact_log2(delta));
758 mtctr(Riterations);
759 }
760
761 // The search loop: Look for locked monitors.
762 {
763 const Register Rcurrent_obj = R0;
764 Label Lloop;
765
766 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
767 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
768 bind(Lloop);
769
770 // Check if current entry is used.
771 cmpdi(CCR0, Rcurrent_obj, 0);
772 bne(CCR0, Lexception);
773 // Preload next iteration's compare value.
774 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
775 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
776 bdnz(Lloop);
777 }
778 // Fell through: Everything's unlocked => finish.
779 b(Lno_unlock);
780
781 // An object is still locked => need to throw exception.
782 bind(Lexception);
783 if (throw_monitor_exception) {
784 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
785 should_not_reach_here();
786 } else {
787 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
788 // Unlock does not block, so don't have to worry about the frame.
789 Register Rmonitor_addr = R11_scratch1;
790 addi(Rmonitor_addr, Rcurrent_obj_addr, -in_bytes(BasicObjectLock::obj_offset()) + delta);
791 unlock_object(Rmonitor_addr);
792 if (install_monitor_exception) {
793 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
794 }
795 b(Lrestart);
796 }
797 }
798
799 align(32, 12);
800 bind(Lno_unlock);
801 pop(state);
802 }
803
804 // Support function for remove_activation & Co.
805 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc,
806 Register Rscratch1, Register Rscratch2) {
807 // Pop interpreter frame.
808 ld(Rscratch1, 0, R1_SP); // *SP
809 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp
810 ld(Rscratch2, 0, Rscratch1); // **SP
811 if (return_pc!=noreg) {
812 ld(return_pc, _abi0(lr), Rscratch1); // LR
813 }
814
815 // Merge top frames.
816 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP
817 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP
818 }
819
820 void InterpreterMacroAssembler::narrow(Register result) {
821 Register ret_type = R11_scratch1;
822 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
823 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1);
824
825 Label notBool, notByte, notChar, done;
826
827 // common case first
828 cmpwi(CCR0, ret_type, T_INT);
829 beq(CCR0, done);
830
831 cmpwi(CCR0, ret_type, T_BOOLEAN);
832 bne(CCR0, notBool);
833 andi(result, result, 0x1);
834 b(done);
835
836 bind(notBool);
837 cmpwi(CCR0, ret_type, T_BYTE);
838 bne(CCR0, notByte);
839 extsb(result, result);
840 b(done);
841
842 bind(notByte);
843 cmpwi(CCR0, ret_type, T_CHAR);
844 bne(CCR0, notChar);
845 andi(result, result, 0xffff);
846 b(done);
847
848 bind(notChar);
849 // cmpwi(CCR0, ret_type, T_SHORT); // all that's left
850 // bne(CCR0, done);
851 extsh(result, result);
852
853 // Nothing to do for T_INT
854 bind(done);
855 }
856
857 // Remove activation.
858 //
859 // Apply stack watermark barrier.
860 // Unlock the receiver if this is a synchronized method.
861 // Unlock any Java monitors from synchronized blocks.
862 // Remove the activation from the stack.
863 //
864 // If there are locked Java monitors
865 // If throw_monitor_exception
866 // throws IllegalMonitorStateException
867 // Else if install_monitor_exception
868 // installs IllegalMonitorStateException
869 // Else
870 // no error processing
871 void InterpreterMacroAssembler::remove_activation(TosState state,
872 bool throw_monitor_exception,
873 bool install_monitor_exception) {
874 BLOCK_COMMENT("remove_activation {");
875
876 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
877 // that would normally not be safe to use. Such bad returns into unsafe territory of
878 // the stack, will call InterpreterRuntime::at_unwind.
879 Label slow_path;
880 Label fast_path;
881 safepoint_poll(slow_path, R11_scratch1, true /* at_return */, false /* in_nmethod */);
882 b(fast_path);
883 bind(slow_path);
884 push(state);
885 set_last_Java_frame(R1_SP, noreg);
886 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), R16_thread);
887 reset_last_Java_frame();
888 pop(state);
889 align(32);
890 bind(fast_path);
891
892 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
893
894 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
895 notify_method_exit(false, state, NotifyJVMTI, true);
896
897 BLOCK_COMMENT("reserved_stack_check:");
898 if (StackReservedPages > 0) {
899 // Test if reserved zone needs to be enabled.
900 Label no_reserved_zone_enabling;
901
902 // check if already enabled - if so no re-enabling needed
903 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
904 lwz(R0, in_bytes(JavaThread::stack_guard_state_offset()), R16_thread);
905 cmpwi(CCR0, R0, StackOverflow::stack_guard_enabled);
906 beq_predict_taken(CCR0, no_reserved_zone_enabling);
907
908 // Compare frame pointers. There is no good stack pointer, as with stack
909 // frame compression we can get different SPs when we do calls. A subsequent
910 // call could have a smaller SP, so that this compare succeeds for an
911 // inner call of the method annotated with ReservedStack.
912 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread);
913 ld_ptr(R11_scratch1, _abi0(callers_sp), R1_SP); // Load frame pointer.
914 cmpld(CCR0, R11_scratch1, R0);
915 blt_predict_taken(CCR0, no_reserved_zone_enabling);
916
917 // Enable reserved zone again, throw stack overflow exception.
918 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread);
919 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
920
921 should_not_reach_here();
922
923 bind(no_reserved_zone_enabling);
924 }
925
926 verify_oop(R17_tos, state);
927
928 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
929 mtlr(R0);
930 pop_cont_fastpath();
931 BLOCK_COMMENT("} remove_activation");
932 }
933
934 // Lock object
935 //
936 // Registers alive
937 // monitor - Address of the BasicObjectLock to be used for locking,
938 // which must be initialized with the object to lock.
939 // object - Address of the object to be locked.
940 //
941 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
942 if (LockingMode == LM_MONITOR) {
943 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
944 } else {
945 // template code (for LM_LEGACY):
946 //
947 // markWord displaced_header = obj->mark().set_unlocked();
948 // monitor->lock()->set_displaced_header(displaced_header);
949 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
950 // // We stored the monitor address into the object's mark word.
951 // } else if (THREAD->is_lock_owned((address)displaced_header))
952 // // Simple recursive case.
953 // monitor->lock()->set_displaced_header(nullptr);
954 // } else {
955 // // Slow path.
956 // InterpreterRuntime::monitorenter(THREAD, monitor);
957 // }
958
959 const Register header = R7_ARG5;
960 const Register object_mark_addr = R8_ARG6;
961 const Register current_header = R9_ARG7;
962 const Register tmp = R10_ARG8;
963
964 Label count_locking, done;
965 Label cas_failed, slow_case;
966
967 assert_different_registers(header, object_mark_addr, current_header, tmp);
968
969 // markWord displaced_header = obj->mark().set_unlocked();
970
971 if (DiagnoseSyncOnValueBasedClasses != 0) {
972 load_klass(tmp, object);
973 lwz(tmp, in_bytes(Klass::access_flags_offset()), tmp);
974 testbitdi(CCR0, R0, tmp, exact_log2(JVM_ACC_IS_VALUE_BASED_CLASS));
975 bne(CCR0, slow_case);
976 }
977
978 if (LockingMode == LM_LIGHTWEIGHT) {
979 lightweight_lock(object, header, tmp, slow_case);
980 b(count_locking);
981 } else if (LockingMode == LM_LEGACY) {
982 // Load markWord from object into header.
983 ld(header, oopDesc::mark_offset_in_bytes(), object);
984
985 // Set displaced_header to be (markWord of object | UNLOCK_VALUE).
986 ori(header, header, markWord::unlocked_value);
987
988 // monitor->lock()->set_displaced_header(displaced_header);
989 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
990 const int mark_offset = lock_offset +
991 BasicLock::displaced_header_offset_in_bytes();
992
993 // Initialize the box (Must happen before we update the object mark!).
994 std(header, mark_offset, monitor);
995
996 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
997
998 // Store stack address of the BasicObjectLock (this is monitor) into object.
999 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
1000
1001 // Must fence, otherwise, preceding store(s) may float below cmpxchg.
1002 // CmpxchgX sets CCR0 to cmpX(current, displaced).
1003 cmpxchgd(/*flag=*/CCR0,
1004 /*current_value=*/current_header,
1005 /*compare_value=*/header, /*exchange_value=*/monitor,
1006 /*where=*/object_mark_addr,
1007 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq,
1008 MacroAssembler::cmpxchgx_hint_acquire_lock(),
1009 noreg,
1010 &cas_failed,
1011 /*check without membar and ldarx first*/true);
1012
1013 // If the compare-and-exchange succeeded, then we found an unlocked
1014 // object and we have now locked it.
1015 b(count_locking);
1016 bind(cas_failed);
1017
1018 // } else if (THREAD->is_lock_owned((address)displaced_header))
1019 // // Simple recursive case.
1020 // monitor->lock()->set_displaced_header(nullptr);
1021
1022 // We did not see an unlocked object so try the fast recursive case.
1023
1024 // Check if owner is self by comparing the value in the markWord of object
1025 // (current_header) with the stack pointer.
1026 sub(current_header, current_header, R1_SP);
1027
1028 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1029 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markWord::lock_mask_in_place);
1030
1031 and_(R0/*==0?*/, current_header, tmp);
1032 // If condition is true we are done and hence we can store 0 in the displaced
1033 // header indicating it is a recursive lock.
1034 bne(CCR0, slow_case);
1035 std(R0/*==0!*/, mark_offset, monitor);
1036 b(count_locking);
1037 }
1038
1039 // } else {
1040 // // Slow path.
1041 // InterpreterRuntime::monitorenter(THREAD, monitor);
1042
1043 // None of the above fast optimizations worked so we have to get into the
1044 // slow case of monitor enter.
1045 bind(slow_case);
1046 if (LockingMode == LM_LIGHTWEIGHT) {
1047 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj), object);
1048 } else {
1049 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
1050 }
1051 b(done);
1052 // }
1053 align(32, 12);
1054 bind(count_locking);
1055 inc_held_monitor_count(current_header /*tmp*/);
1056 bind(done);
1057 }
1058 }
1059
1060 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1061 //
1062 // Registers alive
1063 // monitor - Address of the BasicObjectLock to be used for locking,
1064 // which must be initialized with the object to lock.
1065 //
1066 // Throw IllegalMonitorException if object is not locked by current thread.
1067 void InterpreterMacroAssembler::unlock_object(Register monitor) {
1068 if (LockingMode == LM_MONITOR) {
1069 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1070 } else {
1071
1072 // template code (for LM_LEGACY):
1073 //
1074 // if ((displaced_header = monitor->displaced_header()) == nullptr) {
1075 // // Recursive unlock. Mark the monitor unlocked by setting the object field to null.
1076 // monitor->set_obj(nullptr);
1077 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1078 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1079 // monitor->set_obj(nullptr);
1080 // } else {
1081 // // Slow path.
1082 // InterpreterRuntime::monitorexit(monitor);
1083 // }
1084
1085 const Register object = R7_ARG5;
1086 const Register header = R8_ARG6;
1087 const Register object_mark_addr = R9_ARG7;
1088 const Register current_header = R10_ARG8;
1089
1090 Label free_slot;
1091 Label slow_case;
1092
1093 assert_different_registers(object, header, object_mark_addr, current_header);
1094
1095 if (LockingMode != LM_LIGHTWEIGHT) {
1096 // Test first if we are in the fast recursive case.
1097 ld(header, in_bytes(BasicObjectLock::lock_offset()) +
1098 BasicLock::displaced_header_offset_in_bytes(), monitor);
1099
1100 // If the displaced header is zero, we have a recursive unlock.
1101 cmpdi(CCR0, header, 0);
1102 beq(CCR0, free_slot); // recursive unlock
1103 }
1104
1105 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1106 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1107 // monitor->set_obj(nullptr);
1108
1109 // If we still have a lightweight lock, unlock the object and be done.
1110
1111 // The object address from the monitor is in object.
1112 ld(object, in_bytes(BasicObjectLock::obj_offset()), monitor);
1113
1114 if (LockingMode == LM_LIGHTWEIGHT) {
1115 lightweight_unlock(object, header, slow_case);
1116 } else {
1117 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
1118
1119 // We have the displaced header in displaced_header. If the lock is still
1120 // lightweight, it will contain the monitor address and we'll store the
1121 // displaced header back into the object's mark word.
1122 // CmpxchgX sets CCR0 to cmpX(current, monitor).
1123 cmpxchgd(/*flag=*/CCR0,
1124 /*current_value=*/current_header,
1125 /*compare_value=*/monitor, /*exchange_value=*/header,
1126 /*where=*/object_mark_addr,
1127 MacroAssembler::MemBarRel,
1128 MacroAssembler::cmpxchgx_hint_release_lock(),
1129 noreg,
1130 &slow_case);
1131 }
1132 b(free_slot);
1133
1134 // } else {
1135 // // Slow path.
1136 // InterpreterRuntime::monitorexit(monitor);
1137
1138 // The lock has been converted into a heavy lock and hence
1139 // we need to get into the slow case.
1140 bind(slow_case);
1141 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1142 // }
1143
1144 Label done;
1145 b(done); // Monitor register may be overwritten! Runtime has already freed the slot.
1146
1147 // Exchange worked, do monitor->set_obj(nullptr);
1148 align(32, 12);
1149 bind(free_slot);
1150 li(R0, 0);
1151 std(R0, in_bytes(BasicObjectLock::obj_offset()), monitor);
1152 dec_held_monitor_count(current_header /*tmp*/);
1153 bind(done);
1154 }
1155 }
1156
1157 // Load compiled (i2c) or interpreter entry when calling from interpreted and
1158 // do the call. Centralized so that all interpreter calls will do the same actions.
1159 // If jvmti single stepping is on for a thread we must not call compiled code.
1160 //
1161 // Input:
1162 // - Rtarget_method: method to call
1163 // - Rret_addr: return address
1164 // - 2 scratch regs
1165 //
1166 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr,
1167 Register Rscratch1, Register Rscratch2) {
1168 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr);
1169 // Assume we want to go compiled if available.
1170 const Register Rtarget_addr = Rscratch1;
1171 const Register Rinterp_only = Rscratch2;
1172
1173 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method);
1174
1175 if (JvmtiExport::can_post_interpreter_events()) {
1176 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
1177
1178 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
1179 // compiled code in threads for which the event is enabled. Check here for
1180 // interp_only_mode if these events CAN be enabled.
1181 Label done;
1182 cmpwi(CCR0, Rinterp_only, 0);
1183 beq(CCR0, done);
1184 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method);
1185 align(32, 12);
1186 bind(done);
1187 }
1188
1189 #ifdef ASSERT
1190 {
1191 Label Lok;
1192 cmpdi(CCR0, Rtarget_addr, 0);
1193 bne(CCR0, Lok);
1194 stop("null entry point");
1195 bind(Lok);
1196 }
1197 #endif // ASSERT
1198
1199 mr(R21_sender_SP, R1_SP);
1200
1201 // Calc a precise SP for the call. The SP value we calculated in
1202 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space
1203 // if esp is not max. Also, the i2c adapter extends the stack space without restoring
1204 // our pre-calced value, so repeating calls via i2c would result in stack overflow.
1205 // Since esp already points to an empty slot, we just have to sub 1 additional slot
1206 // to meet the abi scratch requirements.
1207 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in
1208 // the return entry of the interpreter.
1209 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::top_ijava_frame_abi_size);
1210 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address
1211 resize_frame_absolute(Rscratch2, Rscratch2, R0);
1212
1213 mr_if_needed(R19_method, Rtarget_method);
1214 mtctr(Rtarget_addr);
1215 mtlr(Rret_addr);
1216
1217 save_interpreter_state(Rscratch2);
1218 #ifdef ASSERT
1219 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp
1220 sldi(Rscratch1, Rscratch1, Interpreter::logStackElementSize);
1221 add(Rscratch1, Rscratch1, Rscratch2); // Rscratch2 contains fp
1222 // Compare sender_sp with the derelativized top_frame_sp
1223 cmpd(CCR0, R21_sender_SP, Rscratch1);
1224 asm_assert_eq("top_frame_sp incorrect");
1225 #endif
1226
1227 bctr();
1228 }
1229
1230 // Set the method data pointer for the current bcp.
1231 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1232 assert(ProfileInterpreter, "must be profiling interpreter");
1233 Label get_continue;
1234 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method);
1235 test_method_data_pointer(get_continue);
1236 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp);
1237
1238 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1239 add(R28_mdx, R28_mdx, R3_RET);
1240 bind(get_continue);
1241 }
1242
1243 // Test ImethodDataPtr. If it is null, continue at the specified label.
1244 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1245 assert(ProfileInterpreter, "must be profiling interpreter");
1246 cmpdi(CCR0, R28_mdx, 0);
1247 beq(CCR0, zero_continue);
1248 }
1249
1250 void InterpreterMacroAssembler::verify_method_data_pointer() {
1251 assert(ProfileInterpreter, "must be profiling interpreter");
1252 #ifdef ASSERT
1253 Label verify_continue;
1254 test_method_data_pointer(verify_continue);
1255
1256 // If the mdp is valid, it will point to a DataLayout header which is
1257 // consistent with the bcp. The converse is highly probable also.
1258 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx);
1259 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method);
1260 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1261 add(R11_scratch1, R12_scratch2, R12_scratch2);
1262 cmpd(CCR0, R11_scratch1, R14_bcp);
1263 beq(CCR0, verify_continue);
1264
1265 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx);
1266
1267 bind(verify_continue);
1268 #endif
1269 }
1270
1271 // Store a value at some constant offset from the method data pointer.
1272 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1273 assert(ProfileInterpreter, "must be profiling interpreter");
1274
1275 std(value, constant, R28_mdx);
1276 }
1277
1278 // Increment the value at some constant offset from the method data pointer.
1279 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1280 Register counter_addr,
1281 Register Rbumped_count,
1282 bool decrement) {
1283 // Locate the counter at a fixed offset from the mdp:
1284 addi(counter_addr, R28_mdx, constant);
1285 increment_mdp_data_at(counter_addr, Rbumped_count, decrement);
1286 }
1287
1288 // Increment the value at some non-fixed (reg + constant) offset from
1289 // the method data pointer.
1290 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1291 int constant,
1292 Register scratch,
1293 Register Rbumped_count,
1294 bool decrement) {
1295 // Add the constant to reg to get the offset.
1296 add(scratch, R28_mdx, reg);
1297 // Then calculate the counter address.
1298 addi(scratch, scratch, constant);
1299 increment_mdp_data_at(scratch, Rbumped_count, decrement);
1300 }
1301
1302 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr,
1303 Register Rbumped_count,
1304 bool decrement) {
1305 assert(ProfileInterpreter, "must be profiling interpreter");
1306
1307 // Load the counter.
1308 ld(Rbumped_count, 0, counter_addr);
1309
1310 if (decrement) {
1311 // Decrement the register. Set condition codes.
1312 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment);
1313 // Store the decremented counter, if it is still negative.
1314 std(Rbumped_count, 0, counter_addr);
1315 // Note: add/sub overflow check are not ported, since 64 bit
1316 // calculation should never overflow.
1317 } else {
1318 // Increment the register. Set carry flag.
1319 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment);
1320 // Store the incremented counter.
1321 std(Rbumped_count, 0, counter_addr);
1322 }
1323 }
1324
1325 // Set a flag value at the current method data pointer position.
1326 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1327 Register scratch) {
1328 assert(ProfileInterpreter, "must be profiling interpreter");
1329 // Load the data header.
1330 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1331 // Set the flag.
1332 ori(scratch, scratch, flag_constant);
1333 // Store the modified header.
1334 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1335 }
1336
1337 // Test the location at some offset from the method data pointer.
1338 // If it is not equal to value, branch to the not_equal_continue Label.
1339 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1340 Register value,
1341 Label& not_equal_continue,
1342 Register test_out) {
1343 assert(ProfileInterpreter, "must be profiling interpreter");
1344
1345 ld(test_out, offset, R28_mdx);
1346 cmpd(CCR0, value, test_out);
1347 bne(CCR0, not_equal_continue);
1348 }
1349
1350 // Update the method data pointer by the displacement located at some fixed
1351 // offset from the method data pointer.
1352 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1353 Register scratch) {
1354 assert(ProfileInterpreter, "must be profiling interpreter");
1355
1356 ld(scratch, offset_of_disp, R28_mdx);
1357 add(R28_mdx, scratch, R28_mdx);
1358 }
1359
1360 // Update the method data pointer by the displacement located at the
1361 // offset (reg + offset_of_disp).
1362 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1363 int offset_of_disp,
1364 Register scratch) {
1365 assert(ProfileInterpreter, "must be profiling interpreter");
1366
1367 add(scratch, reg, R28_mdx);
1368 ld(scratch, offset_of_disp, scratch);
1369 add(R28_mdx, scratch, R28_mdx);
1370 }
1371
1372 // Update the method data pointer by a simple constant displacement.
1373 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1374 assert(ProfileInterpreter, "must be profiling interpreter");
1375 addi(R28_mdx, R28_mdx, constant);
1376 }
1377
1378 // Update the method data pointer for a _ret bytecode whose target
1379 // was not among our cached targets.
1380 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1381 Register return_bci) {
1382 assert(ProfileInterpreter, "must be profiling interpreter");
1383
1384 push(state);
1385 assert(return_bci->is_nonvolatile(), "need to protect return_bci");
1386 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1387 pop(state);
1388 }
1389
1390 // Increments the backedge counter.
1391 // Returns backedge counter + invocation counter in Rdst.
1392 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst,
1393 const Register Rtmp1, Register Rscratch) {
1394 assert(UseCompiler, "incrementing must be useful");
1395 assert_different_registers(Rdst, Rtmp1);
1396 const Register invocation_counter = Rtmp1;
1397 const Register counter = Rdst;
1398 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size.");
1399
1400 // Load backedge counter.
1401 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1402 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1403 // Load invocation counter.
1404 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) +
1405 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1406
1407 // Add the delta to the backedge counter.
1408 addi(counter, counter, InvocationCounter::count_increment);
1409
1410 // Mask the invocation counter.
1411 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value);
1412
1413 // Store new counter value.
1414 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1415 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1416 // Return invocation counter + backedge counter.
1417 add(counter, counter, invocation_counter);
1418 }
1419
1420 // Count a taken branch in the bytecodes.
1421 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1422 if (ProfileInterpreter) {
1423 Label profile_continue;
1424
1425 // If no method data exists, go to profile_continue.
1426 test_method_data_pointer(profile_continue);
1427
1428 // We are taking a branch. Increment the taken count.
1429 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count);
1430
1431 // The method data pointer needs to be updated to reflect the new target.
1432 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1433 bind (profile_continue);
1434 }
1435 }
1436
1437 // Count a not-taken branch in the bytecodes.
1438 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) {
1439 if (ProfileInterpreter) {
1440 Label profile_continue;
1441
1442 // If no method data exists, go to profile_continue.
1443 test_method_data_pointer(profile_continue);
1444
1445 // We are taking a branch. Increment the not taken count.
1446 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2);
1447
1448 // The method data pointer needs to be updated to correspond to the
1449 // next bytecode.
1450 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1451 bind (profile_continue);
1452 }
1453 }
1454
1455 // Count a non-virtual call in the bytecodes.
1456 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) {
1457 if (ProfileInterpreter) {
1458 Label profile_continue;
1459
1460 // If no method data exists, go to profile_continue.
1461 test_method_data_pointer(profile_continue);
1462
1463 // We are making a call. Increment the count.
1464 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1465
1466 // The method data pointer needs to be updated to reflect the new target.
1467 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1468 bind (profile_continue);
1469 }
1470 }
1471
1472 // Count a final call in the bytecodes.
1473 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) {
1474 if (ProfileInterpreter) {
1475 Label profile_continue;
1476
1477 // If no method data exists, go to profile_continue.
1478 test_method_data_pointer(profile_continue);
1479
1480 // We are making a call. Increment the count.
1481 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1482
1483 // The method data pointer needs to be updated to reflect the new target.
1484 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1485 bind (profile_continue);
1486 }
1487 }
1488
1489 // Count a virtual call in the bytecodes.
1490 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver,
1491 Register Rscratch1,
1492 Register Rscratch2,
1493 bool receiver_can_be_null) {
1494 if (!ProfileInterpreter) { return; }
1495 Label profile_continue;
1496
1497 // If no method data exists, go to profile_continue.
1498 test_method_data_pointer(profile_continue);
1499
1500 Label skip_receiver_profile;
1501 if (receiver_can_be_null) {
1502 Label not_null;
1503 cmpdi(CCR0, Rreceiver, 0);
1504 bne(CCR0, not_null);
1505 // We are making a call. Increment the count for null receiver.
1506 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2);
1507 b(skip_receiver_profile);
1508 bind(not_null);
1509 }
1510
1511 // Record the receiver type.
1512 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2);
1513 bind(skip_receiver_profile);
1514
1515 // The method data pointer needs to be updated to reflect the new target.
1516 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1517 bind (profile_continue);
1518 }
1519
1520 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) {
1521 if (ProfileInterpreter) {
1522 Label profile_continue;
1523
1524 // If no method data exists, go to profile_continue.
1525 test_method_data_pointer(profile_continue);
1526
1527 int mdp_delta = in_bytes(BitData::bit_data_size());
1528 if (TypeProfileCasts) {
1529 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1530
1531 // Record the object type.
1532 record_klass_in_profile(Rklass, Rscratch1, Rscratch2);
1533 }
1534
1535 // The method data pointer needs to be updated.
1536 update_mdp_by_constant(mdp_delta);
1537
1538 bind (profile_continue);
1539 }
1540 }
1541
1542 // Count a ret in the bytecodes.
1543 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci,
1544 Register scratch1, Register scratch2) {
1545 if (ProfileInterpreter) {
1546 Label profile_continue;
1547 uint row;
1548
1549 // If no method data exists, go to profile_continue.
1550 test_method_data_pointer(profile_continue);
1551
1552 // Update the total ret count.
1553 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 );
1554
1555 for (row = 0; row < RetData::row_limit(); row++) {
1556 Label next_test;
1557
1558 // See if return_bci is equal to bci[n]:
1559 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1);
1560
1561 // return_bci is equal to bci[n]. Increment the count.
1562 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2);
1563
1564 // The method data pointer needs to be updated to reflect the new target.
1565 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1);
1566 b(profile_continue);
1567 bind(next_test);
1568 }
1569
1570 update_mdp_for_ret(state, return_bci);
1571
1572 bind (profile_continue);
1573 }
1574 }
1575
1576 // Count the default case of a switch construct.
1577 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) {
1578 if (ProfileInterpreter) {
1579 Label profile_continue;
1580
1581 // If no method data exists, go to profile_continue.
1582 test_method_data_pointer(profile_continue);
1583
1584 // Update the default case count
1585 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1586 scratch1, scratch2);
1587
1588 // The method data pointer needs to be updated.
1589 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()),
1590 scratch1);
1591
1592 bind (profile_continue);
1593 }
1594 }
1595
1596 // Count the index'th case of a switch construct.
1597 void InterpreterMacroAssembler::profile_switch_case(Register index,
1598 Register scratch1,
1599 Register scratch2,
1600 Register scratch3) {
1601 if (ProfileInterpreter) {
1602 assert_different_registers(index, scratch1, scratch2, scratch3);
1603 Label profile_continue;
1604
1605 // If no method data exists, go to profile_continue.
1606 test_method_data_pointer(profile_continue);
1607
1608 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes().
1609 li(scratch3, in_bytes(MultiBranchData::case_array_offset()));
1610
1611 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works");
1612 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1613 add(scratch1, scratch1, scratch3);
1614
1615 // Update the case count.
1616 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3);
1617
1618 // The method data pointer needs to be updated.
1619 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2);
1620
1621 bind (profile_continue);
1622 }
1623 }
1624
1625 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) {
1626 if (ProfileInterpreter) {
1627 assert_different_registers(Rscratch1, Rscratch2);
1628 Label profile_continue;
1629
1630 // If no method data exists, go to profile_continue.
1631 test_method_data_pointer(profile_continue);
1632
1633 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1);
1634
1635 // The method data pointer needs to be updated.
1636 int mdp_delta = in_bytes(BitData::bit_data_size());
1637 if (TypeProfileCasts) {
1638 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1639 }
1640 update_mdp_by_constant(mdp_delta);
1641
1642 bind (profile_continue);
1643 }
1644 }
1645
1646 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver,
1647 Register Rscratch1, Register Rscratch2) {
1648 assert(ProfileInterpreter, "must be profiling");
1649 assert_different_registers(Rreceiver, Rscratch1, Rscratch2);
1650
1651 Label done;
1652 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done);
1653 bind (done);
1654 }
1655
1656 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1657 Register receiver, Register scratch1, Register scratch2,
1658 int start_row, Label& done) {
1659 if (TypeProfileWidth == 0) {
1660 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1661 return;
1662 }
1663
1664 int last_row = VirtualCallData::row_limit() - 1;
1665 assert(start_row <= last_row, "must be work left to do");
1666 // Test this row for both the receiver and for null.
1667 // Take any of three different outcomes:
1668 // 1. found receiver => increment count and goto done
1669 // 2. found null => keep looking for case 1, maybe allocate this cell
1670 // 3. found something else => keep looking for cases 1 and 2
1671 // Case 3 is handled by a recursive call.
1672 for (int row = start_row; row <= last_row; row++) {
1673 Label next_test;
1674 bool test_for_null_also = (row == start_row);
1675
1676 // See if the receiver is receiver[n].
1677 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1678 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1);
1679 // delayed()->tst(scratch);
1680
1681 // The receiver is receiver[n]. Increment count[n].
1682 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1683 increment_mdp_data_at(count_offset, scratch1, scratch2);
1684 b(done);
1685 bind(next_test);
1686
1687 if (test_for_null_also) {
1688 Label found_null;
1689 // Failed the equality check on receiver[n]... Test for null.
1690 if (start_row == last_row) {
1691 // The only thing left to do is handle the null case.
1692 // Scratch1 contains test_out from test_mdp_data_at.
1693 cmpdi(CCR0, scratch1, 0);
1694 beq(CCR0, found_null);
1695 // Receiver did not match any saved receiver and there is no empty row for it.
1696 // Increment total counter to indicate polymorphic case.
1697 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1698 b(done);
1699 bind(found_null);
1700 break;
1701 }
1702 // Since null is rare, make it be the branch-taken case.
1703 cmpdi(CCR0, scratch1, 0);
1704 beq(CCR0, found_null);
1705
1706 // Put all the "Case 3" tests here.
1707 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done);
1708
1709 // Found a null. Keep searching for a matching receiver,
1710 // but remember that this is an empty (unused) slot.
1711 bind(found_null);
1712 }
1713 }
1714
1715 // In the fall-through case, we found no matching receiver, but we
1716 // observed the receiver[start_row] is null.
1717
1718 // Fill in the receiver field and increment the count.
1719 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1720 set_mdp_data_at(recvr_offset, receiver);
1721 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1722 li(scratch1, DataLayout::counter_increment);
1723 set_mdp_data_at(count_offset, scratch1);
1724 if (start_row > 0) {
1725 b(done);
1726 }
1727 }
1728
1729 // Argument and return type profilig.
1730 // kills: tmp, tmp2, R0, CR0, CR1
1731 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base,
1732 RegisterOrConstant mdo_addr_offs,
1733 Register tmp, Register tmp2) {
1734 Label do_nothing, do_update;
1735
1736 // tmp2 = obj is allowed
1737 assert_different_registers(obj, mdo_addr_base, tmp, R0);
1738 assert_different_registers(tmp2, mdo_addr_base, tmp, R0);
1739 const Register klass = tmp2;
1740
1741 verify_oop(obj);
1742
1743 ld(tmp, mdo_addr_offs, mdo_addr_base);
1744
1745 // Set null_seen if obj is 0.
1746 cmpdi(CCR0, obj, 0);
1747 ori(R0, tmp, TypeEntries::null_seen);
1748 beq(CCR0, do_update);
1749
1750 load_klass(klass, obj);
1751
1752 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
1753 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
1754 cmpd(CCR1, R0, klass);
1755 // Klass seen before, nothing to do (regardless of unknown bit).
1756 //beq(CCR1, do_nothing);
1757
1758 andi_(R0, tmp, TypeEntries::type_unknown);
1759 // Already unknown. Nothing to do anymore.
1760 //bne(CCR0, do_nothing);
1761 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne
1762 beq(CCR0, do_nothing);
1763
1764 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
1765 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1766 beq(CCR0, do_update); // First time here. Set profile type.
1767
1768 // Different than before. Cannot keep accurate profile.
1769 ori(R0, tmp, TypeEntries::type_unknown);
1770
1771 bind(do_update);
1772 // update profile
1773 std(R0, mdo_addr_offs, mdo_addr_base);
1774
1775 align(32, 12);
1776 bind(do_nothing);
1777 }
1778
1779 void InterpreterMacroAssembler::profile_arguments_type(Register callee,
1780 Register tmp1, Register tmp2,
1781 bool is_virtual) {
1782 if (!ProfileInterpreter) {
1783 return;
1784 }
1785
1786 assert_different_registers(callee, tmp1, tmp2, R28_mdx);
1787
1788 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1789 Label profile_continue;
1790
1791 test_method_data_pointer(profile_continue);
1792
1793 int off_to_start = is_virtual ?
1794 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1795
1796 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx);
1797 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1798 bne(CCR0, profile_continue);
1799
1800 if (MethodData::profile_arguments()) {
1801 Label done;
1802 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1803 addi(R28_mdx, R28_mdx, off_to_args);
1804
1805 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1806 if (i > 0 || MethodData::profile_return()) {
1807 // If return value type is profiled we may have no argument to profile.
1808 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1809 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count());
1810 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count());
1811 blt(CCR0, done);
1812 }
1813 ld(tmp1, in_bytes(Method::const_offset()), callee);
1814 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1);
1815 // Stack offset o (zero based) from the start of the argument
1816 // list, for n arguments translates into offset n - o - 1 from
1817 // the end of the argument list. But there's an extra slot at
1818 // the top of the stack. So the offset is n - o from Lesp.
1819 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx);
1820 subf(tmp1, tmp2, tmp1);
1821
1822 sldi(tmp1, tmp1, Interpreter::logStackElementSize);
1823 ldx(tmp1, tmp1, R15_esp);
1824
1825 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1);
1826
1827 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1828 addi(R28_mdx, R28_mdx, to_add);
1829 off_to_args += to_add;
1830 }
1831
1832 if (MethodData::profile_return()) {
1833 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1834 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1835 }
1836
1837 bind(done);
1838
1839 if (MethodData::profile_return()) {
1840 // We're right after the type profile for the last
1841 // argument. tmp1 is the number of cells left in the
1842 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1843 // if there's a return to profile.
1844 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(),
1845 "can't move past ret type");
1846 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size));
1847 add(R28_mdx, tmp1, R28_mdx);
1848 }
1849 } else {
1850 assert(MethodData::profile_return(), "either profile call args or call ret");
1851 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
1852 }
1853
1854 // Mdp points right after the end of the
1855 // CallTypeData/VirtualCallTypeData, right after the cells for the
1856 // return value type if there's one.
1857 align(32, 12);
1858 bind(profile_continue);
1859 }
1860 }
1861
1862 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
1863 assert_different_registers(ret, tmp1, tmp2);
1864 if (ProfileInterpreter && MethodData::profile_return()) {
1865 Label profile_continue;
1866
1867 test_method_data_pointer(profile_continue);
1868
1869 if (MethodData::profile_return_jsr292_only()) {
1870 // If we don't profile all invoke bytecodes we must make sure
1871 // it's a bytecode we indeed profile. We can't go back to the
1872 // beginning of the ProfileData we intend to update to check its
1873 // type because we're right after it and we don't known its
1874 // length.
1875 lbz(tmp1, 0, R14_bcp);
1876 lbz(tmp2, in_bytes(Method::intrinsic_id_offset()), R19_method);
1877 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic);
1878 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle);
1879 cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
1880 cmpwi(CCR1, tmp2, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1881 cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
1882 bne(CCR0, profile_continue);
1883 }
1884
1885 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2);
1886
1887 align(32, 12);
1888 bind(profile_continue);
1889 }
1890 }
1891
1892 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2,
1893 Register tmp3, Register tmp4) {
1894 if (ProfileInterpreter && MethodData::profile_parameters()) {
1895 Label profile_continue, done;
1896
1897 test_method_data_pointer(profile_continue);
1898
1899 // Load the offset of the area within the MDO used for
1900 // parameters. If it's negative we're not profiling any parameters.
1901 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx);
1902 cmpwi(CCR0, tmp1, 0);
1903 blt(CCR0, profile_continue);
1904
1905 // Compute a pointer to the area for parameters from the offset
1906 // and move the pointer to the slot for the last
1907 // parameters. Collect profiling from last parameter down.
1908 // mdo start + parameters offset + array length - 1
1909
1910 // Pointer to the parameter area in the MDO.
1911 const Register mdp = tmp1;
1912 add(mdp, tmp1, R28_mdx);
1913
1914 // Offset of the current profile entry to update.
1915 const Register entry_offset = tmp2;
1916 // entry_offset = array len in number of cells
1917 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp);
1918
1919 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1920 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
1921
1922 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
1923 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size));
1924 // entry_offset in bytes
1925 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1926
1927 Label loop;
1928 align(32, 12);
1929 bind(loop);
1930
1931 // Load offset on the stack from the slot for this parameter.
1932 ld(tmp3, entry_offset, mdp);
1933 sldi(tmp3, tmp3, Interpreter::logStackElementSize);
1934 neg(tmp3, tmp3);
1935 // Read the parameter from the local area.
1936 ldx(tmp3, tmp3, R18_locals);
1937
1938 // Make entry_offset now point to the type field for this parameter.
1939 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1940 assert(type_base > off_base, "unexpected");
1941 addi(entry_offset, entry_offset, type_base - off_base);
1942
1943 // Profile the parameter.
1944 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3);
1945
1946 // Go to next parameter.
1947 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base);
1948 cmpdi(CCR0, entry_offset, off_base + delta);
1949 addi(entry_offset, entry_offset, -delta);
1950 bge(CCR0, loop);
1951
1952 align(32, 12);
1953 bind(profile_continue);
1954 }
1955 }
1956
1957 // Add a monitor (see frame_ppc.hpp).
1958 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) {
1959
1960 // Very-local scratch registers.
1961 const Register esp = Rtemp1;
1962 const Register slot = Rtemp2;
1963
1964 // Extracted monitor_size.
1965 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes();
1966 assert(Assembler::is_aligned((unsigned int)monitor_size,
1967 (unsigned int)frame::alignment_in_bytes),
1968 "size of a monitor must respect alignment of SP");
1969
1970 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor
1971 subf(Rtemp2, esp, R1_SP); // esp contains fp
1972 sradi(Rtemp2, Rtemp2, Interpreter::logStackElementSize);
1973 // Store relativized top_frame_sp
1974 std(Rtemp2, _ijava_state_neg(top_frame_sp), esp); // esp contains fp
1975
1976 // Shuffle expression stack down. Recall that stack_base points
1977 // just above the new expression stack bottom. Old_tos and new_tos
1978 // are used to scan thru the old and new expression stacks.
1979 if (!stack_is_empty) {
1980 Label copy_slot, copy_slot_finished;
1981 const Register n_slots = slot;
1982
1983 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack).
1984 subf(n_slots, esp, R26_monitor);
1985 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy.
1986 assert(LogBytesPerWord == 3, "conflicts assembler instructions");
1987 beq(CCR0, copy_slot_finished); // Nothing to copy.
1988
1989 mtctr(n_slots);
1990
1991 // loop
1992 bind(copy_slot);
1993 ld(slot, 0, esp); // Move expression stack down.
1994 std(slot, -monitor_size, esp); // distance = monitor_size
1995 addi(esp, esp, BytesPerWord);
1996 bdnz(copy_slot);
1997
1998 bind(copy_slot_finished);
1999 }
2000
2001 addi(R15_esp, R15_esp, -monitor_size);
2002 addi(R26_monitor, R26_monitor, -monitor_size);
2003
2004 // Restart interpreter
2005 }
2006
2007 // ============================================================================
2008 // Java locals access
2009
2010 // Load a local variable at index in Rindex into register Rdst_value.
2011 // Also puts address of local into Rdst_address as a service.
2012 // Kills:
2013 // - Rdst_value
2014 // - Rdst_address
2015 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) {
2016 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2017 subf(Rdst_address, Rdst_address, R18_locals);
2018 lwz(Rdst_value, 0, Rdst_address);
2019 }
2020
2021 // Load a local variable at index in Rindex into register Rdst_value.
2022 // Also puts address of local into Rdst_address as a service.
2023 // Kills:
2024 // - Rdst_value
2025 // - Rdst_address
2026 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) {
2027 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2028 subf(Rdst_address, Rdst_address, R18_locals);
2029 ld(Rdst_value, -8, Rdst_address);
2030 }
2031
2032 // Load a local variable at index in Rindex into register Rdst_value.
2033 // Also puts address of local into Rdst_address as a service.
2034 // Input:
2035 // - Rindex: slot nr of local variable
2036 // Kills:
2037 // - Rdst_value
2038 // - Rdst_address
2039 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value,
2040 Register Rdst_address,
2041 Register Rindex) {
2042 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2043 subf(Rdst_address, Rdst_address, R18_locals);
2044 ld(Rdst_value, 0, Rdst_address);
2045 }
2046
2047 // Load a local variable at index in Rindex into register Rdst_value.
2048 // Also puts address of local into Rdst_address as a service.
2049 // Kills:
2050 // - Rdst_value
2051 // - Rdst_address
2052 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value,
2053 Register Rdst_address,
2054 Register Rindex) {
2055 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2056 subf(Rdst_address, Rdst_address, R18_locals);
2057 lfs(Rdst_value, 0, Rdst_address);
2058 }
2059
2060 // Load a local variable at index in Rindex into register Rdst_value.
2061 // Also puts address of local into Rdst_address as a service.
2062 // Kills:
2063 // - Rdst_value
2064 // - Rdst_address
2065 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value,
2066 Register Rdst_address,
2067 Register Rindex) {
2068 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2069 subf(Rdst_address, Rdst_address, R18_locals);
2070 lfd(Rdst_value, -8, Rdst_address);
2071 }
2072
2073 // Store an int value at local variable slot Rindex.
2074 // Kills:
2075 // - Rindex
2076 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) {
2077 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2078 subf(Rindex, Rindex, R18_locals);
2079 stw(Rvalue, 0, Rindex);
2080 }
2081
2082 // Store a long value at local variable slot Rindex.
2083 // Kills:
2084 // - Rindex
2085 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) {
2086 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2087 subf(Rindex, Rindex, R18_locals);
2088 std(Rvalue, -8, Rindex);
2089 }
2090
2091 // Store an oop value at local variable slot Rindex.
2092 // Kills:
2093 // - Rindex
2094 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) {
2095 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2096 subf(Rindex, Rindex, R18_locals);
2097 std(Rvalue, 0, Rindex);
2098 }
2099
2100 // Store an int value at local variable slot Rindex.
2101 // Kills:
2102 // - Rindex
2103 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) {
2104 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2105 subf(Rindex, Rindex, R18_locals);
2106 stfs(Rvalue, 0, Rindex);
2107 }
2108
2109 // Store an int value at local variable slot Rindex.
2110 // Kills:
2111 // - Rindex
2112 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) {
2113 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2114 subf(Rindex, Rindex, R18_locals);
2115 stfd(Rvalue, -8, Rindex);
2116 }
2117
2118 // Read pending exception from thread and jump to interpreter.
2119 // Throw exception entry if one if pending. Fall through otherwise.
2120 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) {
2121 assert_different_registers(Rscratch1, Rscratch2, R3);
2122 Register Rexception = Rscratch1;
2123 Register Rtmp = Rscratch2;
2124 Label Ldone;
2125 // Get pending exception oop.
2126 ld(Rexception, thread_(pending_exception));
2127 cmpdi(CCR0, Rexception, 0);
2128 beq(CCR0, Ldone);
2129 li(Rtmp, 0);
2130 mr_if_needed(R3, Rexception);
2131 std(Rtmp, thread_(pending_exception)); // Clear exception in thread
2132 if (Interpreter::rethrow_exception_entry() != nullptr) {
2133 // Already got entry address.
2134 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry());
2135 } else {
2136 // Dynamically load entry address.
2137 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true);
2138 ld(Rtmp, simm16_rest, Rtmp);
2139 }
2140 mtctr(Rtmp);
2141 save_interpreter_state(Rtmp);
2142 bctr();
2143
2144 align(32, 12);
2145 bind(Ldone);
2146 }
2147
2148 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) {
2149 save_interpreter_state(R11_scratch1);
2150
2151 MacroAssembler::call_VM(oop_result, entry_point, false);
2152
2153 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2154
2155 check_and_handle_popframe(R11_scratch1);
2156 check_and_handle_earlyret(R11_scratch1);
2157 // Now check exceptions manually.
2158 if (check_exceptions) {
2159 check_and_forward_exception(R11_scratch1, R12_scratch2);
2160 }
2161 }
2162
2163 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2164 Register arg_1, bool check_exceptions) {
2165 // ARG1 is reserved for the thread.
2166 mr_if_needed(R4_ARG2, arg_1);
2167 call_VM(oop_result, entry_point, check_exceptions);
2168 }
2169
2170 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2171 Register arg_1, Register arg_2,
2172 bool check_exceptions) {
2173 // ARG1 is reserved for the thread.
2174 mr_if_needed(R4_ARG2, arg_1);
2175 assert(arg_2 != R4_ARG2, "smashed argument");
2176 mr_if_needed(R5_ARG3, arg_2);
2177 call_VM(oop_result, entry_point, check_exceptions);
2178 }
2179
2180 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2181 Register arg_1, Register arg_2, Register arg_3,
2182 bool check_exceptions) {
2183 // ARG1 is reserved for the thread.
2184 mr_if_needed(R4_ARG2, arg_1);
2185 assert(arg_2 != R4_ARG2, "smashed argument");
2186 mr_if_needed(R5_ARG3, arg_2);
2187 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument");
2188 mr_if_needed(R6_ARG4, arg_3);
2189 call_VM(oop_result, entry_point, check_exceptions);
2190 }
2191
2192 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) {
2193 ld(scratch, 0, R1_SP);
2194 subf(R0, scratch, R15_esp);
2195 sradi(R0, R0, Interpreter::logStackElementSize);
2196 std(R0, _ijava_state_neg(esp), scratch);
2197 std(R14_bcp, _ijava_state_neg(bcp), scratch);
2198 subf(R0, scratch, R26_monitor);
2199 sradi(R0, R0, Interpreter::logStackElementSize);
2200 std(R0, _ijava_state_neg(monitors), scratch);
2201 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); }
2202 // Other entries should be unchanged.
2203 }
2204
2205 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only, bool restore_top_frame_sp) {
2206 ld_ptr(scratch, _abi0(callers_sp), R1_SP); // Load frame pointer.
2207 if (restore_top_frame_sp) {
2208 // After thawing the top frame of a continuation we reach here with frame::java_abi.
2209 // therefore we have to restore top_frame_sp before the assertion below.
2210 assert(!bcp_and_mdx_only, "chose other registers");
2211 Register tfsp = R18_locals;
2212 Register scratch2 = R26_monitor;
2213 ld(tfsp, _ijava_state_neg(top_frame_sp), scratch);
2214 // Derelativize top_frame_sp
2215 sldi(tfsp, tfsp, Interpreter::logStackElementSize);
2216 add(tfsp, tfsp, scratch);
2217 resize_frame_absolute(tfsp, scratch2, R0);
2218 }
2219 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception).
2220 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code.
2221 if (!bcp_and_mdx_only) {
2222 // Following ones are Metadata.
2223 ld(R19_method, _ijava_state_neg(method), scratch);
2224 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch);
2225 // Following ones are stack addresses and don't require reload.
2226 // Derelativize esp
2227 ld(R15_esp, _ijava_state_neg(esp), scratch);
2228 sldi(R15_esp, R15_esp, Interpreter::logStackElementSize);
2229 add(R15_esp, R15_esp, scratch);
2230 ld(R18_locals, _ijava_state_neg(locals), scratch);
2231 sldi(R18_locals, R18_locals, Interpreter::logStackElementSize);
2232 add(R18_locals, R18_locals, scratch);
2233 ld(R26_monitor, _ijava_state_neg(monitors), scratch);
2234 // Derelativize monitors
2235 sldi(R26_monitor, R26_monitor, Interpreter::logStackElementSize);
2236 add(R26_monitor, R26_monitor, scratch);
2237 }
2238 #ifdef ASSERT
2239 {
2240 Label Lok;
2241 subf(R0, R1_SP, scratch);
2242 cmpdi(CCR0, R0, frame::top_ijava_frame_abi_size + frame::ijava_state_size);
2243 bge(CCR0, Lok);
2244 stop("frame too small (restore istate)");
2245 bind(Lok);
2246 }
2247 #endif
2248 }
2249
2250 void InterpreterMacroAssembler::get_method_counters(Register method,
2251 Register Rcounters,
2252 Label& skip) {
2253 BLOCK_COMMENT("Load and ev. allocate counter object {");
2254 Label has_counters;
2255 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2256 cmpdi(CCR0, Rcounters, 0);
2257 bne(CCR0, has_counters);
2258 call_VM(noreg, CAST_FROM_FN_PTR(address,
2259 InterpreterRuntime::build_method_counters), method);
2260 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2261 cmpdi(CCR0, Rcounters, 0);
2262 beq(CCR0, skip); // No MethodCounters, OutOfMemory.
2263 BLOCK_COMMENT("} Load and ev. allocate counter object");
2264
2265 bind(has_counters);
2266 }
2267
2268 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters,
2269 Register iv_be_count,
2270 Register Rtmp_r0) {
2271 assert(UseCompiler, "incrementing must be useful");
2272 Register invocation_count = iv_be_count;
2273 Register backedge_count = Rtmp_r0;
2274 int delta = InvocationCounter::count_increment;
2275
2276 // Load each counter in a register.
2277 // ld(inv_counter, Rtmp);
2278 // ld(be_counter, Rtmp2);
2279 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() +
2280 InvocationCounter::counter_offset());
2281 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() +
2282 InvocationCounter::counter_offset());
2283
2284 BLOCK_COMMENT("Increment profiling counters {");
2285
2286 // Load the backedge counter.
2287 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int
2288 // Mask the backedge counter.
2289 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value);
2290
2291 // Load the invocation counter.
2292 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int
2293 // Add the delta to the invocation counter and store the result.
2294 addi(invocation_count, invocation_count, delta);
2295 // Store value.
2296 stw(invocation_count, inv_counter_offset, Rcounters);
2297
2298 // Add invocation counter + backedge counter.
2299 add(iv_be_count, backedge_count, invocation_count);
2300
2301 // Note that this macro must leave the backedge_count + invocation_count in
2302 // register iv_be_count!
2303 BLOCK_COMMENT("} Increment profiling counters");
2304 }
2305
2306 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2307 if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); }
2308 }
2309
2310 // Local helper function for the verify_oop_or_return_address macro.
2311 static bool verify_return_address(Method* m, int bci) {
2312 #ifndef PRODUCT
2313 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci;
2314 // Assume it is a valid return address if it is inside m and is preceded by a jsr.
2315 if (!m->contains(pc)) return false;
2316 address jsr_pc;
2317 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2318 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2319 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2320 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2321 #endif // PRODUCT
2322 return false;
2323 }
2324
2325 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2326 if (VerifyFPU) {
2327 unimplemented("verfiyFPU");
2328 }
2329 }
2330
2331 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2332 if (!VerifyOops) return;
2333
2334 // The VM documentation for the astore[_wide] bytecode allows
2335 // the TOS to be not only an oop but also a return address.
2336 Label test;
2337 Label skip;
2338 // See if it is an address (in the current method):
2339
2340 const int log2_bytecode_size_limit = 16;
2341 srdi_(Rtmp, reg, log2_bytecode_size_limit);
2342 bne(CCR0, test);
2343
2344 address fd = CAST_FROM_FN_PTR(address, verify_return_address);
2345 const int nbytes_save = MacroAssembler::num_volatile_regs * 8;
2346 save_volatile_gprs(R1_SP, -nbytes_save); // except R0
2347 save_LR_CR(Rtmp); // Save in old frame.
2348 push_frame_reg_args(nbytes_save, Rtmp);
2349
2350 load_const_optimized(Rtmp, fd, R0);
2351 mr_if_needed(R4_ARG2, reg);
2352 mr(R3_ARG1, R19_method);
2353 call_c(Rtmp); // call C
2354
2355 pop_frame();
2356 restore_LR_CR(Rtmp);
2357 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
2358 b(skip);
2359
2360 // Perform a more elaborate out-of-line call.
2361 // Not an address; verify it:
2362 bind(test);
2363 verify_oop(reg);
2364 bind(skip);
2365 }
2366
2367 // Inline assembly for:
2368 //
2369 // if (thread is in interp_only_mode) {
2370 // InterpreterRuntime::post_method_entry();
2371 // }
2372 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) ||
2373 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) {
2374 // SharedRuntime::jvmpi_method_entry(method, receiver);
2375 // }
2376 void InterpreterMacroAssembler::notify_method_entry() {
2377 // JVMTI
2378 // Whenever JVMTI puts a thread in interp_only_mode, method
2379 // entry/exit events are sent for that thread to track stack
2380 // depth. If it is possible to enter interp_only_mode we add
2381 // the code to check if the event should be sent.
2382 if (JvmtiExport::can_post_interpreter_events()) {
2383 Label jvmti_post_done;
2384
2385 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2386 cmpwi(CCR0, R0, 0);
2387 beq(CCR0, jvmti_post_done);
2388 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2389
2390 bind(jvmti_post_done);
2391 }
2392 }
2393
2394 // Inline assembly for:
2395 //
2396 // if (thread is in interp_only_mode) {
2397 // // save result
2398 // InterpreterRuntime::post_method_exit();
2399 // // restore result
2400 // }
2401 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) {
2402 // // save result
2403 // SharedRuntime::jvmpi_method_exit();
2404 // // restore result
2405 // }
2406 //
2407 // Native methods have their result stored in d_tmp and l_tmp.
2408 // Java methods have their result stored in the expression stack.
2409 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state,
2410 NotifyMethodExitMode mode, bool check_exceptions) {
2411 // JVMTI
2412 // Whenever JVMTI puts a thread in interp_only_mode, method
2413 // entry/exit events are sent for that thread to track stack
2414 // depth. If it is possible to enter interp_only_mode we add
2415 // the code to check if the event should be sent.
2416 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2417 Label jvmti_post_done;
2418
2419 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2420 cmpwi(CCR0, R0, 0);
2421 beq(CCR0, jvmti_post_done);
2422 if (!is_native_method) { push(state); } // Expose tos to GC.
2423 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), check_exceptions);
2424 if (!is_native_method) { pop(state); }
2425
2426 align(32, 12);
2427 bind(jvmti_post_done);
2428 }
2429
2430 // Dtrace support not implemented.
2431 }