1 /*
2 * Copyright (c) 2003, 2024, 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 lbz(tmp, in_bytes(Klass::misc_flags_offset()), tmp);
974 testbitdi(CCR0, R0, tmp, exact_log2(KlassFlags::_misc_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 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
1047 b(done);
1048 // }
1049 align(32, 12);
1050 bind(count_locking);
1051 inc_held_monitor_count(current_header /*tmp*/);
1052 bind(done);
1053 }
1054 }
1055
1056 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1057 //
1058 // Registers alive
1059 // monitor - Address of the BasicObjectLock to be used for locking,
1060 // which must be initialized with the object to lock.
1061 //
1062 // Throw IllegalMonitorException if object is not locked by current thread.
1063 void InterpreterMacroAssembler::unlock_object(Register monitor) {
1064 if (LockingMode == LM_MONITOR) {
1065 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1066 } else {
1067
1068 // template code (for LM_LEGACY):
1069 //
1070 // if ((displaced_header = monitor->displaced_header()) == nullptr) {
1071 // // Recursive unlock. Mark the monitor unlocked by setting the object field to null.
1072 // monitor->set_obj(nullptr);
1073 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1074 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1075 // monitor->set_obj(nullptr);
1076 // } else {
1077 // // Slow path.
1078 // InterpreterRuntime::monitorexit(monitor);
1079 // }
1080
1081 const Register object = R7_ARG5;
1082 const Register header = R8_ARG6;
1083 const Register object_mark_addr = R9_ARG7;
1084 const Register current_header = R10_ARG8;
1085
1086 Label free_slot;
1087 Label slow_case;
1088
1089 assert_different_registers(object, header, object_mark_addr, current_header);
1090
1091 if (LockingMode != LM_LIGHTWEIGHT) {
1092 // Test first if we are in the fast recursive case.
1093 ld(header, in_bytes(BasicObjectLock::lock_offset()) +
1094 BasicLock::displaced_header_offset_in_bytes(), monitor);
1095
1096 // If the displaced header is zero, we have a recursive unlock.
1097 cmpdi(CCR0, header, 0);
1098 beq(CCR0, free_slot); // recursive unlock
1099 }
1100
1101 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1102 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1103 // monitor->set_obj(nullptr);
1104
1105 // If we still have a lightweight lock, unlock the object and be done.
1106
1107 // The object address from the monitor is in object.
1108 ld(object, in_bytes(BasicObjectLock::obj_offset()), monitor);
1109
1110 if (LockingMode == LM_LIGHTWEIGHT) {
1111 lightweight_unlock(object, header, slow_case);
1112 } else {
1113 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
1114
1115 // We have the displaced header in displaced_header. If the lock is still
1116 // lightweight, it will contain the monitor address and we'll store the
1117 // displaced header back into the object's mark word.
1118 // CmpxchgX sets CCR0 to cmpX(current, monitor).
1119 cmpxchgd(/*flag=*/CCR0,
1120 /*current_value=*/current_header,
1121 /*compare_value=*/monitor, /*exchange_value=*/header,
1122 /*where=*/object_mark_addr,
1123 MacroAssembler::MemBarRel,
1124 MacroAssembler::cmpxchgx_hint_release_lock(),
1125 noreg,
1126 &slow_case);
1127 }
1128 b(free_slot);
1129
1130 // } else {
1131 // // Slow path.
1132 // InterpreterRuntime::monitorexit(monitor);
1133
1134 // The lock has been converted into a heavy lock and hence
1135 // we need to get into the slow case.
1136 bind(slow_case);
1137 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1138 // }
1139
1140 Label done;
1141 b(done); // Monitor register may be overwritten! Runtime has already freed the slot.
1142
1143 // Exchange worked, do monitor->set_obj(nullptr);
1144 align(32, 12);
1145 bind(free_slot);
1146 li(R0, 0);
1147 std(R0, in_bytes(BasicObjectLock::obj_offset()), monitor);
1148 dec_held_monitor_count(current_header /*tmp*/);
1149 bind(done);
1150 }
1151 }
1152
1153 // Load compiled (i2c) or interpreter entry when calling from interpreted and
1154 // do the call. Centralized so that all interpreter calls will do the same actions.
1155 // If jvmti single stepping is on for a thread we must not call compiled code.
1156 //
1157 // Input:
1158 // - Rtarget_method: method to call
1159 // - Rret_addr: return address
1160 // - 2 scratch regs
1161 //
1162 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr,
1163 Register Rscratch1, Register Rscratch2) {
1164 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr);
1165 // Assume we want to go compiled if available.
1166 const Register Rtarget_addr = Rscratch1;
1167 const Register Rinterp_only = Rscratch2;
1168
1169 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method);
1170
1171 if (JvmtiExport::can_post_interpreter_events()) {
1172 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
1173
1174 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
1175 // compiled code in threads for which the event is enabled. Check here for
1176 // interp_only_mode if these events CAN be enabled.
1177 Label done;
1178 cmpwi(CCR0, Rinterp_only, 0);
1179 beq(CCR0, done);
1180 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method);
1181 align(32, 12);
1182 bind(done);
1183 }
1184
1185 #ifdef ASSERT
1186 {
1187 Label Lok;
1188 cmpdi(CCR0, Rtarget_addr, 0);
1189 bne(CCR0, Lok);
1190 stop("null entry point");
1191 bind(Lok);
1192 }
1193 #endif // ASSERT
1194
1195 mr(R21_sender_SP, R1_SP);
1196
1197 // Calc a precise SP for the call. The SP value we calculated in
1198 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space
1199 // if esp is not max. Also, the i2c adapter extends the stack space without restoring
1200 // our pre-calced value, so repeating calls via i2c would result in stack overflow.
1201 // Since esp already points to an empty slot, we just have to sub 1 additional slot
1202 // to meet the abi scratch requirements.
1203 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in
1204 // the return entry of the interpreter.
1205 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::top_ijava_frame_abi_size);
1206 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address
1207 resize_frame_absolute(Rscratch2, Rscratch2, R0);
1208
1209 mr_if_needed(R19_method, Rtarget_method);
1210 mtctr(Rtarget_addr);
1211 mtlr(Rret_addr);
1212
1213 save_interpreter_state(Rscratch2);
1214 #ifdef ASSERT
1215 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp
1216 sldi(Rscratch1, Rscratch1, Interpreter::logStackElementSize);
1217 add(Rscratch1, Rscratch1, Rscratch2); // Rscratch2 contains fp
1218 // Compare sender_sp with the derelativized top_frame_sp
1219 cmpd(CCR0, R21_sender_SP, Rscratch1);
1220 asm_assert_eq("top_frame_sp incorrect");
1221 #endif
1222
1223 bctr();
1224 }
1225
1226 // Set the method data pointer for the current bcp.
1227 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1228 assert(ProfileInterpreter, "must be profiling interpreter");
1229 Label get_continue;
1230 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method);
1231 test_method_data_pointer(get_continue);
1232 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp);
1233
1234 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1235 add(R28_mdx, R28_mdx, R3_RET);
1236 bind(get_continue);
1237 }
1238
1239 // Test ImethodDataPtr. If it is null, continue at the specified label.
1240 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1241 assert(ProfileInterpreter, "must be profiling interpreter");
1242 cmpdi(CCR0, R28_mdx, 0);
1243 beq(CCR0, zero_continue);
1244 }
1245
1246 void InterpreterMacroAssembler::verify_method_data_pointer() {
1247 assert(ProfileInterpreter, "must be profiling interpreter");
1248 #ifdef ASSERT
1249 Label verify_continue;
1250 test_method_data_pointer(verify_continue);
1251
1252 // If the mdp is valid, it will point to a DataLayout header which is
1253 // consistent with the bcp. The converse is highly probable also.
1254 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx);
1255 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method);
1256 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1257 add(R11_scratch1, R12_scratch2, R12_scratch2);
1258 cmpd(CCR0, R11_scratch1, R14_bcp);
1259 beq(CCR0, verify_continue);
1260
1261 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx);
1262
1263 bind(verify_continue);
1264 #endif
1265 }
1266
1267 // Store a value at some constant offset from the method data pointer.
1268 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1269 assert(ProfileInterpreter, "must be profiling interpreter");
1270
1271 std(value, constant, R28_mdx);
1272 }
1273
1274 // Increment the value at some constant offset from the method data pointer.
1275 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1276 Register counter_addr,
1277 Register Rbumped_count,
1278 bool decrement) {
1279 // Locate the counter at a fixed offset from the mdp:
1280 addi(counter_addr, R28_mdx, constant);
1281 increment_mdp_data_at(counter_addr, Rbumped_count, decrement);
1282 }
1283
1284 // Increment the value at some non-fixed (reg + constant) offset from
1285 // the method data pointer.
1286 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1287 int constant,
1288 Register scratch,
1289 Register Rbumped_count,
1290 bool decrement) {
1291 // Add the constant to reg to get the offset.
1292 add(scratch, R28_mdx, reg);
1293 // Then calculate the counter address.
1294 addi(scratch, scratch, constant);
1295 increment_mdp_data_at(scratch, Rbumped_count, decrement);
1296 }
1297
1298 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr,
1299 Register Rbumped_count,
1300 bool decrement) {
1301 assert(ProfileInterpreter, "must be profiling interpreter");
1302
1303 // Load the counter.
1304 ld(Rbumped_count, 0, counter_addr);
1305
1306 if (decrement) {
1307 // Decrement the register. Set condition codes.
1308 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment);
1309 // Store the decremented counter, if it is still negative.
1310 std(Rbumped_count, 0, counter_addr);
1311 // Note: add/sub overflow check are not ported, since 64 bit
1312 // calculation should never overflow.
1313 } else {
1314 // Increment the register. Set carry flag.
1315 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment);
1316 // Store the incremented counter.
1317 std(Rbumped_count, 0, counter_addr);
1318 }
1319 }
1320
1321 // Set a flag value at the current method data pointer position.
1322 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1323 Register scratch) {
1324 assert(ProfileInterpreter, "must be profiling interpreter");
1325 // Load the data header.
1326 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1327 // Set the flag.
1328 ori(scratch, scratch, flag_constant);
1329 // Store the modified header.
1330 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1331 }
1332
1333 // Test the location at some offset from the method data pointer.
1334 // If it is not equal to value, branch to the not_equal_continue Label.
1335 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1336 Register value,
1337 Label& not_equal_continue,
1338 Register test_out) {
1339 assert(ProfileInterpreter, "must be profiling interpreter");
1340
1341 ld(test_out, offset, R28_mdx);
1342 cmpd(CCR0, value, test_out);
1343 bne(CCR0, not_equal_continue);
1344 }
1345
1346 // Update the method data pointer by the displacement located at some fixed
1347 // offset from the method data pointer.
1348 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1349 Register scratch) {
1350 assert(ProfileInterpreter, "must be profiling interpreter");
1351
1352 ld(scratch, offset_of_disp, R28_mdx);
1353 add(R28_mdx, scratch, R28_mdx);
1354 }
1355
1356 // Update the method data pointer by the displacement located at the
1357 // offset (reg + offset_of_disp).
1358 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1359 int offset_of_disp,
1360 Register scratch) {
1361 assert(ProfileInterpreter, "must be profiling interpreter");
1362
1363 add(scratch, reg, R28_mdx);
1364 ld(scratch, offset_of_disp, scratch);
1365 add(R28_mdx, scratch, R28_mdx);
1366 }
1367
1368 // Update the method data pointer by a simple constant displacement.
1369 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1370 assert(ProfileInterpreter, "must be profiling interpreter");
1371 addi(R28_mdx, R28_mdx, constant);
1372 }
1373
1374 // Update the method data pointer for a _ret bytecode whose target
1375 // was not among our cached targets.
1376 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1377 Register return_bci) {
1378 assert(ProfileInterpreter, "must be profiling interpreter");
1379
1380 push(state);
1381 assert(return_bci->is_nonvolatile(), "need to protect return_bci");
1382 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1383 pop(state);
1384 }
1385
1386 // Increments the backedge counter.
1387 // Returns backedge counter + invocation counter in Rdst.
1388 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst,
1389 const Register Rtmp1, Register Rscratch) {
1390 assert(UseCompiler, "incrementing must be useful");
1391 assert_different_registers(Rdst, Rtmp1);
1392 const Register invocation_counter = Rtmp1;
1393 const Register counter = Rdst;
1394 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size.");
1395
1396 // Load backedge counter.
1397 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1398 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1399 // Load invocation counter.
1400 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) +
1401 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1402
1403 // Add the delta to the backedge counter.
1404 addi(counter, counter, InvocationCounter::count_increment);
1405
1406 // Mask the invocation counter.
1407 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value);
1408
1409 // Store new counter value.
1410 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1411 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1412 // Return invocation counter + backedge counter.
1413 add(counter, counter, invocation_counter);
1414 }
1415
1416 // Count a taken branch in the bytecodes.
1417 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1418 if (ProfileInterpreter) {
1419 Label profile_continue;
1420
1421 // If no method data exists, go to profile_continue.
1422 test_method_data_pointer(profile_continue);
1423
1424 // We are taking a branch. Increment the taken count.
1425 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count);
1426
1427 // The method data pointer needs to be updated to reflect the new target.
1428 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1429 bind (profile_continue);
1430 }
1431 }
1432
1433 // Count a not-taken branch in the bytecodes.
1434 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) {
1435 if (ProfileInterpreter) {
1436 Label profile_continue;
1437
1438 // If no method data exists, go to profile_continue.
1439 test_method_data_pointer(profile_continue);
1440
1441 // We are taking a branch. Increment the not taken count.
1442 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2);
1443
1444 // The method data pointer needs to be updated to correspond to the
1445 // next bytecode.
1446 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1447 bind (profile_continue);
1448 }
1449 }
1450
1451 // Count a non-virtual call in the bytecodes.
1452 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) {
1453 if (ProfileInterpreter) {
1454 Label profile_continue;
1455
1456 // If no method data exists, go to profile_continue.
1457 test_method_data_pointer(profile_continue);
1458
1459 // We are making a call. Increment the count.
1460 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1461
1462 // The method data pointer needs to be updated to reflect the new target.
1463 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1464 bind (profile_continue);
1465 }
1466 }
1467
1468 // Count a final call in the bytecodes.
1469 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) {
1470 if (ProfileInterpreter) {
1471 Label profile_continue;
1472
1473 // If no method data exists, go to profile_continue.
1474 test_method_data_pointer(profile_continue);
1475
1476 // We are making a call. Increment the count.
1477 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1478
1479 // The method data pointer needs to be updated to reflect the new target.
1480 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1481 bind (profile_continue);
1482 }
1483 }
1484
1485 // Count a virtual call in the bytecodes.
1486 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver,
1487 Register Rscratch1,
1488 Register Rscratch2,
1489 bool receiver_can_be_null) {
1490 if (!ProfileInterpreter) { return; }
1491 Label profile_continue;
1492
1493 // If no method data exists, go to profile_continue.
1494 test_method_data_pointer(profile_continue);
1495
1496 Label skip_receiver_profile;
1497 if (receiver_can_be_null) {
1498 Label not_null;
1499 cmpdi(CCR0, Rreceiver, 0);
1500 bne(CCR0, not_null);
1501 // We are making a call. Increment the count for null receiver.
1502 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2);
1503 b(skip_receiver_profile);
1504 bind(not_null);
1505 }
1506
1507 // Record the receiver type.
1508 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2);
1509 bind(skip_receiver_profile);
1510
1511 // The method data pointer needs to be updated to reflect the new target.
1512 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1513 bind (profile_continue);
1514 }
1515
1516 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) {
1517 if (ProfileInterpreter) {
1518 Label profile_continue;
1519
1520 // If no method data exists, go to profile_continue.
1521 test_method_data_pointer(profile_continue);
1522
1523 int mdp_delta = in_bytes(BitData::bit_data_size());
1524 if (TypeProfileCasts) {
1525 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1526
1527 // Record the object type.
1528 record_klass_in_profile(Rklass, Rscratch1, Rscratch2);
1529 }
1530
1531 // The method data pointer needs to be updated.
1532 update_mdp_by_constant(mdp_delta);
1533
1534 bind (profile_continue);
1535 }
1536 }
1537
1538 // Count a ret in the bytecodes.
1539 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci,
1540 Register scratch1, Register scratch2) {
1541 if (ProfileInterpreter) {
1542 Label profile_continue;
1543 uint row;
1544
1545 // If no method data exists, go to profile_continue.
1546 test_method_data_pointer(profile_continue);
1547
1548 // Update the total ret count.
1549 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 );
1550
1551 for (row = 0; row < RetData::row_limit(); row++) {
1552 Label next_test;
1553
1554 // See if return_bci is equal to bci[n]:
1555 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1);
1556
1557 // return_bci is equal to bci[n]. Increment the count.
1558 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2);
1559
1560 // The method data pointer needs to be updated to reflect the new target.
1561 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1);
1562 b(profile_continue);
1563 bind(next_test);
1564 }
1565
1566 update_mdp_for_ret(state, return_bci);
1567
1568 bind (profile_continue);
1569 }
1570 }
1571
1572 // Count the default case of a switch construct.
1573 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) {
1574 if (ProfileInterpreter) {
1575 Label profile_continue;
1576
1577 // If no method data exists, go to profile_continue.
1578 test_method_data_pointer(profile_continue);
1579
1580 // Update the default case count
1581 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1582 scratch1, scratch2);
1583
1584 // The method data pointer needs to be updated.
1585 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()),
1586 scratch1);
1587
1588 bind (profile_continue);
1589 }
1590 }
1591
1592 // Count the index'th case of a switch construct.
1593 void InterpreterMacroAssembler::profile_switch_case(Register index,
1594 Register scratch1,
1595 Register scratch2,
1596 Register scratch3) {
1597 if (ProfileInterpreter) {
1598 assert_different_registers(index, scratch1, scratch2, scratch3);
1599 Label profile_continue;
1600
1601 // If no method data exists, go to profile_continue.
1602 test_method_data_pointer(profile_continue);
1603
1604 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes().
1605 li(scratch3, in_bytes(MultiBranchData::case_array_offset()));
1606
1607 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works");
1608 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1609 add(scratch1, scratch1, scratch3);
1610
1611 // Update the case count.
1612 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3);
1613
1614 // The method data pointer needs to be updated.
1615 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2);
1616
1617 bind (profile_continue);
1618 }
1619 }
1620
1621 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) {
1622 if (ProfileInterpreter) {
1623 assert_different_registers(Rscratch1, Rscratch2);
1624 Label profile_continue;
1625
1626 // If no method data exists, go to profile_continue.
1627 test_method_data_pointer(profile_continue);
1628
1629 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1);
1630
1631 // The method data pointer needs to be updated.
1632 int mdp_delta = in_bytes(BitData::bit_data_size());
1633 if (TypeProfileCasts) {
1634 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1635 }
1636 update_mdp_by_constant(mdp_delta);
1637
1638 bind (profile_continue);
1639 }
1640 }
1641
1642 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver,
1643 Register Rscratch1, Register Rscratch2) {
1644 assert(ProfileInterpreter, "must be profiling");
1645 assert_different_registers(Rreceiver, Rscratch1, Rscratch2);
1646
1647 Label done;
1648 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done);
1649 bind (done);
1650 }
1651
1652 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1653 Register receiver, Register scratch1, Register scratch2,
1654 int start_row, Label& done) {
1655 if (TypeProfileWidth == 0) {
1656 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1657 return;
1658 }
1659
1660 int last_row = VirtualCallData::row_limit() - 1;
1661 assert(start_row <= last_row, "must be work left to do");
1662 // Test this row for both the receiver and for null.
1663 // Take any of three different outcomes:
1664 // 1. found receiver => increment count and goto done
1665 // 2. found null => keep looking for case 1, maybe allocate this cell
1666 // 3. found something else => keep looking for cases 1 and 2
1667 // Case 3 is handled by a recursive call.
1668 for (int row = start_row; row <= last_row; row++) {
1669 Label next_test;
1670 bool test_for_null_also = (row == start_row);
1671
1672 // See if the receiver is receiver[n].
1673 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1674 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1);
1675 // delayed()->tst(scratch);
1676
1677 // The receiver is receiver[n]. Increment count[n].
1678 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1679 increment_mdp_data_at(count_offset, scratch1, scratch2);
1680 b(done);
1681 bind(next_test);
1682
1683 if (test_for_null_also) {
1684 Label found_null;
1685 // Failed the equality check on receiver[n]... Test for null.
1686 if (start_row == last_row) {
1687 // The only thing left to do is handle the null case.
1688 // Scratch1 contains test_out from test_mdp_data_at.
1689 cmpdi(CCR0, scratch1, 0);
1690 beq(CCR0, found_null);
1691 // Receiver did not match any saved receiver and there is no empty row for it.
1692 // Increment total counter to indicate polymorphic case.
1693 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1694 b(done);
1695 bind(found_null);
1696 break;
1697 }
1698 // Since null is rare, make it be the branch-taken case.
1699 cmpdi(CCR0, scratch1, 0);
1700 beq(CCR0, found_null);
1701
1702 // Put all the "Case 3" tests here.
1703 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done);
1704
1705 // Found a null. Keep searching for a matching receiver,
1706 // but remember that this is an empty (unused) slot.
1707 bind(found_null);
1708 }
1709 }
1710
1711 // In the fall-through case, we found no matching receiver, but we
1712 // observed the receiver[start_row] is null.
1713
1714 // Fill in the receiver field and increment the count.
1715 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1716 set_mdp_data_at(recvr_offset, receiver);
1717 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1718 li(scratch1, DataLayout::counter_increment);
1719 set_mdp_data_at(count_offset, scratch1);
1720 if (start_row > 0) {
1721 b(done);
1722 }
1723 }
1724
1725 // Argument and return type profilig.
1726 // kills: tmp, tmp2, R0, CR0, CR1
1727 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base,
1728 RegisterOrConstant mdo_addr_offs,
1729 Register tmp, Register tmp2) {
1730 Label do_nothing, do_update;
1731
1732 // tmp2 = obj is allowed
1733 assert_different_registers(obj, mdo_addr_base, tmp, R0);
1734 assert_different_registers(tmp2, mdo_addr_base, tmp, R0);
1735 const Register klass = tmp2;
1736
1737 verify_oop(obj);
1738
1739 ld(tmp, mdo_addr_offs, mdo_addr_base);
1740
1741 // Set null_seen if obj is 0.
1742 cmpdi(CCR0, obj, 0);
1743 ori(R0, tmp, TypeEntries::null_seen);
1744 beq(CCR0, do_update);
1745
1746 load_klass(klass, obj);
1747
1748 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
1749 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
1750 cmpd(CCR1, R0, klass);
1751 // Klass seen before, nothing to do (regardless of unknown bit).
1752 //beq(CCR1, do_nothing);
1753
1754 andi_(R0, tmp, TypeEntries::type_unknown);
1755 // Already unknown. Nothing to do anymore.
1756 //bne(CCR0, do_nothing);
1757 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne
1758 beq(CCR0, do_nothing);
1759
1760 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
1761 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1762 beq(CCR0, do_update); // First time here. Set profile type.
1763
1764 // Different than before. Cannot keep accurate profile.
1765 ori(R0, tmp, TypeEntries::type_unknown);
1766
1767 bind(do_update);
1768 // update profile
1769 std(R0, mdo_addr_offs, mdo_addr_base);
1770
1771 align(32, 12);
1772 bind(do_nothing);
1773 }
1774
1775 void InterpreterMacroAssembler::profile_arguments_type(Register callee,
1776 Register tmp1, Register tmp2,
1777 bool is_virtual) {
1778 if (!ProfileInterpreter) {
1779 return;
1780 }
1781
1782 assert_different_registers(callee, tmp1, tmp2, R28_mdx);
1783
1784 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1785 Label profile_continue;
1786
1787 test_method_data_pointer(profile_continue);
1788
1789 int off_to_start = is_virtual ?
1790 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1791
1792 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx);
1793 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1794 bne(CCR0, profile_continue);
1795
1796 if (MethodData::profile_arguments()) {
1797 Label done;
1798 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1799 addi(R28_mdx, R28_mdx, off_to_args);
1800
1801 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1802 if (i > 0 || MethodData::profile_return()) {
1803 // If return value type is profiled we may have no argument to profile.
1804 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1805 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count());
1806 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count());
1807 blt(CCR0, done);
1808 }
1809 ld(tmp1, in_bytes(Method::const_offset()), callee);
1810 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1);
1811 // Stack offset o (zero based) from the start of the argument
1812 // list, for n arguments translates into offset n - o - 1 from
1813 // the end of the argument list. But there's an extra slot at
1814 // the top of the stack. So the offset is n - o from Lesp.
1815 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx);
1816 subf(tmp1, tmp2, tmp1);
1817
1818 sldi(tmp1, tmp1, Interpreter::logStackElementSize);
1819 ldx(tmp1, tmp1, R15_esp);
1820
1821 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1);
1822
1823 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1824 addi(R28_mdx, R28_mdx, to_add);
1825 off_to_args += to_add;
1826 }
1827
1828 if (MethodData::profile_return()) {
1829 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1830 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1831 }
1832
1833 bind(done);
1834
1835 if (MethodData::profile_return()) {
1836 // We're right after the type profile for the last
1837 // argument. tmp1 is the number of cells left in the
1838 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1839 // if there's a return to profile.
1840 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(),
1841 "can't move past ret type");
1842 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size));
1843 add(R28_mdx, tmp1, R28_mdx);
1844 }
1845 } else {
1846 assert(MethodData::profile_return(), "either profile call args or call ret");
1847 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
1848 }
1849
1850 // Mdp points right after the end of the
1851 // CallTypeData/VirtualCallTypeData, right after the cells for the
1852 // return value type if there's one.
1853 align(32, 12);
1854 bind(profile_continue);
1855 }
1856 }
1857
1858 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
1859 assert_different_registers(ret, tmp1, tmp2);
1860 if (ProfileInterpreter && MethodData::profile_return()) {
1861 Label profile_continue;
1862
1863 test_method_data_pointer(profile_continue);
1864
1865 if (MethodData::profile_return_jsr292_only()) {
1866 // If we don't profile all invoke bytecodes we must make sure
1867 // it's a bytecode we indeed profile. We can't go back to the
1868 // beginning of the ProfileData we intend to update to check its
1869 // type because we're right after it and we don't known its
1870 // length.
1871 lbz(tmp1, 0, R14_bcp);
1872 lbz(tmp2, in_bytes(Method::intrinsic_id_offset()), R19_method);
1873 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic);
1874 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle);
1875 cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
1876 cmpwi(CCR1, tmp2, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1877 cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
1878 bne(CCR0, profile_continue);
1879 }
1880
1881 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2);
1882
1883 align(32, 12);
1884 bind(profile_continue);
1885 }
1886 }
1887
1888 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2,
1889 Register tmp3, Register tmp4) {
1890 if (ProfileInterpreter && MethodData::profile_parameters()) {
1891 Label profile_continue, done;
1892
1893 test_method_data_pointer(profile_continue);
1894
1895 // Load the offset of the area within the MDO used for
1896 // parameters. If it's negative we're not profiling any parameters.
1897 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx);
1898 cmpwi(CCR0, tmp1, 0);
1899 blt(CCR0, profile_continue);
1900
1901 // Compute a pointer to the area for parameters from the offset
1902 // and move the pointer to the slot for the last
1903 // parameters. Collect profiling from last parameter down.
1904 // mdo start + parameters offset + array length - 1
1905
1906 // Pointer to the parameter area in the MDO.
1907 const Register mdp = tmp1;
1908 add(mdp, tmp1, R28_mdx);
1909
1910 // Offset of the current profile entry to update.
1911 const Register entry_offset = tmp2;
1912 // entry_offset = array len in number of cells
1913 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp);
1914
1915 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1916 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
1917
1918 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
1919 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size));
1920 // entry_offset in bytes
1921 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1922
1923 Label loop;
1924 align(32, 12);
1925 bind(loop);
1926
1927 // Load offset on the stack from the slot for this parameter.
1928 ld(tmp3, entry_offset, mdp);
1929 sldi(tmp3, tmp3, Interpreter::logStackElementSize);
1930 neg(tmp3, tmp3);
1931 // Read the parameter from the local area.
1932 ldx(tmp3, tmp3, R18_locals);
1933
1934 // Make entry_offset now point to the type field for this parameter.
1935 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1936 assert(type_base > off_base, "unexpected");
1937 addi(entry_offset, entry_offset, type_base - off_base);
1938
1939 // Profile the parameter.
1940 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3);
1941
1942 // Go to next parameter.
1943 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base);
1944 cmpdi(CCR0, entry_offset, off_base + delta);
1945 addi(entry_offset, entry_offset, -delta);
1946 bge(CCR0, loop);
1947
1948 align(32, 12);
1949 bind(profile_continue);
1950 }
1951 }
1952
1953 // Add a monitor (see frame_ppc.hpp).
1954 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) {
1955
1956 // Very-local scratch registers.
1957 const Register esp = Rtemp1;
1958 const Register slot = Rtemp2;
1959
1960 // Extracted monitor_size.
1961 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes();
1962 assert(Assembler::is_aligned((unsigned int)monitor_size,
1963 (unsigned int)frame::alignment_in_bytes),
1964 "size of a monitor must respect alignment of SP");
1965
1966 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor
1967 subf(Rtemp2, esp, R1_SP); // esp contains fp
1968 sradi(Rtemp2, Rtemp2, Interpreter::logStackElementSize);
1969 // Store relativized top_frame_sp
1970 std(Rtemp2, _ijava_state_neg(top_frame_sp), esp); // esp contains fp
1971
1972 // Shuffle expression stack down. Recall that stack_base points
1973 // just above the new expression stack bottom. Old_tos and new_tos
1974 // are used to scan thru the old and new expression stacks.
1975 if (!stack_is_empty) {
1976 Label copy_slot, copy_slot_finished;
1977 const Register n_slots = slot;
1978
1979 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack).
1980 subf(n_slots, esp, R26_monitor);
1981 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy.
1982 assert(LogBytesPerWord == 3, "conflicts assembler instructions");
1983 beq(CCR0, copy_slot_finished); // Nothing to copy.
1984
1985 mtctr(n_slots);
1986
1987 // loop
1988 bind(copy_slot);
1989 ld(slot, 0, esp); // Move expression stack down.
1990 std(slot, -monitor_size, esp); // distance = monitor_size
1991 addi(esp, esp, BytesPerWord);
1992 bdnz(copy_slot);
1993
1994 bind(copy_slot_finished);
1995 }
1996
1997 addi(R15_esp, R15_esp, -monitor_size);
1998 addi(R26_monitor, R26_monitor, -monitor_size);
1999
2000 // Restart interpreter
2001 }
2002
2003 // ============================================================================
2004 // Java locals access
2005
2006 // Load a local variable at index in Rindex into register Rdst_value.
2007 // Also puts address of local into Rdst_address as a service.
2008 // Kills:
2009 // - Rdst_value
2010 // - Rdst_address
2011 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) {
2012 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2013 subf(Rdst_address, Rdst_address, R18_locals);
2014 lwz(Rdst_value, 0, Rdst_address);
2015 }
2016
2017 // Load a local variable at index in Rindex into register Rdst_value.
2018 // Also puts address of local into Rdst_address as a service.
2019 // Kills:
2020 // - Rdst_value
2021 // - Rdst_address
2022 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) {
2023 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2024 subf(Rdst_address, Rdst_address, R18_locals);
2025 ld(Rdst_value, -8, Rdst_address);
2026 }
2027
2028 // Load a local variable at index in Rindex into register Rdst_value.
2029 // Also puts address of local into Rdst_address as a service.
2030 // Input:
2031 // - Rindex: slot nr of local variable
2032 // Kills:
2033 // - Rdst_value
2034 // - Rdst_address
2035 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value,
2036 Register Rdst_address,
2037 Register Rindex) {
2038 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2039 subf(Rdst_address, Rdst_address, R18_locals);
2040 ld(Rdst_value, 0, Rdst_address);
2041 }
2042
2043 // Load a local variable at index in Rindex into register Rdst_value.
2044 // Also puts address of local into Rdst_address as a service.
2045 // Kills:
2046 // - Rdst_value
2047 // - Rdst_address
2048 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value,
2049 Register Rdst_address,
2050 Register Rindex) {
2051 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2052 subf(Rdst_address, Rdst_address, R18_locals);
2053 lfs(Rdst_value, 0, Rdst_address);
2054 }
2055
2056 // Load a local variable at index in Rindex into register Rdst_value.
2057 // Also puts address of local into Rdst_address as a service.
2058 // Kills:
2059 // - Rdst_value
2060 // - Rdst_address
2061 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value,
2062 Register Rdst_address,
2063 Register Rindex) {
2064 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2065 subf(Rdst_address, Rdst_address, R18_locals);
2066 lfd(Rdst_value, -8, Rdst_address);
2067 }
2068
2069 // Store an int value at local variable slot Rindex.
2070 // Kills:
2071 // - Rindex
2072 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) {
2073 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2074 subf(Rindex, Rindex, R18_locals);
2075 stw(Rvalue, 0, Rindex);
2076 }
2077
2078 // Store a long value at local variable slot Rindex.
2079 // Kills:
2080 // - Rindex
2081 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) {
2082 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2083 subf(Rindex, Rindex, R18_locals);
2084 std(Rvalue, -8, Rindex);
2085 }
2086
2087 // Store an oop value at local variable slot Rindex.
2088 // Kills:
2089 // - Rindex
2090 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) {
2091 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2092 subf(Rindex, Rindex, R18_locals);
2093 std(Rvalue, 0, Rindex);
2094 }
2095
2096 // Store an int value at local variable slot Rindex.
2097 // Kills:
2098 // - Rindex
2099 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) {
2100 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2101 subf(Rindex, Rindex, R18_locals);
2102 stfs(Rvalue, 0, Rindex);
2103 }
2104
2105 // Store an int value at local variable slot Rindex.
2106 // Kills:
2107 // - Rindex
2108 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) {
2109 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2110 subf(Rindex, Rindex, R18_locals);
2111 stfd(Rvalue, -8, Rindex);
2112 }
2113
2114 // Read pending exception from thread and jump to interpreter.
2115 // Throw exception entry if one if pending. Fall through otherwise.
2116 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) {
2117 assert_different_registers(Rscratch1, Rscratch2, R3);
2118 Register Rexception = Rscratch1;
2119 Register Rtmp = Rscratch2;
2120 Label Ldone;
2121 // Get pending exception oop.
2122 ld(Rexception, thread_(pending_exception));
2123 cmpdi(CCR0, Rexception, 0);
2124 beq(CCR0, Ldone);
2125 li(Rtmp, 0);
2126 mr_if_needed(R3, Rexception);
2127 std(Rtmp, thread_(pending_exception)); // Clear exception in thread
2128 if (Interpreter::rethrow_exception_entry() != nullptr) {
2129 // Already got entry address.
2130 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry());
2131 } else {
2132 // Dynamically load entry address.
2133 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true);
2134 ld(Rtmp, simm16_rest, Rtmp);
2135 }
2136 mtctr(Rtmp);
2137 save_interpreter_state(Rtmp);
2138 bctr();
2139
2140 align(32, 12);
2141 bind(Ldone);
2142 }
2143
2144 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) {
2145 save_interpreter_state(R11_scratch1);
2146
2147 MacroAssembler::call_VM(oop_result, entry_point, false);
2148
2149 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2150
2151 check_and_handle_popframe(R11_scratch1);
2152 check_and_handle_earlyret(R11_scratch1);
2153 // Now check exceptions manually.
2154 if (check_exceptions) {
2155 check_and_forward_exception(R11_scratch1, R12_scratch2);
2156 }
2157 }
2158
2159 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2160 Register arg_1, bool check_exceptions) {
2161 // ARG1 is reserved for the thread.
2162 mr_if_needed(R4_ARG2, arg_1);
2163 call_VM(oop_result, entry_point, check_exceptions);
2164 }
2165
2166 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2167 Register arg_1, Register arg_2,
2168 bool check_exceptions) {
2169 // ARG1 is reserved for the thread.
2170 mr_if_needed(R4_ARG2, arg_1);
2171 assert(arg_2 != R4_ARG2, "smashed argument");
2172 mr_if_needed(R5_ARG3, arg_2);
2173 call_VM(oop_result, entry_point, check_exceptions);
2174 }
2175
2176 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2177 Register arg_1, Register arg_2, Register arg_3,
2178 bool check_exceptions) {
2179 // ARG1 is reserved for the thread.
2180 mr_if_needed(R4_ARG2, arg_1);
2181 assert(arg_2 != R4_ARG2, "smashed argument");
2182 mr_if_needed(R5_ARG3, arg_2);
2183 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument");
2184 mr_if_needed(R6_ARG4, arg_3);
2185 call_VM(oop_result, entry_point, check_exceptions);
2186 }
2187
2188 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) {
2189 ld(scratch, 0, R1_SP);
2190 subf(R0, scratch, R15_esp);
2191 sradi(R0, R0, Interpreter::logStackElementSize);
2192 std(R0, _ijava_state_neg(esp), scratch);
2193 std(R14_bcp, _ijava_state_neg(bcp), scratch);
2194 subf(R0, scratch, R26_monitor);
2195 sradi(R0, R0, Interpreter::logStackElementSize);
2196 std(R0, _ijava_state_neg(monitors), scratch);
2197 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); }
2198 // Other entries should be unchanged.
2199 }
2200
2201 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only, bool restore_top_frame_sp) {
2202 ld_ptr(scratch, _abi0(callers_sp), R1_SP); // Load frame pointer.
2203 if (restore_top_frame_sp) {
2204 // After thawing the top frame of a continuation we reach here with frame::java_abi.
2205 // therefore we have to restore top_frame_sp before the assertion below.
2206 assert(!bcp_and_mdx_only, "chose other registers");
2207 Register tfsp = R18_locals;
2208 Register scratch2 = R26_monitor;
2209 ld(tfsp, _ijava_state_neg(top_frame_sp), scratch);
2210 // Derelativize top_frame_sp
2211 sldi(tfsp, tfsp, Interpreter::logStackElementSize);
2212 add(tfsp, tfsp, scratch);
2213 resize_frame_absolute(tfsp, scratch2, R0);
2214 }
2215 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception).
2216 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code.
2217 if (!bcp_and_mdx_only) {
2218 // Following ones are Metadata.
2219 ld(R19_method, _ijava_state_neg(method), scratch);
2220 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch);
2221 // Following ones are stack addresses and don't require reload.
2222 // Derelativize esp
2223 ld(R15_esp, _ijava_state_neg(esp), scratch);
2224 sldi(R15_esp, R15_esp, Interpreter::logStackElementSize);
2225 add(R15_esp, R15_esp, scratch);
2226 ld(R18_locals, _ijava_state_neg(locals), scratch);
2227 sldi(R18_locals, R18_locals, Interpreter::logStackElementSize);
2228 add(R18_locals, R18_locals, scratch);
2229 ld(R26_monitor, _ijava_state_neg(monitors), scratch);
2230 // Derelativize monitors
2231 sldi(R26_monitor, R26_monitor, Interpreter::logStackElementSize);
2232 add(R26_monitor, R26_monitor, scratch);
2233 }
2234 #ifdef ASSERT
2235 {
2236 Label Lok;
2237 subf(R0, R1_SP, scratch);
2238 cmpdi(CCR0, R0, frame::top_ijava_frame_abi_size + frame::ijava_state_size);
2239 bge(CCR0, Lok);
2240 stop("frame too small (restore istate)");
2241 bind(Lok);
2242 }
2243 #endif
2244 }
2245
2246 void InterpreterMacroAssembler::get_method_counters(Register method,
2247 Register Rcounters,
2248 Label& skip) {
2249 BLOCK_COMMENT("Load and ev. allocate counter object {");
2250 Label has_counters;
2251 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2252 cmpdi(CCR0, Rcounters, 0);
2253 bne(CCR0, has_counters);
2254 call_VM(noreg, CAST_FROM_FN_PTR(address,
2255 InterpreterRuntime::build_method_counters), method);
2256 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2257 cmpdi(CCR0, Rcounters, 0);
2258 beq(CCR0, skip); // No MethodCounters, OutOfMemory.
2259 BLOCK_COMMENT("} Load and ev. allocate counter object");
2260
2261 bind(has_counters);
2262 }
2263
2264 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters,
2265 Register iv_be_count,
2266 Register Rtmp_r0) {
2267 assert(UseCompiler, "incrementing must be useful");
2268 Register invocation_count = iv_be_count;
2269 Register backedge_count = Rtmp_r0;
2270 int delta = InvocationCounter::count_increment;
2271
2272 // Load each counter in a register.
2273 // ld(inv_counter, Rtmp);
2274 // ld(be_counter, Rtmp2);
2275 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() +
2276 InvocationCounter::counter_offset());
2277 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() +
2278 InvocationCounter::counter_offset());
2279
2280 BLOCK_COMMENT("Increment profiling counters {");
2281
2282 // Load the backedge counter.
2283 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int
2284 // Mask the backedge counter.
2285 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value);
2286
2287 // Load the invocation counter.
2288 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int
2289 // Add the delta to the invocation counter and store the result.
2290 addi(invocation_count, invocation_count, delta);
2291 // Store value.
2292 stw(invocation_count, inv_counter_offset, Rcounters);
2293
2294 // Add invocation counter + backedge counter.
2295 add(iv_be_count, backedge_count, invocation_count);
2296
2297 // Note that this macro must leave the backedge_count + invocation_count in
2298 // register iv_be_count!
2299 BLOCK_COMMENT("} Increment profiling counters");
2300 }
2301
2302 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2303 if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); }
2304 }
2305
2306 // Local helper function for the verify_oop_or_return_address macro.
2307 static bool verify_return_address(Method* m, int bci) {
2308 #ifndef PRODUCT
2309 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci;
2310 // Assume it is a valid return address if it is inside m and is preceded by a jsr.
2311 if (!m->contains(pc)) return false;
2312 address jsr_pc;
2313 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2314 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2315 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2316 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2317 #endif // PRODUCT
2318 return false;
2319 }
2320
2321 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2322 if (VerifyFPU) {
2323 unimplemented("verfiyFPU");
2324 }
2325 }
2326
2327 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2328 if (!VerifyOops) return;
2329
2330 // The VM documentation for the astore[_wide] bytecode allows
2331 // the TOS to be not only an oop but also a return address.
2332 Label test;
2333 Label skip;
2334 // See if it is an address (in the current method):
2335
2336 const int log2_bytecode_size_limit = 16;
2337 srdi_(Rtmp, reg, log2_bytecode_size_limit);
2338 bne(CCR0, test);
2339
2340 address fd = CAST_FROM_FN_PTR(address, verify_return_address);
2341 const int nbytes_save = MacroAssembler::num_volatile_regs * 8;
2342 save_volatile_gprs(R1_SP, -nbytes_save); // except R0
2343 save_LR_CR(Rtmp); // Save in old frame.
2344 push_frame_reg_args(nbytes_save, Rtmp);
2345
2346 load_const_optimized(Rtmp, fd, R0);
2347 mr_if_needed(R4_ARG2, reg);
2348 mr(R3_ARG1, R19_method);
2349 call_c(Rtmp); // call C
2350
2351 pop_frame();
2352 restore_LR_CR(Rtmp);
2353 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
2354 b(skip);
2355
2356 // Perform a more elaborate out-of-line call.
2357 // Not an address; verify it:
2358 bind(test);
2359 verify_oop(reg);
2360 bind(skip);
2361 }
2362
2363 // Inline assembly for:
2364 //
2365 // if (thread is in interp_only_mode) {
2366 // InterpreterRuntime::post_method_entry();
2367 // }
2368 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) ||
2369 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) {
2370 // SharedRuntime::jvmpi_method_entry(method, receiver);
2371 // }
2372 void InterpreterMacroAssembler::notify_method_entry() {
2373 // JVMTI
2374 // Whenever JVMTI puts a thread in interp_only_mode, method
2375 // entry/exit events are sent for that thread to track stack
2376 // depth. If it is possible to enter interp_only_mode we add
2377 // the code to check if the event should be sent.
2378 if (JvmtiExport::can_post_interpreter_events()) {
2379 Label jvmti_post_done;
2380
2381 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2382 cmpwi(CCR0, R0, 0);
2383 beq(CCR0, jvmti_post_done);
2384 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2385
2386 bind(jvmti_post_done);
2387 }
2388 }
2389
2390 // Inline assembly for:
2391 //
2392 // if (thread is in interp_only_mode) {
2393 // // save result
2394 // InterpreterRuntime::post_method_exit();
2395 // // restore result
2396 // }
2397 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) {
2398 // // save result
2399 // SharedRuntime::jvmpi_method_exit();
2400 // // restore result
2401 // }
2402 //
2403 // Native methods have their result stored in d_tmp and l_tmp.
2404 // Java methods have their result stored in the expression stack.
2405 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state,
2406 NotifyMethodExitMode mode, bool check_exceptions) {
2407 // JVMTI
2408 // Whenever JVMTI puts a thread in interp_only_mode, method
2409 // entry/exit events are sent for that thread to track stack
2410 // depth. If it is possible to enter interp_only_mode we add
2411 // the code to check if the event should be sent.
2412 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2413 Label jvmti_post_done;
2414
2415 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2416 cmpwi(CCR0, R0, 0);
2417 beq(CCR0, jvmti_post_done);
2418 if (!is_native_method) { push(state); } // Expose tos to GC.
2419 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), check_exceptions);
2420 if (!is_native_method) { pop(state); }
2421
2422 align(32, 12);
2423 bind(jvmti_post_done);
2424 }
2425
2426 // Dtrace support not implemented.
2427 }