rev 55858 : 8228649: [PPC64] SA reads wrong slots from interpreter frames
Summary: Make frame layout consistent between dbg and product build and implement offsets accordingly.
Reviewed-by: goetz, gromero
1 /*
2 * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2012, 2018 SAP SE. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26
27 #include "precompiled.hpp"
28 #include "asm/macroAssembler.inline.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interp_masm_ppc.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "prims/jvmtiThreadState.hpp"
34 #include "runtime/frame.inline.hpp"
35 #include "runtime/safepointMechanism.hpp"
36 #include "runtime/sharedRuntime.hpp"
37
38 // Implementation of InterpreterMacroAssembler.
39
40 // This file specializes the assembler with interpreter-specific macros.
41
42 #ifdef PRODUCT
43 #define BLOCK_COMMENT(str) // nothing
44 #else
45 #define BLOCK_COMMENT(str) block_comment(str)
46 #endif
47
48 void InterpreterMacroAssembler::null_check_throw(Register a, int offset, Register temp_reg) {
49 address exception_entry = Interpreter::throw_NullPointerException_entry();
50 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry);
51 }
52
53 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) {
54 assert(entry, "Entry must have been generated by now");
55 if (is_within_range_of_b(entry, pc())) {
56 b(entry);
57 } else {
58 load_const_optimized(Rscratch, entry, R0);
59 mtctr(Rscratch);
60 bctr();
61 }
62 }
63
64 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) {
65 Register bytecode = R12_scratch2;
66 if (bcp_incr != 0) {
67 lbzu(bytecode, bcp_incr, R14_bcp);
68 } else {
69 lbz(bytecode, 0, R14_bcp);
70 }
71
72 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state), generate_poll);
73 }
74
75 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
76 // Load current bytecode.
77 Register bytecode = R12_scratch2;
78 lbz(bytecode, 0, R14_bcp);
79 dispatch_Lbyte_code(state, bytecode, table);
80 }
81
82 // Dispatch code executed in the prolog of a bytecode which does not do it's
83 // own dispatch. The dispatch address is computed and placed in R24_dispatch_addr.
84 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
85 Register bytecode = R12_scratch2;
86 lbz(bytecode, bcp_incr, R14_bcp);
87
88 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state));
89
90 sldi(bytecode, bytecode, LogBytesPerWord);
91 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode);
92 }
93
94 // Dispatch code executed in the epilog of a bytecode which does not do it's
95 // own dispatch. The dispatch address in R24_dispatch_addr is used for the
96 // dispatch.
97 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
98 if (bcp_incr) { addi(R14_bcp, R14_bcp, bcp_incr); }
99 mtctr(R24_dispatch_addr);
100 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
101 }
102
103 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
104 assert(scratch_reg != R0, "can't use R0 as scratch_reg here");
105 if (JvmtiExport::can_pop_frame()) {
106 Label L;
107
108 // Check the "pending popframe condition" flag in the current thread.
109 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
110
111 // Initiate popframe handling only if it is not already being
112 // processed. If the flag has the popframe_processing bit set, it
113 // means that this code is called *during* popframe handling - we
114 // don't want to reenter.
115 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit);
116 beq(CCR0, L);
117
118 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit);
119 bne(CCR0, L);
120
121 // Call the Interpreter::remove_activation_preserving_args_entry()
122 // func to get the address of the same-named entrypoint in the
123 // generated interpreter code.
124 #if defined(ABI_ELFv2)
125 call_c(CAST_FROM_FN_PTR(address,
126 Interpreter::remove_activation_preserving_args_entry),
127 relocInfo::none);
128 #else
129 call_c(CAST_FROM_FN_PTR(FunctionDescriptor*,
130 Interpreter::remove_activation_preserving_args_entry),
131 relocInfo::none);
132 #endif
133
134 // Jump to Interpreter::_remove_activation_preserving_args_entry.
135 mtctr(R3_RET);
136 bctr();
137
138 align(32, 12);
139 bind(L);
140 }
141 }
142
143 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
144 const Register Rthr_state_addr = scratch_reg;
145 if (JvmtiExport::can_force_early_return()) {
146 Label Lno_early_ret;
147 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
148 cmpdi(CCR0, Rthr_state_addr, 0);
149 beq(CCR0, Lno_early_ret);
150
151 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr);
152 cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending);
153 bne(CCR0, Lno_early_ret);
154
155 // Jump to Interpreter::_earlyret_entry.
156 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr);
157 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry));
158 mtlr(R3_RET);
159 blr();
160
161 align(32, 12);
162 bind(Lno_early_ret);
163 }
164 }
165
166 void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) {
167 const Register RjvmtiState = Rscratch1;
168 const Register Rscratch2 = R0;
169
170 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
171 li(Rscratch2, 0);
172
173 switch (state) {
174 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
175 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
176 break;
177 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
178 break;
179 case btos: // fall through
180 case ztos: // fall through
181 case ctos: // fall through
182 case stos: // fall through
183 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
184 break;
185 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
186 break;
187 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
188 break;
189 case vtos: break;
190 default : ShouldNotReachHere();
191 }
192
193 // Clean up tos value in the jvmti thread state.
194 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
195 // Set tos state field to illegal value.
196 li(Rscratch2, ilgl);
197 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState);
198 }
199
200 // Common code to dispatch and dispatch_only.
201 // Dispatch value in Lbyte_code and increment Lbcp.
202
203 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) {
204 address table_base = (address)Interpreter::dispatch_table((TosState)0);
205 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base;
206 if (is_simm16(table_offs)) {
207 addi(dst, R25_templateTableBase, (int)table_offs);
208 } else {
209 load_const_optimized(dst, table, R0);
210 }
211 }
212
213 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode,
214 address* table, bool generate_poll) {
215 assert_different_registers(bytecode, R11_scratch1);
216
217 // Calc dispatch table address.
218 load_dispatch_table(R11_scratch1, table);
219
220 if (SafepointMechanism::uses_thread_local_poll() && generate_poll) {
221 address *sfpt_tbl = Interpreter::safept_table(state);
222 if (table != sfpt_tbl) {
223 Label dispatch;
224 ld(R0, in_bytes(Thread::polling_page_offset()), R16_thread);
225 // Armed page has poll_bit set, if poll bit is cleared just continue.
226 andi_(R0, R0, SafepointMechanism::poll_bit());
227 beq(CCR0, dispatch);
228 load_dispatch_table(R11_scratch1, sfpt_tbl);
229 align(32, 16);
230 bind(dispatch);
231 }
232 }
233
234 sldi(R12_scratch2, bytecode, LogBytesPerWord);
235 ldx(R11_scratch1, R11_scratch1, R12_scratch2);
236
237 // Jump off!
238 mtctr(R11_scratch1);
239 bcctr(bcondAlways, 0, bhintbhBCCTRisNotPredictable);
240 }
241
242 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) {
243 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize);
244 ldx(Rrecv_dst, Rrecv_dst, R15_esp);
245 }
246
247 // helpers for expression stack
248
249 void InterpreterMacroAssembler::pop_i(Register r) {
250 lwzu(r, Interpreter::stackElementSize, R15_esp);
251 }
252
253 void InterpreterMacroAssembler::pop_ptr(Register r) {
254 ldu(r, Interpreter::stackElementSize, R15_esp);
255 }
256
257 void InterpreterMacroAssembler::pop_l(Register r) {
258 ld(r, Interpreter::stackElementSize, R15_esp);
259 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
260 }
261
262 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
263 lfsu(f, Interpreter::stackElementSize, R15_esp);
264 }
265
266 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
267 lfd(f, Interpreter::stackElementSize, R15_esp);
268 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
269 }
270
271 void InterpreterMacroAssembler::push_i(Register r) {
272 stw(r, 0, R15_esp);
273 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
274 }
275
276 void InterpreterMacroAssembler::push_ptr(Register r) {
277 std(r, 0, R15_esp);
278 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
279 }
280
281 void InterpreterMacroAssembler::push_l(Register r) {
282 // Clear unused slot.
283 load_const_optimized(R0, 0L);
284 std(R0, 0, R15_esp);
285 std(r, - Interpreter::stackElementSize, R15_esp);
286 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
287 }
288
289 void InterpreterMacroAssembler::push_f(FloatRegister f) {
290 stfs(f, 0, R15_esp);
291 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
292 }
293
294 void InterpreterMacroAssembler::push_d(FloatRegister f) {
295 stfd(f, - Interpreter::stackElementSize, R15_esp);
296 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
297 }
298
299 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) {
300 std(first, 0, R15_esp);
301 std(second, -Interpreter::stackElementSize, R15_esp);
302 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
303 }
304
305 void InterpreterMacroAssembler::move_l_to_d(Register l, FloatRegister d) {
306 if (VM_Version::has_mtfprd()) {
307 mtfprd(d, l);
308 } else {
309 std(l, 0, R15_esp);
310 lfd(d, 0, R15_esp);
311 }
312 }
313
314 void InterpreterMacroAssembler::move_d_to_l(FloatRegister d, Register l) {
315 if (VM_Version::has_mtfprd()) {
316 mffprd(l, d);
317 } else {
318 stfd(d, 0, R15_esp);
319 ld(l, 0, R15_esp);
320 }
321 }
322
323 void InterpreterMacroAssembler::push(TosState state) {
324 switch (state) {
325 case atos: push_ptr(); break;
326 case btos:
327 case ztos:
328 case ctos:
329 case stos:
330 case itos: push_i(); break;
331 case ltos: push_l(); break;
332 case ftos: push_f(); break;
333 case dtos: push_d(); break;
334 case vtos: /* nothing to do */ break;
335 default : ShouldNotReachHere();
336 }
337 }
338
339 void InterpreterMacroAssembler::pop(TosState state) {
340 switch (state) {
341 case atos: pop_ptr(); break;
342 case btos:
343 case ztos:
344 case ctos:
345 case stos:
346 case itos: pop_i(); break;
347 case ltos: pop_l(); break;
348 case ftos: pop_f(); break;
349 case dtos: pop_d(); break;
350 case vtos: /* nothing to do */ break;
351 default : ShouldNotReachHere();
352 }
353 verify_oop(R17_tos, state);
354 }
355
356 void InterpreterMacroAssembler::empty_expression_stack() {
357 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
358 }
359
360 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset,
361 Register Rdst,
362 signedOrNot is_signed) {
363 #if defined(VM_LITTLE_ENDIAN)
364 if (bcp_offset) {
365 load_const_optimized(Rdst, bcp_offset);
366 lhbrx(Rdst, R14_bcp, Rdst);
367 } else {
368 lhbrx(Rdst, R14_bcp);
369 }
370 if (is_signed == Signed) {
371 extsh(Rdst, Rdst);
372 }
373 #else
374 // Read Java big endian format.
375 if (is_signed == Signed) {
376 lha(Rdst, bcp_offset, R14_bcp);
377 } else {
378 lhz(Rdst, bcp_offset, R14_bcp);
379 }
380 #endif
381 }
382
383 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset,
384 Register Rdst,
385 signedOrNot is_signed) {
386 #if defined(VM_LITTLE_ENDIAN)
387 if (bcp_offset) {
388 load_const_optimized(Rdst, bcp_offset);
389 lwbrx(Rdst, R14_bcp, Rdst);
390 } else {
391 lwbrx(Rdst, R14_bcp);
392 }
393 if (is_signed == Signed) {
394 extsw(Rdst, Rdst);
395 }
396 #else
397 // Read Java big endian format.
398 if (bcp_offset & 3) { // Offset unaligned?
399 load_const_optimized(Rdst, bcp_offset);
400 if (is_signed == Signed) {
401 lwax(Rdst, R14_bcp, Rdst);
402 } else {
403 lwzx(Rdst, R14_bcp, Rdst);
404 }
405 } else {
406 if (is_signed == Signed) {
407 lwa(Rdst, bcp_offset, R14_bcp);
408 } else {
409 lwz(Rdst, bcp_offset, R14_bcp);
410 }
411 }
412 #endif
413 }
414
415
416 // Load the constant pool cache index from the bytecode stream.
417 //
418 // Kills / writes:
419 // - Rdst, Rscratch
420 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset,
421 size_t index_size) {
422 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
423 // Cache index is always in the native format, courtesy of Rewriter.
424 if (index_size == sizeof(u2)) {
425 lhz(Rdst, bcp_offset, R14_bcp);
426 } else if (index_size == sizeof(u4)) {
427 if (bcp_offset & 3) {
428 load_const_optimized(Rdst, bcp_offset);
429 lwax(Rdst, R14_bcp, Rdst);
430 } else {
431 lwa(Rdst, bcp_offset, R14_bcp);
432 }
433 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
434 nand(Rdst, Rdst, Rdst); // convert to plain index
435 } else if (index_size == sizeof(u1)) {
436 lbz(Rdst, bcp_offset, R14_bcp);
437 } else {
438 ShouldNotReachHere();
439 }
440 // Rdst now contains cp cache index.
441 }
442
443 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset,
444 size_t index_size) {
445 get_cache_index_at_bcp(cache, bcp_offset, index_size);
446 sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord));
447 add(cache, R27_constPoolCache, cache);
448 }
449
450 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format)
451 // from (Rsrc)+offset.
452 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset,
453 signedOrNot is_signed) {
454 #if defined(VM_LITTLE_ENDIAN)
455 if (offset) {
456 load_const_optimized(Rdst, offset);
457 lwbrx(Rdst, Rdst, Rsrc);
458 } else {
459 lwbrx(Rdst, Rsrc);
460 }
461 if (is_signed == Signed) {
462 extsw(Rdst, Rdst);
463 }
464 #else
465 if (is_signed == Signed) {
466 lwa(Rdst, offset, Rsrc);
467 } else {
468 lwz(Rdst, offset, Rsrc);
469 }
470 #endif
471 }
472
473 // Load object from cpool->resolved_references(index).
474 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index, Label *L_handle_null) {
475 assert_different_registers(result, index);
476 get_constant_pool(result);
477
478 // Convert from field index to resolved_references() index and from
479 // word index to byte offset. Since this is a java object, it can be compressed.
480 Register tmp = index; // reuse
481 sldi(tmp, index, LogBytesPerHeapOop);
482 // Load pointer for resolved_references[] objArray.
483 ld(result, ConstantPool::cache_offset_in_bytes(), result);
484 ld(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result);
485 resolve_oop_handle(result);
486 #ifdef ASSERT
487 Label index_ok;
488 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result);
489 sldi(R0, R0, LogBytesPerHeapOop);
490 cmpd(CCR0, tmp, R0);
491 blt(CCR0, index_ok);
492 stop("resolved reference index out of bounds", 0x09256);
493 bind(index_ok);
494 #endif
495 // Add in the index.
496 add(result, tmp, result);
497 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result, tmp, R0, false, 0, L_handle_null);
498 }
499
500 // load cpool->resolved_klass_at(index)
501 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool, Register Roffset, Register Rklass) {
502 // int value = *(Rcpool->int_at_addr(which));
503 // int resolved_klass_index = extract_low_short_from_int(value);
504 add(Roffset, Rcpool, Roffset);
505 #if defined(VM_LITTLE_ENDIAN)
506 lhz(Roffset, sizeof(ConstantPool), Roffset); // Roffset = resolved_klass_index
507 #else
508 lhz(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index
509 #endif
510
511 ld(Rklass, ConstantPool::resolved_klasses_offset_in_bytes(), Rcpool); // Rklass = Rcpool->_resolved_klasses
512
513 sldi(Roffset, Roffset, LogBytesPerWord);
514 addi(Roffset, Roffset, Array<Klass*>::base_offset_in_bytes());
515 isync(); // Order load of instance Klass wrt. tags.
516 ldx(Rklass, Rklass, Roffset);
517 }
518
519 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
520 Register cache,
521 Register method) {
522 const int method_offset = in_bytes(
523 ConstantPoolCache::base_offset() +
524 ((byte_no == TemplateTable::f2_byte)
525 ? ConstantPoolCacheEntry::f2_offset()
526 : ConstantPoolCacheEntry::f1_offset()));
527
528 ld(method, method_offset, cache); // get f1 Method*
529 }
530
531 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
532 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2.
533 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1,
534 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) {
535 // Profile the not-null value's klass.
536 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2);
537 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
538 profile_typecheck_failed(Rtmp1, Rtmp2);
539 }
540
541 // Separate these two to allow for delay slot in middle.
542 // These are used to do a test and full jump to exception-throwing code.
543
544 // Check that index is in range for array, then shift index by index_shift,
545 // and put arrayOop + shifted_index into res.
546 // Note: res is still shy of address by array offset into object.
547
548 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex,
549 int index_shift, Register Rtmp, Register Rres) {
550 // Check that index is in range for array, then shift index by index_shift,
551 // and put arrayOop + shifted_index into res.
552 // Note: res is still shy of address by array offset into object.
553 // Kills:
554 // - Rindex
555 // Writes:
556 // - Rres: Address that corresponds to the array index if check was successful.
557 verify_oop(Rarray);
558 const Register Rlength = R0;
559 const Register RsxtIndex = Rtmp;
560 Label LisNull, LnotOOR;
561
562 // Array nullcheck
563 if (!ImplicitNullChecks) {
564 cmpdi(CCR0, Rarray, 0);
565 beq(CCR0, LisNull);
566 } else {
567 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex);
568 }
569
570 // Rindex might contain garbage in upper bits (remember that we don't sign extend
571 // during integer arithmetic operations). So kill them and put value into same register
572 // where ArrayIndexOutOfBounds would expect the index in.
573 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit
574
575 // Index check
576 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray);
577 cmplw(CCR0, Rindex, Rlength);
578 sldi(RsxtIndex, RsxtIndex, index_shift);
579 blt(CCR0, LnotOOR);
580 // Index should be in R17_tos, array should be in R4_ARG2.
581 mr_if_needed(R17_tos, Rindex);
582 mr_if_needed(R4_ARG2, Rarray);
583 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
584 mtctr(Rtmp);
585 bctr();
586
587 if (!ImplicitNullChecks) {
588 bind(LisNull);
589 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry);
590 mtctr(Rtmp);
591 bctr();
592 }
593
594 align(32, 16);
595 bind(LnotOOR);
596
597 // Calc address
598 add(Rres, RsxtIndex, Rarray);
599 }
600
601 void InterpreterMacroAssembler::index_check(Register array, Register index,
602 int index_shift, Register tmp, Register res) {
603 // pop array
604 pop_ptr(array);
605
606 // check array
607 index_check_without_pop(array, index, index_shift, tmp, res);
608 }
609
610 void InterpreterMacroAssembler::get_const(Register Rdst) {
611 ld(Rdst, in_bytes(Method::const_offset()), R19_method);
612 }
613
614 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
615 get_const(Rdst);
616 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
617 }
618
619 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
620 get_constant_pool(Rdst);
621 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst);
622 }
623
624 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
625 get_constant_pool(Rcpool);
626 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool);
627 }
628
629 // Unlock if synchronized method.
630 //
631 // Unlock the receiver if this is a synchronized method.
632 // Unlock any Java monitors from synchronized blocks.
633 //
634 // If there are locked Java monitors
635 // If throw_monitor_exception
636 // throws IllegalMonitorStateException
637 // Else if install_monitor_exception
638 // installs IllegalMonitorStateException
639 // Else
640 // no error processing
641 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
642 bool throw_monitor_exception,
643 bool install_monitor_exception) {
644 Label Lunlocked, Lno_unlock;
645 {
646 Register Rdo_not_unlock_flag = R11_scratch1;
647 Register Raccess_flags = R12_scratch2;
648
649 // Check if synchronized method or unlocking prevented by
650 // JavaThread::do_not_unlock_if_synchronized flag.
651 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
652 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method);
653 li(R0, 0);
654 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag
655
656 push(state);
657
658 // Skip if we don't have to unlock.
659 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0.
660 beq(CCR0, Lunlocked);
661
662 cmpwi(CCR0, Rdo_not_unlock_flag, 0);
663 bne(CCR0, Lno_unlock);
664 }
665
666 // Unlock
667 {
668 Register Rmonitor_base = R11_scratch1;
669
670 Label Lunlock;
671 // If it's still locked, everything is ok, unlock it.
672 ld(Rmonitor_base, 0, R1_SP);
673 addi(Rmonitor_base, Rmonitor_base,
674 -(frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
675
676 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base);
677 cmpdi(CCR0, R0, 0);
678 bne(CCR0, Lunlock);
679
680 // If it's already unlocked, throw exception.
681 if (throw_monitor_exception) {
682 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
683 should_not_reach_here();
684 } else {
685 if (install_monitor_exception) {
686 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
687 b(Lunlocked);
688 }
689 }
690
691 bind(Lunlock);
692 unlock_object(Rmonitor_base);
693 }
694
695 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not.
696 bind(Lunlocked);
697 {
698 Label Lexception, Lrestart;
699 Register Rcurrent_obj_addr = R11_scratch1;
700 const int delta = frame::interpreter_frame_monitor_size_in_bytes();
701 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords");
702
703 bind(Lrestart);
704 // Set up search loop: Calc num of iterations.
705 {
706 Register Riterations = R12_scratch2;
707 Register Rmonitor_base = Rcurrent_obj_addr;
708 ld(Rmonitor_base, 0, R1_SP);
709 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base
710
711 subf_(Riterations, R26_monitor, Rmonitor_base);
712 ble(CCR0, Lno_unlock);
713
714 addi(Rcurrent_obj_addr, Rmonitor_base,
715 BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes());
716 // Check if any monitor is on stack, bail out if not
717 srdi(Riterations, Riterations, exact_log2(delta));
718 mtctr(Riterations);
719 }
720
721 // The search loop: Look for locked monitors.
722 {
723 const Register Rcurrent_obj = R0;
724 Label Lloop;
725
726 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
727 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
728 bind(Lloop);
729
730 // Check if current entry is used.
731 cmpdi(CCR0, Rcurrent_obj, 0);
732 bne(CCR0, Lexception);
733 // Preload next iteration's compare value.
734 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
735 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
736 bdnz(Lloop);
737 }
738 // Fell through: Everything's unlocked => finish.
739 b(Lno_unlock);
740
741 // An object is still locked => need to throw exception.
742 bind(Lexception);
743 if (throw_monitor_exception) {
744 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
745 should_not_reach_here();
746 } else {
747 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
748 // Unlock does not block, so don't have to worry about the frame.
749 Register Rmonitor_addr = R11_scratch1;
750 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta);
751 unlock_object(Rmonitor_addr);
752 if (install_monitor_exception) {
753 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
754 }
755 b(Lrestart);
756 }
757 }
758
759 align(32, 12);
760 bind(Lno_unlock);
761 pop(state);
762 }
763
764 // Support function for remove_activation & Co.
765 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc,
766 Register Rscratch1, Register Rscratch2) {
767 // Pop interpreter frame.
768 ld(Rscratch1, 0, R1_SP); // *SP
769 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp
770 ld(Rscratch2, 0, Rscratch1); // **SP
771 #ifdef ASSERT
772 {
773 Label Lok;
774 ld(R0, _ijava_state_neg(ijava_reserved), Rscratch1);
775 cmpdi(CCR0, R0, 0x5afe);
776 beq(CCR0, Lok);
777 stop("frame corrupted (remove activation)", 0x5afe);
778 bind(Lok);
779 }
780 #endif
781 if (return_pc!=noreg) {
782 ld(return_pc, _abi(lr), Rscratch1); // LR
783 }
784
785 // Merge top frames.
786 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP
787 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP
788 }
789
790 void InterpreterMacroAssembler::narrow(Register result) {
791 Register ret_type = R11_scratch1;
792 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
793 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1);
794
795 Label notBool, notByte, notChar, done;
796
797 // common case first
798 cmpwi(CCR0, ret_type, T_INT);
799 beq(CCR0, done);
800
801 cmpwi(CCR0, ret_type, T_BOOLEAN);
802 bne(CCR0, notBool);
803 andi(result, result, 0x1);
804 b(done);
805
806 bind(notBool);
807 cmpwi(CCR0, ret_type, T_BYTE);
808 bne(CCR0, notByte);
809 extsb(result, result);
810 b(done);
811
812 bind(notByte);
813 cmpwi(CCR0, ret_type, T_CHAR);
814 bne(CCR0, notChar);
815 andi(result, result, 0xffff);
816 b(done);
817
818 bind(notChar);
819 // cmpwi(CCR0, ret_type, T_SHORT); // all that's left
820 // bne(CCR0, done);
821 extsh(result, result);
822
823 // Nothing to do for T_INT
824 bind(done);
825 }
826
827 // Remove activation.
828 //
829 // Unlock the receiver if this is a synchronized method.
830 // Unlock any Java monitors from synchronized blocks.
831 // Remove the activation from the stack.
832 //
833 // If there are locked Java monitors
834 // If throw_monitor_exception
835 // throws IllegalMonitorStateException
836 // Else if install_monitor_exception
837 // installs IllegalMonitorStateException
838 // Else
839 // no error processing
840 void InterpreterMacroAssembler::remove_activation(TosState state,
841 bool throw_monitor_exception,
842 bool install_monitor_exception) {
843 BLOCK_COMMENT("remove_activation {");
844 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
845
846 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
847 notify_method_exit(false, state, NotifyJVMTI, true);
848
849 BLOCK_COMMENT("reserved_stack_check:");
850 if (StackReservedPages > 0) {
851 // Test if reserved zone needs to be enabled.
852 Label no_reserved_zone_enabling;
853
854 // Compare frame pointers. There is no good stack pointer, as with stack
855 // frame compression we can get different SPs when we do calls. A subsequent
856 // call could have a smaller SP, so that this compare succeeds for an
857 // inner call of the method annotated with ReservedStack.
858 ld_ptr(R0, JavaThread::reserved_stack_activation_offset(), R16_thread);
859 ld_ptr(R11_scratch1, _abi(callers_sp), R1_SP); // Load frame pointer.
860 cmpld(CCR0, R11_scratch1, R0);
861 blt_predict_taken(CCR0, no_reserved_zone_enabling);
862
863 // Enable reserved zone again, throw stack overflow exception.
864 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), R16_thread);
865 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
866
867 should_not_reach_here();
868
869 bind(no_reserved_zone_enabling);
870 }
871
872 verify_oop(R17_tos, state);
873 verify_thread();
874
875 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
876 mtlr(R0);
877 BLOCK_COMMENT("} remove_activation");
878 }
879
880 // Lock object
881 //
882 // Registers alive
883 // monitor - Address of the BasicObjectLock to be used for locking,
884 // which must be initialized with the object to lock.
885 // object - Address of the object to be locked.
886 //
887 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
888 if (UseHeavyMonitors) {
889 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
890 monitor, /*check_for_exceptions=*/true);
891 } else {
892 // template code:
893 //
894 // markOop displaced_header = obj->mark().set_unlocked();
895 // monitor->lock()->set_displaced_header(displaced_header);
896 // if (Atomic::cmpxchg(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
897 // // We stored the monitor address into the object's mark word.
898 // } else if (THREAD->is_lock_owned((address)displaced_header))
899 // // Simple recursive case.
900 // monitor->lock()->set_displaced_header(NULL);
901 // } else {
902 // // Slow path.
903 // InterpreterRuntime::monitorenter(THREAD, monitor);
904 // }
905
906 const Register displaced_header = R7_ARG5;
907 const Register object_mark_addr = R8_ARG6;
908 const Register current_header = R9_ARG7;
909 const Register tmp = R10_ARG8;
910
911 Label done;
912 Label cas_failed, slow_case;
913
914 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp);
915
916 // markOop displaced_header = obj->mark().set_unlocked();
917
918 // Load markOop from object into displaced_header.
919 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object);
920
921 if (UseBiasedLocking) {
922 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case);
923 }
924
925 // Set displaced_header to be (markOop of object | UNLOCK_VALUE).
926 ori(displaced_header, displaced_header, markOopDesc::unlocked_value);
927
928 // monitor->lock()->set_displaced_header(displaced_header);
929
930 // Initialize the box (Must happen before we update the object mark!).
931 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() +
932 BasicLock::displaced_header_offset_in_bytes(), monitor);
933
934 // if (Atomic::cmpxchg(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
935
936 // Store stack address of the BasicObjectLock (this is monitor) into object.
937 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
938
939 // Must fence, otherwise, preceding store(s) may float below cmpxchg.
940 // CmpxchgX sets CCR0 to cmpX(current, displaced).
941 cmpxchgd(/*flag=*/CCR0,
942 /*current_value=*/current_header,
943 /*compare_value=*/displaced_header, /*exchange_value=*/monitor,
944 /*where=*/object_mark_addr,
945 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq,
946 MacroAssembler::cmpxchgx_hint_acquire_lock(),
947 noreg,
948 &cas_failed,
949 /*check without membar and ldarx first*/true);
950
951 // If the compare-and-exchange succeeded, then we found an unlocked
952 // object and we have now locked it.
953 b(done);
954 bind(cas_failed);
955
956 // } else if (THREAD->is_lock_owned((address)displaced_header))
957 // // Simple recursive case.
958 // monitor->lock()->set_displaced_header(NULL);
959
960 // We did not see an unlocked object so try the fast recursive case.
961
962 // Check if owner is self by comparing the value in the markOop of object
963 // (current_header) with the stack pointer.
964 sub(current_header, current_header, R1_SP);
965
966 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
967 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place);
968
969 and_(R0/*==0?*/, current_header, tmp);
970 // If condition is true we are done and hence we can store 0 in the displaced
971 // header indicating it is a recursive lock.
972 bne(CCR0, slow_case);
973 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() +
974 BasicLock::displaced_header_offset_in_bytes(), monitor);
975 b(done);
976
977 // } else {
978 // // Slow path.
979 // InterpreterRuntime::monitorenter(THREAD, monitor);
980
981 // None of the above fast optimizations worked so we have to get into the
982 // slow case of monitor enter.
983 bind(slow_case);
984 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
985 monitor, /*check_for_exceptions=*/true);
986 // }
987 align(32, 12);
988 bind(done);
989 }
990 }
991
992 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
993 //
994 // Registers alive
995 // monitor - Address of the BasicObjectLock to be used for locking,
996 // which must be initialized with the object to lock.
997 //
998 // Throw IllegalMonitorException if object is not locked by current thread.
999 void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) {
1000 if (UseHeavyMonitors) {
1001 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1002 monitor, check_for_exceptions);
1003 } else {
1004
1005 // template code:
1006 //
1007 // if ((displaced_header = monitor->displaced_header()) == NULL) {
1008 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL.
1009 // monitor->set_obj(NULL);
1010 // } else if (Atomic::cmpxchg(displaced_header, obj->mark_addr(), monitor) == monitor) {
1011 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1012 // monitor->set_obj(NULL);
1013 // } else {
1014 // // Slow path.
1015 // InterpreterRuntime::monitorexit(THREAD, monitor);
1016 // }
1017
1018 const Register object = R7_ARG5;
1019 const Register displaced_header = R8_ARG6;
1020 const Register object_mark_addr = R9_ARG7;
1021 const Register current_header = R10_ARG8;
1022
1023 Label free_slot;
1024 Label slow_case;
1025
1026 assert_different_registers(object, displaced_header, object_mark_addr, current_header);
1027
1028 if (UseBiasedLocking) {
1029 // The object address from the monitor is in object.
1030 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor);
1031 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0");
1032 biased_locking_exit(CCR0, object, displaced_header, free_slot);
1033 }
1034
1035 // Test first if we are in the fast recursive case.
1036 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() +
1037 BasicLock::displaced_header_offset_in_bytes(), monitor);
1038
1039 // If the displaced header is zero, we have a recursive unlock.
1040 cmpdi(CCR0, displaced_header, 0);
1041 beq(CCR0, free_slot); // recursive unlock
1042
1043 // } else if (Atomic::cmpxchg(displaced_header, obj->mark_addr(), monitor) == monitor) {
1044 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1045 // monitor->set_obj(NULL);
1046
1047 // If we still have a lightweight lock, unlock the object and be done.
1048
1049 // The object address from the monitor is in object.
1050 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); }
1051 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
1052
1053 // We have the displaced header in displaced_header. If the lock is still
1054 // lightweight, it will contain the monitor address and we'll store the
1055 // displaced header back into the object's mark word.
1056 // CmpxchgX sets CCR0 to cmpX(current, monitor).
1057 cmpxchgd(/*flag=*/CCR0,
1058 /*current_value=*/current_header,
1059 /*compare_value=*/monitor, /*exchange_value=*/displaced_header,
1060 /*where=*/object_mark_addr,
1061 MacroAssembler::MemBarRel,
1062 MacroAssembler::cmpxchgx_hint_release_lock(),
1063 noreg,
1064 &slow_case);
1065 b(free_slot);
1066
1067 // } else {
1068 // // Slow path.
1069 // InterpreterRuntime::monitorexit(THREAD, monitor);
1070
1071 // The lock has been converted into a heavy lock and hence
1072 // we need to get into the slow case.
1073 bind(slow_case);
1074 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1075 monitor, check_for_exceptions);
1076 // }
1077
1078 Label done;
1079 b(done); // Monitor register may be overwritten! Runtime has already freed the slot.
1080
1081 // Exchange worked, do monitor->set_obj(NULL);
1082 align(32, 12);
1083 bind(free_slot);
1084 li(R0, 0);
1085 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor);
1086 bind(done);
1087 }
1088 }
1089
1090 // Load compiled (i2c) or interpreter entry when calling from interpreted and
1091 // do the call. Centralized so that all interpreter calls will do the same actions.
1092 // If jvmti single stepping is on for a thread we must not call compiled code.
1093 //
1094 // Input:
1095 // - Rtarget_method: method to call
1096 // - Rret_addr: return address
1097 // - 2 scratch regs
1098 //
1099 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr,
1100 Register Rscratch1, Register Rscratch2) {
1101 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr);
1102 // Assume we want to go compiled if available.
1103 const Register Rtarget_addr = Rscratch1;
1104 const Register Rinterp_only = Rscratch2;
1105
1106 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method);
1107
1108 if (JvmtiExport::can_post_interpreter_events()) {
1109 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
1110
1111 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
1112 // compiled code in threads for which the event is enabled. Check here for
1113 // interp_only_mode if these events CAN be enabled.
1114 Label done;
1115 verify_thread();
1116 cmpwi(CCR0, Rinterp_only, 0);
1117 beq(CCR0, done);
1118 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method);
1119 align(32, 12);
1120 bind(done);
1121 }
1122
1123 #ifdef ASSERT
1124 {
1125 Label Lok;
1126 cmpdi(CCR0, Rtarget_addr, 0);
1127 bne(CCR0, Lok);
1128 stop("null entry point");
1129 bind(Lok);
1130 }
1131 #endif // ASSERT
1132
1133 mr(R21_sender_SP, R1_SP);
1134
1135 // Calc a precise SP for the call. The SP value we calculated in
1136 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space
1137 // if esp is not max. Also, the i2c adapter extends the stack space without restoring
1138 // our pre-calced value, so repeating calls via i2c would result in stack overflow.
1139 // Since esp already points to an empty slot, we just have to sub 1 additional slot
1140 // to meet the abi scratch requirements.
1141 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in
1142 // the return entry of the interpreter.
1143 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size);
1144 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address
1145 resize_frame_absolute(Rscratch2, Rscratch2, R0);
1146
1147 mr_if_needed(R19_method, Rtarget_method);
1148 mtctr(Rtarget_addr);
1149 mtlr(Rret_addr);
1150
1151 save_interpreter_state(Rscratch2);
1152 #ifdef ASSERT
1153 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp
1154 cmpd(CCR0, R21_sender_SP, Rscratch1);
1155 asm_assert_eq("top_frame_sp incorrect", 0x951);
1156 #endif
1157
1158 bctr();
1159 }
1160
1161 // Set the method data pointer for the current bcp.
1162 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1163 assert(ProfileInterpreter, "must be profiling interpreter");
1164 Label get_continue;
1165 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method);
1166 test_method_data_pointer(get_continue);
1167 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp);
1168
1169 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1170 add(R28_mdx, R28_mdx, R3_RET);
1171 bind(get_continue);
1172 }
1173
1174 // Test ImethodDataPtr. If it is null, continue at the specified label.
1175 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1176 assert(ProfileInterpreter, "must be profiling interpreter");
1177 cmpdi(CCR0, R28_mdx, 0);
1178 beq(CCR0, zero_continue);
1179 }
1180
1181 void InterpreterMacroAssembler::verify_method_data_pointer() {
1182 assert(ProfileInterpreter, "must be profiling interpreter");
1183 #ifdef ASSERT
1184 Label verify_continue;
1185 test_method_data_pointer(verify_continue);
1186
1187 // If the mdp is valid, it will point to a DataLayout header which is
1188 // consistent with the bcp. The converse is highly probable also.
1189 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx);
1190 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method);
1191 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1192 add(R11_scratch1, R12_scratch2, R12_scratch2);
1193 cmpd(CCR0, R11_scratch1, R14_bcp);
1194 beq(CCR0, verify_continue);
1195
1196 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx);
1197
1198 bind(verify_continue);
1199 #endif
1200 }
1201
1202 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1203 Register method_counters,
1204 Register Rscratch,
1205 Label &profile_continue) {
1206 assert(ProfileInterpreter, "must be profiling interpreter");
1207 // Control will flow to "profile_continue" if the counter is less than the
1208 // limit or if we call profile_method().
1209 Label done;
1210
1211 // If no method data exists, and the counter is high enough, make one.
1212 lwz(Rscratch, in_bytes(MethodCounters::interpreter_profile_limit_offset()), method_counters);
1213
1214 cmpdi(CCR0, R28_mdx, 0);
1215 // Test to see if we should create a method data oop.
1216 cmpd(CCR1, Rscratch, invocation_count);
1217 bne(CCR0, done);
1218 bge(CCR1, profile_continue);
1219
1220 // Build it now.
1221 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1222 set_method_data_pointer_for_bcp();
1223 b(profile_continue);
1224
1225 align(32, 12);
1226 bind(done);
1227 }
1228
1229 void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register method_counters,
1230 Register target_bcp, Register disp, Register Rtmp) {
1231 assert_different_registers(backedge_count, target_bcp, disp, Rtmp, R4_ARG2);
1232 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
1233
1234 Label did_not_overflow;
1235 Label overflow_with_error;
1236
1237 lwz(Rtmp, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()), method_counters);
1238 cmpw(CCR0, backedge_count, Rtmp);
1239
1240 blt(CCR0, did_not_overflow);
1241
1242 // When ProfileInterpreter is on, the backedge_count comes from the
1243 // methodDataOop, which value does not get reset on the call to
1244 // frequency_counter_overflow(). To avoid excessive calls to the overflow
1245 // routine while the method is being compiled, add a second test to make sure
1246 // the overflow function is called only once every overflow_frequency.
1247 if (ProfileInterpreter) {
1248 const int overflow_frequency = 1024;
1249 andi_(Rtmp, backedge_count, overflow_frequency-1);
1250 bne(CCR0, did_not_overflow);
1251 }
1252
1253 // Overflow in loop, pass branch bytecode.
1254 subf(R4_ARG2, disp, target_bcp); // Compute branch bytecode (previous bcp).
1255 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
1256
1257 // Was an OSR adapter generated?
1258 cmpdi(CCR0, R3_RET, 0);
1259 beq(CCR0, overflow_with_error);
1260
1261 // Has the nmethod been invalidated already?
1262 lbz(Rtmp, nmethod::state_offset(), R3_RET);
1263 cmpwi(CCR0, Rtmp, nmethod::in_use);
1264 bne(CCR0, overflow_with_error);
1265
1266 // Migrate the interpreter frame off of the stack.
1267 // We can use all registers because we will not return to interpreter from this point.
1268
1269 // Save nmethod.
1270 const Register osr_nmethod = R31;
1271 mr(osr_nmethod, R3_RET);
1272 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1273 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1274 reset_last_Java_frame();
1275 // OSR buffer is in ARG1
1276
1277 // Remove the interpreter frame.
1278 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1279
1280 // Jump to the osr code.
1281 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1282 mtlr(R0);
1283 mtctr(R11_scratch1);
1284 bctr();
1285
1286 align(32, 12);
1287 bind(overflow_with_error);
1288 bind(did_not_overflow);
1289 }
1290
1291 // Store a value at some constant offset from the method data pointer.
1292 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1293 assert(ProfileInterpreter, "must be profiling interpreter");
1294
1295 std(value, constant, R28_mdx);
1296 }
1297
1298 // Increment the value at some constant offset from the method data pointer.
1299 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1300 Register counter_addr,
1301 Register Rbumped_count,
1302 bool decrement) {
1303 // Locate the counter at a fixed offset from the mdp:
1304 addi(counter_addr, R28_mdx, constant);
1305 increment_mdp_data_at(counter_addr, Rbumped_count, decrement);
1306 }
1307
1308 // Increment the value at some non-fixed (reg + constant) offset from
1309 // the method data pointer.
1310 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1311 int constant,
1312 Register scratch,
1313 Register Rbumped_count,
1314 bool decrement) {
1315 // Add the constant to reg to get the offset.
1316 add(scratch, R28_mdx, reg);
1317 // Then calculate the counter address.
1318 addi(scratch, scratch, constant);
1319 increment_mdp_data_at(scratch, Rbumped_count, decrement);
1320 }
1321
1322 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr,
1323 Register Rbumped_count,
1324 bool decrement) {
1325 assert(ProfileInterpreter, "must be profiling interpreter");
1326
1327 // Load the counter.
1328 ld(Rbumped_count, 0, counter_addr);
1329
1330 if (decrement) {
1331 // Decrement the register. Set condition codes.
1332 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment);
1333 // Store the decremented counter, if it is still negative.
1334 std(Rbumped_count, 0, counter_addr);
1335 // Note: add/sub overflow check are not ported, since 64 bit
1336 // calculation should never overflow.
1337 } else {
1338 // Increment the register. Set carry flag.
1339 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment);
1340 // Store the incremented counter.
1341 std(Rbumped_count, 0, counter_addr);
1342 }
1343 }
1344
1345 // Set a flag value at the current method data pointer position.
1346 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1347 Register scratch) {
1348 assert(ProfileInterpreter, "must be profiling interpreter");
1349 // Load the data header.
1350 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1351 // Set the flag.
1352 ori(scratch, scratch, flag_constant);
1353 // Store the modified header.
1354 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1355 }
1356
1357 // Test the location at some offset from the method data pointer.
1358 // If it is not equal to value, branch to the not_equal_continue Label.
1359 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1360 Register value,
1361 Label& not_equal_continue,
1362 Register test_out) {
1363 assert(ProfileInterpreter, "must be profiling interpreter");
1364
1365 ld(test_out, offset, R28_mdx);
1366 cmpd(CCR0, value, test_out);
1367 bne(CCR0, not_equal_continue);
1368 }
1369
1370 // Update the method data pointer by the displacement located at some fixed
1371 // offset from the method data pointer.
1372 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1373 Register scratch) {
1374 assert(ProfileInterpreter, "must be profiling interpreter");
1375
1376 ld(scratch, offset_of_disp, R28_mdx);
1377 add(R28_mdx, scratch, R28_mdx);
1378 }
1379
1380 // Update the method data pointer by the displacement located at the
1381 // offset (reg + offset_of_disp).
1382 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1383 int offset_of_disp,
1384 Register scratch) {
1385 assert(ProfileInterpreter, "must be profiling interpreter");
1386
1387 add(scratch, reg, R28_mdx);
1388 ld(scratch, offset_of_disp, scratch);
1389 add(R28_mdx, scratch, R28_mdx);
1390 }
1391
1392 // Update the method data pointer by a simple constant displacement.
1393 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1394 assert(ProfileInterpreter, "must be profiling interpreter");
1395 addi(R28_mdx, R28_mdx, constant);
1396 }
1397
1398 // Update the method data pointer for a _ret bytecode whose target
1399 // was not among our cached targets.
1400 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1401 Register return_bci) {
1402 assert(ProfileInterpreter, "must be profiling interpreter");
1403
1404 push(state);
1405 assert(return_bci->is_nonvolatile(), "need to protect return_bci");
1406 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1407 pop(state);
1408 }
1409
1410 // Increments the backedge counter.
1411 // Returns backedge counter + invocation counter in Rdst.
1412 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst,
1413 const Register Rtmp1, Register Rscratch) {
1414 assert(UseCompiler, "incrementing must be useful");
1415 assert_different_registers(Rdst, Rtmp1);
1416 const Register invocation_counter = Rtmp1;
1417 const Register counter = Rdst;
1418 // TODO: PPC port: assert(4 == InvocationCounter::sz_counter(), "unexpected field size.");
1419
1420 // Load backedge counter.
1421 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1422 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1423 // Load invocation counter.
1424 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) +
1425 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1426
1427 // Add the delta to the backedge counter.
1428 addi(counter, counter, InvocationCounter::count_increment);
1429
1430 // Mask the invocation counter.
1431 andi(invocation_counter, invocation_counter, InvocationCounter::count_mask_value);
1432
1433 // Store new counter value.
1434 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1435 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1436 // Return invocation counter + backedge counter.
1437 add(counter, counter, invocation_counter);
1438 }
1439
1440 // Count a taken branch in the bytecodes.
1441 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1442 if (ProfileInterpreter) {
1443 Label profile_continue;
1444
1445 // If no method data exists, go to profile_continue.
1446 test_method_data_pointer(profile_continue);
1447
1448 // We are taking a branch. Increment the taken count.
1449 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count);
1450
1451 // The method data pointer needs to be updated to reflect the new target.
1452 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1453 bind (profile_continue);
1454 }
1455 }
1456
1457 // Count a not-taken branch in the bytecodes.
1458 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) {
1459 if (ProfileInterpreter) {
1460 Label profile_continue;
1461
1462 // If no method data exists, go to profile_continue.
1463 test_method_data_pointer(profile_continue);
1464
1465 // We are taking a branch. Increment the not taken count.
1466 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2);
1467
1468 // The method data pointer needs to be updated to correspond to the
1469 // next bytecode.
1470 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1471 bind (profile_continue);
1472 }
1473 }
1474
1475 // Count a non-virtual call in the bytecodes.
1476 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) {
1477 if (ProfileInterpreter) {
1478 Label profile_continue;
1479
1480 // If no method data exists, go to profile_continue.
1481 test_method_data_pointer(profile_continue);
1482
1483 // We are making a call. Increment the count.
1484 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1485
1486 // The method data pointer needs to be updated to reflect the new target.
1487 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1488 bind (profile_continue);
1489 }
1490 }
1491
1492 // Count a final call in the bytecodes.
1493 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) {
1494 if (ProfileInterpreter) {
1495 Label profile_continue;
1496
1497 // If no method data exists, go to profile_continue.
1498 test_method_data_pointer(profile_continue);
1499
1500 // We are making a call. Increment the count.
1501 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1502
1503 // The method data pointer needs to be updated to reflect the new target.
1504 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1505 bind (profile_continue);
1506 }
1507 }
1508
1509 // Count a virtual call in the bytecodes.
1510 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver,
1511 Register Rscratch1,
1512 Register Rscratch2,
1513 bool receiver_can_be_null) {
1514 if (!ProfileInterpreter) { return; }
1515 Label profile_continue;
1516
1517 // If no method data exists, go to profile_continue.
1518 test_method_data_pointer(profile_continue);
1519
1520 Label skip_receiver_profile;
1521 if (receiver_can_be_null) {
1522 Label not_null;
1523 cmpdi(CCR0, Rreceiver, 0);
1524 bne(CCR0, not_null);
1525 // We are making a call. Increment the count for null receiver.
1526 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2);
1527 b(skip_receiver_profile);
1528 bind(not_null);
1529 }
1530
1531 // Record the receiver type.
1532 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true);
1533 bind(skip_receiver_profile);
1534
1535 // The method data pointer needs to be updated to reflect the new target.
1536 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1537 bind (profile_continue);
1538 }
1539
1540 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) {
1541 if (ProfileInterpreter) {
1542 Label profile_continue;
1543
1544 // If no method data exists, go to profile_continue.
1545 test_method_data_pointer(profile_continue);
1546
1547 int mdp_delta = in_bytes(BitData::bit_data_size());
1548 if (TypeProfileCasts) {
1549 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1550
1551 // Record the object type.
1552 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false);
1553 }
1554
1555 // The method data pointer needs to be updated.
1556 update_mdp_by_constant(mdp_delta);
1557
1558 bind (profile_continue);
1559 }
1560 }
1561
1562 void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) {
1563 if (ProfileInterpreter && TypeProfileCasts) {
1564 Label profile_continue;
1565
1566 // If no method data exists, go to profile_continue.
1567 test_method_data_pointer(profile_continue);
1568
1569 int count_offset = in_bytes(CounterData::count_offset());
1570 // Back up the address, since we have already bumped the mdp.
1571 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1572
1573 // *Decrement* the counter. We expect to see zero or small negatives.
1574 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true);
1575
1576 bind (profile_continue);
1577 }
1578 }
1579
1580 // Count a ret in the bytecodes.
1581 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci,
1582 Register scratch1, Register scratch2) {
1583 if (ProfileInterpreter) {
1584 Label profile_continue;
1585 uint row;
1586
1587 // If no method data exists, go to profile_continue.
1588 test_method_data_pointer(profile_continue);
1589
1590 // Update the total ret count.
1591 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 );
1592
1593 for (row = 0; row < RetData::row_limit(); row++) {
1594 Label next_test;
1595
1596 // See if return_bci is equal to bci[n]:
1597 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1);
1598
1599 // return_bci is equal to bci[n]. Increment the count.
1600 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2);
1601
1602 // The method data pointer needs to be updated to reflect the new target.
1603 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1);
1604 b(profile_continue);
1605 bind(next_test);
1606 }
1607
1608 update_mdp_for_ret(state, return_bci);
1609
1610 bind (profile_continue);
1611 }
1612 }
1613
1614 // Count the default case of a switch construct.
1615 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) {
1616 if (ProfileInterpreter) {
1617 Label profile_continue;
1618
1619 // If no method data exists, go to profile_continue.
1620 test_method_data_pointer(profile_continue);
1621
1622 // Update the default case count
1623 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1624 scratch1, scratch2);
1625
1626 // The method data pointer needs to be updated.
1627 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()),
1628 scratch1);
1629
1630 bind (profile_continue);
1631 }
1632 }
1633
1634 // Count the index'th case of a switch construct.
1635 void InterpreterMacroAssembler::profile_switch_case(Register index,
1636 Register scratch1,
1637 Register scratch2,
1638 Register scratch3) {
1639 if (ProfileInterpreter) {
1640 assert_different_registers(index, scratch1, scratch2, scratch3);
1641 Label profile_continue;
1642
1643 // If no method data exists, go to profile_continue.
1644 test_method_data_pointer(profile_continue);
1645
1646 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes().
1647 li(scratch3, in_bytes(MultiBranchData::case_array_offset()));
1648
1649 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works");
1650 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1651 add(scratch1, scratch1, scratch3);
1652
1653 // Update the case count.
1654 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3);
1655
1656 // The method data pointer needs to be updated.
1657 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2);
1658
1659 bind (profile_continue);
1660 }
1661 }
1662
1663 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) {
1664 if (ProfileInterpreter) {
1665 assert_different_registers(Rscratch1, Rscratch2);
1666 Label profile_continue;
1667
1668 // If no method data exists, go to profile_continue.
1669 test_method_data_pointer(profile_continue);
1670
1671 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1);
1672
1673 // The method data pointer needs to be updated.
1674 int mdp_delta = in_bytes(BitData::bit_data_size());
1675 if (TypeProfileCasts) {
1676 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1677 }
1678 update_mdp_by_constant(mdp_delta);
1679
1680 bind (profile_continue);
1681 }
1682 }
1683
1684 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver,
1685 Register Rscratch1, Register Rscratch2,
1686 bool is_virtual_call) {
1687 assert(ProfileInterpreter, "must be profiling");
1688 assert_different_registers(Rreceiver, Rscratch1, Rscratch2);
1689
1690 Label done;
1691 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call);
1692 bind (done);
1693 }
1694
1695 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1696 Register receiver, Register scratch1, Register scratch2,
1697 int start_row, Label& done, bool is_virtual_call) {
1698 if (TypeProfileWidth == 0) {
1699 if (is_virtual_call) {
1700 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1701 }
1702 return;
1703 }
1704
1705 int last_row = VirtualCallData::row_limit() - 1;
1706 assert(start_row <= last_row, "must be work left to do");
1707 // Test this row for both the receiver and for null.
1708 // Take any of three different outcomes:
1709 // 1. found receiver => increment count and goto done
1710 // 2. found null => keep looking for case 1, maybe allocate this cell
1711 // 3. found something else => keep looking for cases 1 and 2
1712 // Case 3 is handled by a recursive call.
1713 for (int row = start_row; row <= last_row; row++) {
1714 Label next_test;
1715 bool test_for_null_also = (row == start_row);
1716
1717 // See if the receiver is receiver[n].
1718 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1719 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1);
1720 // delayed()->tst(scratch);
1721
1722 // The receiver is receiver[n]. Increment count[n].
1723 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1724 increment_mdp_data_at(count_offset, scratch1, scratch2);
1725 b(done);
1726 bind(next_test);
1727
1728 if (test_for_null_also) {
1729 Label found_null;
1730 // Failed the equality check on receiver[n]... Test for null.
1731 if (start_row == last_row) {
1732 // The only thing left to do is handle the null case.
1733 if (is_virtual_call) {
1734 // Scratch1 contains test_out from test_mdp_data_at.
1735 cmpdi(CCR0, scratch1, 0);
1736 beq(CCR0, found_null);
1737 // Receiver did not match any saved receiver and there is no empty row for it.
1738 // Increment total counter to indicate polymorphic case.
1739 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1740 b(done);
1741 bind(found_null);
1742 } else {
1743 cmpdi(CCR0, scratch1, 0);
1744 bne(CCR0, done);
1745 }
1746 break;
1747 }
1748 // Since null is rare, make it be the branch-taken case.
1749 cmpdi(CCR0, scratch1, 0);
1750 beq(CCR0, found_null);
1751
1752 // Put all the "Case 3" tests here.
1753 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call);
1754
1755 // Found a null. Keep searching for a matching receiver,
1756 // but remember that this is an empty (unused) slot.
1757 bind(found_null);
1758 }
1759 }
1760
1761 // In the fall-through case, we found no matching receiver, but we
1762 // observed the receiver[start_row] is NULL.
1763
1764 // Fill in the receiver field and increment the count.
1765 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1766 set_mdp_data_at(recvr_offset, receiver);
1767 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1768 li(scratch1, DataLayout::counter_increment);
1769 set_mdp_data_at(count_offset, scratch1);
1770 if (start_row > 0) {
1771 b(done);
1772 }
1773 }
1774
1775 // Argument and return type profilig.
1776 // kills: tmp, tmp2, R0, CR0, CR1
1777 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base,
1778 RegisterOrConstant mdo_addr_offs,
1779 Register tmp, Register tmp2) {
1780 Label do_nothing, do_update;
1781
1782 // tmp2 = obj is allowed
1783 assert_different_registers(obj, mdo_addr_base, tmp, R0);
1784 assert_different_registers(tmp2, mdo_addr_base, tmp, R0);
1785 const Register klass = tmp2;
1786
1787 verify_oop(obj);
1788
1789 ld(tmp, mdo_addr_offs, mdo_addr_base);
1790
1791 // Set null_seen if obj is 0.
1792 cmpdi(CCR0, obj, 0);
1793 ori(R0, tmp, TypeEntries::null_seen);
1794 beq(CCR0, do_update);
1795
1796 load_klass(klass, obj);
1797
1798 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
1799 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
1800 cmpd(CCR1, R0, klass);
1801 // Klass seen before, nothing to do (regardless of unknown bit).
1802 //beq(CCR1, do_nothing);
1803
1804 andi_(R0, klass, TypeEntries::type_unknown);
1805 // Already unknown. Nothing to do anymore.
1806 //bne(CCR0, do_nothing);
1807 crorc(CCR0, Assembler::equal, CCR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne
1808 beq(CCR0, do_nothing);
1809
1810 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
1811 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1812 beq(CCR0, do_update); // First time here. Set profile type.
1813
1814 // Different than before. Cannot keep accurate profile.
1815 ori(R0, tmp, TypeEntries::type_unknown);
1816
1817 bind(do_update);
1818 // update profile
1819 std(R0, mdo_addr_offs, mdo_addr_base);
1820
1821 align(32, 12);
1822 bind(do_nothing);
1823 }
1824
1825 void InterpreterMacroAssembler::profile_arguments_type(Register callee,
1826 Register tmp1, Register tmp2,
1827 bool is_virtual) {
1828 if (!ProfileInterpreter) {
1829 return;
1830 }
1831
1832 assert_different_registers(callee, tmp1, tmp2, R28_mdx);
1833
1834 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1835 Label profile_continue;
1836
1837 test_method_data_pointer(profile_continue);
1838
1839 int off_to_start = is_virtual ?
1840 in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1841
1842 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx);
1843 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1844 bne(CCR0, profile_continue);
1845
1846 if (MethodData::profile_arguments()) {
1847 Label done;
1848 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1849 add(R28_mdx, off_to_args, R28_mdx);
1850
1851 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1852 if (i > 0 || MethodData::profile_return()) {
1853 // If return value type is profiled we may have no argument to profile.
1854 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1855 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count());
1856 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count());
1857 blt(CCR0, done);
1858 }
1859 ld(tmp1, in_bytes(Method::const_offset()), callee);
1860 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1);
1861 // Stack offset o (zero based) from the start of the argument
1862 // list, for n arguments translates into offset n - o - 1 from
1863 // the end of the argument list. But there's an extra slot at
1864 // the top of the stack. So the offset is n - o from Lesp.
1865 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx);
1866 subf(tmp1, tmp2, tmp1);
1867
1868 sldi(tmp1, tmp1, Interpreter::logStackElementSize);
1869 ldx(tmp1, tmp1, R15_esp);
1870
1871 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1);
1872
1873 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1874 addi(R28_mdx, R28_mdx, to_add);
1875 off_to_args += to_add;
1876 }
1877
1878 if (MethodData::profile_return()) {
1879 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1880 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1881 }
1882
1883 bind(done);
1884
1885 if (MethodData::profile_return()) {
1886 // We're right after the type profile for the last
1887 // argument. tmp1 is the number of cells left in the
1888 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1889 // if there's a return to profile.
1890 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(),
1891 "can't move past ret type");
1892 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size));
1893 add(R28_mdx, tmp1, R28_mdx);
1894 }
1895 } else {
1896 assert(MethodData::profile_return(), "either profile call args or call ret");
1897 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
1898 }
1899
1900 // Mdp points right after the end of the
1901 // CallTypeData/VirtualCallTypeData, right after the cells for the
1902 // return value type if there's one.
1903 align(32, 12);
1904 bind(profile_continue);
1905 }
1906 }
1907
1908 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
1909 assert_different_registers(ret, tmp1, tmp2);
1910 if (ProfileInterpreter && MethodData::profile_return()) {
1911 Label profile_continue;
1912
1913 test_method_data_pointer(profile_continue);
1914
1915 if (MethodData::profile_return_jsr292_only()) {
1916 // If we don't profile all invoke bytecodes we must make sure
1917 // it's a bytecode we indeed profile. We can't go back to the
1918 // begining of the ProfileData we intend to update to check its
1919 // type because we're right after it and we don't known its
1920 // length.
1921 lbz(tmp1, 0, R14_bcp);
1922 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method);
1923 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic);
1924 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle);
1925 cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
1926 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm);
1927 cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
1928 bne(CCR0, profile_continue);
1929 }
1930
1931 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2);
1932
1933 align(32, 12);
1934 bind(profile_continue);
1935 }
1936 }
1937
1938 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2,
1939 Register tmp3, Register tmp4) {
1940 if (ProfileInterpreter && MethodData::profile_parameters()) {
1941 Label profile_continue, done;
1942
1943 test_method_data_pointer(profile_continue);
1944
1945 // Load the offset of the area within the MDO used for
1946 // parameters. If it's negative we're not profiling any parameters.
1947 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx);
1948 cmpwi(CCR0, tmp1, 0);
1949 blt(CCR0, profile_continue);
1950
1951 // Compute a pointer to the area for parameters from the offset
1952 // and move the pointer to the slot for the last
1953 // parameters. Collect profiling from last parameter down.
1954 // mdo start + parameters offset + array length - 1
1955
1956 // Pointer to the parameter area in the MDO.
1957 const Register mdp = tmp1;
1958 add(mdp, tmp1, R28_mdx);
1959
1960 // Offset of the current profile entry to update.
1961 const Register entry_offset = tmp2;
1962 // entry_offset = array len in number of cells
1963 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp);
1964
1965 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1966 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
1967
1968 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
1969 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size));
1970 // entry_offset in bytes
1971 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1972
1973 Label loop;
1974 align(32, 12);
1975 bind(loop);
1976
1977 // Load offset on the stack from the slot for this parameter.
1978 ld(tmp3, entry_offset, mdp);
1979 sldi(tmp3, tmp3, Interpreter::logStackElementSize);
1980 neg(tmp3, tmp3);
1981 // Read the parameter from the local area.
1982 ldx(tmp3, tmp3, R18_locals);
1983
1984 // Make entry_offset now point to the type field for this parameter.
1985 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1986 assert(type_base > off_base, "unexpected");
1987 addi(entry_offset, entry_offset, type_base - off_base);
1988
1989 // Profile the parameter.
1990 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3);
1991
1992 // Go to next parameter.
1993 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base);
1994 cmpdi(CCR0, entry_offset, off_base + delta);
1995 addi(entry_offset, entry_offset, -delta);
1996 bge(CCR0, loop);
1997
1998 align(32, 12);
1999 bind(profile_continue);
2000 }
2001 }
2002
2003 // Add a InterpMonitorElem to stack (see frame_sparc.hpp).
2004 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) {
2005
2006 // Very-local scratch registers.
2007 const Register esp = Rtemp1;
2008 const Register slot = Rtemp2;
2009
2010 // Extracted monitor_size.
2011 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes();
2012 assert(Assembler::is_aligned((unsigned int)monitor_size,
2013 (unsigned int)frame::alignment_in_bytes),
2014 "size of a monitor must respect alignment of SP");
2015
2016 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor
2017 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp
2018
2019 // Shuffle expression stack down. Recall that stack_base points
2020 // just above the new expression stack bottom. Old_tos and new_tos
2021 // are used to scan thru the old and new expression stacks.
2022 if (!stack_is_empty) {
2023 Label copy_slot, copy_slot_finished;
2024 const Register n_slots = slot;
2025
2026 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack).
2027 subf(n_slots, esp, R26_monitor);
2028 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy.
2029 assert(LogBytesPerWord == 3, "conflicts assembler instructions");
2030 beq(CCR0, copy_slot_finished); // Nothing to copy.
2031
2032 mtctr(n_slots);
2033
2034 // loop
2035 bind(copy_slot);
2036 ld(slot, 0, esp); // Move expression stack down.
2037 std(slot, -monitor_size, esp); // distance = monitor_size
2038 addi(esp, esp, BytesPerWord);
2039 bdnz(copy_slot);
2040
2041 bind(copy_slot_finished);
2042 }
2043
2044 addi(R15_esp, R15_esp, -monitor_size);
2045 addi(R26_monitor, R26_monitor, -monitor_size);
2046
2047 // Restart interpreter
2048 }
2049
2050 // ============================================================================
2051 // Java locals access
2052
2053 // Load a local variable at index in Rindex into register Rdst_value.
2054 // Also puts address of local into Rdst_address as a service.
2055 // Kills:
2056 // - Rdst_value
2057 // - Rdst_address
2058 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) {
2059 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2060 subf(Rdst_address, Rdst_address, R18_locals);
2061 lwz(Rdst_value, 0, Rdst_address);
2062 }
2063
2064 // Load a local variable at index in Rindex into register Rdst_value.
2065 // Also puts address of local into Rdst_address as a service.
2066 // Kills:
2067 // - Rdst_value
2068 // - Rdst_address
2069 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) {
2070 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2071 subf(Rdst_address, Rdst_address, R18_locals);
2072 ld(Rdst_value, -8, Rdst_address);
2073 }
2074
2075 // Load a local variable at index in Rindex into register Rdst_value.
2076 // Also puts address of local into Rdst_address as a service.
2077 // Input:
2078 // - Rindex: slot nr of local variable
2079 // Kills:
2080 // - Rdst_value
2081 // - Rdst_address
2082 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value,
2083 Register Rdst_address,
2084 Register Rindex) {
2085 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2086 subf(Rdst_address, Rdst_address, R18_locals);
2087 ld(Rdst_value, 0, Rdst_address);
2088 }
2089
2090 // Load a local variable at index in Rindex into register Rdst_value.
2091 // Also puts address of local into Rdst_address as a service.
2092 // Kills:
2093 // - Rdst_value
2094 // - Rdst_address
2095 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value,
2096 Register Rdst_address,
2097 Register Rindex) {
2098 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2099 subf(Rdst_address, Rdst_address, R18_locals);
2100 lfs(Rdst_value, 0, Rdst_address);
2101 }
2102
2103 // Load a local variable at index in Rindex into register Rdst_value.
2104 // Also puts address of local into Rdst_address as a service.
2105 // Kills:
2106 // - Rdst_value
2107 // - Rdst_address
2108 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value,
2109 Register Rdst_address,
2110 Register Rindex) {
2111 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2112 subf(Rdst_address, Rdst_address, R18_locals);
2113 lfd(Rdst_value, -8, Rdst_address);
2114 }
2115
2116 // Store an int value at local variable slot Rindex.
2117 // Kills:
2118 // - Rindex
2119 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) {
2120 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2121 subf(Rindex, Rindex, R18_locals);
2122 stw(Rvalue, 0, Rindex);
2123 }
2124
2125 // Store a long value at local variable slot Rindex.
2126 // Kills:
2127 // - Rindex
2128 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) {
2129 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2130 subf(Rindex, Rindex, R18_locals);
2131 std(Rvalue, -8, Rindex);
2132 }
2133
2134 // Store an oop value at local variable slot Rindex.
2135 // Kills:
2136 // - Rindex
2137 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) {
2138 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2139 subf(Rindex, Rindex, R18_locals);
2140 std(Rvalue, 0, Rindex);
2141 }
2142
2143 // Store an int value at local variable slot Rindex.
2144 // Kills:
2145 // - Rindex
2146 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) {
2147 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2148 subf(Rindex, Rindex, R18_locals);
2149 stfs(Rvalue, 0, Rindex);
2150 }
2151
2152 // Store an int value at local variable slot Rindex.
2153 // Kills:
2154 // - Rindex
2155 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) {
2156 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2157 subf(Rindex, Rindex, R18_locals);
2158 stfd(Rvalue, -8, Rindex);
2159 }
2160
2161 // Read pending exception from thread and jump to interpreter.
2162 // Throw exception entry if one if pending. Fall through otherwise.
2163 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) {
2164 assert_different_registers(Rscratch1, Rscratch2, R3);
2165 Register Rexception = Rscratch1;
2166 Register Rtmp = Rscratch2;
2167 Label Ldone;
2168 // Get pending exception oop.
2169 ld(Rexception, thread_(pending_exception));
2170 cmpdi(CCR0, Rexception, 0);
2171 beq(CCR0, Ldone);
2172 li(Rtmp, 0);
2173 mr_if_needed(R3, Rexception);
2174 std(Rtmp, thread_(pending_exception)); // Clear exception in thread
2175 if (Interpreter::rethrow_exception_entry() != NULL) {
2176 // Already got entry address.
2177 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry());
2178 } else {
2179 // Dynamically load entry address.
2180 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true);
2181 ld(Rtmp, simm16_rest, Rtmp);
2182 }
2183 mtctr(Rtmp);
2184 save_interpreter_state(Rtmp);
2185 bctr();
2186
2187 align(32, 12);
2188 bind(Ldone);
2189 }
2190
2191 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) {
2192 save_interpreter_state(R11_scratch1);
2193
2194 MacroAssembler::call_VM(oop_result, entry_point, false);
2195
2196 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2197
2198 check_and_handle_popframe(R11_scratch1);
2199 check_and_handle_earlyret(R11_scratch1);
2200 // Now check exceptions manually.
2201 if (check_exceptions) {
2202 check_and_forward_exception(R11_scratch1, R12_scratch2);
2203 }
2204 }
2205
2206 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2207 Register arg_1, bool check_exceptions) {
2208 // ARG1 is reserved for the thread.
2209 mr_if_needed(R4_ARG2, arg_1);
2210 call_VM(oop_result, entry_point, check_exceptions);
2211 }
2212
2213 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2214 Register arg_1, Register arg_2,
2215 bool check_exceptions) {
2216 // ARG1 is reserved for the thread.
2217 mr_if_needed(R4_ARG2, arg_1);
2218 assert(arg_2 != R4_ARG2, "smashed argument");
2219 mr_if_needed(R5_ARG3, arg_2);
2220 call_VM(oop_result, entry_point, check_exceptions);
2221 }
2222
2223 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point,
2224 Register arg_1, Register arg_2, Register arg_3,
2225 bool check_exceptions) {
2226 // ARG1 is reserved for the thread.
2227 mr_if_needed(R4_ARG2, arg_1);
2228 assert(arg_2 != R4_ARG2, "smashed argument");
2229 mr_if_needed(R5_ARG3, arg_2);
2230 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument");
2231 mr_if_needed(R6_ARG4, arg_3);
2232 call_VM(oop_result, entry_point, check_exceptions);
2233 }
2234
2235 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) {
2236 ld(scratch, 0, R1_SP);
2237 std(R15_esp, _ijava_state_neg(esp), scratch);
2238 std(R14_bcp, _ijava_state_neg(bcp), scratch);
2239 std(R26_monitor, _ijava_state_neg(monitors), scratch);
2240 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); }
2241 // Other entries should be unchanged.
2242 }
2243
2244 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) {
2245 ld(scratch, 0, R1_SP);
2246 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception).
2247 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code.
2248 if (!bcp_and_mdx_only) {
2249 // Following ones are Metadata.
2250 ld(R19_method, _ijava_state_neg(method), scratch);
2251 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch);
2252 // Following ones are stack addresses and don't require reload.
2253 ld(R15_esp, _ijava_state_neg(esp), scratch);
2254 ld(R18_locals, _ijava_state_neg(locals), scratch);
2255 ld(R26_monitor, _ijava_state_neg(monitors), scratch);
2256 }
2257 #ifdef ASSERT
2258 {
2259 Label Lok;
2260 subf(R0, R1_SP, scratch);
2261 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size);
2262 bge(CCR0, Lok);
2263 stop("frame too small (restore istate)", 0x5432);
2264 bind(Lok);
2265 }
2266 {
2267 Label Lok;
2268 ld(R0, _ijava_state_neg(ijava_reserved), scratch);
2269 cmpdi(CCR0, R0, 0x5afe);
2270 beq(CCR0, Lok);
2271 stop("frame corrupted (restore istate)", 0x5afe);
2272 bind(Lok);
2273 }
2274 #endif
2275 }
2276
2277 void InterpreterMacroAssembler::get_method_counters(Register method,
2278 Register Rcounters,
2279 Label& skip) {
2280 BLOCK_COMMENT("Load and ev. allocate counter object {");
2281 Label has_counters;
2282 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2283 cmpdi(CCR0, Rcounters, 0);
2284 bne(CCR0, has_counters);
2285 call_VM(noreg, CAST_FROM_FN_PTR(address,
2286 InterpreterRuntime::build_method_counters), method, false);
2287 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2288 cmpdi(CCR0, Rcounters, 0);
2289 beq(CCR0, skip); // No MethodCounters, OutOfMemory.
2290 BLOCK_COMMENT("} Load and ev. allocate counter object");
2291
2292 bind(has_counters);
2293 }
2294
2295 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters,
2296 Register iv_be_count,
2297 Register Rtmp_r0) {
2298 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful");
2299 Register invocation_count = iv_be_count;
2300 Register backedge_count = Rtmp_r0;
2301 int delta = InvocationCounter::count_increment;
2302
2303 // Load each counter in a register.
2304 // ld(inv_counter, Rtmp);
2305 // ld(be_counter, Rtmp2);
2306 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() +
2307 InvocationCounter::counter_offset());
2308 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() +
2309 InvocationCounter::counter_offset());
2310
2311 BLOCK_COMMENT("Increment profiling counters {");
2312
2313 // Load the backedge counter.
2314 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int
2315 // Mask the backedge counter.
2316 andi(backedge_count, backedge_count, InvocationCounter::count_mask_value);
2317
2318 // Load the invocation counter.
2319 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int
2320 // Add the delta to the invocation counter and store the result.
2321 addi(invocation_count, invocation_count, delta);
2322 // Store value.
2323 stw(invocation_count, inv_counter_offset, Rcounters);
2324
2325 // Add invocation counter + backedge counter.
2326 add(iv_be_count, backedge_count, invocation_count);
2327
2328 // Note that this macro must leave the backedge_count + invocation_count in
2329 // register iv_be_count!
2330 BLOCK_COMMENT("} Increment profiling counters");
2331 }
2332
2333 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2334 if (state == atos) { MacroAssembler::verify_oop(reg); }
2335 }
2336
2337 // Local helper function for the verify_oop_or_return_address macro.
2338 static bool verify_return_address(Method* m, int bci) {
2339 #ifndef PRODUCT
2340 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci;
2341 // Assume it is a valid return address if it is inside m and is preceded by a jsr.
2342 if (!m->contains(pc)) return false;
2343 address jsr_pc;
2344 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2345 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2346 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2347 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2348 #endif // PRODUCT
2349 return false;
2350 }
2351
2352 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2353 if (VerifyFPU) {
2354 unimplemented("verfiyFPU");
2355 }
2356 }
2357
2358 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2359 if (!VerifyOops) return;
2360
2361 // The VM documentation for the astore[_wide] bytecode allows
2362 // the TOS to be not only an oop but also a return address.
2363 Label test;
2364 Label skip;
2365 // See if it is an address (in the current method):
2366
2367 const int log2_bytecode_size_limit = 16;
2368 srdi_(Rtmp, reg, log2_bytecode_size_limit);
2369 bne(CCR0, test);
2370
2371 address fd = CAST_FROM_FN_PTR(address, verify_return_address);
2372 const int nbytes_save = MacroAssembler::num_volatile_regs * 8;
2373 save_volatile_gprs(R1_SP, -nbytes_save); // except R0
2374 save_LR_CR(Rtmp); // Save in old frame.
2375 push_frame_reg_args(nbytes_save, Rtmp);
2376
2377 load_const_optimized(Rtmp, fd, R0);
2378 mr_if_needed(R4_ARG2, reg);
2379 mr(R3_ARG1, R19_method);
2380 call_c(Rtmp); // call C
2381
2382 pop_frame();
2383 restore_LR_CR(Rtmp);
2384 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
2385 b(skip);
2386
2387 // Perform a more elaborate out-of-line call.
2388 // Not an address; verify it:
2389 bind(test);
2390 verify_oop(reg);
2391 bind(skip);
2392 }
2393
2394 // Inline assembly for:
2395 //
2396 // if (thread is in interp_only_mode) {
2397 // InterpreterRuntime::post_method_entry();
2398 // }
2399 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) ||
2400 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) {
2401 // SharedRuntime::jvmpi_method_entry(method, receiver);
2402 // }
2403 void InterpreterMacroAssembler::notify_method_entry() {
2404 // JVMTI
2405 // Whenever JVMTI puts a thread in interp_only_mode, method
2406 // entry/exit events are sent for that thread to track stack
2407 // depth. If it is possible to enter interp_only_mode we add
2408 // the code to check if the event should be sent.
2409 if (JvmtiExport::can_post_interpreter_events()) {
2410 Label jvmti_post_done;
2411
2412 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2413 cmpwi(CCR0, R0, 0);
2414 beq(CCR0, jvmti_post_done);
2415 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry),
2416 /*check_exceptions=*/true);
2417
2418 bind(jvmti_post_done);
2419 }
2420 }
2421
2422 // Inline assembly for:
2423 //
2424 // if (thread is in interp_only_mode) {
2425 // // save result
2426 // InterpreterRuntime::post_method_exit();
2427 // // restore result
2428 // }
2429 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) {
2430 // // save result
2431 // SharedRuntime::jvmpi_method_exit();
2432 // // restore result
2433 // }
2434 //
2435 // Native methods have their result stored in d_tmp and l_tmp.
2436 // Java methods have their result stored in the expression stack.
2437 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state,
2438 NotifyMethodExitMode mode, bool check_exceptions) {
2439 // JVMTI
2440 // Whenever JVMTI puts a thread in interp_only_mode, method
2441 // entry/exit events are sent for that thread to track stack
2442 // depth. If it is possible to enter interp_only_mode we add
2443 // the code to check if the event should be sent.
2444 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2445 Label jvmti_post_done;
2446
2447 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2448 cmpwi(CCR0, R0, 0);
2449 beq(CCR0, jvmti_post_done);
2450 if (!is_native_method) { push(state); } // Expose tos to GC.
2451 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit),
2452 /*check_exceptions=*/check_exceptions);
2453 if (!is_native_method) { pop(state); }
2454
2455 align(32, 12);
2456 bind(jvmti_post_done);
2457 }
2458
2459 // Dtrace support not implemented.
2460 }
--- EOF ---