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