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 mtfprd(d, l);
312 }
313
314 void InterpreterMacroAssembler::move_d_to_l(FloatRegister d, Register l) {
315 mffprd(l, d);
316 }
317
318 void InterpreterMacroAssembler::push(TosState state) {
319 switch (state) {
320 case atos: push_ptr(); break;
321 case btos:
322 case ztos:
323 case ctos:
324 case stos:
325 case itos: push_i(); break;
326 case ltos: push_l(); break;
327 case ftos: push_f(); break;
328 case dtos: push_d(); break;
329 case vtos: /* nothing to do */ break;
330 default : ShouldNotReachHere();
331 }
332 }
333
334 void InterpreterMacroAssembler::pop(TosState state) {
335 switch (state) {
336 case atos: pop_ptr(); break;
337 case btos:
338 case ztos:
339 case ctos:
340 case stos:
341 case itos: pop_i(); break;
342 case ltos: pop_l(); break;
343 case ftos: pop_f(); break;
344 case dtos: pop_d(); break;
345 case vtos: /* nothing to do */ break;
346 default : ShouldNotReachHere();
347 }
348 verify_oop(R17_tos, state);
349 }
350
351 void InterpreterMacroAssembler::empty_expression_stack() {
352 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
353 }
354
355 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset,
356 Register Rdst,
357 signedOrNot is_signed) {
358 #if defined(VM_LITTLE_ENDIAN)
359 if (bcp_offset) {
360 load_const_optimized(Rdst, bcp_offset);
361 lhbrx(Rdst, R14_bcp, Rdst);
362 } else {
363 lhbrx(Rdst, R14_bcp);
364 }
365 if (is_signed == Signed) {
366 extsh(Rdst, Rdst);
367 }
368 #else
369 // Read Java big endian format.
370 if (is_signed == Signed) {
371 lha(Rdst, bcp_offset, R14_bcp);
372 } else {
373 lhz(Rdst, bcp_offset, R14_bcp);
374 }
375 #endif
376 }
377
378 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset,
379 Register Rdst,
380 signedOrNot is_signed) {
381 #if defined(VM_LITTLE_ENDIAN)
382 if (bcp_offset) {
383 load_const_optimized(Rdst, bcp_offset);
384 lwbrx(Rdst, R14_bcp, Rdst);
385 } else {
386 lwbrx(Rdst, R14_bcp);
387 }
388 if (is_signed == Signed) {
389 extsw(Rdst, Rdst);
390 }
391 #else
392 // Read Java big endian format.
393 if (bcp_offset & 3) { // Offset unaligned?
394 load_const_optimized(Rdst, bcp_offset);
395 if (is_signed == Signed) {
396 lwax(Rdst, R14_bcp, Rdst);
397 } else {
398 lwzx(Rdst, R14_bcp, Rdst);
399 }
400 } else {
401 if (is_signed == Signed) {
402 lwa(Rdst, bcp_offset, R14_bcp);
403 } else {
404 lwz(Rdst, bcp_offset, R14_bcp);
405 }
406 }
407 #endif
408 }
409
410
411 // Load the constant pool cache index from the bytecode stream.
412 //
413 // Kills / writes:
414 // - Rdst, Rscratch
415 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset,
416 size_t index_size) {
417 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
418 // Cache index is always in the native format, courtesy of Rewriter.
419 if (index_size == sizeof(u2)) {
420 lhz(Rdst, bcp_offset, R14_bcp);
421 } else if (index_size == sizeof(u4)) {
422 if (bcp_offset & 3) {
423 load_const_optimized(Rdst, bcp_offset);
424 lwax(Rdst, R14_bcp, Rdst);
425 } else {
426 lwa(Rdst, bcp_offset, R14_bcp);
427 }
428 } else if (index_size == sizeof(u1)) {
429 lbz(Rdst, bcp_offset, R14_bcp);
430 } else {
431 ShouldNotReachHere();
432 }
433 // Rdst now contains cp cache index.
434 }
435
436 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format)
437 // from (Rsrc)+offset.
438 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset,
439 signedOrNot is_signed) {
440 #if defined(VM_LITTLE_ENDIAN)
441 if (offset) {
442 load_const_optimized(Rdst, offset);
443 lwbrx(Rdst, Rdst, Rsrc);
444 } else {
445 lwbrx(Rdst, Rsrc);
446 }
447 if (is_signed == Signed) {
448 extsw(Rdst, Rdst);
449 }
450 #else
451 if (is_signed == Signed) {
452 lwa(Rdst, offset, Rsrc);
453 } else {
454 lwz(Rdst, offset, Rsrc);
455 }
456 #endif
457 }
458
459 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
460 // Get index out of bytecode pointer
461 get_cache_index_at_bcp(index, 1, sizeof(u4));
462
463 // Get address of invokedynamic array
464 ld_ptr(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset()), R27_constPoolCache);
465 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
466 sldi(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
467 addi(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
468 add(cache, cache, index);
469 }
470
471 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
472 // Get index out of bytecode pointer
473 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
474 // Take shortcut if the size is a power of 2
475 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
476 // Scale index by power of 2
477 sldi(index, index, log2i_exact(sizeof(ResolvedFieldEntry)));
478 } else {
479 // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
480 mulli(index, index, sizeof(ResolvedFieldEntry));
481 }
482 // Get address of field entries array
483 ld_ptr(cache, in_bytes(ConstantPoolCache::field_entries_offset()), R27_constPoolCache);
484 addi(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
485 add(cache, cache, index);
486 }
487
488 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
489 // Get index out of bytecode pointer
490 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
491 // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
492 mulli(index, index, sizeof(ResolvedMethodEntry));
493
494 // Get address of field entries array
495 ld_ptr(cache, ConstantPoolCache::method_entries_offset(), R27_constPoolCache);
496 addi(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
497 add(cache, cache, index); // method_entries + base_offset + scaled index
498 }
499
500 // Load object from cpool->resolved_references(index).
501 // Kills:
502 // - index
503 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index,
504 Register tmp1, Register tmp2,
505 Label *L_handle_null) {
506 assert_different_registers(result, index, tmp1, tmp2);
507 assert(index->is_nonvolatile(), "needs to survive C-call in resolve_oop_handle");
508 get_constant_pool(result);
509
510 // Convert from field index to resolved_references() index and from
511 // word index to byte offset. Since this is a java object, it can be compressed.
512 sldi(index, index, LogBytesPerHeapOop);
513 // Load pointer for resolved_references[] objArray.
514 ld(result, ConstantPool::cache_offset(), result);
515 ld(result, ConstantPoolCache::resolved_references_offset(), result);
516 resolve_oop_handle(result, tmp1, tmp2, MacroAssembler::PRESERVATION_NONE);
517 #ifdef ASSERT
518 Label index_ok;
519 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result);
520 sldi(R0, R0, LogBytesPerHeapOop);
521 cmpd(CR0, index, R0);
522 blt(CR0, index_ok);
523 stop("resolved reference index out of bounds");
524 bind(index_ok);
525 #endif
526 // Add in the index.
527 add(result, index, result);
528 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result,
529 tmp1, tmp2,
530 MacroAssembler::PRESERVATION_NONE,
531 0, L_handle_null);
532 }
533
534 // load cpool->resolved_klass_at(index)
535 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool, Register Roffset, Register Rklass) {
536 // int value = *(Rcpool->int_at_addr(which));
537 // int resolved_klass_index = extract_low_short_from_int(value);
538 add(Roffset, Rcpool, Roffset);
539 #if defined(VM_LITTLE_ENDIAN)
540 lhz(Roffset, sizeof(ConstantPool), Roffset); // Roffset = resolved_klass_index
541 #else
542 lhz(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index
543 #endif
544
545 ld(Rklass, ConstantPool::resolved_klasses_offset(), Rcpool); // Rklass = Rcpool->_resolved_klasses
546
547 sldi(Roffset, Roffset, LogBytesPerWord);
548 addi(Roffset, Roffset, Array<Klass*>::base_offset_in_bytes());
549 isync(); // Order load of instance Klass wrt. tags.
550 ldx(Rklass, Rklass, Roffset);
551 }
552
553 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
554 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2.
555 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1,
556 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) {
557 // Profile the not-null value's klass.
558 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2);
559 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
560 }
561
562 // Separate these two to allow for delay slot in middle.
563 // These are used to do a test and full jump to exception-throwing code.
564
565 // Check that index is in range for array, then shift index by index_shift,
566 // and put arrayOop + shifted_index into res.
567 // Note: res is still shy of address by array offset into object.
568
569 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex,
570 int index_shift, Register Rtmp, Register Rres) {
571 // Check that index is in range for array, then shift index by index_shift,
572 // and put arrayOop + shifted_index into res.
573 // Note: res is still shy of address by array offset into object.
574 // Kills:
575 // - Rindex
576 // Writes:
577 // - Rres: Address that corresponds to the array index if check was successful.
578 verify_oop(Rarray);
579 const Register Rlength = R0;
580 const Register RsxtIndex = Rtmp;
581 Label LisNull, LnotOOR;
582
583 // Array nullcheck
584 if (!ImplicitNullChecks) {
585 cmpdi(CR0, Rarray, 0);
586 beq(CR0, LisNull);
587 } else {
588 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex);
589 }
590
591 // Rindex might contain garbage in upper bits (remember that we don't sign extend
592 // during integer arithmetic operations). So kill them and put value into same register
593 // where ArrayIndexOutOfBounds would expect the index in.
594 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit
595
596 // Index check
597 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray);
598 cmplw(CR0, Rindex, Rlength);
599 sldi(RsxtIndex, RsxtIndex, index_shift);
600 blt(CR0, LnotOOR);
601 // Index should be in R17_tos, array should be in R4_ARG2.
602 mr_if_needed(R17_tos, Rindex);
603 mr_if_needed(R4_ARG2, Rarray);
604 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
605 mtctr(Rtmp);
606 bctr();
607
608 if (!ImplicitNullChecks) {
609 bind(LisNull);
610 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry);
611 mtctr(Rtmp);
612 bctr();
613 }
614
615 align(32, 16);
616 bind(LnotOOR);
617
618 // Calc address
619 add(Rres, RsxtIndex, Rarray);
620 }
621
622 void InterpreterMacroAssembler::index_check(Register array, Register index,
623 int index_shift, Register tmp, Register res) {
624 // pop array
625 pop_ptr(array);
626
627 // check array
628 index_check_without_pop(array, index, index_shift, tmp, res);
629 }
630
631 void InterpreterMacroAssembler::get_const(Register Rdst) {
632 ld(Rdst, in_bytes(Method::const_offset()), R19_method);
633 }
634
635 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
636 get_const(Rdst);
637 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
638 }
639
640 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
641 get_constant_pool(Rdst);
642 ld(Rdst, ConstantPool::cache_offset(), Rdst);
643 }
644
645 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
646 get_constant_pool(Rcpool);
647 ld(Rtags, ConstantPool::tags_offset(), Rcpool);
648 }
649
650 // Unlock if synchronized method.
651 //
652 // Unlock the receiver if this is a synchronized method.
653 // Unlock any Java monitors from synchronized blocks.
654 //
655 // If there are locked Java monitors
656 // If throw_monitor_exception
657 // throws IllegalMonitorStateException
658 // Else if install_monitor_exception
659 // installs IllegalMonitorStateException
660 // Else
661 // no error processing
662 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
663 bool throw_monitor_exception,
664 bool install_monitor_exception) {
665 Label Lunlocked, Lno_unlock;
666 {
667 Register Rdo_not_unlock_flag = R11_scratch1;
668 Register Raccess_flags = R12_scratch2;
669
670 // Check if synchronized method or unlocking prevented by
671 // JavaThread::do_not_unlock_if_synchronized flag.
672 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
673 lhz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method);
674 li(R0, 0);
675 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag
676
677 push(state);
678
679 // Skip if we don't have to unlock.
680 testbitdi(CR0, R0, Raccess_flags, JVM_ACC_SYNCHRONIZED_BIT);
681 beq(CR0, Lunlocked);
682
683 cmpwi(CR0, Rdo_not_unlock_flag, 0);
684 bne(CR0, Lno_unlock);
685 }
686
687 // Unlock
688 {
689 Register Rmonitor_base = R11_scratch1;
690
691 Label Lunlock;
692 // If it's still locked, everything is ok, unlock it.
693 ld(Rmonitor_base, 0, R1_SP);
694 addi(Rmonitor_base, Rmonitor_base,
695 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
696
697 ld(R0, BasicObjectLock::obj_offset(), Rmonitor_base);
698 cmpdi(CR0, R0, 0);
699 bne(CR0, Lunlock);
700
701 // If it's already unlocked, throw exception.
702 if (throw_monitor_exception) {
703 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
704 should_not_reach_here();
705 } else {
706 if (install_monitor_exception) {
707 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
708 b(Lunlocked);
709 }
710 }
711
712 bind(Lunlock);
713 unlock_object(Rmonitor_base);
714 }
715
716 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not.
717 bind(Lunlocked);
718 {
719 Label Lexception, Lrestart;
720 Register Rcurrent_obj_addr = R11_scratch1;
721 const int delta = frame::interpreter_frame_monitor_size_in_bytes();
722 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords");
723
724 bind(Lrestart);
725 // Set up search loop: Calc num of iterations.
726 {
727 Register Riterations = R12_scratch2;
728 Register Rmonitor_base = Rcurrent_obj_addr;
729 ld(Rmonitor_base, 0, R1_SP);
730 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base
731
732 subf_(Riterations, R26_monitor, Rmonitor_base);
733 ble(CR0, Lno_unlock);
734
735 addi(Rcurrent_obj_addr, Rmonitor_base,
736 in_bytes(BasicObjectLock::obj_offset()) - frame::interpreter_frame_monitor_size_in_bytes());
737 // Check if any monitor is on stack, bail out if not
738 srdi(Riterations, Riterations, exact_log2(delta));
739 mtctr(Riterations);
740 }
741
742 // The search loop: Look for locked monitors.
743 {
744 const Register Rcurrent_obj = R0;
745 Label Lloop;
746
747 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
748 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
749 bind(Lloop);
750
751 // Check if current entry is used.
752 cmpdi(CR0, Rcurrent_obj, 0);
753 bne(CR0, Lexception);
754 // Preload next iteration's compare value.
755 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
756 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
757 bdnz(Lloop);
758 }
759 // Fell through: Everything's unlocked => finish.
760 b(Lno_unlock);
761
762 // An object is still locked => need to throw exception.
763 bind(Lexception);
764 if (throw_monitor_exception) {
765 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
766 should_not_reach_here();
767 } else {
768 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
769 // Unlock does not block, so don't have to worry about the frame.
770 Register Rmonitor_addr = R11_scratch1;
771 addi(Rmonitor_addr, Rcurrent_obj_addr, -in_bytes(BasicObjectLock::obj_offset()) + delta);
772 unlock_object(Rmonitor_addr);
773 if (install_monitor_exception) {
774 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
775 }
776 b(Lrestart);
777 }
778 }
779
780 align(32, 12);
781 bind(Lno_unlock);
782 pop(state);
783 }
784
785 // Support function for remove_activation & Co.
786 void InterpreterMacroAssembler::load_fp(Register fp) {
787 ld(fp, _abi0(callers_sp), R1_SP); // *SP
788 }
789
790 void InterpreterMacroAssembler::remove_top_frame_given_fp(Register fp, Register sender_sp, Register sender_fp,
791 Register return_pc, Register temp) {
792 assert_different_registers(sender_sp, sender_fp, return_pc, temp);
793 ld(sender_sp, _ijava_state_neg(sender_sp), fp);
794 ld(sender_fp, _abi0(callers_sp), fp); // **SP
795 if (return_pc != noreg) {
796 ld(return_pc, _abi0(lr), fp); // last usage of fp, register can be reused
797 }
798 subf(temp, R1_SP, sender_sp); // sender_sp - SP
799 stdux(sender_fp, R1_SP, temp); // atomically set *(SP = sender_sp) = sender_fp
800 }
801
802 void InterpreterMacroAssembler::merge_frames(Register sender_sp, Register return_pc,
803 Register temp1, Register temp2) {
804 Register fp = temp1, sender_fp = temp2;
805 load_fp(fp);
806 remove_top_frame_given_fp(fp, sender_sp, sender_fp, return_pc, /* temp */ fp);
807 }
808
809 void InterpreterMacroAssembler::narrow(Register result) {
810 Register ret_type = R11_scratch1;
811 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
812 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1);
813
814 Label notBool, notByte, notChar, done;
815
816 // common case first
817 cmpwi(CR0, ret_type, T_INT);
818 beq(CR0, done);
819
820 cmpwi(CR0, ret_type, T_BOOLEAN);
821 bne(CR0, notBool);
822 andi(result, result, 0x1);
823 b(done);
824
825 bind(notBool);
826 cmpwi(CR0, ret_type, T_BYTE);
827 bne(CR0, notByte);
828 extsb(result, result);
829 b(done);
830
831 bind(notByte);
832 cmpwi(CR0, ret_type, T_CHAR);
833 bne(CR0, notChar);
834 andi(result, result, 0xffff);
835 b(done);
836
837 bind(notChar);
838 // cmpwi(CR0, ret_type, T_SHORT); // all that's left
839 // bne(CR0, done);
840 extsh(result, result);
841
842 // Nothing to do for T_INT
843 bind(done);
844 }
845
846 // Remove activation.
847 //
848 // Apply stack watermark barrier.
849 // Unlock the receiver if this is a synchronized method.
850 // Unlock any Java monitors from synchronized blocks.
851 // Remove the activation from the stack.
852 //
853 // If there are locked Java monitors
854 // If throw_monitor_exception
855 // throws IllegalMonitorStateException
856 // Else if install_monitor_exception
857 // installs IllegalMonitorStateException
858 // Else
859 // no error processing
860 void InterpreterMacroAssembler::remove_activation(TosState state,
861 bool throw_monitor_exception,
862 bool install_monitor_exception) {
863 BLOCK_COMMENT("remove_activation {");
864
865 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
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, fast_path;
871 Register fp = R22_tmp2;
872 load_fp(fp);
873
874 JFR_ONLY(enter_jfr_critical_section();)
875 safepoint_poll(slow_path, R11_scratch1, true /* at_return */, false /* in_nmethod */);
876 b(fast_path);
877 bind(slow_path);
878 push(state);
879 set_last_Java_frame(R1_SP, noreg);
880 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), R16_thread);
881 reset_last_Java_frame();
882 pop(state);
883 align(32);
884 bind(fast_path);
885
886 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
887 notify_method_exit(false, state, NotifyJVMTI, true);
888
889 BLOCK_COMMENT("reserved_stack_check:");
890 if (StackReservedPages > 0) {
891 // Test if reserved zone needs to be enabled.
892 Label no_reserved_zone_enabling;
893
894 // check if already enabled - if so no re-enabling needed
895 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
896 lwz(R0, in_bytes(JavaThread::stack_guard_state_offset()), R16_thread);
897 cmpwi(CR0, R0, StackOverflow::stack_guard_enabled);
898 beq_predict_taken(CR0, no_reserved_zone_enabling);
899
900 // Compare frame pointers. There is no good stack pointer, as with stack
901 // frame compression we can get different SPs when we do calls. A subsequent
902 // call could have a smaller SP, so that this compare succeeds for an
903 // inner call of the method annotated with ReservedStack.
904 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread);
905 cmpld(CR0, fp, R0);
906 blt_predict_taken(CR0, no_reserved_zone_enabling);
907
908 JFR_ONLY(leave_jfr_critical_section();)
909
910 // Enable reserved zone again, throw stack overflow exception.
911 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread);
912 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
913
914 should_not_reach_here();
915
916 bind(no_reserved_zone_enabling);
917 }
918
919 verify_oop(R17_tos, state);
920
921 remove_top_frame_given_fp(fp, R21_sender_SP, R23_tmp3, /*return_pc*/ R0, R11_scratch1);
922 mtlr(R0);
923 pop_cont_fastpath();
924 JFR_ONLY(leave_jfr_critical_section();)
925
926 BLOCK_COMMENT("} remove_activation");
927 }
928
929 #if INCLUDE_JFR
930 void InterpreterMacroAssembler::enter_jfr_critical_section() {
931 li(R0, 1);
932 stb(R0, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR), R16_thread);
933 }
934
935 void InterpreterMacroAssembler::leave_jfr_critical_section() {
936 li(R0, 0);
937 stb(R0, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR), R16_thread);
938 }
939 #endif // INCLUDE_JFR
940
941 // Lock object
942 //
943 // Registers alive
944 // monitor - Address of the BasicObjectLock to be used for locking,
945 // which must be initialized with the object to lock.
946 // object - Address of the object to be locked.
947 //
948 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
949 if (LockingMode == LM_MONITOR) {
950 call_VM_preemptable(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
951 } else {
952 // template code (for LM_LEGACY):
953 //
954 // markWord displaced_header = obj->mark().set_unlocked();
955 // monitor->lock()->set_displaced_header(displaced_header);
956 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
957 // // We stored the monitor address into the object's mark word.
958 // } else if (THREAD->is_lock_owned((address)displaced_header))
959 // // Simple recursive case.
960 // monitor->lock()->set_displaced_header(nullptr);
961 // } else {
962 // // Slow path.
963 // InterpreterRuntime::monitorenter(THREAD, monitor);
964 // }
965
966 const Register header = R7_ARG5;
967 const Register object_mark_addr = R8_ARG6;
968 const Register current_header = R9_ARG7;
969 const Register tmp = R10_ARG8;
970
971 Label count_locking, done, slow_case, cas_failed;
972
973 assert_different_registers(header, object_mark_addr, current_header, tmp);
974
975 // markWord displaced_header = obj->mark().set_unlocked();
976
977 if (LockingMode == LM_LIGHTWEIGHT) {
978 lightweight_lock(monitor, object, header, tmp, slow_case);
979 b(done);
980 } else if (LockingMode == LM_LEGACY) {
981
982 if (DiagnoseSyncOnValueBasedClasses != 0) {
983 load_klass(tmp, object);
984 lbz(tmp, in_bytes(Klass::misc_flags_offset()), tmp);
985 testbitdi(CR0, R0, tmp, exact_log2(KlassFlags::_misc_is_value_based_class));
986 bne(CR0, slow_case);
987 }
988
989 // Load markWord from object into header.
990 ld(header, oopDesc::mark_offset_in_bytes(), object);
991
992 // Set displaced_header to be (markWord of object | UNLOCK_VALUE).
993 ori(header, header, markWord::unlocked_value);
994
995 // monitor->lock()->set_displaced_header(displaced_header);
996 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
997 const int mark_offset = lock_offset +
998 BasicLock::displaced_header_offset_in_bytes();
999
1000 // Initialize the box (Must happen before we update the object mark!).
1001 std(header, mark_offset, monitor);
1002
1003 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
1004
1005 // Store stack address of the BasicObjectLock (this is monitor) into object.
1006 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
1007
1008 // Must fence, otherwise, preceding store(s) may float below cmpxchg.
1009 // CmpxchgX sets CR0 to cmpX(current, displaced).
1010 cmpxchgd(/*flag=*/CR0,
1011 /*current_value=*/current_header,
1012 /*compare_value=*/header, /*exchange_value=*/monitor,
1013 /*where=*/object_mark_addr,
1014 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq,
1015 MacroAssembler::cmpxchgx_hint_acquire_lock(),
1016 noreg,
1017 &cas_failed,
1018 /*check without membar and ldarx first*/true);
1019
1020 // If the compare-and-exchange succeeded, then we found an unlocked
1021 // object and we have now locked it.
1022 b(count_locking);
1023 bind(cas_failed);
1024
1025 // } else if (THREAD->is_lock_owned((address)displaced_header))
1026 // // Simple recursive case.
1027 // monitor->lock()->set_displaced_header(nullptr);
1028
1029 // We did not see an unlocked object so try the fast recursive case.
1030
1031 // Check if owner is self by comparing the value in the markWord of object
1032 // (current_header) with the stack pointer.
1033 sub(current_header, current_header, R1_SP);
1034
1035 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1036 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markWord::lock_mask_in_place);
1037
1038 and_(R0/*==0?*/, current_header, tmp);
1039 // If condition is true we are done and hence we can store 0 in the displaced
1040 // header indicating it is a recursive lock.
1041 bne(CR0, slow_case);
1042 std(R0/*==0!*/, mark_offset, monitor);
1043 b(count_locking);
1044 }
1045
1046 // } else {
1047 // // Slow path.
1048 // InterpreterRuntime::monitorenter(THREAD, monitor);
1049
1050 // None of the above fast optimizations worked so we have to get into the
1051 // slow case of monitor enter.
1052 bind(slow_case);
1053 call_VM_preemptable(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
1054 // }
1055
1056 if (LockingMode == LM_LEGACY) {
1057 b(done);
1058 align(32, 12);
1059 bind(count_locking);
1060 inc_held_monitor_count(current_header /*tmp*/);
1061 }
1062 bind(done);
1063 }
1064 }
1065
1066 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1067 //
1068 // Registers alive
1069 // monitor - Address of the BasicObjectLock to be used for locking,
1070 // which must be initialized with the object to lock.
1071 //
1072 // Throw IllegalMonitorException if object is not locked by current thread.
1073 void InterpreterMacroAssembler::unlock_object(Register monitor) {
1074 if (LockingMode == LM_MONITOR) {
1075 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1076 } else {
1077
1078 // template code (for LM_LEGACY):
1079 //
1080 // if ((displaced_header = monitor->displaced_header()) == nullptr) {
1081 // // Recursive unlock. Mark the monitor unlocked by setting the object field to null.
1082 // monitor->set_obj(nullptr);
1083 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1084 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1085 // monitor->set_obj(nullptr);
1086 // } else {
1087 // // Slow path.
1088 // InterpreterRuntime::monitorexit(monitor);
1089 // }
1090
1091 const Register object = R7_ARG5;
1092 const Register header = R8_ARG6;
1093 const Register object_mark_addr = R9_ARG7;
1094 const Register current_header = R10_ARG8;
1095
1096 Label free_slot;
1097 Label slow_case;
1098
1099 assert_different_registers(object, header, object_mark_addr, current_header);
1100
1101 if (LockingMode != LM_LIGHTWEIGHT) {
1102 // Test first if we are in the fast recursive case.
1103 ld(header, in_bytes(BasicObjectLock::lock_offset()) +
1104 BasicLock::displaced_header_offset_in_bytes(), monitor);
1105
1106 // If the displaced header is zero, we have a recursive unlock.
1107 cmpdi(CR0, header, 0);
1108 beq(CR0, free_slot); // recursive unlock
1109 }
1110
1111 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1112 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1113 // monitor->set_obj(nullptr);
1114
1115 // If we still have a lightweight lock, unlock the object and be done.
1116
1117 // The object address from the monitor is in object.
1118 ld(object, in_bytes(BasicObjectLock::obj_offset()), monitor);
1119
1120 if (LockingMode == LM_LIGHTWEIGHT) {
1121 lightweight_unlock(object, header, slow_case);
1122 } else {
1123 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
1124
1125 // We have the displaced header in displaced_header. If the lock is still
1126 // lightweight, it will contain the monitor address and we'll store the
1127 // displaced header back into the object's mark word.
1128 // CmpxchgX sets CR0 to cmpX(current, monitor).
1129 cmpxchgd(/*flag=*/CR0,
1130 /*current_value=*/current_header,
1131 /*compare_value=*/monitor, /*exchange_value=*/header,
1132 /*where=*/object_mark_addr,
1133 MacroAssembler::MemBarRel,
1134 MacroAssembler::cmpxchgx_hint_release_lock(),
1135 noreg,
1136 &slow_case);
1137 }
1138 b(free_slot);
1139
1140 // } else {
1141 // // Slow path.
1142 // InterpreterRuntime::monitorexit(monitor);
1143
1144 // The lock has been converted into a heavy lock and hence
1145 // we need to get into the slow case.
1146 bind(slow_case);
1147 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1148 // }
1149
1150 Label done;
1151 b(done); // Monitor register may be overwritten! Runtime has already freed the slot.
1152
1153 // Exchange worked, do monitor->set_obj(nullptr);
1154 align(32, 12);
1155 bind(free_slot);
1156 li(R0, 0);
1157 std(R0, in_bytes(BasicObjectLock::obj_offset()), monitor);
1158 if (LockingMode == LM_LEGACY) {
1159 dec_held_monitor_count(current_header /*tmp*/);
1160 }
1161 bind(done);
1162 }
1163 }
1164
1165 // Load compiled (i2c) or interpreter entry when calling from interpreted and
1166 // do the call. Centralized so that all interpreter calls will do the same actions.
1167 // If jvmti single stepping is on for a thread we must not call compiled code.
1168 //
1169 // Input:
1170 // - Rtarget_method: method to call
1171 // - Rret_addr: return address
1172 // - 2 scratch regs
1173 //
1174 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr,
1175 Register Rscratch1, Register Rscratch2) {
1176 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr);
1177 // Assume we want to go compiled if available.
1178 const Register Rtarget_addr = Rscratch1;
1179 const Register Rinterp_only = Rscratch2;
1180
1181 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method);
1182
1183 if (JvmtiExport::can_post_interpreter_events()) {
1184 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
1185
1186 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
1187 // compiled code in threads for which the event is enabled. Check here for
1188 // interp_only_mode if these events CAN be enabled.
1189 Label done;
1190 cmpwi(CR0, Rinterp_only, 0);
1191 beq(CR0, done);
1192 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method);
1193 align(32, 12);
1194 bind(done);
1195 }
1196
1197 #ifdef ASSERT
1198 {
1199 Label Lok;
1200 cmpdi(CR0, Rtarget_addr, 0);
1201 bne(CR0, Lok);
1202 stop("null entry point");
1203 bind(Lok);
1204 }
1205 #endif // ASSERT
1206
1207 mr(R21_sender_SP, R1_SP);
1208
1209 // Calc a precise SP for the call. The SP value we calculated in
1210 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space
1211 // if esp is not max. Also, the i2c adapter extends the stack space without restoring
1212 // our pre-calced value, so repeating calls via i2c would result in stack overflow.
1213 // Since esp already points to an empty slot, we just have to sub 1 additional slot
1214 // to meet the abi scratch requirements.
1215 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in
1216 // the return entry of the interpreter.
1217 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::top_ijava_frame_abi_size);
1218 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address
1219 resize_frame_absolute(Rscratch2, Rscratch2, R0);
1220
1221 mr_if_needed(R19_method, Rtarget_method);
1222 mtctr(Rtarget_addr);
1223 mtlr(Rret_addr);
1224
1225 save_interpreter_state(Rscratch2);
1226 #ifdef ASSERT
1227 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp
1228 sldi(Rscratch1, Rscratch1, Interpreter::logStackElementSize);
1229 add(Rscratch1, Rscratch1, Rscratch2); // Rscratch2 contains fp
1230 // Compare sender_sp with the derelativized top_frame_sp
1231 cmpd(CR0, R21_sender_SP, Rscratch1);
1232 asm_assert_eq("top_frame_sp incorrect");
1233 #endif
1234
1235 bctr();
1236 }
1237
1238 // Set the method data pointer for the current bcp.
1239 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1240 assert(ProfileInterpreter, "must be profiling interpreter");
1241 Label get_continue;
1242 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method);
1243 test_method_data_pointer(get_continue);
1244 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp);
1245
1246 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1247 add(R28_mdx, R28_mdx, R3_RET);
1248 bind(get_continue);
1249 }
1250
1251 // Test ImethodDataPtr. If it is null, continue at the specified label.
1252 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1253 assert(ProfileInterpreter, "must be profiling interpreter");
1254 cmpdi(CR0, R28_mdx, 0);
1255 beq(CR0, zero_continue);
1256 }
1257
1258 void InterpreterMacroAssembler::verify_method_data_pointer() {
1259 assert(ProfileInterpreter, "must be profiling interpreter");
1260 #ifdef ASSERT
1261 Label verify_continue;
1262 test_method_data_pointer(verify_continue);
1263
1264 // If the mdp is valid, it will point to a DataLayout header which is
1265 // consistent with the bcp. The converse is highly probable also.
1266 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx);
1267 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method);
1268 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1269 add(R11_scratch1, R12_scratch2, R12_scratch2);
1270 cmpd(CR0, R11_scratch1, R14_bcp);
1271 beq(CR0, verify_continue);
1272
1273 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx);
1274
1275 bind(verify_continue);
1276 #endif
1277 }
1278
1279 // Store a value at some constant offset from the method data pointer.
1280 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1281 assert(ProfileInterpreter, "must be profiling interpreter");
1282
1283 std(value, constant, R28_mdx);
1284 }
1285
1286 // Increment the value at some constant offset from the method data pointer.
1287 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1288 Register counter_addr,
1289 Register Rbumped_count,
1290 bool decrement) {
1291 // Locate the counter at a fixed offset from the mdp:
1292 addi(counter_addr, R28_mdx, constant);
1293 increment_mdp_data_at(counter_addr, Rbumped_count, decrement);
1294 }
1295
1296 // Increment the value at some non-fixed (reg + constant) offset from
1297 // the method data pointer.
1298 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1299 int constant,
1300 Register scratch,
1301 Register Rbumped_count,
1302 bool decrement) {
1303 // Add the constant to reg to get the offset.
1304 add(scratch, R28_mdx, reg);
1305 // Then calculate the counter address.
1306 addi(scratch, scratch, constant);
1307 increment_mdp_data_at(scratch, Rbumped_count, decrement);
1308 }
1309
1310 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr,
1311 Register Rbumped_count,
1312 bool decrement) {
1313 assert(ProfileInterpreter, "must be profiling interpreter");
1314
1315 // Load the counter.
1316 ld(Rbumped_count, 0, counter_addr);
1317
1318 if (decrement) {
1319 // Decrement the register. Set condition codes.
1320 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment);
1321 // Store the decremented counter, if it is still negative.
1322 std(Rbumped_count, 0, counter_addr);
1323 // Note: add/sub overflow check are not ported, since 64 bit
1324 // calculation should never overflow.
1325 } else {
1326 // Increment the register. Set carry flag.
1327 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment);
1328 // Store the incremented counter.
1329 std(Rbumped_count, 0, counter_addr);
1330 }
1331 }
1332
1333 // Set a flag value at the current method data pointer position.
1334 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1335 Register scratch) {
1336 assert(ProfileInterpreter, "must be profiling interpreter");
1337 // Load the data header.
1338 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1339 // Set the flag.
1340 ori(scratch, scratch, flag_constant);
1341 // Store the modified header.
1342 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1343 }
1344
1345 // Test the location at some offset from the method data pointer.
1346 // If it is not equal to value, branch to the not_equal_continue Label.
1347 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1348 Register value,
1349 Label& not_equal_continue,
1350 Register test_out) {
1351 assert(ProfileInterpreter, "must be profiling interpreter");
1352
1353 ld(test_out, offset, R28_mdx);
1354 cmpd(CR0, value, test_out);
1355 bne(CR0, not_equal_continue);
1356 }
1357
1358 // Update the method data pointer by the displacement located at some fixed
1359 // offset from the method data pointer.
1360 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1361 Register scratch) {
1362 assert(ProfileInterpreter, "must be profiling interpreter");
1363
1364 ld(scratch, offset_of_disp, R28_mdx);
1365 add(R28_mdx, scratch, R28_mdx);
1366 }
1367
1368 // Update the method data pointer by the displacement located at the
1369 // offset (reg + offset_of_disp).
1370 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1371 int offset_of_disp,
1372 Register scratch) {
1373 assert(ProfileInterpreter, "must be profiling interpreter");
1374
1375 add(scratch, reg, R28_mdx);
1376 ld(scratch, offset_of_disp, scratch);
1377 add(R28_mdx, scratch, R28_mdx);
1378 }
1379
1380 // Update the method data pointer by a simple constant displacement.
1381 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1382 assert(ProfileInterpreter, "must be profiling interpreter");
1383 addi(R28_mdx, R28_mdx, constant);
1384 }
1385
1386 // Update the method data pointer for a _ret bytecode whose target
1387 // was not among our cached targets.
1388 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1389 Register return_bci) {
1390 assert(ProfileInterpreter, "must be profiling interpreter");
1391
1392 push(state);
1393 assert(return_bci->is_nonvolatile(), "need to protect return_bci");
1394 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1395 pop(state);
1396 }
1397
1398 // Increments the backedge counter.
1399 // Returns backedge counter + invocation counter in Rdst.
1400 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst,
1401 const Register Rtmp1, Register Rscratch) {
1402 assert(UseCompiler, "incrementing must be useful");
1403 assert_different_registers(Rdst, Rtmp1);
1404 const Register invocation_counter = Rtmp1;
1405 const Register counter = Rdst;
1406 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size.");
1407
1408 // Load backedge counter.
1409 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1410 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1411 // Load invocation counter.
1412 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) +
1413 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1414
1415 // Add the delta to the backedge counter.
1416 addi(counter, counter, InvocationCounter::count_increment);
1417
1418 // Mask the invocation counter.
1419 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value);
1420
1421 // Store new counter value.
1422 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1423 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1424 // Return invocation counter + backedge counter.
1425 add(counter, counter, invocation_counter);
1426 }
1427
1428 // Count a taken branch in the bytecodes.
1429 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
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 taken count.
1437 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count);
1438
1439 // The method data pointer needs to be updated to reflect the new target.
1440 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1441 bind (profile_continue);
1442 }
1443 }
1444
1445 // Count a not-taken branch in the bytecodes.
1446 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) {
1447 if (ProfileInterpreter) {
1448 Label profile_continue;
1449
1450 // If no method data exists, go to profile_continue.
1451 test_method_data_pointer(profile_continue);
1452
1453 // We are taking a branch. Increment the not taken count.
1454 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2);
1455
1456 // The method data pointer needs to be updated to correspond to the
1457 // next bytecode.
1458 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1459 bind (profile_continue);
1460 }
1461 }
1462
1463 // Count a non-virtual call in the bytecodes.
1464 void InterpreterMacroAssembler::profile_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(CounterData::counter_data_size()));
1476 bind (profile_continue);
1477 }
1478 }
1479
1480 // Count a final call in the bytecodes.
1481 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) {
1482 if (ProfileInterpreter) {
1483 Label profile_continue;
1484
1485 // If no method data exists, go to profile_continue.
1486 test_method_data_pointer(profile_continue);
1487
1488 // We are making a call. Increment the count.
1489 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1490
1491 // The method data pointer needs to be updated to reflect the new target.
1492 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1493 bind (profile_continue);
1494 }
1495 }
1496
1497 // Count a virtual call in the bytecodes.
1498 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver,
1499 Register Rscratch1,
1500 Register Rscratch2,
1501 bool receiver_can_be_null) {
1502 if (!ProfileInterpreter) { return; }
1503 Label profile_continue;
1504
1505 // If no method data exists, go to profile_continue.
1506 test_method_data_pointer(profile_continue);
1507
1508 Label skip_receiver_profile;
1509 if (receiver_can_be_null) {
1510 Label not_null;
1511 cmpdi(CR0, Rreceiver, 0);
1512 bne(CR0, not_null);
1513 // We are making a call. Increment the count for null receiver.
1514 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2);
1515 b(skip_receiver_profile);
1516 bind(not_null);
1517 }
1518
1519 // Record the receiver type.
1520 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2);
1521 bind(skip_receiver_profile);
1522
1523 // The method data pointer needs to be updated to reflect the new target.
1524 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1525 bind (profile_continue);
1526 }
1527
1528 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) {
1529 if (ProfileInterpreter) {
1530 Label profile_continue;
1531
1532 // If no method data exists, go to profile_continue.
1533 test_method_data_pointer(profile_continue);
1534
1535 int mdp_delta = in_bytes(BitData::bit_data_size());
1536 if (TypeProfileCasts) {
1537 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1538
1539 // Record the object type.
1540 record_klass_in_profile(Rklass, Rscratch1, Rscratch2);
1541 }
1542
1543 // The method data pointer needs to be updated.
1544 update_mdp_by_constant(mdp_delta);
1545
1546 bind (profile_continue);
1547 }
1548 }
1549
1550 // Count a ret in the bytecodes.
1551 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci,
1552 Register scratch1, Register scratch2) {
1553 if (ProfileInterpreter) {
1554 Label profile_continue;
1555 uint row;
1556
1557 // If no method data exists, go to profile_continue.
1558 test_method_data_pointer(profile_continue);
1559
1560 // Update the total ret count.
1561 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 );
1562
1563 for (row = 0; row < RetData::row_limit(); row++) {
1564 Label next_test;
1565
1566 // See if return_bci is equal to bci[n]:
1567 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1);
1568
1569 // return_bci is equal to bci[n]. Increment the count.
1570 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2);
1571
1572 // The method data pointer needs to be updated to reflect the new target.
1573 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1);
1574 b(profile_continue);
1575 bind(next_test);
1576 }
1577
1578 update_mdp_for_ret(state, return_bci);
1579
1580 bind (profile_continue);
1581 }
1582 }
1583
1584 // Count the default case of a switch construct.
1585 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) {
1586 if (ProfileInterpreter) {
1587 Label profile_continue;
1588
1589 // If no method data exists, go to profile_continue.
1590 test_method_data_pointer(profile_continue);
1591
1592 // Update the default case count
1593 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1594 scratch1, scratch2);
1595
1596 // The method data pointer needs to be updated.
1597 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()),
1598 scratch1);
1599
1600 bind (profile_continue);
1601 }
1602 }
1603
1604 // Count the index'th case of a switch construct.
1605 void InterpreterMacroAssembler::profile_switch_case(Register index,
1606 Register scratch1,
1607 Register scratch2,
1608 Register scratch3) {
1609 if (ProfileInterpreter) {
1610 assert_different_registers(index, scratch1, scratch2, scratch3);
1611 Label profile_continue;
1612
1613 // If no method data exists, go to profile_continue.
1614 test_method_data_pointer(profile_continue);
1615
1616 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes().
1617 li(scratch3, in_bytes(MultiBranchData::case_array_offset()));
1618
1619 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works");
1620 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1621 add(scratch1, scratch1, scratch3);
1622
1623 // Update the case count.
1624 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3);
1625
1626 // The method data pointer needs to be updated.
1627 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2);
1628
1629 bind (profile_continue);
1630 }
1631 }
1632
1633 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) {
1634 if (ProfileInterpreter) {
1635 assert_different_registers(Rscratch1, Rscratch2);
1636 Label profile_continue;
1637
1638 // If no method data exists, go to profile_continue.
1639 test_method_data_pointer(profile_continue);
1640
1641 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1);
1642
1643 // The method data pointer needs to be updated.
1644 int mdp_delta = in_bytes(BitData::bit_data_size());
1645 if (TypeProfileCasts) {
1646 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1647 }
1648 update_mdp_by_constant(mdp_delta);
1649
1650 bind (profile_continue);
1651 }
1652 }
1653
1654 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver,
1655 Register Rscratch1, Register Rscratch2) {
1656 assert(ProfileInterpreter, "must be profiling");
1657 assert_different_registers(Rreceiver, Rscratch1, Rscratch2);
1658
1659 Label done;
1660 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done);
1661 bind (done);
1662 }
1663
1664 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1665 Register receiver, Register scratch1, Register scratch2,
1666 int start_row, Label& done) {
1667 if (TypeProfileWidth == 0) {
1668 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1669 return;
1670 }
1671
1672 int last_row = VirtualCallData::row_limit() - 1;
1673 assert(start_row <= last_row, "must be work left to do");
1674 // Test this row for both the receiver and for null.
1675 // Take any of three different outcomes:
1676 // 1. found receiver => increment count and goto done
1677 // 2. found null => keep looking for case 1, maybe allocate this cell
1678 // 3. found something else => keep looking for cases 1 and 2
1679 // Case 3 is handled by a recursive call.
1680 for (int row = start_row; row <= last_row; row++) {
1681 Label next_test;
1682 bool test_for_null_also = (row == start_row);
1683
1684 // See if the receiver is receiver[n].
1685 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1686 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1);
1687 // delayed()->tst(scratch);
1688
1689 // The receiver is receiver[n]. Increment count[n].
1690 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1691 increment_mdp_data_at(count_offset, scratch1, scratch2);
1692 b(done);
1693 bind(next_test);
1694
1695 if (test_for_null_also) {
1696 Label found_null;
1697 // Failed the equality check on receiver[n]... Test for null.
1698 if (start_row == last_row) {
1699 // The only thing left to do is handle the null case.
1700 // Scratch1 contains test_out from test_mdp_data_at.
1701 cmpdi(CR0, scratch1, 0);
1702 beq(CR0, found_null);
1703 // Receiver did not match any saved receiver and there is no empty row for it.
1704 // Increment total counter to indicate polymorphic case.
1705 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1706 b(done);
1707 bind(found_null);
1708 break;
1709 }
1710 // Since null is rare, make it be the branch-taken case.
1711 cmpdi(CR0, scratch1, 0);
1712 beq(CR0, found_null);
1713
1714 // Put all the "Case 3" tests here.
1715 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done);
1716
1717 // Found a null. Keep searching for a matching receiver,
1718 // but remember that this is an empty (unused) slot.
1719 bind(found_null);
1720 }
1721 }
1722
1723 // In the fall-through case, we found no matching receiver, but we
1724 // observed the receiver[start_row] is null.
1725
1726 // Fill in the receiver field and increment the count.
1727 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1728 set_mdp_data_at(recvr_offset, receiver);
1729 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1730 li(scratch1, DataLayout::counter_increment);
1731 set_mdp_data_at(count_offset, scratch1);
1732 if (start_row > 0) {
1733 b(done);
1734 }
1735 }
1736
1737 // Argument and return type profilig.
1738 // kills: tmp, tmp2, R0, CR0, CR1
1739 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base,
1740 RegisterOrConstant mdo_addr_offs,
1741 Register tmp, Register tmp2) {
1742 Label do_nothing, do_update;
1743
1744 // tmp2 = obj is allowed
1745 assert_different_registers(obj, mdo_addr_base, tmp, R0);
1746 assert_different_registers(tmp2, mdo_addr_base, tmp, R0);
1747 const Register klass = tmp2;
1748
1749 verify_oop(obj);
1750
1751 ld(tmp, mdo_addr_offs, mdo_addr_base);
1752
1753 // Set null_seen if obj is 0.
1754 cmpdi(CR0, obj, 0);
1755 ori(R0, tmp, TypeEntries::null_seen);
1756 beq(CR0, do_update);
1757
1758 load_klass(klass, obj);
1759
1760 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
1761 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
1762 cmpd(CR1, R0, klass);
1763 // Klass seen before, nothing to do (regardless of unknown bit).
1764 //beq(CR1, do_nothing);
1765
1766 andi_(R0, tmp, TypeEntries::type_unknown);
1767 // Already unknown. Nothing to do anymore.
1768 //bne(CR0, do_nothing);
1769 crorc(CR0, Assembler::equal, CR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne
1770 beq(CR0, do_nothing);
1771
1772 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
1773 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1774 beq(CR0, do_update); // First time here. Set profile type.
1775
1776 // Different than before. Cannot keep accurate profile.
1777 ori(R0, tmp, TypeEntries::type_unknown);
1778
1779 bind(do_update);
1780 // update profile
1781 std(R0, mdo_addr_offs, mdo_addr_base);
1782
1783 align(32, 12);
1784 bind(do_nothing);
1785 }
1786
1787 void InterpreterMacroAssembler::profile_arguments_type(Register callee,
1788 Register tmp1, Register tmp2,
1789 bool is_virtual) {
1790 if (!ProfileInterpreter) {
1791 return;
1792 }
1793
1794 assert_different_registers(callee, tmp1, tmp2, R28_mdx);
1795
1796 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1797 Label profile_continue;
1798
1799 test_method_data_pointer(profile_continue);
1800
1801 int off_to_start = is_virtual ?
1802 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1803
1804 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx);
1805 cmpwi(CR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1806 bne(CR0, profile_continue);
1807
1808 if (MethodData::profile_arguments()) {
1809 Label done;
1810 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1811 addi(R28_mdx, R28_mdx, off_to_args);
1812
1813 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1814 if (i > 0 || MethodData::profile_return()) {
1815 // If return value type is profiled we may have no argument to profile.
1816 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1817 cmpdi(CR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count());
1818 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count());
1819 blt(CR0, done);
1820 }
1821 ld(tmp1, in_bytes(Method::const_offset()), callee);
1822 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1);
1823 // Stack offset o (zero based) from the start of the argument
1824 // list, for n arguments translates into offset n - o - 1 from
1825 // the end of the argument list. But there's an extra slot at
1826 // the top of the stack. So the offset is n - o from Lesp.
1827 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx);
1828 subf(tmp1, tmp2, tmp1);
1829
1830 sldi(tmp1, tmp1, Interpreter::logStackElementSize);
1831 ldx(tmp1, tmp1, R15_esp);
1832
1833 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1);
1834
1835 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1836 addi(R28_mdx, R28_mdx, to_add);
1837 off_to_args += to_add;
1838 }
1839
1840 if (MethodData::profile_return()) {
1841 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1842 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1843 }
1844
1845 bind(done);
1846
1847 if (MethodData::profile_return()) {
1848 // We're right after the type profile for the last
1849 // argument. tmp1 is the number of cells left in the
1850 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1851 // if there's a return to profile.
1852 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(),
1853 "can't move past ret type");
1854 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size));
1855 add(R28_mdx, tmp1, R28_mdx);
1856 }
1857 } else {
1858 assert(MethodData::profile_return(), "either profile call args or call ret");
1859 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
1860 }
1861
1862 // Mdp points right after the end of the
1863 // CallTypeData/VirtualCallTypeData, right after the cells for the
1864 // return value type if there's one.
1865 align(32, 12);
1866 bind(profile_continue);
1867 }
1868 }
1869
1870 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
1871 assert_different_registers(ret, tmp1, tmp2);
1872 if (ProfileInterpreter && MethodData::profile_return()) {
1873 Label profile_continue;
1874
1875 test_method_data_pointer(profile_continue);
1876
1877 if (MethodData::profile_return_jsr292_only()) {
1878 // If we don't profile all invoke bytecodes we must make sure
1879 // it's a bytecode we indeed profile. We can't go back to the
1880 // beginning of the ProfileData we intend to update to check its
1881 // type because we're right after it and we don't known its
1882 // length.
1883 lbz(tmp1, 0, R14_bcp);
1884 lbz(tmp2, in_bytes(Method::intrinsic_id_offset()), R19_method);
1885 cmpwi(CR0, tmp1, Bytecodes::_invokedynamic);
1886 cmpwi(CR1, tmp1, Bytecodes::_invokehandle);
1887 cror(CR0, Assembler::equal, CR1, Assembler::equal);
1888 cmpwi(CR1, tmp2, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1889 cror(CR0, Assembler::equal, CR1, Assembler::equal);
1890 bne(CR0, profile_continue);
1891 }
1892
1893 profile_obj_type(ret, R28_mdx, -in_bytes(SingleTypeEntry::size()), tmp1, tmp2);
1894
1895 align(32, 12);
1896 bind(profile_continue);
1897 }
1898 }
1899
1900 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2,
1901 Register tmp3, Register tmp4) {
1902 if (ProfileInterpreter && MethodData::profile_parameters()) {
1903 Label profile_continue, done;
1904
1905 test_method_data_pointer(profile_continue);
1906
1907 // Load the offset of the area within the MDO used for
1908 // parameters. If it's negative we're not profiling any parameters.
1909 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx);
1910 cmpwi(CR0, tmp1, 0);
1911 blt(CR0, profile_continue);
1912
1913 // Compute a pointer to the area for parameters from the offset
1914 // and move the pointer to the slot for the last
1915 // parameters. Collect profiling from last parameter down.
1916 // mdo start + parameters offset + array length - 1
1917
1918 // Pointer to the parameter area in the MDO.
1919 const Register mdp = tmp1;
1920 add(mdp, tmp1, R28_mdx);
1921
1922 // Offset of the current profile entry to update.
1923 const Register entry_offset = tmp2;
1924 // entry_offset = array len in number of cells
1925 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp);
1926
1927 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1928 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
1929
1930 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
1931 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size));
1932 // entry_offset in bytes
1933 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1934
1935 Label loop;
1936 align(32, 12);
1937 bind(loop);
1938
1939 // Load offset on the stack from the slot for this parameter.
1940 ld(tmp3, entry_offset, mdp);
1941 sldi(tmp3, tmp3, Interpreter::logStackElementSize);
1942 neg(tmp3, tmp3);
1943 // Read the parameter from the local area.
1944 ldx(tmp3, tmp3, R18_locals);
1945
1946 // Make entry_offset now point to the type field for this parameter.
1947 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1948 assert(type_base > off_base, "unexpected");
1949 addi(entry_offset, entry_offset, type_base - off_base);
1950
1951 // Profile the parameter.
1952 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3);
1953
1954 // Go to next parameter.
1955 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base);
1956 cmpdi(CR0, entry_offset, off_base + delta);
1957 addi(entry_offset, entry_offset, -delta);
1958 bge(CR0, loop);
1959
1960 align(32, 12);
1961 bind(profile_continue);
1962 }
1963 }
1964
1965 // Add a monitor (see frame_ppc.hpp).
1966 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) {
1967
1968 // Very-local scratch registers.
1969 const Register esp = Rtemp1;
1970 const Register slot = Rtemp2;
1971
1972 // Extracted monitor_size.
1973 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes();
1974 assert(Assembler::is_aligned((unsigned int)monitor_size,
1975 (unsigned int)frame::alignment_in_bytes),
1976 "size of a monitor must respect alignment of SP");
1977
1978 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor
1979 subf(Rtemp2, esp, R1_SP); // esp contains fp
1980 sradi(Rtemp2, Rtemp2, Interpreter::logStackElementSize);
1981 // Store relativized top_frame_sp
1982 std(Rtemp2, _ijava_state_neg(top_frame_sp), esp); // esp contains fp
1983
1984 // Shuffle expression stack down. Recall that stack_base points
1985 // just above the new expression stack bottom. Old_tos and new_tos
1986 // are used to scan thru the old and new expression stacks.
1987 if (!stack_is_empty) {
1988 Label copy_slot, copy_slot_finished;
1989 const Register n_slots = slot;
1990
1991 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack).
1992 subf(n_slots, esp, R26_monitor);
1993 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy.
1994 assert(LogBytesPerWord == 3, "conflicts assembler instructions");
1995 beq(CR0, copy_slot_finished); // Nothing to copy.
1996
1997 mtctr(n_slots);
1998
1999 // loop
2000 bind(copy_slot);
2001 ld(slot, 0, esp); // Move expression stack down.
2002 std(slot, -monitor_size, esp); // distance = monitor_size
2003 addi(esp, esp, BytesPerWord);
2004 bdnz(copy_slot);
2005
2006 bind(copy_slot_finished);
2007 }
2008
2009 addi(R15_esp, R15_esp, -monitor_size);
2010 addi(R26_monitor, R26_monitor, -monitor_size);
2011
2012 // Restart interpreter
2013 }
2014
2015 // ============================================================================
2016 // Java locals access
2017
2018 // Load a local variable at index in Rindex into register Rdst_value.
2019 // Also puts address of local into Rdst_address as a service.
2020 // Kills:
2021 // - Rdst_value
2022 // - Rdst_address
2023 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) {
2024 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2025 subf(Rdst_address, Rdst_address, R18_locals);
2026 lwz(Rdst_value, 0, Rdst_address);
2027 }
2028
2029 // Load a local variable at index in Rindex into register Rdst_value.
2030 // Also puts address of local into Rdst_address as a service.
2031 // Kills:
2032 // - Rdst_value
2033 // - Rdst_address
2034 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) {
2035 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2036 subf(Rdst_address, Rdst_address, R18_locals);
2037 ld(Rdst_value, -8, Rdst_address);
2038 }
2039
2040 // Load a local variable at index in Rindex into register Rdst_value.
2041 // Also puts address of local into Rdst_address as a service.
2042 // Input:
2043 // - Rindex: slot nr of local variable
2044 // Kills:
2045 // - Rdst_value
2046 // - Rdst_address
2047 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value,
2048 Register Rdst_address,
2049 Register Rindex) {
2050 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2051 subf(Rdst_address, Rdst_address, R18_locals);
2052 ld(Rdst_value, 0, Rdst_address);
2053 }
2054
2055 // Load a local variable at index in Rindex into register Rdst_value.
2056 // Also puts address of local into Rdst_address as a service.
2057 // Kills:
2058 // - Rdst_value
2059 // - Rdst_address
2060 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value,
2061 Register Rdst_address,
2062 Register Rindex) {
2063 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2064 subf(Rdst_address, Rdst_address, R18_locals);
2065 lfs(Rdst_value, 0, Rdst_address);
2066 }
2067
2068 // Load a local variable at index in Rindex into register Rdst_value.
2069 // Also puts address of local into Rdst_address as a service.
2070 // Kills:
2071 // - Rdst_value
2072 // - Rdst_address
2073 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value,
2074 Register Rdst_address,
2075 Register Rindex) {
2076 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2077 subf(Rdst_address, Rdst_address, R18_locals);
2078 lfd(Rdst_value, -8, Rdst_address);
2079 }
2080
2081 // Store an int value at local variable slot Rindex.
2082 // Kills:
2083 // - Rindex
2084 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) {
2085 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2086 subf(Rindex, Rindex, R18_locals);
2087 stw(Rvalue, 0, Rindex);
2088 }
2089
2090 // Store a long value at local variable slot Rindex.
2091 // Kills:
2092 // - Rindex
2093 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) {
2094 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2095 subf(Rindex, Rindex, R18_locals);
2096 std(Rvalue, -8, Rindex);
2097 }
2098
2099 // Store an oop value at local variable slot Rindex.
2100 // Kills:
2101 // - Rindex
2102 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) {
2103 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2104 subf(Rindex, Rindex, R18_locals);
2105 std(Rvalue, 0, Rindex);
2106 }
2107
2108 // Store an int value at local variable slot Rindex.
2109 // Kills:
2110 // - Rindex
2111 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) {
2112 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2113 subf(Rindex, Rindex, R18_locals);
2114 stfs(Rvalue, 0, Rindex);
2115 }
2116
2117 // Store an int value at local variable slot Rindex.
2118 // Kills:
2119 // - Rindex
2120 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) {
2121 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2122 subf(Rindex, Rindex, R18_locals);
2123 stfd(Rvalue, -8, Rindex);
2124 }
2125
2126 // Read pending exception from thread and jump to interpreter.
2127 // Throw exception entry if one if pending. Fall through otherwise.
2128 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) {
2129 assert_different_registers(Rscratch1, Rscratch2, R3);
2130 Register Rexception = Rscratch1;
2131 Register Rtmp = Rscratch2;
2132 Label Ldone;
2133 // Get pending exception oop.
2134 ld(Rexception, thread_(pending_exception));
2135 cmpdi(CR0, Rexception, 0);
2136 beq(CR0, Ldone);
2137 li(Rtmp, 0);
2138 mr_if_needed(R3, Rexception);
2139 std(Rtmp, thread_(pending_exception)); // Clear exception in thread
2140 if (Interpreter::rethrow_exception_entry() != nullptr) {
2141 // Already got entry address.
2142 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry());
2143 } else {
2144 // Dynamically load entry address.
2145 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true);
2146 ld(Rtmp, simm16_rest, Rtmp);
2147 }
2148 mtctr(Rtmp);
2149 save_interpreter_state(Rtmp);
2150 bctr();
2151
2152 align(32, 12);
2153 bind(Ldone);
2154 }
2155
2156 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions, Label* last_java_pc) {
2157 save_interpreter_state(R11_scratch1);
2158
2159 MacroAssembler::call_VM(oop_result, entry_point, false /*check_exceptions*/, last_java_pc);
2160
2161 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2162
2163 check_and_handle_popframe(R11_scratch1);
2164 check_and_handle_earlyret(R11_scratch1);
2165 // Now check exceptions manually.
2166 if (check_exceptions) {
2167 check_and_forward_exception(R11_scratch1, R12_scratch2);
2168 }
2169 }
2170
2171 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2172 Register arg_1, bool check_exceptions) {
2173 // ARG1 is reserved for the thread.
2174 mr_if_needed(R4_ARG2, arg_1);
2175 call_VM(oop_result, entry_point, check_exceptions);
2176 }
2177
2178 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result, address entry_point,
2179 Register arg_1, bool check_exceptions) {
2180 if (!Continuations::enabled()) {
2181 call_VM(oop_result, entry_point, arg_1, check_exceptions);
2182 return;
2183 }
2184
2185 Label resume_pc, not_preempted;
2186
2187 DEBUG_ONLY(ld(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread));
2188 DEBUG_ONLY(cmpdi(CR0, R0, 0));
2189 asm_assert_eq("Should not have alternate return address set");
2190
2191 // Preserve 2 registers
2192 assert(nonvolatile_accross_vthread_preemtion(R31) && nonvolatile_accross_vthread_preemtion(R22), "");
2193 ld(R3_ARG1, _abi0(callers_sp), R1_SP); // load FP
2194 std(R31, _ijava_state_neg(lresult), R3_ARG1);
2195 std(R22, _ijava_state_neg(fresult), R3_ARG1);
2196
2197 // We set resume_pc as last java pc. It will be saved if the vthread gets preempted.
2198 // Later execution will continue right there.
2199 mr_if_needed(R4_ARG2, arg_1);
2200 push_cont_fastpath();
2201 call_VM(oop_result, entry_point, false /*check_exceptions*/, &resume_pc /* last_java_pc */);
2202 pop_cont_fastpath();
2203
2204 // Jump to handler if the call was preempted
2205 ld(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
2206 cmpdi(CR0, R0, 0);
2207 beq(CR0, not_preempted);
2208 mtlr(R0);
2209 li(R0, 0);
2210 std(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
2211 blr();
2212
2213 bind(resume_pc); // Location to resume execution
2214 restore_after_resume(noreg /* fp */);
2215 bind(not_preempted);
2216 }
2217
2218 void InterpreterMacroAssembler::restore_after_resume(Register fp) {
2219 if (!Continuations::enabled()) return;
2220
2221 const address resume_adapter = TemplateInterpreter::cont_resume_interpreter_adapter();
2222 add_const_optimized(R31, R29_TOC, MacroAssembler::offset_to_global_toc(resume_adapter));
2223 mtctr(R31);
2224 bctrl();
2225 // Restore registers that are preserved across vthread preemption
2226 assert(nonvolatile_accross_vthread_preemtion(R31) && nonvolatile_accross_vthread_preemtion(R22), "");
2227 ld(R3_ARG1, _abi0(callers_sp), R1_SP); // load FP
2228 ld(R31, _ijava_state_neg(lresult), R3_ARG1);
2229 ld(R22, _ijava_state_neg(fresult), R3_ARG1);
2230 #ifdef ASSERT
2231 // Assert FP is in R11_scratch1 (see generate_cont_resume_interpreter_adapter())
2232 {
2233 Label ok;
2234 ld(R12_scratch2, 0, R1_SP); // load fp
2235 cmpd(CR0, R12_scratch2, R11_scratch1);
2236 beq(CR0, ok);
2237 stop(FILE_AND_LINE ": FP is expected in R11_scratch1");
2238 bind(ok);
2239 }
2240 #endif
2241 if (fp != noreg && fp != R11_scratch1) {
2242 mr(fp, R11_scratch1);
2243 }
2244 }
2245
2246 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2247 Register arg_1, Register arg_2,
2248 bool check_exceptions) {
2249 // ARG1 is reserved for the thread.
2250 mr_if_needed(R4_ARG2, arg_1);
2251 assert(arg_2 != R4_ARG2, "smashed argument");
2252 mr_if_needed(R5_ARG3, arg_2);
2253 call_VM(oop_result, entry_point, check_exceptions);
2254 }
2255
2256 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2257 Register arg_1, Register arg_2, Register arg_3,
2258 bool check_exceptions) {
2259 // ARG1 is reserved for the thread.
2260 mr_if_needed(R4_ARG2, arg_1);
2261 assert(arg_2 != R4_ARG2, "smashed argument");
2262 mr_if_needed(R5_ARG3, arg_2);
2263 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument");
2264 mr_if_needed(R6_ARG4, arg_3);
2265 call_VM(oop_result, entry_point, check_exceptions);
2266 }
2267
2268 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) {
2269 ld(scratch, 0, R1_SP);
2270 subf(R0, scratch, R15_esp);
2271 sradi(R0, R0, Interpreter::logStackElementSize);
2272 std(R0, _ijava_state_neg(esp), scratch);
2273 std(R14_bcp, _ijava_state_neg(bcp), scratch);
2274 subf(R0, scratch, R26_monitor);
2275 sradi(R0, R0, Interpreter::logStackElementSize);
2276 std(R0, _ijava_state_neg(monitors), scratch);
2277 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); }
2278 // Other entries should be unchanged.
2279 }
2280
2281 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only, bool restore_top_frame_sp) {
2282 ld_ptr(scratch, _abi0(callers_sp), R1_SP); // Load frame pointer.
2283 if (restore_top_frame_sp) {
2284 // After thawing the top frame of a continuation we reach here with frame::java_abi.
2285 // therefore we have to restore top_frame_sp before the assertion below.
2286 assert(!bcp_and_mdx_only, "chose other registers");
2287 Register tfsp = R18_locals;
2288 Register scratch2 = R26_monitor;
2289 ld(tfsp, _ijava_state_neg(top_frame_sp), scratch);
2290 // Derelativize top_frame_sp
2291 sldi(tfsp, tfsp, Interpreter::logStackElementSize);
2292 add(tfsp, tfsp, scratch);
2293 resize_frame_absolute(tfsp, scratch2, R0);
2294 }
2295 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception).
2296 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code.
2297 if (!bcp_and_mdx_only) {
2298 // Following ones are Metadata.
2299 ld(R19_method, _ijava_state_neg(method), scratch);
2300 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch);
2301 // Following ones are stack addresses and don't require reload.
2302 // Derelativize esp
2303 ld(R15_esp, _ijava_state_neg(esp), scratch);
2304 sldi(R15_esp, R15_esp, Interpreter::logStackElementSize);
2305 add(R15_esp, R15_esp, scratch);
2306 ld(R18_locals, _ijava_state_neg(locals), scratch);
2307 sldi(R18_locals, R18_locals, Interpreter::logStackElementSize);
2308 add(R18_locals, R18_locals, scratch);
2309 ld(R26_monitor, _ijava_state_neg(monitors), scratch);
2310 // Derelativize monitors
2311 sldi(R26_monitor, R26_monitor, Interpreter::logStackElementSize);
2312 add(R26_monitor, R26_monitor, scratch);
2313 }
2314 #ifdef ASSERT
2315 {
2316 Label Lok;
2317 subf(R0, R1_SP, scratch);
2318 cmpdi(CR0, R0, frame::top_ijava_frame_abi_size + frame::ijava_state_size);
2319 bge(CR0, Lok);
2320 stop("frame too small (restore istate)");
2321 bind(Lok);
2322 }
2323 #endif
2324 }
2325
2326 void InterpreterMacroAssembler::get_method_counters(Register method,
2327 Register Rcounters,
2328 Label& skip) {
2329 BLOCK_COMMENT("Load and ev. allocate counter object {");
2330 Label has_counters;
2331 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2332 cmpdi(CR0, Rcounters, 0);
2333 bne(CR0, has_counters);
2334 call_VM(noreg, CAST_FROM_FN_PTR(address,
2335 InterpreterRuntime::build_method_counters), method);
2336 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2337 cmpdi(CR0, Rcounters, 0);
2338 beq(CR0, skip); // No MethodCounters, OutOfMemory.
2339 BLOCK_COMMENT("} Load and ev. allocate counter object");
2340
2341 bind(has_counters);
2342 }
2343
2344 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters,
2345 Register iv_be_count,
2346 Register Rtmp_r0) {
2347 assert(UseCompiler, "incrementing must be useful");
2348 Register invocation_count = iv_be_count;
2349 Register backedge_count = Rtmp_r0;
2350 int delta = InvocationCounter::count_increment;
2351
2352 // Load each counter in a register.
2353 // ld(inv_counter, Rtmp);
2354 // ld(be_counter, Rtmp2);
2355 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() +
2356 InvocationCounter::counter_offset());
2357 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() +
2358 InvocationCounter::counter_offset());
2359
2360 BLOCK_COMMENT("Increment profiling counters {");
2361
2362 // Load the backedge counter.
2363 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int
2364 // Mask the backedge counter.
2365 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value);
2366
2367 // Load the invocation counter.
2368 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int
2369 // Add the delta to the invocation counter and store the result.
2370 addi(invocation_count, invocation_count, delta);
2371 // Store value.
2372 stw(invocation_count, inv_counter_offset, Rcounters);
2373
2374 // Add invocation counter + backedge counter.
2375 add(iv_be_count, backedge_count, invocation_count);
2376
2377 // Note that this macro must leave the backedge_count + invocation_count in
2378 // register iv_be_count!
2379 BLOCK_COMMENT("} Increment profiling counters");
2380 }
2381
2382 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2383 if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); }
2384 }
2385
2386 // Local helper function for the verify_oop_or_return_address macro.
2387 static bool verify_return_address(Method* m, int bci) {
2388 #ifndef PRODUCT
2389 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci;
2390 // Assume it is a valid return address if it is inside m and is preceded by a jsr.
2391 if (!m->contains(pc)) return false;
2392 address jsr_pc;
2393 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2394 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2395 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2396 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2397 #endif // PRODUCT
2398 return false;
2399 }
2400
2401 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2402 if (!VerifyOops) return;
2403
2404 // The VM documentation for the astore[_wide] bytecode allows
2405 // the TOS to be not only an oop but also a return address.
2406 Label test;
2407 Label skip;
2408 // See if it is an address (in the current method):
2409
2410 const int log2_bytecode_size_limit = 16;
2411 srdi_(Rtmp, reg, log2_bytecode_size_limit);
2412 bne(CR0, test);
2413
2414 address fd = CAST_FROM_FN_PTR(address, verify_return_address);
2415 const int nbytes_save = MacroAssembler::num_volatile_regs * 8;
2416 save_volatile_gprs(R1_SP, -nbytes_save); // except R0
2417 save_LR_CR(Rtmp); // Save in old frame.
2418 push_frame_reg_args(nbytes_save, Rtmp);
2419
2420 load_const_optimized(Rtmp, fd, R0);
2421 mr_if_needed(R4_ARG2, reg);
2422 mr(R3_ARG1, R19_method);
2423 call_c(Rtmp); // call C
2424
2425 pop_frame();
2426 restore_LR_CR(Rtmp);
2427 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
2428 b(skip);
2429
2430 // Perform a more elaborate out-of-line call.
2431 // Not an address; verify it:
2432 bind(test);
2433 verify_oop(reg);
2434 bind(skip);
2435 }
2436
2437 // Inline assembly for:
2438 //
2439 // if (thread is in interp_only_mode) {
2440 // InterpreterRuntime::post_method_entry();
2441 // }
2442 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) ||
2443 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) {
2444 // SharedRuntime::jvmpi_method_entry(method, receiver);
2445 // }
2446 void InterpreterMacroAssembler::notify_method_entry() {
2447 // JVMTI
2448 // Whenever JVMTI puts a thread in interp_only_mode, method
2449 // entry/exit events are sent for that thread to track stack
2450 // depth. If it is possible to enter interp_only_mode we add
2451 // the code to check if the event should be sent.
2452 if (JvmtiExport::can_post_interpreter_events()) {
2453 Label jvmti_post_done;
2454
2455 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2456 cmpwi(CR0, R0, 0);
2457 beq(CR0, jvmti_post_done);
2458 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2459
2460 bind(jvmti_post_done);
2461 }
2462 }
2463
2464 // Inline assembly for:
2465 //
2466 // if (thread is in interp_only_mode) {
2467 // // save result
2468 // InterpreterRuntime::post_method_exit();
2469 // // restore result
2470 // }
2471 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) {
2472 // // save result
2473 // SharedRuntime::jvmpi_method_exit();
2474 // // restore result
2475 // }
2476 //
2477 // Native methods have their result stored in d_tmp and l_tmp.
2478 // Java methods have their result stored in the expression stack.
2479 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state,
2480 NotifyMethodExitMode mode, bool check_exceptions) {
2481 // JVMTI
2482 // Whenever JVMTI puts a thread in interp_only_mode, method
2483 // entry/exit events are sent for that thread to track stack
2484 // depth. If it is possible to enter interp_only_mode we add
2485 // the code to check if the event should be sent.
2486 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2487 Label jvmti_post_done;
2488
2489 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2490 cmpwi(CR0, R0, 0);
2491 beq(CR0, jvmti_post_done);
2492 if (!is_native_method) { push(state); } // Expose tos to GC.
2493 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), check_exceptions);
2494 if (!is_native_method) { pop(state); }
2495
2496 align(32, 12);
2497 bind(jvmti_post_done);
2498 }
2499
2500 // Dtrace support not implemented.
2501 }