1 /*
2 * Copyright (c) 2016, 2023, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2016, 2023 SAP SE. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 // Major contributions by AHa, AS, JL, ML.
27
28 #include "precompiled.hpp"
29 #include "asm/macroAssembler.inline.hpp"
30 #include "gc/shared/barrierSet.hpp"
31 #include "gc/shared/barrierSetAssembler.hpp"
32 #include "interp_masm_s390.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "interpreter/interpreterRuntime.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/markWord.hpp"
37 #include "oops/methodCounters.hpp"
38 #include "oops/methodData.hpp"
39 #include "oops/resolvedFieldEntry.hpp"
40 #include "oops/resolvedIndyEntry.hpp"
41 #include "oops/resolvedMethodEntry.hpp"
42 #include "prims/jvmtiExport.hpp"
43 #include "prims/jvmtiThreadState.hpp"
44 #include "runtime/basicLock.hpp"
45 #include "runtime/frame.inline.hpp"
46 #include "runtime/javaThread.hpp"
47 #include "runtime/safepointMechanism.hpp"
48 #include "runtime/sharedRuntime.hpp"
49 #include "utilities/macros.hpp"
50 #include "utilities/powerOfTwo.hpp"
51
52 // Implementation of InterpreterMacroAssembler.
53 // This file specializes the assembler with interpreter-specific macros.
54
55 #ifdef PRODUCT
56 #define BLOCK_COMMENT(str)
57 #define BIND(label) bind(label);
58 #else
59 #define BLOCK_COMMENT(str) block_comment(str)
60 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
61 #endif
62
63 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) {
64 assert(entry != nullptr, "Entry must have been generated by now");
65 assert(Rscratch != Z_R0, "Can't use R0 for addressing");
66 branch_optimized(Assembler::bcondAlways, entry);
67 }
68
69 void InterpreterMacroAssembler::empty_expression_stack(void) {
70 get_monitors(Z_R1_scratch);
71 add2reg(Z_esp, -Interpreter::stackElementSize, Z_R1_scratch);
72 }
73
74 // Dispatch code executed in the prolog of a bytecode which does not do it's
75 // own dispatch.
76 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
77 // On z/Architecture we are short on registers, therefore we do not preload the
78 // dispatch address of the next bytecode.
79 }
80
81 // Dispatch code executed in the epilog of a bytecode which does not do it's
82 // own dispatch.
83 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
84 dispatch_next(state, step);
85 }
86
87 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) {
88 z_llgc(Z_bytecode, bcp_incr, Z_R0, Z_bcp); // Load next bytecode.
89 add2reg(Z_bcp, bcp_incr); // Advance bcp. Add2reg produces optimal code.
90 dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
91 }
92
93 // Common code to dispatch and dispatch_only.
94 // Dispatch value in Lbyte_code and increment Lbcp.
95
96 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table, bool generate_poll) {
97 verify_FPU(1, state);
98
99 #ifdef ASSERT
100 address reentry = nullptr;
101 { Label OK;
102 // Check if the frame pointer in Z_fp is correct.
103 z_cg(Z_fp, 0, Z_SP);
104 z_bre(OK);
105 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp: " FILE_AND_LINE);
106 bind(OK);
107 }
108 { Label OK;
109 // check if the locals pointer in Z_locals is correct
110 z_cg(Z_locals, _z_ijava_state_neg(locals), Z_fp);
111 z_bre(OK);
112 reentry = stop_chain_static(reentry, "invalid locals pointer Z_locals: " FILE_AND_LINE);
113 bind(OK);
114 }
115 #endif
116
117 // TODO: Maybe implement +VerifyActivationFrameSize here.
118 verify_oop(Z_tos, state);
119
120 // Dispatch table to use.
121 load_absolute_address(Z_tmp_1, (address)table); // Z_tmp_1 = table;
122
123 if (generate_poll) {
124 address *sfpt_tbl = Interpreter::safept_table(state);
125 if (table != sfpt_tbl) {
126 Label dispatch;
127 const Address poll_byte_addr(Z_thread, in_bytes(JavaThread::polling_word_offset()) + 7 /* Big Endian */);
128 // Armed page has poll_bit set, if poll bit is cleared just continue.
129 z_tm(poll_byte_addr, SafepointMechanism::poll_bit());
130 z_braz(dispatch);
131 load_absolute_address(Z_tmp_1, (address)sfpt_tbl); // Z_tmp_1 = table;
132 bind(dispatch);
133 }
134 }
135
136 // 0 <= Z_bytecode < 256 => Use a 32 bit shift, because it is shorter than sllg.
137 // Z_bytecode must have been loaded zero-extended for this approach to be correct.
138 z_sll(Z_bytecode, LogBytesPerWord, Z_R0); // Multiply by wordSize.
139 z_lg(Z_tmp_1, 0, Z_bytecode, Z_tmp_1); // Get entry addr.
140
141 z_br(Z_tmp_1);
142 }
143
144 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
145 dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
146 }
147
148 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
149 dispatch_base(state, Interpreter::normal_table(state));
150 }
151
152 void InterpreterMacroAssembler::dispatch_via(TosState state, address *table) {
153 // Load current bytecode.
154 z_llgc(Z_bytecode, Address(Z_bcp, (intptr_t)0));
155 dispatch_base(state, table);
156 }
157
158 // The following call_VM*_base() methods overload and mask the respective
159 // declarations/definitions in class MacroAssembler. They are meant as a "detour"
160 // to perform additional, template interpreter specific tasks before actually
161 // calling their MacroAssembler counterparts.
162
163 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point) {
164 bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated.
165 // interpreter specific
166 // Note: No need to save/restore bcp (Z_R13) pointer since these are callee
167 // saved registers and no blocking/ GC can happen in leaf calls.
168
169 // super call
170 MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation);
171 }
172
173 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, bool allow_relocation) {
174 // interpreter specific
175 // Note: No need to save/restore bcp (Z_R13) pointer since these are callee
176 // saved registers and no blocking/ GC can happen in leaf calls.
177
178 // super call
179 MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation);
180 }
181
182 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp,
183 address entry_point, bool check_exceptions) {
184 bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated.
185 // interpreter specific
186
187 save_bcp();
188 save_esp();
189 // super call
190 MacroAssembler::call_VM_base(oop_result, last_java_sp,
191 entry_point, allow_relocation, check_exceptions);
192 restore_bcp();
193 }
194
195 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp,
196 address entry_point, bool allow_relocation,
197 bool check_exceptions) {
198 // interpreter specific
199
200 save_bcp();
201 save_esp();
202 // super call
203 MacroAssembler::call_VM_base(oop_result, last_java_sp,
204 entry_point, allow_relocation, check_exceptions);
205 restore_bcp();
206 }
207
208 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
209 if (JvmtiExport::can_pop_frame()) {
210 BLOCK_COMMENT("check_and_handle_popframe {");
211 Label L;
212 // Initiate popframe handling only if it is not already being
213 // processed. If the flag has the popframe_processing bit set, it
214 // means that this code is called *during* popframe handling - we
215 // don't want to reenter.
216 // TODO: Check if all four state combinations could be visible.
217 // If (processing and !pending) is an invisible/impossible state,
218 // there is optimization potential by testing both bits at once.
219 // Then, All_Zeroes and All_Ones means skip, Mixed means doit.
220 testbit(Address(Z_thread, JavaThread::popframe_condition_offset()),
221 exact_log2(JavaThread::popframe_pending_bit));
222 z_bfalse(L);
223 testbit(Address(Z_thread, JavaThread::popframe_condition_offset()),
224 exact_log2(JavaThread::popframe_processing_bit));
225 z_btrue(L);
226
227 // Call Interpreter::remove_activation_preserving_args_entry() to get the
228 // address of the same-named entrypoint in the generated interpreter code.
229 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
230 // The above call should (as its only effect) return the contents of the field
231 // _remove_activation_preserving_args_entry in Z_RET.
232 // We just jump there to have the work done.
233 z_br(Z_RET);
234 // There is no way for control to fall thru here.
235
236 bind(L);
237 BLOCK_COMMENT("} check_and_handle_popframe");
238 }
239 }
240
241
242 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
243 Register RjvmtiState = Z_R1_scratch;
244 int tos_off = in_bytes(JvmtiThreadState::earlyret_tos_offset());
245 int oop_off = in_bytes(JvmtiThreadState::earlyret_oop_offset());
246 int val_off = in_bytes(JvmtiThreadState::earlyret_value_offset());
247 int state_off = in_bytes(JavaThread::jvmti_thread_state_offset());
248
249 z_lg(RjvmtiState, state_off, Z_thread);
250
251 switch (state) {
252 case atos: z_lg(Z_tos, oop_off, RjvmtiState);
253 store_const(Address(RjvmtiState, oop_off), 0L, 8, 8, Z_R0_scratch);
254 break;
255 case ltos: z_lg(Z_tos, val_off, RjvmtiState); break;
256 case btos: // fall through
257 case ztos: // fall through
258 case ctos: // fall through
259 case stos: // fall through
260 case itos: z_llgf(Z_tos, val_off, RjvmtiState); break;
261 case ftos: z_le(Z_ftos, val_off, RjvmtiState); break;
262 case dtos: z_ld(Z_ftos, val_off, RjvmtiState); break;
263 case vtos: /* nothing to do */ break;
264 default : ShouldNotReachHere();
265 }
266
267 // Clean up tos value in the jvmti thread state.
268 store_const(Address(RjvmtiState, val_off), 0L, 8, 8, Z_R0_scratch);
269 // Set tos state field to illegal value.
270 store_const(Address(RjvmtiState, tos_off), ilgl, 4, 1, Z_R0_scratch);
271 }
272
273 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
274 if (JvmtiExport::can_force_early_return()) {
275 BLOCK_COMMENT("check_and_handle_earlyret {");
276 Label L;
277 // arg regs are save, because we are just behind the call in call_VM_base
278 Register jvmti_thread_state = Z_ARG2;
279 Register tmp = Z_ARG3;
280 load_and_test_long(jvmti_thread_state, Address(Z_thread, JavaThread::jvmti_thread_state_offset()));
281 z_bre(L); // if (thread->jvmti_thread_state() == nullptr) exit;
282
283 // Initiate earlyret handling only if it is not already being processed.
284 // If the flag has the earlyret_processing bit set, it means that this code
285 // is called *during* earlyret handling - we don't want to reenter.
286
287 assert((JvmtiThreadState::earlyret_pending != 0) && (JvmtiThreadState::earlyret_inactive == 0),
288 "must fix this check, when changing the values of the earlyret enum");
289 assert(JvmtiThreadState::earlyret_pending == 1, "must fix this check, when changing the values of the earlyret enum");
290
291 load_and_test_int(tmp, Address(jvmti_thread_state, JvmtiThreadState::earlyret_state_offset()));
292 z_brz(L); // if (thread->jvmti_thread_state()->_earlyret_state != JvmtiThreadState::earlyret_pending) exit;
293
294 // Call Interpreter::remove_activation_early_entry() to get the address of the
295 // same-named entrypoint in the generated interpreter code.
296 assert(sizeof(TosState) == 4, "unexpected size");
297 z_l(Z_ARG1, Address(jvmti_thread_state, JvmtiThreadState::earlyret_tos_offset()));
298 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Z_ARG1);
299 // The above call should (as its only effect) return the contents of the field
300 // _remove_activation_preserving_args_entry in Z_RET.
301 // We just jump there to have the work done.
302 z_br(Z_RET);
303 // There is no way for control to fall thru here.
304
305 bind(L);
306 BLOCK_COMMENT("} check_and_handle_earlyret");
307 }
308 }
309
310 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2) {
311 lgr_if_needed(Z_ARG1, arg_1);
312 assert(arg_2 != Z_ARG1, "smashed argument");
313 lgr_if_needed(Z_ARG2, arg_2);
314 MacroAssembler::call_VM_leaf_base(entry_point, true);
315 }
316
317 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, int bcp_offset, size_t index_size) {
318 Address param(Z_bcp, bcp_offset);
319
320 BLOCK_COMMENT("get_cache_index_at_bcp {");
321 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
322 if (index_size == sizeof(u2)) {
323 load_sized_value(index, param, 2, false /*signed*/);
324 } else if (index_size == sizeof(u4)) {
325
326 load_sized_value(index, param, 4, false);
327
328 // Check if the secondary index definition is still ~x, otherwise
329 // we have to change the following assembler code to calculate the
330 // plain index.
331 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
332 not_(index); // Convert to plain index.
333 } else if (index_size == sizeof(u1)) {
334 z_llgc(index, param);
335 } else {
336 ShouldNotReachHere();
337 }
338 BLOCK_COMMENT("}");
339 }
340
341 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
342 // Get index out of bytecode pointer.
343 get_cache_index_at_bcp(index, 1, sizeof(u4));
344
345 // Get the address of the ResolvedIndyEntry array
346 get_constant_pool_cache(cache);
347 z_lg(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
348
349 // Scale the index to form a byte offset into the ResolvedIndyEntry array
350 size_t entry_size = sizeof(ResolvedIndyEntry);
351 if (is_power_of_2(entry_size)) {
352 z_sllg(index, index, exact_log2(entry_size));
353 } else {
354 z_mghi(index, entry_size);
355 }
356
357 // Calculate the final field address.
358 z_la(cache, Array<ResolvedIndyEntry>::base_offset_in_bytes(), index, cache);
359 }
360
361 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
362 // Get field index out of bytecode pointer.
363 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
364
365 // Get the address of the ResolvedFieldEntry array.
366 get_constant_pool_cache(cache);
367 z_lg(cache, Address(cache, in_bytes(ConstantPoolCache::field_entries_offset())));
368
369 // Scale the index to form a byte offset into the ResolvedFieldEntry array
370 size_t entry_size = sizeof(ResolvedFieldEntry);
371 if (is_power_of_2(entry_size)) {
372 z_sllg(index, index, exact_log2(entry_size));
373 } else {
374 z_mghi(index, entry_size);
375 }
376
377 // Calculate the final field address.
378 z_la(cache, Array<ResolvedFieldEntry>::base_offset_in_bytes(), index, cache);
379 }
380
381 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
382 // Get field index out of bytecode pointer.
383 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
384
385 // Get the address of the ResolvedMethodEntry array.
386 get_constant_pool_cache(cache);
387 z_lg(cache, Address(cache, in_bytes(ConstantPoolCache::method_entries_offset())));
388
389 // Scale the index to form a byte offset into the ResolvedMethodEntry array
390 size_t entry_size = sizeof(ResolvedMethodEntry);
391 if (is_power_of_2(entry_size)) {
392 z_sllg(index, index, exact_log2(entry_size));
393 } else {
394 z_mghi(index, entry_size);
395 }
396
397 // Calculate the final field address.
398 z_la(cache, Array<ResolvedMethodEntry>::base_offset_in_bytes(), index, cache);
399 }
400
401 // Load object from cpool->resolved_references(index).
402 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) {
403 assert_different_registers(result, index);
404 get_constant_pool(result);
405
406 // Convert
407 // - from field index to resolved_references() index and
408 // - from word index to byte offset.
409 // Since this is a java object, it is potentially compressed.
410 Register tmp = index; // reuse
411 z_sllg(index, index, LogBytesPerHeapOop); // Offset into resolved references array.
412 // Load pointer for resolved_references[] objArray.
413 z_lg(result, in_bytes(ConstantPool::cache_offset()), result);
414 z_lg(result, in_bytes(ConstantPoolCache::resolved_references_offset()), result);
415 resolve_oop_handle(result); // Load resolved references array itself.
416 #ifdef ASSERT
417 NearLabel index_ok;
418 z_lgf(Z_R0, Address(result, arrayOopDesc::length_offset_in_bytes()));
419 z_sllg(Z_R0, Z_R0, LogBytesPerHeapOop);
420 compare64_and_branch(tmp, Z_R0, Assembler::bcondLow, index_ok);
421 stop("resolved reference index out of bounds", 0x09256);
422 bind(index_ok);
423 #endif
424 z_agr(result, index); // Address of indexed array element.
425 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp, noreg);
426 }
427
428 // load cpool->resolved_klass_at(index)
429 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register cpool, Register offset, Register iklass) {
430 // int value = *(Rcpool->int_at_addr(which));
431 // int resolved_klass_index = extract_low_short_from_int(value);
432 z_llgh(offset, Address(cpool, offset, sizeof(ConstantPool) + 2)); // offset = resolved_klass_index (s390 is big-endian)
433 z_sllg(offset, offset, LogBytesPerWord); // Convert 'index' to 'offset'
434 z_lg(iklass, Address(cpool, ConstantPool::resolved_klasses_offset())); // iklass = cpool->_resolved_klasses
435 z_lg(iklass, Address(iklass, offset, Array<Klass*>::base_offset_in_bytes()));
436 }
437
438 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
439 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2.
440 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
441 Register Rsuper_klass,
442 Register Rtmp1,
443 Register Rtmp2,
444 Label &ok_is_subtype) {
445 // Profile the not-null value's klass.
446 profile_typecheck(Rtmp1, Rsub_klass, Rtmp2);
447
448 // Do the check.
449 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
450
451 // Profile the failure of the check.
452 profile_typecheck_failed(Rtmp1, Rtmp2);
453 }
454
455 // Pop topmost element from stack. It just disappears.
456 // Useful if consumed previously by access via stackTop().
457 void InterpreterMacroAssembler::popx(int len) {
458 add2reg(Z_esp, len*Interpreter::stackElementSize);
459 debug_only(verify_esp(Z_esp, Z_R1_scratch));
460 }
461
462 // Get Address object of stack top. No checks. No pop.
463 // Purpose: - Provide address of stack operand to exploit reg-mem operations.
464 // - Avoid RISC-like mem2reg - reg-reg-op sequence.
465 Address InterpreterMacroAssembler::stackTop() {
466 return Address(Z_esp, Interpreter::expr_offset_in_bytes(0));
467 }
468
469 void InterpreterMacroAssembler::pop_i(Register r) {
470 z_l(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
471 add2reg(Z_esp, Interpreter::stackElementSize);
472 assert_different_registers(r, Z_R1_scratch);
473 debug_only(verify_esp(Z_esp, Z_R1_scratch));
474 }
475
476 void InterpreterMacroAssembler::pop_ptr(Register r) {
477 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
478 add2reg(Z_esp, Interpreter::stackElementSize);
479 assert_different_registers(r, Z_R1_scratch);
480 debug_only(verify_esp(Z_esp, Z_R1_scratch));
481 }
482
483 void InterpreterMacroAssembler::pop_l(Register r) {
484 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
485 add2reg(Z_esp, 2*Interpreter::stackElementSize);
486 assert_different_registers(r, Z_R1_scratch);
487 debug_only(verify_esp(Z_esp, Z_R1_scratch));
488 }
489
490 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
491 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), false);
492 add2reg(Z_esp, Interpreter::stackElementSize);
493 debug_only(verify_esp(Z_esp, Z_R1_scratch));
494 }
495
496 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
497 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), true);
498 add2reg(Z_esp, 2*Interpreter::stackElementSize);
499 debug_only(verify_esp(Z_esp, Z_R1_scratch));
500 }
501
502 void InterpreterMacroAssembler::push_i(Register r) {
503 assert_different_registers(r, Z_R1_scratch);
504 debug_only(verify_esp(Z_esp, Z_R1_scratch));
505 z_st(r, Address(Z_esp));
506 add2reg(Z_esp, -Interpreter::stackElementSize);
507 }
508
509 void InterpreterMacroAssembler::push_ptr(Register r) {
510 z_stg(r, Address(Z_esp));
511 add2reg(Z_esp, -Interpreter::stackElementSize);
512 }
513
514 void InterpreterMacroAssembler::push_l(Register r) {
515 assert_different_registers(r, Z_R1_scratch);
516 debug_only(verify_esp(Z_esp, Z_R1_scratch));
517 int offset = -Interpreter::stackElementSize;
518 z_stg(r, Address(Z_esp, offset));
519 clear_mem(Address(Z_esp), Interpreter::stackElementSize);
520 add2reg(Z_esp, 2 * offset);
521 }
522
523 void InterpreterMacroAssembler::push_f(FloatRegister f) {
524 debug_only(verify_esp(Z_esp, Z_R1_scratch));
525 freg2mem_opt(f, Address(Z_esp), false);
526 add2reg(Z_esp, -Interpreter::stackElementSize);
527 }
528
529 void InterpreterMacroAssembler::push_d(FloatRegister d) {
530 debug_only(verify_esp(Z_esp, Z_R1_scratch));
531 int offset = -Interpreter::stackElementSize;
532 freg2mem_opt(d, Address(Z_esp, offset));
533 add2reg(Z_esp, 2 * offset);
534 }
535
536 void InterpreterMacroAssembler::push(TosState state) {
537 verify_oop(Z_tos, state);
538 switch (state) {
539 case atos: push_ptr(); break;
540 case btos: push_i(); break;
541 case ztos:
542 case ctos:
543 case stos: push_i(); break;
544 case itos: push_i(); break;
545 case ltos: push_l(); break;
546 case ftos: push_f(); break;
547 case dtos: push_d(); break;
548 case vtos: /* nothing to do */ break;
549 default : ShouldNotReachHere();
550 }
551 }
552
553 void InterpreterMacroAssembler::pop(TosState state) {
554 switch (state) {
555 case atos: pop_ptr(Z_tos); break;
556 case btos: pop_i(Z_tos); break;
557 case ztos:
558 case ctos:
559 case stos: pop_i(Z_tos); break;
560 case itos: pop_i(Z_tos); break;
561 case ltos: pop_l(Z_tos); break;
562 case ftos: pop_f(Z_ftos); break;
563 case dtos: pop_d(Z_ftos); break;
564 case vtos: /* nothing to do */ break;
565 default : ShouldNotReachHere();
566 }
567 verify_oop(Z_tos, state);
568 }
569
570 // Helpers for swap and dup.
571 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
572 z_lg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n)));
573 }
574
575 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
576 z_stg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n)));
577 }
578
579 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted(Register method) {
580 // Satisfy interpreter calling convention (see generate_normal_entry()).
581 z_lgr(Z_R10, Z_SP); // Set sender sp (aka initial caller sp, aka unextended sp).
582 // Record top_frame_sp, because the callee might modify it, if it's compiled.
583 z_stg(Z_SP, _z_ijava_state_neg(top_frame_sp), Z_fp);
584 save_bcp();
585 save_esp();
586 z_lgr(Z_method, method); // Set Z_method (kills Z_fp!).
587 }
588
589 // Jump to from_interpreted entry of a call unless single stepping is possible
590 // in this thread in which case we must call the i2i entry.
591 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
592 assert_different_registers(method, Z_R10 /*used for initial_caller_sp*/, temp);
593 prepare_to_jump_from_interpreted(method);
594
595 if (JvmtiExport::can_post_interpreter_events()) {
596 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
597 // compiled code in threads for which the event is enabled. Check here for
598 // interp_only_mode if these events CAN be enabled.
599 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
600 MacroAssembler::load_and_test_int(Z_R0_scratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
601 z_bcr(bcondEqual, Z_R1_scratch); // Run compiled code if zero.
602 // Run interpreted.
603 z_lg(Z_R1_scratch, Address(method, Method::interpreter_entry_offset()));
604 z_br(Z_R1_scratch);
605 } else {
606 // Run compiled code.
607 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
608 z_br(Z_R1_scratch);
609 }
610 }
611
612 #ifdef ASSERT
613 void InterpreterMacroAssembler::verify_esp(Register Resp, Register Rtemp) {
614 // About to read or write Resp[0].
615 // Make sure it is not in the monitors or the TOP_IJAVA_FRAME_ABI.
616 address reentry = nullptr;
617
618 {
619 // Check if the frame pointer in Z_fp is correct.
620 NearLabel OK;
621 z_cg(Z_fp, 0, Z_SP);
622 z_bre(OK);
623 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp");
624 bind(OK);
625 }
626 {
627 // Resp must not point into or below the operand stack,
628 // i.e. IJAVA_STATE.monitors > Resp.
629 NearLabel OK;
630 Register Rmonitors = Rtemp;
631 z_lg(Rmonitors, _z_ijava_state_neg(monitors), Z_fp);
632 compareU64_and_branch(Rmonitors, Resp, bcondHigh, OK);
633 reentry = stop_chain_static(reentry, "too many pops: Z_esp points into monitor area");
634 bind(OK);
635 }
636 {
637 // Resp may point to the last word of TOP_IJAVA_FRAME_ABI, but not below
638 // i.e. !(Z_SP + frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize > Resp).
639 NearLabel OK;
640 Register Rabi_bottom = Rtemp;
641 add2reg(Rabi_bottom, frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize, Z_SP);
642 compareU64_and_branch(Rabi_bottom, Resp, bcondNotHigh, OK);
643 reentry = stop_chain_static(reentry, "too many pushes: Z_esp points into TOP_IJAVA_FRAME_ABI");
644 bind(OK);
645 }
646 }
647
648 void InterpreterMacroAssembler::asm_assert_ijava_state_magic(Register tmp) {
649 Label magic_ok;
650 load_const_optimized(tmp, frame::z_istate_magic_number);
651 z_cg(tmp, Address(Z_fp, _z_ijava_state_neg(magic)));
652 z_bre(magic_ok);
653 stop_static("error: wrong magic number in ijava_state access");
654 bind(magic_ok);
655 }
656 #endif // ASSERT
657
658 void InterpreterMacroAssembler::save_bcp() {
659 z_stg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
660 asm_assert_ijava_state_magic(Z_bcp);
661 NOT_PRODUCT(z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))));
662 }
663
664 void InterpreterMacroAssembler::restore_bcp() {
665 asm_assert_ijava_state_magic(Z_bcp);
666 z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
667 }
668
669 void InterpreterMacroAssembler::save_esp() {
670 z_stg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp)));
671 }
672
673 void InterpreterMacroAssembler::restore_esp() {
674 asm_assert_ijava_state_magic(Z_esp);
675 z_lg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp)));
676 }
677
678 void InterpreterMacroAssembler::get_monitors(Register reg) {
679 asm_assert_ijava_state_magic(reg);
680 mem2reg_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
681 }
682
683 void InterpreterMacroAssembler::save_monitors(Register reg) {
684 reg2mem_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
685 }
686
687 void InterpreterMacroAssembler::get_mdp(Register mdp) {
688 z_lg(mdp, _z_ijava_state_neg(mdx), Z_fp);
689 }
690
691 void InterpreterMacroAssembler::save_mdp(Register mdp) {
692 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
693 }
694
695 // Values that are only read (besides initialization).
696 void InterpreterMacroAssembler::restore_locals() {
697 asm_assert_ijava_state_magic(Z_locals);
698 z_lg(Z_locals, Address(Z_fp, _z_ijava_state_neg(locals)));
699 }
700
701 void InterpreterMacroAssembler::get_method(Register reg) {
702 asm_assert_ijava_state_magic(reg);
703 z_lg(reg, Address(Z_fp, _z_ijava_state_neg(method)));
704 }
705
706 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(Register Rdst, int bcp_offset,
707 signedOrNot is_signed) {
708 // Rdst is an 8-byte return value!!!
709
710 // Unaligned loads incur only a small penalty on z/Architecture. The penalty
711 // is a few (2..3) ticks, even when the load crosses a cache line
712 // boundary. In case of a cache miss, the stall could, of course, be
713 // much longer.
714
715 switch (is_signed) {
716 case Signed:
717 z_lgh(Rdst, bcp_offset, Z_R0, Z_bcp);
718 break;
719 case Unsigned:
720 z_llgh(Rdst, bcp_offset, Z_R0, Z_bcp);
721 break;
722 default:
723 ShouldNotReachHere();
724 }
725 }
726
727
728 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(Register Rdst, int bcp_offset,
729 setCCOrNot set_cc) {
730 // Rdst is an 8-byte return value!!!
731
732 // Unaligned loads incur only a small penalty on z/Architecture. The penalty
733 // is a few (2..3) ticks, even when the load crosses a cache line
734 // boundary. In case of a cache miss, the stall could, of course, be
735 // much longer.
736
737 // Both variants implement a sign-extending int2long load.
738 if (set_cc == set_CC) {
739 load_and_test_int2long(Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
740 } else {
741 mem2reg_signed_opt( Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
742 }
743 }
744
745 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
746 get_method(Rdst);
747 mem2reg_opt(Rdst, Address(Rdst, Method::const_offset()));
748 mem2reg_opt(Rdst, Address(Rdst, ConstMethod::constants_offset()));
749 }
750
751 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
752 get_constant_pool(Rdst);
753 mem2reg_opt(Rdst, Address(Rdst, ConstantPool::cache_offset()));
754 }
755
756 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
757 get_constant_pool(Rcpool);
758 mem2reg_opt(Rtags, Address(Rcpool, ConstantPool::tags_offset()));
759 }
760
761 // Unlock if synchronized method.
762 //
763 // Unlock the receiver if this is a synchronized method.
764 // Unlock any Java monitors from synchronized blocks.
765 //
766 // If there are locked Java monitors
767 // If throw_monitor_exception
768 // throws IllegalMonitorStateException
769 // Else if install_monitor_exception
770 // installs IllegalMonitorStateException
771 // Else
772 // no error processing
773 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
774 bool throw_monitor_exception,
775 bool install_monitor_exception) {
776 NearLabel unlocked, unlock, no_unlock;
777
778 {
779 Register R_method = Z_ARG2;
780 Register R_do_not_unlock_if_synchronized = Z_ARG3;
781
782 // Get the value of _do_not_unlock_if_synchronized into G1_scratch.
783 const Address do_not_unlock_if_synchronized(Z_thread,
784 JavaThread::do_not_unlock_if_synchronized_offset());
785 load_sized_value(R_do_not_unlock_if_synchronized, do_not_unlock_if_synchronized, 1, false /*unsigned*/);
786 z_mvi(do_not_unlock_if_synchronized, false); // Reset the flag.
787
788 // Check if synchronized method.
789 get_method(R_method);
790 verify_oop(Z_tos, state);
791 push(state); // Save tos/result.
792 testbit(method2_(R_method, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
793 z_bfalse(unlocked);
794
795 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
796 // is set.
797 compareU64_and_branch(R_do_not_unlock_if_synchronized, (intptr_t)0L, bcondNotEqual, no_unlock);
798 }
799
800 // unlock monitor
801
802 // BasicObjectLock will be first in list, since this is a
803 // synchronized method. However, need to check that the object has
804 // not been unlocked by an explicit monitorexit bytecode.
805 const Address monitor(Z_fp, -(frame::z_ijava_state_size + (int) sizeof(BasicObjectLock)));
806 // We use Z_ARG2 so that if we go slow path it will be the correct
807 // register for unlock_object to pass to VM directly.
808 load_address(Z_ARG2, monitor); // Address of first monitor.
809 z_lg(Z_ARG3, Address(Z_ARG2, BasicObjectLock::obj_offset()));
810 compareU64_and_branch(Z_ARG3, (intptr_t)0L, bcondNotEqual, unlock);
811
812 if (throw_monitor_exception) {
813 // Entry already unlocked need to throw an exception.
814 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
815 should_not_reach_here();
816 } else {
817 // Monitor already unlocked during a stack unroll.
818 // If requested, install an illegal_monitor_state_exception.
819 // Continue with stack unrolling.
820 if (install_monitor_exception) {
821 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
822 }
823 z_bru(unlocked);
824 }
825
826 bind(unlock);
827
828 unlock_object(Z_ARG2);
829
830 bind(unlocked);
831
832 // I0, I1: Might contain return value
833
834 // Check that all monitors are unlocked.
835 {
836 NearLabel loop, exception, entry, restart;
837 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
838 // We use Z_ARG2 so that if we go slow path it will be the correct
839 // register for unlock_object to pass to VM directly.
840 Register R_current_monitor = Z_ARG2;
841 Register R_monitor_block_bot = Z_ARG1;
842 const Address monitor_block_top(Z_fp, _z_ijava_state_neg(monitors));
843 const Address monitor_block_bot(Z_fp, -frame::z_ijava_state_size);
844
845 bind(restart);
846 // Starting with top-most entry.
847 z_lg(R_current_monitor, monitor_block_top);
848 // Points to word before bottom of monitor block.
849 load_address(R_monitor_block_bot, monitor_block_bot);
850 z_bru(entry);
851
852 // Entry already locked, need to throw exception.
853 bind(exception);
854
855 if (throw_monitor_exception) {
856 // Throw exception.
857 MacroAssembler::call_VM(noreg,
858 CAST_FROM_FN_PTR(address, InterpreterRuntime::
859 throw_illegal_monitor_state_exception));
860 should_not_reach_here();
861 } else {
862 // Stack unrolling. Unlock object and install illegal_monitor_exception.
863 // Unlock does not block, so don't have to worry about the frame.
864 // We don't have to preserve c_rarg1 since we are going to throw an exception.
865 unlock_object(R_current_monitor);
866 if (install_monitor_exception) {
867 call_VM(noreg, CAST_FROM_FN_PTR(address,
868 InterpreterRuntime::
869 new_illegal_monitor_state_exception));
870 }
871 z_bru(restart);
872 }
873
874 bind(loop);
875 // Check if current entry is used.
876 load_and_test_long(Z_R0_scratch, Address(R_current_monitor, BasicObjectLock::obj_offset()));
877 z_brne(exception);
878
879 add2reg(R_current_monitor, entry_size); // Otherwise advance to next entry.
880 bind(entry);
881 compareU64_and_branch(R_current_monitor, R_monitor_block_bot, bcondNotEqual, loop);
882 }
883
884 bind(no_unlock);
885 pop(state);
886 verify_oop(Z_tos, state);
887 }
888
889 void InterpreterMacroAssembler::narrow(Register result, Register ret_type) {
890 get_method(ret_type);
891 z_lg(ret_type, Address(ret_type, in_bytes(Method::const_offset())));
892 z_lb(ret_type, Address(ret_type, in_bytes(ConstMethod::result_type_offset())));
893
894 Label notBool, notByte, notChar, done;
895
896 // common case first
897 compareU32_and_branch(ret_type, T_INT, bcondEqual, done);
898
899 compareU32_and_branch(ret_type, T_BOOLEAN, bcondNotEqual, notBool);
900 z_nilf(result, 0x1);
901 z_bru(done);
902
903 bind(notBool);
904 compareU32_and_branch(ret_type, T_BYTE, bcondNotEqual, notByte);
905 z_lbr(result, result);
906 z_bru(done);
907
908 bind(notByte);
909 compareU32_and_branch(ret_type, T_CHAR, bcondNotEqual, notChar);
910 z_nilf(result, 0xffff);
911 z_bru(done);
912
913 bind(notChar);
914 // compareU32_and_branch(ret_type, T_SHORT, bcondNotEqual, notShort);
915 z_lhr(result, result);
916
917 // Nothing to do for T_INT
918 bind(done);
919 }
920
921 // remove activation
922 //
923 // Unlock the receiver if this is a synchronized method.
924 // Unlock any Java monitors from synchronized blocks.
925 // Remove the activation from the stack.
926 //
927 // If there are locked Java monitors
928 // If throw_monitor_exception
929 // throws IllegalMonitorStateException
930 // Else if install_monitor_exception
931 // installs IllegalMonitorStateException
932 // Else
933 // no error processing
934 void InterpreterMacroAssembler::remove_activation(TosState state,
935 Register return_pc,
936 bool throw_monitor_exception,
937 bool install_monitor_exception,
938 bool notify_jvmti) {
939 BLOCK_COMMENT("remove_activation {");
940 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
941
942 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
943 notify_method_exit(false, state, notify_jvmti ? NotifyJVMTI : SkipNotifyJVMTI);
944
945 if (StackReservedPages > 0) {
946 BLOCK_COMMENT("reserved_stack_check:");
947 // Test if reserved zone needs to be enabled.
948 Label no_reserved_zone_enabling;
949
950 // check if already enabled - if so no re-enabling needed
951 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
952 z_ly(Z_R0, Address(Z_thread, JavaThread::stack_guard_state_offset()));
953 compare32_and_branch(Z_R0, StackOverflow::stack_guard_enabled, bcondEqual, no_reserved_zone_enabling);
954
955 // Compare frame pointers. There is no good stack pointer, as with stack
956 // frame compression we can get different SPs when we do calls. A subsequent
957 // call could have a smaller SP, so that this compare succeeds for an
958 // inner call of the method annotated with ReservedStack.
959 z_lg(Z_R0, Address(Z_SP, (intptr_t)_z_abi(callers_sp)));
960 z_clg(Z_R0, Address(Z_thread, JavaThread::reserved_stack_activation_offset())); // Compare with frame pointer in memory.
961 z_brl(no_reserved_zone_enabling);
962
963 // Enable reserved zone again, throw stack overflow exception.
964 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), Z_thread);
965 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
966
967 should_not_reach_here();
968
969 bind(no_reserved_zone_enabling);
970 }
971
972 verify_oop(Z_tos, state);
973
974 pop_interpreter_frame(return_pc, Z_ARG2, Z_ARG3);
975 BLOCK_COMMENT("} remove_activation");
976 }
977
978 // lock object
979 //
980 // Registers alive
981 // monitor (Z_R10) - Address of the BasicObjectLock to be used for locking,
982 // which must be initialized with the object to lock.
983 // object (Z_R11, Z_R2) - Address of the object to be locked.
984 // templateTable (monitorenter) is using Z_R2 for object
985 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
986
987 if (LockingMode == LM_MONITOR) {
988 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
989 return;
990 }
991
992 // template code: (for LM_LEGACY)
993 //
994 // markWord displaced_header = obj->mark().set_unlocked();
995 // monitor->lock()->set_displaced_header(displaced_header);
996 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
997 // // We stored the monitor address into the object's mark word.
998 // } else if (THREAD->is_lock_owned((address)displaced_header))
999 // // Simple recursive case.
1000 // monitor->lock()->set_displaced_header(nullptr);
1001 // } else {
1002 // // Slow path.
1003 // InterpreterRuntime::monitorenter(THREAD, monitor);
1004 // }
1005
1006 const int hdr_offset = oopDesc::mark_offset_in_bytes();
1007
1008 const Register header = Z_ARG5;
1009 const Register object_mark_addr = Z_ARG4;
1010 const Register current_header = Z_ARG5;
1011 const Register tmp = Z_R1_scratch;
1012
1013 NearLabel done, slow_case;
1014
1015 // markWord header = obj->mark().set_unlocked();
1016
1017 // Load markWord from object into header.
1018 z_lg(header, hdr_offset, object);
1019
1020 if (DiagnoseSyncOnValueBasedClasses != 0) {
1021 load_klass(tmp, object);
1022 testbit(Address(tmp, Klass::access_flags_offset()), exact_log2(JVM_ACC_IS_VALUE_BASED_CLASS));
1023 z_btrue(slow_case);
1024 }
1025
1026 if (LockingMode == LM_LIGHTWEIGHT) {
1027 lightweight_lock(object, /* mark word */ header, tmp, slow_case);
1028 } else if (LockingMode == LM_LEGACY) {
1029
1030 // Set header to be (markWord of object | UNLOCK_VALUE).
1031 // This will not change anything if it was unlocked before.
1032 z_oill(header, markWord::unlocked_value);
1033
1034 // monitor->lock()->set_displaced_header(displaced_header);
1035 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1036 const int mark_offset = lock_offset + BasicLock::displaced_header_offset_in_bytes();
1037
1038 // Initialize the box (Must happen before we update the object mark!).
1039 z_stg(header, mark_offset, monitor);
1040
1041 // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
1042
1043 // not necessary, use offset in instruction directly.
1044 // add2reg(object_mark_addr, hdr_offset, object);
1045
1046 // Store stack address of the BasicObjectLock (this is monitor) into object.
1047 z_csg(header, monitor, hdr_offset, object);
1048 assert(current_header == header,
1049 "must be same register"); // Identified two registers from z/Architecture.
1050
1051 z_bre(done);
1052
1053 // } else if (THREAD->is_lock_owned((address)displaced_header))
1054 // // Simple recursive case.
1055 // monitor->lock()->set_displaced_header(nullptr);
1056
1057 // We did not see an unlocked object so try the fast recursive case.
1058
1059 // Check if owner is self by comparing the value in the markWord of object
1060 // (current_header) with the stack pointer.
1061 z_sgr(current_header, Z_SP);
1062
1063 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1064
1065 // The prior sequence "LGR, NGR, LTGR" can be done better
1066 // (Z_R1 is temp and not used after here).
1067 load_const_optimized(Z_R0, (~(os::vm_page_size() - 1) | markWord::lock_mask_in_place));
1068 z_ngr(Z_R0, current_header); // AND sets CC (result eq/ne 0)
1069
1070 // If condition is true we are done and hence we can store 0 in the displaced
1071 // header indicating it is a recursive lock and be done.
1072 z_brne(slow_case);
1073 z_release(); // Member unnecessary on zarch AND because the above csg does a sync before and after.
1074 z_stg(Z_R0/*==0!*/, mark_offset, monitor);
1075 }
1076 z_bru(done);
1077 // } else {
1078 // // Slow path.
1079 // InterpreterRuntime::monitorenter(THREAD, monitor);
1080
1081 // None of the above fast optimizations worked so we have to get into the
1082 // slow case of monitor enter.
1083 bind(slow_case);
1084 if (LockingMode == LM_LIGHTWEIGHT) {
1085 // for lightweight locking we need to use monitorenter_obj, see interpreterRuntime.cpp
1086 call_VM(noreg,
1087 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj),
1088 object);
1089 } else {
1090 call_VM(noreg,
1091 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1092 monitor);
1093 }
1094 // }
1095
1096 bind(done);
1097 }
1098
1099 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1100 //
1101 // Registers alive
1102 // monitor - address of the BasicObjectLock to be used for locking,
1103 // which must be initialized with the object to lock.
1104 //
1105 // Throw IllegalMonitorException if object is not locked by current thread.
1106 void InterpreterMacroAssembler::unlock_object(Register monitor, Register object) {
1107
1108 if (LockingMode == LM_MONITOR) {
1109 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1110 return;
1111 }
1112
1113 // else {
1114 // template code: (for LM_LEGACY):
1115 //
1116 // if ((displaced_header = monitor->displaced_header()) == nullptr) {
1117 // // Recursive unlock. Mark the monitor unlocked by setting the object field to null.
1118 // monitor->set_obj(nullptr);
1119 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1120 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1121 // monitor->set_obj(nullptr);
1122 // } else {
1123 // // Slow path.
1124 // InterpreterRuntime::monitorexit(monitor);
1125 // }
1126
1127 const int hdr_offset = oopDesc::mark_offset_in_bytes();
1128
1129 const Register header = Z_ARG4;
1130 const Register current_header = Z_R1_scratch;
1131 Address obj_entry(monitor, BasicObjectLock::obj_offset());
1132 Label done, slow_case;
1133
1134 if (object == noreg) {
1135 // In the template interpreter, we must assure that the object
1136 // entry in the monitor is cleared on all paths. Thus we move
1137 // loading up to here, and clear the entry afterwards.
1138 object = Z_ARG3; // Use Z_ARG3 if caller didn't pass object.
1139 z_lg(object, obj_entry);
1140 }
1141
1142 assert_different_registers(monitor, object, header, current_header);
1143
1144 // if ((displaced_header = monitor->displaced_header()) == nullptr) {
1145 // // Recursive unlock. Mark the monitor unlocked by setting the object field to null.
1146 // monitor->set_obj(nullptr);
1147
1148 // monitor->lock()->set_displaced_header(displaced_header);
1149 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1150 const int mark_offset = lock_offset + BasicLock::displaced_header_offset_in_bytes();
1151
1152 clear_mem(obj_entry, sizeof(oop));
1153 if (LockingMode != LM_LIGHTWEIGHT) {
1154 // Test first if we are in the fast recursive case.
1155 MacroAssembler::load_and_test_long(header, Address(monitor, mark_offset));
1156 z_bre(done); // header == 0 -> goto done
1157 }
1158
1159 // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1160 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1161 // monitor->set_obj(nullptr);
1162
1163 // If we still have a lightweight lock, unlock the object and be done.
1164 if (LockingMode == LM_LIGHTWEIGHT) {
1165 // Check for non-symmetric locking. This is allowed by the spec and the interpreter
1166 // must handle it.
1167
1168 Register tmp = current_header;
1169
1170 // First check for lock-stack underflow.
1171 z_lgf(tmp, Address(Z_thread, JavaThread::lock_stack_top_offset()));
1172 compareU32_and_branch(tmp, (unsigned)LockStack::start_offset(), Assembler::bcondNotHigh, slow_case);
1173
1174 // Then check if the top of the lock-stack matches the unlocked object.
1175 z_aghi(tmp, -oopSize);
1176 z_lg(tmp, Address(Z_thread, tmp));
1177 compare64_and_branch(tmp, object, Assembler::bcondNotEqual, slow_case);
1178
1179 z_lg(header, Address(object, hdr_offset));
1180 z_lgr(tmp, header);
1181 z_nill(tmp, markWord::monitor_value);
1182 z_brne(slow_case);
1183
1184 lightweight_unlock(object, header, tmp, slow_case);
1185
1186 z_bru(done);
1187 } else {
1188 // The markword is expected to be at offset 0.
1189 // This is not required on s390, at least not here.
1190 assert(hdr_offset == 0, "unlock_object: review code below");
1191
1192 // We have the displaced header in header. If the lock is still
1193 // lightweight, it will contain the monitor address and we'll store the
1194 // displaced header back into the object's mark word.
1195 z_lgr(current_header, monitor);
1196 z_csg(current_header, header, hdr_offset, object);
1197 z_bre(done);
1198 }
1199
1200 // } else {
1201 // // Slow path.
1202 // InterpreterRuntime::monitorexit(monitor);
1203
1204 // The lock has been converted into a heavy lock and hence
1205 // we need to get into the slow case.
1206 bind(slow_case);
1207 z_stg(object, obj_entry); // Restore object entry, has been cleared above.
1208 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1209
1210 // }
1211
1212 bind(done);
1213 }
1214
1215 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) {
1216 assert(ProfileInterpreter, "must be profiling interpreter");
1217 load_and_test_long(mdp, Address(Z_fp, _z_ijava_state_neg(mdx)));
1218 z_brz(zero_continue);
1219 }
1220
1221 // Set the method data pointer for the current bcp.
1222 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1223 assert(ProfileInterpreter, "must be profiling interpreter");
1224 Label set_mdp;
1225 Register mdp = Z_ARG4;
1226 Register method = Z_ARG5;
1227
1228 get_method(method);
1229 // Test MDO to avoid the call if it is null.
1230 load_and_test_long(mdp, method2_(method, method_data));
1231 z_brz(set_mdp);
1232
1233 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), method, Z_bcp);
1234 // Z_RET: mdi
1235 // Mdo is guaranteed to be non-zero here, we checked for it before the call.
1236 assert(method->is_nonvolatile(), "choose nonvolatile reg or reload from frame");
1237 z_lg(mdp, method2_(method, method_data)); // Must reload, mdp is volatile reg.
1238 add2reg_with_index(mdp, in_bytes(MethodData::data_offset()), Z_RET, mdp);
1239
1240 bind(set_mdp);
1241 save_mdp(mdp);
1242 }
1243
1244 void InterpreterMacroAssembler::verify_method_data_pointer() {
1245 assert(ProfileInterpreter, "must be profiling interpreter");
1246 #ifdef ASSERT
1247 NearLabel verify_continue;
1248 Register bcp_expected = Z_ARG3;
1249 Register mdp = Z_ARG4;
1250 Register method = Z_ARG5;
1251
1252 test_method_data_pointer(mdp, verify_continue); // If mdp is zero, continue
1253 get_method(method);
1254
1255 // If the mdp is valid, it will point to a DataLayout header which is
1256 // consistent with the bcp. The converse is highly probable also.
1257 load_sized_value(bcp_expected, Address(mdp, DataLayout::bci_offset()), 2, false /*signed*/);
1258 z_ag(bcp_expected, Address(method, Method::const_offset()));
1259 load_address(bcp_expected, Address(bcp_expected, ConstMethod::codes_offset()));
1260 compareU64_and_branch(bcp_expected, Z_bcp, bcondEqual, verify_continue);
1261 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), method, Z_bcp, mdp);
1262 bind(verify_continue);
1263 #endif // ASSERT
1264 }
1265
1266 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) {
1267 assert(ProfileInterpreter, "must be profiling interpreter");
1268 z_stg(value, constant, mdp_in);
1269 }
1270
1271 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1272 int constant,
1273 Register tmp,
1274 bool decrement) {
1275 assert_different_registers(mdp_in, tmp);
1276 // counter address
1277 Address data(mdp_in, constant);
1278 const int delta = decrement ? -DataLayout::counter_increment : DataLayout::counter_increment;
1279 add2mem_64(Address(mdp_in, constant), delta, tmp);
1280 }
1281
1282 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1283 int flag_byte_constant) {
1284 assert(ProfileInterpreter, "must be profiling interpreter");
1285 // Set the flag.
1286 z_oi(Address(mdp_in, DataLayout::flags_offset()), flag_byte_constant);
1287 }
1288
1289 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1290 int offset,
1291 Register value,
1292 Register test_value_out,
1293 Label& not_equal_continue) {
1294 assert(ProfileInterpreter, "must be profiling interpreter");
1295 if (test_value_out == noreg) {
1296 z_cg(value, Address(mdp_in, offset));
1297 z_brne(not_equal_continue);
1298 } else {
1299 // Put the test value into a register, so caller can use it:
1300 z_lg(test_value_out, Address(mdp_in, offset));
1301 compareU64_and_branch(test_value_out, value, bcondNotEqual, not_equal_continue);
1302 }
1303 }
1304
1305 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) {
1306 update_mdp_by_offset(mdp_in, noreg, offset_of_disp);
1307 }
1308
1309 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1310 Register dataidx,
1311 int offset_of_disp) {
1312 assert(ProfileInterpreter, "must be profiling interpreter");
1313 Address disp_address(mdp_in, dataidx, offset_of_disp);
1314 Assembler::z_ag(mdp_in, disp_address);
1315 save_mdp(mdp_in);
1316 }
1317
1318 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) {
1319 assert(ProfileInterpreter, "must be profiling interpreter");
1320 add2reg(mdp_in, constant);
1321 save_mdp(mdp_in);
1322 }
1323
1324 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1325 assert(ProfileInterpreter, "must be profiling interpreter");
1326 assert(return_bci->is_nonvolatile(), "choose nonvolatile reg or save/restore");
1327 call_VM(noreg,
1328 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1329 return_bci);
1330 }
1331
1332 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) {
1333 if (ProfileInterpreter) {
1334 Label profile_continue;
1335
1336 // If no method data exists, go to profile_continue.
1337 // Otherwise, assign to mdp.
1338 test_method_data_pointer(mdp, profile_continue);
1339
1340 // We are taking a branch. Increment the taken count.
1341 // We inline increment_mdp_data_at to return bumped_count in a register
1342 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1343 Address data(mdp, JumpData::taken_offset());
1344 z_lg(bumped_count, data);
1345 // 64-bit overflow is very unlikely. Saturation to 32-bit values is
1346 // performed when reading the counts.
1347 add2reg(bumped_count, DataLayout::counter_increment);
1348 z_stg(bumped_count, data); // Store back out
1349
1350 // The method data pointer needs to be updated to reflect the new target.
1351 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1352 bind(profile_continue);
1353 }
1354 }
1355
1356 // Kills Z_R1_scratch.
1357 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1358 if (ProfileInterpreter) {
1359 Label profile_continue;
1360
1361 // If no method data exists, go to profile_continue.
1362 test_method_data_pointer(mdp, profile_continue);
1363
1364 // We are taking a branch. Increment the not taken count.
1365 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()), Z_R1_scratch);
1366
1367 // The method data pointer needs to be updated to correspond to
1368 // the next bytecode.
1369 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1370 bind(profile_continue);
1371 }
1372 }
1373
1374 // Kills: Z_R1_scratch.
1375 void InterpreterMacroAssembler::profile_call(Register mdp) {
1376 if (ProfileInterpreter) {
1377 Label profile_continue;
1378
1379 // If no method data exists, go to profile_continue.
1380 test_method_data_pointer(mdp, profile_continue);
1381
1382 // We are making a call. Increment the count.
1383 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1384
1385 // The method data pointer needs to be updated to reflect the new target.
1386 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1387 bind(profile_continue);
1388 }
1389 }
1390
1391 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1392 if (ProfileInterpreter) {
1393 Label profile_continue;
1394
1395 // If no method data exists, go to profile_continue.
1396 test_method_data_pointer(mdp, profile_continue);
1397
1398 // We are making a call. Increment the count.
1399 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1400
1401 // The method data pointer needs to be updated to reflect the new target.
1402 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1403 bind(profile_continue);
1404 }
1405 }
1406
1407 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1408 Register mdp,
1409 Register reg2,
1410 bool receiver_can_be_null) {
1411 if (ProfileInterpreter) {
1412 NearLabel profile_continue;
1413
1414 // If no method data exists, go to profile_continue.
1415 test_method_data_pointer(mdp, profile_continue);
1416
1417 NearLabel skip_receiver_profile;
1418 if (receiver_can_be_null) {
1419 NearLabel not_null;
1420 compareU64_and_branch(receiver, (intptr_t)0L, bcondNotEqual, not_null);
1421 // We are making a call. Increment the count for null receiver.
1422 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1423 z_bru(skip_receiver_profile);
1424 bind(not_null);
1425 }
1426
1427 // Record the receiver type.
1428 record_klass_in_profile(receiver, mdp, reg2, true);
1429 bind(skip_receiver_profile);
1430
1431 // The method data pointer needs to be updated to reflect the new target.
1432 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1433 bind(profile_continue);
1434 }
1435 }
1436
1437 // This routine creates a state machine for updating the multi-row
1438 // type profile at a virtual call site (or other type-sensitive bytecode).
1439 // The machine visits each row (of receiver/count) until the receiver type
1440 // is found, or until it runs out of rows. At the same time, it remembers
1441 // the location of the first empty row. (An empty row records null for its
1442 // receiver, and can be allocated for a newly-observed receiver type.)
1443 // Because there are two degrees of freedom in the state, a simple linear
1444 // search will not work; it must be a decision tree. Hence this helper
1445 // function is recursive, to generate the required tree structured code.
1446 // It's the interpreter, so we are trading off code space for speed.
1447 // See below for example code.
1448 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1449 Register receiver, Register mdp,
1450 Register reg2, int start_row,
1451 Label& done, bool is_virtual_call) {
1452 if (TypeProfileWidth == 0) {
1453 if (is_virtual_call) {
1454 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1455 }
1456 return;
1457 }
1458
1459 int last_row = VirtualCallData::row_limit() - 1;
1460 assert(start_row <= last_row, "must be work left to do");
1461 // Test this row for both the receiver and for null.
1462 // Take any of three different outcomes:
1463 // 1. found receiver => increment count and goto done
1464 // 2. found null => keep looking for case 1, maybe allocate this cell
1465 // 3. found something else => keep looking for cases 1 and 2
1466 // Case 3 is handled by a recursive call.
1467 for (int row = start_row; row <= last_row; row++) {
1468 NearLabel next_test;
1469 bool test_for_null_also = (row == start_row);
1470
1471 // See if the receiver is receiver[n].
1472 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1473 test_mdp_data_at(mdp, recvr_offset, receiver,
1474 (test_for_null_also ? reg2 : noreg),
1475 next_test);
1476 // (Reg2 now contains the receiver from the CallData.)
1477
1478 // The receiver is receiver[n]. Increment count[n].
1479 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1480 increment_mdp_data_at(mdp, count_offset);
1481 z_bru(done);
1482 bind(next_test);
1483
1484 if (test_for_null_also) {
1485 Label found_null;
1486 // Failed the equality check on receiver[n]... Test for null.
1487 z_ltgr(reg2, reg2);
1488 if (start_row == last_row) {
1489 // The only thing left to do is handle the null case.
1490 if (is_virtual_call) {
1491 z_brz(found_null);
1492 // Receiver did not match any saved receiver and there is no empty row for it.
1493 // Increment total counter to indicate polymorphic case.
1494 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1495 z_bru(done);
1496 bind(found_null);
1497 } else {
1498 z_brnz(done);
1499 }
1500 break;
1501 }
1502 // Since null is rare, make it be the branch-taken case.
1503 z_brz(found_null);
1504
1505 // Put all the "Case 3" tests here.
1506 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
1507
1508 // Found a null. Keep searching for a matching receiver,
1509 // but remember that this is an empty (unused) slot.
1510 bind(found_null);
1511 }
1512 }
1513
1514 // In the fall-through case, we found no matching receiver, but we
1515 // observed the receiver[start_row] is null.
1516
1517 // Fill in the receiver field and increment the count.
1518 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1519 set_mdp_data_at(mdp, recvr_offset, receiver);
1520 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1521 load_const_optimized(reg2, DataLayout::counter_increment);
1522 set_mdp_data_at(mdp, count_offset, reg2);
1523 if (start_row > 0) {
1524 z_bru(done);
1525 }
1526 }
1527
1528 // Example state machine code for three profile rows:
1529 // // main copy of decision tree, rooted at row[1]
1530 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1531 // if (row[0].rec != nullptr) {
1532 // // inner copy of decision tree, rooted at row[1]
1533 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1534 // if (row[1].rec != nullptr) {
1535 // // degenerate decision tree, rooted at row[2]
1536 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1537 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1538 // row[2].init(rec); goto done;
1539 // } else {
1540 // // remember row[1] is empty
1541 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1542 // row[1].init(rec); goto done;
1543 // }
1544 // } else {
1545 // // remember row[0] is empty
1546 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1547 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1548 // row[0].init(rec); goto done;
1549 // }
1550 // done:
1551
1552 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1553 Register mdp, Register reg2,
1554 bool is_virtual_call) {
1555 assert(ProfileInterpreter, "must be profiling");
1556 Label done;
1557
1558 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1559
1560 bind (done);
1561 }
1562
1563 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
1564 if (ProfileInterpreter) {
1565 NearLabel profile_continue;
1566 uint row;
1567
1568 // If no method data exists, go to profile_continue.
1569 test_method_data_pointer(mdp, profile_continue);
1570
1571 // Update the total ret count.
1572 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1573
1574 for (row = 0; row < RetData::row_limit(); row++) {
1575 NearLabel next_test;
1576
1577 // See if return_bci is equal to bci[n]:
1578 test_mdp_data_at(mdp,
1579 in_bytes(RetData::bci_offset(row)),
1580 return_bci, noreg,
1581 next_test);
1582
1583 // Return_bci is equal to bci[n]. Increment the count.
1584 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1585
1586 // The method data pointer needs to be updated to reflect the new target.
1587 update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row)));
1588 z_bru(profile_continue);
1589 bind(next_test);
1590 }
1591
1592 update_mdp_for_ret(return_bci);
1593
1594 bind(profile_continue);
1595 }
1596 }
1597
1598 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1599 if (ProfileInterpreter) {
1600 Label profile_continue;
1601
1602 // If no method data exists, go to profile_continue.
1603 test_method_data_pointer(mdp, profile_continue);
1604
1605 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1606
1607 // The method data pointer needs to be updated.
1608 int mdp_delta = in_bytes(BitData::bit_data_size());
1609 if (TypeProfileCasts) {
1610 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1611 }
1612 update_mdp_by_constant(mdp, mdp_delta);
1613
1614 bind(profile_continue);
1615 }
1616 }
1617
1618 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp, Register tmp) {
1619 if (ProfileInterpreter && TypeProfileCasts) {
1620 Label profile_continue;
1621
1622 // If no method data exists, go to profile_continue.
1623 test_method_data_pointer(mdp, profile_continue);
1624
1625 int count_offset = in_bytes(CounterData::count_offset());
1626 // Back up the address, since we have already bumped the mdp.
1627 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1628
1629 // *Decrement* the counter. We expect to see zero or small negatives.
1630 increment_mdp_data_at(mdp, count_offset, tmp, true);
1631
1632 bind (profile_continue);
1633 }
1634 }
1635
1636 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1637 if (ProfileInterpreter) {
1638 Label profile_continue;
1639
1640 // If no method data exists, go to profile_continue.
1641 test_method_data_pointer(mdp, profile_continue);
1642
1643 // The method data pointer needs to be updated.
1644 int mdp_delta = in_bytes(BitData::bit_data_size());
1645 if (TypeProfileCasts) {
1646 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1647
1648 // Record the object type.
1649 record_klass_in_profile(klass, mdp, reg2, false);
1650 }
1651 update_mdp_by_constant(mdp, mdp_delta);
1652
1653 bind(profile_continue);
1654 }
1655 }
1656
1657 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1658 if (ProfileInterpreter) {
1659 Label profile_continue;
1660
1661 // If no method data exists, go to profile_continue.
1662 test_method_data_pointer(mdp, profile_continue);
1663
1664 // Update the default case count.
1665 increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset()));
1666
1667 // The method data pointer needs to be updated.
1668 update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset()));
1669
1670 bind(profile_continue);
1671 }
1672 }
1673
1674 // Kills: index, scratch1, scratch2.
1675 void InterpreterMacroAssembler::profile_switch_case(Register index,
1676 Register mdp,
1677 Register scratch1,
1678 Register scratch2) {
1679 if (ProfileInterpreter) {
1680 Label profile_continue;
1681 assert_different_registers(index, mdp, scratch1, scratch2);
1682
1683 // If no method data exists, go to profile_continue.
1684 test_method_data_pointer(mdp, profile_continue);
1685
1686 // Build the base (index * per_case_size_in_bytes()) +
1687 // case_array_offset_in_bytes().
1688 z_sllg(index, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1689 add2reg(index, in_bytes(MultiBranchData::case_array_offset()));
1690
1691 // Add the calculated base to the mdp -> address of the case' data.
1692 Address case_data_addr(mdp, index);
1693 Register case_data = scratch1;
1694 load_address(case_data, case_data_addr);
1695
1696 // Update the case count.
1697 increment_mdp_data_at(case_data,
1698 in_bytes(MultiBranchData::relative_count_offset()),
1699 scratch2);
1700
1701 // The method data pointer needs to be updated.
1702 update_mdp_by_offset(mdp,
1703 index,
1704 in_bytes(MultiBranchData::relative_displacement_offset()));
1705
1706 bind(profile_continue);
1707 }
1708 }
1709
1710 // kills: R0, R1, flags, loads klass from obj (if not null)
1711 void InterpreterMacroAssembler::profile_obj_type(Register obj, Address mdo_addr, Register klass, bool cmp_done) {
1712 NearLabel null_seen, init_klass, do_nothing, do_update;
1713
1714 // Klass = obj is allowed.
1715 const Register tmp = Z_R1;
1716 assert_different_registers(obj, mdo_addr.base(), tmp, Z_R0);
1717 assert_different_registers(klass, mdo_addr.base(), tmp, Z_R0);
1718
1719 z_lg(tmp, mdo_addr);
1720 if (cmp_done) {
1721 z_brz(null_seen);
1722 } else {
1723 compareU64_and_branch(obj, (intptr_t)0, Assembler::bcondEqual, null_seen);
1724 }
1725
1726 MacroAssembler::verify_oop(obj, FILE_AND_LINE);
1727 load_klass(klass, obj);
1728
1729 // Klass seen before, nothing to do (regardless of unknown bit).
1730 z_lgr(Z_R0, tmp);
1731 assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
1732 z_nill(Z_R0, TypeEntries::type_klass_mask & 0xFFFF);
1733 compareU64_and_branch(Z_R0, klass, Assembler::bcondEqual, do_nothing);
1734
1735 // Already unknown. Nothing to do anymore.
1736 z_tmll(tmp, TypeEntries::type_unknown);
1737 z_brc(Assembler::bcondAllOne, do_nothing);
1738
1739 z_lgr(Z_R0, tmp);
1740 assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
1741 z_nill(Z_R0, TypeEntries::type_mask & 0xFFFF);
1742 compareU64_and_branch(Z_R0, (intptr_t)0, Assembler::bcondEqual, init_klass);
1743
1744 // Different than before. Cannot keep accurate profile.
1745 z_oill(tmp, TypeEntries::type_unknown);
1746 z_bru(do_update);
1747
1748 bind(init_klass);
1749 // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1750 z_ogr(tmp, klass);
1751 z_bru(do_update);
1752
1753 bind(null_seen);
1754 // Set null_seen if obj is 0.
1755 z_oill(tmp, TypeEntries::null_seen);
1756 // fallthru: z_bru(do_update);
1757
1758 bind(do_update);
1759 z_stg(tmp, mdo_addr);
1760
1761 bind(do_nothing);
1762 }
1763
1764 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1765 if (!ProfileInterpreter) {
1766 return;
1767 }
1768
1769 assert_different_registers(mdp, callee, tmp);
1770
1771 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1772 Label profile_continue;
1773
1774 test_method_data_pointer(mdp, profile_continue);
1775
1776 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1777
1778 z_cliy(in_bytes(DataLayout::tag_offset()) - off_to_start, mdp,
1779 is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1780 z_brne(profile_continue);
1781
1782 if (MethodData::profile_arguments()) {
1783 NearLabel done;
1784 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1785 add2reg(mdp, off_to_args);
1786
1787 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1788 if (i > 0 || MethodData::profile_return()) {
1789 // If return value type is profiled we may have no argument to profile.
1790 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1791 add2reg(tmp, -i*TypeStackSlotEntries::per_arg_count());
1792 compare64_and_branch(tmp, TypeStackSlotEntries::per_arg_count(), Assembler::bcondLow, done);
1793 }
1794 z_lg(tmp, Address(callee, Method::const_offset()));
1795 z_lgh(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1796 // Stack offset o (zero based) from the start of the argument
1797 // list. For n arguments translates into offset n - o - 1 from
1798 // the end of the argument list. But there is an extra slot at
1799 // the top of the stack. So the offset is n - o from Lesp.
1800 z_sg(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
1801 z_sllg(tmp, tmp, Interpreter::logStackElementSize);
1802 Address stack_slot_addr(tmp, Z_esp);
1803 z_ltg(tmp, stack_slot_addr);
1804
1805 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
1806 profile_obj_type(tmp, mdo_arg_addr, tmp, /*ltg did compare to 0*/ true);
1807
1808 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1809 add2reg(mdp, to_add);
1810 off_to_args += to_add;
1811 }
1812
1813 if (MethodData::profile_return()) {
1814 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1815 add2reg(tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1816 }
1817
1818 bind(done);
1819
1820 if (MethodData::profile_return()) {
1821 // We're right after the type profile for the last
1822 // argument. Tmp is the number of cells left in the
1823 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1824 // if there's a return to profile.
1825 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1826 z_sllg(tmp, tmp, exact_log2(DataLayout::cell_size));
1827 z_agr(mdp, tmp);
1828 }
1829 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
1830 } else {
1831 assert(MethodData::profile_return(), "either profile call args or call ret");
1832 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1833 }
1834
1835 // Mdp points right after the end of the
1836 // CallTypeData/VirtualCallTypeData, right after the cells for the
1837 // return value type if there's one.
1838 bind(profile_continue);
1839 }
1840 }
1841
1842 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1843 assert_different_registers(mdp, ret, tmp);
1844 if (ProfileInterpreter && MethodData::profile_return()) {
1845 Label profile_continue;
1846
1847 test_method_data_pointer(mdp, profile_continue);
1848
1849 if (MethodData::profile_return_jsr292_only()) {
1850 // If we don't profile all invoke bytecodes we must make sure
1851 // it's a bytecode we indeed profile. We can't go back to the
1852 // beginning of the ProfileData we intend to update to check its
1853 // type because we're right after it and we don't known its
1854 // length.
1855 NearLabel do_profile;
1856 Address bc(Z_bcp);
1857 z_lb(tmp, bc);
1858 compare32_and_branch(tmp, Bytecodes::_invokedynamic, Assembler::bcondEqual, do_profile);
1859 compare32_and_branch(tmp, Bytecodes::_invokehandle, Assembler::bcondEqual, do_profile);
1860 get_method(tmp);
1861 // Supplement to 8139891: _intrinsic_id exceeded 1-byte size limit.
1862 if (Method::intrinsic_id_size_in_bytes() == 1) {
1863 z_cli(in_bytes(Method::intrinsic_id_offset()), tmp, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1864 } else {
1865 assert(Method::intrinsic_id_size_in_bytes() == 2, "size error: check Method::_intrinsic_id");
1866 z_lh(tmp, in_bytes(Method::intrinsic_id_offset()), Z_R0, tmp);
1867 z_chi(tmp, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1868 }
1869 z_brne(profile_continue);
1870
1871 bind(do_profile);
1872 }
1873
1874 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
1875 profile_obj_type(ret, mdo_ret_addr, tmp);
1876
1877 bind(profile_continue);
1878 }
1879 }
1880
1881 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1882 if (ProfileInterpreter && MethodData::profile_parameters()) {
1883 Label profile_continue, done;
1884
1885 test_method_data_pointer(mdp, profile_continue);
1886
1887 // Load the offset of the area within the MDO used for
1888 // parameters. If it's negative we're not profiling any parameters.
1889 Address parm_di_addr(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()));
1890 load_and_test_int2long(tmp1, parm_di_addr);
1891 z_brl(profile_continue);
1892
1893 // Compute a pointer to the area for parameters from the offset
1894 // and move the pointer to the slot for the last
1895 // parameters. Collect profiling from last parameter down.
1896 // mdo start + parameters offset + array length - 1
1897
1898 // Pointer to the parameter area in the MDO.
1899 z_agr(mdp, tmp1);
1900
1901 // Offset of the current profile entry to update.
1902 const Register entry_offset = tmp1;
1903 // entry_offset = array len in number of cells.
1904 z_lg(entry_offset, Address(mdp, ArrayData::array_len_offset()));
1905 // entry_offset (number of cells) = array len - size of 1 entry
1906 add2reg(entry_offset, -TypeStackSlotEntries::per_arg_count());
1907 // entry_offset in bytes
1908 z_sllg(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1909
1910 Label loop;
1911 bind(loop);
1912
1913 Address arg_off(mdp, entry_offset, ParametersTypeData::stack_slot_offset(0));
1914 Address arg_type(mdp, entry_offset, ParametersTypeData::type_offset(0));
1915
1916 // Load offset on the stack from the slot for this parameter.
1917 z_lg(tmp2, arg_off);
1918 z_sllg(tmp2, tmp2, Interpreter::logStackElementSize);
1919 z_lcgr(tmp2); // Negate.
1920
1921 // Profile the parameter.
1922 z_ltg(tmp2, Address(Z_locals, tmp2));
1923 profile_obj_type(tmp2, arg_type, tmp2, /*ltg did compare to 0*/ true);
1924
1925 // Go to next parameter.
1926 z_aghi(entry_offset, -TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size);
1927 z_brnl(loop);
1928
1929 bind(profile_continue);
1930 }
1931 }
1932
1933 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1934 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1935 int increment,
1936 Address mask,
1937 Register scratch,
1938 bool preloaded,
1939 branch_condition cond,
1940 Label *where) {
1941 assert_different_registers(counter_addr.base(), scratch);
1942 if (preloaded) {
1943 add2reg(scratch, increment);
1944 reg2mem_opt(scratch, counter_addr, false);
1945 } else {
1946 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment) && counter_addr.is_RSYform()) {
1947 z_alsi(counter_addr.disp20(), counter_addr.base(), increment);
1948 mem2reg_signed_opt(scratch, counter_addr);
1949 } else {
1950 mem2reg_signed_opt(scratch, counter_addr);
1951 add2reg(scratch, increment);
1952 reg2mem_opt(scratch, counter_addr, false);
1953 }
1954 }
1955 z_n(scratch, mask);
1956 if (where) { z_brc(cond, *where); }
1957 }
1958
1959 // Get MethodCounters object for given method. Lazily allocated if necessary.
1960 // method - Ptr to Method object.
1961 // Rcounters - Ptr to MethodCounters object associated with Method object.
1962 // skip - Exit point if MethodCounters object can't be created (OOM condition).
1963 void InterpreterMacroAssembler::get_method_counters(Register Rmethod,
1964 Register Rcounters,
1965 Label& skip) {
1966 assert_different_registers(Rmethod, Rcounters);
1967
1968 BLOCK_COMMENT("get MethodCounters object {");
1969
1970 Label has_counters;
1971 load_and_test_long(Rcounters, Address(Rmethod, Method::method_counters_offset()));
1972 z_brnz(has_counters);
1973
1974 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), Rmethod);
1975 z_ltgr(Rcounters, Z_RET); // Runtime call returns MethodCounters object.
1976 z_brz(skip); // No MethodCounters, out of memory.
1977
1978 bind(has_counters);
1979
1980 BLOCK_COMMENT("} get MethodCounters object");
1981 }
1982
1983 // Increment invocation counter in MethodCounters object.
1984 // Return (invocation_counter+backedge_counter) as "result" in RctrSum.
1985 // Counter values are all unsigned.
1986 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register RctrSum) {
1987 assert(UseCompiler, "incrementing must be useful");
1988 assert_different_registers(Rcounters, RctrSum);
1989
1990 int increment = InvocationCounter::count_increment;
1991 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
1992 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset());
1993
1994 BLOCK_COMMENT("Increment invocation counter {");
1995
1996 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
1997 // Increment the invocation counter in place,
1998 // then add the incremented value to the backedge counter.
1999 z_l(RctrSum, be_counter_offset, Rcounters);
2000 z_alsi(inv_counter_offset, Rcounters, increment); // Atomic increment @no extra cost!
2001 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
2002 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
2003 } else {
2004 // This path is optimized for low register consumption
2005 // at the cost of somewhat higher operand delays.
2006 // It does not need an extra temp register.
2007
2008 // Update the invocation counter.
2009 z_l(RctrSum, inv_counter_offset, Rcounters);
2010 if (RctrSum == Z_R0) {
2011 z_ahi(RctrSum, increment);
2012 } else {
2013 add2reg(RctrSum, increment);
2014 }
2015 z_st(RctrSum, inv_counter_offset, Rcounters);
2016
2017 // Mask off the state bits.
2018 z_nilf(RctrSum, InvocationCounter::count_mask_value);
2019
2020 // Add the backedge counter to the updated invocation counter to
2021 // form the result.
2022 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
2023 }
2024
2025 BLOCK_COMMENT("} Increment invocation counter");
2026
2027 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2028 }
2029
2030
2031 // increment backedge counter in MethodCounters object.
2032 // return (invocation_counter+backedge_counter) as "result" in RctrSum
2033 // counter values are all unsigned!
2034 void InterpreterMacroAssembler::increment_backedge_counter(Register Rcounters, Register RctrSum) {
2035 assert(UseCompiler, "incrementing must be useful");
2036 assert_different_registers(Rcounters, RctrSum);
2037
2038 int increment = InvocationCounter::count_increment;
2039 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
2040 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset());
2041
2042 BLOCK_COMMENT("Increment backedge counter {");
2043
2044 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
2045 // Increment the invocation counter in place,
2046 // then add the incremented value to the backedge counter.
2047 z_l(RctrSum, inv_counter_offset, Rcounters);
2048 z_alsi(be_counter_offset, Rcounters, increment); // Atomic increment @no extra cost!
2049 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
2050 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
2051 } else {
2052 // This path is optimized for low register consumption
2053 // at the cost of somewhat higher operand delays.
2054 // It does not need an extra temp register.
2055
2056 // Update the invocation counter.
2057 z_l(RctrSum, be_counter_offset, Rcounters);
2058 if (RctrSum == Z_R0) {
2059 z_ahi(RctrSum, increment);
2060 } else {
2061 add2reg(RctrSum, increment);
2062 }
2063 z_st(RctrSum, be_counter_offset, Rcounters);
2064
2065 // Mask off the state bits.
2066 z_nilf(RctrSum, InvocationCounter::count_mask_value);
2067
2068 // Add the backedge counter to the updated invocation counter to
2069 // form the result.
2070 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
2071 }
2072
2073 BLOCK_COMMENT("} Increment backedge counter");
2074
2075 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2076 }
2077
2078 // Add an InterpMonitorElem to stack (see frame_s390.hpp).
2079 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty,
2080 Register Rtemp1,
2081 Register Rtemp2,
2082 Register Rtemp3) {
2083
2084 const Register Rcurr_slot = Rtemp1;
2085 const Register Rlimit = Rtemp2;
2086 const jint delta = -frame::interpreter_frame_monitor_size_in_bytes();
2087
2088 assert((delta & LongAlignmentMask) == 0,
2089 "sizeof BasicObjectLock must be even number of doublewords");
2090 assert(2 * wordSize == -delta, "this works only as long as delta == -2*wordSize");
2091 assert(Rcurr_slot != Z_R0, "Register must be usable as base register");
2092 assert_different_registers(Rlimit, Rcurr_slot, Rtemp3);
2093
2094 get_monitors(Rlimit);
2095
2096 // Adjust stack pointer for additional monitor entry.
2097 resize_frame(RegisterOrConstant((intptr_t) delta), Z_fp, false);
2098
2099 if (!stack_is_empty) {
2100 // Must copy stack contents down.
2101 NearLabel next, done;
2102
2103 // Rtemp := addr(Tos), Z_esp is pointing below it!
2104 add2reg(Rcurr_slot, wordSize, Z_esp);
2105
2106 // Nothing to do, if already at monitor area.
2107 compareU64_and_branch(Rcurr_slot, Rlimit, bcondNotLow, done);
2108
2109 bind(next);
2110
2111 // Move one stack slot.
2112 mem2reg_opt(Rtemp3, Address(Rcurr_slot));
2113 reg2mem_opt(Rtemp3, Address(Rcurr_slot, delta));
2114 add2reg(Rcurr_slot, wordSize);
2115 compareU64_and_branch(Rcurr_slot, Rlimit, bcondLow, next); // Are we done?
2116
2117 bind(done);
2118 // Done copying stack.
2119 }
2120
2121 // Adjust expression stack and monitor pointers.
2122 add2reg(Z_esp, delta);
2123 add2reg(Rlimit, delta);
2124 save_monitors(Rlimit);
2125 }
2126
2127 // Note: Index holds the offset in bytes afterwards.
2128 // You can use this to store a new value (with Llocals as the base).
2129 void InterpreterMacroAssembler::access_local_int(Register index, Register dst) {
2130 z_sllg(index, index, LogBytesPerWord);
2131 mem2reg_opt(dst, Address(Z_locals, index), false);
2132 }
2133
2134 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2135 if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); }
2136 }
2137
2138 // Inline assembly for:
2139 //
2140 // if (thread is in interp_only_mode) {
2141 // InterpreterRuntime::post_method_entry();
2142 // }
2143
2144 void InterpreterMacroAssembler::notify_method_entry() {
2145
2146 // JVMTI
2147 // Whenever JVMTI puts a thread in interp_only_mode, method
2148 // entry/exit events are sent for that thread to track stack
2149 // depth. If it is possible to enter interp_only_mode we add
2150 // the code to check if the event should be sent.
2151 if (JvmtiExport::can_post_interpreter_events()) {
2152 Label jvmti_post_done;
2153 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2154 z_bre(jvmti_post_done);
2155 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2156 bind(jvmti_post_done);
2157 }
2158 }
2159
2160 // Inline assembly for:
2161 //
2162 // if (thread is in interp_only_mode) {
2163 // if (!native_method) save result
2164 // InterpreterRuntime::post_method_exit();
2165 // if (!native_method) restore result
2166 // }
2167 // if (DTraceMethodProbes) {
2168 // SharedRuntime::dtrace_method_exit(thread, method);
2169 // }
2170 //
2171 // For native methods their result is stored in z_ijava_state.lresult
2172 // and z_ijava_state.fresult before coming here.
2173 // Java methods have their result stored in the expression stack.
2174 //
2175 // Notice the dependency to frame::interpreter_frame_result().
2176 void InterpreterMacroAssembler::notify_method_exit(bool native_method,
2177 TosState state,
2178 NotifyMethodExitMode mode) {
2179 // JVMTI
2180 // Whenever JVMTI puts a thread in interp_only_mode, method
2181 // entry/exit events are sent for that thread to track stack
2182 // depth. If it is possible to enter interp_only_mode we add
2183 // the code to check if the event should be sent.
2184 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2185 Label jvmti_post_done;
2186 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2187 z_bre(jvmti_post_done);
2188 if (!native_method) push(state); // see frame::interpreter_frame_result()
2189 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2190 if (!native_method) pop(state);
2191 bind(jvmti_post_done);
2192 }
2193
2194 #if 0
2195 // Dtrace currently not supported on z/Architecture.
2196 {
2197 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2198 push(state);
2199 get_method(c_rarg1);
2200 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2201 r15_thread, c_rarg1);
2202 pop(state);
2203 }
2204 #endif
2205 }
2206
2207 void InterpreterMacroAssembler::skip_if_jvmti_mode(Label &Lskip, Register Rscratch) {
2208 if (!JvmtiExport::can_post_interpreter_events()) {
2209 return;
2210 }
2211
2212 load_and_test_int(Rscratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2213 z_brnz(Lskip);
2214
2215 }
2216
2217 // Pop the topmost TOP_IJAVA_FRAME and set it's sender_sp as new Z_SP.
2218 // The return pc is loaded into the register return_pc.
2219 //
2220 // Registers updated:
2221 // return_pc - The return pc of the calling frame.
2222 // tmp1, tmp2 - scratch
2223 void InterpreterMacroAssembler::pop_interpreter_frame(Register return_pc, Register tmp1, Register tmp2) {
2224 // F0 Z_SP -> caller_sp (F1's)
2225 // ...
2226 // sender_sp (F1's)
2227 // ...
2228 // F1 Z_fp -> caller_sp (F2's)
2229 // return_pc (Continuation after return from F0.)
2230 // ...
2231 // F2 caller_sp
2232
2233 // Remove F0's activation. Restoring Z_SP to sender_sp reverts modifications
2234 // (a) by a c2i adapter and (b) by generate_fixed_frame().
2235 // In case (a) the new top frame F1 is an unextended compiled frame.
2236 // In case (b) F1 is converted from PARENT_IJAVA_FRAME to TOP_IJAVA_FRAME.
2237
2238 // Case (b) seems to be redundant when returning to a interpreted caller,
2239 // because then the caller's top_frame_sp is installed as sp (see
2240 // TemplateInterpreterGenerator::generate_return_entry_for ()). But
2241 // pop_interpreter_frame() is also used in exception handling and there the
2242 // frame type of the caller is unknown, therefore top_frame_sp cannot be used,
2243 // so it is important that sender_sp is the caller's sp as TOP_IJAVA_FRAME.
2244
2245 Register R_f1_sender_sp = tmp1;
2246 Register R_f2_sp = tmp2;
2247
2248 // First check for the interpreter frame's magic.
2249 asm_assert_ijava_state_magic(R_f2_sp/*tmp*/);
2250 z_lg(R_f2_sp, _z_parent_ijava_frame_abi(callers_sp), Z_fp);
2251 z_lg(R_f1_sender_sp, _z_ijava_state_neg(sender_sp), Z_fp);
2252 if (return_pc->is_valid())
2253 z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2254 // Pop F0 by resizing to R_f1_sender_sp and using R_f2_sp as fp.
2255 resize_frame_absolute(R_f1_sender_sp, R_f2_sp, false/*load fp*/);
2256
2257 #ifdef ASSERT
2258 // The return_pc in the new top frame is dead... at least that's my
2259 // current understanding; to assert this I overwrite it.
2260 load_const_optimized(Z_ARG3, 0xb00b1);
2261 z_stg(Z_ARG3, _z_parent_ijava_frame_abi(return_pc), Z_SP);
2262 #endif
2263 }
2264
2265 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2266 if (VerifyFPU) {
2267 unimplemented("verifyFPU");
2268 }
2269 }