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