Tue, 17 Oct 2017 12:58:25 +0800
merge
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 */
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"
33 #ifdef PRODUCT
34 #define BLOCK_COMMENT(str) // nothing
35 #else
36 #define BLOCK_COMMENT(str) block_comment(str)
37 #endif
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 }
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 }
59 #ifndef CC_INTERP
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 }
69 dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state));
70 }
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 }
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);
85 load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state));
87 sldi(bytecode, bytecode, LogBytesPerWord);
88 ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode);
89 }
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 }
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;
105 // Check the "pending popframe condition" flag in the current thread.
106 lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
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);
115 andi_(R0, scratch_reg, JavaThread::popframe_processing_bit);
116 bne(CCR0, L);
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
131 // Jump to Interpreter::_remove_activation_preserving_args_entry.
132 mtctr(R3_RET);
133 bctr();
135 align(32, 12);
136 bind(L);
137 }
138 }
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);
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);
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();
158 align(32, 12);
159 bind(Lno_early_ret);
160 }
161 }
163 void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) {
164 const Register RjvmtiState = Rscratch1;
165 const Register Rscratch2 = R0;
167 ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
168 li(Rscratch2, 0);
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 ctos: // fall through
178 case stos: // fall through
179 case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
180 break;
181 case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
182 break;
183 case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
184 break;
185 case vtos: break;
186 default : ShouldNotReachHere();
187 }
189 // Clean up tos value in the jvmti thread state.
190 std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState);
191 // Set tos state field to illegal value.
192 li(Rscratch2, ilgl);
193 stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState);
194 }
196 // Common code to dispatch and dispatch_only.
197 // Dispatch value in Lbyte_code and increment Lbcp.
199 void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) {
200 address table_base = (address)Interpreter::dispatch_table((TosState)0);
201 intptr_t table_offs = (intptr_t)table - (intptr_t)table_base;
202 if (is_simm16(table_offs)) {
203 addi(dst, R25_templateTableBase, (int)table_offs);
204 } else {
205 load_const_optimized(dst, table, R0);
206 }
207 }
209 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode, address* table, bool verify) {
210 if (verify) {
211 unimplemented("dispatch_Lbyte_code: verify"); // See Sparc Implementation to implement this
212 }
214 #ifdef FAST_DISPATCH
215 unimplemented("dispatch_Lbyte_code FAST_DISPATCH");
216 #else
217 assert_different_registers(bytecode, R11_scratch1);
219 // Calc dispatch table address.
220 load_dispatch_table(R11_scratch1, table);
222 sldi(R12_scratch2, bytecode, LogBytesPerWord);
223 ldx(R11_scratch1, R11_scratch1, R12_scratch2);
225 // Jump off!
226 mtctr(R11_scratch1);
227 bctr();
228 #endif
229 }
231 void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) {
232 sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize);
233 ldx(Rrecv_dst, Rrecv_dst, R15_esp);
234 }
236 // helpers for expression stack
238 void InterpreterMacroAssembler::pop_i(Register r) {
239 lwzu(r, Interpreter::stackElementSize, R15_esp);
240 }
242 void InterpreterMacroAssembler::pop_ptr(Register r) {
243 ldu(r, Interpreter::stackElementSize, R15_esp);
244 }
246 void InterpreterMacroAssembler::pop_l(Register r) {
247 ld(r, Interpreter::stackElementSize, R15_esp);
248 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
249 }
251 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
252 lfsu(f, Interpreter::stackElementSize, R15_esp);
253 }
255 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
256 lfd(f, Interpreter::stackElementSize, R15_esp);
257 addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize);
258 }
260 void InterpreterMacroAssembler::push_i(Register r) {
261 stw(r, 0, R15_esp);
262 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
263 }
265 void InterpreterMacroAssembler::push_ptr(Register r) {
266 std(r, 0, R15_esp);
267 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
268 }
270 void InterpreterMacroAssembler::push_l(Register r) {
271 std(r, - Interpreter::stackElementSize, R15_esp);
272 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
273 }
275 void InterpreterMacroAssembler::push_f(FloatRegister f) {
276 stfs(f, 0, R15_esp);
277 addi(R15_esp, R15_esp, - Interpreter::stackElementSize );
278 }
280 void InterpreterMacroAssembler::push_d(FloatRegister f) {
281 stfd(f, - Interpreter::stackElementSize, R15_esp);
282 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
283 }
285 void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) {
286 std(first, 0, R15_esp);
287 std(second, -Interpreter::stackElementSize, R15_esp);
288 addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize );
289 }
291 void InterpreterMacroAssembler::push_l_pop_d(Register l, FloatRegister d) {
292 std(l, 0, R15_esp);
293 lfd(d, 0, R15_esp);
294 }
296 void InterpreterMacroAssembler::push_d_pop_l(FloatRegister d, Register l) {
297 stfd(d, 0, R15_esp);
298 ld(l, 0, R15_esp);
299 }
301 void InterpreterMacroAssembler::push(TosState state) {
302 switch (state) {
303 case atos: push_ptr(); break;
304 case btos:
305 case ctos:
306 case stos:
307 case itos: push_i(); break;
308 case ltos: push_l(); break;
309 case ftos: push_f(); break;
310 case dtos: push_d(); break;
311 case vtos: /* nothing to do */ break;
312 default : ShouldNotReachHere();
313 }
314 }
316 void InterpreterMacroAssembler::pop(TosState state) {
317 switch (state) {
318 case atos: pop_ptr(); break;
319 case btos:
320 case ctos:
321 case stos:
322 case itos: pop_i(); break;
323 case ltos: pop_l(); break;
324 case ftos: pop_f(); break;
325 case dtos: pop_d(); break;
326 case vtos: /* nothing to do */ break;
327 default : ShouldNotReachHere();
328 }
329 verify_oop(R17_tos, state);
330 }
332 void InterpreterMacroAssembler::empty_expression_stack() {
333 addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
334 }
336 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset,
337 Register Rdst,
338 signedOrNot is_signed) {
339 #if defined(VM_LITTLE_ENDIAN)
340 if (bcp_offset) {
341 load_const_optimized(Rdst, bcp_offset);
342 lhbrx(Rdst, R14_bcp, Rdst);
343 } else {
344 lhbrx(Rdst, R14_bcp);
345 }
346 if (is_signed == Signed) {
347 extsh(Rdst, Rdst);
348 }
349 #else
350 // Read Java big endian format.
351 if (is_signed == Signed) {
352 lha(Rdst, bcp_offset, R14_bcp);
353 } else {
354 lhz(Rdst, bcp_offset, R14_bcp);
355 }
356 #endif
357 }
359 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset,
360 Register Rdst,
361 signedOrNot is_signed) {
362 #if defined(VM_LITTLE_ENDIAN)
363 if (bcp_offset) {
364 load_const_optimized(Rdst, bcp_offset);
365 lwbrx(Rdst, R14_bcp, Rdst);
366 } else {
367 lwbrx(Rdst, R14_bcp);
368 }
369 if (is_signed == Signed) {
370 extsw(Rdst, Rdst);
371 }
372 #else
373 // Read Java big endian format.
374 if (bcp_offset & 3) { // Offset unaligned?
375 load_const_optimized(Rdst, bcp_offset);
376 if (is_signed == Signed) {
377 lwax(Rdst, R14_bcp, Rdst);
378 } else {
379 lwzx(Rdst, R14_bcp, Rdst);
380 }
381 } else {
382 if (is_signed == Signed) {
383 lwa(Rdst, bcp_offset, R14_bcp);
384 } else {
385 lwz(Rdst, bcp_offset, R14_bcp);
386 }
387 }
388 #endif
389 }
392 // Load the constant pool cache index from the bytecode stream.
393 //
394 // Kills / writes:
395 // - Rdst, Rscratch
396 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, size_t index_size) {
397 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
398 // Cache index is always in the native format, courtesy of Rewriter.
399 if (index_size == sizeof(u2)) {
400 lhz(Rdst, bcp_offset, R14_bcp);
401 } else if (index_size == sizeof(u4)) {
402 assert(EnableInvokeDynamic, "giant index used only for JSR 292");
403 if (bcp_offset & 3) {
404 load_const_optimized(Rdst, bcp_offset);
405 lwax(Rdst, R14_bcp, Rdst);
406 } else {
407 lwa(Rdst, bcp_offset, R14_bcp);
408 }
409 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
410 nand(Rdst, Rdst, Rdst); // convert to plain index
411 } else if (index_size == sizeof(u1)) {
412 lbz(Rdst, bcp_offset, R14_bcp);
413 } else {
414 ShouldNotReachHere();
415 }
416 // Rdst now contains cp cache index.
417 }
419 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset, size_t index_size) {
420 get_cache_index_at_bcp(cache, bcp_offset, index_size);
421 sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord));
422 add(cache, R27_constPoolCache, cache);
423 }
425 // Load 4-byte signed or unsigned integer in Java format (that is, big-endian format)
426 // from (Rsrc)+offset.
427 void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset,
428 signedOrNot is_signed) {
429 #if defined(VM_LITTLE_ENDIAN)
430 if (offset) {
431 load_const_optimized(Rdst, offset);
432 lwbrx(Rdst, Rdst, Rsrc);
433 } else {
434 lwbrx(Rdst, Rsrc);
435 }
436 if (is_signed == Signed) {
437 extsw(Rdst, Rdst);
438 }
439 #else
440 if (is_signed == Signed) {
441 lwa(Rdst, offset, Rsrc);
442 } else {
443 lwz(Rdst, offset, Rsrc);
444 }
445 #endif
446 }
448 // Load object from cpool->resolved_references(index).
449 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) {
450 assert_different_registers(result, index);
451 get_constant_pool(result);
453 // Convert from field index to resolved_references() index and from
454 // word index to byte offset. Since this is a java object, it can be compressed.
455 Register tmp = index; // reuse
456 sldi(tmp, index, LogBytesPerHeapOop);
457 // Load pointer for resolved_references[] objArray.
458 ld(result, ConstantPool::resolved_references_offset_in_bytes(), result);
459 // JNIHandles::resolve(result)
460 ld(result, 0, result);
461 #ifdef ASSERT
462 Label index_ok;
463 lwa(R0, arrayOopDesc::length_offset_in_bytes(), result);
464 sldi(R0, R0, LogBytesPerHeapOop);
465 cmpd(CCR0, tmp, R0);
466 blt(CCR0, index_ok);
467 stop("resolved reference index out of bounds", 0x09256);
468 bind(index_ok);
469 #endif
470 // Add in the index.
471 add(result, tmp, result);
472 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result);
473 }
475 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
476 // a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2.
477 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1,
478 Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) {
479 // Profile the not-null value's klass.
480 profile_typecheck(Rsub_klass, Rtmp1, Rtmp2);
481 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
482 profile_typecheck_failed(Rtmp1, Rtmp2);
483 }
485 void InterpreterMacroAssembler::generate_stack_overflow_check_with_compare_and_throw(Register Rmem_frame_size, Register Rscratch1) {
486 Label done;
487 sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
488 ld(Rscratch1, thread_(stack_overflow_limit));
489 cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
490 bgt(CCR0/*is_stack_overflow*/, done);
492 // Load target address of the runtime stub.
493 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order");
494 load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0);
495 mtctr(Rscratch1);
496 // Restore caller_sp.
497 #ifdef ASSERT
498 ld(Rscratch1, 0, R1_SP);
499 ld(R0, 0, R21_sender_SP);
500 cmpd(CCR0, R0, Rscratch1);
501 asm_assert_eq("backlink", 0x547);
502 #endif // ASSERT
503 mr(R1_SP, R21_sender_SP);
504 bctr();
506 align(32, 12);
507 bind(done);
508 }
510 // Separate these two to allow for delay slot in middle.
511 // These are used to do a test and full jump to exception-throwing code.
513 // Check that index is in range for array, then shift index by index_shift,
514 // and put arrayOop + shifted_index into res.
515 // Note: res is still shy of address by array offset into object.
517 void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, int index_shift, Register Rtmp, Register Rres) {
518 // Check that index is in range for array, then shift index by index_shift,
519 // and put arrayOop + shifted_index into res.
520 // Note: res is still shy of address by array offset into object.
521 // Kills:
522 // - Rindex
523 // Writes:
524 // - Rres: Address that corresponds to the array index if check was successful.
525 verify_oop(Rarray);
526 const Register Rlength = R0;
527 const Register RsxtIndex = Rtmp;
528 Label LisNull, LnotOOR;
530 // Array nullcheck
531 if (!ImplicitNullChecks) {
532 cmpdi(CCR0, Rarray, 0);
533 beq(CCR0, LisNull);
534 } else {
535 null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex);
536 }
538 // Rindex might contain garbage in upper bits (remember that we don't sign extend
539 // during integer arithmetic operations). So kill them and put value into same register
540 // where ArrayIndexOutOfBounds would expect the index in.
541 rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit
543 // Index check
544 lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray);
545 cmplw(CCR0, Rindex, Rlength);
546 sldi(RsxtIndex, RsxtIndex, index_shift);
547 blt(CCR0, LnotOOR);
548 // Index should be in R17_tos, array should be in R4_ARG2.
549 mr(R17_tos, Rindex);
550 mr(R4_ARG2, Rarray);
551 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
552 mtctr(Rtmp);
553 bctr();
555 if (!ImplicitNullChecks) {
556 bind(LisNull);
557 load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry);
558 mtctr(Rtmp);
559 bctr();
560 }
562 align(32, 16);
563 bind(LnotOOR);
565 // Calc address
566 add(Rres, RsxtIndex, Rarray);
567 }
569 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) {
570 // pop array
571 pop_ptr(array);
573 // check array
574 index_check_without_pop(array, index, index_shift, tmp, res);
575 }
577 void InterpreterMacroAssembler::get_const(Register Rdst) {
578 ld(Rdst, in_bytes(Method::const_offset()), R19_method);
579 }
581 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
582 get_const(Rdst);
583 ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
584 }
586 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
587 get_constant_pool(Rdst);
588 ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst);
589 }
591 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
592 get_constant_pool(Rcpool);
593 ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool);
594 }
596 // Unlock if synchronized method.
597 //
598 // Unlock the receiver if this is a synchronized method.
599 // Unlock any Java monitors from synchronized blocks.
600 //
601 // If there are locked Java monitors
602 // If throw_monitor_exception
603 // throws IllegalMonitorStateException
604 // Else if install_monitor_exception
605 // installs IllegalMonitorStateException
606 // Else
607 // no error processing
608 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
609 bool throw_monitor_exception,
610 bool install_monitor_exception) {
611 Label Lunlocked, Lno_unlock;
612 {
613 Register Rdo_not_unlock_flag = R11_scratch1;
614 Register Raccess_flags = R12_scratch2;
616 // Check if synchronized method or unlocking prevented by
617 // JavaThread::do_not_unlock_if_synchronized flag.
618 lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
619 lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method);
620 li(R0, 0);
621 stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag
623 push(state);
625 // Skip if we don't have to unlock.
626 rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0.
627 beq(CCR0, Lunlocked);
629 cmpwi(CCR0, Rdo_not_unlock_flag, 0);
630 bne(CCR0, Lno_unlock);
631 }
633 // Unlock
634 {
635 Register Rmonitor_base = R11_scratch1;
637 Label Lunlock;
638 // If it's still locked, everything is ok, unlock it.
639 ld(Rmonitor_base, 0, R1_SP);
640 addi(Rmonitor_base, Rmonitor_base, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
642 ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base);
643 cmpdi(CCR0, R0, 0);
644 bne(CCR0, Lunlock);
646 // If it's already unlocked, throw exception.
647 if (throw_monitor_exception) {
648 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
649 should_not_reach_here();
650 } else {
651 if (install_monitor_exception) {
652 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
653 b(Lunlocked);
654 }
655 }
657 bind(Lunlock);
658 unlock_object(Rmonitor_base);
659 }
661 // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not.
662 bind(Lunlocked);
663 {
664 Label Lexception, Lrestart;
665 Register Rcurrent_obj_addr = R11_scratch1;
666 const int delta = frame::interpreter_frame_monitor_size_in_bytes();
667 assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords");
669 bind(Lrestart);
670 // Set up search loop: Calc num of iterations.
671 {
672 Register Riterations = R12_scratch2;
673 Register Rmonitor_base = Rcurrent_obj_addr;
674 ld(Rmonitor_base, 0, R1_SP);
675 addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base
677 subf_(Riterations, R26_monitor, Rmonitor_base);
678 ble(CCR0, Lno_unlock);
680 addi(Rcurrent_obj_addr, Rmonitor_base, BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes());
681 // Check if any monitor is on stack, bail out if not
682 srdi(Riterations, Riterations, exact_log2(delta));
683 mtctr(Riterations);
684 }
686 // The search loop: Look for locked monitors.
687 {
688 const Register Rcurrent_obj = R0;
689 Label Lloop;
691 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
692 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
693 bind(Lloop);
695 // Check if current entry is used.
696 cmpdi(CCR0, Rcurrent_obj, 0);
697 bne(CCR0, Lexception);
698 // Preload next iteration's compare value.
699 ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
700 addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta);
701 bdnz(Lloop);
702 }
703 // Fell through: Everything's unlocked => finish.
704 b(Lno_unlock);
706 // An object is still locked => need to throw exception.
707 bind(Lexception);
708 if (throw_monitor_exception) {
709 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
710 should_not_reach_here();
711 } else {
712 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
713 // Unlock does not block, so don't have to worry about the frame.
714 Register Rmonitor_addr = R11_scratch1;
715 addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta);
716 unlock_object(Rmonitor_addr);
717 if (install_monitor_exception) {
718 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
719 }
720 b(Lrestart);
721 }
722 }
724 align(32, 12);
725 bind(Lno_unlock);
726 pop(state);
727 }
729 // Support function for remove_activation & Co.
730 void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, Register Rscratch1, Register Rscratch2) {
731 // Pop interpreter frame.
732 ld(Rscratch1, 0, R1_SP); // *SP
733 ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp
734 ld(Rscratch2, 0, Rscratch1); // **SP
735 #ifdef ASSERT
736 {
737 Label Lok;
738 ld(R0, _ijava_state_neg(ijava_reserved), Rscratch1);
739 cmpdi(CCR0, R0, 0x5afe);
740 beq(CCR0, Lok);
741 stop("frame corrupted (remove activation)", 0x5afe);
742 bind(Lok);
743 }
744 #endif
745 if (return_pc!=noreg) {
746 ld(return_pc, _abi(lr), Rscratch1); // LR
747 }
749 // Merge top frames.
750 subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP
751 stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP
752 }
754 // Remove activation.
755 //
756 // Unlock the receiver if this is a synchronized method.
757 // Unlock any Java monitors from synchronized blocks.
758 // Remove the activation from the stack.
759 //
760 // If there are locked Java monitors
761 // If throw_monitor_exception
762 // throws IllegalMonitorStateException
763 // Else if install_monitor_exception
764 // installs IllegalMonitorStateException
765 // Else
766 // no error processing
767 void InterpreterMacroAssembler::remove_activation(TosState state,
768 bool throw_monitor_exception,
769 bool install_monitor_exception) {
770 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
772 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
773 notify_method_exit(false, state, NotifyJVMTI, true);
775 verify_oop(R17_tos, state);
776 verify_thread();
778 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
779 mtlr(R0);
780 }
782 #endif // !CC_INTERP
784 // Lock object
785 //
786 // Registers alive
787 // monitor - Address of the BasicObjectLock to be used for locking,
788 // which must be initialized with the object to lock.
789 // object - Address of the object to be locked.
790 //
791 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
792 if (UseHeavyMonitors) {
793 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
794 monitor, /*check_for_exceptions=*/true CC_INTERP_ONLY(&& false));
795 } else {
796 // template code:
797 //
798 // markOop displaced_header = obj->mark().set_unlocked();
799 // monitor->lock()->set_displaced_header(displaced_header);
800 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
801 // // We stored the monitor address into the object's mark word.
802 // } else if (THREAD->is_lock_owned((address)displaced_header))
803 // // Simple recursive case.
804 // monitor->lock()->set_displaced_header(NULL);
805 // } else {
806 // // Slow path.
807 // InterpreterRuntime::monitorenter(THREAD, monitor);
808 // }
810 const Register displaced_header = R7_ARG5;
811 const Register object_mark_addr = R8_ARG6;
812 const Register current_header = R9_ARG7;
813 const Register tmp = R10_ARG8;
815 Label done;
816 Label cas_failed, slow_case;
818 assert_different_registers(displaced_header, object_mark_addr, current_header, tmp);
820 // markOop displaced_header = obj->mark().set_unlocked();
822 // Load markOop from object into displaced_header.
823 ld(displaced_header, oopDesc::mark_offset_in_bytes(), object);
825 if (UseBiasedLocking) {
826 biased_locking_enter(CCR0, object, displaced_header, tmp, current_header, done, &slow_case);
827 }
829 // Set displaced_header to be (markOop of object | UNLOCK_VALUE).
830 ori(displaced_header, displaced_header, markOopDesc::unlocked_value);
832 // monitor->lock()->set_displaced_header(displaced_header);
834 // Initialize the box (Must happen before we update the object mark!).
835 std(displaced_header, BasicObjectLock::lock_offset_in_bytes() +
836 BasicLock::displaced_header_offset_in_bytes(), monitor);
838 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
840 // Store stack address of the BasicObjectLock (this is monitor) into object.
841 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
843 // Must fence, otherwise, preceding store(s) may float below cmpxchg.
844 // CmpxchgX sets CCR0 to cmpX(current, displaced).
845 fence(); // TODO: replace by MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq ?
846 cmpxchgd(/*flag=*/CCR0,
847 /*current_value=*/current_header,
848 /*compare_value=*/displaced_header, /*exchange_value=*/monitor,
849 /*where=*/object_mark_addr,
850 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq,
851 MacroAssembler::cmpxchgx_hint_acquire_lock(),
852 noreg,
853 &cas_failed);
855 // If the compare-and-exchange succeeded, then we found an unlocked
856 // object and we have now locked it.
857 b(done);
858 bind(cas_failed);
860 // } else if (THREAD->is_lock_owned((address)displaced_header))
861 // // Simple recursive case.
862 // monitor->lock()->set_displaced_header(NULL);
864 // We did not see an unlocked object so try the fast recursive case.
866 // Check if owner is self by comparing the value in the markOop of object
867 // (current_header) with the stack pointer.
868 sub(current_header, current_header, R1_SP);
870 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
871 load_const_optimized(tmp,
872 (address) (~(os::vm_page_size()-1) |
873 markOopDesc::lock_mask_in_place));
875 and_(R0/*==0?*/, current_header, tmp);
876 // If condition is true we are done and hence we can store 0 in the displaced
877 // header indicating it is a recursive lock.
878 bne(CCR0, slow_case);
879 release();
880 std(R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() +
881 BasicLock::displaced_header_offset_in_bytes(), monitor);
882 b(done);
884 // } else {
885 // // Slow path.
886 // InterpreterRuntime::monitorenter(THREAD, monitor);
888 // None of the above fast optimizations worked so we have to get into the
889 // slow case of monitor enter.
890 bind(slow_case);
891 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
892 monitor, /*check_for_exceptions=*/true CC_INTERP_ONLY(&& false));
893 // }
894 align(32, 12);
895 bind(done);
896 }
897 }
899 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
900 //
901 // Registers alive
902 // monitor - Address of the BasicObjectLock to be used for locking,
903 // which must be initialized with the object to lock.
904 //
905 // Throw IllegalMonitorException if object is not locked by current thread.
906 void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) {
907 if (UseHeavyMonitors) {
908 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
909 monitor, check_for_exceptions CC_INTERP_ONLY(&& false));
910 } else {
912 // template code:
913 //
914 // if ((displaced_header = monitor->displaced_header()) == NULL) {
915 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL.
916 // monitor->set_obj(NULL);
917 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) {
918 // // We swapped the unlocked mark in displaced_header into the object's mark word.
919 // monitor->set_obj(NULL);
920 // } else {
921 // // Slow path.
922 // InterpreterRuntime::monitorexit(THREAD, monitor);
923 // }
925 const Register object = R7_ARG5;
926 const Register displaced_header = R8_ARG6;
927 const Register object_mark_addr = R9_ARG7;
928 const Register current_header = R10_ARG8;
930 Label free_slot;
931 Label slow_case;
933 assert_different_registers(object, displaced_header, object_mark_addr, current_header);
935 if (UseBiasedLocking) {
936 // The object address from the monitor is in object.
937 ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor);
938 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0");
939 biased_locking_exit(CCR0, object, displaced_header, free_slot);
940 }
942 // Test first if we are in the fast recursive case.
943 ld(displaced_header, BasicObjectLock::lock_offset_in_bytes() +
944 BasicLock::displaced_header_offset_in_bytes(), monitor);
946 // If the displaced header is zero, we have a recursive unlock.
947 cmpdi(CCR0, displaced_header, 0);
948 beq(CCR0, free_slot); // recursive unlock
950 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) {
951 // // We swapped the unlocked mark in displaced_header into the object's mark word.
952 // monitor->set_obj(NULL);
954 // If we still have a lightweight lock, unlock the object and be done.
956 // The object address from the monitor is in object.
957 if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); }
958 addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes());
960 // We have the displaced header in displaced_header. If the lock is still
961 // lightweight, it will contain the monitor address and we'll store the
962 // displaced header back into the object's mark word.
963 // CmpxchgX sets CCR0 to cmpX(current, monitor).
964 cmpxchgd(/*flag=*/CCR0,
965 /*current_value=*/current_header,
966 /*compare_value=*/monitor, /*exchange_value=*/displaced_header,
967 /*where=*/object_mark_addr,
968 MacroAssembler::MemBarRel,
969 MacroAssembler::cmpxchgx_hint_release_lock(),
970 noreg,
971 &slow_case);
972 b(free_slot);
974 // } else {
975 // // Slow path.
976 // InterpreterRuntime::monitorexit(THREAD, monitor);
978 // The lock has been converted into a heavy lock and hence
979 // we need to get into the slow case.
980 bind(slow_case);
981 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
982 monitor, check_for_exceptions CC_INTERP_ONLY(&& false));
983 // }
985 Label done;
986 b(done); // Monitor register may be overwritten! Runtime has already freed the slot.
988 // Exchange worked, do monitor->set_obj(NULL);
989 align(32, 12);
990 bind(free_slot);
991 li(R0, 0);
992 std(R0, BasicObjectLock::obj_offset_in_bytes(), monitor);
993 bind(done);
994 }
995 }
997 #ifndef CC_INTERP
999 // Load compiled (i2c) or interpreter entry when calling from interpreted and
1000 // do the call. Centralized so that all interpreter calls will do the same actions.
1001 // If jvmti single stepping is on for a thread we must not call compiled code.
1002 //
1003 // Input:
1004 // - Rtarget_method: method to call
1005 // - Rret_addr: return address
1006 // - 2 scratch regs
1007 //
1008 void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, Register Rscratch1, Register Rscratch2) {
1009 assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr);
1010 // Assume we want to go compiled if available.
1011 const Register Rtarget_addr = Rscratch1;
1012 const Register Rinterp_only = Rscratch2;
1014 ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method);
1016 if (JvmtiExport::can_post_interpreter_events()) {
1017 lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
1019 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
1020 // compiled code in threads for which the event is enabled. Check here for
1021 // interp_only_mode if these events CAN be enabled.
1022 Label done;
1023 verify_thread();
1024 cmpwi(CCR0, Rinterp_only, 0);
1025 beq(CCR0, done);
1026 ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method);
1027 align(32, 12);
1028 bind(done);
1029 }
1031 #ifdef ASSERT
1032 {
1033 Label Lok;
1034 cmpdi(CCR0, Rtarget_addr, 0);
1035 bne(CCR0, Lok);
1036 stop("null entry point");
1037 bind(Lok);
1038 }
1039 #endif // ASSERT
1041 mr(R21_sender_SP, R1_SP);
1043 // Calc a precise SP for the call. The SP value we calculated in
1044 // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space
1045 // if esp is not max. Also, the i2c adapter extends the stack space without restoring
1046 // our pre-calced value, so repeating calls via i2c would result in stack overflow.
1047 // Since esp already points to an empty slot, we just have to sub 1 additional slot
1048 // to meet the abi scratch requirements.
1049 // The max_stack pointer will get restored by means of the GR_Lmax_stack local in
1050 // the return entry of the interpreter.
1051 addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size);
1052 clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address
1053 resize_frame_absolute(Rscratch2, Rscratch2, R0);
1055 mr_if_needed(R19_method, Rtarget_method);
1056 mtctr(Rtarget_addr);
1057 mtlr(Rret_addr);
1059 save_interpreter_state(Rscratch2);
1060 #ifdef ASSERT
1061 ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp
1062 cmpd(CCR0, R21_sender_SP, Rscratch1);
1063 asm_assert_eq("top_frame_sp incorrect", 0x951);
1064 #endif
1066 bctr();
1067 }
1069 // Set the method data pointer for the current bcp.
1070 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1071 assert(ProfileInterpreter, "must be profiling interpreter");
1072 Label get_continue;
1073 ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method);
1074 test_method_data_pointer(get_continue);
1075 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp);
1077 addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1078 add(R28_mdx, R28_mdx, R3_RET);
1079 bind(get_continue);
1080 }
1082 // Test ImethodDataPtr. If it is null, continue at the specified label.
1083 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1084 assert(ProfileInterpreter, "must be profiling interpreter");
1085 cmpdi(CCR0, R28_mdx, 0);
1086 beq(CCR0, zero_continue);
1087 }
1089 void InterpreterMacroAssembler::verify_method_data_pointer() {
1090 assert(ProfileInterpreter, "must be profiling interpreter");
1091 #ifdef ASSERT
1092 Label verify_continue;
1093 test_method_data_pointer(verify_continue);
1095 // If the mdp is valid, it will point to a DataLayout header which is
1096 // consistent with the bcp. The converse is highly probable also.
1097 lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx);
1098 ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method);
1099 addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1100 add(R11_scratch1, R12_scratch2, R12_scratch2);
1101 cmpd(CCR0, R11_scratch1, R14_bcp);
1102 beq(CCR0, verify_continue);
1104 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx);
1106 bind(verify_continue);
1107 #endif
1108 }
1110 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1111 Register Rscratch,
1112 Label &profile_continue) {
1113 assert(ProfileInterpreter, "must be profiling interpreter");
1114 // Control will flow to "profile_continue" if the counter is less than the
1115 // limit or if we call profile_method().
1116 Label done;
1118 // If no method data exists, and the counter is high enough, make one.
1119 int ipl_offs = load_const_optimized(Rscratch, &InvocationCounter::InterpreterProfileLimit, R0, true);
1120 lwz(Rscratch, ipl_offs, Rscratch);
1122 cmpdi(CCR0, R28_mdx, 0);
1123 // Test to see if we should create a method data oop.
1124 cmpd(CCR1, Rscratch /* InterpreterProfileLimit */, invocation_count);
1125 bne(CCR0, done);
1126 bge(CCR1, profile_continue);
1128 // Build it now.
1129 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1130 set_method_data_pointer_for_bcp();
1131 b(profile_continue);
1133 align(32, 12);
1134 bind(done);
1135 }
1137 void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register branch_bcp, Register Rtmp) {
1138 assert_different_registers(backedge_count, Rtmp, branch_bcp);
1139 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
1141 Label did_not_overflow;
1142 Label overflow_with_error;
1144 int ibbl_offs = load_const_optimized(Rtmp, &InvocationCounter::InterpreterBackwardBranchLimit, R0, true);
1145 lwz(Rtmp, ibbl_offs, Rtmp);
1146 cmpw(CCR0, backedge_count, Rtmp);
1148 blt(CCR0, did_not_overflow);
1150 // When ProfileInterpreter is on, the backedge_count comes from the
1151 // methodDataOop, which value does not get reset on the call to
1152 // frequency_counter_overflow(). To avoid excessive calls to the overflow
1153 // routine while the method is being compiled, add a second test to make sure
1154 // the overflow function is called only once every overflow_frequency.
1155 if (ProfileInterpreter) {
1156 const int overflow_frequency = 1024;
1157 li(Rtmp, overflow_frequency-1);
1158 andr(Rtmp, Rtmp, backedge_count);
1159 cmpwi(CCR0, Rtmp, 0);
1160 bne(CCR0, did_not_overflow);
1161 }
1163 // Overflow in loop, pass branch bytecode.
1164 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, true);
1166 // Was an OSR adapter generated?
1167 // O0 = osr nmethod
1168 cmpdi(CCR0, R3_RET, 0);
1169 beq(CCR0, overflow_with_error);
1171 // Has the nmethod been invalidated already?
1172 lwz(Rtmp, nmethod::entry_bci_offset(), R3_RET);
1173 cmpwi(CCR0, Rtmp, InvalidOSREntryBci);
1174 beq(CCR0, overflow_with_error);
1176 // Migrate the interpreter frame off of the stack.
1177 // We can use all registers because we will not return to interpreter from this point.
1179 // Save nmethod.
1180 const Register osr_nmethod = R31;
1181 mr(osr_nmethod, R3_RET);
1182 set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1183 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1184 reset_last_Java_frame();
1185 // OSR buffer is in ARG1
1187 // Remove the interpreter frame.
1188 merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1190 // Jump to the osr code.
1191 ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1192 mtlr(R0);
1193 mtctr(R11_scratch1);
1194 bctr();
1196 align(32, 12);
1197 bind(overflow_with_error);
1198 bind(did_not_overflow);
1199 }
1201 // Store a value at some constant offset from the method data pointer.
1202 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1203 assert(ProfileInterpreter, "must be profiling interpreter");
1205 std(value, constant, R28_mdx);
1206 }
1208 // Increment the value at some constant offset from the method data pointer.
1209 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1210 Register counter_addr,
1211 Register Rbumped_count,
1212 bool decrement) {
1213 // Locate the counter at a fixed offset from the mdp:
1214 addi(counter_addr, R28_mdx, constant);
1215 increment_mdp_data_at(counter_addr, Rbumped_count, decrement);
1216 }
1218 // Increment the value at some non-fixed (reg + constant) offset from
1219 // the method data pointer.
1220 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1221 int constant,
1222 Register scratch,
1223 Register Rbumped_count,
1224 bool decrement) {
1225 // Add the constant to reg to get the offset.
1226 add(scratch, R28_mdx, reg);
1227 // Then calculate the counter address.
1228 addi(scratch, scratch, constant);
1229 increment_mdp_data_at(scratch, Rbumped_count, decrement);
1230 }
1232 void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr,
1233 Register Rbumped_count,
1234 bool decrement) {
1235 assert(ProfileInterpreter, "must be profiling interpreter");
1237 // Load the counter.
1238 ld(Rbumped_count, 0, counter_addr);
1240 if (decrement) {
1241 // Decrement the register. Set condition codes.
1242 addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment);
1243 // Store the decremented counter, if it is still negative.
1244 std(Rbumped_count, 0, counter_addr);
1245 // Note: add/sub overflow check are not ported, since 64 bit
1246 // calculation should never overflow.
1247 } else {
1248 // Increment the register. Set carry flag.
1249 addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment);
1250 // Store the incremented counter.
1251 std(Rbumped_count, 0, counter_addr);
1252 }
1253 }
1255 // Set a flag value at the current method data pointer position.
1256 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1257 Register scratch) {
1258 assert(ProfileInterpreter, "must be profiling interpreter");
1259 // Load the data header.
1260 lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1261 // Set the flag.
1262 ori(scratch, scratch, flag_constant);
1263 // Store the modified header.
1264 stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx);
1265 }
1267 // Test the location at some offset from the method data pointer.
1268 // If it is not equal to value, branch to the not_equal_continue Label.
1269 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1270 Register value,
1271 Label& not_equal_continue,
1272 Register test_out) {
1273 assert(ProfileInterpreter, "must be profiling interpreter");
1275 ld(test_out, offset, R28_mdx);
1276 cmpd(CCR0, value, test_out);
1277 bne(CCR0, not_equal_continue);
1278 }
1280 // Update the method data pointer by the displacement located at some fixed
1281 // offset from the method data pointer.
1282 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1283 Register scratch) {
1284 assert(ProfileInterpreter, "must be profiling interpreter");
1286 ld(scratch, offset_of_disp, R28_mdx);
1287 add(R28_mdx, scratch, R28_mdx);
1288 }
1290 // Update the method data pointer by the displacement located at the
1291 // offset (reg + offset_of_disp).
1292 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1293 int offset_of_disp,
1294 Register scratch) {
1295 assert(ProfileInterpreter, "must be profiling interpreter");
1297 add(scratch, reg, R28_mdx);
1298 ld(scratch, offset_of_disp, scratch);
1299 add(R28_mdx, scratch, R28_mdx);
1300 }
1302 // Update the method data pointer by a simple constant displacement.
1303 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1304 assert(ProfileInterpreter, "must be profiling interpreter");
1305 addi(R28_mdx, R28_mdx, constant);
1306 }
1308 // Update the method data pointer for a _ret bytecode whose target
1309 // was not among our cached targets.
1310 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1311 Register return_bci) {
1312 assert(ProfileInterpreter, "must be profiling interpreter");
1314 push(state);
1315 assert(return_bci->is_nonvolatile(), "need to protect return_bci");
1316 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1317 pop(state);
1318 }
1320 // Increments the backedge counter.
1321 // Returns backedge counter + invocation counter in Rdst.
1322 void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst,
1323 const Register Rtmp1, Register Rscratch) {
1324 assert(UseCompiler, "incrementing must be useful");
1325 assert_different_registers(Rdst, Rtmp1);
1326 const Register invocation_counter = Rtmp1;
1327 const Register counter = Rdst;
1328 // TODO ppc port assert(4 == InvocationCounter::sz_counter(), "unexpected field size.");
1330 // Load backedge counter.
1331 lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1332 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1333 // Load invocation counter.
1334 lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) +
1335 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1337 // Add the delta to the backedge counter.
1338 addi(counter, counter, InvocationCounter::count_increment);
1340 // Mask the invocation counter.
1341 li(Rscratch, InvocationCounter::count_mask_value);
1342 andr(invocation_counter, invocation_counter, Rscratch);
1344 // Store new counter value.
1345 stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) +
1346 in_bytes(InvocationCounter::counter_offset()), Rcounters);
1347 // Return invocation counter + backedge counter.
1348 add(counter, counter, invocation_counter);
1349 }
1351 // Count a taken branch in the bytecodes.
1352 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1353 if (ProfileInterpreter) {
1354 Label profile_continue;
1356 // If no method data exists, go to profile_continue.
1357 test_method_data_pointer(profile_continue);
1359 // We are taking a branch. Increment the taken count.
1360 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count);
1362 // The method data pointer needs to be updated to reflect the new target.
1363 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1364 bind (profile_continue);
1365 }
1366 }
1368 // Count a not-taken branch in the bytecodes.
1369 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) {
1370 if (ProfileInterpreter) {
1371 Label profile_continue;
1373 // If no method data exists, go to profile_continue.
1374 test_method_data_pointer(profile_continue);
1376 // We are taking a branch. Increment the not taken count.
1377 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2);
1379 // The method data pointer needs to be updated to correspond to the
1380 // next bytecode.
1381 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1382 bind (profile_continue);
1383 }
1384 }
1386 // Count a non-virtual call in the bytecodes.
1387 void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) {
1388 if (ProfileInterpreter) {
1389 Label profile_continue;
1391 // If no method data exists, go to profile_continue.
1392 test_method_data_pointer(profile_continue);
1394 // We are making a call. Increment the count.
1395 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1397 // The method data pointer needs to be updated to reflect the new target.
1398 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1399 bind (profile_continue);
1400 }
1401 }
1403 // Count a final call in the bytecodes.
1404 void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) {
1405 if (ProfileInterpreter) {
1406 Label profile_continue;
1408 // If no method data exists, go to profile_continue.
1409 test_method_data_pointer(profile_continue);
1411 // We are making a call. Increment the count.
1412 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1414 // The method data pointer needs to be updated to reflect the new target.
1415 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1416 bind (profile_continue);
1417 }
1418 }
1420 // Count a virtual call in the bytecodes.
1421 void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver,
1422 Register Rscratch1,
1423 Register Rscratch2,
1424 bool receiver_can_be_null) {
1425 if (!ProfileInterpreter) { return; }
1426 Label profile_continue;
1428 // If no method data exists, go to profile_continue.
1429 test_method_data_pointer(profile_continue);
1431 Label skip_receiver_profile;
1432 if (receiver_can_be_null) {
1433 Label not_null;
1434 cmpdi(CCR0, Rreceiver, 0);
1435 bne(CCR0, not_null);
1436 // We are making a call. Increment the count for null receiver.
1437 increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2);
1438 b(skip_receiver_profile);
1439 bind(not_null);
1440 }
1442 // Record the receiver type.
1443 record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true);
1444 bind(skip_receiver_profile);
1446 // The method data pointer needs to be updated to reflect the new target.
1447 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1448 bind (profile_continue);
1449 }
1451 void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) {
1452 if (ProfileInterpreter) {
1453 Label profile_continue;
1455 // If no method data exists, go to profile_continue.
1456 test_method_data_pointer(profile_continue);
1458 int mdp_delta = in_bytes(BitData::bit_data_size());
1459 if (TypeProfileCasts) {
1460 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1462 // Record the object type.
1463 record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false);
1464 }
1466 // The method data pointer needs to be updated.
1467 update_mdp_by_constant(mdp_delta);
1469 bind (profile_continue);
1470 }
1471 }
1473 void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) {
1474 if (ProfileInterpreter && TypeProfileCasts) {
1475 Label profile_continue;
1477 // If no method data exists, go to profile_continue.
1478 test_method_data_pointer(profile_continue);
1480 int count_offset = in_bytes(CounterData::count_offset());
1481 // Back up the address, since we have already bumped the mdp.
1482 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1484 // *Decrement* the counter. We expect to see zero or small negatives.
1485 increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true);
1487 bind (profile_continue);
1488 }
1489 }
1491 // Count a ret in the bytecodes.
1492 void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, Register scratch1, Register scratch2) {
1493 if (ProfileInterpreter) {
1494 Label profile_continue;
1495 uint row;
1497 // If no method data exists, go to profile_continue.
1498 test_method_data_pointer(profile_continue);
1500 // Update the total ret count.
1501 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 );
1503 for (row = 0; row < RetData::row_limit(); row++) {
1504 Label next_test;
1506 // See if return_bci is equal to bci[n]:
1507 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1);
1509 // return_bci is equal to bci[n]. Increment the count.
1510 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2);
1512 // The method data pointer needs to be updated to reflect the new target.
1513 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1);
1514 b(profile_continue);
1515 bind(next_test);
1516 }
1518 update_mdp_for_ret(state, return_bci);
1520 bind (profile_continue);
1521 }
1522 }
1524 // Count the default case of a switch construct.
1525 void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) {
1526 if (ProfileInterpreter) {
1527 Label profile_continue;
1529 // If no method data exists, go to profile_continue.
1530 test_method_data_pointer(profile_continue);
1532 // Update the default case count
1533 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1534 scratch1, scratch2);
1536 // The method data pointer needs to be updated.
1537 update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()),
1538 scratch1);
1540 bind (profile_continue);
1541 }
1542 }
1544 // Count the index'th case of a switch construct.
1545 void InterpreterMacroAssembler::profile_switch_case(Register index,
1546 Register scratch1,
1547 Register scratch2,
1548 Register scratch3) {
1549 if (ProfileInterpreter) {
1550 assert_different_registers(index, scratch1, scratch2, scratch3);
1551 Label profile_continue;
1553 // If no method data exists, go to profile_continue.
1554 test_method_data_pointer(profile_continue);
1556 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes().
1557 li(scratch3, in_bytes(MultiBranchData::case_array_offset()));
1559 assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works");
1560 sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1561 add(scratch1, scratch1, scratch3);
1563 // Update the case count.
1564 increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3);
1566 // The method data pointer needs to be updated.
1567 update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2);
1569 bind (profile_continue);
1570 }
1571 }
1573 void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) {
1574 if (ProfileInterpreter) {
1575 assert_different_registers(Rscratch1, Rscratch2);
1576 Label profile_continue;
1578 // If no method data exists, go to profile_continue.
1579 test_method_data_pointer(profile_continue);
1581 set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1);
1583 // The method data pointer needs to be updated.
1584 int mdp_delta = in_bytes(BitData::bit_data_size());
1585 if (TypeProfileCasts) {
1586 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1587 }
1588 update_mdp_by_constant(mdp_delta);
1590 bind (profile_continue);
1591 }
1592 }
1594 void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver,
1595 Register Rscratch1, Register Rscratch2,
1596 bool is_virtual_call) {
1597 assert(ProfileInterpreter, "must be profiling");
1598 assert_different_registers(Rreceiver, Rscratch1, Rscratch2);
1600 Label done;
1601 record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call);
1602 bind (done);
1603 }
1605 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1606 Register receiver, Register scratch1, Register scratch2,
1607 int start_row, Label& done, bool is_virtual_call) {
1608 if (TypeProfileWidth == 0) {
1609 if (is_virtual_call) {
1610 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1611 }
1612 return;
1613 }
1615 int last_row = VirtualCallData::row_limit() - 1;
1616 assert(start_row <= last_row, "must be work left to do");
1617 // Test this row for both the receiver and for null.
1618 // Take any of three different outcomes:
1619 // 1. found receiver => increment count and goto done
1620 // 2. found null => keep looking for case 1, maybe allocate this cell
1621 // 3. found something else => keep looking for cases 1 and 2
1622 // Case 3 is handled by a recursive call.
1623 for (int row = start_row; row <= last_row; row++) {
1624 Label next_test;
1625 bool test_for_null_also = (row == start_row);
1627 // See if the receiver is receiver[n].
1628 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1629 test_mdp_data_at(recvr_offset, receiver, next_test, scratch1);
1630 // delayed()->tst(scratch);
1632 // The receiver is receiver[n]. Increment count[n].
1633 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1634 increment_mdp_data_at(count_offset, scratch1, scratch2);
1635 b(done);
1636 bind(next_test);
1638 if (test_for_null_also) {
1639 Label found_null;
1640 // Failed the equality check on receiver[n]... Test for null.
1641 if (start_row == last_row) {
1642 // The only thing left to do is handle the null case.
1643 if (is_virtual_call) {
1644 // Scratch1 contains test_out from test_mdp_data_at.
1645 cmpdi(CCR0, scratch1, 0);
1646 beq(CCR0, found_null);
1647 // Receiver did not match any saved receiver and there is no empty row for it.
1648 // Increment total counter to indicate polymorphic case.
1649 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2);
1650 b(done);
1651 bind(found_null);
1652 } else {
1653 cmpdi(CCR0, scratch1, 0);
1654 bne(CCR0, done);
1655 }
1656 break;
1657 }
1658 // Since null is rare, make it be the branch-taken case.
1659 cmpdi(CCR0, scratch1, 0);
1660 beq(CCR0, found_null);
1662 // Put all the "Case 3" tests here.
1663 record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call);
1665 // Found a null. Keep searching for a matching receiver,
1666 // but remember that this is an empty (unused) slot.
1667 bind(found_null);
1668 }
1669 }
1671 // In the fall-through case, we found no matching receiver, but we
1672 // observed the receiver[start_row] is NULL.
1674 // Fill in the receiver field and increment the count.
1675 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1676 set_mdp_data_at(recvr_offset, receiver);
1677 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1678 li(scratch1, DataLayout::counter_increment);
1679 set_mdp_data_at(count_offset, scratch1);
1680 if (start_row > 0) {
1681 b(done);
1682 }
1683 }
1685 // Argument and return type profilig.
1686 // kills: tmp, tmp2, R0, CR0, CR1
1687 void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base,
1688 RegisterOrConstant mdo_addr_offs, Register tmp, Register tmp2) {
1689 Label do_nothing, do_update;
1691 // tmp2 = obj is allowed
1692 assert_different_registers(obj, mdo_addr_base, tmp, R0);
1693 assert_different_registers(tmp2, mdo_addr_base, tmp, R0);
1694 const Register klass = tmp2;
1696 verify_oop(obj);
1698 ld(tmp, mdo_addr_offs, mdo_addr_base);
1700 // Set null_seen if obj is 0.
1701 cmpdi(CCR0, obj, 0);
1702 ori(R0, tmp, TypeEntries::null_seen);
1703 beq(CCR0, do_update);
1705 load_klass(klass, obj);
1707 clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
1708 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
1709 cmpd(CCR1, R0, klass);
1710 // Klass seen before, nothing to do (regardless of unknown bit).
1711 //beq(CCR1, do_nothing);
1713 andi_(R0, klass, TypeEntries::type_unknown);
1714 // Already unknown. Nothing to do anymore.
1715 //bne(CCR0, do_nothing);
1716 crorc(/*CCR0 eq*/2, /*CCR1 eq*/4+2, /*CCR0 eq*/2); // cr0 eq = cr1 eq or cr0 ne
1717 beq(CCR0, do_nothing);
1719 clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
1720 orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1721 beq(CCR0, do_update); // First time here. Set profile type.
1723 // Different than before. Cannot keep accurate profile.
1724 ori(R0, tmp, TypeEntries::type_unknown);
1726 bind(do_update);
1727 // update profile
1728 std(R0, mdo_addr_offs, mdo_addr_base);
1730 align(32, 12);
1731 bind(do_nothing);
1732 }
1734 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) {
1735 if (!ProfileInterpreter) {
1736 return;
1737 }
1739 assert_different_registers(callee, tmp1, tmp2, R28_mdx);
1741 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1742 Label profile_continue;
1744 test_method_data_pointer(profile_continue);
1746 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1748 lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx);
1749 cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1750 bne(CCR0, profile_continue);
1752 if (MethodData::profile_arguments()) {
1753 Label done;
1754 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1755 add(R28_mdx, off_to_args, R28_mdx);
1757 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1758 if (i > 0 || MethodData::profile_return()) {
1759 // If return value type is profiled we may have no argument to profile.
1760 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1761 cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count());
1762 addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count());
1763 blt(CCR0, done);
1764 }
1765 ld(tmp1, in_bytes(Method::const_offset()), callee);
1766 lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1);
1767 // Stack offset o (zero based) from the start of the argument
1768 // list, for n arguments translates into offset n - o - 1 from
1769 // the end of the argument list. But there's an extra slot at
1770 // the top of the stack. So the offset is n - o from Lesp.
1771 ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx);
1772 subf(tmp1, tmp2, tmp1);
1774 sldi(tmp1, tmp1, Interpreter::logStackElementSize);
1775 ldx(tmp1, tmp1, R15_esp);
1777 profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1);
1779 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1780 addi(R28_mdx, R28_mdx, to_add);
1781 off_to_args += to_add;
1782 }
1784 if (MethodData::profile_return()) {
1785 ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
1786 addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1787 }
1789 bind(done);
1791 if (MethodData::profile_return()) {
1792 // We're right after the type profile for the last
1793 // argument. tmp1 is the number of cells left in the
1794 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1795 // if there's a return to profile.
1796 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1797 sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size));
1798 add(R28_mdx, tmp1, R28_mdx);
1799 }
1800 } else {
1801 assert(MethodData::profile_return(), "either profile call args or call ret");
1802 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
1803 }
1805 // Mdp points right after the end of the
1806 // CallTypeData/VirtualCallTypeData, right after the cells for the
1807 // return value type if there's one.
1808 align(32, 12);
1809 bind(profile_continue);
1810 }
1811 }
1813 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
1814 assert_different_registers(ret, tmp1, tmp2);
1815 if (ProfileInterpreter && MethodData::profile_return()) {
1816 Label profile_continue;
1818 test_method_data_pointer(profile_continue);
1820 if (MethodData::profile_return_jsr292_only()) {
1821 // If we don't profile all invoke bytecodes we must make sure
1822 // it's a bytecode we indeed profile. We can't go back to the
1823 // begining of the ProfileData we intend to update to check its
1824 // type because we're right after it and we don't known its
1825 // length.
1826 lbz(tmp1, 0, R14_bcp);
1827 lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method);
1828 cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic);
1829 cmpwi(CCR1, tmp1, Bytecodes::_invokehandle);
1830 cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
1831 cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm);
1832 cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
1833 bne(CCR0, profile_continue);
1834 }
1836 profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2);
1838 align(32, 12);
1839 bind(profile_continue);
1840 }
1841 }
1843 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
1844 if (ProfileInterpreter && MethodData::profile_parameters()) {
1845 Label profile_continue, done;
1847 test_method_data_pointer(profile_continue);
1849 // Load the offset of the area within the MDO used for
1850 // parameters. If it's negative we're not profiling any parameters.
1851 lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx);
1852 cmpwi(CCR0, tmp1, 0);
1853 blt(CCR0, profile_continue);
1855 // Compute a pointer to the area for parameters from the offset
1856 // and move the pointer to the slot for the last
1857 // parameters. Collect profiling from last parameter down.
1858 // mdo start + parameters offset + array length - 1
1860 // Pointer to the parameter area in the MDO.
1861 const Register mdp = tmp1;
1862 add(mdp, tmp1, R28_mdx);
1864 // Pffset of the current profile entry to update.
1865 const Register entry_offset = tmp2;
1866 // entry_offset = array len in number of cells
1867 ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp);
1869 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1870 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
1872 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
1873 addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size));
1874 // entry_offset in bytes
1875 sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1877 Label loop;
1878 align(32, 12);
1879 bind(loop);
1881 // Load offset on the stack from the slot for this parameter.
1882 ld(tmp3, entry_offset, mdp);
1883 sldi(tmp3, tmp3, Interpreter::logStackElementSize);
1884 neg(tmp3, tmp3);
1885 // Read the parameter from the local area.
1886 ldx(tmp3, tmp3, R18_locals);
1888 // Make entry_offset now point to the type field for this parameter.
1889 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1890 assert(type_base > off_base, "unexpected");
1891 addi(entry_offset, entry_offset, type_base - off_base);
1893 // Profile the parameter.
1894 profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3);
1896 // Go to next parameter.
1897 int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base);
1898 cmpdi(CCR0, entry_offset, off_base + delta);
1899 addi(entry_offset, entry_offset, -delta);
1900 bge(CCR0, loop);
1902 align(32, 12);
1903 bind(profile_continue);
1904 }
1905 }
1907 // Add a InterpMonitorElem to stack (see frame_sparc.hpp).
1908 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) {
1910 // Very-local scratch registers.
1911 const Register esp = Rtemp1;
1912 const Register slot = Rtemp2;
1914 // Extracted monitor_size.
1915 int monitor_size = frame::interpreter_frame_monitor_size_in_bytes();
1916 assert(Assembler::is_aligned((unsigned int)monitor_size,
1917 (unsigned int)frame::alignment_in_bytes),
1918 "size of a monitor must respect alignment of SP");
1920 resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor
1921 std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp
1923 // Shuffle expression stack down. Recall that stack_base points
1924 // just above the new expression stack bottom. Old_tos and new_tos
1925 // are used to scan thru the old and new expression stacks.
1926 if (!stack_is_empty) {
1927 Label copy_slot, copy_slot_finished;
1928 const Register n_slots = slot;
1930 addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack).
1931 subf(n_slots, esp, R26_monitor);
1932 srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy.
1933 assert(LogBytesPerWord == 3, "conflicts assembler instructions");
1934 beq(CCR0, copy_slot_finished); // Nothing to copy.
1936 mtctr(n_slots);
1938 // loop
1939 bind(copy_slot);
1940 ld(slot, 0, esp); // Move expression stack down.
1941 std(slot, -monitor_size, esp); // distance = monitor_size
1942 addi(esp, esp, BytesPerWord);
1943 bdnz(copy_slot);
1945 bind(copy_slot_finished);
1946 }
1948 addi(R15_esp, R15_esp, -monitor_size);
1949 addi(R26_monitor, R26_monitor, -monitor_size);
1951 // Restart interpreter
1952 }
1954 // ============================================================================
1955 // Java locals access
1957 // Load a local variable at index in Rindex into register Rdst_value.
1958 // Also puts address of local into Rdst_address as a service.
1959 // Kills:
1960 // - Rdst_value
1961 // - Rdst_address
1962 void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) {
1963 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
1964 subf(Rdst_address, Rdst_address, R18_locals);
1965 lwz(Rdst_value, 0, Rdst_address);
1966 }
1968 // Load a local variable at index in Rindex into register Rdst_value.
1969 // Also puts address of local into Rdst_address as a service.
1970 // Kills:
1971 // - Rdst_value
1972 // - Rdst_address
1973 void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) {
1974 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
1975 subf(Rdst_address, Rdst_address, R18_locals);
1976 ld(Rdst_value, -8, Rdst_address);
1977 }
1979 // Load a local variable at index in Rindex into register Rdst_value.
1980 // Also puts address of local into Rdst_address as a service.
1981 // Input:
1982 // - Rindex: slot nr of local variable
1983 // Kills:
1984 // - Rdst_value
1985 // - Rdst_address
1986 void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, Register Rdst_address, Register Rindex) {
1987 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
1988 subf(Rdst_address, Rdst_address, R18_locals);
1989 ld(Rdst_value, 0, Rdst_address);
1990 }
1992 // Load a local variable at index in Rindex into register Rdst_value.
1993 // Also puts address of local into Rdst_address as a service.
1994 // Kills:
1995 // - Rdst_value
1996 // - Rdst_address
1997 void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) {
1998 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
1999 subf(Rdst_address, Rdst_address, R18_locals);
2000 lfs(Rdst_value, 0, Rdst_address);
2001 }
2003 // Load a local variable at index in Rindex into register Rdst_value.
2004 // Also puts address of local into Rdst_address as a service.
2005 // Kills:
2006 // - Rdst_value
2007 // - Rdst_address
2008 void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) {
2009 sldi(Rdst_address, Rindex, Interpreter::logStackElementSize);
2010 subf(Rdst_address, Rdst_address, R18_locals);
2011 lfd(Rdst_value, -8, Rdst_address);
2012 }
2014 // Store an int value at local variable slot Rindex.
2015 // Kills:
2016 // - Rindex
2017 void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) {
2018 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2019 subf(Rindex, Rindex, R18_locals);
2020 stw(Rvalue, 0, Rindex);
2021 }
2023 // Store a long value at local variable slot Rindex.
2024 // Kills:
2025 // - Rindex
2026 void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) {
2027 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2028 subf(Rindex, Rindex, R18_locals);
2029 std(Rvalue, -8, Rindex);
2030 }
2032 // Store an oop value at local variable slot Rindex.
2033 // Kills:
2034 // - Rindex
2035 void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) {
2036 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2037 subf(Rindex, Rindex, R18_locals);
2038 std(Rvalue, 0, Rindex);
2039 }
2041 // Store an int value at local variable slot Rindex.
2042 // Kills:
2043 // - Rindex
2044 void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) {
2045 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2046 subf(Rindex, Rindex, R18_locals);
2047 stfs(Rvalue, 0, Rindex);
2048 }
2050 // Store an int value at local variable slot Rindex.
2051 // Kills:
2052 // - Rindex
2053 void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) {
2054 sldi(Rindex, Rindex, Interpreter::logStackElementSize);
2055 subf(Rindex, Rindex, R18_locals);
2056 stfd(Rvalue, -8, Rindex);
2057 }
2059 // Read pending exception from thread and jump to interpreter.
2060 // Throw exception entry if one if pending. Fall through otherwise.
2061 void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) {
2062 assert_different_registers(Rscratch1, Rscratch2, R3);
2063 Register Rexception = Rscratch1;
2064 Register Rtmp = Rscratch2;
2065 Label Ldone;
2066 // Get pending exception oop.
2067 ld(Rexception, thread_(pending_exception));
2068 cmpdi(CCR0, Rexception, 0);
2069 beq(CCR0, Ldone);
2070 li(Rtmp, 0);
2071 mr_if_needed(R3, Rexception);
2072 std(Rtmp, thread_(pending_exception)); // Clear exception in thread
2073 if (Interpreter::rethrow_exception_entry() != NULL) {
2074 // Already got entry address.
2075 load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry());
2076 } else {
2077 // Dynamically load entry address.
2078 int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true);
2079 ld(Rtmp, simm16_rest, Rtmp);
2080 }
2081 mtctr(Rtmp);
2082 save_interpreter_state(Rtmp);
2083 bctr();
2085 align(32, 12);
2086 bind(Ldone);
2087 }
2089 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) {
2090 save_interpreter_state(R11_scratch1);
2092 MacroAssembler::call_VM(oop_result, entry_point, false);
2094 restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2096 check_and_handle_popframe(R11_scratch1);
2097 check_and_handle_earlyret(R11_scratch1);
2098 // Now check exceptions manually.
2099 if (check_exceptions) {
2100 check_and_forward_exception(R11_scratch1, R12_scratch2);
2101 }
2102 }
2104 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions) {
2105 // ARG1 is reserved for the thread.
2106 mr_if_needed(R4_ARG2, arg_1);
2107 call_VM(oop_result, entry_point, check_exceptions);
2108 }
2110 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) {
2111 // ARG1 is reserved for the thread.
2112 mr_if_needed(R4_ARG2, arg_1);
2113 assert(arg_2 != R4_ARG2, "smashed argument");
2114 mr_if_needed(R5_ARG3, arg_2);
2115 call_VM(oop_result, entry_point, check_exceptions);
2116 }
2118 void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) {
2119 // ARG1 is reserved for the thread.
2120 mr_if_needed(R4_ARG2, arg_1);
2121 assert(arg_2 != R4_ARG2, "smashed argument");
2122 mr_if_needed(R5_ARG3, arg_2);
2123 assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument");
2124 mr_if_needed(R6_ARG4, arg_3);
2125 call_VM(oop_result, entry_point, check_exceptions);
2126 }
2128 void InterpreterMacroAssembler::save_interpreter_state(Register scratch) {
2129 ld(scratch, 0, R1_SP);
2130 std(R15_esp, _ijava_state_neg(esp), scratch);
2131 std(R14_bcp, _ijava_state_neg(bcp), scratch);
2132 std(R26_monitor, _ijava_state_neg(monitors), scratch);
2133 if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); }
2134 // Other entries should be unchanged.
2135 }
2137 void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) {
2138 ld(scratch, 0, R1_SP);
2139 ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception).
2140 if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code.
2141 if (!bcp_and_mdx_only) {
2142 // Following ones are Metadata.
2143 ld(R19_method, _ijava_state_neg(method), scratch);
2144 ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch);
2145 // Following ones are stack addresses and don't require reload.
2146 ld(R15_esp, _ijava_state_neg(esp), scratch);
2147 ld(R18_locals, _ijava_state_neg(locals), scratch);
2148 ld(R26_monitor, _ijava_state_neg(monitors), scratch);
2149 }
2150 #ifdef ASSERT
2151 {
2152 Label Lok;
2153 subf(R0, R1_SP, scratch);
2154 cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size);
2155 bge(CCR0, Lok);
2156 stop("frame too small (restore istate)", 0x5432);
2157 bind(Lok);
2158 }
2159 {
2160 Label Lok;
2161 ld(R0, _ijava_state_neg(ijava_reserved), scratch);
2162 cmpdi(CCR0, R0, 0x5afe);
2163 beq(CCR0, Lok);
2164 stop("frame corrupted (restore istate)", 0x5afe);
2165 bind(Lok);
2166 }
2167 #endif
2168 }
2170 #endif // !CC_INTERP
2172 void InterpreterMacroAssembler::get_method_counters(Register method,
2173 Register Rcounters,
2174 Label& skip) {
2175 BLOCK_COMMENT("Load and ev. allocate counter object {");
2176 Label has_counters;
2177 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2178 cmpdi(CCR0, Rcounters, 0);
2179 bne(CCR0, has_counters);
2180 call_VM(noreg, CAST_FROM_FN_PTR(address,
2181 InterpreterRuntime::build_method_counters), method, false);
2182 ld(Rcounters, in_bytes(Method::method_counters_offset()), method);
2183 cmpdi(CCR0, Rcounters, 0);
2184 beq(CCR0, skip); // No MethodCounters, OutOfMemory.
2185 BLOCK_COMMENT("} Load and ev. allocate counter object");
2187 bind(has_counters);
2188 }
2190 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register iv_be_count, Register Rtmp_r0) {
2191 assert(UseCompiler, "incrementing must be useful");
2192 Register invocation_count = iv_be_count;
2193 Register backedge_count = Rtmp_r0;
2194 int delta = InvocationCounter::count_increment;
2196 // Load each counter in a register.
2197 // ld(inv_counter, Rtmp);
2198 // ld(be_counter, Rtmp2);
2199 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() +
2200 InvocationCounter::counter_offset());
2201 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() +
2202 InvocationCounter::counter_offset());
2204 BLOCK_COMMENT("Increment profiling counters {");
2206 // Load the backedge counter.
2207 lwz(backedge_count, be_counter_offset, Rcounters); // is unsigned int
2208 // Mask the backedge counter.
2209 Register tmp = invocation_count;
2210 li(tmp, InvocationCounter::count_mask_value);
2211 andr(backedge_count, tmp, backedge_count); // Cannot use andi, need sign extension of count_mask_value.
2213 // Load the invocation counter.
2214 lwz(invocation_count, inv_counter_offset, Rcounters); // is unsigned int
2215 // Add the delta to the invocation counter and store the result.
2216 addi(invocation_count, invocation_count, delta);
2217 // Store value.
2218 stw(invocation_count, inv_counter_offset, Rcounters);
2220 // Add invocation counter + backedge counter.
2221 add(iv_be_count, backedge_count, invocation_count);
2223 // Note that this macro must leave the backedge_count + invocation_count in
2224 // register iv_be_count!
2225 BLOCK_COMMENT("} Increment profiling counters");
2226 }
2228 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2229 if (state == atos) { MacroAssembler::verify_oop(reg); }
2230 }
2232 #ifndef CC_INTERP
2233 // Local helper function for the verify_oop_or_return_address macro.
2234 static bool verify_return_address(Method* m, int bci) {
2235 #ifndef PRODUCT
2236 address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci;
2237 // Assume it is a valid return address if it is inside m and is preceded by a jsr.
2238 if (!m->contains(pc)) return false;
2239 address jsr_pc;
2240 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2241 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2242 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2243 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2244 #endif // PRODUCT
2245 return false;
2246 }
2248 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2249 if (VerifyFPU) {
2250 unimplemented("verfiyFPU");
2251 }
2252 }
2254 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2255 if (!VerifyOops) return;
2257 // The VM documentation for the astore[_wide] bytecode allows
2258 // the TOS to be not only an oop but also a return address.
2259 Label test;
2260 Label skip;
2261 // See if it is an address (in the current method):
2263 const int log2_bytecode_size_limit = 16;
2264 srdi_(Rtmp, reg, log2_bytecode_size_limit);
2265 bne(CCR0, test);
2267 address fd = CAST_FROM_FN_PTR(address, verify_return_address);
2268 const int nbytes_save = 11*8; // volatile gprs except R0
2269 save_volatile_gprs(R1_SP, -nbytes_save); // except R0
2270 save_LR_CR(Rtmp); // Save in old frame.
2271 push_frame_reg_args(nbytes_save, Rtmp);
2273 load_const_optimized(Rtmp, fd, R0);
2274 mr_if_needed(R4_ARG2, reg);
2275 mr(R3_ARG1, R19_method);
2276 call_c(Rtmp); // call C
2278 pop_frame();
2279 restore_LR_CR(Rtmp);
2280 restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
2281 b(skip);
2283 // Perform a more elaborate out-of-line call.
2284 // Not an address; verify it:
2285 bind(test);
2286 verify_oop(reg);
2287 bind(skip);
2288 }
2289 #endif // !CC_INTERP
2291 // Inline assembly for:
2292 //
2293 // if (thread is in interp_only_mode) {
2294 // InterpreterRuntime::post_method_entry();
2295 // }
2296 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY ) ||
2297 // *jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_ENTRY2) ) {
2298 // SharedRuntime::jvmpi_method_entry(method, receiver);
2299 // }
2300 void InterpreterMacroAssembler::notify_method_entry() {
2301 // JVMTI
2302 // Whenever JVMTI puts a thread in interp_only_mode, method
2303 // entry/exit events are sent for that thread to track stack
2304 // depth. If it is possible to enter interp_only_mode we add
2305 // the code to check if the event should be sent.
2306 if (JvmtiExport::can_post_interpreter_events()) {
2307 Label jvmti_post_done;
2309 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2310 cmpwi(CCR0, R0, 0);
2311 beq(CCR0, jvmti_post_done);
2312 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry),
2313 /*check_exceptions=*/true CC_INTERP_ONLY(&& false));
2315 bind(jvmti_post_done);
2316 }
2317 }
2319 // Inline assembly for:
2320 //
2321 // if (thread is in interp_only_mode) {
2322 // // save result
2323 // InterpreterRuntime::post_method_exit();
2324 // // restore result
2325 // }
2326 // if (*jvmpi::event_flags_array_at_addr(JVMPI_EVENT_METHOD_EXIT)) {
2327 // // save result
2328 // SharedRuntime::jvmpi_method_exit();
2329 // // restore result
2330 // }
2331 //
2332 // Native methods have their result stored in d_tmp and l_tmp.
2333 // Java methods have their result stored in the expression stack.
2334 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state,
2335 NotifyMethodExitMode mode, bool check_exceptions) {
2336 // JVMTI
2337 // Whenever JVMTI puts a thread in interp_only_mode, method
2338 // entry/exit events are sent for that thread to track stack
2339 // depth. If it is possible to enter interp_only_mode we add
2340 // the code to check if the event should be sent.
2341 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2342 Label jvmti_post_done;
2344 lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
2345 cmpwi(CCR0, R0, 0);
2346 beq(CCR0, jvmti_post_done);
2347 CC_INTERP_ONLY(assert(is_native_method && !check_exceptions, "must not push state"));
2348 if (!is_native_method) push(state); // Expose tos to GC.
2349 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit),
2350 /*check_exceptions=*/check_exceptions);
2351 if (!is_native_method) pop(state);
2353 align(32, 12);
2354 bind(jvmti_post_done);
2355 }
2357 // Dtrace support not implemented.
2358 }
2360 #ifdef CC_INTERP
2361 // Convert the current TOP_IJAVA_FRAME into a PARENT_IJAVA_FRAME
2362 // (using parent_frame_resize) and push a new interpreter
2363 // TOP_IJAVA_FRAME (using frame_size).
2364 void InterpreterMacroAssembler::push_interpreter_frame(Register top_frame_size, Register parent_frame_resize,
2365 Register tmp1, Register tmp2, Register tmp3,
2366 Register tmp4, Register pc) {
2367 assert_different_registers(top_frame_size, parent_frame_resize, tmp1, tmp2, tmp3, tmp4);
2368 ld(tmp1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
2369 mr(tmp2/*top_frame_sp*/, R1_SP);
2370 // Move initial_caller_sp.
2371 ld(tmp4, _top_ijava_frame_abi(initial_caller_sp), R1_SP);
2372 neg(parent_frame_resize, parent_frame_resize);
2373 resize_frame(parent_frame_resize/*-parent_frame_resize*/, tmp3);
2375 // Set LR in new parent frame.
2376 std(tmp1, _abi(lr), R1_SP);
2377 // Set top_frame_sp info for new parent frame.
2378 std(tmp2, _parent_ijava_frame_abi(top_frame_sp), R1_SP);
2379 std(tmp4, _parent_ijava_frame_abi(initial_caller_sp), R1_SP);
2381 // Push new TOP_IJAVA_FRAME.
2382 push_frame(top_frame_size, tmp2);
2384 get_PC_trash_LR(tmp3);
2385 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
2386 // Used for non-initial callers by unextended_sp().
2387 std(R1_SP, _top_ijava_frame_abi(initial_caller_sp), R1_SP);
2388 }
2390 // Pop the topmost TOP_IJAVA_FRAME and convert the previous
2391 // PARENT_IJAVA_FRAME back into a TOP_IJAVA_FRAME.
2392 void InterpreterMacroAssembler::pop_interpreter_frame(Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
2393 assert_different_registers(tmp1, tmp2, tmp3, tmp4);
2395 ld(tmp1/*caller's sp*/, _abi(callers_sp), R1_SP);
2396 ld(tmp3, _abi(lr), tmp1);
2398 ld(tmp4, _parent_ijava_frame_abi(initial_caller_sp), tmp1);
2400 ld(tmp2/*caller's caller's sp*/, _abi(callers_sp), tmp1);
2401 // Merge top frame.
2402 std(tmp2, _abi(callers_sp), R1_SP);
2404 ld(tmp2, _parent_ijava_frame_abi(top_frame_sp), tmp1);
2406 // Update C stack pointer to caller's top_abi.
2407 resize_frame_absolute(tmp2/*addr*/, tmp1/*tmp*/, tmp2/*tmp*/);
2409 // Update LR in top_frame.
2410 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
2412 std(tmp4, _top_ijava_frame_abi(initial_caller_sp), R1_SP);
2414 // Store the top-frame stack-pointer for c2i adapters.
2415 std(R1_SP, _top_ijava_frame_abi(top_frame_sp), R1_SP);
2416 }
2418 // Turn state's interpreter frame into the current TOP_IJAVA_FRAME.
2419 void InterpreterMacroAssembler::pop_interpreter_frame_to_state(Register state, Register tmp1, Register tmp2, Register tmp3) {
2420 assert_different_registers(R14_state, R15_prev_state, tmp1, tmp2, tmp3);
2422 if (state == R14_state) {
2423 ld(tmp1/*state's fp*/, state_(_last_Java_fp));
2424 ld(tmp2/*state's sp*/, state_(_last_Java_sp));
2425 } else if (state == R15_prev_state) {
2426 ld(tmp1/*state's fp*/, prev_state_(_last_Java_fp));
2427 ld(tmp2/*state's sp*/, prev_state_(_last_Java_sp));
2428 } else {
2429 ShouldNotReachHere();
2430 }
2432 // Merge top frames.
2433 std(tmp1, _abi(callers_sp), R1_SP);
2435 // Tmp2 is new SP.
2436 // Tmp1 is parent's SP.
2437 resize_frame_absolute(tmp2/*addr*/, tmp1/*tmp*/, tmp2/*tmp*/);
2439 // Update LR in top_frame.
2440 // Must be interpreter frame.
2441 get_PC_trash_LR(tmp3);
2442 std(tmp3, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
2443 // Used for non-initial callers by unextended_sp().
2444 std(R1_SP, _top_ijava_frame_abi(initial_caller_sp), R1_SP);
2445 }
2447 // Set SP to initial caller's sp, but before fix the back chain.
2448 void InterpreterMacroAssembler::resize_frame_to_initial_caller(Register tmp1, Register tmp2) {
2449 ld(tmp1, _parent_ijava_frame_abi(initial_caller_sp), R1_SP);
2450 ld(tmp2, _parent_ijava_frame_abi(callers_sp), R1_SP);
2451 std(tmp2, _parent_ijava_frame_abi(callers_sp), tmp1); // Fix back chain ...
2452 mr(R1_SP, tmp1); // ... and resize to initial caller.
2453 }
2455 // Pop the current interpreter state (without popping the correspoding
2456 // frame) and restore R14_state and R15_prev_state accordingly.
2457 // Use prev_state_may_be_0 to indicate whether prev_state may be 0
2458 // in order to generate an extra check before retrieving prev_state_(_prev_link).
2459 void InterpreterMacroAssembler::pop_interpreter_state(bool prev_state_may_be_0)
2460 {
2461 // Move prev_state to state and restore prev_state from state_(_prev_link).
2462 Label prev_state_is_0;
2463 mr(R14_state, R15_prev_state);
2465 // Don't retrieve /*state==*/prev_state_(_prev_link)
2466 // if /*state==*/prev_state is 0.
2467 if (prev_state_may_be_0) {
2468 cmpdi(CCR0, R15_prev_state, 0);
2469 beq(CCR0, prev_state_is_0);
2470 }
2472 ld(R15_prev_state, /*state==*/prev_state_(_prev_link));
2473 bind(prev_state_is_0);
2474 }
2476 void InterpreterMacroAssembler::restore_prev_state() {
2477 // _prev_link is private, but cInterpreter is a friend.
2478 ld(R15_prev_state, state_(_prev_link));
2479 }
2480 #endif // CC_INTERP