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