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