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