1 /*
2 * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2012, 2014 SAP AG. 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_64.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "prims/jvmtiThreadState.hpp"
32
33 #ifdef PRODUCT
34 #define BLOCK_COMMENT(str) // nothing
35 #else
36 #define BLOCK_COMMENT(str) block_comment(str)
37 #endif
38
39 void InterpreterMacroAssembler::null_check_throw(Register a, int offset, Register temp_reg) {
40 #ifdef CC_INTERP
41 address exception_entry = StubRoutines::throw_NullPointerException_at_call_entry();
42 #else
43 address exception_entry = Interpreter::throw_NullPointerException_entry();
44 #endif
45 MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry);
46 }
47
48 void InterpreterMacroAssembler::branch_to_entry(address entry, Register Rscratch) {
49 assert(entry, "Entry must have been generated by now");
50 if (is_within_range_of_b(entry, pc())) {
51 b(entry);
52 } else {
53 load_const_optimized(Rscratch, entry, R0);
54 mtctr(Rscratch);
55 bctr();
56 }
57 }
58
59 #ifndef CC_INTERP
60
61 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) {
62 Register bytecode = R12_scratch2;
63 if (bcp_incr != 0) {
64 lbzu(bytecode, bcp_incr, R14_bcp);
65 } else {
66 lbz(bytecode, 0, R14_bcp);
67 }
68
69 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state));
70 }
71
72 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
73 // Load current bytecode.
74 Register bytecode = R12_scratch2;
75 lbz(bytecode, 0, R14_bcp);
76 dispatch_Lbyte_code(state, bytecode, table);
77 }
78
79 // Dispatch code executed in the prolog of a bytecode which does not do it's
80 // own dispatch. The dispatch address is computed and placed in R24_dispatch_addr.
81 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
82 Register bytecode = R12_scratch2;
83 lbz(bytecode, bcp_incr, R14_bcp);
84
85 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state));
86
87 sldi(bytecode, bytecode, LogBytesPerWord);
88 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode);
89 }
90
91 // Dispatch code executed in the epilog of a bytecode which does not do it's
92 // own dispatch. The dispatch address in R24_dispatch_addr is used for the
93 // dispatch.
94 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
95 mtctr(R24_dispatch_addr);
96 addi(R14_bcp, R14_bcp, bcp_incr);
97 bctr();
98 }
99
100 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
101 assert(scratch_reg != R0, "can't use R0 as scratch_reg here");
102 if (JvmtiExport::can_pop_frame()) {
103 Label L;
104
105 // Check the "pending popframe condition" flag in the current thread.
106 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
107
108 // Initiate popframe handling only if it is not already being
109 // processed. If the flag has the popframe_processing bit set, it
110 // means that this code is called *during* popframe handling - we
111 // don't want to reenter.
112 andi_(R0, scratch_reg, JavaThread::popframe_pending_bit);
113 beq(CCR0, L);
114
115 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit);
116 bne(CCR0, L);
117
118 // Call the Interpreter::remove_activation_preserving_args_entry()
119 // func to get the address of the same-named entrypoint in the
120 // generated interpreter code.
121 #if defined(ABI_ELFv2)
122 call_c(CAST_FROM_FN_PTR(address,
123 Interpreter::remove_activation_preserving_args_entry),
124 relocInfo::none);
125 #else
126 call_c(CAST_FROM_FN_PTR(FunctionDescriptor*,
127 Interpreter::remove_activation_preserving_args_entry),
128 relocInfo::none);
129 #endif
130
131 // Jump to Interpreter::_remove_activation_preserving_args_entry.
132 mtctr(R3_RET);
133 bctr();
134
135 align(32, 12);
136 bind(L);
137 }
138 }
139
140 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
141 const Register Rthr_state_addr = scratch_reg;
142 if (JvmtiExport::can_force_early_return()) {
143 Label Lno_early_ret;
144 ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
145 cmpdi(CCR0, Rthr_state_addr, 0);
146 beq(CCR0, Lno_early_ret);
147
148 lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr);
149 cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending);
150 bne(CCR0, Lno_early_ret);
151
152 // Jump to Interpreter::_earlyret_entry.
153 lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr);
154 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry));
155 mtlr(R3_RET);
156 blr();
157
158 align(32, 12);
159 bind(Lno_early_ret);
160 }
161 }
162
163 void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) {
164 const Register RjvmtiState = Rscratch1;
165 const Register Rscratch2 = R0;
166
167 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
168 li(Rscratch2, 0);
169
170 switch (state) {
171 case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
172 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState);
173 break;
174 case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
175 break;
176 case btos: // fall through
177 case ztos: // fall through
178 case ctos: // fall through
179 case stos: // fall through
180 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
181 break;
182 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
183 break;
184 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
185 break;
186 case vtos: break;
187 default : ShouldNotReachHere();
188 }
189
190 // Clean up tos value in the jvmti thread state.
191 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
192 // Set tos state field to illegal value.
193 li(Rscratch2, ilgl);
194 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState);
195 }
196
197 // Common code to dispatch and dispatch_only.
198 // Dispatch value in Lbyte_code and increment Lbcp.
199
200 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) {
201 address table_base = (address)Interpreter::dispatch_table((TosState)0);
202 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base;
203 if (is_simm16(table_offs)) {
204 addi(dst, R25_templateTableBase, (int)table_offs);
205 } else {
206 load_const_optimized(dst, table, R0);
207 }
208 }
209
210 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode, address* table, bool verify) {
211 if (verify) {
212 unimplemented("dispatch_Lbyte_code: verify"); // See Sparc Implementation to implement this
213 }
214
215 #ifdef FAST_DISPATCH
216 unimplemented("dispatch_Lbyte_code FAST_DISPATCH");
217 #else
218 assert_different_registers(bytecode, R11_scratch1);
219
220 // Calc dispatch table address.
221 load_dispatch_table(R11_scratch1, table);
222
223 sldi(R12_scratch2, bytecode, LogBytesPerWord);
224 ldx(R11_scratch1, R11_scratch1, R12_scratch2);
225
226 // Jump off!
227 mtctr(R11_scratch1);
228 bctr();
229 #endif
230 }
231
232 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) {
233 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize);
234 ldx(Rrecv_dst, Rrecv_dst, R15_esp);
235 }
236
237 // helpers for expression stack
238
239 void InterpreterMacroAssembler::pop_i(Register r) {
240 lwzu(r, Interpreter::stackElementSize, R15_esp);
241 }
242
243 void InterpreterMacroAssembler::pop_ptr(Register r) {
244 ldu(r, Interpreter::stackElementSize, R15_esp);
245 }
246
247 void InterpreterMacroAssembler::pop_l(Register r) {
248 ld(r, Interpreter::stackElementSize, R15_esp);
249 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
250 }
251
252 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
253 lfsu(f, Interpreter::stackElementSize, R15_esp);
254 }
255
256 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
257 lfd(f, Interpreter::stackElementSize, R15_esp);
258 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
259 }
260
261 void InterpreterMacroAssembler::push_i(Register r) {
262 stw(r, 0, R15_esp);
263 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
264 }
265
266 void InterpreterMacroAssembler::push_ptr(Register r) {
267 std(r, 0, R15_esp);
268 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
269 }
270
271 void InterpreterMacroAssembler::push_l(Register r) {
272 std(r, - Interpreter::stackElementSize, R15_esp);
273 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
274 }
275
276 void InterpreterMacroAssembler::push_f(FloatRegister f) {
277 stfs(f, 0, R15_esp);
278 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
279 }
280
281 void InterpreterMacroAssembler::push_d(FloatRegister f) {
282 stfd(f, - Interpreter::stackElementSize, R15_esp);
283 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
284 }
285
286 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) {
287 std(first, 0, R15_esp);
288 std(second, -Interpreter::stackElementSize, R15_esp);
289 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
290 }
291
292 void InterpreterMacroAssembler::push_l_pop_d(Register l, FloatRegister d) {
293 std(l, 0, R15_esp);
294 lfd(d, 0, R15_esp);
295 }
296
297 void InterpreterMacroAssembler::push_d_pop_l(FloatRegister d, Register l) {
298 stfd(d, 0, R15_esp);
299 ld(l, 0, R15_esp);
300 }
301
302 void InterpreterMacroAssembler::push(TosState state) {
303 switch (state) {
304 case atos: push_ptr(); break;
305 case btos:
306 case ztos:
307 case ctos:
308 case stos:
309 case itos: push_i(); break;
310 case ltos: push_l(); break;
311 case ftos: push_f(); break;
312 case dtos: push_d(); break;
313 case vtos: /* nothing to do */ break;
314 default : ShouldNotReachHere();
315 }
316 }
317
318 void InterpreterMacroAssembler::pop(TosState state) {
319 switch (state) {
320 case atos: pop_ptr(); break;
321 case btos:
322 case ztos:
323 case ctos:
324 case stos:
325 case itos: pop_i(); break;
326 case ltos: pop_l(); break;
327 case ftos: pop_f(); break;
328 case dtos: pop_d(); break;
329 case vtos: /* nothing to do */ break;
330 default : ShouldNotReachHere();
331 }
332 verify_oop(R17_tos, state);
333 }
334
335 void InterpreterMacroAssembler::empty_expression_stack() {
336 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
337 }
338
339 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset,
340 Register Rdst,
341 signedOrNot is_signed) {
342 #if defined(VM_LITTLE_ENDIAN)
343 if (bcp_offset) {
344 load_const_optimized(Rdst, bcp_offset);
345 lhbrx(Rdst, R14_bcp, Rdst);
346 } else {
347 lhbrx(Rdst, R14_bcp);
348 }
349 if (is_signed == Signed) {
350 extsh(Rdst, Rdst);
351 }
352 #else
353 // Read Java big endian format.
354 if (is_signed == Signed) {
355 lha(Rdst, bcp_offset, R14_bcp);
356 } else {
357 lhz(Rdst, bcp_offset, R14_bcp);
358 }
359 #endif
360 }
361
362 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset,
363 Register Rdst,
364 signedOrNot is_signed) {
365 #if defined(VM_LITTLE_ENDIAN)
366 if (bcp_offset) {
367 load_const_optimized(Rdst, bcp_offset);
368 lwbrx(Rdst, R14_bcp, Rdst);
369 } else {
370 lwbrx(Rdst, R14_bcp);
371 }
372 if (is_signed == Signed) {
373 extsw(Rdst, Rdst);
374 }
375 #else
376 // Read Java big endian format.
377 if (bcp_offset & 3) { // Offset unaligned?
378 load_const_optimized(Rdst, bcp_offset);
379 if (is_signed == Signed) {
380 lwax(Rdst, R14_bcp, Rdst);
381 } else {
382 lwzx(Rdst, R14_bcp, Rdst);
383 }
384 } else {
385 if (is_signed == Signed) {
386 lwa(Rdst, bcp_offset, R14_bcp);
387 } else {
388 lwz(Rdst, bcp_offset, R14_bcp);
389 }
390 }
391 #endif
392 }
393
394
395 // Load the constant pool cache index from the bytecode stream.
396 //
397 // Kills / writes:
398 // - Rdst, Rscratch
399 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, size_t index_size) {
400 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
401 // Cache index is always in the native format, courtesy of Rewriter.
402 if (index_size == sizeof(u2)) {
403 lhz(Rdst, bcp_offset, R14_bcp);
404 } else if (index_size == sizeof(u4)) {
405 assert(EnableInvokeDynamic, "giant index used only for JSR 292");
406 if (bcp_offset & 3) {
407 load_const_optimized(Rdst, bcp_offset);
408 lwax(Rdst, R14_bcp, Rdst);
409 } else {
410 lwa(Rdst, bcp_offset, R14_bcp);
411 }
412 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
413 nand(Rdst, Rdst, Rdst); // convert to plain index
414 } else if (index_size == sizeof(u1)) {
415 lbz(Rdst, bcp_offset, R14_bcp);
416 } else {
417 ShouldNotReachHere();
418 }
419 // Rdst now contains cp cache index.
420 }
421
422 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset, size_t index_size) {
423 get_cache_index_at_bcp(cache, bcp_offset, index_size);
424 sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord));
425 add(cache, R27_constPoolCache, cache);
426 }
427
428 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format)
429 // from (Rsrc)+offset.
430 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset,
431 signedOrNot is_signed) {
432 #if defined(VM_LITTLE_ENDIAN)
433 if (offset) {
434 load_const_optimized(Rdst, offset);
435 lwbrx(Rdst, Rdst, Rsrc);
436 } else {
437 lwbrx(Rdst, Rsrc);
438 }
439 if (is_signed == Signed) {
440 extsw(Rdst, Rdst);
441 }
442 #else
443 if (is_signed == Signed) {
444 lwa(Rdst, offset, Rsrc);
445 } else {
446 lwz(Rdst, offset, Rsrc);
447 }
448 #endif
449 }
450
451 // Load object from cpool->resolved_references(index).
452 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) {
453 assert_different_registers(result, index);
454 get_constant_pool(result);
455
456 // Convert from field index to resolved_references() index and from
457 // word index to byte offset. Since this is a java object, it can be compressed.
458 Register tmp = index; // reuse
459 sldi(tmp, index, LogBytesPerHeapOop);
460 // Load pointer for resolved_references[] objArray.
461 ld(result, ConstantPool::resolved_references_offset_in_bytes(), result);
462 // JNIHandles::resolve(result)
463 ld(result, 0, result);
464 #ifdef ASSERT
465 Label index_ok;
466 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result);
467 sldi(R0, R0, LogBytesPerHeapOop);
468 cmpd(CCR0, tmp, R0);
469 blt(CCR0, index_ok);
470 stop("resolved reference index out of bounds", 0x09256);
471 bind(index_ok);
472 #endif
473 // Add in the index.
474 add(result, tmp, result);
475 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result);
476 }
477
478 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
479 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2.
480 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1,
481 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) {
482 // Profile the not-null value's klass.
483 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2);
484 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
485 profile_typecheck_failed(Rtmp1, Rtmp2);
486 }
487
488 void InterpreterMacroAssembler::generate_stack_overflow_check_with_compare_and_throw(Register Rmem_frame_size, Register Rscratch1) {
489 Label done;
490 sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
491 ld(Rscratch1, thread_(stack_overflow_limit));
492 cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
493 bgt(CCR0/*is_stack_overflow*/, done);
494
495 // Load target address of the runtime stub.
496 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order");
497 load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0);
498 mtctr(Rscratch1);
499 // Restore caller_sp.
500 #ifdef ASSERT
501 ld(Rscratch1, 0, R1_SP);
502 ld(R0, 0, R21_sender_SP);
503 cmpd(CCR0, R0, Rscratch1);
504 asm_assert_eq("backlink", 0x547);
505 #endif // ASSERT
506 mr(R1_SP, R21_sender_SP);
507 bctr();
508
509 align(32, 12);
510 bind(done);
511 }
512
513 // Separate these two to allow for delay slot in middle.
514 // These are used to do a test and full jump to exception-throwing code.
515
516 // Check that index is in range for array, then shift index by index_shift,
517 // and put arrayOop + shifted_index into res.
518 // Note: res is still shy of address by array offset into object.
519
520 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, 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(R17_tos, Rindex);
553 mr(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, int index_shift, Register tmp, Register res) {
573 // pop array
574 pop_ptr(array);
575
576 // check array
577 index_check_without_pop(array, index, index_shift, tmp, res);
578 }
579
580 void InterpreterMacroAssembler::get_const(Register Rdst) {
581 ld(Rdst, in_bytes(Method::const_offset()), R19_method);
582 }
583
584 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
585 get_const(Rdst);
586 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
587 }
588
589 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
590 get_constant_pool(Rdst);
591 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst);
592 }
593
594 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
595 get_constant_pool(Rcpool);
596 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool);
597 }
598
599 // Unlock if synchronized method.
600 //
601 // Unlock the receiver if this is a synchronized method.
602 // Unlock any Java monitors from synchronized blocks.
603 //
604 // If there are locked Java monitors
605 // If throw_monitor_exception
606 // throws IllegalMonitorStateException
607 // Else if install_monitor_exception
608 // installs IllegalMonitorStateException
609 // Else
610 // no error processing
611 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
612 bool throw_monitor_exception,
613 bool install_monitor_exception) {
614 Label Lunlocked, Lno_unlock;
615 {
616 Register Rdo_not_unlock_flag = R11_scratch1;
617 Register Raccess_flags = R12_scratch2;
618
619 // Check if synchronized method or unlocking prevented by
620 // JavaThread::do_not_unlock_if_synchronized flag.
621 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
622 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method);
623 li(R0, 0);
624 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag
625
626 push(state);
627
628 // Skip if we don't have to unlock.
629 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0.
630 beq(CCR0, Lunlocked);
631
632 cmpwi(CCR0, Rdo_not_unlock_flag, 0);
633 bne(CCR0, Lno_unlock);
634 }
635
636 // Unlock
637 {
638 Register Rmonitor_base = R11_scratch1;
639
640 Label Lunlock;
641 // If it's still locked, everything is ok, unlock it.
642 ld(Rmonitor_base, 0, R1_SP);
643 addi(Rmonitor_base, Rmonitor_base, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
644
645 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base);
646 cmpdi(CCR0, R0, 0);
647 bne(CCR0, Lunlock);
648
649 // If it's already unlocked, throw exception.
650 if (throw_monitor_exception) {
651 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
652 should_not_reach_here();
653 } else {
654 if (install_monitor_exception) {
655 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
656 b(Lunlocked);
657 }
658 }
659
660 bind(Lunlock);
661 unlock_object(Rmonitor_base);
662 }
663
664 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not.
665 bind(Lunlocked);
666 {
667 Label Lexception, Lrestart;
668 Register Rcurrent_obj_addr = R11_scratch1;
669 const int delta = frame::interpreter_frame_monitor_size_in_bytes();
670 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords");
671
672 bind(Lrestart);
673 // Set up search loop: Calc num of iterations.
674 {
675 Register Riterations = R12_scratch2;
676 Register Rmonitor_base = Rcurrent_obj_addr;
677 ld(Rmonitor_base, 0, R1_SP);
678 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base
679
680 subf_(Riterations, R26_monitor, Rmonitor_base);
681 ble(CCR0, Lno_unlock);
682
683 addi(Rcurrent_obj_addr, Rmonitor_base, BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes());
684 // Check if any monitor is on stack, bail out if not
685 srdi(Riterations, Riterations, exact_log2(delta));
686 mtctr(Riterations);
687 }
688
689 // The search loop: Look for locked monitors.
690 {
691 const Register Rcurrent_obj = R0;
692 Label Lloop;
693
694 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
695 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
696 bind(Lloop);
697
698 // Check if current entry is used.
699 cmpdi(CCR0, Rcurrent_obj, 0);
700 bne(CCR0, Lexception);
701 // Preload next iteration's compare value.
702 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
703 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
704 bdnz(Lloop);
705 }
706 // Fell through: Everything's unlocked => finish.
707 b(Lno_unlock);
708
709 // An object is still locked => need to throw exception.
710 bind(Lexception);
711 if (throw_monitor_exception) {
712 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
713 should_not_reach_here();
714 } else {
715 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
716 // Unlock does not block, so don't have to worry about the frame.
717 Register Rmonitor_addr = R11_scratch1;
718 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta);
719 unlock_object(Rmonitor_addr);
720 if (install_monitor_exception) {
721 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
722 }
723 b(Lrestart);
724 }
725 }
726
727 align(32, 12);
728 bind(Lno_unlock);
729 pop(state);
730 }
731
732 // Support function for remove_activation & Co.
733 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, Register Rscratch1, Register Rscratch2) {
734 // Pop interpreter frame.
735 ld(Rscratch1, 0, R1_SP); // *SP
736 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp
737 ld(Rscratch2, 0, Rscratch1); // **SP
738 #ifdef ASSERT
739 {
740 Label Lok;
741 ld(R0, _ijava_state_neg(ijava_reserved), Rscratch1);
742 cmpdi(CCR0, R0, 0x5afe);
743 beq(CCR0, Lok);
744 stop("frame corrupted (remove activation)", 0x5afe);
745 bind(Lok);
746 }
747 #endif
748 if (return_pc!=noreg) {
749 ld(return_pc, _abi(lr), Rscratch1); // LR
750 }
751
752 // Merge top frames.
753 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP
754 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP
755 }
756
757 void InterpreterMacroAssembler::narrow(Register result) {
758 Register ret_type = R11_scratch1;
759 ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
760 lbz(ret_type, in_bytes(ConstMethod::result_type_offset()), R11_scratch1);
761
762 Label notBool, notByte, notChar, done;
763
764 // common case first
765 cmpwi(CCR0, ret_type, T_INT);
766 beq(CCR0, done);
767
768 cmpwi(CCR0, ret_type, T_BOOLEAN);
769 bne(CCR0, notBool);
770 andi(result, result, 0x1);
771 b(done);
772
773 bind(notBool);
774 cmpwi(CCR0, ret_type, T_BYTE);
775 bne(CCR0, notByte);
776 extsb(result, result);
777 b(done);
778
779 bind(notByte);
780 cmpwi(CCR0, ret_type, T_CHAR);
781 bne(CCR0, notChar);
782 andi(result, result, 0xffff);
783 b(done);
784
785 bind(notChar);
786 // cmpwi(CCR0, ret_type, T_SHORT); // all that's left
787 // bne(CCR0, done);
788 extsh(result, result);
789
790 // Nothing to do for T_INT
791 bind(done);
792 }
793
794 // Remove activation.
795 //
796 // Unlock the receiver if this is a synchronized method.
797 // Unlock any Java monitors from synchronized blocks.
798 // Remove the activation from the stack.
799 //
800 // If there are locked Java monitors
801 // If throw_monitor_exception
802 // throws IllegalMonitorStateException
803 // Else if install_monitor_exception
804 // installs IllegalMonitorStateException
805 // Else
806 // no error processing
807 void InterpreterMacroAssembler::remove_activation(TosState state,
808 bool throw_monitor_exception,
809 bool install_monitor_exception) {
810 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
811
812 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
813 notify_method_exit(false, state, NotifyJVMTI, true);
814
815 verify_oop(R17_tos, state);
816 verify_thread();
817
818 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
819 mtlr(R0);
820 }
821
822 #endif // !CC_INTERP
823
824 // Lock object
825 //
826 // Registers alive
827 // monitor - Address of the BasicObjectLock to be used for locking,
828 // which must be initialized with the object to lock.
829 // object - Address of the object to be locked.
830 //
831 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
832 if (UseHeavyMonitors) {
833 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
834 monitor, /*check_for_exceptions=*/true CC_INTERP_ONLY(&& false));
835 } else {
836 // template code:
837 //
838 // markOop displaced_header = obj->mark().set_unlocked();
839 // monitor->lock()->set_displaced_header(displaced_header);
840 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
841 // // We stored the monitor address into the object's mark word.
842 // } else if (THREAD->is_lock_owned((address)displaced_header))
843 // // Simple recursive case.
844 // monitor->lock()->set_displaced_header(NULL);
845 // } else {
846 // // Slow path.
847 // InterpreterRuntime::monitorenter(THREAD, monitor);
848 // }
849
850 const Register displaced_header = R7_ARG5;
851 const Register object_mark_addr = R8_ARG6;
852 const Register current_header = R9_ARG7;
853 const Register tmp = R10_ARG8;
854
855 Label done;
856 Label cas_failed, slow_case;
857
858 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp);
859
860 // markOop displaced_header = obj->mark().set_unlocked();
861
862 // Load markOop from object into displaced_header.
863 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object);
864
865 if (UseBiasedLocking) {
866 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case);
867 }
868
869 // Set displaced_header to be (markOop of object | UNLOCK_VALUE).
870 ori(displaced_header, displaced_header, markOopDesc::unlocked_value);
871
872 // monitor->lock()->set_displaced_header(displaced_header);
873
874 // Initialize the box (Must happen before we update the object mark!).
875 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() +
876 BasicLock::displaced_header_offset_in_bytes(), monitor);
877
878 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
879
880 // Store stack address of the BasicObjectLock (this is monitor) into object.
881 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
882
883 // Must fence, otherwise, preceding store(s) may float below cmpxchg.
884 // CmpxchgX sets CCR0 to cmpX(current, displaced).
885 fence(); // TODO: replace by MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq ?
886 cmpxchgd(/*flag=*/CCR0,
887 /*current_value=*/current_header,
888 /*compare_value=*/displaced_header, /*exchange_value=*/monitor,
889 /*where=*/object_mark_addr,
890 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq,
891 MacroAssembler::cmpxchgx_hint_acquire_lock(),
892 noreg,
893 &cas_failed);
894
895 // If the compare-and-exchange succeeded, then we found an unlocked
896 // object and we have now locked it.
897 b(done);
898 bind(cas_failed);
899
900 // } else if (THREAD->is_lock_owned((address)displaced_header))
901 // // Simple recursive case.
902 // monitor->lock()->set_displaced_header(NULL);
903
904 // We did not see an unlocked object so try the fast recursive case.
905
906 // Check if owner is self by comparing the value in the markOop of object
907 // (current_header) with the stack pointer.
908 sub(current_header, current_header, R1_SP);
909
910 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
911 load_const_optimized(tmp, ~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place);
912
913 and_(R0/*==0?*/, current_header, tmp);
914 // If condition is true we are done and hence we can store 0 in the displaced
915 // header indicating it is a recursive lock.
916 bne(CCR0, slow_case);
917 release();
918 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() +
919 BasicLock::displaced_header_offset_in_bytes(), monitor);
920 b(done);
921
922 // } else {
923 // // Slow path.
924 // InterpreterRuntime::monitorenter(THREAD, monitor);
925
926 // None of the above fast optimizations worked so we have to get into the
927 // slow case of monitor enter.
928 bind(slow_case);
929 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
930 monitor, /*check_for_exceptions=*/true CC_INTERP_ONLY(&& false));
931 // }
932 align(32, 12);
933 bind(done);
934 }
935 }
936
937 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
938 //
939 // Registers alive
940 // monitor - Address of the BasicObjectLock to be used for locking,
941 // which must be initialized with the object to lock.
942 //
943 // Throw IllegalMonitorException if object is not locked by current thread.
944 void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) {
945 if (UseHeavyMonitors) {
946 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
947 monitor, check_for_exceptions CC_INTERP_ONLY(&& false));
948 } else {
949
950 // template code:
951 //
952 // if ((displaced_header = monitor->displaced_header()) == NULL) {
953 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL.
954 // monitor->set_obj(NULL);
955 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) {
956 // // We swapped the unlocked mark in displaced_header into the object's mark word.
957 // monitor->set_obj(NULL);
958 // } else {
959 // // Slow path.
960 // InterpreterRuntime::monitorexit(THREAD, monitor);
961 // }
962
963 const Register object = R7_ARG5;
964 const Register displaced_header = R8_ARG6;
965 const Register object_mark_addr = R9_ARG7;
966 const Register current_header = R10_ARG8;
967
968 Label free_slot;
969 Label slow_case;
970
971 assert_different_registers(object, displaced_header, object_mark_addr, current_header);
972
973 if (UseBiasedLocking) {
974 // The object address from the monitor is in object.
975 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor);
976 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0");
977 biased_locking_exit(CCR0, object, displaced_header, free_slot);
978 }
979
980 // Test first if we are in the fast recursive case.
981 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() +
982 BasicLock::displaced_header_offset_in_bytes(), monitor);
983
984 // If the displaced header is zero, we have a recursive unlock.
985 cmpdi(CCR0, displaced_header, 0);
986 beq(CCR0, free_slot); // recursive unlock
987
988 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) {
989 // // We swapped the unlocked mark in displaced_header into the object's mark word.
990 // monitor->set_obj(NULL);
991
992 // If we still have a lightweight lock, unlock the object and be done.
993
994 // The object address from the monitor is in object.
995 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); }
996 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
997
998 // We have the displaced header in displaced_header. If the lock is still
999 // lightweight, it will contain the monitor address and we'll store the
1000 // displaced header back into the object's mark word.
1001 // CmpxchgX sets CCR0 to cmpX(current, monitor).
1002 cmpxchgd(/*flag=*/CCR0,
1003 /*current_value=*/current_header,
1004 /*compare_value=*/monitor, /*exchange_value=*/displaced_header,
1005 /*where=*/object_mark_addr,
1006 MacroAssembler::MemBarRel,
1007 MacroAssembler::cmpxchgx_hint_release_lock(),
1008 noreg,
1009 &slow_case);
1010 b(free_slot);
1011
1012 // } else {
1013 // // Slow path.
1014 // InterpreterRuntime::monitorexit(THREAD, monitor);
1015
1016 // The lock has been converted into a heavy lock and hence
1017 // we need to get into the slow case.
1018 bind(slow_case);
1019 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1020 monitor, check_for_exceptions CC_INTERP_ONLY(&& false));
1021 // }
1022
1023 Label done;
1024 b(done); // Monitor register may be overwritten! Runtime has already freed the slot.
1025
1026 // Exchange worked, do monitor->set_obj(NULL);
1027 align(32, 12);
1028 bind(free_slot);
1029 li(R0, 0);
1030 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor);
1031 bind(done);
1032 }
1033 }
1034
1035 #ifndef CC_INTERP
1036
1037 // Load compiled (i2c) or interpreter entry when calling from interpreted and
1038 // do the call. Centralized so that all interpreter calls will do the same actions.
1039 // If jvmti single stepping is on for a thread we must not call compiled code.
1040 //
1041 // Input:
1042 // - Rtarget_method: method to call
1043 // - Rret_addr: return address
1044 // - 2 scratch regs
1045 //
1046 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, Register Rscratch1, Register Rscratch2) {
1047 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr);
1048 // Assume we want to go compiled if available.
1049 const Register Rtarget_addr = Rscratch1;
1050 const Register Rinterp_only = Rscratch2;
1051
1052 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method);
1053
1054 if (JvmtiExport::can_post_interpreter_events()) {
1055 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
1056
1057 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
1058 // compiled code in threads for which the event is enabled. Check here for
1059 // interp_only_mode if these events CAN be enabled.
1060 Label done;
1061 verify_thread();
1062 cmpwi(CCR0, Rinterp_only, 0);
1063 beq(CCR0, done);
1064 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method);
1065 align(32, 12);
1066 bind(done);
1067 }
1068
1069 #ifdef ASSERT
1070 {
1071 Label Lok;
1072 cmpdi(CCR0, Rtarget_addr, 0);
1073 bne(CCR0, Lok);
1074 stop("null entry point");
1075 bind(Lok);
1076 }
1077 #endif // ASSERT
1078
1079 mr(R21_sender_SP, R1_SP);
1080
1081 // Calc a precise SP for the call. The SP value we calculated in
1082 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space
1083 // if esp is not max. Also, the i2c adapter extends the stack space without restoring
1084 // our pre-calced value, so repeating calls via i2c would result in stack overflow.
1085 // Since esp already points to an empty slot, we just have to sub 1 additional slot
1086 // to meet the abi scratch requirements.
1087 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in
1088 // the return entry of the interpreter.
1089 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size);
1090 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address
1091 resize_frame_absolute(Rscratch2, Rscratch2, R0);
1092
1093 mr_if_needed(R19_method, Rtarget_method);
1094 mtctr(Rtarget_addr);
1095 mtlr(Rret_addr);
1096
1097 save_interpreter_state(Rscratch2);
1098 #ifdef ASSERT
1099 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp
1100 cmpd(CCR0, R21_sender_SP, Rscratch1);
1101 asm_assert_eq("top_frame_sp incorrect", 0x951);
1102 #endif
1103
1104 bctr();
1105 }
1106
1107 // Set the method data pointer for the current bcp.
1108 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1109 assert(ProfileInterpreter, "must be profiling interpreter");
1110 Label get_continue;
1111 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method);
1112 test_method_data_pointer(get_continue);
1113 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp);
1114
1115 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1116 add(R28_mdx, R28_mdx, R3_RET);
1117 bind(get_continue);
1118 }
1119
1120 // Test ImethodDataPtr. If it is null, continue at the specified label.
1121 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1122 assert(ProfileInterpreter, "must be profiling interpreter");
1123 cmpdi(CCR0, R28_mdx, 0);
1124 beq(CCR0, zero_continue);
1125 }
1126
1127 void InterpreterMacroAssembler::verify_method_data_pointer() {
1128 assert(ProfileInterpreter, "must be profiling interpreter");
1129 #ifdef ASSERT
1130 Label verify_continue;
1131 test_method_data_pointer(verify_continue);
1132
1133 // If the mdp is valid, it will point to a DataLayout header which is
1134 // consistent with the bcp. The converse is highly probable also.
1135 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx);
1136 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method);
1137 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1138 add(R11_scratch1, R12_scratch2, R12_scratch2);
1139 cmpd(CCR0, R11_scratch1, R14_bcp);
1140 beq(CCR0, verify_continue);
1141
1142 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx);
1143
1144 bind(verify_continue);
1145 #endif
1146 }
1147
1148 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1149 Register Rscratch,
1150 Label &profile_continue) {
1151 assert(ProfileInterpreter, "must be profiling interpreter");
1152 // Control will flow to "profile_continue" if the counter is less than the
1153 // limit or if we call profile_method().
1154 Label done;
1155
1156 // If no method data exists, and the counter is high enough, make one.
1157 int ipl_offs = load_const_optimized(Rscratch, &InvocationCounter::InterpreterProfileLimit, R0, true);
1158 lwz(Rscratch, ipl_offs, Rscratch);
1159
1160 cmpdi(CCR0, R28_mdx, 0);
1161 // Test to see if we should create a method data oop.
1162 cmpd(CCR1, Rscratch /* InterpreterProfileLimit */, invocation_count);
1163 bne(CCR0, done);
1164 bge(CCR1, profile_continue);
1165
1166 // Build it now.
1167 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1168 set_method_data_pointer_for_bcp();
1169 b(profile_continue);
1170
1171 align(32, 12);
1172 bind(done);
1173 }
1174
1175 void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register branch_bcp, Register Rtmp) {
1176 assert_different_registers(backedge_count, Rtmp, branch_bcp);
1177 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
1178
1179 Label did_not_overflow;
1180 Label overflow_with_error;
1181
1182 int ibbl_offs = load_const_optimized(Rtmp, &InvocationCounter::InterpreterBackwardBranchLimit, R0, true);
1183 lwz(Rtmp, ibbl_offs, Rtmp);
1184 cmpw(CCR0, backedge_count, Rtmp);
1185
1186 blt(CCR0, did_not_overflow);
1187
1188 // When ProfileInterpreter is on, the backedge_count comes from the
1189 // methodDataOop, which value does not get reset on the call to
1190 // frequency_counter_overflow(). To avoid excessive calls to the overflow
1191 // routine while the method is being compiled, add a second test to make sure
1192 // the overflow function is called only once every overflow_frequency.
1193 if (ProfileInterpreter) {
1194 const int overflow_frequency = 1024;
1195 li(Rtmp, overflow_frequency-1);
1196 andr(Rtmp, Rtmp, backedge_count);
1197 cmpwi(CCR0, Rtmp, 0);
1198 bne(CCR0, did_not_overflow);
1199 }
1200
1201 // Overflow in loop, pass branch bytecode.
1202 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, true);
1203
1204 // Was an OSR adapter generated?
1205 // O0 = osr nmethod
1206 cmpdi(CCR0, R3_RET, 0);
1207 beq(CCR0, overflow_with_error);
1208
1209 // Has the nmethod been invalidated already?
1210 lwz(Rtmp, nmethod::entry_bci_offset(), R3_RET);
1211 cmpwi(CCR0, Rtmp, InvalidOSREntryBci);
1212 beq(CCR0, overflow_with_error);
1213
1214 // Migrate the interpreter frame off of the stack.
1215 // We can use all registers because we will not return to interpreter from this point.
1216
1217 // Save nmethod.
1218 const Register osr_nmethod = R31;
1219 mr(osr_nmethod, R3_RET);
1220 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1221 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1222 reset_last_Java_frame();
1223 // OSR buffer is in ARG1
1224
1225 // Remove the interpreter frame.
1226 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1227
1228 // Jump to the osr code.
1229 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1230 mtlr(R0);
1231 mtctr(R11_scratch1);
1232 bctr();
1233
1234 align(32, 12);
1235 bind(overflow_with_error);
1236 bind(did_not_overflow);
1237 }
1238
1239 // Store a value at some constant offset from the method data pointer.
1240 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1241 assert(ProfileInterpreter, "must be profiling interpreter");
1242
1243 std(value, constant, R28_mdx);
1244 }
1245
1246 // Increment the value at some constant offset from the method data pointer.
1247 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1248 Register counter_addr,
1249 Register Rbumped_count,
1250 bool decrement) {
1251 // Locate the counter at a fixed offset from the mdp:
1252 addi(counter_addr, R28_mdx, constant);
1253 increment_mdp_data_at(counter_addr, Rbumped_count, decrement);
1254 }
1255
1256 // Increment the value at some non-fixed (reg + constant) offset from
1257 // the method data pointer.
1258 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1259 int constant,
1260 Register scratch,
1261 Register Rbumped_count,
1262 bool decrement) {
1263 // Add the constant to reg to get the offset.
1264 add(scratch, R28_mdx, reg);
1265 // Then calculate the counter address.
1266 addi(scratch, scratch, constant);
1267 increment_mdp_data_at(scratch, Rbumped_count, decrement);
1268 }
1269
1270 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr,
1271 Register Rbumped_count,
1272 bool decrement) {
1273 assert(ProfileInterpreter, "must be profiling interpreter");
1274
1275 // Load the counter.
1276 ld(Rbumped_count, 0, counter_addr);
1277
1278 if (decrement) {
1279 // Decrement the register. Set condition codes.
1280 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment);
1281 // Store the decremented counter, if it is still negative.
1282 std(Rbumped_count, 0, counter_addr);
1283 // Note: add/sub overflow check are not ported, since 64 bit
1284 // calculation should never overflow.
1285 } else {
1286 // Increment the register. Set carry flag.
1287 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment);
1288 // Store the incremented counter.
1289 std(Rbumped_count, 0, counter_addr);
1290 }
1291 }
1292
1293 // Set a flag value at the current method data pointer position.
1294 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1295 Register scratch) {
1296 assert(ProfileInterpreter, "must be profiling interpreter");
1297 // Load the data header.
1298 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1299 // Set the flag.
1300 ori(scratch, scratch, flag_constant);
1301 // Store the modified header.
1302 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1303 }
1304
1305 // Test the location at some offset from the method data pointer.
1306 // If it is not equal to value, branch to the not_equal_continue Label.
1307 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1308 Register value,
1309 Label& not_equal_continue,
1310 Register test_out) {
1311 assert(ProfileInterpreter, "must be profiling interpreter");
1312
1313 ld(test_out, offset, R28_mdx);
1314 cmpd(CCR0, value, test_out);
1315 bne(CCR0, not_equal_continue);
1316 }
1317
1318 // Update the method data pointer by the displacement located at some fixed
1319 // offset from the method data pointer.
1320 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1321 Register scratch) {
1322 assert(ProfileInterpreter, "must be profiling interpreter");
1323
1324 ld(scratch, offset_of_disp, R28_mdx);
1325 add(R28_mdx, scratch, R28_mdx);
1326 }
1327
1328 // Update the method data pointer by the displacement located at the
1329 // offset (reg + offset_of_disp).
1330 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1331 int offset_of_disp,
1332 Register scratch) {
1333 assert(ProfileInterpreter, "must be profiling interpreter");
1334
1335 add(scratch, reg, R28_mdx);
1336 ld(scratch, offset_of_disp, scratch);
1337 add(R28_mdx, scratch, R28_mdx);
1338 }
1339
1340 // Update the method data pointer by a simple constant displacement.
1341 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1342 assert(ProfileInterpreter, "must be profiling interpreter");
1343 addi(R28_mdx, R28_mdx, constant);
1344 }
1345
1346 // Update the method data pointer for a _ret bytecode whose target
1347 // was not among our cached targets.
1348 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1349 Register return_bci) {
1350 assert(ProfileInterpreter, "must be profiling interpreter");
1351
1352 push(state);
1353 assert(return_bci->is_nonvolatile(), "need to protect return_bci");
1354 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1355 pop(state);
1356 }
1357
1358 // Increments the backedge counter.
1359 // Returns backedge counter + invocation counter in Rdst.
1360 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst,
1361 const Register Rtmp1, Register Rscratch) {
1362 assert(UseCompiler, "incrementing must be useful");
1363 assert_different_registers(Rdst, Rtmp1);
1364 const Register invocation_counter = Rtmp1;
1365 const Register counter = Rdst;
1366 // TODO ppc port assert(4 == InvocationCounter::sz_counter(), "unexpected field size.");
1367
1368 // Load backedge counter.
1369 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1370 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1371 // Load invocation counter.
1372 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) +
1373 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1374
1375 // Add the delta to the backedge counter.
1376 addi(counter, counter, InvocationCounter::count_increment);
1377
1378 // Mask the invocation counter.
1379 li(Rscratch, InvocationCounter::count_mask_value);
1380 andr(invocation_counter, invocation_counter, Rscratch);
1381
1382 // Store new counter value.
1383 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1384 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1385 // Return invocation counter + backedge counter.
1386 add(counter, counter, invocation_counter);
1387 }
1388
1389 // Count a taken branch in the bytecodes.
1390 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1391 if (ProfileInterpreter) {
1392 Label profile_continue;
1393
1394 // If no method data exists, go to profile_continue.
1395 test_method_data_pointer(profile_continue);
1396
1397 // We are taking a branch. Increment the taken count.
1398 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count);
1399
1400 // The method data pointer needs to be updated to reflect the new target.
1401 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1402 bind (profile_continue);
1403 }
1404 }
1405
1406 // Count a not-taken branch in the bytecodes.
1407 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) {
1408 if (ProfileInterpreter) {
1409 Label profile_continue;
1410
1411 // If no method data exists, go to profile_continue.
1412 test_method_data_pointer(profile_continue);
1413
1414 // We are taking a branch. Increment the not taken count.
1415 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2);
1416
1417 // The method data pointer needs to be updated to correspond to the
1418 // next bytecode.
1419 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1420 bind (profile_continue);
1421 }
1422 }
1423
1424 // Count a non-virtual call in the bytecodes.
1425 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) {
1426 if (ProfileInterpreter) {
1427 Label profile_continue;
1428
1429 // If no method data exists, go to profile_continue.
1430 test_method_data_pointer(profile_continue);
1431
1432 // We are making a call. Increment the count.
1433 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1434
1435 // The method data pointer needs to be updated to reflect the new target.
1436 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1437 bind (profile_continue);
1438 }
1439 }
1440
1441 // Count a final call in the bytecodes.
1442 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) {
1443 if (ProfileInterpreter) {
1444 Label profile_continue;
1445
1446 // If no method data exists, go to profile_continue.
1447 test_method_data_pointer(profile_continue);
1448
1449 // We are making a call. Increment the count.
1450 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1451
1452 // The method data pointer needs to be updated to reflect the new target.
1453 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1454 bind (profile_continue);
1455 }
1456 }
1457
1458 // Count a virtual call in the bytecodes.
1459 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver,
1460 Register Rscratch1,
1461 Register Rscratch2,
1462 bool receiver_can_be_null) {
1463 if (!ProfileInterpreter) { return; }
1464 Label profile_continue;
1465
1466 // If no method data exists, go to profile_continue.
1467 test_method_data_pointer(profile_continue);
1468
1469 Label skip_receiver_profile;
1470 if (receiver_can_be_null) {
1471 Label not_null;
1472 cmpdi(CCR0, Rreceiver, 0);
1473 bne(CCR0, not_null);
1474 // We are making a call. Increment the count for null receiver.
1475 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2);
1476 b(skip_receiver_profile);
1477 bind(not_null);
1478 }
1479
1480 // Record the receiver type.
1481 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true);
1482 bind(skip_receiver_profile);
1483
1484 // The method data pointer needs to be updated to reflect the new target.
1485 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1486 bind (profile_continue);
1487 }
1488
1489 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) {
1490 if (ProfileInterpreter) {
1491 Label profile_continue;
1492
1493 // If no method data exists, go to profile_continue.
1494 test_method_data_pointer(profile_continue);
1495
1496 int mdp_delta = in_bytes(BitData::bit_data_size());
1497 if (TypeProfileCasts) {
1498 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1499
1500 // Record the object type.
1501 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false);
1502 }
1503
1504 // The method data pointer needs to be updated.
1505 update_mdp_by_constant(mdp_delta);
1506
1507 bind (profile_continue);
1508 }
1509 }
1510
1511 void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) {
1512 if (ProfileInterpreter && TypeProfileCasts) {
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 count_offset = in_bytes(CounterData::count_offset());
1519 // Back up the address, since we have already bumped the mdp.
1520 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1521
1522 // *Decrement* the counter. We expect to see zero or small negatives.
1523 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true);
1524
1525 bind (profile_continue);
1526 }
1527 }
1528
1529 // Count a ret in the bytecodes.
1530 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, Register scratch1, Register scratch2) {
1531 if (ProfileInterpreter) {
1532 Label profile_continue;
1533 uint row;
1534
1535 // If no method data exists, go to profile_continue.
1536 test_method_data_pointer(profile_continue);
1537
1538 // Update the total ret count.
1539 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 );
1540
1541 for (row = 0; row < RetData::row_limit(); row++) {
1542 Label next_test;
1543
1544 // See if return_bci is equal to bci[n]:
1545 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1);
1546
1547 // return_bci is equal to bci[n]. Increment the count.
1548 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2);
1549
1550 // The method data pointer needs to be updated to reflect the new target.
1551 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1);
1552 b(profile_continue);
1553 bind(next_test);
1554 }
1555
1556 update_mdp_for_ret(state, return_bci);
1557
1558 bind (profile_continue);
1559 }
1560 }
1561
1562 // Count the default case of a switch construct.
1563 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) {
1564 if (ProfileInterpreter) {
1565 Label profile_continue;
1566
1567 // If no method data exists, go to profile_continue.
1568 test_method_data_pointer(profile_continue);
1569
1570 // Update the default case count
1571 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1572 scratch1, scratch2);
1573
1574 // The method data pointer needs to be updated.
1575 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()),
1576 scratch1);
1577
1578 bind (profile_continue);
1579 }
1580 }
1581
1582 // Count the index'th case of a switch construct.
1583 void InterpreterMacroAssembler::profile_switch_case(Register index,
1584 Register scratch1,
1585 Register scratch2,
1586 Register scratch3) {
1587 if (ProfileInterpreter) {
1588 assert_different_registers(index, scratch1, scratch2, scratch3);
1589 Label profile_continue;
1590
1591 // If no method data exists, go to profile_continue.
1592 test_method_data_pointer(profile_continue);
1593
1594 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes().
1595 li(scratch3, in_bytes(MultiBranchData::case_array_offset()));
1596
1597 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works");
1598 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1599 add(scratch1, scratch1, scratch3);
1600
1601 // Update the case count.
1602 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3);
1603
1604 // The method data pointer needs to be updated.
1605 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2);
1606
1607 bind (profile_continue);
1608 }
1609 }
1610
1611 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) {
1612 if (ProfileInterpreter) {
1613 assert_different_registers(Rscratch1, Rscratch2);
1614 Label profile_continue;
1615
1616 // If no method data exists, go to profile_continue.
1617 test_method_data_pointer(profile_continue);
1618
1619 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1);
1620
1621 // The method data pointer needs to be updated.
1622 int mdp_delta = in_bytes(BitData::bit_data_size());
1623 if (TypeProfileCasts) {
1624 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1625 }
1626 update_mdp_by_constant(mdp_delta);
1627
1628 bind (profile_continue);
1629 }
1630 }
1631
1632 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver,
1633 Register Rscratch1, Register Rscratch2,
1634 bool is_virtual_call) {
1635 assert(ProfileInterpreter, "must be profiling");
1636 assert_different_registers(Rreceiver, Rscratch1, Rscratch2);
1637
1638 Label done;
1639 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call);
1640 bind (done);
1641 }
1642
1643 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1644 Register receiver, Register scratch1, Register scratch2,
1645 int start_row, Label& done, bool is_virtual_call) {
1646 if (TypeProfileWidth == 0) {
1647 if (is_virtual_call) {
1648 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1649 }
1650 return;
1651 }
1652
1653 int last_row = VirtualCallData::row_limit() - 1;
1654 assert(start_row <= last_row, "must be work left to do");
1655 // Test this row for both the receiver and for null.
1656 // Take any of three different outcomes:
1657 // 1. found receiver => increment count and goto done
1658 // 2. found null => keep looking for case 1, maybe allocate this cell
1659 // 3. found something else => keep looking for cases 1 and 2
1660 // Case 3 is handled by a recursive call.
1661 for (int row = start_row; row <= last_row; row++) {
1662 Label next_test;
1663 bool test_for_null_also = (row == start_row);
1664
1665 // See if the receiver is receiver[n].
1666 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1667 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1);
1668 // delayed()->tst(scratch);
1669
1670 // The receiver is receiver[n]. Increment count[n].
1671 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1672 increment_mdp_data_at(count_offset, scratch1, scratch2);
1673 b(done);
1674 bind(next_test);
1675
1676 if (test_for_null_also) {
1677 Label found_null;
1678 // Failed the equality check on receiver[n]... Test for null.
1679 if (start_row == last_row) {
1680 // The only thing left to do is handle the null case.
1681 if (is_virtual_call) {
1682 // Scratch1 contains test_out from test_mdp_data_at.
1683 cmpdi(CCR0, scratch1, 0);
1684 beq(CCR0, found_null);
1685 // Receiver did not match any saved receiver and there is no empty row for it.
1686 // Increment total counter to indicate polymorphic case.
1687 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1688 b(done);
1689 bind(found_null);
1690 } else {
1691 cmpdi(CCR0, scratch1, 0);
1692 bne(CCR0, done);
1693 }
1694 break;
1695 }
1696 // Since null is rare, make it be the branch-taken case.
1697 cmpdi(CCR0, scratch1, 0);
1698 beq(CCR0, found_null);
1699
1700 // Put all the "Case 3" tests here.
1701 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call);
1702
1703 // Found a null. Keep searching for a matching receiver,
1704 // but remember that this is an empty (unused) slot.
1705 bind(found_null);
1706 }
1707 }
1708
1709 // In the fall-through case, we found no matching receiver, but we
1710 // observed the receiver[start_row] is NULL.
1711
1712 // Fill in the receiver field and increment the count.
1713 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1714 set_mdp_data_at(recvr_offset, receiver);
1715 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1716 li(scratch1, DataLayout::counter_increment);
1717 set_mdp_data_at(count_offset, scratch1);
1718 if (start_row > 0) {
1719 b(done);
1720 }
1721 }
1722
1723 // Argument and return type profilig.
1724 // kills: tmp, tmp2, R0, CR0, CR1
1725 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base,
1726 RegisterOrConstant mdo_addr_offs, Register tmp, Register tmp2) {
1727 Label do_nothing, do_update;
1728
1729 // tmp2 = obj is allowed
1730 assert_different_registers(obj, mdo_addr_base, tmp, R0);
1731 assert_different_registers(tmp2, mdo_addr_base, tmp, R0);
1732 const Register klass = tmp2;
1733
1734 verify_oop(obj);
1735
1736 ld(tmp, mdo_addr_offs, mdo_addr_base);
1737
1738 // Set null_seen if obj is 0.
1739 cmpdi(CCR0, obj, 0);
1740 ori(R0, tmp, TypeEntries::null_seen);
1741 beq(CCR0, do_update);
1742
1743 load_klass(klass, obj);
1744
1745 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
1746 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
1747 cmpd(CCR1, R0, klass);
1748 // Klass seen before, nothing to do (regardless of unknown bit).
1749 //beq(CCR1, do_nothing);
1750
1751 andi_(R0, klass, TypeEntries::type_unknown);
1752 // Already unknown. Nothing to do anymore.
1753 //bne(CCR0, do_nothing);
1754 crorc(/*CCR0 eq*/2, /*CCR1 eq*/4+2, /*CCR0 eq*/2); // cr0 eq = cr1 eq or cr0 ne
1755 beq(CCR0, do_nothing);
1756
1757 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
1758 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1759 beq(CCR0, do_update); // First time here. Set profile type.
1760
1761 // Different than before. Cannot keep accurate profile.
1762 ori(R0, tmp, TypeEntries::type_unknown);
1763
1764 bind(do_update);
1765 // update profile
1766 std(R0, mdo_addr_offs, mdo_addr_base);
1767
1768 align(32, 12);
1769 bind(do_nothing);
1770 }
1771
1772 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) {
1773 if (!ProfileInterpreter) {
1774 return;
1775 }
1776
1777 assert_different_registers(callee, tmp1, tmp2, R28_mdx);
1778
1779 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1780 Label profile_continue;
1781
1782 test_method_data_pointer(profile_continue);
1783
1784 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1785
1786 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx);
1787 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1788 bne(CCR0, profile_continue);
1789
1790 if (MethodData::profile_arguments()) {
1791 Label done;
1792 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1793 add(R28_mdx, off_to_args, R28_mdx);
1794
1795 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1796 if (i > 0 || MethodData::profile_return()) {
1797 // If return value type is profiled we may have no argument to profile.
1798 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1799 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count());
1800 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count());
1801 blt(CCR0, done);
1802 }
1803 ld(tmp1, in_bytes(Method::const_offset()), callee);
1804 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1);
1805 // Stack offset o (zero based) from the start of the argument
1806 // list, for n arguments translates into offset n - o - 1 from
1807 // the end of the argument list. But there's an extra slot at
1808 // the top of the stack. So the offset is n - o from Lesp.
1809 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx);
1810 subf(tmp1, tmp2, tmp1);
1811
1812 sldi(tmp1, tmp1, Interpreter::logStackElementSize);
1813 ldx(tmp1, tmp1, R15_esp);
1814
1815 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1);
1816
1817 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1818 addi(R28_mdx, R28_mdx, to_add);
1819 off_to_args += to_add;
1820 }
1821
1822 if (MethodData::profile_return()) {
1823 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1824 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1825 }
1826
1827 bind(done);
1828
1829 if (MethodData::profile_return()) {
1830 // We're right after the type profile for the last
1831 // argument. tmp1 is the number of cells left in the
1832 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1833 // if there's a return to profile.
1834 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1835 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size));
1836 add(R28_mdx, tmp1, R28_mdx);
1837 }
1838 } else {
1839 assert(MethodData::profile_return(), "either profile call args or call ret");
1840 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
1841 }
1842
1843 // Mdp points right after the end of the
1844 // CallTypeData/VirtualCallTypeData, right after the cells for the
1845 // return value type if there's one.
1846 align(32, 12);
1847 bind(profile_continue);
1848 }
1849 }
1850
1851 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
1852 assert_different_registers(ret, tmp1, tmp2);
1853 if (ProfileInterpreter && MethodData::profile_return()) {
1854 Label profile_continue;
1855
1856 test_method_data_pointer(profile_continue);
1857
1858 if (MethodData::profile_return_jsr292_only()) {
1859 // If we don't profile all invoke bytecodes we must make sure
1860 // it's a bytecode we indeed profile. We can't go back to the
1861 // begining of the ProfileData we intend to update to check its
1862 // type because we're right after it and we don't known its
1863 // length.
1864 lbz(tmp1, 0, R14_bcp);
1865 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method);
1866 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic);
1867 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle);
1868 cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
1869 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm);
1870 cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
1871 bne(CCR0, profile_continue);
1872 }
1873
1874 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2);
1875
1876 align(32, 12);
1877 bind(profile_continue);
1878 }
1879 }
1880
1881 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
1882 if (ProfileInterpreter && MethodData::profile_parameters()) {
1883 Label profile_continue, done;
1884
1885 test_method_data_pointer(profile_continue);
1886
1887 // Load the offset of the area within the MDO used for
1888 // parameters. If it's negative we're not profiling any parameters.
1889 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx);
1890 cmpwi(CCR0, tmp1, 0);
1891 blt(CCR0, profile_continue);
1892
1893 // Compute a pointer to the area for parameters from the offset
1894 // and move the pointer to the slot for the last
1895 // parameters. Collect profiling from last parameter down.
1896 // mdo start + parameters offset + array length - 1
1897
1898 // Pointer to the parameter area in the MDO.
1899 const Register mdp = tmp1;
1900 add(mdp, tmp1, R28_mdx);
1901
1902 // Pffset of the current profile entry to update.
1903 const Register entry_offset = tmp2;
1904 // entry_offset = array len in number of cells
1905 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp);
1906
1907 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1908 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
1909
1910 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
1911 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size));
1912 // entry_offset in bytes
1913 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1914
1915 Label loop;
1916 align(32, 12);
1917 bind(loop);
1918
1919 // Load offset on the stack from the slot for this parameter.
1920 ld(tmp3, entry_offset, mdp);
1921 sldi(tmp3, tmp3, Interpreter::logStackElementSize);
1922 neg(tmp3, tmp3);
1923 // Read the parameter from the local area.
1924 ldx(tmp3, tmp3, R18_locals);
1925
1926 // Make entry_offset now point to the type field for this parameter.
1927 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1928 assert(type_base > off_base, "unexpected");
1929 addi(entry_offset, entry_offset, type_base - off_base);
1930
1931 // Profile the parameter.
1932 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3);
1933
1934 // Go to next parameter.
1935 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base);
1936 cmpdi(CCR0, entry_offset, off_base + delta);
1937 addi(entry_offset, entry_offset, -delta);
1938 bge(CCR0, loop);
1939
1940 align(32, 12);
1941 bind(profile_continue);
1942 }
1943 }
1944
1945 // Add a InterpMonitorElem to stack (see frame_sparc.hpp).
1946 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) {
1947
1948 // Very-local scratch registers.
1949 const Register esp = Rtemp1;
1950 const Register slot = Rtemp2;
1951
1952 // Extracted monitor_size.
1953 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes();
1954 assert(Assembler::is_aligned((unsigned int)monitor_size,
1955 (unsigned int)frame::alignment_in_bytes),
1956 "size of a monitor must respect alignment of SP");
1957
1958 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor
1959 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp
1960
1961 // Shuffle expression stack down. Recall that stack_base points
1962 // just above the new expression stack bottom. Old_tos and new_tos
1963 // are used to scan thru the old and new expression stacks.
1964 if (!stack_is_empty) {
1965 Label copy_slot, copy_slot_finished;
1966 const Register n_slots = slot;
1967
1968 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack).
1969 subf(n_slots, esp, R26_monitor);
1970 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy.
1971 assert(LogBytesPerWord == 3, "conflicts assembler instructions");
1972 beq(CCR0, copy_slot_finished); // Nothing to copy.
1973
1974 mtctr(n_slots);
1975
1976 // loop
1977 bind(copy_slot);
1978 ld(slot, 0, esp); // Move expression stack down.
1979 std(slot, -monitor_size, esp); // distance = monitor_size
1980 addi(esp, esp, BytesPerWord);
1981 bdnz(copy_slot);
1982
1983 bind(copy_slot_finished);
1984 }
1985
1986 addi(R15_esp, R15_esp, -monitor_size);
1987 addi(R26_monitor, R26_monitor, -monitor_size);
1988
1989 // Restart interpreter
1990 }
1991
1992 // ============================================================================
1993 // Java locals access
1994
1995 // Load a local variable at index in Rindex into register Rdst_value.
1996 // Also puts address of local into Rdst_address as a service.
1997 // Kills:
1998 // - Rdst_value
1999 // - Rdst_address
2000 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) {
2001 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2002 subf(Rdst_address, Rdst_address, R18_locals);
2003 lwz(Rdst_value, 0, Rdst_address);
2004 }
2005
2006 // Load a local variable at index in Rindex into register Rdst_value.
2007 // Also puts address of local into Rdst_address as a service.
2008 // Kills:
2009 // - Rdst_value
2010 // - Rdst_address
2011 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) {
2012 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2013 subf(Rdst_address, Rdst_address, R18_locals);
2014 ld(Rdst_value, -8, Rdst_address);
2015 }
2016
2017 // Load a local variable at index in Rindex into register Rdst_value.
2018 // Also puts address of local into Rdst_address as a service.
2019 // Input:
2020 // - Rindex: slot nr of local variable
2021 // Kills:
2022 // - Rdst_value
2023 // - Rdst_address
2024 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, Register Rdst_address, Register Rindex) {
2025 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2026 subf(Rdst_address, Rdst_address, R18_locals);
2027 ld(Rdst_value, 0, Rdst_address);
2028 }
2029
2030 // Load a local variable at index in Rindex into register Rdst_value.
2031 // Also puts address of local into Rdst_address as a service.
2032 // Kills:
2033 // - Rdst_value
2034 // - Rdst_address
2035 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) {
2036 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2037 subf(Rdst_address, Rdst_address, R18_locals);
2038 lfs(Rdst_value, 0, Rdst_address);
2039 }
2040
2041 // Load a local variable at index in Rindex into register Rdst_value.
2042 // Also puts address of local into Rdst_address as a service.
2043 // Kills:
2044 // - Rdst_value
2045 // - Rdst_address
2046 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) {
2047 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2048 subf(Rdst_address, Rdst_address, R18_locals);
2049 lfd(Rdst_value, -8, Rdst_address);
2050 }
2051
2052 // Store an int value at local variable slot Rindex.
2053 // Kills:
2054 // - Rindex
2055 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) {
2056 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2057 subf(Rindex, Rindex, R18_locals);
2058 stw(Rvalue, 0, Rindex);
2059 }
2060
2061 // Store a long value at local variable slot Rindex.
2062 // Kills:
2063 // - Rindex
2064 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) {
2065 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2066 subf(Rindex, Rindex, R18_locals);
2067 std(Rvalue, -8, Rindex);
2068 }
2069
2070 // Store an oop value at local variable slot Rindex.
2071 // Kills:
2072 // - Rindex
2073 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) {
2074 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2075 subf(Rindex, Rindex, R18_locals);
2076 std(Rvalue, 0, Rindex);
2077 }
2078
2079 // Store an int value at local variable slot Rindex.
2080 // Kills:
2081 // - Rindex
2082 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) {
2083 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2084 subf(Rindex, Rindex, R18_locals);
2085 stfs(Rvalue, 0, Rindex);
2086 }
2087
2088 // Store an int value at local variable slot Rindex.
2089 // Kills:
2090 // - Rindex
2091 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) {
2092 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2093 subf(Rindex, Rindex, R18_locals);
2094 stfd(Rvalue, -8, Rindex);
2095 }
2096
2097 // Read pending exception from thread and jump to interpreter.
2098 // Throw exception entry if one if pending. Fall through otherwise.
2099 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) {
2100 assert_different_registers(Rscratch1, Rscratch2, R3);
2101 Register Rexception = Rscratch1;
2102 Register Rtmp = Rscratch2;
2103 Label Ldone;
2104 // Get pending exception oop.
2105 ld(Rexception, thread_(pending_exception));
2106 cmpdi(CCR0, Rexception, 0);
2107 beq(CCR0, Ldone);
2108 li(Rtmp, 0);
2109 mr_if_needed(R3, Rexception);
2110 std(Rtmp, thread_(pending_exception)); // Clear exception in thread
2111 if (Interpreter::rethrow_exception_entry() != NULL) {
2112 // Already got entry address.
2113 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry());
2114 } else {
2115 // Dynamically load entry address.
2116 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true);
2117 ld(Rtmp, simm16_rest, Rtmp);
2118 }
2119 mtctr(Rtmp);
2120 save_interpreter_state(Rtmp);
2121 bctr();
2122
2123 align(32, 12);
2124 bind(Ldone);
2125 }
2126
2127 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) {
2128 save_interpreter_state(R11_scratch1);
2129
2130 MacroAssembler::call_VM(oop_result, entry_point, false);
2131
2132 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2133
2134 check_and_handle_popframe(R11_scratch1);
2135 check_and_handle_earlyret(R11_scratch1);
2136 // Now check exceptions manually.
2137 if (check_exceptions) {
2138 check_and_forward_exception(R11_scratch1, R12_scratch2);
2139 }
2140 }
2141
2142 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions) {
2143 // ARG1 is reserved for the thread.
2144 mr_if_needed(R4_ARG2, arg_1);
2145 call_VM(oop_result, entry_point, check_exceptions);
2146 }
2147
2148 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) {
2149 // ARG1 is reserved for the thread.
2150 mr_if_needed(R4_ARG2, arg_1);
2151 assert(arg_2 != R4_ARG2, "smashed argument");
2152 mr_if_needed(R5_ARG3, arg_2);
2153 call_VM(oop_result, entry_point, check_exceptions);
2154 }
2155
2156 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) {
2157 // ARG1 is reserved for the thread.
2158 mr_if_needed(R4_ARG2, arg_1);
2159 assert(arg_2 != R4_ARG2, "smashed argument");
2160 mr_if_needed(R5_ARG3, arg_2);
2161 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument");
2162 mr_if_needed(R6_ARG4, arg_3);
2163 call_VM(oop_result, entry_point, check_exceptions);
2164 }
2165
2166 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) {
2167 ld(scratch, 0, R1_SP);
2168 std(R15_esp, _ijava_state_neg(esp), scratch);
2169 std(R14_bcp, _ijava_state_neg(bcp), scratch);
2170 std(R26_monitor, _ijava_state_neg(monitors), scratch);
2171 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); }
2172 // Other entries should be unchanged.
2173 }
2174
2175 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) {
2176 ld(scratch, 0, R1_SP);
2177 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception).
2178 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code.
2179 if (!bcp_and_mdx_only) {
2180 // Following ones are Metadata.
2181 ld(R19_method, _ijava_state_neg(method), scratch);
2182 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch);
2183 // Following ones are stack addresses and don't require reload.
2184 ld(R15_esp, _ijava_state_neg(esp), scratch);
2185 ld(R18_locals, _ijava_state_neg(locals), scratch);
2186 ld(R26_monitor, _ijava_state_neg(monitors), scratch);
2187 }
2188 #ifdef ASSERT
2189 {
2190 Label Lok;
2191 subf(R0, R1_SP, scratch);
2192 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size);
2193 bge(CCR0, Lok);
2194 stop("frame too small (restore istate)", 0x5432);
2195 bind(Lok);
2196 }
2197 {
2198 Label Lok;
2199 ld(R0, _ijava_state_neg(ijava_reserved), scratch);
2200 cmpdi(CCR0, R0, 0x5afe);
2201 beq(CCR0, Lok);
2202 stop("frame corrupted (restore istate)", 0x5afe);
2203 bind(Lok);
2204 }
2205 #endif
2206 }
2207
2208 #endif // !CC_INTERP
2209
2210 void InterpreterMacroAssembler::get_method_counters(Register method,
2211 Register Rcounters,
2212 Label& skip) {
2213 BLOCK_COMMENT("Load and ev. allocate counter object {");
2214 Label has_counters;
2215 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2216 cmpdi(CCR0, Rcounters, 0);
2217 bne(CCR0, has_counters);
2218 call_VM(noreg, CAST_FROM_FN_PTR(address,
2219 InterpreterRuntime::build_method_counters), method, false);
2220 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2221 cmpdi(CCR0, Rcounters, 0);
2222 beq(CCR0, skip); // No MethodCounters, OutOfMemory.
2223 BLOCK_COMMENT("} Load and ev. allocate counter object");
2224
2225 bind(has_counters);
2226 }
2227
2228 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register iv_be_count, Register Rtmp_r0) {
2229 assert(UseCompiler, "incrementing must be useful");
2230 Register invocation_count = iv_be_count;
2231 Register backedge_count = Rtmp_r0;
2232 int delta = InvocationCounter::count_increment;
2233
2234 // Load each counter in a register.
2235 // ld(inv_counter, Rtmp);
2236 // ld(be_counter, Rtmp2);
2237 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() +
2238 InvocationCounter::counter_offset());
2239 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() +
2240 InvocationCounter::counter_offset());
2241
2242 BLOCK_COMMENT("Increment profiling counters {");
2243
2244 // Load the backedge counter.
2245 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int
2246 // Mask the backedge counter.
2247 Register tmp = invocation_count;
2248 li(tmp, InvocationCounter::count_mask_value);
2249 andr(backedge_count, tmp, backedge_count); // Cannot use andi, need sign extension of count_mask_value.
2250
2251 // Load the invocation counter.
2252 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int
2253 // Add the delta to the invocation counter and store the result.
2254 addi(invocation_count, invocation_count, delta);
2255 // Store value.
2256 stw(invocation_count, inv_counter_offset, Rcounters);
2257
2258 // Add invocation counter + backedge counter.
2259 add(iv_be_count, backedge_count, invocation_count);
2260
2261 // Note that this macro must leave the backedge_count + invocation_count in
2262 // register iv_be_count!
2263 BLOCK_COMMENT("} Increment profiling counters");
2264 }
2265
2266 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2267 if (state == atos) { MacroAssembler::verify_oop(reg); }
2268 }
2269
2270 #ifndef CC_INTERP
2271 // Local helper function for the verify_oop_or_return_address macro.
2272 static bool verify_return_address(Method* m, int bci) {
2273 #ifndef PRODUCT
2274 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci;
2275 // Assume it is a valid return address if it is inside m and is preceded by a jsr.
2276 if (!m->contains(pc)) return false;
2277 address jsr_pc;
2278 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2279 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2280 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2281 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2282 #endif // PRODUCT
2283 return false;
2284 }
2285
2286 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2287 if (VerifyFPU) {
2288 unimplemented("verfiyFPU");
2289 }
2290 }
2291
2292 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2293 if (!VerifyOops) return;
2294
2295 // The VM documentation for the astore[_wide] bytecode allows
2296 // the TOS to be not only an oop but also a return address.
2297 Label test;
2298 Label skip;
2299 // See if it is an address (in the current method):
2300
2301 const int log2_bytecode_size_limit = 16;
2302 srdi_(Rtmp, reg, log2_bytecode_size_limit);
2303 bne(CCR0, test);
2304
2305 address fd = CAST_FROM_FN_PTR(address, verify_return_address);
2306 const int nbytes_save = 11*8; // volatile gprs except R0
2307 save_volatile_gprs(R1_SP, -nbytes_save); // except R0
2308 save_LR_CR(Rtmp); // Save in old frame.
2309 push_frame_reg_args(nbytes_save, Rtmp);
2310
2311 load_const_optimized(Rtmp, fd, R0);
2312 mr_if_needed(R4_ARG2, reg);
2313 mr(R3_ARG1, R19_method);
2314 call_c(Rtmp); // call C
2315
2316 pop_frame();
2317 restore_LR_CR(Rtmp);
2318 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
2319 b(skip);
2320
2321 // Perform a more elaborate out-of-line call.
2322 // Not an address; verify it:
2323 bind(test);
2324 verify_oop(reg);
2325 bind(skip);
2326 }
2327 #endif // !CC_INTERP
2328
2329 // Inline assembly for:
2330 //
2331 // if (thread is in interp_only_mode) {
2332 // InterpreterRuntime::post_method_entry();
2333 // }
2334 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) ||
2335 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) {
2336 // SharedRuntime::jvmpi_method_entry(method, receiver);
2337 // }
2338 void InterpreterMacroAssembler::notify_method_entry() {
2339 // JVMTI
2340 // Whenever JVMTI puts a thread in interp_only_mode, method
2341 // entry/exit events are sent for that thread to track stack
2342 // depth. If it is possible to enter interp_only_mode we add
2343 // the code to check if the event should be sent.
2344 if (JvmtiExport::can_post_interpreter_events()) {
2345 Label jvmti_post_done;
2346
2347 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2348 cmpwi(CCR0, R0, 0);
2349 beq(CCR0, jvmti_post_done);
2350 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry),
2351 /*check_exceptions=*/true CC_INTERP_ONLY(&& false));
2352
2353 bind(jvmti_post_done);
2354 }
2355 }
2356
2357 // Inline assembly for:
2358 //
2359 // if (thread is in interp_only_mode) {
2360 // // save result
2361 // InterpreterRuntime::post_method_exit();
2362 // // restore result
2363 // }
2364 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) {
2365 // // save result
2366 // SharedRuntime::jvmpi_method_exit();
2367 // // restore result
2368 // }
2369 //
2370 // Native methods have their result stored in d_tmp and l_tmp.
2371 // Java methods have their result stored in the expression stack.
2372 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state,
2373 NotifyMethodExitMode mode, bool check_exceptions) {
2374 // JVMTI
2375 // Whenever JVMTI puts a thread in interp_only_mode, method
2376 // entry/exit events are sent for that thread to track stack
2377 // depth. If it is possible to enter interp_only_mode we add
2378 // the code to check if the event should be sent.
2379 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2380 Label jvmti_post_done;
2381
2382 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2383 cmpwi(CCR0, R0, 0);
2384 beq(CCR0, jvmti_post_done);
2385 CC_INTERP_ONLY(assert(is_native_method && !check_exceptions, "must not push state"));
2386 if (!is_native_method) push(state); // Expose tos to GC.
2387 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit),
2388 /*check_exceptions=*/check_exceptions);
2389 if (!is_native_method) pop(state);
2390
2391 align(32, 12);
2392 bind(jvmti_post_done);
2393 }
2394
2395 // Dtrace support not implemented.
2396 }
2397
2398 #ifdef CC_INTERP
2399 // Convert the current TOP_IJAVA_FRAME into a PARENT_IJAVA_FRAME
2400 // (using parent_frame_resize) and push a new interpreter
2401 // TOP_IJAVA_FRAME (using frame_size).
2402 void InterpreterMacroAssembler::push_interpreter_frame(Register top_frame_size, Register parent_frame_resize,
2403 Register tmp1, Register tmp2, Register tmp3,
2404 Register tmp4, Register pc) {
2405 assert_different_registers(top_frame_size, parent_frame_resize, tmp1, tmp2, tmp3, tmp4);
2406 ld(tmp1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
2407 mr(tmp2/*top_frame_sp*/, R1_SP);
2408 // Move initial_caller_sp.
2409 ld(tmp4, _top_ijava_frame_abi(initial_caller_sp), R1_SP);
2410 neg(parent_frame_resize, parent_frame_resize);
2411 resize_frame(parent_frame_resize/*-parent_frame_resize*/, tmp3);
2412
2413 // Set LR in new parent frame.
2414 std(tmp1, _abi(lr), R1_SP);
2415 // Set top_frame_sp info for new parent frame.
2416 std(tmp2, _parent_ijava_frame_abi(top_frame_sp), R1_SP);
2417 std(tmp4, _parent_ijava_frame_abi(initial_caller_sp), R1_SP);
2418
2419 // Push new TOP_IJAVA_FRAME.
2420 push_frame(top_frame_size, tmp2);
2421
2422 get_PC_trash_LR(tmp3);
2423 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
2424 // Used for non-initial callers by unextended_sp().
2425 std(R1_SP, _top_ijava_frame_abi(initial_caller_sp), R1_SP);
2426 }
2427
2428 // Pop the topmost TOP_IJAVA_FRAME and convert the previous
2429 // PARENT_IJAVA_FRAME back into a TOP_IJAVA_FRAME.
2430 void InterpreterMacroAssembler::pop_interpreter_frame(Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
2431 assert_different_registers(tmp1, tmp2, tmp3, tmp4);
2432
2433 ld(tmp1/*caller's sp*/, _abi(callers_sp), R1_SP);
2434 ld(tmp3, _abi(lr), tmp1);
2435
2436 ld(tmp4, _parent_ijava_frame_abi(initial_caller_sp), tmp1);
2437
2438 ld(tmp2/*caller's caller's sp*/, _abi(callers_sp), tmp1);
2439 // Merge top frame.
2440 std(tmp2, _abi(callers_sp), R1_SP);
2441
2442 ld(tmp2, _parent_ijava_frame_abi(top_frame_sp), tmp1);
2443
2444 // Update C stack pointer to caller's top_abi.
2445 resize_frame_absolute(tmp2/*addr*/, tmp1/*tmp*/, tmp2/*tmp*/);
2446
2447 // Update LR in top_frame.
2448 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
2449
2450 std(tmp4, _top_ijava_frame_abi(initial_caller_sp), R1_SP);
2451
2452 // Store the top-frame stack-pointer for c2i adapters.
2453 std(R1_SP, _top_ijava_frame_abi(top_frame_sp), R1_SP);
2454 }
2455
2456 // Turn state's interpreter frame into the current TOP_IJAVA_FRAME.
2457 void InterpreterMacroAssembler::pop_interpreter_frame_to_state(Register state, Register tmp1, Register tmp2, Register tmp3) {
2458 assert_different_registers(R14_state, R15_prev_state, tmp1, tmp2, tmp3);
2459
2460 if (state == R14_state) {
2461 ld(tmp1/*state's fp*/, state_(_last_Java_fp));
2462 ld(tmp2/*state's sp*/, state_(_last_Java_sp));
2463 } else if (state == R15_prev_state) {
2464 ld(tmp1/*state's fp*/, prev_state_(_last_Java_fp));
2465 ld(tmp2/*state's sp*/, prev_state_(_last_Java_sp));
2466 } else {
2467 ShouldNotReachHere();
2468 }
2469
2470 // Merge top frames.
2471 std(tmp1, _abi(callers_sp), R1_SP);
2472
2473 // Tmp2 is new SP.
2474 // Tmp1 is parent's SP.
2475 resize_frame_absolute(tmp2/*addr*/, tmp1/*tmp*/, tmp2/*tmp*/);
2476
2477 // Update LR in top_frame.
2478 // Must be interpreter frame.
2479 get_PC_trash_LR(tmp3);
2480 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
2481 // Used for non-initial callers by unextended_sp().
2482 std(R1_SP, _top_ijava_frame_abi(initial_caller_sp), R1_SP);
2483 }
2484
2485 // Set SP to initial caller's sp, but before fix the back chain.
2486 void InterpreterMacroAssembler::resize_frame_to_initial_caller(Register tmp1, Register tmp2) {
2487 ld(tmp1, _parent_ijava_frame_abi(initial_caller_sp), R1_SP);
2488 ld(tmp2, _parent_ijava_frame_abi(callers_sp), R1_SP);
2489 std(tmp2, _parent_ijava_frame_abi(callers_sp), tmp1); // Fix back chain ...
2490 mr(R1_SP, tmp1); // ... and resize to initial caller.
2491 }
2492
2493 // Pop the current interpreter state (without popping the correspoding
2494 // frame) and restore R14_state and R15_prev_state accordingly.
2495 // Use prev_state_may_be_0 to indicate whether prev_state may be 0
2496 // in order to generate an extra check before retrieving prev_state_(_prev_link).
2497 void InterpreterMacroAssembler::pop_interpreter_state(bool prev_state_may_be_0)
2498 {
2499 // Move prev_state to state and restore prev_state from state_(_prev_link).
2500 Label prev_state_is_0;
2501 mr(R14_state, R15_prev_state);
2502
2503 // Don't retrieve /*state==*/prev_state_(_prev_link)
2504 // if /*state==*/prev_state is 0.
2505 if (prev_state_may_be_0) {
2506 cmpdi(CCR0, R15_prev_state, 0);
2507 beq(CCR0, prev_state_is_0);
2508 }
2509
2510 ld(R15_prev_state, /*state==*/prev_state_(_prev_link));
2511 bind(prev_state_is_0);
2512 }
2513
2514 void InterpreterMacroAssembler::restore_prev_state() {
2515 // _prev_link is private, but cInterpreter is a friend.
2516 ld(R15_prev_state, state_(_prev_link));
2517 }
2518 #endif // CC_INTERP