1 /*
2 * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2012, 2025 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 "asm/macroAssembler.inline.hpp"
28 #include "gc/shared/barrierSet.hpp"
29 #include "gc/shared/barrierSetAssembler.hpp"
30 #include "interp_masm_ppc.hpp"
31 #include "interpreter/interpreterRuntime.hpp"
32 #include "oops/methodCounters.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/resolvedFieldEntry.hpp"
35 #include "oops/resolvedIndyEntry.hpp"
36 #include "oops/resolvedMethodEntry.hpp"
37 #include "prims/jvmtiExport.hpp"
38 #include "prims/jvmtiThreadState.hpp"
39 #include "runtime/frame.inline.hpp"
40 #include "runtime/safepointMechanism.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/vm_version.hpp"
43 #include "utilities/macros.hpp"
44 #include "utilities/powerOfTwo.hpp"
45
46 // Implementation of InterpreterMacroAssembler.
47
48 // This file specializes the assembler with interpreter-specific macros.
49
50 #ifdef PRODUCT
51 #define BLOCK_COMMENT(str) // nothing
52 #else
53 #define BLOCK_COMMENT(str) block_comment(str)
54 #endif
55
56 void InterpreterMacroAssembler::null_check_throw(Register a, int offset, Register temp_reg) {
57 address exception_entry = Interpreter::throw_NullPointerException_entry();
58 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry);
59 }
60
61 void InterpreterMacroAssembler::load_klass_check_null_throw(Register dst, Register src, Register temp_reg) {
62 null_check_throw(src, oopDesc::klass_offset_in_bytes(), temp_reg);
63 load_klass(dst, src);
64 }
65
66 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) {
67 assert(entry, "Entry must have been generated by now");
68 if (is_within_range_of_b(entry, pc())) {
69 b(entry);
70 } else {
71 load_const_optimized(Rscratch, entry, R0);
72 mtctr(Rscratch);
73 bctr();
74 }
75 }
76
77 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) {
78 Register bytecode = R12_scratch2;
79 if (bcp_incr != 0) {
80 lbzu(bytecode, bcp_incr, R14_bcp);
81 } else {
82 lbz(bytecode, 0, R14_bcp);
83 }
84
85 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state), generate_poll);
86 }
87
88 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
89 // Load current bytecode.
90 Register bytecode = R12_scratch2;
91 lbz(bytecode, 0, R14_bcp);
92 dispatch_Lbyte_code(state, bytecode, table);
93 }
94
95 // Dispatch code executed in the prolog of a bytecode which does not do it's
96 // own dispatch. The dispatch address is computed and placed in R24_dispatch_addr.
97 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
98 Register bytecode = R12_scratch2;
99 lbz(bytecode, bcp_incr, R14_bcp);
100
101 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state));
102
103 sldi(bytecode, bytecode, LogBytesPerWord);
104 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode);
105 }
106
107 // Dispatch code executed in the epilog of a bytecode which does not do it's
108 // own dispatch. The dispatch address in R24_dispatch_addr is used for the
109 // dispatch.
110 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
111 if (bcp_incr) { addi(R14_bcp, R14_bcp, bcp_incr); }
112 mtctr(R24_dispatch_addr);
113 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
114 }
115
116 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
117 assert(scratch_reg != R0, "can't use R0 as scratch_reg here");
118 if (JvmtiExport::can_pop_frame()) {
119 Label L;
120
121 // Check the "pending popframe condition" flag in the current thread.
122 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
123
124 // Initiate popframe handling only if it is not already being
125 // processed. If the flag has the popframe_processing bit set, it
126 // means that this code is called *during* popframe handling - we
127 // don't want to reenter.
128 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit);
129 beq(CR0, L);
130
131 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit);
132 bne(CR0, L);
133
134 // Call the Interpreter::remove_activation_preserving_args_entry()
135 // func to get the address of the same-named entrypoint in the
136 // generated interpreter code.
137 call_c(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
138
139 // Jump to Interpreter::_remove_activation_preserving_args_entry.
140 mtctr(R3_RET);
141 bctr();
142
143 align(32, 12);
144 bind(L);
145 }
146 }
147
148 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
149 const Register Rthr_state_addr = scratch_reg;
150 if (JvmtiExport::can_force_early_return()) {
151 Label Lno_early_ret;
152 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
153 cmpdi(CR0, Rthr_state_addr, 0);
154 beq(CR0, Lno_early_ret);
155
156 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr);
157 cmpwi(CR0, R0, JvmtiThreadState::earlyret_pending);
158 bne(CR0, Lno_early_ret);
159
160 // Jump to Interpreter::_earlyret_entry.
161 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr);
162 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry));
163 mtlr(R3_RET);
164 blr();
165
166 align(32, 12);
167 bind(Lno_early_ret);
168 }
169 }
170
171 void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) {
172 const Register RjvmtiState = Rscratch1;
173 const Register Rscratch2 = R0;
174
175 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
176 li(Rscratch2, 0);
177
178 switch (state) {
179 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
180 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
181 break;
182 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
183 break;
184 case btos: // fall through
185 case ztos: // fall through
186 case ctos: // fall through
187 case stos: // fall through
188 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
189 break;
190 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
191 break;
192 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
193 break;
194 case vtos: break;
195 default : ShouldNotReachHere();
196 }
197
198 // Clean up tos value in the jvmti thread state.
199 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
200 // Set tos state field to illegal value.
201 li(Rscratch2, ilgl);
202 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState);
203 }
204
205 // Common code to dispatch and dispatch_only.
206 // Dispatch value in Lbyte_code and increment Lbcp.
207
208 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) {
209 address table_base = (address)Interpreter::dispatch_table((TosState)0);
210 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base;
211 if (is_simm16(table_offs)) {
212 addi(dst, R25_templateTableBase, (int)table_offs);
213 } else {
214 load_const_optimized(dst, table, R0);
215 }
216 }
217
218 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode,
219 address* table, bool generate_poll) {
220 assert_different_registers(bytecode, R11_scratch1);
221
222 // Calc dispatch table address.
223 load_dispatch_table(R11_scratch1, table);
224
225 if (generate_poll) {
226 address *sfpt_tbl = Interpreter::safept_table(state);
227 if (table != sfpt_tbl) {
228 Label dispatch;
229 ld(R0, in_bytes(JavaThread::polling_word_offset()), R16_thread);
230 // Armed page has poll_bit set, if poll bit is cleared just continue.
231 andi_(R0, R0, SafepointMechanism::poll_bit());
232 beq(CR0, dispatch);
233 load_dispatch_table(R11_scratch1, sfpt_tbl);
234 align(32, 16);
235 bind(dispatch);
236 }
237 }
238
239 sldi(R12_scratch2, bytecode, LogBytesPerWord);
240 ldx(R11_scratch1, R11_scratch1, R12_scratch2);
241
242 // Jump off!
243 mtctr(R11_scratch1);
244 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
245 }
246
247 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) {
248 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize);
249 ldx(Rrecv_dst, Rrecv_dst, R15_esp);
250 }
251
252 // helpers for expression stack
253
254 void InterpreterMacroAssembler::pop_i(Register r) {
255 lwzu(r, Interpreter::stackElementSize, R15_esp);
256 }
257
258 void InterpreterMacroAssembler::pop_ptr(Register r) {
259 ldu(r, Interpreter::stackElementSize, R15_esp);
260 }
261
262 void InterpreterMacroAssembler::pop_l(Register r) {
263 ld(r, Interpreter::stackElementSize, R15_esp);
264 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
265 }
266
267 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
268 lfsu(f, Interpreter::stackElementSize, R15_esp);
269 }
270
271 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
272 lfd(f, Interpreter::stackElementSize, R15_esp);
273 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
274 }
275
276 void InterpreterMacroAssembler::push_i(Register r) {
277 stw(r, 0, R15_esp);
278 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
279 }
280
281 void InterpreterMacroAssembler::push_ptr(Register r) {
282 std(r, 0, R15_esp);
283 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
284 }
285
286 void InterpreterMacroAssembler::push_l(Register r) {
287 // Clear unused slot.
288 load_const_optimized(R0, 0L);
289 std(R0, 0, R15_esp);
290 std(r, - Interpreter::stackElementSize, R15_esp);
291 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
292 }
293
294 void InterpreterMacroAssembler::push_f(FloatRegister f) {
295 stfs(f, 0, R15_esp);
296 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
297 }
298
299 void InterpreterMacroAssembler::push_d(FloatRegister f) {
300 stfd(f, - Interpreter::stackElementSize, R15_esp);
301 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
302 }
303
304 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) {
305 std(first, 0, R15_esp);
306 std(second, -Interpreter::stackElementSize, R15_esp);
307 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
308 }
309
310 void InterpreterMacroAssembler::move_l_to_d(Register l, FloatRegister d) {
311 if (VM_Version::has_mtfprd()) {
312 mtfprd(d, l);
313 } else {
314 std(l, 0, R15_esp);
315 lfd(d, 0, R15_esp);
316 }
317 }
318
319 void InterpreterMacroAssembler::move_d_to_l(FloatRegister d, Register l) {
320 if (VM_Version::has_mtfprd()) {
321 mffprd(l, d);
322 } else {
323 stfd(d, 0, R15_esp);
324 ld(l, 0, R15_esp);
325 }
326 }
327
328 void InterpreterMacroAssembler::push(TosState state) {
329 switch (state) {
330 case atos: push_ptr(); break;
331 case btos:
332 case ztos:
333 case ctos:
334 case stos:
335 case itos: push_i(); break;
336 case ltos: push_l(); break;
337 case ftos: push_f(); break;
338 case dtos: push_d(); break;
339 case vtos: /* nothing to do */ break;
340 default : ShouldNotReachHere();
341 }
342 }
343
344 void InterpreterMacroAssembler::pop(TosState state) {
345 switch (state) {
346 case atos: pop_ptr(); break;
347 case btos:
348 case ztos:
349 case ctos:
350 case stos:
351 case itos: pop_i(); break;
352 case ltos: pop_l(); break;
353 case ftos: pop_f(); break;
354 case dtos: pop_d(); break;
355 case vtos: /* nothing to do */ break;
356 default : ShouldNotReachHere();
357 }
358 verify_oop(R17_tos, state);
359 }
360
361 void InterpreterMacroAssembler::empty_expression_stack() {
362 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
363 }
364
365 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset,
366 Register Rdst,
367 signedOrNot is_signed) {
368 #if defined(VM_LITTLE_ENDIAN)
369 if (bcp_offset) {
370 load_const_optimized(Rdst, bcp_offset);
371 lhbrx(Rdst, R14_bcp, Rdst);
372 } else {
373 lhbrx(Rdst, R14_bcp);
374 }
375 if (is_signed == Signed) {
376 extsh(Rdst, Rdst);
377 }
378 #else
379 // Read Java big endian format.
380 if (is_signed == Signed) {
381 lha(Rdst, bcp_offset, R14_bcp);
382 } else {
383 lhz(Rdst, bcp_offset, R14_bcp);
384 }
385 #endif
386 }
387
388 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset,
389 Register Rdst,
390 signedOrNot is_signed) {
391 #if defined(VM_LITTLE_ENDIAN)
392 if (bcp_offset) {
393 load_const_optimized(Rdst, bcp_offset);
394 lwbrx(Rdst, R14_bcp, Rdst);
395 } else {
396 lwbrx(Rdst, R14_bcp);
397 }
398 if (is_signed == Signed) {
399 extsw(Rdst, Rdst);
400 }
401 #else
402 // Read Java big endian format.
403 if (bcp_offset & 3) { // Offset unaligned?
404 load_const_optimized(Rdst, bcp_offset);
405 if (is_signed == Signed) {
406 lwax(Rdst, R14_bcp, Rdst);
407 } else {
408 lwzx(Rdst, R14_bcp, Rdst);
409 }
410 } else {
411 if (is_signed == Signed) {
412 lwa(Rdst, bcp_offset, R14_bcp);
413 } else {
414 lwz(Rdst, bcp_offset, R14_bcp);
415 }
416 }
417 #endif
418 }
419
420
421 // Load the constant pool cache index from the bytecode stream.
422 //
423 // Kills / writes:
424 // - Rdst, Rscratch
425 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset,
426 size_t index_size) {
427 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
428 // Cache index is always in the native format, courtesy of Rewriter.
429 if (index_size == sizeof(u2)) {
430 lhz(Rdst, bcp_offset, R14_bcp);
431 } else if (index_size == sizeof(u4)) {
432 if (bcp_offset & 3) {
433 load_const_optimized(Rdst, bcp_offset);
434 lwax(Rdst, R14_bcp, Rdst);
435 } else {
436 lwa(Rdst, bcp_offset, R14_bcp);
437 }
438 } else if (index_size == sizeof(u1)) {
439 lbz(Rdst, bcp_offset, R14_bcp);
440 } else {
441 ShouldNotReachHere();
442 }
443 // Rdst now contains cp cache index.
444 }
445
446 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format)
447 // from (Rsrc)+offset.
448 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset,
449 signedOrNot is_signed) {
450 #if defined(VM_LITTLE_ENDIAN)
451 if (offset) {
452 load_const_optimized(Rdst, offset);
453 lwbrx(Rdst, Rdst, Rsrc);
454 } else {
455 lwbrx(Rdst, Rsrc);
456 }
457 if (is_signed == Signed) {
458 extsw(Rdst, Rdst);
459 }
460 #else
461 if (is_signed == Signed) {
462 lwa(Rdst, offset, Rsrc);
463 } else {
464 lwz(Rdst, offset, Rsrc);
465 }
466 #endif
467 }
468
469 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
470 // Get index out of bytecode pointer
471 get_cache_index_at_bcp(index, 1, sizeof(u4));
472
473 // Get address of invokedynamic array
474 ld_ptr(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset()), R27_constPoolCache);
475 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
476 sldi(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
477 addi(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
478 add(cache, cache, index);
479 }
480
481 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
482 // Get index out of bytecode pointer
483 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
484 // Take shortcut if the size is a power of 2
485 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
486 // Scale index by power of 2
487 sldi(index, index, log2i_exact(sizeof(ResolvedFieldEntry)));
488 } else {
489 // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
490 mulli(index, index, sizeof(ResolvedFieldEntry));
491 }
492 // Get address of field entries array
493 ld_ptr(cache, in_bytes(ConstantPoolCache::field_entries_offset()), R27_constPoolCache);
494 addi(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
495 add(cache, cache, index);
496 }
497
498 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
499 // Get index out of bytecode pointer
500 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
501 // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
502 mulli(index, index, sizeof(ResolvedMethodEntry));
503
504 // Get address of field entries array
505 ld_ptr(cache, ConstantPoolCache::method_entries_offset(), R27_constPoolCache);
506 addi(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
507 add(cache, cache, index); // method_entries + base_offset + scaled index
508 }
509
510 // Load object from cpool->resolved_references(index).
511 // Kills:
512 // - index
513 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index,
514 Register tmp1, Register tmp2,
515 Label *L_handle_null) {
516 assert_different_registers(result, index, tmp1, tmp2);
517 assert(index->is_nonvolatile(), "needs to survive C-call in resolve_oop_handle");
518 get_constant_pool(result);
519
520 // Convert from field index to resolved_references() index and from
521 // word index to byte offset. Since this is a java object, it can be compressed.
522 sldi(index, index, LogBytesPerHeapOop);
523 // Load pointer for resolved_references[] objArray.
524 ld(result, ConstantPool::cache_offset(), result);
525 ld(result, ConstantPoolCache::resolved_references_offset(), result);
526 resolve_oop_handle(result, tmp1, tmp2, MacroAssembler::PRESERVATION_NONE);
527 #ifdef ASSERT
528 Label index_ok;
529 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result);
530 sldi(R0, R0, LogBytesPerHeapOop);
531 cmpd(CR0, index, R0);
532 blt(CR0, index_ok);
533 stop("resolved reference index out of bounds");
534 bind(index_ok);
535 #endif
536 // Add in the index.
537 add(result, index, result);
538 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result,
539 tmp1, tmp2,
540 MacroAssembler::PRESERVATION_NONE,
541 0, L_handle_null);
542 }
543
544 // load cpool->resolved_klass_at(index)
545 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool, Register Roffset, Register Rklass) {
546 // int value = *(Rcpool->int_at_addr(which));
547 // int resolved_klass_index = extract_low_short_from_int(value);
548 add(Roffset, Rcpool, Roffset);
549 #if defined(VM_LITTLE_ENDIAN)
550 lhz(Roffset, sizeof(ConstantPool), Roffset); // Roffset = resolved_klass_index
551 #else
552 lhz(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index
553 #endif
554
555 ld(Rklass, ConstantPool::resolved_klasses_offset(), Rcpool); // Rklass = Rcpool->_resolved_klasses
556
557 sldi(Roffset, Roffset, LogBytesPerWord);
558 addi(Roffset, Roffset, Array<Klass*>::base_offset_in_bytes());
559 isync(); // Order load of instance Klass wrt. tags.
560 ldx(Rklass, Rklass, Roffset);
561 }
562
563 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
564 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2.
565 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1,
566 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) {
567 // Profile the not-null value's klass.
568 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2);
569 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
570 }
571
572 // Separate these two to allow for delay slot in middle.
573 // These are used to do a test and full jump to exception-throwing code.
574
575 // Check that index is in range for array, then shift index by index_shift,
576 // and put arrayOop + shifted_index into res.
577 // Note: res is still shy of address by array offset into object.
578
579 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex,
580 int index_shift, Register Rtmp, Register Rres) {
581 // Check that index is in range for array, then shift index by index_shift,
582 // and put arrayOop + shifted_index into res.
583 // Note: res is still shy of address by array offset into object.
584 // Kills:
585 // - Rindex
586 // Writes:
587 // - Rres: Address that corresponds to the array index if check was successful.
588 verify_oop(Rarray);
589 const Register Rlength = R0;
590 const Register RsxtIndex = Rtmp;
591 Label LisNull, LnotOOR;
592
593 // Array nullcheck
594 if (!ImplicitNullChecks) {
595 cmpdi(CR0, Rarray, 0);
596 beq(CR0, LisNull);
597 } else {
598 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex);
599 }
600
601 // Rindex might contain garbage in upper bits (remember that we don't sign extend
602 // during integer arithmetic operations). So kill them and put value into same register
603 // where ArrayIndexOutOfBounds would expect the index in.
604 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit
605
606 // Index check
607 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray);
608 cmplw(CR0, Rindex, Rlength);
609 sldi(RsxtIndex, RsxtIndex, index_shift);
610 blt(CR0, LnotOOR);
611 // Index should be in R17_tos, array should be in R4_ARG2.
612 mr_if_needed(R17_tos, Rindex);
613 mr_if_needed(R4_ARG2, Rarray);
614 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
615 mtctr(Rtmp);
616 bctr();
617
618 if (!ImplicitNullChecks) {
619 bind(LisNull);
620 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry);
621 mtctr(Rtmp);
622 bctr();
623 }
624
625 align(32, 16);
626 bind(LnotOOR);
627
628 // Calc address
629 add(Rres, RsxtIndex, Rarray);
630 }
631
632 void InterpreterMacroAssembler::index_check(Register array, Register index,
633 int index_shift, Register tmp, Register res) {
634 // pop array
635 pop_ptr(array);
636
637 // check array
638 index_check_without_pop(array, index, index_shift, tmp, res);
639 }
640
641 void InterpreterMacroAssembler::get_const(Register Rdst) {
642 ld(Rdst, in_bytes(Method::const_offset()), R19_method);
643 }
644
645 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
646 get_const(Rdst);
647 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
648 }
649
650 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
651 get_constant_pool(Rdst);
652 ld(Rdst, ConstantPool::cache_offset(), Rdst);
653 }
654
655 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
656 get_constant_pool(Rcpool);
657 ld(Rtags, ConstantPool::tags_offset(), Rcpool);
658 }
659
660 // Unlock if synchronized method.
661 //
662 // Unlock the receiver if this is a synchronized method.
663 // Unlock any Java monitors from synchronized blocks.
664 //
665 // If there are locked Java monitors
666 // If throw_monitor_exception
667 // throws IllegalMonitorStateException
668 // Else if install_monitor_exception
669 // installs IllegalMonitorStateException
670 // Else
671 // no error processing
672 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
673 bool throw_monitor_exception,
674 bool install_monitor_exception) {
675 Label Lunlocked, Lno_unlock;
676 {
677 Register Rdo_not_unlock_flag = R11_scratch1;
678 Register Raccess_flags = R12_scratch2;
679
680 // Check if synchronized method or unlocking prevented by
681 // JavaThread::do_not_unlock_if_synchronized flag.
682 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
683 lhz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method);
684 li(R0, 0);
685 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag
686
687 push(state);
688
689 // Skip if we don't have to unlock.
690 testbitdi(CR0, R0, Raccess_flags, JVM_ACC_SYNCHRONIZED_BIT);
691 beq(CR0, Lunlocked);
692
693 cmpwi(CR0, Rdo_not_unlock_flag, 0);
694 bne(CR0, Lno_unlock);
695 }
696
697 // Unlock
698 {
699 Register Rmonitor_base = R11_scratch1;
700
701 Label Lunlock;
702 // If it's still locked, everything is ok, unlock it.
703 ld(Rmonitor_base, 0, R1_SP);
704 addi(Rmonitor_base, Rmonitor_base,
705 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
706
707 ld(R0, BasicObjectLock::obj_offset(), Rmonitor_base);
708 cmpdi(CR0, R0, 0);
709 bne(CR0, Lunlock);
710
711 // If it's already unlocked, throw exception.
712 if (throw_monitor_exception) {
713 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
714 should_not_reach_here();
715 } else {
716 if (install_monitor_exception) {
717 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
718 b(Lunlocked);
719 }
720 }
721
722 bind(Lunlock);
723 unlock_object(Rmonitor_base);
724 }
725
726 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not.
727 bind(Lunlocked);
728 {
729 Label Lexception, Lrestart;
730 Register Rcurrent_obj_addr = R11_scratch1;
731 const int delta = frame::interpreter_frame_monitor_size_in_bytes();
732 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords");
733
734 bind(Lrestart);
735 // Set up search loop: Calc num of iterations.
736 {
737 Register Riterations = R12_scratch2;
738 Register Rmonitor_base = Rcurrent_obj_addr;
739 ld(Rmonitor_base, 0, R1_SP);
740 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base
741
742 subf_(Riterations, R26_monitor, Rmonitor_base);
743 ble(CR0, Lno_unlock);
744
745 addi(Rcurrent_obj_addr, Rmonitor_base,
746 in_bytes(BasicObjectLock::obj_offset()) - frame::interpreter_frame_monitor_size_in_bytes());
747 // Check if any monitor is on stack, bail out if not
748 srdi(Riterations, Riterations, exact_log2(delta));
749 mtctr(Riterations);
750 }
751
752 // The search loop: Look for locked monitors.
753 {
754 const Register Rcurrent_obj = R0;
755 Label Lloop;
756
757 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
758 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
759 bind(Lloop);
760
761 // Check if current entry is used.
762 cmpdi(CR0, Rcurrent_obj, 0);
763 bne(CR0, Lexception);
764 // Preload next iteration's compare value.
765 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
766 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
767 bdnz(Lloop);
768 }
769 // Fell through: Everything's unlocked => finish.
770 b(Lno_unlock);
771
772 // An object is still locked => need to throw exception.
773 bind(Lexception);
774 if (throw_monitor_exception) {
775 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
776 should_not_reach_here();
777 } else {
778 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
779 // Unlock does not block, so don't have to worry about the frame.
780 Register Rmonitor_addr = R11_scratch1;
781 addi(Rmonitor_addr, Rcurrent_obj_addr, -in_bytes(BasicObjectLock::obj_offset()) + delta);
782 unlock_object(Rmonitor_addr);
783 if (install_monitor_exception) {
784 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
785 }
786 b(Lrestart);
787 }
788 }
789
790 align(32, 12);
791 bind(Lno_unlock);
792 pop(state);
793 }
794
795 // Support function for remove_activation & Co.
796 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc,
797 Register Rscratch1, Register Rscratch2) {
798 // Pop interpreter frame.
799 ld(Rscratch1, 0, R1_SP); // *SP
800 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp
801 ld(Rscratch2, 0, Rscratch1); // **SP
802 if (return_pc!=noreg) {
803 ld(return_pc, _abi0(lr), Rscratch1); // LR
804 }
805
806 // Merge top frames.
807 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP
808 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP
809 }
810
811 void InterpreterMacroAssembler::narrow(Register result) {
812 Register ret_type = R11_scratch1;
813 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
814 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1);
815
816 Label notBool, notByte, notChar, done;
817
818 // common case first
819 cmpwi(CR0, ret_type, T_INT);
820 beq(CR0, done);
821
822 cmpwi(CR0, ret_type, T_BOOLEAN);
823 bne(CR0, notBool);
824 andi(result, result, 0x1);
825 b(done);
826
827 bind(notBool);
828 cmpwi(CR0, ret_type, T_BYTE);
829 bne(CR0, notByte);
830 extsb(result, result);
831 b(done);
832
833 bind(notByte);
834 cmpwi(CR0, ret_type, T_CHAR);
835 bne(CR0, notChar);
836 andi(result, result, 0xffff);
837 b(done);
838
839 bind(notChar);
840 // cmpwi(CR0, ret_type, T_SHORT); // all that's left
841 // bne(CR0, done);
842 extsh(result, result);
843
844 // Nothing to do for T_INT
845 bind(done);
846 }
847
848 // Remove activation.
849 //
850 // Apply stack watermark barrier.
851 // Unlock the receiver if this is a synchronized method.
852 // Unlock any Java monitors from synchronized blocks.
853 // Remove the activation from the stack.
854 //
855 // If there are locked Java monitors
856 // If throw_monitor_exception
857 // throws IllegalMonitorStateException
858 // Else if install_monitor_exception
859 // installs IllegalMonitorStateException
860 // Else
861 // no error processing
862 void InterpreterMacroAssembler::remove_activation(TosState state,
863 bool throw_monitor_exception,
864 bool install_monitor_exception) {
865 BLOCK_COMMENT("remove_activation {");
866
867 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
868 // that would normally not be safe to use. Such bad returns into unsafe territory of
869 // the stack, will call InterpreterRuntime::at_unwind.
870 Label slow_path;
871 Label fast_path;
872 safepoint_poll(slow_path, R11_scratch1, true /* at_return */, false /* in_nmethod */);
873 b(fast_path);
874 bind(slow_path);
875 push(state);
876 set_last_Java_frame(R1_SP, noreg);
877 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), R16_thread);
878 reset_last_Java_frame();
879 pop(state);
880 align(32);
881 bind(fast_path);
882
883 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
884
885 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
886 notify_method_exit(false, state, NotifyJVMTI, true);
887
888 BLOCK_COMMENT("reserved_stack_check:");
889 if (StackReservedPages > 0) {
890 // Test if reserved zone needs to be enabled.
891 Label no_reserved_zone_enabling;
892
893 // check if already enabled - if so no re-enabling needed
894 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
895 lwz(R0, in_bytes(JavaThread::stack_guard_state_offset()), R16_thread);
896 cmpwi(CR0, R0, StackOverflow::stack_guard_enabled);
897 beq_predict_taken(CR0, no_reserved_zone_enabling);
898
899 // Compare frame pointers. There is no good stack pointer, as with stack
900 // frame compression we can get different SPs when we do calls. A subsequent
901 // call could have a smaller SP, so that this compare succeeds for an
902 // inner call of the method annotated with ReservedStack.
903 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread);
904 ld_ptr(R11_scratch1, _abi0(callers_sp), R1_SP); // Load frame pointer.
905 cmpld(CR0, R11_scratch1, R0);
906 blt_predict_taken(CR0, no_reserved_zone_enabling);
907
908 // Enable reserved zone again, throw stack overflow exception.
909 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread);
910 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
911
912 should_not_reach_here();
913
914 bind(no_reserved_zone_enabling);
915 }
916
917 verify_oop(R17_tos, state);
918
919 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
920 mtlr(R0);
921 pop_cont_fastpath();
922 BLOCK_COMMENT("} remove_activation");
923 }
924
925 // Lock object
926 //
927 // Registers alive
928 // monitor - Address of the BasicObjectLock to be used for locking,
929 // which must be initialized with the object to lock.
930 // object - Address of the object to be locked.
931 //
932 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
933 if (LockingMode == LM_MONITOR) {
934 call_VM_preemptable(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
935 } else {
936 // template code (for LM_LEGACY):
937 //
938 // markWord displaced_header = obj->mark().set_unlocked();
939 // monitor->lock()->set_displaced_header(displaced_header);
940 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
941 // // We stored the monitor address into the object's mark word.
942 // } else if (THREAD->is_lock_owned((address)displaced_header))
943 // // Simple recursive case.
944 // monitor->lock()->set_displaced_header(nullptr);
945 // } else {
946 // // Slow path.
947 // InterpreterRuntime::monitorenter(THREAD, monitor);
948 // }
949
950 const Register header = R7_ARG5;
951 const Register object_mark_addr = R8_ARG6;
952 const Register current_header = R9_ARG7;
953 const Register tmp = R10_ARG8;
954
955 Label count_locking, done, slow_case, cas_failed;
956
957 assert_different_registers(header, object_mark_addr, current_header, tmp);
958
959 // markWord displaced_header = obj->mark().set_unlocked();
960
961 if (DiagnoseSyncOnValueBasedClasses != 0) {
962 load_klass(tmp, object);
963 lbz(tmp, in_bytes(Klass::misc_flags_offset()), tmp);
964 testbitdi(CR0, R0, tmp, exact_log2(KlassFlags::_misc_is_value_based_class));
965 bne(CR0, slow_case);
966 }
967
968 if (LockingMode == LM_LIGHTWEIGHT) {
969 lightweight_lock(monitor, object, header, tmp, slow_case);
970 b(done);
971 } else if (LockingMode == LM_LEGACY) {
972 // Load markWord from object into header.
973 ld(header, oopDesc::mark_offset_in_bytes(), object);
974
975 // Set displaced_header to be (markWord of object | UNLOCK_VALUE).
976 ori(header, header, markWord::unlocked_value);
977
978 // monitor->lock()->set_displaced_header(displaced_header);
979 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
980 const int mark_offset = lock_offset +
981 BasicLock::displaced_header_offset_in_bytes();
982
983 // Initialize the box (Must happen before we update the object mark!).
984 std(header, mark_offset, monitor);
985
986 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
987
988 // Store stack address of the BasicObjectLock (this is monitor) into object.
989 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
990
991 // Must fence, otherwise, preceding store(s) may float below cmpxchg.
992 // CmpxchgX sets CR0 to cmpX(current, displaced).
993 cmpxchgd(/*flag=*/CR0,
994 /*current_value=*/current_header,
995 /*compare_value=*/header, /*exchange_value=*/monitor,
996 /*where=*/object_mark_addr,
997 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq,
998 MacroAssembler::cmpxchgx_hint_acquire_lock(),
999 noreg,
1000 &cas_failed,
1001 /*check without membar and ldarx first*/true);
1002
1003 // If the compare-and-exchange succeeded, then we found an unlocked
1004 // object and we have now locked it.
1005 b(count_locking);
1006 bind(cas_failed);
1007
1008 // } else if (THREAD->is_lock_owned((address)displaced_header))
1009 // // Simple recursive case.
1010 // monitor->lock()->set_displaced_header(nullptr);
1011
1012 // We did not see an unlocked object so try the fast recursive case.
1013
1014 // Check if owner is self by comparing the value in the markWord of object
1015 // (current_header) with the stack pointer.
1016 sub(current_header, current_header, R1_SP);
1017
1018 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1019 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markWord::lock_mask_in_place);
1020
1021 and_(R0/*==0?*/, current_header, tmp);
1022 // If condition is true we are done and hence we can store 0 in the displaced
1023 // header indicating it is a recursive lock.
1024 bne(CR0, slow_case);
1025 std(R0/*==0!*/, mark_offset, monitor);
1026 b(count_locking);
1027 }
1028
1029 // } else {
1030 // // Slow path.
1031 // InterpreterRuntime::monitorenter(THREAD, monitor);
1032
1033 // None of the above fast optimizations worked so we have to get into the
1034 // slow case of monitor enter.
1035 bind(slow_case);
1036 call_VM_preemptable(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
1037 // }
1038
1039 if (LockingMode == LM_LEGACY) {
1040 b(done);
1041 align(32, 12);
1042 bind(count_locking);
1043 inc_held_monitor_count(current_header /*tmp*/);
1044 }
1045 bind(done);
1046 }
1047 }
1048
1049 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1050 //
1051 // Registers alive
1052 // monitor - Address of the BasicObjectLock to be used for locking,
1053 // which must be initialized with the object to lock.
1054 //
1055 // Throw IllegalMonitorException if object is not locked by current thread.
1056 void InterpreterMacroAssembler::unlock_object(Register monitor) {
1057 if (LockingMode == LM_MONITOR) {
1058 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1059 } else {
1060
1061 // template code (for LM_LEGACY):
1062 //
1063 // if ((displaced_header = monitor->displaced_header()) == nullptr) {
1064 // // Recursive unlock. Mark the monitor unlocked by setting the object field to null.
1065 // monitor->set_obj(nullptr);
1066 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1067 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1068 // monitor->set_obj(nullptr);
1069 // } else {
1070 // // Slow path.
1071 // InterpreterRuntime::monitorexit(monitor);
1072 // }
1073
1074 const Register object = R7_ARG5;
1075 const Register header = R8_ARG6;
1076 const Register object_mark_addr = R9_ARG7;
1077 const Register current_header = R10_ARG8;
1078
1079 Label free_slot;
1080 Label slow_case;
1081
1082 assert_different_registers(object, header, object_mark_addr, current_header);
1083
1084 if (LockingMode != LM_LIGHTWEIGHT) {
1085 // Test first if we are in the fast recursive case.
1086 ld(header, in_bytes(BasicObjectLock::lock_offset()) +
1087 BasicLock::displaced_header_offset_in_bytes(), monitor);
1088
1089 // If the displaced header is zero, we have a recursive unlock.
1090 cmpdi(CR0, header, 0);
1091 beq(CR0, free_slot); // recursive unlock
1092 }
1093
1094 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1095 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1096 // monitor->set_obj(nullptr);
1097
1098 // If we still have a lightweight lock, unlock the object and be done.
1099
1100 // The object address from the monitor is in object.
1101 ld(object, in_bytes(BasicObjectLock::obj_offset()), monitor);
1102
1103 if (LockingMode == LM_LIGHTWEIGHT) {
1104 lightweight_unlock(object, header, slow_case);
1105 } else {
1106 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
1107
1108 // We have the displaced header in displaced_header. If the lock is still
1109 // lightweight, it will contain the monitor address and we'll store the
1110 // displaced header back into the object's mark word.
1111 // CmpxchgX sets CR0 to cmpX(current, monitor).
1112 cmpxchgd(/*flag=*/CR0,
1113 /*current_value=*/current_header,
1114 /*compare_value=*/monitor, /*exchange_value=*/header,
1115 /*where=*/object_mark_addr,
1116 MacroAssembler::MemBarRel,
1117 MacroAssembler::cmpxchgx_hint_release_lock(),
1118 noreg,
1119 &slow_case);
1120 }
1121 b(free_slot);
1122
1123 // } else {
1124 // // Slow path.
1125 // InterpreterRuntime::monitorexit(monitor);
1126
1127 // The lock has been converted into a heavy lock and hence
1128 // we need to get into the slow case.
1129 bind(slow_case);
1130 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1131 // }
1132
1133 Label done;
1134 b(done); // Monitor register may be overwritten! Runtime has already freed the slot.
1135
1136 // Exchange worked, do monitor->set_obj(nullptr);
1137 align(32, 12);
1138 bind(free_slot);
1139 li(R0, 0);
1140 std(R0, in_bytes(BasicObjectLock::obj_offset()), monitor);
1141 if (LockingMode == LM_LEGACY) {
1142 dec_held_monitor_count(current_header /*tmp*/);
1143 }
1144 bind(done);
1145 }
1146 }
1147
1148 // Load compiled (i2c) or interpreter entry when calling from interpreted and
1149 // do the call. Centralized so that all interpreter calls will do the same actions.
1150 // If jvmti single stepping is on for a thread we must not call compiled code.
1151 //
1152 // Input:
1153 // - Rtarget_method: method to call
1154 // - Rret_addr: return address
1155 // - 2 scratch regs
1156 //
1157 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr,
1158 Register Rscratch1, Register Rscratch2) {
1159 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr);
1160 // Assume we want to go compiled if available.
1161 const Register Rtarget_addr = Rscratch1;
1162 const Register Rinterp_only = Rscratch2;
1163
1164 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method);
1165
1166 if (JvmtiExport::can_post_interpreter_events()) {
1167 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
1168
1169 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
1170 // compiled code in threads for which the event is enabled. Check here for
1171 // interp_only_mode if these events CAN be enabled.
1172 Label done;
1173 cmpwi(CR0, Rinterp_only, 0);
1174 beq(CR0, done);
1175 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method);
1176 align(32, 12);
1177 bind(done);
1178 }
1179
1180 #ifdef ASSERT
1181 {
1182 Label Lok;
1183 cmpdi(CR0, Rtarget_addr, 0);
1184 bne(CR0, Lok);
1185 stop("null entry point");
1186 bind(Lok);
1187 }
1188 #endif // ASSERT
1189
1190 mr(R21_sender_SP, R1_SP);
1191
1192 // Calc a precise SP for the call. The SP value we calculated in
1193 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space
1194 // if esp is not max. Also, the i2c adapter extends the stack space without restoring
1195 // our pre-calced value, so repeating calls via i2c would result in stack overflow.
1196 // Since esp already points to an empty slot, we just have to sub 1 additional slot
1197 // to meet the abi scratch requirements.
1198 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in
1199 // the return entry of the interpreter.
1200 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::top_ijava_frame_abi_size);
1201 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address
1202 resize_frame_absolute(Rscratch2, Rscratch2, R0);
1203
1204 mr_if_needed(R19_method, Rtarget_method);
1205 mtctr(Rtarget_addr);
1206 mtlr(Rret_addr);
1207
1208 save_interpreter_state(Rscratch2);
1209 #ifdef ASSERT
1210 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp
1211 sldi(Rscratch1, Rscratch1, Interpreter::logStackElementSize);
1212 add(Rscratch1, Rscratch1, Rscratch2); // Rscratch2 contains fp
1213 // Compare sender_sp with the derelativized top_frame_sp
1214 cmpd(CR0, R21_sender_SP, Rscratch1);
1215 asm_assert_eq("top_frame_sp incorrect");
1216 #endif
1217
1218 bctr();
1219 }
1220
1221 // Set the method data pointer for the current bcp.
1222 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1223 assert(ProfileInterpreter, "must be profiling interpreter");
1224 Label get_continue;
1225 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method);
1226 test_method_data_pointer(get_continue);
1227 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp);
1228
1229 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1230 add(R28_mdx, R28_mdx, R3_RET);
1231 bind(get_continue);
1232 }
1233
1234 // Test ImethodDataPtr. If it is null, continue at the specified label.
1235 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1236 assert(ProfileInterpreter, "must be profiling interpreter");
1237 cmpdi(CR0, R28_mdx, 0);
1238 beq(CR0, zero_continue);
1239 }
1240
1241 void InterpreterMacroAssembler::verify_method_data_pointer() {
1242 assert(ProfileInterpreter, "must be profiling interpreter");
1243 #ifdef ASSERT
1244 Label verify_continue;
1245 test_method_data_pointer(verify_continue);
1246
1247 // If the mdp is valid, it will point to a DataLayout header which is
1248 // consistent with the bcp. The converse is highly probable also.
1249 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx);
1250 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method);
1251 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1252 add(R11_scratch1, R12_scratch2, R12_scratch2);
1253 cmpd(CR0, R11_scratch1, R14_bcp);
1254 beq(CR0, verify_continue);
1255
1256 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx);
1257
1258 bind(verify_continue);
1259 #endif
1260 }
1261
1262 // Store a value at some constant offset from the method data pointer.
1263 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1264 assert(ProfileInterpreter, "must be profiling interpreter");
1265
1266 std(value, constant, R28_mdx);
1267 }
1268
1269 // Increment the value at some constant offset from the method data pointer.
1270 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1271 Register counter_addr,
1272 Register Rbumped_count,
1273 bool decrement) {
1274 // Locate the counter at a fixed offset from the mdp:
1275 addi(counter_addr, R28_mdx, constant);
1276 increment_mdp_data_at(counter_addr, Rbumped_count, decrement);
1277 }
1278
1279 // Increment the value at some non-fixed (reg + constant) offset from
1280 // the method data pointer.
1281 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1282 int constant,
1283 Register scratch,
1284 Register Rbumped_count,
1285 bool decrement) {
1286 // Add the constant to reg to get the offset.
1287 add(scratch, R28_mdx, reg);
1288 // Then calculate the counter address.
1289 addi(scratch, scratch, constant);
1290 increment_mdp_data_at(scratch, Rbumped_count, decrement);
1291 }
1292
1293 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr,
1294 Register Rbumped_count,
1295 bool decrement) {
1296 assert(ProfileInterpreter, "must be profiling interpreter");
1297
1298 // Load the counter.
1299 ld(Rbumped_count, 0, counter_addr);
1300
1301 if (decrement) {
1302 // Decrement the register. Set condition codes.
1303 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment);
1304 // Store the decremented counter, if it is still negative.
1305 std(Rbumped_count, 0, counter_addr);
1306 // Note: add/sub overflow check are not ported, since 64 bit
1307 // calculation should never overflow.
1308 } else {
1309 // Increment the register. Set carry flag.
1310 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment);
1311 // Store the incremented counter.
1312 std(Rbumped_count, 0, counter_addr);
1313 }
1314 }
1315
1316 // Set a flag value at the current method data pointer position.
1317 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1318 Register scratch) {
1319 assert(ProfileInterpreter, "must be profiling interpreter");
1320 // Load the data header.
1321 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1322 // Set the flag.
1323 ori(scratch, scratch, flag_constant);
1324 // Store the modified header.
1325 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1326 }
1327
1328 // Test the location at some offset from the method data pointer.
1329 // If it is not equal to value, branch to the not_equal_continue Label.
1330 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1331 Register value,
1332 Label& not_equal_continue,
1333 Register test_out) {
1334 assert(ProfileInterpreter, "must be profiling interpreter");
1335
1336 ld(test_out, offset, R28_mdx);
1337 cmpd(CR0, value, test_out);
1338 bne(CR0, not_equal_continue);
1339 }
1340
1341 // Update the method data pointer by the displacement located at some fixed
1342 // offset from the method data pointer.
1343 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1344 Register scratch) {
1345 assert(ProfileInterpreter, "must be profiling interpreter");
1346
1347 ld(scratch, offset_of_disp, R28_mdx);
1348 add(R28_mdx, scratch, R28_mdx);
1349 }
1350
1351 // Update the method data pointer by the displacement located at the
1352 // offset (reg + offset_of_disp).
1353 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1354 int offset_of_disp,
1355 Register scratch) {
1356 assert(ProfileInterpreter, "must be profiling interpreter");
1357
1358 add(scratch, reg, R28_mdx);
1359 ld(scratch, offset_of_disp, scratch);
1360 add(R28_mdx, scratch, R28_mdx);
1361 }
1362
1363 // Update the method data pointer by a simple constant displacement.
1364 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1365 assert(ProfileInterpreter, "must be profiling interpreter");
1366 addi(R28_mdx, R28_mdx, constant);
1367 }
1368
1369 // Update the method data pointer for a _ret bytecode whose target
1370 // was not among our cached targets.
1371 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1372 Register return_bci) {
1373 assert(ProfileInterpreter, "must be profiling interpreter");
1374
1375 push(state);
1376 assert(return_bci->is_nonvolatile(), "need to protect return_bci");
1377 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1378 pop(state);
1379 }
1380
1381 // Increments the backedge counter.
1382 // Returns backedge counter + invocation counter in Rdst.
1383 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst,
1384 const Register Rtmp1, Register Rscratch) {
1385 assert(UseCompiler, "incrementing must be useful");
1386 assert_different_registers(Rdst, Rtmp1);
1387 const Register invocation_counter = Rtmp1;
1388 const Register counter = Rdst;
1389 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size.");
1390
1391 // Load backedge counter.
1392 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1393 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1394 // Load invocation counter.
1395 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) +
1396 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1397
1398 // Add the delta to the backedge counter.
1399 addi(counter, counter, InvocationCounter::count_increment);
1400
1401 // Mask the invocation counter.
1402 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value);
1403
1404 // Store new counter value.
1405 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1406 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1407 // Return invocation counter + backedge counter.
1408 add(counter, counter, invocation_counter);
1409 }
1410
1411 // Count a taken branch in the bytecodes.
1412 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1413 if (ProfileInterpreter) {
1414 Label profile_continue;
1415
1416 // If no method data exists, go to profile_continue.
1417 test_method_data_pointer(profile_continue);
1418
1419 // We are taking a branch. Increment the taken count.
1420 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count);
1421
1422 // The method data pointer needs to be updated to reflect the new target.
1423 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1424 bind (profile_continue);
1425 }
1426 }
1427
1428 // Count a not-taken branch in the bytecodes.
1429 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) {
1430 if (ProfileInterpreter) {
1431 Label profile_continue;
1432
1433 // If no method data exists, go to profile_continue.
1434 test_method_data_pointer(profile_continue);
1435
1436 // We are taking a branch. Increment the not taken count.
1437 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2);
1438
1439 // The method data pointer needs to be updated to correspond to the
1440 // next bytecode.
1441 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1442 bind (profile_continue);
1443 }
1444 }
1445
1446 // Count a non-virtual call in the bytecodes.
1447 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) {
1448 if (ProfileInterpreter) {
1449 Label profile_continue;
1450
1451 // If no method data exists, go to profile_continue.
1452 test_method_data_pointer(profile_continue);
1453
1454 // We are making a call. Increment the count.
1455 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1456
1457 // The method data pointer needs to be updated to reflect the new target.
1458 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1459 bind (profile_continue);
1460 }
1461 }
1462
1463 // Count a final call in the bytecodes.
1464 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) {
1465 if (ProfileInterpreter) {
1466 Label profile_continue;
1467
1468 // If no method data exists, go to profile_continue.
1469 test_method_data_pointer(profile_continue);
1470
1471 // We are making a call. Increment the count.
1472 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1473
1474 // The method data pointer needs to be updated to reflect the new target.
1475 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1476 bind (profile_continue);
1477 }
1478 }
1479
1480 // Count a virtual call in the bytecodes.
1481 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver,
1482 Register Rscratch1,
1483 Register Rscratch2,
1484 bool receiver_can_be_null) {
1485 if (!ProfileInterpreter) { return; }
1486 Label profile_continue;
1487
1488 // If no method data exists, go to profile_continue.
1489 test_method_data_pointer(profile_continue);
1490
1491 Label skip_receiver_profile;
1492 if (receiver_can_be_null) {
1493 Label not_null;
1494 cmpdi(CR0, Rreceiver, 0);
1495 bne(CR0, not_null);
1496 // We are making a call. Increment the count for null receiver.
1497 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2);
1498 b(skip_receiver_profile);
1499 bind(not_null);
1500 }
1501
1502 // Record the receiver type.
1503 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2);
1504 bind(skip_receiver_profile);
1505
1506 // The method data pointer needs to be updated to reflect the new target.
1507 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1508 bind (profile_continue);
1509 }
1510
1511 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) {
1512 if (ProfileInterpreter) {
1513 Label profile_continue;
1514
1515 // If no method data exists, go to profile_continue.
1516 test_method_data_pointer(profile_continue);
1517
1518 int mdp_delta = in_bytes(BitData::bit_data_size());
1519 if (TypeProfileCasts) {
1520 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1521
1522 // Record the object type.
1523 record_klass_in_profile(Rklass, Rscratch1, Rscratch2);
1524 }
1525
1526 // The method data pointer needs to be updated.
1527 update_mdp_by_constant(mdp_delta);
1528
1529 bind (profile_continue);
1530 }
1531 }
1532
1533 // Count a ret in the bytecodes.
1534 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci,
1535 Register scratch1, Register scratch2) {
1536 if (ProfileInterpreter) {
1537 Label profile_continue;
1538 uint row;
1539
1540 // If no method data exists, go to profile_continue.
1541 test_method_data_pointer(profile_continue);
1542
1543 // Update the total ret count.
1544 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 );
1545
1546 for (row = 0; row < RetData::row_limit(); row++) {
1547 Label next_test;
1548
1549 // See if return_bci is equal to bci[n]:
1550 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1);
1551
1552 // return_bci is equal to bci[n]. Increment the count.
1553 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2);
1554
1555 // The method data pointer needs to be updated to reflect the new target.
1556 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1);
1557 b(profile_continue);
1558 bind(next_test);
1559 }
1560
1561 update_mdp_for_ret(state, return_bci);
1562
1563 bind (profile_continue);
1564 }
1565 }
1566
1567 // Count the default case of a switch construct.
1568 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) {
1569 if (ProfileInterpreter) {
1570 Label profile_continue;
1571
1572 // If no method data exists, go to profile_continue.
1573 test_method_data_pointer(profile_continue);
1574
1575 // Update the default case count
1576 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1577 scratch1, scratch2);
1578
1579 // The method data pointer needs to be updated.
1580 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()),
1581 scratch1);
1582
1583 bind (profile_continue);
1584 }
1585 }
1586
1587 // Count the index'th case of a switch construct.
1588 void InterpreterMacroAssembler::profile_switch_case(Register index,
1589 Register scratch1,
1590 Register scratch2,
1591 Register scratch3) {
1592 if (ProfileInterpreter) {
1593 assert_different_registers(index, scratch1, scratch2, scratch3);
1594 Label profile_continue;
1595
1596 // If no method data exists, go to profile_continue.
1597 test_method_data_pointer(profile_continue);
1598
1599 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes().
1600 li(scratch3, in_bytes(MultiBranchData::case_array_offset()));
1601
1602 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works");
1603 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1604 add(scratch1, scratch1, scratch3);
1605
1606 // Update the case count.
1607 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3);
1608
1609 // The method data pointer needs to be updated.
1610 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2);
1611
1612 bind (profile_continue);
1613 }
1614 }
1615
1616 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) {
1617 if (ProfileInterpreter) {
1618 assert_different_registers(Rscratch1, Rscratch2);
1619 Label profile_continue;
1620
1621 // If no method data exists, go to profile_continue.
1622 test_method_data_pointer(profile_continue);
1623
1624 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1);
1625
1626 // The method data pointer needs to be updated.
1627 int mdp_delta = in_bytes(BitData::bit_data_size());
1628 if (TypeProfileCasts) {
1629 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1630 }
1631 update_mdp_by_constant(mdp_delta);
1632
1633 bind (profile_continue);
1634 }
1635 }
1636
1637 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver,
1638 Register Rscratch1, Register Rscratch2) {
1639 assert(ProfileInterpreter, "must be profiling");
1640 assert_different_registers(Rreceiver, Rscratch1, Rscratch2);
1641
1642 Label done;
1643 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done);
1644 bind (done);
1645 }
1646
1647 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1648 Register receiver, Register scratch1, Register scratch2,
1649 int start_row, Label& done) {
1650 if (TypeProfileWidth == 0) {
1651 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1652 return;
1653 }
1654
1655 int last_row = VirtualCallData::row_limit() - 1;
1656 assert(start_row <= last_row, "must be work left to do");
1657 // Test this row for both the receiver and for null.
1658 // Take any of three different outcomes:
1659 // 1. found receiver => increment count and goto done
1660 // 2. found null => keep looking for case 1, maybe allocate this cell
1661 // 3. found something else => keep looking for cases 1 and 2
1662 // Case 3 is handled by a recursive call.
1663 for (int row = start_row; row <= last_row; row++) {
1664 Label next_test;
1665 bool test_for_null_also = (row == start_row);
1666
1667 // See if the receiver is receiver[n].
1668 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1669 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1);
1670 // delayed()->tst(scratch);
1671
1672 // The receiver is receiver[n]. Increment count[n].
1673 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1674 increment_mdp_data_at(count_offset, scratch1, scratch2);
1675 b(done);
1676 bind(next_test);
1677
1678 if (test_for_null_also) {
1679 Label found_null;
1680 // Failed the equality check on receiver[n]... Test for null.
1681 if (start_row == last_row) {
1682 // The only thing left to do is handle the null case.
1683 // Scratch1 contains test_out from test_mdp_data_at.
1684 cmpdi(CR0, scratch1, 0);
1685 beq(CR0, found_null);
1686 // Receiver did not match any saved receiver and there is no empty row for it.
1687 // Increment total counter to indicate polymorphic case.
1688 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1689 b(done);
1690 bind(found_null);
1691 break;
1692 }
1693 // Since null is rare, make it be the branch-taken case.
1694 cmpdi(CR0, scratch1, 0);
1695 beq(CR0, found_null);
1696
1697 // Put all the "Case 3" tests here.
1698 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done);
1699
1700 // Found a null. Keep searching for a matching receiver,
1701 // but remember that this is an empty (unused) slot.
1702 bind(found_null);
1703 }
1704 }
1705
1706 // In the fall-through case, we found no matching receiver, but we
1707 // observed the receiver[start_row] is null.
1708
1709 // Fill in the receiver field and increment the count.
1710 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1711 set_mdp_data_at(recvr_offset, receiver);
1712 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1713 li(scratch1, DataLayout::counter_increment);
1714 set_mdp_data_at(count_offset, scratch1);
1715 if (start_row > 0) {
1716 b(done);
1717 }
1718 }
1719
1720 // Argument and return type profilig.
1721 // kills: tmp, tmp2, R0, CR0, CR1
1722 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base,
1723 RegisterOrConstant mdo_addr_offs,
1724 Register tmp, Register tmp2) {
1725 Label do_nothing, do_update;
1726
1727 // tmp2 = obj is allowed
1728 assert_different_registers(obj, mdo_addr_base, tmp, R0);
1729 assert_different_registers(tmp2, mdo_addr_base, tmp, R0);
1730 const Register klass = tmp2;
1731
1732 verify_oop(obj);
1733
1734 ld(tmp, mdo_addr_offs, mdo_addr_base);
1735
1736 // Set null_seen if obj is 0.
1737 cmpdi(CR0, obj, 0);
1738 ori(R0, tmp, TypeEntries::null_seen);
1739 beq(CR0, do_update);
1740
1741 load_klass(klass, obj);
1742
1743 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
1744 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
1745 cmpd(CR1, R0, klass);
1746 // Klass seen before, nothing to do (regardless of unknown bit).
1747 //beq(CR1, do_nothing);
1748
1749 andi_(R0, tmp, TypeEntries::type_unknown);
1750 // Already unknown. Nothing to do anymore.
1751 //bne(CR0, do_nothing);
1752 crorc(CR0, Assembler::equal, CR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne
1753 beq(CR0, do_nothing);
1754
1755 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
1756 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1757 beq(CR0, do_update); // First time here. Set profile type.
1758
1759 // Different than before. Cannot keep accurate profile.
1760 ori(R0, tmp, TypeEntries::type_unknown);
1761
1762 bind(do_update);
1763 // update profile
1764 std(R0, mdo_addr_offs, mdo_addr_base);
1765
1766 align(32, 12);
1767 bind(do_nothing);
1768 }
1769
1770 void InterpreterMacroAssembler::profile_arguments_type(Register callee,
1771 Register tmp1, Register tmp2,
1772 bool is_virtual) {
1773 if (!ProfileInterpreter) {
1774 return;
1775 }
1776
1777 assert_different_registers(callee, tmp1, tmp2, R28_mdx);
1778
1779 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1780 Label profile_continue;
1781
1782 test_method_data_pointer(profile_continue);
1783
1784 int off_to_start = is_virtual ?
1785 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1786
1787 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx);
1788 cmpwi(CR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1789 bne(CR0, profile_continue);
1790
1791 if (MethodData::profile_arguments()) {
1792 Label done;
1793 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1794 addi(R28_mdx, R28_mdx, off_to_args);
1795
1796 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1797 if (i > 0 || MethodData::profile_return()) {
1798 // If return value type is profiled we may have no argument to profile.
1799 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1800 cmpdi(CR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count());
1801 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count());
1802 blt(CR0, done);
1803 }
1804 ld(tmp1, in_bytes(Method::const_offset()), callee);
1805 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1);
1806 // Stack offset o (zero based) from the start of the argument
1807 // list, for n arguments translates into offset n - o - 1 from
1808 // the end of the argument list. But there's an extra slot at
1809 // the top of the stack. So the offset is n - o from Lesp.
1810 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx);
1811 subf(tmp1, tmp2, tmp1);
1812
1813 sldi(tmp1, tmp1, Interpreter::logStackElementSize);
1814 ldx(tmp1, tmp1, R15_esp);
1815
1816 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1);
1817
1818 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1819 addi(R28_mdx, R28_mdx, to_add);
1820 off_to_args += to_add;
1821 }
1822
1823 if (MethodData::profile_return()) {
1824 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1825 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1826 }
1827
1828 bind(done);
1829
1830 if (MethodData::profile_return()) {
1831 // We're right after the type profile for the last
1832 // argument. tmp1 is the number of cells left in the
1833 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1834 // if there's a return to profile.
1835 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(),
1836 "can't move past ret type");
1837 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size));
1838 add(R28_mdx, tmp1, R28_mdx);
1839 }
1840 } else {
1841 assert(MethodData::profile_return(), "either profile call args or call ret");
1842 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
1843 }
1844
1845 // Mdp points right after the end of the
1846 // CallTypeData/VirtualCallTypeData, right after the cells for the
1847 // return value type if there's one.
1848 align(32, 12);
1849 bind(profile_continue);
1850 }
1851 }
1852
1853 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
1854 assert_different_registers(ret, tmp1, tmp2);
1855 if (ProfileInterpreter && MethodData::profile_return()) {
1856 Label profile_continue;
1857
1858 test_method_data_pointer(profile_continue);
1859
1860 if (MethodData::profile_return_jsr292_only()) {
1861 // If we don't profile all invoke bytecodes we must make sure
1862 // it's a bytecode we indeed profile. We can't go back to the
1863 // beginning of the ProfileData we intend to update to check its
1864 // type because we're right after it and we don't known its
1865 // length.
1866 lbz(tmp1, 0, R14_bcp);
1867 lbz(tmp2, in_bytes(Method::intrinsic_id_offset()), R19_method);
1868 cmpwi(CR0, tmp1, Bytecodes::_invokedynamic);
1869 cmpwi(CR1, tmp1, Bytecodes::_invokehandle);
1870 cror(CR0, Assembler::equal, CR1, Assembler::equal);
1871 cmpwi(CR1, tmp2, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1872 cror(CR0, Assembler::equal, CR1, Assembler::equal);
1873 bne(CR0, profile_continue);
1874 }
1875
1876 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2);
1877
1878 align(32, 12);
1879 bind(profile_continue);
1880 }
1881 }
1882
1883 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2,
1884 Register tmp3, Register tmp4) {
1885 if (ProfileInterpreter && MethodData::profile_parameters()) {
1886 Label profile_continue, done;
1887
1888 test_method_data_pointer(profile_continue);
1889
1890 // Load the offset of the area within the MDO used for
1891 // parameters. If it's negative we're not profiling any parameters.
1892 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx);
1893 cmpwi(CR0, tmp1, 0);
1894 blt(CR0, profile_continue);
1895
1896 // Compute a pointer to the area for parameters from the offset
1897 // and move the pointer to the slot for the last
1898 // parameters. Collect profiling from last parameter down.
1899 // mdo start + parameters offset + array length - 1
1900
1901 // Pointer to the parameter area in the MDO.
1902 const Register mdp = tmp1;
1903 add(mdp, tmp1, R28_mdx);
1904
1905 // Offset of the current profile entry to update.
1906 const Register entry_offset = tmp2;
1907 // entry_offset = array len in number of cells
1908 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp);
1909
1910 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1911 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
1912
1913 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
1914 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size));
1915 // entry_offset in bytes
1916 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1917
1918 Label loop;
1919 align(32, 12);
1920 bind(loop);
1921
1922 // Load offset on the stack from the slot for this parameter.
1923 ld(tmp3, entry_offset, mdp);
1924 sldi(tmp3, tmp3, Interpreter::logStackElementSize);
1925 neg(tmp3, tmp3);
1926 // Read the parameter from the local area.
1927 ldx(tmp3, tmp3, R18_locals);
1928
1929 // Make entry_offset now point to the type field for this parameter.
1930 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1931 assert(type_base > off_base, "unexpected");
1932 addi(entry_offset, entry_offset, type_base - off_base);
1933
1934 // Profile the parameter.
1935 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3);
1936
1937 // Go to next parameter.
1938 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base);
1939 cmpdi(CR0, entry_offset, off_base + delta);
1940 addi(entry_offset, entry_offset, -delta);
1941 bge(CR0, loop);
1942
1943 align(32, 12);
1944 bind(profile_continue);
1945 }
1946 }
1947
1948 // Add a monitor (see frame_ppc.hpp).
1949 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) {
1950
1951 // Very-local scratch registers.
1952 const Register esp = Rtemp1;
1953 const Register slot = Rtemp2;
1954
1955 // Extracted monitor_size.
1956 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes();
1957 assert(Assembler::is_aligned((unsigned int)monitor_size,
1958 (unsigned int)frame::alignment_in_bytes),
1959 "size of a monitor must respect alignment of SP");
1960
1961 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor
1962 subf(Rtemp2, esp, R1_SP); // esp contains fp
1963 sradi(Rtemp2, Rtemp2, Interpreter::logStackElementSize);
1964 // Store relativized top_frame_sp
1965 std(Rtemp2, _ijava_state_neg(top_frame_sp), esp); // esp contains fp
1966
1967 // Shuffle expression stack down. Recall that stack_base points
1968 // just above the new expression stack bottom. Old_tos and new_tos
1969 // are used to scan thru the old and new expression stacks.
1970 if (!stack_is_empty) {
1971 Label copy_slot, copy_slot_finished;
1972 const Register n_slots = slot;
1973
1974 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack).
1975 subf(n_slots, esp, R26_monitor);
1976 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy.
1977 assert(LogBytesPerWord == 3, "conflicts assembler instructions");
1978 beq(CR0, copy_slot_finished); // Nothing to copy.
1979
1980 mtctr(n_slots);
1981
1982 // loop
1983 bind(copy_slot);
1984 ld(slot, 0, esp); // Move expression stack down.
1985 std(slot, -monitor_size, esp); // distance = monitor_size
1986 addi(esp, esp, BytesPerWord);
1987 bdnz(copy_slot);
1988
1989 bind(copy_slot_finished);
1990 }
1991
1992 addi(R15_esp, R15_esp, -monitor_size);
1993 addi(R26_monitor, R26_monitor, -monitor_size);
1994
1995 // Restart interpreter
1996 }
1997
1998 // ============================================================================
1999 // Java locals access
2000
2001 // Load a local variable at index in Rindex into register Rdst_value.
2002 // Also puts address of local into Rdst_address as a service.
2003 // Kills:
2004 // - Rdst_value
2005 // - Rdst_address
2006 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) {
2007 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2008 subf(Rdst_address, Rdst_address, R18_locals);
2009 lwz(Rdst_value, 0, Rdst_address);
2010 }
2011
2012 // Load a local variable at index in Rindex into register Rdst_value.
2013 // Also puts address of local into Rdst_address as a service.
2014 // Kills:
2015 // - Rdst_value
2016 // - Rdst_address
2017 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) {
2018 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2019 subf(Rdst_address, Rdst_address, R18_locals);
2020 ld(Rdst_value, -8, Rdst_address);
2021 }
2022
2023 // Load a local variable at index in Rindex into register Rdst_value.
2024 // Also puts address of local into Rdst_address as a service.
2025 // Input:
2026 // - Rindex: slot nr of local variable
2027 // Kills:
2028 // - Rdst_value
2029 // - Rdst_address
2030 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value,
2031 Register Rdst_address,
2032 Register Rindex) {
2033 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2034 subf(Rdst_address, Rdst_address, R18_locals);
2035 ld(Rdst_value, 0, Rdst_address);
2036 }
2037
2038 // Load a local variable at index in Rindex into register Rdst_value.
2039 // Also puts address of local into Rdst_address as a service.
2040 // Kills:
2041 // - Rdst_value
2042 // - Rdst_address
2043 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value,
2044 Register Rdst_address,
2045 Register Rindex) {
2046 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2047 subf(Rdst_address, Rdst_address, R18_locals);
2048 lfs(Rdst_value, 0, Rdst_address);
2049 }
2050
2051 // Load a local variable at index in Rindex into register Rdst_value.
2052 // Also puts address of local into Rdst_address as a service.
2053 // Kills:
2054 // - Rdst_value
2055 // - Rdst_address
2056 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value,
2057 Register Rdst_address,
2058 Register Rindex) {
2059 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2060 subf(Rdst_address, Rdst_address, R18_locals);
2061 lfd(Rdst_value, -8, Rdst_address);
2062 }
2063
2064 // Store an int value at local variable slot Rindex.
2065 // Kills:
2066 // - Rindex
2067 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) {
2068 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2069 subf(Rindex, Rindex, R18_locals);
2070 stw(Rvalue, 0, Rindex);
2071 }
2072
2073 // Store a long value at local variable slot Rindex.
2074 // Kills:
2075 // - Rindex
2076 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) {
2077 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2078 subf(Rindex, Rindex, R18_locals);
2079 std(Rvalue, -8, Rindex);
2080 }
2081
2082 // Store an oop value at local variable slot Rindex.
2083 // Kills:
2084 // - Rindex
2085 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) {
2086 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2087 subf(Rindex, Rindex, R18_locals);
2088 std(Rvalue, 0, Rindex);
2089 }
2090
2091 // Store an int value at local variable slot Rindex.
2092 // Kills:
2093 // - Rindex
2094 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) {
2095 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2096 subf(Rindex, Rindex, R18_locals);
2097 stfs(Rvalue, 0, Rindex);
2098 }
2099
2100 // Store an int value at local variable slot Rindex.
2101 // Kills:
2102 // - Rindex
2103 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) {
2104 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2105 subf(Rindex, Rindex, R18_locals);
2106 stfd(Rvalue, -8, Rindex);
2107 }
2108
2109 // Read pending exception from thread and jump to interpreter.
2110 // Throw exception entry if one if pending. Fall through otherwise.
2111 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) {
2112 assert_different_registers(Rscratch1, Rscratch2, R3);
2113 Register Rexception = Rscratch1;
2114 Register Rtmp = Rscratch2;
2115 Label Ldone;
2116 // Get pending exception oop.
2117 ld(Rexception, thread_(pending_exception));
2118 cmpdi(CR0, Rexception, 0);
2119 beq(CR0, Ldone);
2120 li(Rtmp, 0);
2121 mr_if_needed(R3, Rexception);
2122 std(Rtmp, thread_(pending_exception)); // Clear exception in thread
2123 if (Interpreter::rethrow_exception_entry() != nullptr) {
2124 // Already got entry address.
2125 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry());
2126 } else {
2127 // Dynamically load entry address.
2128 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true);
2129 ld(Rtmp, simm16_rest, Rtmp);
2130 }
2131 mtctr(Rtmp);
2132 save_interpreter_state(Rtmp);
2133 bctr();
2134
2135 align(32, 12);
2136 bind(Ldone);
2137 }
2138
2139 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions, Label* last_java_pc) {
2140 save_interpreter_state(R11_scratch1);
2141
2142 MacroAssembler::call_VM(oop_result, entry_point, false /*check_exceptions*/, last_java_pc);
2143
2144 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2145
2146 check_and_handle_popframe(R11_scratch1);
2147 check_and_handle_earlyret(R11_scratch1);
2148 // Now check exceptions manually.
2149 if (check_exceptions) {
2150 check_and_forward_exception(R11_scratch1, R12_scratch2);
2151 }
2152 }
2153
2154 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2155 Register arg_1, bool check_exceptions) {
2156 // ARG1 is reserved for the thread.
2157 mr_if_needed(R4_ARG2, arg_1);
2158 call_VM(oop_result, entry_point, check_exceptions);
2159 }
2160
2161 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result, address entry_point,
2162 Register arg_1, bool check_exceptions) {
2163 if (!Continuations::enabled()) {
2164 call_VM(oop_result, entry_point, arg_1, check_exceptions);
2165 return;
2166 }
2167
2168 Label resume_pc, not_preempted;
2169
2170 DEBUG_ONLY(ld(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread));
2171 DEBUG_ONLY(cmpdi(CR0, R0, 0));
2172 asm_assert_eq("Should not have alternate return address set");
2173
2174 // Preserve 2 registers
2175 assert(nonvolatile_accross_vthread_preemtion(R31) && nonvolatile_accross_vthread_preemtion(R22), "");
2176 ld(R3_ARG1, _abi0(callers_sp), R1_SP); // load FP
2177 std(R31, _ijava_state_neg(lresult), R3_ARG1);
2178 std(R22, _ijava_state_neg(fresult), R3_ARG1);
2179
2180 // We set resume_pc as last java pc. It will be saved if the vthread gets preempted.
2181 // Later execution will continue right there.
2182 mr_if_needed(R4_ARG2, arg_1);
2183 push_cont_fastpath();
2184 call_VM(oop_result, entry_point, false /*check_exceptions*/, &resume_pc /* last_java_pc */);
2185 pop_cont_fastpath();
2186
2187 // Jump to handler if the call was preempted
2188 ld(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
2189 cmpdi(CR0, R0, 0);
2190 beq(CR0, not_preempted);
2191 mtlr(R0);
2192 li(R0, 0);
2193 std(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
2194 blr();
2195
2196 bind(resume_pc); // Location to resume execution
2197 restore_after_resume(noreg /* fp */);
2198 bind(not_preempted);
2199 }
2200
2201 void InterpreterMacroAssembler::restore_after_resume(Register fp) {
2202 if (!Continuations::enabled()) return;
2203
2204 const address resume_adapter = TemplateInterpreter::cont_resume_interpreter_adapter();
2205 add_const_optimized(R31, R29_TOC, MacroAssembler::offset_to_global_toc(resume_adapter));
2206 mtctr(R31);
2207 bctrl();
2208 // Restore registers that are preserved across vthread preemption
2209 assert(nonvolatile_accross_vthread_preemtion(R31) && nonvolatile_accross_vthread_preemtion(R22), "");
2210 ld(R3_ARG1, _abi0(callers_sp), R1_SP); // load FP
2211 ld(R31, _ijava_state_neg(lresult), R3_ARG1);
2212 ld(R22, _ijava_state_neg(fresult), R3_ARG1);
2213 #ifdef ASSERT
2214 // Assert FP is in R11_scratch1 (see generate_cont_resume_interpreter_adapter())
2215 {
2216 Label ok;
2217 ld(R12_scratch2, 0, R1_SP); // load fp
2218 cmpd(CR0, R12_scratch2, R11_scratch1);
2219 beq(CR0, ok);
2220 stop(FILE_AND_LINE ": FP is expected in R11_scratch1");
2221 bind(ok);
2222 }
2223 #endif
2224 if (fp != noreg && fp != R11_scratch1) {
2225 mr(fp, R11_scratch1);
2226 }
2227 }
2228
2229 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2230 Register arg_1, Register arg_2,
2231 bool check_exceptions) {
2232 // ARG1 is reserved for the thread.
2233 mr_if_needed(R4_ARG2, arg_1);
2234 assert(arg_2 != R4_ARG2, "smashed argument");
2235 mr_if_needed(R5_ARG3, arg_2);
2236 call_VM(oop_result, entry_point, check_exceptions);
2237 }
2238
2239 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2240 Register arg_1, Register arg_2, Register arg_3,
2241 bool check_exceptions) {
2242 // ARG1 is reserved for the thread.
2243 mr_if_needed(R4_ARG2, arg_1);
2244 assert(arg_2 != R4_ARG2, "smashed argument");
2245 mr_if_needed(R5_ARG3, arg_2);
2246 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument");
2247 mr_if_needed(R6_ARG4, arg_3);
2248 call_VM(oop_result, entry_point, check_exceptions);
2249 }
2250
2251 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) {
2252 ld(scratch, 0, R1_SP);
2253 subf(R0, scratch, R15_esp);
2254 sradi(R0, R0, Interpreter::logStackElementSize);
2255 std(R0, _ijava_state_neg(esp), scratch);
2256 std(R14_bcp, _ijava_state_neg(bcp), scratch);
2257 subf(R0, scratch, R26_monitor);
2258 sradi(R0, R0, Interpreter::logStackElementSize);
2259 std(R0, _ijava_state_neg(monitors), scratch);
2260 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); }
2261 // Other entries should be unchanged.
2262 }
2263
2264 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only, bool restore_top_frame_sp) {
2265 ld_ptr(scratch, _abi0(callers_sp), R1_SP); // Load frame pointer.
2266 if (restore_top_frame_sp) {
2267 // After thawing the top frame of a continuation we reach here with frame::java_abi.
2268 // therefore we have to restore top_frame_sp before the assertion below.
2269 assert(!bcp_and_mdx_only, "chose other registers");
2270 Register tfsp = R18_locals;
2271 Register scratch2 = R26_monitor;
2272 ld(tfsp, _ijava_state_neg(top_frame_sp), scratch);
2273 // Derelativize top_frame_sp
2274 sldi(tfsp, tfsp, Interpreter::logStackElementSize);
2275 add(tfsp, tfsp, scratch);
2276 resize_frame_absolute(tfsp, scratch2, R0);
2277 }
2278 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception).
2279 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code.
2280 if (!bcp_and_mdx_only) {
2281 // Following ones are Metadata.
2282 ld(R19_method, _ijava_state_neg(method), scratch);
2283 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch);
2284 // Following ones are stack addresses and don't require reload.
2285 // Derelativize esp
2286 ld(R15_esp, _ijava_state_neg(esp), scratch);
2287 sldi(R15_esp, R15_esp, Interpreter::logStackElementSize);
2288 add(R15_esp, R15_esp, scratch);
2289 ld(R18_locals, _ijava_state_neg(locals), scratch);
2290 sldi(R18_locals, R18_locals, Interpreter::logStackElementSize);
2291 add(R18_locals, R18_locals, scratch);
2292 ld(R26_monitor, _ijava_state_neg(monitors), scratch);
2293 // Derelativize monitors
2294 sldi(R26_monitor, R26_monitor, Interpreter::logStackElementSize);
2295 add(R26_monitor, R26_monitor, scratch);
2296 }
2297 #ifdef ASSERT
2298 {
2299 Label Lok;
2300 subf(R0, R1_SP, scratch);
2301 cmpdi(CR0, R0, frame::top_ijava_frame_abi_size + frame::ijava_state_size);
2302 bge(CR0, Lok);
2303 stop("frame too small (restore istate)");
2304 bind(Lok);
2305 }
2306 #endif
2307 }
2308
2309 void InterpreterMacroAssembler::get_method_counters(Register method,
2310 Register Rcounters,
2311 Label& skip) {
2312 BLOCK_COMMENT("Load and ev. allocate counter object {");
2313 Label has_counters;
2314 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2315 cmpdi(CR0, Rcounters, 0);
2316 bne(CR0, has_counters);
2317 call_VM(noreg, CAST_FROM_FN_PTR(address,
2318 InterpreterRuntime::build_method_counters), method);
2319 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2320 cmpdi(CR0, Rcounters, 0);
2321 beq(CR0, skip); // No MethodCounters, OutOfMemory.
2322 BLOCK_COMMENT("} Load and ev. allocate counter object");
2323
2324 bind(has_counters);
2325 }
2326
2327 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters,
2328 Register iv_be_count,
2329 Register Rtmp_r0) {
2330 assert(UseCompiler, "incrementing must be useful");
2331 Register invocation_count = iv_be_count;
2332 Register backedge_count = Rtmp_r0;
2333 int delta = InvocationCounter::count_increment;
2334
2335 // Load each counter in a register.
2336 // ld(inv_counter, Rtmp);
2337 // ld(be_counter, Rtmp2);
2338 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() +
2339 InvocationCounter::counter_offset());
2340 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() +
2341 InvocationCounter::counter_offset());
2342
2343 BLOCK_COMMENT("Increment profiling counters {");
2344
2345 // Load the backedge counter.
2346 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int
2347 // Mask the backedge counter.
2348 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value);
2349
2350 // Load the invocation counter.
2351 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int
2352 // Add the delta to the invocation counter and store the result.
2353 addi(invocation_count, invocation_count, delta);
2354 // Store value.
2355 stw(invocation_count, inv_counter_offset, Rcounters);
2356
2357 // Add invocation counter + backedge counter.
2358 add(iv_be_count, backedge_count, invocation_count);
2359
2360 // Note that this macro must leave the backedge_count + invocation_count in
2361 // register iv_be_count!
2362 BLOCK_COMMENT("} Increment profiling counters");
2363 }
2364
2365 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2366 if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); }
2367 }
2368
2369 // Local helper function for the verify_oop_or_return_address macro.
2370 static bool verify_return_address(Method* m, int bci) {
2371 #ifndef PRODUCT
2372 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci;
2373 // Assume it is a valid return address if it is inside m and is preceded by a jsr.
2374 if (!m->contains(pc)) return false;
2375 address jsr_pc;
2376 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2377 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2378 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2379 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2380 #endif // PRODUCT
2381 return false;
2382 }
2383
2384 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2385 if (!VerifyOops) return;
2386
2387 // The VM documentation for the astore[_wide] bytecode allows
2388 // the TOS to be not only an oop but also a return address.
2389 Label test;
2390 Label skip;
2391 // See if it is an address (in the current method):
2392
2393 const int log2_bytecode_size_limit = 16;
2394 srdi_(Rtmp, reg, log2_bytecode_size_limit);
2395 bne(CR0, test);
2396
2397 address fd = CAST_FROM_FN_PTR(address, verify_return_address);
2398 const int nbytes_save = MacroAssembler::num_volatile_regs * 8;
2399 save_volatile_gprs(R1_SP, -nbytes_save); // except R0
2400 save_LR_CR(Rtmp); // Save in old frame.
2401 push_frame_reg_args(nbytes_save, Rtmp);
2402
2403 load_const_optimized(Rtmp, fd, R0);
2404 mr_if_needed(R4_ARG2, reg);
2405 mr(R3_ARG1, R19_method);
2406 call_c(Rtmp); // call C
2407
2408 pop_frame();
2409 restore_LR_CR(Rtmp);
2410 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
2411 b(skip);
2412
2413 // Perform a more elaborate out-of-line call.
2414 // Not an address; verify it:
2415 bind(test);
2416 verify_oop(reg);
2417 bind(skip);
2418 }
2419
2420 // Inline assembly for:
2421 //
2422 // if (thread is in interp_only_mode) {
2423 // InterpreterRuntime::post_method_entry();
2424 // }
2425 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) ||
2426 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) {
2427 // SharedRuntime::jvmpi_method_entry(method, receiver);
2428 // }
2429 void InterpreterMacroAssembler::notify_method_entry() {
2430 // JVMTI
2431 // Whenever JVMTI puts a thread in interp_only_mode, method
2432 // entry/exit events are sent for that thread to track stack
2433 // depth. If it is possible to enter interp_only_mode we add
2434 // the code to check if the event should be sent.
2435 if (JvmtiExport::can_post_interpreter_events()) {
2436 Label jvmti_post_done;
2437
2438 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2439 cmpwi(CR0, R0, 0);
2440 beq(CR0, jvmti_post_done);
2441 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2442
2443 bind(jvmti_post_done);
2444 }
2445 }
2446
2447 // Inline assembly for:
2448 //
2449 // if (thread is in interp_only_mode) {
2450 // // save result
2451 // InterpreterRuntime::post_method_exit();
2452 // // restore result
2453 // }
2454 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) {
2455 // // save result
2456 // SharedRuntime::jvmpi_method_exit();
2457 // // restore result
2458 // }
2459 //
2460 // Native methods have their result stored in d_tmp and l_tmp.
2461 // Java methods have their result stored in the expression stack.
2462 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state,
2463 NotifyMethodExitMode mode, bool check_exceptions) {
2464 // JVMTI
2465 // Whenever JVMTI puts a thread in interp_only_mode, method
2466 // entry/exit events are sent for that thread to track stack
2467 // depth. If it is possible to enter interp_only_mode we add
2468 // the code to check if the event should be sent.
2469 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2470 Label jvmti_post_done;
2471
2472 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2473 cmpwi(CR0, R0, 0);
2474 beq(CR0, jvmti_post_done);
2475 if (!is_native_method) { push(state); } // Expose tos to GC.
2476 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), check_exceptions);
2477 if (!is_native_method) { pop(state); }
2478
2479 align(32, 12);
2480 bind(jvmti_post_done);
2481 }
2482
2483 // Dtrace support not implemented.
2484 }