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 assert(nonvolatile_accross_vthread_preemtion(R24_dispatch_addr),
112 "Requirement of field accesses (e.g. putstatic)");
113 if (bcp_incr) { addi(R14_bcp, R14_bcp, bcp_incr); }
114 mtctr(R24_dispatch_addr);
115 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
116 }
117
118 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
119 assert(scratch_reg != R0, "can't use R0 as scratch_reg here");
120 if (JvmtiExport::can_pop_frame()) {
121 Label L;
122
123 // Check the "pending popframe condition" flag in the current thread.
124 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
125
126 // Initiate popframe handling only if it is not already being
127 // processed. If the flag has the popframe_processing bit set, it
128 // means that this code is called *during* popframe handling - we
129 // don't want to reenter.
130 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit);
131 beq(CR0, L);
132
133 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit);
134 bne(CR0, L);
135
136 // Call the Interpreter::remove_activation_preserving_args_entry()
137 // func to get the address of the same-named entrypoint in the
138 // generated interpreter code.
139 call_c(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
140
141 // Jump to Interpreter::_remove_activation_preserving_args_entry.
142 mtctr(R3_RET);
143 bctr();
144
145 align(32, 12);
146 bind(L);
147 }
148 }
149
150 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
151 const Register Rthr_state_addr = scratch_reg;
152 if (JvmtiExport::can_force_early_return()) {
153 Label Lno_early_ret;
154 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
155 cmpdi(CR0, Rthr_state_addr, 0);
156 beq(CR0, Lno_early_ret);
157
158 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr);
159 cmpwi(CR0, R0, JvmtiThreadState::earlyret_pending);
160 bne(CR0, Lno_early_ret);
161
162 // Jump to Interpreter::_earlyret_entry.
163 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr);
164 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry));
165 mtlr(R3_RET);
166 blr();
167
168 align(32, 12);
169 bind(Lno_early_ret);
170 }
171 }
172
173 void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) {
174 const Register RjvmtiState = Rscratch1;
175 const Register Rscratch2 = R0;
176
177 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
178 li(Rscratch2, 0);
179
180 switch (state) {
181 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
182 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
183 break;
184 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
185 break;
186 case btos: // fall through
187 case ztos: // fall through
188 case ctos: // fall through
189 case stos: // fall through
190 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
191 break;
192 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
193 break;
194 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
195 break;
196 case vtos: break;
197 default : ShouldNotReachHere();
198 }
199
200 // Clean up tos value in the jvmti thread state.
201 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
202 // Set tos state field to illegal value.
203 li(Rscratch2, ilgl);
204 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState);
205 }
206
207 // Common code to dispatch and dispatch_only.
208 // Dispatch value in Lbyte_code and increment Lbcp.
209
210 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) {
211 address table_base = (address)Interpreter::dispatch_table((TosState)0);
212 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base;
213 if (is_simm16(table_offs)) {
214 addi(dst, R25_templateTableBase, (int)table_offs);
215 } else {
216 load_const_optimized(dst, table, R0);
217 }
218 }
219
220 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode,
221 address* table, bool generate_poll) {
222 assert_different_registers(bytecode, R11_scratch1);
223
224 // Calc dispatch table address.
225 load_dispatch_table(R11_scratch1, table);
226
227 if (generate_poll) {
228 address *sfpt_tbl = Interpreter::safept_table(state);
229 if (table != sfpt_tbl) {
230 Label dispatch;
231 ld(R0, in_bytes(JavaThread::polling_word_offset()), R16_thread);
232 // Armed page has poll_bit set, if poll bit is cleared just continue.
233 andi_(R0, R0, SafepointMechanism::poll_bit());
234 beq(CR0, dispatch);
235 load_dispatch_table(R11_scratch1, sfpt_tbl);
236 align(32, 16);
237 bind(dispatch);
238 }
239 }
240
241 sldi(R12_scratch2, bytecode, LogBytesPerWord);
242 ldx(R11_scratch1, R11_scratch1, R12_scratch2);
243
244 // Jump off!
245 mtctr(R11_scratch1);
246 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
247 }
248
249 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) {
250 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize);
251 ldx(Rrecv_dst, Rrecv_dst, R15_esp);
252 }
253
254 // helpers for expression stack
255
256 void InterpreterMacroAssembler::pop_i(Register r) {
257 lwzu(r, Interpreter::stackElementSize, R15_esp);
258 }
259
260 void InterpreterMacroAssembler::pop_ptr(Register r) {
261 ldu(r, Interpreter::stackElementSize, R15_esp);
262 }
263
264 void InterpreterMacroAssembler::pop_l(Register r) {
265 ld(r, Interpreter::stackElementSize, R15_esp);
266 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
267 }
268
269 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
270 lfsu(f, Interpreter::stackElementSize, R15_esp);
271 }
272
273 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
274 lfd(f, Interpreter::stackElementSize, R15_esp);
275 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
276 }
277
278 void InterpreterMacroAssembler::push_i(Register r) {
279 stw(r, 0, R15_esp);
280 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
281 }
282
283 void InterpreterMacroAssembler::push_ptr(Register r) {
284 std(r, 0, R15_esp);
285 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
286 }
287
288 void InterpreterMacroAssembler::push_l(Register r) {
289 // Clear unused slot.
290 load_const_optimized(R0, 0L);
291 std(R0, 0, R15_esp);
292 std(r, - Interpreter::stackElementSize, R15_esp);
293 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
294 }
295
296 void InterpreterMacroAssembler::push_f(FloatRegister f) {
297 stfs(f, 0, R15_esp);
298 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
299 }
300
301 void InterpreterMacroAssembler::push_d(FloatRegister f) {
302 stfd(f, - Interpreter::stackElementSize, R15_esp);
303 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
304 }
305
306 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) {
307 std(first, 0, R15_esp);
308 std(second, -Interpreter::stackElementSize, R15_esp);
309 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
310 }
311
312 void InterpreterMacroAssembler::move_l_to_d(Register l, FloatRegister d) {
313 mtfprd(d, l);
314 }
315
316 void InterpreterMacroAssembler::move_d_to_l(FloatRegister d, Register l) {
317 mffprd(l, d);
318 }
319
320 void InterpreterMacroAssembler::push(TosState state) {
321 switch (state) {
322 case atos: push_ptr(); break;
323 case btos:
324 case ztos:
325 case ctos:
326 case stos:
327 case itos: push_i(); break;
328 case ltos: push_l(); break;
329 case ftos: push_f(); break;
330 case dtos: push_d(); break;
331 case vtos: /* nothing to do */ break;
332 default : ShouldNotReachHere();
333 }
334 }
335
336 void InterpreterMacroAssembler::pop(TosState state) {
337 switch (state) {
338 case atos: pop_ptr(); break;
339 case btos:
340 case ztos:
341 case ctos:
342 case stos:
343 case itos: pop_i(); break;
344 case ltos: pop_l(); break;
345 case ftos: pop_f(); break;
346 case dtos: pop_d(); break;
347 case vtos: /* nothing to do */ break;
348 default : ShouldNotReachHere();
349 }
350 verify_oop(R17_tos, state);
351 }
352
353 void InterpreterMacroAssembler::empty_expression_stack() {
354 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
355 }
356
357 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset,
358 Register Rdst,
359 signedOrNot is_signed) {
360 #if defined(VM_LITTLE_ENDIAN)
361 if (bcp_offset) {
362 load_const_optimized(Rdst, bcp_offset);
363 lhbrx(Rdst, R14_bcp, Rdst);
364 } else {
365 lhbrx(Rdst, R14_bcp);
366 }
367 if (is_signed == Signed) {
368 extsh(Rdst, Rdst);
369 }
370 #else
371 // Read Java big endian format.
372 if (is_signed == Signed) {
373 lha(Rdst, bcp_offset, R14_bcp);
374 } else {
375 lhz(Rdst, bcp_offset, R14_bcp);
376 }
377 #endif
378 }
379
380 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset,
381 Register Rdst,
382 signedOrNot is_signed) {
383 #if defined(VM_LITTLE_ENDIAN)
384 if (bcp_offset) {
385 load_const_optimized(Rdst, bcp_offset);
386 lwbrx(Rdst, R14_bcp, Rdst);
387 } else {
388 lwbrx(Rdst, R14_bcp);
389 }
390 if (is_signed == Signed) {
391 extsw(Rdst, Rdst);
392 }
393 #else
394 // Read Java big endian format.
395 if (bcp_offset & 3) { // Offset unaligned?
396 load_const_optimized(Rdst, bcp_offset);
397 if (is_signed == Signed) {
398 lwax(Rdst, R14_bcp, Rdst);
399 } else {
400 lwzx(Rdst, R14_bcp, Rdst);
401 }
402 } else {
403 if (is_signed == Signed) {
404 lwa(Rdst, bcp_offset, R14_bcp);
405 } else {
406 lwz(Rdst, bcp_offset, R14_bcp);
407 }
408 }
409 #endif
410 }
411
412
413 // Load the constant pool cache index from the bytecode stream.
414 //
415 // Kills / writes:
416 // - Rdst, Rscratch
417 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset,
418 size_t index_size) {
419 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
420 // Cache index is always in the native format, courtesy of Rewriter.
421 if (index_size == sizeof(u2)) {
422 lhz(Rdst, bcp_offset, R14_bcp);
423 } else if (index_size == sizeof(u4)) {
424 if (bcp_offset & 3) {
425 load_const_optimized(Rdst, bcp_offset);
426 lwax(Rdst, R14_bcp, Rdst);
427 } else {
428 lwa(Rdst, bcp_offset, R14_bcp);
429 }
430 } else if (index_size == sizeof(u1)) {
431 lbz(Rdst, bcp_offset, R14_bcp);
432 } else {
433 ShouldNotReachHere();
434 }
435 // Rdst now contains cp cache index.
436 }
437
438 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format)
439 // from (Rsrc)+offset.
440 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset,
441 signedOrNot is_signed) {
442 #if defined(VM_LITTLE_ENDIAN)
443 if (offset) {
444 load_const_optimized(Rdst, offset);
445 lwbrx(Rdst, Rdst, Rsrc);
446 } else {
447 lwbrx(Rdst, Rsrc);
448 }
449 if (is_signed == Signed) {
450 extsw(Rdst, Rdst);
451 }
452 #else
453 if (is_signed == Signed) {
454 lwa(Rdst, offset, Rsrc);
455 } else {
456 lwz(Rdst, offset, Rsrc);
457 }
458 #endif
459 }
460
461 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
462 // Get index out of bytecode pointer
463 get_cache_index_at_bcp(index, 1, sizeof(u4));
464
465 // Get address of invokedynamic array
466 ld_ptr(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset()), R27_constPoolCache);
467 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
468 sldi(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
469 addi(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
470 add(cache, cache, index);
471 }
472
473 void InterpreterMacroAssembler::load_field_or_method_entry(bool is_method, Register cache, Register index, int bcp_offset, bool for_fast_bytecode) {
474 const int entry_size = is_method ? sizeof(ResolvedMethodEntry) : sizeof(ResolvedFieldEntry),
475 base_offset = is_method ? Array<ResolvedMethodEntry>::base_offset_in_bytes() : Array<ResolvedFieldEntry>::base_offset_in_bytes(),
476 entries_offset = is_method ? in_bytes(ConstantPoolCache::method_entries_offset()) : in_bytes(ConstantPoolCache::field_entries_offset());
477
478 // Get index out of bytecode pointer
479 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
480 // Take shortcut if the size is a power of 2
481 if (is_power_of_2(entry_size)) {
482 // Scale index by power of 2
483 sldi(index, index, log2i_exact(entry_size));
484 } else {
485 // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
486 mulli(index, index, entry_size);
487 }
488 // Get address of field entries array
489 ld_ptr(cache, entries_offset, R27_constPoolCache);
490 addi(cache, cache, base_offset);
491 add(cache, cache, index);
492
493 if (for_fast_bytecode) {
494 // Prevent speculative loading from ResolvedFieldEntry/ResolvedMethodEntry as it can miss the info written by another thread.
495 // TemplateTable::patch_bytecode uses release-store.
496 // We reached here via control dependency (Bytecode dispatch has used the rewritten Bytecode).
497 // So, we can use control-isync based ordering.
498 isync();
499 }
500 }
501
502 // Load object from cpool->resolved_references(index).
503 // Kills:
504 // - index
505 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index,
506 Register tmp1, Register tmp2,
507 Label *L_handle_null) {
508 assert_different_registers(result, index, tmp1, tmp2);
509 assert(index->is_nonvolatile(), "needs to survive C-call in resolve_oop_handle");
510 get_constant_pool(result);
511
512 // Convert from field index to resolved_references() index and from
513 // word index to byte offset. Since this is a java object, it can be compressed.
514 sldi(index, index, LogBytesPerHeapOop);
515 // Load pointer for resolved_references[] objArray.
516 ld(result, ConstantPool::cache_offset(), result);
517 ld(result, ConstantPoolCache::resolved_references_offset(), result);
518 resolve_oop_handle(result, tmp1, tmp2, MacroAssembler::PRESERVATION_NONE);
519 #ifdef ASSERT
520 Label index_ok;
521 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result);
522 sldi(R0, R0, LogBytesPerHeapOop);
523 cmpd(CR0, index, R0);
524 blt(CR0, index_ok);
525 stop("resolved reference index out of bounds");
526 bind(index_ok);
527 #endif
528 // Add in the index.
529 add(result, index, result);
530 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result,
531 tmp1, tmp2,
532 MacroAssembler::PRESERVATION_NONE,
533 0, L_handle_null);
534 }
535
536 // load cpool->resolved_klass_at(index)
537 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool, Register Roffset, Register Rklass) {
538 // int value = *(Rcpool->int_at_addr(which));
539 // int resolved_klass_index = extract_low_short_from_int(value);
540 add(Roffset, Rcpool, Roffset);
541 #if defined(VM_LITTLE_ENDIAN)
542 lhz(Roffset, sizeof(ConstantPool), Roffset); // Roffset = resolved_klass_index
543 #else
544 lhz(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index
545 #endif
546
547 ld(Rklass, ConstantPool::resolved_klasses_offset(), Rcpool); // Rklass = Rcpool->_resolved_klasses
548
549 sldi(Roffset, Roffset, LogBytesPerWord);
550 addi(Roffset, Roffset, Array<Klass*>::base_offset_in_bytes());
551 isync(); // Order load of instance Klass wrt. tags.
552 ldx(Rklass, Rklass, Roffset);
553 }
554
555 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
556 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2.
557 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1,
558 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) {
559 // Profile the not-null value's klass.
560 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2);
561 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
562 }
563
564 // Separate these two to allow for delay slot in middle.
565 // These are used to do a test and full jump to exception-throwing code.
566
567 // Check that index is in range for array, then shift index by index_shift,
568 // and put arrayOop + shifted_index into res.
569 // Note: res is still shy of address by array offset into object.
570
571 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex,
572 int index_shift, Register Rtmp, Register Rres) {
573 // Check that index is in range for array, then shift index by index_shift,
574 // and put arrayOop + shifted_index into res.
575 // Note: res is still shy of address by array offset into object.
576 // Kills:
577 // - Rindex
578 // Writes:
579 // - Rres: Address that corresponds to the array index if check was successful.
580 verify_oop(Rarray);
581 const Register Rlength = R0;
582 const Register RsxtIndex = Rtmp;
583 Label LisNull, LnotOOR;
584
585 // Array nullcheck
586 if (!ImplicitNullChecks) {
587 cmpdi(CR0, Rarray, 0);
588 beq(CR0, LisNull);
589 } else {
590 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex);
591 }
592
593 // Rindex might contain garbage in upper bits (remember that we don't sign extend
594 // during integer arithmetic operations). So kill them and put value into same register
595 // where ArrayIndexOutOfBounds would expect the index in.
596 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit
597
598 // Index check
599 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray);
600 cmplw(CR0, Rindex, Rlength);
601 sldi(RsxtIndex, RsxtIndex, index_shift);
602 blt(CR0, LnotOOR);
603 // Index should be in R17_tos, array should be in R4_ARG2.
604 mr_if_needed(R17_tos, Rindex);
605 mr_if_needed(R4_ARG2, Rarray);
606 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
607 mtctr(Rtmp);
608 bctr();
609
610 if (!ImplicitNullChecks) {
611 bind(LisNull);
612 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry);
613 mtctr(Rtmp);
614 bctr();
615 }
616
617 align(32, 16);
618 bind(LnotOOR);
619
620 // Calc address
621 add(Rres, RsxtIndex, Rarray);
622 }
623
624 void InterpreterMacroAssembler::index_check(Register array, Register index,
625 int index_shift, Register tmp, Register res) {
626 // pop array
627 pop_ptr(array);
628
629 // check array
630 index_check_without_pop(array, index, index_shift, tmp, res);
631 }
632
633 void InterpreterMacroAssembler::get_const(Register Rdst) {
634 ld(Rdst, in_bytes(Method::const_offset()), R19_method);
635 }
636
637 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
638 get_const(Rdst);
639 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
640 }
641
642 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
643 get_constant_pool(Rdst);
644 ld(Rdst, ConstantPool::cache_offset(), Rdst);
645 }
646
647 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
648 get_constant_pool(Rcpool);
649 ld(Rtags, ConstantPool::tags_offset(), Rcpool);
650 }
651
652 // Unlock if synchronized method.
653 //
654 // Unlock the receiver if this is a synchronized method.
655 // Unlock any Java monitors from synchronized blocks.
656 //
657 // If there are locked Java monitors
658 // If throw_monitor_exception
659 // throws IllegalMonitorStateException
660 // Else if install_monitor_exception
661 // installs IllegalMonitorStateException
662 // Else
663 // no error processing
664 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
665 bool throw_monitor_exception,
666 bool install_monitor_exception) {
667 Label Lunlocked, Lno_unlock;
668 {
669 Register Rdo_not_unlock_flag = R11_scratch1;
670 Register Raccess_flags = R12_scratch2;
671
672 // Check if synchronized method or unlocking prevented by
673 // JavaThread::do_not_unlock_if_synchronized flag.
674 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
675 lhz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method);
676 li(R0, 0);
677 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag
678
679 push(state);
680
681 // Skip if we don't have to unlock.
682 testbitdi(CR0, R0, Raccess_flags, JVM_ACC_SYNCHRONIZED_BIT);
683 beq(CR0, Lunlocked);
684
685 cmpwi(CR0, Rdo_not_unlock_flag, 0);
686 bne(CR0, Lno_unlock);
687 }
688
689 // Unlock
690 {
691 Register Rmonitor_base = R11_scratch1;
692
693 Label Lunlock;
694 // If it's still locked, everything is ok, unlock it.
695 ld(Rmonitor_base, 0, R1_SP);
696 addi(Rmonitor_base, Rmonitor_base,
697 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
698
699 ld(R0, BasicObjectLock::obj_offset(), Rmonitor_base);
700 cmpdi(CR0, R0, 0);
701 bne(CR0, Lunlock);
702
703 // If it's already unlocked, throw exception.
704 if (throw_monitor_exception) {
705 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
706 should_not_reach_here();
707 } else {
708 if (install_monitor_exception) {
709 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
710 b(Lunlocked);
711 }
712 }
713
714 bind(Lunlock);
715 unlock_object(Rmonitor_base);
716 }
717
718 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not.
719 bind(Lunlocked);
720 {
721 Label Lexception, Lrestart;
722 Register Rcurrent_obj_addr = R11_scratch1;
723 const int delta = frame::interpreter_frame_monitor_size_in_bytes();
724 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords");
725
726 bind(Lrestart);
727 // Set up search loop: Calc num of iterations.
728 {
729 Register Riterations = R12_scratch2;
730 Register Rmonitor_base = Rcurrent_obj_addr;
731 ld(Rmonitor_base, 0, R1_SP);
732 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base
733
734 subf_(Riterations, R26_monitor, Rmonitor_base);
735 ble(CR0, Lno_unlock);
736
737 addi(Rcurrent_obj_addr, Rmonitor_base,
738 in_bytes(BasicObjectLock::obj_offset()) - frame::interpreter_frame_monitor_size_in_bytes());
739 // Check if any monitor is on stack, bail out if not
740 srdi(Riterations, Riterations, exact_log2(delta));
741 mtctr(Riterations);
742 }
743
744 // The search loop: Look for locked monitors.
745 {
746 const Register Rcurrent_obj = R0;
747 Label Lloop;
748
749 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
750 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
751 bind(Lloop);
752
753 // Check if current entry is used.
754 cmpdi(CR0, Rcurrent_obj, 0);
755 bne(CR0, Lexception);
756 // Preload next iteration's compare value.
757 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
758 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
759 bdnz(Lloop);
760 }
761 // Fell through: Everything's unlocked => finish.
762 b(Lno_unlock);
763
764 // An object is still locked => need to throw exception.
765 bind(Lexception);
766 if (throw_monitor_exception) {
767 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
768 should_not_reach_here();
769 } else {
770 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
771 // Unlock does not block, so don't have to worry about the frame.
772 Register Rmonitor_addr = R11_scratch1;
773 addi(Rmonitor_addr, Rcurrent_obj_addr, -in_bytes(BasicObjectLock::obj_offset()) + delta);
774 unlock_object(Rmonitor_addr);
775 if (install_monitor_exception) {
776 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
777 }
778 b(Lrestart);
779 }
780 }
781
782 align(32, 12);
783 bind(Lno_unlock);
784 pop(state);
785 }
786
787 // Support function for remove_activation & Co.
788 void InterpreterMacroAssembler::load_fp(Register fp) {
789 ld(fp, _abi0(callers_sp), R1_SP); // *SP
790 }
791
792 void InterpreterMacroAssembler::remove_top_frame_given_fp(Register fp, Register sender_sp, Register sender_fp,
793 Register return_pc, Register temp) {
794 assert_different_registers(sender_sp, sender_fp, return_pc, temp);
795 ld(sender_sp, _ijava_state_neg(sender_sp), fp);
796 ld(sender_fp, _abi0(callers_sp), fp); // **SP
797 if (return_pc != noreg) {
798 ld(return_pc, _abi0(lr), fp); // last usage of fp, register can be reused
799 }
800 subf(temp, R1_SP, sender_sp); // sender_sp - SP
801 stdux(sender_fp, R1_SP, temp); // atomically set *(SP = sender_sp) = sender_fp
802 }
803
804 void InterpreterMacroAssembler::merge_frames(Register sender_sp, Register return_pc,
805 Register temp1, Register temp2) {
806 Register fp = temp1, sender_fp = temp2;
807 load_fp(fp);
808 remove_top_frame_given_fp(fp, sender_sp, sender_fp, return_pc, /* temp */ fp);
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 asm_assert_mem8_is_zero(in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread,
868 "remove_activation: should not have alternate return address set");
869
870 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
871
872 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
873 // that would normally not be safe to use. Such bad returns into unsafe territory of
874 // the stack, will call InterpreterRuntime::at_unwind.
875 Label slow_path, fast_path;
876 Register fp = R22_tmp2;
877 load_fp(fp);
878
879 JFR_ONLY(enter_jfr_critical_section();)
880 safepoint_poll(slow_path, R11_scratch1, true /* at_return */, false /* in_nmethod */);
881 b(fast_path);
882 bind(slow_path);
883 push(state);
884 set_last_Java_frame(R1_SP, noreg);
885 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), R16_thread);
886 reset_last_Java_frame();
887 pop(state);
888 align(32);
889 bind(fast_path);
890
891 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
892 notify_method_exit(false, state, NotifyJVMTI, true);
893
894 BLOCK_COMMENT("reserved_stack_check:");
895 if (StackReservedPages > 0) {
896 // Test if reserved zone needs to be enabled.
897 Label no_reserved_zone_enabling;
898
899 // check if already enabled - if so no re-enabling needed
900 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
901 lwz(R0, in_bytes(JavaThread::stack_guard_state_offset()), R16_thread);
902 cmpwi(CR0, R0, StackOverflow::stack_guard_enabled);
903 beq_predict_taken(CR0, no_reserved_zone_enabling);
904
905 // Compare frame pointers. There is no good stack pointer, as with stack
906 // frame compression we can get different SPs when we do calls. A subsequent
907 // call could have a smaller SP, so that this compare succeeds for an
908 // inner call of the method annotated with ReservedStack.
909 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread);
910 cmpld(CR0, fp, R0);
911 blt_predict_taken(CR0, no_reserved_zone_enabling);
912
913 JFR_ONLY(leave_jfr_critical_section();)
914
915 // Enable reserved zone again, throw stack overflow exception.
916 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread);
917 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
918
919 should_not_reach_here();
920
921 bind(no_reserved_zone_enabling);
922 }
923
924 verify_oop(R17_tos, state);
925
926 remove_top_frame_given_fp(fp, R21_sender_SP, R23_tmp3, /*return_pc*/ R0, R11_scratch1);
927 mtlr(R0);
928 pop_cont_fastpath();
929 JFR_ONLY(leave_jfr_critical_section();)
930
931 BLOCK_COMMENT("} remove_activation");
932 }
933
934 #if INCLUDE_JFR
935 void InterpreterMacroAssembler::enter_jfr_critical_section() {
936 li(R0, 1);
937 stb(R0, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR), R16_thread);
938 }
939
940 void InterpreterMacroAssembler::leave_jfr_critical_section() {
941 li(R0, 0);
942 stb(R0, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR), R16_thread);
943 }
944 #endif // INCLUDE_JFR
945
946 // Lock object
947 //
948 // Registers alive
949 // monitor - Address of the BasicObjectLock to be used for locking,
950 // which must be initialized with the object to lock.
951 // object - Address of the object to be locked.
952 //
953 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
954 const Register header = R7_ARG5;
955 const Register tmp = R8_ARG6;
956
957 Label done, slow_case;
958
959 assert_different_registers(header, tmp);
960
961 lightweight_lock(monitor, object, header, tmp, slow_case);
962 b(done);
963
964 bind(slow_case);
965 call_VM_preemptable(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
966
967 bind(done);
968 }
969
970 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
971 //
972 // Registers alive
973 // monitor - Address of the BasicObjectLock to be used for locking,
974 // which must be initialized with the object to lock.
975 //
976 // Throw IllegalMonitorException if object is not locked by current thread.
977 void InterpreterMacroAssembler::unlock_object(Register monitor) {
978 const Register object = R7_ARG5;
979 const Register header = R8_ARG6;
980 const Register current_header = R10_ARG8;
981
982 Label free_slot;
983 Label slow_case;
984
985 assert_different_registers(object, header, current_header);
986
987 // The object address from the monitor is in object.
988 ld(object, in_bytes(BasicObjectLock::obj_offset()), monitor);
989
990 lightweight_unlock(object, header, slow_case);
991
992 b(free_slot);
993
994 bind(slow_case);
995 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
996
997 Label done;
998 b(done); // Monitor register may be overwritten! Runtime has already freed the slot.
999
1000 // Do monitor->set_obj(nullptr);
1001 align(32, 12);
1002 bind(free_slot);
1003 li(R0, 0);
1004 std(R0, in_bytes(BasicObjectLock::obj_offset()), monitor);
1005 bind(done);
1006 }
1007
1008 // Load compiled (i2c) or interpreter entry when calling from interpreted and
1009 // do the call. Centralized so that all interpreter calls will do the same actions.
1010 // If jvmti single stepping is on for a thread we must not call compiled code.
1011 //
1012 // Input:
1013 // - Rtarget_method: method to call
1014 // - Rret_addr: return address
1015 // - 2 scratch regs
1016 //
1017 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr,
1018 Register Rscratch1, Register Rscratch2) {
1019 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr);
1020 // Assume we want to go compiled if available.
1021 const Register Rtarget_addr = Rscratch1;
1022 const Register Rinterp_only = Rscratch2;
1023
1024 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method);
1025
1026 if (JvmtiExport::can_post_interpreter_events()) {
1027 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
1028
1029 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
1030 // compiled code in threads for which the event is enabled. Check here for
1031 // interp_only_mode if these events CAN be enabled.
1032 Label done;
1033 cmpwi(CR0, Rinterp_only, 0);
1034 beq(CR0, done);
1035 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method);
1036 align(32, 12);
1037 bind(done);
1038 }
1039
1040 #ifdef ASSERT
1041 {
1042 Label Lok;
1043 cmpdi(CR0, Rtarget_addr, 0);
1044 bne(CR0, Lok);
1045 stop("null entry point");
1046 bind(Lok);
1047 }
1048 #endif // ASSERT
1049
1050 mr(R21_sender_SP, R1_SP);
1051
1052 // Calc a precise SP for the call. The SP value we calculated in
1053 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space
1054 // if esp is not max. Also, the i2c adapter extends the stack space without restoring
1055 // our pre-calced value, so repeating calls via i2c would result in stack overflow.
1056 // Since esp already points to an empty slot, we just have to sub 1 additional slot
1057 // to meet the abi scratch requirements.
1058 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in
1059 // the return entry of the interpreter.
1060 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::top_ijava_frame_abi_size);
1061 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address
1062 resize_frame_absolute(Rscratch2, Rscratch2, R0);
1063
1064 mr_if_needed(R19_method, Rtarget_method);
1065 mtctr(Rtarget_addr);
1066 mtlr(Rret_addr);
1067
1068 save_interpreter_state(Rscratch2);
1069 #ifdef ASSERT
1070 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp
1071 sldi(Rscratch1, Rscratch1, Interpreter::logStackElementSize);
1072 add(Rscratch1, Rscratch1, Rscratch2); // Rscratch2 contains fp
1073 // Compare sender_sp with the derelativized top_frame_sp
1074 cmpd(CR0, R21_sender_SP, Rscratch1);
1075 asm_assert_eq("top_frame_sp incorrect");
1076 #endif
1077
1078 bctr();
1079 }
1080
1081 // Set the method data pointer for the current bcp.
1082 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1083 assert(ProfileInterpreter, "must be profiling interpreter");
1084 Label get_continue;
1085 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method);
1086 test_method_data_pointer(get_continue);
1087 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp);
1088
1089 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1090 add(R28_mdx, R28_mdx, R3_RET);
1091 bind(get_continue);
1092 }
1093
1094 // Test ImethodDataPtr. If it is null, continue at the specified label.
1095 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1096 assert(ProfileInterpreter, "must be profiling interpreter");
1097 cmpdi(CR0, R28_mdx, 0);
1098 beq(CR0, zero_continue);
1099 }
1100
1101 void InterpreterMacroAssembler::verify_method_data_pointer() {
1102 assert(ProfileInterpreter, "must be profiling interpreter");
1103 #ifdef ASSERT
1104 Label verify_continue;
1105 test_method_data_pointer(verify_continue);
1106
1107 // If the mdp is valid, it will point to a DataLayout header which is
1108 // consistent with the bcp. The converse is highly probable also.
1109 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx);
1110 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method);
1111 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1112 add(R11_scratch1, R12_scratch2, R12_scratch2);
1113 cmpd(CR0, R11_scratch1, R14_bcp);
1114 beq(CR0, verify_continue);
1115
1116 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx);
1117
1118 bind(verify_continue);
1119 #endif
1120 }
1121
1122 // Store a value at some constant offset from the method data pointer.
1123 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1124 assert(ProfileInterpreter, "must be profiling interpreter");
1125
1126 std(value, constant, R28_mdx);
1127 }
1128
1129 // Increment the value at some constant offset from the method data pointer.
1130 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1131 Register counter_addr,
1132 Register Rbumped_count,
1133 bool decrement) {
1134 // Locate the counter at a fixed offset from the mdp:
1135 addi(counter_addr, R28_mdx, constant);
1136 increment_mdp_data_at(counter_addr, Rbumped_count, decrement);
1137 }
1138
1139 // Increment the value at some non-fixed (reg + constant) offset from
1140 // the method data pointer.
1141 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1142 int constant,
1143 Register scratch,
1144 Register Rbumped_count,
1145 bool decrement) {
1146 // Add the constant to reg to get the offset.
1147 add(scratch, R28_mdx, reg);
1148 // Then calculate the counter address.
1149 addi(scratch, scratch, constant);
1150 increment_mdp_data_at(scratch, Rbumped_count, decrement);
1151 }
1152
1153 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr,
1154 Register Rbumped_count,
1155 bool decrement) {
1156 assert(ProfileInterpreter, "must be profiling interpreter");
1157
1158 // Load the counter.
1159 ld(Rbumped_count, 0, counter_addr);
1160
1161 if (decrement) {
1162 // Decrement the register. Set condition codes.
1163 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment);
1164 // Store the decremented counter, if it is still negative.
1165 std(Rbumped_count, 0, counter_addr);
1166 // Note: add/sub overflow check are not ported, since 64 bit
1167 // calculation should never overflow.
1168 } else {
1169 // Increment the register. Set carry flag.
1170 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment);
1171 // Store the incremented counter.
1172 std(Rbumped_count, 0, counter_addr);
1173 }
1174 }
1175
1176 // Set a flag value at the current method data pointer position.
1177 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1178 Register scratch) {
1179 assert(ProfileInterpreter, "must be profiling interpreter");
1180 // Load the data header.
1181 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1182 // Set the flag.
1183 ori(scratch, scratch, flag_constant);
1184 // Store the modified header.
1185 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1186 }
1187
1188 // Test the location at some offset from the method data pointer.
1189 // If it is not equal to value, branch to the not_equal_continue Label.
1190 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1191 Register value,
1192 Label& not_equal_continue,
1193 Register test_out) {
1194 assert(ProfileInterpreter, "must be profiling interpreter");
1195
1196 ld(test_out, offset, R28_mdx);
1197 cmpd(CR0, value, test_out);
1198 bne(CR0, not_equal_continue);
1199 }
1200
1201 // Update the method data pointer by the displacement located at some fixed
1202 // offset from the method data pointer.
1203 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1204 Register scratch) {
1205 assert(ProfileInterpreter, "must be profiling interpreter");
1206
1207 ld(scratch, offset_of_disp, R28_mdx);
1208 add(R28_mdx, scratch, R28_mdx);
1209 }
1210
1211 // Update the method data pointer by the displacement located at the
1212 // offset (reg + offset_of_disp).
1213 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1214 int offset_of_disp,
1215 Register scratch) {
1216 assert(ProfileInterpreter, "must be profiling interpreter");
1217
1218 add(scratch, reg, R28_mdx);
1219 ld(scratch, offset_of_disp, scratch);
1220 add(R28_mdx, scratch, R28_mdx);
1221 }
1222
1223 // Update the method data pointer by a simple constant displacement.
1224 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1225 assert(ProfileInterpreter, "must be profiling interpreter");
1226 addi(R28_mdx, R28_mdx, constant);
1227 }
1228
1229 // Update the method data pointer for a _ret bytecode whose target
1230 // was not among our cached targets.
1231 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1232 Register return_bci) {
1233 assert(ProfileInterpreter, "must be profiling interpreter");
1234
1235 push(state);
1236 assert(return_bci->is_nonvolatile(), "need to protect return_bci");
1237 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1238 pop(state);
1239 }
1240
1241 // Increments the backedge counter.
1242 // Returns backedge counter + invocation counter in Rdst.
1243 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst,
1244 const Register Rtmp1, Register Rscratch) {
1245 assert(UseCompiler, "incrementing must be useful");
1246 assert_different_registers(Rdst, Rtmp1);
1247 const Register invocation_counter = Rtmp1;
1248 const Register counter = Rdst;
1249 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size.");
1250
1251 // Load backedge counter.
1252 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1253 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1254 // Load invocation counter.
1255 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) +
1256 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1257
1258 // Add the delta to the backedge counter.
1259 addi(counter, counter, InvocationCounter::count_increment);
1260
1261 // Mask the invocation counter.
1262 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value);
1263
1264 // Store new counter value.
1265 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1266 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1267 // Return invocation counter + backedge counter.
1268 add(counter, counter, invocation_counter);
1269 }
1270
1271 // Count a taken branch in the bytecodes.
1272 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1273 if (ProfileInterpreter) {
1274 Label profile_continue;
1275
1276 // If no method data exists, go to profile_continue.
1277 test_method_data_pointer(profile_continue);
1278
1279 // We are taking a branch. Increment the taken count.
1280 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count);
1281
1282 // The method data pointer needs to be updated to reflect the new target.
1283 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1284 bind (profile_continue);
1285 }
1286 }
1287
1288 // Count a not-taken branch in the bytecodes.
1289 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) {
1290 if (ProfileInterpreter) {
1291 Label profile_continue;
1292
1293 // If no method data exists, go to profile_continue.
1294 test_method_data_pointer(profile_continue);
1295
1296 // We are taking a branch. Increment the not taken count.
1297 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2);
1298
1299 // The method data pointer needs to be updated to correspond to the
1300 // next bytecode.
1301 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1302 bind (profile_continue);
1303 }
1304 }
1305
1306 // Count a non-virtual call in the bytecodes.
1307 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) {
1308 if (ProfileInterpreter) {
1309 Label profile_continue;
1310
1311 // If no method data exists, go to profile_continue.
1312 test_method_data_pointer(profile_continue);
1313
1314 // We are making a call. Increment the count.
1315 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1316
1317 // The method data pointer needs to be updated to reflect the new target.
1318 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1319 bind (profile_continue);
1320 }
1321 }
1322
1323 // Count a final call in the bytecodes.
1324 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) {
1325 if (ProfileInterpreter) {
1326 Label profile_continue;
1327
1328 // If no method data exists, go to profile_continue.
1329 test_method_data_pointer(profile_continue);
1330
1331 // We are making a call. Increment the count.
1332 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1333
1334 // The method data pointer needs to be updated to reflect the new target.
1335 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1336 bind (profile_continue);
1337 }
1338 }
1339
1340 // Count a virtual call in the bytecodes.
1341 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver,
1342 Register Rscratch1,
1343 Register Rscratch2,
1344 bool receiver_can_be_null) {
1345 if (!ProfileInterpreter) { return; }
1346 Label profile_continue;
1347
1348 // If no method data exists, go to profile_continue.
1349 test_method_data_pointer(profile_continue);
1350
1351 Label skip_receiver_profile;
1352 if (receiver_can_be_null) {
1353 Label not_null;
1354 cmpdi(CR0, Rreceiver, 0);
1355 bne(CR0, not_null);
1356 // We are making a call. Increment the count for null receiver.
1357 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2);
1358 b(skip_receiver_profile);
1359 bind(not_null);
1360 }
1361
1362 // Record the receiver type.
1363 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2);
1364 bind(skip_receiver_profile);
1365
1366 // The method data pointer needs to be updated to reflect the new target.
1367 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1368 bind (profile_continue);
1369 }
1370
1371 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) {
1372 if (ProfileInterpreter) {
1373 Label profile_continue;
1374
1375 // If no method data exists, go to profile_continue.
1376 test_method_data_pointer(profile_continue);
1377
1378 int mdp_delta = in_bytes(BitData::bit_data_size());
1379 if (TypeProfileCasts) {
1380 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1381
1382 // Record the object type.
1383 record_klass_in_profile(Rklass, Rscratch1, Rscratch2);
1384 }
1385
1386 // The method data pointer needs to be updated.
1387 update_mdp_by_constant(mdp_delta);
1388
1389 bind (profile_continue);
1390 }
1391 }
1392
1393 // Count a ret in the bytecodes.
1394 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci,
1395 Register scratch1, Register scratch2) {
1396 if (ProfileInterpreter) {
1397 Label profile_continue;
1398 uint row;
1399
1400 // If no method data exists, go to profile_continue.
1401 test_method_data_pointer(profile_continue);
1402
1403 // Update the total ret count.
1404 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 );
1405
1406 for (row = 0; row < RetData::row_limit(); row++) {
1407 Label next_test;
1408
1409 // See if return_bci is equal to bci[n]:
1410 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1);
1411
1412 // return_bci is equal to bci[n]. Increment the count.
1413 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2);
1414
1415 // The method data pointer needs to be updated to reflect the new target.
1416 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1);
1417 b(profile_continue);
1418 bind(next_test);
1419 }
1420
1421 update_mdp_for_ret(state, return_bci);
1422
1423 bind (profile_continue);
1424 }
1425 }
1426
1427 // Count the default case of a switch construct.
1428 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) {
1429 if (ProfileInterpreter) {
1430 Label profile_continue;
1431
1432 // If no method data exists, go to profile_continue.
1433 test_method_data_pointer(profile_continue);
1434
1435 // Update the default case count
1436 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1437 scratch1, scratch2);
1438
1439 // The method data pointer needs to be updated.
1440 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()),
1441 scratch1);
1442
1443 bind (profile_continue);
1444 }
1445 }
1446
1447 // Count the index'th case of a switch construct.
1448 void InterpreterMacroAssembler::profile_switch_case(Register index,
1449 Register scratch1,
1450 Register scratch2,
1451 Register scratch3) {
1452 if (ProfileInterpreter) {
1453 assert_different_registers(index, scratch1, scratch2, scratch3);
1454 Label profile_continue;
1455
1456 // If no method data exists, go to profile_continue.
1457 test_method_data_pointer(profile_continue);
1458
1459 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes().
1460 li(scratch3, in_bytes(MultiBranchData::case_array_offset()));
1461
1462 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works");
1463 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1464 add(scratch1, scratch1, scratch3);
1465
1466 // Update the case count.
1467 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3);
1468
1469 // The method data pointer needs to be updated.
1470 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2);
1471
1472 bind (profile_continue);
1473 }
1474 }
1475
1476 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) {
1477 if (ProfileInterpreter) {
1478 assert_different_registers(Rscratch1, Rscratch2);
1479 Label profile_continue;
1480
1481 // If no method data exists, go to profile_continue.
1482 test_method_data_pointer(profile_continue);
1483
1484 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1);
1485
1486 // The method data pointer needs to be updated.
1487 int mdp_delta = in_bytes(BitData::bit_data_size());
1488 if (TypeProfileCasts) {
1489 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1490 }
1491 update_mdp_by_constant(mdp_delta);
1492
1493 bind (profile_continue);
1494 }
1495 }
1496
1497 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver,
1498 Register Rscratch1, Register Rscratch2) {
1499 assert(ProfileInterpreter, "must be profiling");
1500 assert_different_registers(Rreceiver, Rscratch1, Rscratch2);
1501
1502 Label done;
1503 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done);
1504 bind (done);
1505 }
1506
1507 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1508 Register receiver, Register scratch1, Register scratch2,
1509 int start_row, Label& done) {
1510 if (TypeProfileWidth == 0) {
1511 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1512 return;
1513 }
1514
1515 int last_row = VirtualCallData::row_limit() - 1;
1516 assert(start_row <= last_row, "must be work left to do");
1517 // Test this row for both the receiver and for null.
1518 // Take any of three different outcomes:
1519 // 1. found receiver => increment count and goto done
1520 // 2. found null => keep looking for case 1, maybe allocate this cell
1521 // 3. found something else => keep looking for cases 1 and 2
1522 // Case 3 is handled by a recursive call.
1523 for (int row = start_row; row <= last_row; row++) {
1524 Label next_test;
1525 bool test_for_null_also = (row == start_row);
1526
1527 // See if the receiver is receiver[n].
1528 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1529 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1);
1530 // delayed()->tst(scratch);
1531
1532 // The receiver is receiver[n]. Increment count[n].
1533 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1534 increment_mdp_data_at(count_offset, scratch1, scratch2);
1535 b(done);
1536 bind(next_test);
1537
1538 if (test_for_null_also) {
1539 Label found_null;
1540 // Failed the equality check on receiver[n]... Test for null.
1541 if (start_row == last_row) {
1542 // The only thing left to do is handle the null case.
1543 // Scratch1 contains test_out from test_mdp_data_at.
1544 cmpdi(CR0, scratch1, 0);
1545 beq(CR0, found_null);
1546 // Receiver did not match any saved receiver and there is no empty row for it.
1547 // Increment total counter to indicate polymorphic case.
1548 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1549 b(done);
1550 bind(found_null);
1551 break;
1552 }
1553 // Since null is rare, make it be the branch-taken case.
1554 cmpdi(CR0, scratch1, 0);
1555 beq(CR0, found_null);
1556
1557 // Put all the "Case 3" tests here.
1558 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done);
1559
1560 // Found a null. Keep searching for a matching receiver,
1561 // but remember that this is an empty (unused) slot.
1562 bind(found_null);
1563 }
1564 }
1565
1566 // In the fall-through case, we found no matching receiver, but we
1567 // observed the receiver[start_row] is null.
1568
1569 // Fill in the receiver field and increment the count.
1570 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1571 set_mdp_data_at(recvr_offset, receiver);
1572 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1573 li(scratch1, DataLayout::counter_increment);
1574 set_mdp_data_at(count_offset, scratch1);
1575 if (start_row > 0) {
1576 b(done);
1577 }
1578 }
1579
1580 // Argument and return type profilig.
1581 // kills: tmp, tmp2, R0, CR0, CR1
1582 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base,
1583 RegisterOrConstant mdo_addr_offs,
1584 Register tmp, Register tmp2) {
1585 Label do_nothing, do_update;
1586
1587 // tmp2 = obj is allowed
1588 assert_different_registers(obj, mdo_addr_base, tmp, R0);
1589 assert_different_registers(tmp2, mdo_addr_base, tmp, R0);
1590 const Register klass = tmp2;
1591
1592 verify_oop(obj);
1593
1594 ld(tmp, mdo_addr_offs, mdo_addr_base);
1595
1596 // Set null_seen if obj is 0.
1597 cmpdi(CR0, obj, 0);
1598 ori(R0, tmp, TypeEntries::null_seen);
1599 beq(CR0, do_update);
1600
1601 load_klass(klass, obj);
1602
1603 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
1604 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
1605 cmpd(CR1, R0, klass);
1606 // Klass seen before, nothing to do (regardless of unknown bit).
1607 //beq(CR1, do_nothing);
1608
1609 andi_(R0, tmp, TypeEntries::type_unknown);
1610 // Already unknown. Nothing to do anymore.
1611 //bne(CR0, do_nothing);
1612 crorc(CR0, Assembler::equal, CR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne
1613 beq(CR0, do_nothing);
1614
1615 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
1616 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1617 beq(CR0, do_update); // First time here. Set profile type.
1618
1619 // Different than before. Cannot keep accurate profile.
1620 ori(R0, tmp, TypeEntries::type_unknown);
1621
1622 bind(do_update);
1623 // update profile
1624 std(R0, mdo_addr_offs, mdo_addr_base);
1625
1626 align(32, 12);
1627 bind(do_nothing);
1628 }
1629
1630 void InterpreterMacroAssembler::profile_arguments_type(Register callee,
1631 Register tmp1, Register tmp2,
1632 bool is_virtual) {
1633 if (!ProfileInterpreter) {
1634 return;
1635 }
1636
1637 assert_different_registers(callee, tmp1, tmp2, R28_mdx);
1638
1639 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1640 Label profile_continue;
1641
1642 test_method_data_pointer(profile_continue);
1643
1644 int off_to_start = is_virtual ?
1645 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1646
1647 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx);
1648 cmpwi(CR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1649 bne(CR0, profile_continue);
1650
1651 if (MethodData::profile_arguments()) {
1652 Label done;
1653 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1654 addi(R28_mdx, R28_mdx, off_to_args);
1655
1656 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1657 if (i > 0 || MethodData::profile_return()) {
1658 // If return value type is profiled we may have no argument to profile.
1659 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1660 cmpdi(CR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count());
1661 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count());
1662 blt(CR0, done);
1663 }
1664 ld(tmp1, in_bytes(Method::const_offset()), callee);
1665 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1);
1666 // Stack offset o (zero based) from the start of the argument
1667 // list, for n arguments translates into offset n - o - 1 from
1668 // the end of the argument list. But there's an extra slot at
1669 // the top of the stack. So the offset is n - o from Lesp.
1670 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx);
1671 subf(tmp1, tmp2, tmp1);
1672
1673 sldi(tmp1, tmp1, Interpreter::logStackElementSize);
1674 ldx(tmp1, tmp1, R15_esp);
1675
1676 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1);
1677
1678 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1679 addi(R28_mdx, R28_mdx, to_add);
1680 off_to_args += to_add;
1681 }
1682
1683 if (MethodData::profile_return()) {
1684 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1685 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1686 }
1687
1688 bind(done);
1689
1690 if (MethodData::profile_return()) {
1691 // We're right after the type profile for the last
1692 // argument. tmp1 is the number of cells left in the
1693 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1694 // if there's a return to profile.
1695 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(),
1696 "can't move past ret type");
1697 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size));
1698 add(R28_mdx, tmp1, R28_mdx);
1699 }
1700 } else {
1701 assert(MethodData::profile_return(), "either profile call args or call ret");
1702 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
1703 }
1704
1705 // Mdp points right after the end of the
1706 // CallTypeData/VirtualCallTypeData, right after the cells for the
1707 // return value type if there's one.
1708 align(32, 12);
1709 bind(profile_continue);
1710 }
1711 }
1712
1713 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
1714 assert_different_registers(ret, tmp1, tmp2);
1715 if (ProfileInterpreter && MethodData::profile_return()) {
1716 Label profile_continue;
1717
1718 test_method_data_pointer(profile_continue);
1719
1720 if (MethodData::profile_return_jsr292_only()) {
1721 // If we don't profile all invoke bytecodes we must make sure
1722 // it's a bytecode we indeed profile. We can't go back to the
1723 // beginning of the ProfileData we intend to update to check its
1724 // type because we're right after it and we don't known its
1725 // length.
1726 lbz(tmp1, 0, R14_bcp);
1727 lbz(tmp2, in_bytes(Method::intrinsic_id_offset()), R19_method);
1728 cmpwi(CR0, tmp1, Bytecodes::_invokedynamic);
1729 cmpwi(CR1, tmp1, Bytecodes::_invokehandle);
1730 cror(CR0, Assembler::equal, CR1, Assembler::equal);
1731 cmpwi(CR1, tmp2, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1732 cror(CR0, Assembler::equal, CR1, Assembler::equal);
1733 bne(CR0, profile_continue);
1734 }
1735
1736 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2);
1737
1738 align(32, 12);
1739 bind(profile_continue);
1740 }
1741 }
1742
1743 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2,
1744 Register tmp3, Register tmp4) {
1745 if (ProfileInterpreter && MethodData::profile_parameters()) {
1746 Label profile_continue, done;
1747
1748 test_method_data_pointer(profile_continue);
1749
1750 // Load the offset of the area within the MDO used for
1751 // parameters. If it's negative we're not profiling any parameters.
1752 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx);
1753 cmpwi(CR0, tmp1, 0);
1754 blt(CR0, profile_continue);
1755
1756 // Compute a pointer to the area for parameters from the offset
1757 // and move the pointer to the slot for the last
1758 // parameters. Collect profiling from last parameter down.
1759 // mdo start + parameters offset + array length - 1
1760
1761 // Pointer to the parameter area in the MDO.
1762 const Register mdp = tmp1;
1763 add(mdp, tmp1, R28_mdx);
1764
1765 // Offset of the current profile entry to update.
1766 const Register entry_offset = tmp2;
1767 // entry_offset = array len in number of cells
1768 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp);
1769
1770 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1771 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
1772
1773 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
1774 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size));
1775 // entry_offset in bytes
1776 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1777
1778 Label loop;
1779 align(32, 12);
1780 bind(loop);
1781
1782 // Load offset on the stack from the slot for this parameter.
1783 ld(tmp3, entry_offset, mdp);
1784 sldi(tmp3, tmp3, Interpreter::logStackElementSize);
1785 neg(tmp3, tmp3);
1786 // Read the parameter from the local area.
1787 ldx(tmp3, tmp3, R18_locals);
1788
1789 // Make entry_offset now point to the type field for this parameter.
1790 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1791 assert(type_base > off_base, "unexpected");
1792 addi(entry_offset, entry_offset, type_base - off_base);
1793
1794 // Profile the parameter.
1795 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3);
1796
1797 // Go to next parameter.
1798 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base);
1799 cmpdi(CR0, entry_offset, off_base + delta);
1800 addi(entry_offset, entry_offset, -delta);
1801 bge(CR0, loop);
1802
1803 align(32, 12);
1804 bind(profile_continue);
1805 }
1806 }
1807
1808 // Add a monitor (see frame_ppc.hpp).
1809 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) {
1810
1811 // Very-local scratch registers.
1812 const Register esp = Rtemp1;
1813 const Register slot = Rtemp2;
1814
1815 // Extracted monitor_size.
1816 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes();
1817 assert(Assembler::is_aligned((unsigned int)monitor_size,
1818 (unsigned int)frame::alignment_in_bytes),
1819 "size of a monitor must respect alignment of SP");
1820
1821 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor
1822 subf(Rtemp2, esp, R1_SP); // esp contains fp
1823 sradi(Rtemp2, Rtemp2, Interpreter::logStackElementSize);
1824 // Store relativized top_frame_sp
1825 std(Rtemp2, _ijava_state_neg(top_frame_sp), esp); // esp contains fp
1826
1827 // Shuffle expression stack down. Recall that stack_base points
1828 // just above the new expression stack bottom. Old_tos and new_tos
1829 // are used to scan thru the old and new expression stacks.
1830 if (!stack_is_empty) {
1831 Label copy_slot, copy_slot_finished;
1832 const Register n_slots = slot;
1833
1834 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack).
1835 subf(n_slots, esp, R26_monitor);
1836 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy.
1837 assert(LogBytesPerWord == 3, "conflicts assembler instructions");
1838 beq(CR0, copy_slot_finished); // Nothing to copy.
1839
1840 mtctr(n_slots);
1841
1842 // loop
1843 bind(copy_slot);
1844 ld(slot, 0, esp); // Move expression stack down.
1845 std(slot, -monitor_size, esp); // distance = monitor_size
1846 addi(esp, esp, BytesPerWord);
1847 bdnz(copy_slot);
1848
1849 bind(copy_slot_finished);
1850 }
1851
1852 addi(R15_esp, R15_esp, -monitor_size);
1853 addi(R26_monitor, R26_monitor, -monitor_size);
1854
1855 // Restart interpreter
1856 }
1857
1858 // ============================================================================
1859 // Java locals access
1860
1861 // Load a local variable at index in Rindex into register Rdst_value.
1862 // Also puts address of local into Rdst_address as a service.
1863 // Kills:
1864 // - Rdst_value
1865 // - Rdst_address
1866 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) {
1867 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
1868 subf(Rdst_address, Rdst_address, R18_locals);
1869 lwz(Rdst_value, 0, Rdst_address);
1870 }
1871
1872 // Load a local variable at index in Rindex into register Rdst_value.
1873 // Also puts address of local into Rdst_address as a service.
1874 // Kills:
1875 // - Rdst_value
1876 // - Rdst_address
1877 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) {
1878 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
1879 subf(Rdst_address, Rdst_address, R18_locals);
1880 ld(Rdst_value, -8, Rdst_address);
1881 }
1882
1883 // Load a local variable at index in Rindex into register Rdst_value.
1884 // Also puts address of local into Rdst_address as a service.
1885 // Input:
1886 // - Rindex: slot nr of local variable
1887 // Kills:
1888 // - Rdst_value
1889 // - Rdst_address
1890 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value,
1891 Register Rdst_address,
1892 Register Rindex) {
1893 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
1894 subf(Rdst_address, Rdst_address, R18_locals);
1895 ld(Rdst_value, 0, Rdst_address);
1896 }
1897
1898 // Load a local variable at index in Rindex into register Rdst_value.
1899 // Also puts address of local into Rdst_address as a service.
1900 // Kills:
1901 // - Rdst_value
1902 // - Rdst_address
1903 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value,
1904 Register Rdst_address,
1905 Register Rindex) {
1906 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
1907 subf(Rdst_address, Rdst_address, R18_locals);
1908 lfs(Rdst_value, 0, Rdst_address);
1909 }
1910
1911 // Load a local variable at index in Rindex into register Rdst_value.
1912 // Also puts address of local into Rdst_address as a service.
1913 // Kills:
1914 // - Rdst_value
1915 // - Rdst_address
1916 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value,
1917 Register Rdst_address,
1918 Register Rindex) {
1919 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
1920 subf(Rdst_address, Rdst_address, R18_locals);
1921 lfd(Rdst_value, -8, Rdst_address);
1922 }
1923
1924 // Store an int value at local variable slot Rindex.
1925 // Kills:
1926 // - Rindex
1927 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) {
1928 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
1929 subf(Rindex, Rindex, R18_locals);
1930 stw(Rvalue, 0, Rindex);
1931 }
1932
1933 // Store a long value at local variable slot Rindex.
1934 // Kills:
1935 // - Rindex
1936 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) {
1937 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
1938 subf(Rindex, Rindex, R18_locals);
1939 std(Rvalue, -8, Rindex);
1940 }
1941
1942 // Store an oop value at local variable slot Rindex.
1943 // Kills:
1944 // - Rindex
1945 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) {
1946 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
1947 subf(Rindex, Rindex, R18_locals);
1948 std(Rvalue, 0, Rindex);
1949 }
1950
1951 // Store an int value at local variable slot Rindex.
1952 // Kills:
1953 // - Rindex
1954 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) {
1955 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
1956 subf(Rindex, Rindex, R18_locals);
1957 stfs(Rvalue, 0, Rindex);
1958 }
1959
1960 // Store an int value at local variable slot Rindex.
1961 // Kills:
1962 // - Rindex
1963 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) {
1964 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
1965 subf(Rindex, Rindex, R18_locals);
1966 stfd(Rvalue, -8, Rindex);
1967 }
1968
1969 // Read pending exception from thread and jump to interpreter.
1970 // Throw exception entry if one if pending. Fall through otherwise.
1971 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) {
1972 assert_different_registers(Rscratch1, Rscratch2, R3);
1973 Register Rexception = Rscratch1;
1974 Register Rtmp = Rscratch2;
1975 Label Ldone;
1976 // Get pending exception oop.
1977 ld(Rexception, thread_(pending_exception));
1978 cmpdi(CR0, Rexception, 0);
1979 beq(CR0, Ldone);
1980 li(Rtmp, 0);
1981 mr_if_needed(R3, Rexception);
1982 std(Rtmp, thread_(pending_exception)); // Clear exception in thread
1983 if (Interpreter::rethrow_exception_entry() != nullptr) {
1984 // Already got entry address.
1985 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry());
1986 } else {
1987 // Dynamically load entry address.
1988 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true);
1989 ld(Rtmp, simm16_rest, Rtmp);
1990 }
1991 mtctr(Rtmp);
1992 save_interpreter_state(Rtmp);
1993 bctr();
1994
1995 align(32, 12);
1996 bind(Ldone);
1997 }
1998
1999 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions, Label* last_java_pc) {
2000 save_interpreter_state(R11_scratch1);
2001
2002 MacroAssembler::call_VM(oop_result, entry_point, false /*check_exceptions*/, last_java_pc);
2003
2004 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2005
2006 check_and_handle_popframe(R11_scratch1);
2007 check_and_handle_earlyret(R11_scratch1);
2008 // Now check exceptions manually.
2009 if (check_exceptions) {
2010 check_and_forward_exception(R11_scratch1, R12_scratch2);
2011 }
2012 }
2013
2014 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2015 Register arg_1, bool check_exceptions) {
2016 // ARG1 is reserved for the thread.
2017 mr_if_needed(R4_ARG2, arg_1);
2018 call_VM(oop_result, entry_point, check_exceptions);
2019 }
2020
2021 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result, address entry_point,
2022 Register arg_1,
2023 bool check_exceptions) {
2024 if (!Continuations::enabled()) {
2025 call_VM(oop_result, entry_point, arg_1, check_exceptions);
2026 return;
2027 }
2028 call_VM_preemptable(oop_result, entry_point, arg_1, noreg /* arg_2 */, check_exceptions);
2029 }
2030
2031 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result, address entry_point,
2032 Register arg_1, Register arg_2,
2033 bool check_exceptions) {
2034 if (!Continuations::enabled()) {
2035 call_VM(oop_result, entry_point, arg_1, arg_2, check_exceptions);
2036 return;
2037 }
2038
2039 Label resume_pc, not_preempted;
2040 Register tmp = R11_scratch1;
2041 assert_different_registers(arg_1, tmp);
2042 assert_different_registers(arg_2, tmp);
2043
2044 #ifdef ASSERT
2045 asm_assert_mem8_is_zero(in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread,
2046 "Should not have alternate return address set");
2047 // We check this counter in patch_return_pc_with_preempt_stub() during freeze.
2048 lwa(tmp, in_bytes(JavaThread::interp_at_preemptable_vmcall_cnt_offset()), R16_thread);
2049 addi(tmp, tmp, 1);
2050 cmpwi(CR0, tmp, 0);
2051 stw(tmp, in_bytes(JavaThread::interp_at_preemptable_vmcall_cnt_offset()), R16_thread);
2052 asm_assert(gt, "call_VM_preemptable: should be > 0");
2053 #endif // ASSERT
2054
2055 // Preserve 2 registers
2056 assert(nonvolatile_accross_vthread_preemtion(R31) && nonvolatile_accross_vthread_preemtion(R24), "");
2057 ld(R3_ARG1, _abi0(callers_sp), R1_SP); // load FP
2058 std(R31, _ijava_state_neg(lresult), R3_ARG1);
2059 std(R24, _ijava_state_neg(fresult), R3_ARG1);
2060
2061 // We set resume_pc as last java pc. It will be saved if the vthread gets preempted.
2062 // Later execution will continue right there.
2063 mr_if_needed(R4_ARG2, arg_1);
2064 assert(arg_2 != R4_ARG2, "smashed argument");
2065 mr_if_needed(R5_ARG3, arg_2, true /* allow_noreg */);
2066 push_cont_fastpath();
2067 call_VM(noreg /* oop_result */, entry_point, false /*check_exceptions*/, &resume_pc /* last_java_pc */);
2068 pop_cont_fastpath();
2069
2070 #ifdef ASSERT
2071 lwa(tmp, in_bytes(JavaThread::interp_at_preemptable_vmcall_cnt_offset()), R16_thread);
2072 addi(tmp, tmp, -1);
2073 cmpwi(CR0, tmp, 0);
2074 stw(tmp, in_bytes(JavaThread::interp_at_preemptable_vmcall_cnt_offset()), R16_thread);
2075 asm_assert(ge, "call_VM_preemptable: should be >= 0");
2076 #endif // ASSERT
2077
2078 // Jump to handler if the call was preempted
2079 ld(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
2080 cmpdi(CR0, R0, 0);
2081 beq(CR0, not_preempted);
2082 // Preempted. Frames are already frozen on heap.
2083 mtlr(R0);
2084 li(R0, 0);
2085 std(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
2086 blr();
2087
2088 bind(resume_pc); // Location to resume execution
2089 restore_after_resume(noreg /* fp */);
2090
2091 bind(not_preempted);
2092 if (check_exceptions) {
2093 check_and_forward_exception(R11_scratch1, R12_scratch2);
2094 }
2095 if (oop_result->is_valid()) {
2096 get_vm_result_oop(oop_result);
2097 }
2098 }
2099
2100 void InterpreterMacroAssembler::restore_after_resume(Register fp) {
2101 const address resume_adapter = TemplateInterpreter::cont_resume_interpreter_adapter();
2102 add_const_optimized(R31, R29_TOC, MacroAssembler::offset_to_global_toc(resume_adapter));
2103 mtctr(R31);
2104 bctrl();
2105 #ifdef ASSERT
2106 // Assert FP is in R11_scratch1 (see generate_cont_resume_interpreter_adapter())
2107 {
2108 Label ok;
2109 ld(R12_scratch2, 0, R1_SP); // load fp
2110 cmpd(CR0, R12_scratch2, R11_scratch1);
2111 beq(CR0, ok);
2112 stop(FILE_AND_LINE ": FP is expected in R11_scratch1");
2113 bind(ok);
2114 }
2115 #endif
2116 if (fp != noreg && fp != R11_scratch1) {
2117 mr(fp, R11_scratch1);
2118 }
2119 }
2120
2121 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2122 Register arg_1, Register arg_2,
2123 bool check_exceptions) {
2124 // ARG1 is reserved for the thread.
2125 mr_if_needed(R4_ARG2, arg_1);
2126 assert(arg_2 != R4_ARG2, "smashed argument");
2127 mr_if_needed(R5_ARG3, arg_2);
2128 call_VM(oop_result, entry_point, check_exceptions);
2129 }
2130
2131 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2132 Register arg_1, Register arg_2, Register arg_3,
2133 bool check_exceptions) {
2134 // ARG1 is reserved for the thread.
2135 mr_if_needed(R4_ARG2, arg_1);
2136 assert(arg_2 != R4_ARG2, "smashed argument");
2137 mr_if_needed(R5_ARG3, arg_2);
2138 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument");
2139 mr_if_needed(R6_ARG4, arg_3);
2140 call_VM(oop_result, entry_point, check_exceptions);
2141 }
2142
2143 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) {
2144 ld(scratch, 0, R1_SP);
2145 subf(R0, scratch, R15_esp);
2146 sradi(R0, R0, Interpreter::logStackElementSize);
2147 std(R0, _ijava_state_neg(esp), scratch);
2148 std(R14_bcp, _ijava_state_neg(bcp), scratch);
2149 subf(R0, scratch, R26_monitor);
2150 sradi(R0, R0, Interpreter::logStackElementSize);
2151 std(R0, _ijava_state_neg(monitors), scratch);
2152 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); }
2153 // Other entries should be unchanged.
2154 }
2155
2156 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only, bool restore_top_frame_sp) {
2157 ld_ptr(scratch, _abi0(callers_sp), R1_SP); // Load frame pointer.
2158 if (restore_top_frame_sp) {
2159 // After thawing the top frame of a continuation we reach here with frame::java_abi.
2160 // therefore we have to restore top_frame_sp before the assertion below.
2161 assert(!bcp_and_mdx_only, "chose other registers");
2162 Register tfsp = R18_locals;
2163 Register scratch2 = R26_monitor;
2164 ld(tfsp, _ijava_state_neg(top_frame_sp), scratch);
2165 // Derelativize top_frame_sp
2166 sldi(tfsp, tfsp, Interpreter::logStackElementSize);
2167 add(tfsp, tfsp, scratch);
2168 resize_frame_absolute(tfsp, scratch2, R0);
2169 }
2170 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception).
2171 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code.
2172 if (!bcp_and_mdx_only) {
2173 // Following ones are Metadata.
2174 ld(R19_method, _ijava_state_neg(method), scratch);
2175 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch);
2176 // Following ones are stack addresses and don't require reload.
2177 // Derelativize esp
2178 ld(R15_esp, _ijava_state_neg(esp), scratch);
2179 sldi(R15_esp, R15_esp, Interpreter::logStackElementSize);
2180 add(R15_esp, R15_esp, scratch);
2181 ld(R18_locals, _ijava_state_neg(locals), scratch);
2182 sldi(R18_locals, R18_locals, Interpreter::logStackElementSize);
2183 add(R18_locals, R18_locals, scratch);
2184 ld(R26_monitor, _ijava_state_neg(monitors), scratch);
2185 // Derelativize monitors
2186 sldi(R26_monitor, R26_monitor, Interpreter::logStackElementSize);
2187 add(R26_monitor, R26_monitor, scratch);
2188 }
2189 #ifdef ASSERT
2190 {
2191 Label Lok;
2192 subf(R0, R1_SP, scratch);
2193 cmpdi(CR0, R0, frame::top_ijava_frame_abi_size + frame::ijava_state_size);
2194 bge(CR0, Lok);
2195 stop("frame too small (restore istate)");
2196 bind(Lok);
2197 }
2198 #endif
2199 }
2200
2201 void InterpreterMacroAssembler::get_method_counters(Register method,
2202 Register Rcounters,
2203 Label& skip) {
2204 BLOCK_COMMENT("Load and ev. allocate counter object {");
2205 Label has_counters;
2206 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2207 cmpdi(CR0, Rcounters, 0);
2208 bne(CR0, has_counters);
2209 call_VM(noreg, CAST_FROM_FN_PTR(address,
2210 InterpreterRuntime::build_method_counters), method);
2211 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2212 cmpdi(CR0, Rcounters, 0);
2213 beq(CR0, skip); // No MethodCounters, OutOfMemory.
2214 BLOCK_COMMENT("} Load and ev. allocate counter object");
2215
2216 bind(has_counters);
2217 }
2218
2219 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters,
2220 Register iv_be_count,
2221 Register Rtmp_r0) {
2222 assert(UseCompiler, "incrementing must be useful");
2223 Register invocation_count = iv_be_count;
2224 Register backedge_count = Rtmp_r0;
2225 int delta = InvocationCounter::count_increment;
2226
2227 // Load each counter in a register.
2228 // ld(inv_counter, Rtmp);
2229 // ld(be_counter, Rtmp2);
2230 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() +
2231 InvocationCounter::counter_offset());
2232 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() +
2233 InvocationCounter::counter_offset());
2234
2235 BLOCK_COMMENT("Increment profiling counters {");
2236
2237 // Load the backedge counter.
2238 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int
2239 // Mask the backedge counter.
2240 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value);
2241
2242 // Load the invocation counter.
2243 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int
2244 // Add the delta to the invocation counter and store the result.
2245 addi(invocation_count, invocation_count, delta);
2246 // Store value.
2247 stw(invocation_count, inv_counter_offset, Rcounters);
2248
2249 // Add invocation counter + backedge counter.
2250 add(iv_be_count, backedge_count, invocation_count);
2251
2252 // Note that this macro must leave the backedge_count + invocation_count in
2253 // register iv_be_count!
2254 BLOCK_COMMENT("} Increment profiling counters");
2255 }
2256
2257 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2258 if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); }
2259 }
2260
2261 // Local helper function for the verify_oop_or_return_address macro.
2262 static bool verify_return_address(Method* m, int bci) {
2263 #ifndef PRODUCT
2264 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci;
2265 // Assume it is a valid return address if it is inside m and is preceded by a jsr.
2266 if (!m->contains(pc)) return false;
2267 address jsr_pc;
2268 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2269 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2270 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2271 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2272 #endif // PRODUCT
2273 return false;
2274 }
2275
2276 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2277 if (!VerifyOops) return;
2278
2279 // The VM documentation for the astore[_wide] bytecode allows
2280 // the TOS to be not only an oop but also a return address.
2281 Label test;
2282 Label skip;
2283 // See if it is an address (in the current method):
2284
2285 const int log2_bytecode_size_limit = 16;
2286 srdi_(Rtmp, reg, log2_bytecode_size_limit);
2287 bne(CR0, test);
2288
2289 address fd = CAST_FROM_FN_PTR(address, verify_return_address);
2290 const int nbytes_save = MacroAssembler::num_volatile_regs * 8;
2291 save_volatile_gprs(R1_SP, -nbytes_save); // except R0
2292 save_LR_CR(Rtmp); // Save in old frame.
2293 push_frame_reg_args(nbytes_save, Rtmp);
2294
2295 load_const_optimized(Rtmp, fd, R0);
2296 mr_if_needed(R4_ARG2, reg);
2297 mr(R3_ARG1, R19_method);
2298 call_c(Rtmp); // call C
2299
2300 pop_frame();
2301 restore_LR_CR(Rtmp);
2302 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
2303 b(skip);
2304
2305 // Perform a more elaborate out-of-line call.
2306 // Not an address; verify it:
2307 bind(test);
2308 verify_oop(reg);
2309 bind(skip);
2310 }
2311
2312 // Inline assembly for:
2313 //
2314 // if (thread is in interp_only_mode) {
2315 // InterpreterRuntime::post_method_entry();
2316 // }
2317 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) ||
2318 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) {
2319 // SharedRuntime::jvmpi_method_entry(method, receiver);
2320 // }
2321 void InterpreterMacroAssembler::notify_method_entry() {
2322 // JVMTI
2323 // Whenever JVMTI puts a thread in interp_only_mode, method
2324 // entry/exit events are sent for that thread to track stack
2325 // depth. If it is possible to enter interp_only_mode we add
2326 // the code to check if the event should be sent.
2327 if (JvmtiExport::can_post_interpreter_events()) {
2328 Label jvmti_post_done;
2329
2330 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2331 cmpwi(CR0, R0, 0);
2332 beq(CR0, jvmti_post_done);
2333 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2334
2335 bind(jvmti_post_done);
2336 }
2337 }
2338
2339 // Inline assembly for:
2340 //
2341 // if (thread is in interp_only_mode) {
2342 // // save result
2343 // InterpreterRuntime::post_method_exit();
2344 // // restore result
2345 // }
2346 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) {
2347 // // save result
2348 // SharedRuntime::jvmpi_method_exit();
2349 // // restore result
2350 // }
2351 //
2352 // Native methods have their result stored in d_tmp and l_tmp.
2353 // Java methods have their result stored in the expression stack.
2354 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state,
2355 NotifyMethodExitMode mode, bool check_exceptions) {
2356 // JVMTI
2357 // Whenever JVMTI puts a thread in interp_only_mode, method
2358 // entry/exit events are sent for that thread to track stack
2359 // depth. If it is possible to enter interp_only_mode we add
2360 // the code to check if the event should be sent.
2361 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2362 Label jvmti_post_done;
2363
2364 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2365 cmpwi(CR0, R0, 0);
2366 beq(CR0, jvmti_post_done);
2367 if (!is_native_method) { push(state); } // Expose tos to GC.
2368 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), check_exceptions);
2369 if (!is_native_method) { pop(state); }
2370
2371 align(32, 12);
2372 bind(jvmti_post_done);
2373 }
2374
2375 // Dtrace support not implemented.
2376 }
--- EOF ---