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 #ifdef ASSERT
579 NearLabel ok;
580 Register tmp = Z_R1;
581 z_lg(tmp, Address(Z_fp, _z_ijava_state_neg(top_frame_sp)));
582 z_slag(tmp, tmp, Interpreter::logStackElementSize);
583 z_agr(tmp, Z_fp);
584 z_cgr(tmp, Z_SP);
585 z_bre(ok);
586 stop("corrupted top_frame_sp");
587 bind(ok);
588 #endif
589 save_bcp();
590 save_esp();
591 z_lgr(Z_method, method); // Set Z_method (kills Z_fp!).
592 }
593
594 // Jump to from_interpreted entry of a call unless single stepping is possible
595 // in this thread in which case we must call the i2i entry.
596 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
597 assert_different_registers(method, Z_R10 /*used for initial_caller_sp*/, temp);
598 prepare_to_jump_from_interpreted(method);
599
600 if (JvmtiExport::can_post_interpreter_events()) {
601 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
602 // compiled code in threads for which the event is enabled. Check here for
603 // interp_only_mode if these events CAN be enabled.
604 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
605 MacroAssembler::load_and_test_int(Z_R0_scratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
606 z_bcr(bcondEqual, Z_R1_scratch); // Run compiled code if zero.
607 // Run interpreted.
608 z_lg(Z_R1_scratch, Address(method, Method::interpreter_entry_offset()));
609 z_br(Z_R1_scratch);
610 } else {
611 // Run compiled code.
612 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
613 z_br(Z_R1_scratch);
614 }
615 }
616
617 #ifdef ASSERT
618 void InterpreterMacroAssembler::verify_esp(Register Resp, Register Rtemp) {
619 // About to read or write Resp[0].
620 // Make sure it is not in the monitors or the TOP_IJAVA_FRAME_ABI.
621 address reentry = nullptr;
622
623 {
624 // Check if the frame pointer in Z_fp is correct.
625 NearLabel OK;
626 z_cg(Z_fp, 0, Z_SP);
627 z_bre(OK);
628 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp");
629 bind(OK);
630 }
631 {
632 // Resp must not point into or below the operand stack,
633 // i.e. IJAVA_STATE.monitors > Resp.
634 NearLabel OK;
635 Register Rmonitors = Rtemp;
636 get_monitors(Rmonitors);
637 compareU64_and_branch(Rmonitors, Resp, bcondHigh, OK);
638 reentry = stop_chain_static(reentry, "too many pops: Z_esp points into monitor area");
639 bind(OK);
640 }
641 {
642 // Resp may point to the last word of TOP_IJAVA_FRAME_ABI, but not below
643 // i.e. !(Z_SP + frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize > Resp).
644 NearLabel OK;
645 Register Rabi_bottom = Rtemp;
646 add2reg(Rabi_bottom, frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize, Z_SP);
647 compareU64_and_branch(Rabi_bottom, Resp, bcondNotHigh, OK);
648 reentry = stop_chain_static(reentry, "too many pushes: Z_esp points into TOP_IJAVA_FRAME_ABI");
649 bind(OK);
650 }
651 }
652
653 void InterpreterMacroAssembler::asm_assert_ijava_state_magic(Register tmp) {
654 Label magic_ok;
655 load_const_optimized(tmp, frame::z_istate_magic_number);
656 z_cg(tmp, Address(Z_fp, _z_ijava_state_neg(magic)));
657 z_bre(magic_ok);
658 stop_static("error: wrong magic number in ijava_state access");
659 bind(magic_ok);
660 }
661 #endif // ASSERT
662
663 void InterpreterMacroAssembler::save_bcp() {
664 z_stg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
665 asm_assert_ijava_state_magic(Z_bcp);
666 NOT_PRODUCT(z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))));
667 }
668
669 void InterpreterMacroAssembler::restore_bcp() {
670 asm_assert_ijava_state_magic(Z_bcp);
671 z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
672 }
673
674 void InterpreterMacroAssembler::save_esp(Register fp) {
675 if (fp == noreg) {
676 fp = Z_fp;
677 }
678 z_sgrk(Z_R0, Z_esp, fp);
679 z_srag(Z_R0, Z_R0, Interpreter::logStackElementSize);
680 z_stg(Z_R0, Address(fp, _z_ijava_state_neg(esp)));
681 }
682
683 void InterpreterMacroAssembler::restore_esp() {
684 asm_assert_ijava_state_magic(Z_esp);
685 z_lg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp)));
686 z_slag(Z_esp, Z_esp, Interpreter::logStackElementSize);
687 z_agr(Z_esp, Z_fp);
688 }
689
690 void InterpreterMacroAssembler::get_monitors(Register reg) {
691 asm_assert_ijava_state_magic(reg);
692 #ifdef ASSERT
693 NearLabel ok;
694 z_cg(Z_fp, 0, Z_SP);
695 z_bre(ok);
696 stop("Z_fp is corrupted");
697 bind(ok);
698 #endif // ASSERT
699 mem2reg_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
700 z_slag(reg, reg, Interpreter::logStackElementSize);
701 z_agr(reg, Z_fp);
702 }
703
704 void InterpreterMacroAssembler::save_monitors(Register reg) {
705 #ifdef ASSERT
706 NearLabel ok;
707 z_cg(Z_fp, 0, Z_SP);
708 z_bre(ok);
709 stop("Z_fp is corrupted");
710 bind(ok);
711 #endif // ASSERT
712 z_sgr(reg, Z_fp);
713 z_srag(reg, reg, Interpreter::logStackElementSize);
714 reg2mem_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
715 }
716
717 void InterpreterMacroAssembler::get_mdp(Register mdp) {
718 z_lg(mdp, _z_ijava_state_neg(mdx), Z_fp);
719 }
720
721 void InterpreterMacroAssembler::save_mdp(Register mdp) {
722 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
723 }
724
725 // Values that are only read (besides initialization).
726 void InterpreterMacroAssembler::restore_locals() {
727 asm_assert_ijava_state_magic(Z_locals);
728 z_lg(Z_locals, Address(Z_fp, _z_ijava_state_neg(locals)));
729 z_sllg(Z_locals, Z_locals, Interpreter::logStackElementSize);
730 z_agr(Z_locals, Z_fp);
731 }
732
733 void InterpreterMacroAssembler::get_method(Register reg) {
734 asm_assert_ijava_state_magic(reg);
735 z_lg(reg, Address(Z_fp, _z_ijava_state_neg(method)));
736 }
737
738 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(Register Rdst, int bcp_offset,
739 signedOrNot is_signed) {
740 // Rdst is an 8-byte return value!!!
741
742 // Unaligned loads incur only a small penalty on z/Architecture. The penalty
743 // is a few (2..3) ticks, even when the load crosses a cache line
744 // boundary. In case of a cache miss, the stall could, of course, be
745 // much longer.
746
747 switch (is_signed) {
748 case Signed:
749 z_lgh(Rdst, bcp_offset, Z_R0, Z_bcp);
750 break;
751 case Unsigned:
752 z_llgh(Rdst, bcp_offset, Z_R0, Z_bcp);
753 break;
754 default:
755 ShouldNotReachHere();
756 }
757 }
758
759
760 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(Register Rdst, int bcp_offset,
761 setCCOrNot set_cc) {
762 // Rdst is an 8-byte return value!!!
763
764 // Unaligned loads incur only a small penalty on z/Architecture. The penalty
765 // is a few (2..3) ticks, even when the load crosses a cache line
766 // boundary. In case of a cache miss, the stall could, of course, be
767 // much longer.
768
769 // Both variants implement a sign-extending int2long load.
770 if (set_cc == set_CC) {
771 load_and_test_int2long(Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
772 } else {
773 mem2reg_signed_opt( Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
774 }
775 }
776
777 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
778 get_method(Rdst);
779 mem2reg_opt(Rdst, Address(Rdst, Method::const_offset()));
780 mem2reg_opt(Rdst, Address(Rdst, ConstMethod::constants_offset()));
781 }
782
783 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
784 get_constant_pool(Rdst);
785 mem2reg_opt(Rdst, Address(Rdst, ConstantPool::cache_offset()));
786 }
787
788 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
789 get_constant_pool(Rcpool);
790 mem2reg_opt(Rtags, Address(Rcpool, ConstantPool::tags_offset()));
791 }
792
793 // Unlock if synchronized method.
794 //
795 // Unlock the receiver if this is a synchronized method.
796 // Unlock any Java monitors from synchronized blocks.
797 //
798 // If there are locked Java monitors
799 // If throw_monitor_exception
800 // throws IllegalMonitorStateException
801 // Else if install_monitor_exception
802 // installs IllegalMonitorStateException
803 // Else
804 // no error processing
805 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
806 bool throw_monitor_exception,
807 bool install_monitor_exception) {
808 NearLabel unlocked, unlock, no_unlock;
809
810 {
811 Register R_method = Z_ARG2;
812 Register R_do_not_unlock_if_synchronized = Z_ARG3;
813
814 // Get the value of _do_not_unlock_if_synchronized into G1_scratch.
815 const Address do_not_unlock_if_synchronized(Z_thread,
816 JavaThread::do_not_unlock_if_synchronized_offset());
817 load_sized_value(R_do_not_unlock_if_synchronized, do_not_unlock_if_synchronized, 1, false /*unsigned*/);
818 z_mvi(do_not_unlock_if_synchronized, false); // Reset the flag.
819
820 // Check if synchronized method.
821 get_method(R_method);
822 verify_oop(Z_tos, state);
823 push(state); // Save tos/result.
824 testbit_ushort(method2_(R_method, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
825 z_bfalse(unlocked);
826
827 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
828 // is set.
829 compareU64_and_branch(R_do_not_unlock_if_synchronized, (intptr_t)0L, bcondNotEqual, no_unlock);
830 }
831
832 // unlock monitor
833
834 // BasicObjectLock will be first in list, since this is a
835 // synchronized method. However, need to check that the object has
836 // not been unlocked by an explicit monitorexit bytecode.
837 const Address monitor(Z_fp, -(frame::z_ijava_state_size + (int) sizeof(BasicObjectLock)));
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 load_address(Z_ARG2, monitor); // Address of first monitor.
841 z_lg(Z_ARG3, Address(Z_ARG2, BasicObjectLock::obj_offset()));
842 compareU64_and_branch(Z_ARG3, (intptr_t)0L, bcondNotEqual, unlock);
843
844 if (throw_monitor_exception) {
845 // Entry already unlocked need to throw an exception.
846 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
847 should_not_reach_here();
848 } else {
849 // Monitor already unlocked during a stack unroll.
850 // If requested, install an illegal_monitor_state_exception.
851 // Continue with stack unrolling.
852 if (install_monitor_exception) {
853 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
854 }
855 z_bru(unlocked);
856 }
857
858 bind(unlock);
859
860 unlock_object(Z_ARG2);
861
862 bind(unlocked);
863
864 // I0, I1: Might contain return value
865
866 // Check that all monitors are unlocked.
867 {
868 NearLabel loop, exception, entry, restart;
869 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
870 // We use Z_ARG2 so that if we go slow path it will be the correct
871 // register for unlock_object to pass to VM directly.
872 Register R_current_monitor = Z_ARG2;
873 Register R_monitor_block_bot = Z_ARG1;
874 const Address monitor_block_bot(Z_fp, -frame::z_ijava_state_size);
875
876 bind(restart);
877 // Starting with top-most entry.
878 get_monitors(R_current_monitor);
879 // Points to word before bottom of monitor block.
880 load_address(R_monitor_block_bot, monitor_block_bot);
881 z_bru(entry);
882
883 // Entry already locked, need to throw exception.
884 bind(exception);
885
886 if (throw_monitor_exception) {
887 // Throw exception.
888 MacroAssembler::call_VM(noreg,
889 CAST_FROM_FN_PTR(address, InterpreterRuntime::
890 throw_illegal_monitor_state_exception));
891 should_not_reach_here();
892 } else {
893 // Stack unrolling. Unlock object and install illegal_monitor_exception.
894 // Unlock does not block, so don't have to worry about the frame.
895 // We don't have to preserve c_rarg1 since we are going to throw an exception.
896 unlock_object(R_current_monitor);
897 if (install_monitor_exception) {
898 call_VM(noreg, CAST_FROM_FN_PTR(address,
899 InterpreterRuntime::
900 new_illegal_monitor_state_exception));
901 }
902 z_bru(restart);
903 }
904
905 bind(loop);
906 // Check if current entry is used.
907 load_and_test_long(Z_R0_scratch, Address(R_current_monitor, BasicObjectLock::obj_offset()));
908 z_brne(exception);
909
910 add2reg(R_current_monitor, entry_size); // Otherwise advance to next entry.
911 bind(entry);
912 compareU64_and_branch(R_current_monitor, R_monitor_block_bot, bcondNotEqual, loop);
913 }
914
915 bind(no_unlock);
916 pop(state);
917 verify_oop(Z_tos, state);
918 }
919
920 void InterpreterMacroAssembler::narrow(Register result, Register ret_type) {
921 get_method(ret_type);
922 z_lg(ret_type, Address(ret_type, in_bytes(Method::const_offset())));
923 z_lb(ret_type, Address(ret_type, in_bytes(ConstMethod::result_type_offset())));
924
925 Label notBool, notByte, notChar, done;
926
927 // common case first
928 compareU32_and_branch(ret_type, T_INT, bcondEqual, done);
929
930 compareU32_and_branch(ret_type, T_BOOLEAN, bcondNotEqual, notBool);
931 z_nilf(result, 0x1);
932 z_bru(done);
933
934 bind(notBool);
935 compareU32_and_branch(ret_type, T_BYTE, bcondNotEqual, notByte);
936 z_lbr(result, result);
937 z_bru(done);
938
939 bind(notByte);
940 compareU32_and_branch(ret_type, T_CHAR, bcondNotEqual, notChar);
941 z_nilf(result, 0xffff);
942 z_bru(done);
943
944 bind(notChar);
945 // compareU32_and_branch(ret_type, T_SHORT, bcondNotEqual, notShort);
946 z_lhr(result, result);
947
948 // Nothing to do for T_INT
949 bind(done);
950 }
951
952 // remove activation
953 //
954 // Unlock the receiver if this is a synchronized method.
955 // Unlock any Java monitors from synchronized blocks.
956 // Remove the activation from the stack.
957 //
958 // If there are locked Java monitors
959 // If throw_monitor_exception
960 // throws IllegalMonitorStateException
961 // Else if install_monitor_exception
962 // installs IllegalMonitorStateException
963 // Else
964 // no error processing
965 void InterpreterMacroAssembler::remove_activation(TosState state,
966 Register return_pc,
967 bool throw_monitor_exception,
968 bool install_monitor_exception,
969 bool notify_jvmti) {
970 BLOCK_COMMENT("remove_activation {");
971 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
972
973 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
974 notify_method_exit(false, state, notify_jvmti ? NotifyJVMTI : SkipNotifyJVMTI);
975
976 if (StackReservedPages > 0) {
977 BLOCK_COMMENT("reserved_stack_check:");
978 // Test if reserved zone needs to be enabled.
979 Label no_reserved_zone_enabling;
980
981 // check if already enabled - if so no re-enabling needed
982 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
983 z_ly(Z_R0, Address(Z_thread, JavaThread::stack_guard_state_offset()));
984 compare32_and_branch(Z_R0, StackOverflow::stack_guard_enabled, bcondEqual, no_reserved_zone_enabling);
985
986 // Compare frame pointers. There is no good stack pointer, as with stack
987 // frame compression we can get different SPs when we do calls. A subsequent
988 // call could have a smaller SP, so that this compare succeeds for an
989 // inner call of the method annotated with ReservedStack.
990 z_lg(Z_R0, Address(Z_SP, (intptr_t)_z_abi(callers_sp)));
991 z_clg(Z_R0, Address(Z_thread, JavaThread::reserved_stack_activation_offset())); // Compare with frame pointer in memory.
992 z_brl(no_reserved_zone_enabling);
993
994 // Enable reserved zone again, throw stack overflow exception.
995 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), Z_thread);
996 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
997
998 should_not_reach_here();
999
1000 bind(no_reserved_zone_enabling);
1001 }
1002
1003 verify_oop(Z_tos, state);
1004
1005 pop_interpreter_frame(return_pc, Z_ARG2, Z_ARG3);
1006 BLOCK_COMMENT("} remove_activation");
1007 }
1008
1009 // lock object
1010 //
1011 // Registers alive
1012 // monitor (Z_R10) - Address of the BasicObjectLock to be used for locking,
1013 // which must be initialized with the object to lock.
1014 // object (Z_R11, Z_R2) - Address of the object to be locked.
1015 // templateTable (monitorenter) is using Z_R2 for object
1016 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
1017 const Register header = Z_ARG5;
1018 const Register tmp = Z_R1_scratch;
1019
1020 NearLabel done, slow_case;
1021
1022 lightweight_lock(monitor, object, header, tmp, slow_case);
1023 z_bru(done);
1024
1025 bind(slow_case);
1026 call_VM(noreg,
1027 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1028 monitor);
1029 bind(done);
1030 }
1031
1032 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1033 //
1034 // Registers alive
1035 // monitor - address of the BasicObjectLock to be used for locking,
1036 // which must be initialized with the object to lock.
1037 //
1038 // Throw IllegalMonitorException if object is not locked by current thread.
1039 void InterpreterMacroAssembler::unlock_object(Register monitor, Register object) {
1040 const Register header = Z_ARG4;
1041 const Register current_header = Z_R1_scratch;
1042 Address obj_entry(monitor, BasicObjectLock::obj_offset());
1043 Label done, slow_case;
1044
1045 if (object == noreg) {
1046 // In the template interpreter, we must assure that the object
1047 // entry in the monitor is cleared on all paths. Thus we move
1048 // loading up to here, and clear the entry afterwards.
1049 object = Z_ARG3; // Use Z_ARG3 if caller didn't pass object.
1050 z_lg(object, obj_entry);
1051 }
1052
1053 assert_different_registers(monitor, object, header, current_header);
1054
1055 clear_mem(obj_entry, sizeof(oop));
1056
1057 lightweight_unlock(object, header, current_header, slow_case);
1058 z_bru(done);
1059
1060 // The lock has been converted into a heavy lock and hence
1061 // we need to get into the slow case.
1062 bind(slow_case);
1063 z_stg(object, obj_entry); // Restore object entry, has been cleared above.
1064 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1065 bind(done);
1066 }
1067
1068 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) {
1069 assert(ProfileInterpreter, "must be profiling interpreter");
1070 load_and_test_long(mdp, Address(Z_fp, _z_ijava_state_neg(mdx)));
1071 z_brz(zero_continue);
1072 }
1073
1074 // Set the method data pointer for the current bcp.
1075 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1076 assert(ProfileInterpreter, "must be profiling interpreter");
1077 Label set_mdp;
1078 Register mdp = Z_ARG4;
1079 Register method = Z_ARG5;
1080
1081 get_method(method);
1082 // Test MDO to avoid the call if it is null.
1083 load_and_test_long(mdp, method2_(method, method_data));
1084 z_brz(set_mdp);
1085
1086 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), method, Z_bcp);
1087 // Z_RET: mdi
1088 // Mdo is guaranteed to be non-zero here, we checked for it before the call.
1089 assert(method->is_nonvolatile(), "choose nonvolatile reg or reload from frame");
1090 z_lg(mdp, method2_(method, method_data)); // Must reload, mdp is volatile reg.
1091 add2reg_with_index(mdp, in_bytes(MethodData::data_offset()), Z_RET, mdp);
1092
1093 bind(set_mdp);
1094 save_mdp(mdp);
1095 }
1096
1097 void InterpreterMacroAssembler::verify_method_data_pointer() {
1098 assert(ProfileInterpreter, "must be profiling interpreter");
1099 #ifdef ASSERT
1100 NearLabel verify_continue;
1101 Register bcp_expected = Z_ARG3;
1102 Register mdp = Z_ARG4;
1103 Register method = Z_ARG5;
1104
1105 test_method_data_pointer(mdp, verify_continue); // If mdp is zero, continue
1106 get_method(method);
1107
1108 // If the mdp is valid, it will point to a DataLayout header which is
1109 // consistent with the bcp. The converse is highly probable also.
1110 load_sized_value(bcp_expected, Address(mdp, DataLayout::bci_offset()), 2, false /*signed*/);
1111 z_ag(bcp_expected, Address(method, Method::const_offset()));
1112 load_address(bcp_expected, Address(bcp_expected, ConstMethod::codes_offset()));
1113 compareU64_and_branch(bcp_expected, Z_bcp, bcondEqual, verify_continue);
1114 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), method, Z_bcp, mdp);
1115 bind(verify_continue);
1116 #endif // ASSERT
1117 }
1118
1119 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) {
1120 assert(ProfileInterpreter, "must be profiling interpreter");
1121 z_stg(value, constant, mdp_in);
1122 }
1123
1124 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1125 int constant,
1126 Register tmp,
1127 bool decrement) {
1128 assert_different_registers(mdp_in, tmp);
1129 // counter address
1130 Address data(mdp_in, constant);
1131 const int delta = decrement ? -DataLayout::counter_increment : DataLayout::counter_increment;
1132 add2mem_64(Address(mdp_in, constant), delta, tmp);
1133 }
1134
1135 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1136 int flag_byte_constant) {
1137 assert(ProfileInterpreter, "must be profiling interpreter");
1138 // Set the flag.
1139 z_oi(Address(mdp_in, DataLayout::flags_offset()), flag_byte_constant);
1140 }
1141
1142 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1143 int offset,
1144 Register value,
1145 Register test_value_out,
1146 Label& not_equal_continue) {
1147 assert(ProfileInterpreter, "must be profiling interpreter");
1148 if (test_value_out == noreg) {
1149 z_cg(value, Address(mdp_in, offset));
1150 z_brne(not_equal_continue);
1151 } else {
1152 // Put the test value into a register, so caller can use it:
1153 z_lg(test_value_out, Address(mdp_in, offset));
1154 compareU64_and_branch(test_value_out, value, bcondNotEqual, not_equal_continue);
1155 }
1156 }
1157
1158 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) {
1159 update_mdp_by_offset(mdp_in, noreg, offset_of_disp);
1160 }
1161
1162 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1163 Register dataidx,
1164 int offset_of_disp) {
1165 assert(ProfileInterpreter, "must be profiling interpreter");
1166 Address disp_address(mdp_in, dataidx, offset_of_disp);
1167 Assembler::z_ag(mdp_in, disp_address);
1168 save_mdp(mdp_in);
1169 }
1170
1171 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) {
1172 assert(ProfileInterpreter, "must be profiling interpreter");
1173 add2reg(mdp_in, constant);
1174 save_mdp(mdp_in);
1175 }
1176
1177 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1178 assert(ProfileInterpreter, "must be profiling interpreter");
1179 assert(return_bci->is_nonvolatile(), "choose nonvolatile reg or save/restore");
1180 call_VM(noreg,
1181 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1182 return_bci);
1183 }
1184
1185 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) {
1186 if (ProfileInterpreter) {
1187 Label profile_continue;
1188
1189 // If no method data exists, go to profile_continue.
1190 // Otherwise, assign to mdp.
1191 test_method_data_pointer(mdp, profile_continue);
1192
1193 // We are taking a branch. Increment the taken count.
1194 // We inline increment_mdp_data_at to return bumped_count in a register
1195 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1196 Address data(mdp, JumpData::taken_offset());
1197 z_lg(bumped_count, data);
1198 // 64-bit overflow is very unlikely. Saturation to 32-bit values is
1199 // performed when reading the counts.
1200 add2reg(bumped_count, DataLayout::counter_increment);
1201 z_stg(bumped_count, data); // Store back out
1202
1203 // The method data pointer needs to be updated to reflect the new target.
1204 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1205 bind(profile_continue);
1206 }
1207 }
1208
1209 // Kills Z_R1_scratch.
1210 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1211 if (ProfileInterpreter) {
1212 Label profile_continue;
1213
1214 // If no method data exists, go to profile_continue.
1215 test_method_data_pointer(mdp, profile_continue);
1216
1217 // We are taking a branch. Increment the not taken count.
1218 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()), Z_R1_scratch);
1219
1220 // The method data pointer needs to be updated to correspond to
1221 // the next bytecode.
1222 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1223 bind(profile_continue);
1224 }
1225 }
1226
1227 // Kills: Z_R1_scratch.
1228 void InterpreterMacroAssembler::profile_call(Register mdp) {
1229 if (ProfileInterpreter) {
1230 Label profile_continue;
1231
1232 // If no method data exists, go to profile_continue.
1233 test_method_data_pointer(mdp, profile_continue);
1234
1235 // We are making a call. Increment the count.
1236 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1237
1238 // The method data pointer needs to be updated to reflect the new target.
1239 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1240 bind(profile_continue);
1241 }
1242 }
1243
1244 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1245 if (ProfileInterpreter) {
1246 Label profile_continue;
1247
1248 // If no method data exists, go to profile_continue.
1249 test_method_data_pointer(mdp, profile_continue);
1250
1251 // We are making a call. Increment the count.
1252 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1253
1254 // The method data pointer needs to be updated to reflect the new target.
1255 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1256 bind(profile_continue);
1257 }
1258 }
1259
1260 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1261 Register mdp,
1262 Register reg2,
1263 bool receiver_can_be_null) {
1264 if (ProfileInterpreter) {
1265 NearLabel profile_continue;
1266
1267 // If no method data exists, go to profile_continue.
1268 test_method_data_pointer(mdp, profile_continue);
1269
1270 NearLabel skip_receiver_profile;
1271 if (receiver_can_be_null) {
1272 NearLabel not_null;
1273 compareU64_and_branch(receiver, (intptr_t)0L, bcondNotEqual, not_null);
1274 // We are making a call. Increment the count for null receiver.
1275 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1276 z_bru(skip_receiver_profile);
1277 bind(not_null);
1278 }
1279
1280 // Record the receiver type.
1281 record_klass_in_profile(receiver, mdp, reg2);
1282 bind(skip_receiver_profile);
1283
1284 // The method data pointer needs to be updated to reflect the new target.
1285 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1286 bind(profile_continue);
1287 }
1288 }
1289
1290 // This routine creates a state machine for updating the multi-row
1291 // type profile at a virtual call site (or other type-sensitive bytecode).
1292 // The machine visits each row (of receiver/count) until the receiver type
1293 // is found, or until it runs out of rows. At the same time, it remembers
1294 // the location of the first empty row. (An empty row records null for its
1295 // receiver, and can be allocated for a newly-observed receiver type.)
1296 // Because there are two degrees of freedom in the state, a simple linear
1297 // search will not work; it must be a decision tree. Hence this helper
1298 // function is recursive, to generate the required tree structured code.
1299 // It's the interpreter, so we are trading off code space for speed.
1300 // See below for example code.
1301 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1302 Register receiver, Register mdp,
1303 Register reg2, int start_row,
1304 Label& done) {
1305 if (TypeProfileWidth == 0) {
1306 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1307 return;
1308 }
1309
1310 int last_row = VirtualCallData::row_limit() - 1;
1311 assert(start_row <= last_row, "must be work left to do");
1312 // Test this row for both the receiver and for null.
1313 // Take any of three different outcomes:
1314 // 1. found receiver => increment count and goto done
1315 // 2. found null => keep looking for case 1, maybe allocate this cell
1316 // 3. found something else => keep looking for cases 1 and 2
1317 // Case 3 is handled by a recursive call.
1318 for (int row = start_row; row <= last_row; row++) {
1319 NearLabel next_test;
1320 bool test_for_null_also = (row == start_row);
1321
1322 // See if the receiver is receiver[n].
1323 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1324 test_mdp_data_at(mdp, recvr_offset, receiver,
1325 (test_for_null_also ? reg2 : noreg),
1326 next_test);
1327 // (Reg2 now contains the receiver from the CallData.)
1328
1329 // The receiver is receiver[n]. Increment count[n].
1330 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1331 increment_mdp_data_at(mdp, count_offset);
1332 z_bru(done);
1333 bind(next_test);
1334
1335 if (test_for_null_also) {
1336 Label found_null;
1337 // Failed the equality check on receiver[n]... Test for null.
1338 z_ltgr(reg2, reg2);
1339 if (start_row == last_row) {
1340 // The only thing left to do is handle the null case.
1341 z_brz(found_null);
1342 // Receiver did not match any saved receiver and there is no empty row for it.
1343 // Increment total counter to indicate polymorphic case.
1344 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1345 z_bru(done);
1346 bind(found_null);
1347 break;
1348 }
1349 // Since null is rare, make it be the branch-taken case.
1350 z_brz(found_null);
1351
1352 // Put all the "Case 3" tests here.
1353 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done);
1354
1355 // Found a null. Keep searching for a matching receiver,
1356 // but remember that this is an empty (unused) slot.
1357 bind(found_null);
1358 }
1359 }
1360
1361 // In the fall-through case, we found no matching receiver, but we
1362 // observed the receiver[start_row] is null.
1363
1364 // Fill in the receiver field and increment the count.
1365 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1366 set_mdp_data_at(mdp, recvr_offset, receiver);
1367 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1368 load_const_optimized(reg2, DataLayout::counter_increment);
1369 set_mdp_data_at(mdp, count_offset, reg2);
1370 if (start_row > 0) {
1371 z_bru(done);
1372 }
1373 }
1374
1375 // Example state machine code for three profile rows:
1376 // // main copy of decision tree, rooted at row[1]
1377 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1378 // if (row[0].rec != nullptr) {
1379 // // inner copy of decision tree, rooted at row[1]
1380 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1381 // if (row[1].rec != nullptr) {
1382 // // degenerate decision tree, rooted at row[2]
1383 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1384 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1385 // row[2].init(rec); goto done;
1386 // } else {
1387 // // remember row[1] is empty
1388 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1389 // row[1].init(rec); goto done;
1390 // }
1391 // } else {
1392 // // remember row[0] is empty
1393 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1394 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1395 // row[0].init(rec); goto done;
1396 // }
1397 // done:
1398
1399 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1400 Register mdp, Register reg2) {
1401 assert(ProfileInterpreter, "must be profiling");
1402 Label done;
1403
1404 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1405
1406 bind (done);
1407 }
1408
1409 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
1410 if (ProfileInterpreter) {
1411 NearLabel profile_continue;
1412 uint row;
1413
1414 // If no method data exists, go to profile_continue.
1415 test_method_data_pointer(mdp, profile_continue);
1416
1417 // Update the total ret count.
1418 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1419
1420 for (row = 0; row < RetData::row_limit(); row++) {
1421 NearLabel next_test;
1422
1423 // See if return_bci is equal to bci[n]:
1424 test_mdp_data_at(mdp,
1425 in_bytes(RetData::bci_offset(row)),
1426 return_bci, noreg,
1427 next_test);
1428
1429 // Return_bci is equal to bci[n]. Increment the count.
1430 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1431
1432 // The method data pointer needs to be updated to reflect the new target.
1433 update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row)));
1434 z_bru(profile_continue);
1435 bind(next_test);
1436 }
1437
1438 update_mdp_for_ret(return_bci);
1439
1440 bind(profile_continue);
1441 }
1442 }
1443
1444 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1445 if (ProfileInterpreter) {
1446 Label profile_continue;
1447
1448 // If no method data exists, go to profile_continue.
1449 test_method_data_pointer(mdp, profile_continue);
1450
1451 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1452
1453 // The method data pointer needs to be updated.
1454 int mdp_delta = in_bytes(BitData::bit_data_size());
1455 if (TypeProfileCasts) {
1456 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1457 }
1458 update_mdp_by_constant(mdp, mdp_delta);
1459
1460 bind(profile_continue);
1461 }
1462 }
1463
1464 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1465 if (ProfileInterpreter) {
1466 Label profile_continue;
1467
1468 // If no method data exists, go to profile_continue.
1469 test_method_data_pointer(mdp, profile_continue);
1470
1471 // The method data pointer needs to be updated.
1472 int mdp_delta = in_bytes(BitData::bit_data_size());
1473 if (TypeProfileCasts) {
1474 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1475
1476 // Record the object type.
1477 record_klass_in_profile(klass, mdp, reg2);
1478 }
1479 update_mdp_by_constant(mdp, mdp_delta);
1480
1481 bind(profile_continue);
1482 }
1483 }
1484
1485 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1486 if (ProfileInterpreter) {
1487 Label profile_continue;
1488
1489 // If no method data exists, go to profile_continue.
1490 test_method_data_pointer(mdp, profile_continue);
1491
1492 // Update the default case count.
1493 increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset()));
1494
1495 // The method data pointer needs to be updated.
1496 update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset()));
1497
1498 bind(profile_continue);
1499 }
1500 }
1501
1502 // Kills: index, scratch1, scratch2.
1503 void InterpreterMacroAssembler::profile_switch_case(Register index,
1504 Register mdp,
1505 Register scratch1,
1506 Register scratch2) {
1507 if (ProfileInterpreter) {
1508 Label profile_continue;
1509 assert_different_registers(index, mdp, scratch1, scratch2);
1510
1511 // If no method data exists, go to profile_continue.
1512 test_method_data_pointer(mdp, profile_continue);
1513
1514 // Build the base (index * per_case_size_in_bytes()) +
1515 // case_array_offset_in_bytes().
1516 z_sllg(index, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1517 add2reg(index, in_bytes(MultiBranchData::case_array_offset()));
1518
1519 // Add the calculated base to the mdp -> address of the case' data.
1520 Address case_data_addr(mdp, index);
1521 Register case_data = scratch1;
1522 load_address(case_data, case_data_addr);
1523
1524 // Update the case count.
1525 increment_mdp_data_at(case_data,
1526 in_bytes(MultiBranchData::relative_count_offset()),
1527 scratch2);
1528
1529 // The method data pointer needs to be updated.
1530 update_mdp_by_offset(mdp,
1531 index,
1532 in_bytes(MultiBranchData::relative_displacement_offset()));
1533
1534 bind(profile_continue);
1535 }
1536 }
1537
1538 // kills: R0, R1, flags, loads klass from obj (if not null)
1539 void InterpreterMacroAssembler::profile_obj_type(Register obj, Address mdo_addr, Register klass, bool cmp_done) {
1540 NearLabel null_seen, init_klass, do_nothing, do_update;
1541
1542 // Klass = obj is allowed.
1543 const Register tmp = Z_R1;
1544 assert_different_registers(obj, mdo_addr.base(), tmp, Z_R0);
1545 assert_different_registers(klass, mdo_addr.base(), tmp, Z_R0);
1546
1547 z_lg(tmp, mdo_addr);
1548 if (cmp_done) {
1549 z_brz(null_seen);
1550 } else {
1551 compareU64_and_branch(obj, (intptr_t)0, Assembler::bcondEqual, null_seen);
1552 }
1553
1554 MacroAssembler::verify_oop(obj, FILE_AND_LINE);
1555 load_klass(klass, obj);
1556
1557 // Klass seen before, nothing to do (regardless of unknown bit).
1558 z_lgr(Z_R0, tmp);
1559 assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
1560 z_nill(Z_R0, TypeEntries::type_klass_mask & 0xFFFF);
1561 compareU64_and_branch(Z_R0, klass, Assembler::bcondEqual, do_nothing);
1562
1563 // Already unknown. Nothing to do anymore.
1564 z_tmll(tmp, TypeEntries::type_unknown);
1565 z_brc(Assembler::bcondAllOne, do_nothing);
1566
1567 z_lgr(Z_R0, tmp);
1568 assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
1569 z_nill(Z_R0, TypeEntries::type_mask & 0xFFFF);
1570 compareU64_and_branch(Z_R0, (intptr_t)0, Assembler::bcondEqual, init_klass);
1571
1572 // Different than before. Cannot keep accurate profile.
1573 z_oill(tmp, TypeEntries::type_unknown);
1574 z_bru(do_update);
1575
1576 bind(init_klass);
1577 // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1578 z_ogr(tmp, klass);
1579 z_bru(do_update);
1580
1581 bind(null_seen);
1582 // Set null_seen if obj is 0.
1583 z_oill(tmp, TypeEntries::null_seen);
1584 // fallthru: z_bru(do_update);
1585
1586 bind(do_update);
1587 z_stg(tmp, mdo_addr);
1588
1589 bind(do_nothing);
1590 }
1591
1592 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1593 if (!ProfileInterpreter) {
1594 return;
1595 }
1596
1597 assert_different_registers(mdp, callee, tmp);
1598
1599 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1600 Label profile_continue;
1601
1602 test_method_data_pointer(mdp, profile_continue);
1603
1604 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1605
1606 z_cliy(in_bytes(DataLayout::tag_offset()) - off_to_start, mdp,
1607 is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1608 z_brne(profile_continue);
1609
1610 if (MethodData::profile_arguments()) {
1611 NearLabel done;
1612 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1613 add2reg(mdp, off_to_args);
1614
1615 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1616 if (i > 0 || MethodData::profile_return()) {
1617 // If return value type is profiled we may have no argument to profile.
1618 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1619 add2reg(tmp, -i*TypeStackSlotEntries::per_arg_count());
1620 compare64_and_branch(tmp, TypeStackSlotEntries::per_arg_count(), Assembler::bcondLow, done);
1621 }
1622 z_lg(tmp, Address(callee, Method::const_offset()));
1623 z_lgh(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1624 // Stack offset o (zero based) from the start of the argument
1625 // list. For n arguments translates into offset n - o - 1 from
1626 // the end of the argument list. But there is an extra slot at
1627 // the top of the stack. So the offset is n - o from Lesp.
1628 z_sg(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
1629 z_sllg(tmp, tmp, Interpreter::logStackElementSize);
1630 Address stack_slot_addr(tmp, Z_esp);
1631 z_ltg(tmp, stack_slot_addr);
1632
1633 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
1634 profile_obj_type(tmp, mdo_arg_addr, tmp, /*ltg did compare to 0*/ true);
1635
1636 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1637 add2reg(mdp, to_add);
1638 off_to_args += to_add;
1639 }
1640
1641 if (MethodData::profile_return()) {
1642 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1643 add2reg(tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1644 }
1645
1646 bind(done);
1647
1648 if (MethodData::profile_return()) {
1649 // We're right after the type profile for the last
1650 // argument. Tmp is the number of cells left in the
1651 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1652 // if there's a return to profile.
1653 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1654 z_sllg(tmp, tmp, exact_log2(DataLayout::cell_size));
1655 z_agr(mdp, tmp);
1656 }
1657 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
1658 } else {
1659 assert(MethodData::profile_return(), "either profile call args or call ret");
1660 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1661 }
1662
1663 // Mdp points right after the end of the
1664 // CallTypeData/VirtualCallTypeData, right after the cells for the
1665 // return value type if there's one.
1666 bind(profile_continue);
1667 }
1668 }
1669
1670 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1671 assert_different_registers(mdp, ret, tmp);
1672 if (ProfileInterpreter && MethodData::profile_return()) {
1673 Label profile_continue;
1674
1675 test_method_data_pointer(mdp, profile_continue);
1676
1677 if (MethodData::profile_return_jsr292_only()) {
1678 // If we don't profile all invoke bytecodes we must make sure
1679 // it's a bytecode we indeed profile. We can't go back to the
1680 // beginning of the ProfileData we intend to update to check its
1681 // type because we're right after it and we don't known its
1682 // length.
1683 NearLabel do_profile;
1684 Address bc(Z_bcp);
1685 z_lb(tmp, bc);
1686 compare32_and_branch(tmp, Bytecodes::_invokedynamic, Assembler::bcondEqual, do_profile);
1687 compare32_and_branch(tmp, Bytecodes::_invokehandle, Assembler::bcondEqual, do_profile);
1688 get_method(tmp);
1689 // Supplement to 8139891: _intrinsic_id exceeded 1-byte size limit.
1690 if (Method::intrinsic_id_size_in_bytes() == 1) {
1691 z_cli(in_bytes(Method::intrinsic_id_offset()), tmp, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1692 } else {
1693 assert(Method::intrinsic_id_size_in_bytes() == 2, "size error: check Method::_intrinsic_id");
1694 z_lh(tmp, in_bytes(Method::intrinsic_id_offset()), Z_R0, tmp);
1695 z_chi(tmp, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1696 }
1697 z_brne(profile_continue);
1698
1699 bind(do_profile);
1700 }
1701
1702 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
1703 profile_obj_type(ret, mdo_ret_addr, tmp);
1704
1705 bind(profile_continue);
1706 }
1707 }
1708
1709 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1710 if (ProfileInterpreter && MethodData::profile_parameters()) {
1711 Label profile_continue, done;
1712
1713 test_method_data_pointer(mdp, profile_continue);
1714
1715 // Load the offset of the area within the MDO used for
1716 // parameters. If it's negative we're not profiling any parameters.
1717 Address parm_di_addr(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()));
1718 load_and_test_int2long(tmp1, parm_di_addr);
1719 z_brl(profile_continue);
1720
1721 // Compute a pointer to the area for parameters from the offset
1722 // and move the pointer to the slot for the last
1723 // parameters. Collect profiling from last parameter down.
1724 // mdo start + parameters offset + array length - 1
1725
1726 // Pointer to the parameter area in the MDO.
1727 z_agr(mdp, tmp1);
1728
1729 // Offset of the current profile entry to update.
1730 const Register entry_offset = tmp1;
1731 // entry_offset = array len in number of cells.
1732 z_lg(entry_offset, Address(mdp, ArrayData::array_len_offset()));
1733 // entry_offset (number of cells) = array len - size of 1 entry
1734 add2reg(entry_offset, -TypeStackSlotEntries::per_arg_count());
1735 // entry_offset in bytes
1736 z_sllg(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1737
1738 Label loop;
1739 bind(loop);
1740
1741 Address arg_off(mdp, entry_offset, ParametersTypeData::stack_slot_offset(0));
1742 Address arg_type(mdp, entry_offset, ParametersTypeData::type_offset(0));
1743
1744 // Load offset on the stack from the slot for this parameter.
1745 z_lg(tmp2, arg_off);
1746 z_sllg(tmp2, tmp2, Interpreter::logStackElementSize);
1747 z_lcgr(tmp2); // Negate.
1748
1749 // Profile the parameter.
1750 z_ltg(tmp2, Address(Z_locals, tmp2));
1751 profile_obj_type(tmp2, arg_type, tmp2, /*ltg did compare to 0*/ true);
1752
1753 // Go to next parameter.
1754 z_aghi(entry_offset, -TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size);
1755 z_brnl(loop);
1756
1757 bind(profile_continue);
1758 }
1759 }
1760
1761 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1762 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1763 int increment,
1764 Address mask,
1765 Register scratch,
1766 bool preloaded,
1767 branch_condition cond,
1768 Label *where) {
1769 assert_different_registers(counter_addr.base(), scratch);
1770 if (preloaded) {
1771 add2reg(scratch, increment);
1772 reg2mem_opt(scratch, counter_addr, false);
1773 } else {
1774 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment) && counter_addr.is_RSYform()) {
1775 z_alsi(counter_addr.disp20(), counter_addr.base(), increment);
1776 mem2reg_signed_opt(scratch, counter_addr);
1777 } else {
1778 mem2reg_signed_opt(scratch, counter_addr);
1779 add2reg(scratch, increment);
1780 reg2mem_opt(scratch, counter_addr, false);
1781 }
1782 }
1783 z_n(scratch, mask);
1784 if (where) { z_brc(cond, *where); }
1785 }
1786
1787 // Get MethodCounters object for given method. Lazily allocated if necessary.
1788 // method - Ptr to Method object.
1789 // Rcounters - Ptr to MethodCounters object associated with Method object.
1790 // skip - Exit point if MethodCounters object can't be created (OOM condition).
1791 void InterpreterMacroAssembler::get_method_counters(Register Rmethod,
1792 Register Rcounters,
1793 Label& skip) {
1794 assert_different_registers(Rmethod, Rcounters);
1795
1796 BLOCK_COMMENT("get MethodCounters object {");
1797
1798 Label has_counters;
1799 load_and_test_long(Rcounters, Address(Rmethod, Method::method_counters_offset()));
1800 z_brnz(has_counters);
1801
1802 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), Rmethod);
1803 z_ltgr(Rcounters, Z_RET); // Runtime call returns MethodCounters object.
1804 z_brz(skip); // No MethodCounters, out of memory.
1805
1806 bind(has_counters);
1807
1808 BLOCK_COMMENT("} get MethodCounters object");
1809 }
1810
1811 // Increment invocation counter in MethodCounters object.
1812 // Return (invocation_counter+backedge_counter) as "result" in RctrSum.
1813 // Counter values are all unsigned.
1814 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register RctrSum) {
1815 assert(UseCompiler, "incrementing must be useful");
1816 assert_different_registers(Rcounters, RctrSum);
1817
1818 int increment = InvocationCounter::count_increment;
1819 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
1820 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset());
1821
1822 BLOCK_COMMENT("Increment invocation counter {");
1823
1824 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
1825 // Increment the invocation counter in place,
1826 // then add the incremented value to the backedge counter.
1827 z_l(RctrSum, be_counter_offset, Rcounters);
1828 z_alsi(inv_counter_offset, Rcounters, increment); // Atomic increment @no extra cost!
1829 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
1830 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
1831 } else {
1832 // This path is optimized for low register consumption
1833 // at the cost of somewhat higher operand delays.
1834 // It does not need an extra temp register.
1835
1836 // Update the invocation counter.
1837 z_l(RctrSum, inv_counter_offset, Rcounters);
1838 if (RctrSum == Z_R0) {
1839 z_ahi(RctrSum, increment);
1840 } else {
1841 add2reg(RctrSum, increment);
1842 }
1843 z_st(RctrSum, inv_counter_offset, Rcounters);
1844
1845 // Mask off the state bits.
1846 z_nilf(RctrSum, InvocationCounter::count_mask_value);
1847
1848 // Add the backedge counter to the updated invocation counter to
1849 // form the result.
1850 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
1851 }
1852
1853 BLOCK_COMMENT("} Increment invocation counter");
1854
1855 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
1856 }
1857
1858
1859 // increment backedge counter in MethodCounters object.
1860 // return (invocation_counter+backedge_counter) as "result" in RctrSum
1861 // counter values are all unsigned!
1862 void InterpreterMacroAssembler::increment_backedge_counter(Register Rcounters, Register RctrSum) {
1863 assert(UseCompiler, "incrementing must be useful");
1864 assert_different_registers(Rcounters, RctrSum);
1865
1866 int increment = InvocationCounter::count_increment;
1867 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
1868 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset());
1869
1870 BLOCK_COMMENT("Increment backedge counter {");
1871
1872 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
1873 // Increment the invocation counter in place,
1874 // then add the incremented value to the backedge counter.
1875 z_l(RctrSum, inv_counter_offset, Rcounters);
1876 z_alsi(be_counter_offset, Rcounters, increment); // Atomic increment @no extra cost!
1877 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
1878 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
1879 } else {
1880 // This path is optimized for low register consumption
1881 // at the cost of somewhat higher operand delays.
1882 // It does not need an extra temp register.
1883
1884 // Update the invocation counter.
1885 z_l(RctrSum, be_counter_offset, Rcounters);
1886 if (RctrSum == Z_R0) {
1887 z_ahi(RctrSum, increment);
1888 } else {
1889 add2reg(RctrSum, increment);
1890 }
1891 z_st(RctrSum, be_counter_offset, Rcounters);
1892
1893 // Mask off the state bits.
1894 z_nilf(RctrSum, InvocationCounter::count_mask_value);
1895
1896 // Add the backedge counter to the updated invocation counter to
1897 // form the result.
1898 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
1899 }
1900
1901 BLOCK_COMMENT("} Increment backedge counter");
1902
1903 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
1904 }
1905
1906 // Add an InterpMonitorElem to stack (see frame_s390.hpp).
1907 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty,
1908 Register Rtemp1,
1909 Register Rtemp2,
1910 Register Rtemp3) {
1911
1912 const Register Rcurr_slot = Rtemp1;
1913 const Register Rlimit = Rtemp2;
1914 const jint delta = -frame::interpreter_frame_monitor_size_in_bytes();
1915
1916 assert((delta & LongAlignmentMask) == 0,
1917 "sizeof BasicObjectLock must be even number of doublewords");
1918 assert(2 * wordSize == -delta, "this works only as long as delta == -2*wordSize");
1919 assert(Rcurr_slot != Z_R0, "Register must be usable as base register");
1920 assert_different_registers(Rlimit, Rcurr_slot, Rtemp3);
1921
1922 get_monitors(Rlimit);
1923
1924 // Adjust stack pointer for additional monitor entry.
1925 resize_frame(RegisterOrConstant((intptr_t) delta), Z_fp, false);
1926
1927 // Rtemp3 is free at this point, use it to store top_frame_sp
1928 z_sgrk(Rtemp3, Z_SP, Z_fp);
1929 z_srag(Rtemp3, Rtemp3, Interpreter::logStackElementSize);
1930 reg2mem_opt(Rtemp3, Address(Z_fp, _z_ijava_state_neg(top_frame_sp)));
1931
1932 if (!stack_is_empty) {
1933 // Must copy stack contents down.
1934 NearLabel next, done;
1935
1936 // Rtemp := addr(Tos), Z_esp is pointing below it!
1937 add2reg(Rcurr_slot, wordSize, Z_esp);
1938
1939 // Nothing to do, if already at monitor area.
1940 compareU64_and_branch(Rcurr_slot, Rlimit, bcondNotLow, done);
1941
1942 bind(next);
1943
1944 // Move one stack slot.
1945 mem2reg_opt(Rtemp3, Address(Rcurr_slot));
1946 reg2mem_opt(Rtemp3, Address(Rcurr_slot, delta));
1947 add2reg(Rcurr_slot, wordSize);
1948 compareU64_and_branch(Rcurr_slot, Rlimit, bcondLow, next); // Are we done?
1949
1950 bind(done);
1951 // Done copying stack.
1952 }
1953
1954 // Adjust expression stack and monitor pointers.
1955 add2reg(Z_esp, delta);
1956 add2reg(Rlimit, delta);
1957 save_monitors(Rlimit);
1958 }
1959
1960 // Note: Index holds the offset in bytes afterwards.
1961 // You can use this to store a new value (with Llocals as the base).
1962 void InterpreterMacroAssembler::access_local_int(Register index, Register dst) {
1963 z_sllg(index, index, LogBytesPerWord);
1964 mem2reg_opt(dst, Address(Z_locals, index), false);
1965 }
1966
1967 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1968 if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); }
1969 }
1970
1971 // Inline assembly for:
1972 //
1973 // if (thread is in interp_only_mode) {
1974 // InterpreterRuntime::post_method_entry();
1975 // }
1976
1977 void InterpreterMacroAssembler::notify_method_entry() {
1978
1979 // JVMTI
1980 // Whenever JVMTI puts a thread in interp_only_mode, method
1981 // entry/exit events are sent for that thread to track stack
1982 // depth. If it is possible to enter interp_only_mode we add
1983 // the code to check if the event should be sent.
1984 if (JvmtiExport::can_post_interpreter_events()) {
1985 Label jvmti_post_done;
1986 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
1987 z_bre(jvmti_post_done);
1988 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
1989 bind(jvmti_post_done);
1990 }
1991 }
1992
1993 // Inline assembly for:
1994 //
1995 // if (thread is in interp_only_mode) {
1996 // if (!native_method) save result
1997 // InterpreterRuntime::post_method_exit();
1998 // if (!native_method) restore result
1999 // }
2000 // if (DTraceMethodProbes) {
2001 // SharedRuntime::dtrace_method_exit(thread, method);
2002 // }
2003 //
2004 // For native methods their result is stored in z_ijava_state.lresult
2005 // and z_ijava_state.fresult before coming here.
2006 // Java methods have their result stored in the expression stack.
2007 //
2008 // Notice the dependency to frame::interpreter_frame_result().
2009 void InterpreterMacroAssembler::notify_method_exit(bool native_method,
2010 TosState state,
2011 NotifyMethodExitMode mode) {
2012 // JVMTI
2013 // Whenever JVMTI puts a thread in interp_only_mode, method
2014 // entry/exit events are sent for that thread to track stack
2015 // depth. If it is possible to enter interp_only_mode we add
2016 // the code to check if the event should be sent.
2017 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2018 Label jvmti_post_done;
2019 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2020 z_bre(jvmti_post_done);
2021 if (!native_method) push(state); // see frame::interpreter_frame_result()
2022 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2023 if (!native_method) pop(state);
2024 bind(jvmti_post_done);
2025 }
2026 }
2027
2028 void InterpreterMacroAssembler::skip_if_jvmti_mode(Label &Lskip, Register Rscratch) {
2029 if (!JvmtiExport::can_post_interpreter_events()) {
2030 return;
2031 }
2032
2033 load_and_test_int(Rscratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2034 z_brnz(Lskip);
2035
2036 }
2037
2038 // Pop the topmost TOP_IJAVA_FRAME and set it's sender_sp as new Z_SP.
2039 // The return pc is loaded into the register return_pc.
2040 //
2041 // Registers updated:
2042 // return_pc - The return pc of the calling frame.
2043 // tmp1, tmp2 - scratch
2044 void InterpreterMacroAssembler::pop_interpreter_frame(Register return_pc, Register tmp1, Register tmp2) {
2045 // F0 Z_SP -> caller_sp (F1's)
2046 // ...
2047 // sender_sp (F1's)
2048 // ...
2049 // F1 Z_fp -> caller_sp (F2's)
2050 // return_pc (Continuation after return from F0.)
2051 // ...
2052 // F2 caller_sp
2053
2054 // Remove F0's activation. Restoring Z_SP to sender_sp reverts modifications
2055 // (a) by a c2i adapter and (b) by generate_fixed_frame().
2056 // In case (a) the new top frame F1 is an unextended compiled frame.
2057 // In case (b) F1 is converted from PARENT_IJAVA_FRAME to TOP_IJAVA_FRAME.
2058
2059 // Case (b) seems to be redundant when returning to a interpreted caller,
2060 // because then the caller's top_frame_sp is installed as sp (see
2061 // TemplateInterpreterGenerator::generate_return_entry_for ()). But
2062 // pop_interpreter_frame() is also used in exception handling and there the
2063 // frame type of the caller is unknown, therefore top_frame_sp cannot be used,
2064 // so it is important that sender_sp is the caller's sp as TOP_IJAVA_FRAME.
2065
2066 Register R_f1_sender_sp = tmp1;
2067 Register R_f2_sp = tmp2;
2068
2069 // First check for the interpreter frame's magic.
2070 asm_assert_ijava_state_magic(R_f2_sp/*tmp*/);
2071 z_lg(R_f2_sp, _z_parent_ijava_frame_abi(callers_sp), Z_fp);
2072 z_lg(R_f1_sender_sp, _z_ijava_state_neg(sender_sp), Z_fp);
2073 if (return_pc->is_valid())
2074 z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2075 // Pop F0 by resizing to R_f1_sender_sp and using R_f2_sp as fp.
2076 resize_frame_absolute(R_f1_sender_sp, R_f2_sp, false/*load fp*/);
2077
2078 #ifdef ASSERT
2079 // The return_pc in the new top frame is dead... at least that's my
2080 // current understanding; to assert this I overwrite it.
2081 load_const_optimized(Z_ARG3, 0xb00b1);
2082 z_stg(Z_ARG3, _z_parent_ijava_frame_abi(return_pc), Z_SP);
2083 #endif
2084 }
--- EOF ---