1 /*
2 * Copyright (c) 2016, 2017, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2016, 2017 SAP SE. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 // Major contributions by AHa, AS, JL, ML.
27
28 #include "precompiled.hpp"
29 #include "asm/macroAssembler.inline.hpp"
30 #include "interp_masm_s390.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "oops/arrayOop.hpp"
34 #include "oops/markOop.hpp"
35 #include "prims/jvmtiExport.hpp"
36 #include "prims/jvmtiThreadState.hpp"
37 #include "runtime/basicLock.hpp"
38 #include "runtime/biasedLocking.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/thread.inline.hpp"
41
42 // Implementation of InterpreterMacroAssembler.
43 // This file specializes the assember with interpreter-specific macros.
44
45 #ifdef PRODUCT
46 #define BLOCK_COMMENT(str)
47 #define BIND(label) bind(label);
48 #else
49 #define BLOCK_COMMENT(str) block_comment(str)
50 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
51 #endif
52
53 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) {
54 assert(entry != NULL, "Entry must have been generated by now");
55 assert(Rscratch != Z_R0, "Can't use R0 for addressing");
56 branch_optimized(Assembler::bcondAlways, entry);
57 }
58
59 void InterpreterMacroAssembler::empty_expression_stack(void) {
60 get_monitors(Z_R1_scratch);
61 add2reg(Z_esp, -Interpreter::stackElementSize, Z_R1_scratch);
62 }
63
64 // Dispatch code executed in the prolog of a bytecode which does not do it's
65 // own dispatch.
66 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
67 // On z/Architecture we are short on registers, therefore we do not preload the
68 // dispatch address of the next bytecode.
69 }
70
71 // Dispatch code executed in the epilog of a bytecode which does not do it's
72 // own dispatch.
73 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
74 dispatch_next(state, step);
75 }
76
77 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) {
78 z_llgc(Z_bytecode, bcp_incr, Z_R0, Z_bcp); // Load next bytecode.
79 add2reg(Z_bcp, bcp_incr); // Advance bcp. Add2reg produces optimal code.
80 dispatch_base(state, Interpreter::dispatch_table(state));
81 }
82
83 // Common code to dispatch and dispatch_only.
84 // Dispatch value in Lbyte_code and increment Lbcp.
85
86 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) {
87 verify_FPU(1, state);
88
89 #ifdef ASSERT
90 address reentry = NULL;
91 { Label OK;
92 // Check if the frame pointer in Z_fp is correct.
93 z_cg(Z_fp, 0, Z_SP);
94 z_bre(OK);
95 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp: " FILE_AND_LINE);
96 bind(OK);
97 }
98 { Label OK;
99 // check if the locals pointer in Z_locals is correct
100 z_cg(Z_locals, _z_ijava_state_neg(locals), Z_fp);
101 z_bre(OK);
102 reentry = stop_chain_static(reentry, "invalid locals pointer Z_locals: " FILE_AND_LINE);
103 bind(OK);
104 }
105 #endif
106
107 // TODO: Maybe implement +VerifyActivationFrameSize here.
108 // verify_thread(); // Too slow. We will just verify on method entry & exit.
109 verify_oop(Z_tos, state);
110 #ifdef FAST_DISPATCH
111 if (table == Interpreter::dispatch_table(state)) {
112 // Use IdispatchTables.
113 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code);
114 // Add offset to correct dispatch table.
115 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // Multiply by wordSize.
116 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // Get entry addr.
117 } else
118 #endif
119 {
120 // Dispatch table to use.
121 load_absolute_address(Z_tmp_1, (address) table); // Z_tmp_1 = table;
122
123 // 0 <= Z_bytecode < 256 => Use a 32 bit shift, because it is shorter than sllg.
124 // Z_bytecode must have been loaded zero-extended for this approach to be correct.
125 z_sll(Z_bytecode, LogBytesPerWord, Z_R0); // Multiply by wordSize.
126 z_lg(Z_tmp_1, 0, Z_bytecode, Z_tmp_1); // Get entry addr.
127 }
128 z_br(Z_tmp_1);
129 }
130
131 void InterpreterMacroAssembler::dispatch_only(TosState state) {
132 dispatch_base(state, Interpreter::dispatch_table(state));
133 }
134
135 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
136 dispatch_base(state, Interpreter::normal_table(state));
137 }
138
139 void InterpreterMacroAssembler::dispatch_via(TosState state, address *table) {
140 // Load current bytecode.
141 z_llgc(Z_bytecode, Address(Z_bcp, (intptr_t)0));
142 dispatch_base(state, table);
143 }
144
145 // The following call_VM*_base() methods overload and mask the respective
146 // declarations/definitions in class MacroAssembler. They are meant as a "detour"
147 // to perform additional, template interpreter specific tasks before actually
148 // calling their MacroAssembler counterparts.
149
150 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point) {
151 bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated.
152 // interpreter specific
153 // Note: No need to save/restore bcp (Z_R13) pointer since these are callee
154 // saved registers and no blocking/ GC can happen in leaf calls.
155
156 // super call
157 MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation);
158 }
159
160 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, bool allow_relocation) {
161 // interpreter specific
162 // Note: No need to save/restore bcp (Z_R13) pointer since these are callee
163 // saved registers and no blocking/ GC can happen in leaf calls.
164
165 // super call
166 MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation);
167 }
168
169 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp,
170 address entry_point, bool check_exceptions) {
171 bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated.
172 // interpreter specific
173
174 save_bcp();
175 save_esp();
176 // super call
177 MacroAssembler::call_VM_base(oop_result, last_java_sp,
178 entry_point, allow_relocation, check_exceptions);
179 restore_bcp();
180 }
181
182 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp,
183 address entry_point, bool allow_relocation,
184 bool check_exceptions) {
185 // interpreter specific
186
187 save_bcp();
188 save_esp();
189 // super call
190 MacroAssembler::call_VM_base(oop_result, last_java_sp,
191 entry_point, allow_relocation, check_exceptions);
192 restore_bcp();
193 }
194
195 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
196 if (JvmtiExport::can_pop_frame()) {
197 BLOCK_COMMENT("check_and_handle_popframe {");
198 Label L;
199 // Initiate popframe handling only if it is not already being
200 // processed. If the flag has the popframe_processing bit set, it
201 // means that this code is called *during* popframe handling - we
202 // don't want to reenter.
203 // TODO: Check if all four state combinations could be visible.
204 // If (processing and !pending) is an invisible/impossible state,
205 // there is optimization potential by testing both bits at once.
206 // Then, All_Zeroes and All_Ones means skip, Mixed means doit.
207 testbit(Address(Z_thread, JavaThread::popframe_condition_offset()),
208 exact_log2(JavaThread::popframe_pending_bit));
209 z_bfalse(L);
210 testbit(Address(Z_thread, JavaThread::popframe_condition_offset()),
211 exact_log2(JavaThread::popframe_processing_bit));
212 z_btrue(L);
213
214 // Call Interpreter::remove_activation_preserving_args_entry() to get the
215 // address of the same-named entrypoint in the generated interpreter code.
216 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
217 // The above call should (as its only effect) return the contents of the field
218 // _remove_activation_preserving_args_entry in Z_RET.
219 // We just jump there to have the work done.
220 z_br(Z_RET);
221 // There is no way for control to fall thru here.
222
223 bind(L);
224 BLOCK_COMMENT("} check_and_handle_popframe");
225 }
226 }
227
228
229 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
230 Register RjvmtiState = Z_R1_scratch;
231 int tos_off = in_bytes(JvmtiThreadState::earlyret_tos_offset());
232 int oop_off = in_bytes(JvmtiThreadState::earlyret_oop_offset());
233 int val_off = in_bytes(JvmtiThreadState::earlyret_value_offset());
234 int state_off = in_bytes(JavaThread::jvmti_thread_state_offset());
235
236 z_lg(RjvmtiState, state_off, Z_thread);
237
238 switch (state) {
239 case atos: z_lg(Z_tos, oop_off, RjvmtiState);
240 store_const(Address(RjvmtiState, oop_off), 0L, 8, 8, Z_R0_scratch);
241 break;
242 case ltos: z_lg(Z_tos, val_off, RjvmtiState); break;
243 case btos: // fall through
244 case ztos: // fall through
245 case ctos: // fall through
246 case stos: // fall through
247 case itos: z_llgf(Z_tos, val_off, RjvmtiState); break;
248 case ftos: z_le(Z_ftos, val_off, RjvmtiState); break;
249 case dtos: z_ld(Z_ftos, val_off, RjvmtiState); break;
250 case vtos: /* nothing to do */ break;
251 default : ShouldNotReachHere();
252 }
253
254 // Clean up tos value in the jvmti thread state.
255 store_const(Address(RjvmtiState, val_off), 0L, 8, 8, Z_R0_scratch);
256 // Set tos state field to illegal value.
257 store_const(Address(RjvmtiState, tos_off), ilgl, 4, 1, Z_R0_scratch);
258 }
259
260 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
261 if (JvmtiExport::can_force_early_return()) {
262 BLOCK_COMMENT("check_and_handle_earlyret {");
263 Label L;
264 // arg regs are save, because we are just behind the call in call_VM_base
265 Register jvmti_thread_state = Z_ARG2;
266 Register tmp = Z_ARG3;
267 load_and_test_long(jvmti_thread_state, Address(Z_thread, JavaThread::jvmti_thread_state_offset()));
268 z_bre(L); // if (thread->jvmti_thread_state() == NULL) exit;
269
270 // Initiate earlyret handling only if it is not already being processed.
271 // If the flag has the earlyret_processing bit set, it means that this code
272 // is called *during* earlyret handling - we don't want to reenter.
273
274 assert((JvmtiThreadState::earlyret_pending != 0) && (JvmtiThreadState::earlyret_inactive == 0),
275 "must fix this check, when changing the values of the earlyret enum");
276 assert(JvmtiThreadState::earlyret_pending == 1, "must fix this check, when changing the values of the earlyret enum");
277
278 load_and_test_int(tmp, Address(jvmti_thread_state, JvmtiThreadState::earlyret_state_offset()));
279 z_brz(L); // if (thread->jvmti_thread_state()->_earlyret_state != JvmtiThreadState::earlyret_pending) exit;
280
281 // Call Interpreter::remove_activation_early_entry() to get the address of the
282 // same-named entrypoint in the generated interpreter code.
283 assert(sizeof(TosState) == 4, "unexpected size");
284 z_l(Z_ARG1, Address(jvmti_thread_state, JvmtiThreadState::earlyret_tos_offset()));
285 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Z_ARG1);
286 // The above call should (as its only effect) return the contents of the field
287 // _remove_activation_preserving_args_entry in Z_RET.
288 // We just jump there to have the work done.
289 z_br(Z_RET);
290 // There is no way for control to fall thru here.
291
292 bind(L);
293 BLOCK_COMMENT("} check_and_handle_earlyret");
294 }
295 }
296
297 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2) {
298 lgr_if_needed(Z_ARG1, arg_1);
299 assert(arg_2 != Z_ARG1, "smashed argument");
300 lgr_if_needed(Z_ARG2, arg_2);
301 MacroAssembler::call_VM_leaf_base(entry_point, true);
302 }
303
304 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, int bcp_offset, size_t index_size) {
305 Address param(Z_bcp, bcp_offset);
306
307 BLOCK_COMMENT("get_cache_index_at_bcp {");
308 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
309 if (index_size == sizeof(u2)) {
310 load_sized_value(index, param, 2, false /*signed*/);
311 } else if (index_size == sizeof(u4)) {
312
313 load_sized_value(index, param, 4, false);
314
315 // Check if the secondary index definition is still ~x, otherwise
316 // we have to change the following assembler code to calculate the
317 // plain index.
318 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
319 not_(index); // Convert to plain index.
320 } else if (index_size == sizeof(u1)) {
321 z_llgc(index, param);
322 } else {
323 ShouldNotReachHere();
324 }
325 BLOCK_COMMENT("}");
326 }
327
328
329 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register cpe_offset,
330 int bcp_offset, size_t index_size) {
331 BLOCK_COMMENT("get_cache_and_index_at_bcp {");
332 assert_different_registers(cache, cpe_offset);
333 get_cache_index_at_bcp(cpe_offset, bcp_offset, index_size);
334 z_lg(cache, Address(Z_fp, _z_ijava_state_neg(cpoolCache)));
335 // Convert from field index to ConstantPoolCache offset in bytes.
336 z_sllg(cpe_offset, cpe_offset, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord));
337 BLOCK_COMMENT("}");
338 }
339
340 // Kills Z_R0_scratch.
341 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
342 Register cpe_offset,
343 Register bytecode,
344 int byte_no,
345 int bcp_offset,
346 size_t index_size) {
347 BLOCK_COMMENT("get_cache_and_index_and_bytecode_at_bcp {");
348 get_cache_and_index_at_bcp(cache, cpe_offset, bcp_offset, index_size);
349
350 // We want to load (from CP cache) the bytecode that corresponds to the passed-in byte_no.
351 // It is located at (cache + cpe_offset + base_offset + indices_offset + (8-1) (last byte in DW) - (byte_no+1).
352 // Instead of loading, shifting and masking a DW, we just load that one byte of interest with z_llgc (unsigned).
353 const int base_ix_off = in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset());
354 const int off_in_DW = (8-1) - (1+byte_no);
355 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
356 assert(ConstantPoolCacheEntry::bytecode_1_mask == 0xff, "");
357 load_sized_value(bytecode, Address(cache, cpe_offset, base_ix_off+off_in_DW), 1, false /*signed*/);
358
359 BLOCK_COMMENT("}");
360 }
361
362 // Load object from cpool->resolved_references(index).
363 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) {
364 assert_different_registers(result, index);
365 get_constant_pool(result);
366
367 // Convert
368 // - from field index to resolved_references() index and
369 // - from word index to byte offset.
370 // Since this is a java object, it is potentially compressed.
371 Register tmp = index; // reuse
372 z_sllg(index, index, LogBytesPerHeapOop); // Offset into resolved references array.
373 // Load pointer for resolved_references[] objArray.
374 z_lg(result, ConstantPool::cache_offset_in_bytes(), result);
375 z_lg(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result);
376 // JNIHandles::resolve(result)
377 z_lg(result, 0, result); // Load resolved references array itself.
378 #ifdef ASSERT
379 NearLabel index_ok;
380 z_lgf(Z_R0, Address(result, arrayOopDesc::length_offset_in_bytes()));
381 z_sllg(Z_R0, Z_R0, LogBytesPerHeapOop);
382 compare64_and_branch(tmp, Z_R0, Assembler::bcondLow, index_ok);
383 stop("resolved reference index out of bounds", 0x09256);
384 bind(index_ok);
385 #endif
386 z_agr(result, index); // Address of indexed array element.
387 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result);
388 }
389
390 // load cpool->resolved_klass_at(index)
391 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register cpool, Register offset, Register iklass) {
392 // int value = *(Rcpool->int_at_addr(which));
393 // int resolved_klass_index = extract_low_short_from_int(value);
394 z_llgh(offset, Address(cpool, offset, sizeof(ConstantPool) + 2)); // offset = resolved_klass_index (s390 is big-endian)
395 z_sllg(offset, offset, LogBytesPerWord); // Convert 'index' to 'offset'
396 z_lg(iklass, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); // iklass = cpool->_resolved_klasses
397 z_lg(iklass, Address(iklass, offset, Array<Klass*>::base_offset_in_bytes()));
398 }
399
400 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
401 Register tmp,
402 int bcp_offset,
403 size_t index_size) {
404 BLOCK_COMMENT("get_cache_entry_pointer_at_bcp {");
405 get_cache_and_index_at_bcp(cache, tmp, bcp_offset, index_size);
406 add2reg_with_index(cache, in_bytes(ConstantPoolCache::base_offset()), tmp, cache);
407 BLOCK_COMMENT("}");
408 }
409
410 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
411 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2.
412 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
413 Register Rsuper_klass,
414 Register Rtmp1,
415 Register Rtmp2,
416 Label &ok_is_subtype) {
417 // Profile the not-null value's klass.
418 profile_typecheck(Rtmp1, Rsub_klass, Rtmp2);
419
420 // Do the check.
421 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
422
423 // Profile the failure of the check.
424 profile_typecheck_failed(Rtmp1, Rtmp2);
425 }
426
427 // Pop topmost element from stack. It just disappears.
428 // Useful if consumed previously by access via stackTop().
429 void InterpreterMacroAssembler::popx(int len) {
430 add2reg(Z_esp, len*Interpreter::stackElementSize);
431 debug_only(verify_esp(Z_esp, Z_R1_scratch));
432 }
433
434 // Get Address object of stack top. No checks. No pop.
435 // Purpose: - Provide address of stack operand to exploit reg-mem operations.
436 // - Avoid RISC-like mem2reg - reg-reg-op sequence.
437 Address InterpreterMacroAssembler::stackTop() {
438 return Address(Z_esp, Interpreter::expr_offset_in_bytes(0));
439 }
440
441 void InterpreterMacroAssembler::pop_i(Register r) {
442 z_l(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
443 add2reg(Z_esp, Interpreter::stackElementSize);
444 assert_different_registers(r, Z_R1_scratch);
445 debug_only(verify_esp(Z_esp, Z_R1_scratch));
446 }
447
448 void InterpreterMacroAssembler::pop_ptr(Register r) {
449 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
450 add2reg(Z_esp, Interpreter::stackElementSize);
451 assert_different_registers(r, Z_R1_scratch);
452 debug_only(verify_esp(Z_esp, Z_R1_scratch));
453 }
454
455 void InterpreterMacroAssembler::pop_l(Register r) {
456 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
457 add2reg(Z_esp, 2*Interpreter::stackElementSize);
458 assert_different_registers(r, Z_R1_scratch);
459 debug_only(verify_esp(Z_esp, Z_R1_scratch));
460 }
461
462 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
463 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), false);
464 add2reg(Z_esp, Interpreter::stackElementSize);
465 debug_only(verify_esp(Z_esp, Z_R1_scratch));
466 }
467
468 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
469 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), true);
470 add2reg(Z_esp, 2*Interpreter::stackElementSize);
471 debug_only(verify_esp(Z_esp, Z_R1_scratch));
472 }
473
474 void InterpreterMacroAssembler::push_i(Register r) {
475 assert_different_registers(r, Z_R1_scratch);
476 debug_only(verify_esp(Z_esp, Z_R1_scratch));
477 z_st(r, Address(Z_esp));
478 add2reg(Z_esp, -Interpreter::stackElementSize);
479 }
480
481 void InterpreterMacroAssembler::push_ptr(Register r) {
482 z_stg(r, Address(Z_esp));
483 add2reg(Z_esp, -Interpreter::stackElementSize);
484 }
485
486 void InterpreterMacroAssembler::push_l(Register r) {
487 assert_different_registers(r, Z_R1_scratch);
488 debug_only(verify_esp(Z_esp, Z_R1_scratch));
489 int offset = -Interpreter::stackElementSize;
490 z_stg(r, Address(Z_esp, offset));
491 clear_mem(Address(Z_esp), Interpreter::stackElementSize);
492 add2reg(Z_esp, 2 * offset);
493 }
494
495 void InterpreterMacroAssembler::push_f(FloatRegister f) {
496 debug_only(verify_esp(Z_esp, Z_R1_scratch));
497 freg2mem_opt(f, Address(Z_esp), false);
498 add2reg(Z_esp, -Interpreter::stackElementSize);
499 }
500
501 void InterpreterMacroAssembler::push_d(FloatRegister d) {
502 debug_only(verify_esp(Z_esp, Z_R1_scratch));
503 int offset = -Interpreter::stackElementSize;
504 freg2mem_opt(d, Address(Z_esp, offset));
505 add2reg(Z_esp, 2 * offset);
506 }
507
508 void InterpreterMacroAssembler::push(TosState state) {
509 verify_oop(Z_tos, state);
510 switch (state) {
511 case atos: push_ptr(); break;
512 case btos: push_i(); break;
513 case ztos:
514 case ctos:
515 case stos: push_i(); break;
516 case itos: push_i(); break;
517 case ltos: push_l(); break;
518 case ftos: push_f(); break;
519 case dtos: push_d(); break;
520 case vtos: /* nothing to do */ break;
521 default : ShouldNotReachHere();
522 }
523 }
524
525 void InterpreterMacroAssembler::pop(TosState state) {
526 switch (state) {
527 case atos: pop_ptr(Z_tos); break;
528 case btos: pop_i(Z_tos); break;
529 case ztos:
530 case ctos:
531 case stos: pop_i(Z_tos); break;
532 case itos: pop_i(Z_tos); break;
533 case ltos: pop_l(Z_tos); break;
534 case ftos: pop_f(Z_ftos); break;
535 case dtos: pop_d(Z_ftos); break;
536 case vtos: /* nothing to do */ break;
537 default : ShouldNotReachHere();
538 }
539 verify_oop(Z_tos, state);
540 }
541
542 // Helpers for swap and dup.
543 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
544 z_lg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n)));
545 }
546
547 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
548 z_stg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n)));
549 }
550
551 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted(Register method) {
552 // Satisfy interpreter calling convention (see generate_normal_entry()).
553 z_lgr(Z_R10, Z_SP); // Set sender sp (aka initial caller sp, aka unextended sp).
554 // Record top_frame_sp, because the callee might modify it, if it's compiled.
555 z_stg(Z_SP, _z_ijava_state_neg(top_frame_sp), Z_fp);
556 save_bcp();
557 save_esp();
558 z_lgr(Z_method, method); // Set Z_method (kills Z_fp!).
559 }
560
561 // Jump to from_interpreted entry of a call unless single stepping is possible
562 // in this thread in which case we must call the i2i entry.
563 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
564 assert_different_registers(method, Z_R10 /*used for initial_caller_sp*/, temp);
565 prepare_to_jump_from_interpreted(method);
566
567 if (JvmtiExport::can_post_interpreter_events()) {
568 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
569 // compiled code in threads for which the event is enabled. Check here for
570 // interp_only_mode if these events CAN be enabled.
571 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
572 MacroAssembler::load_and_test_int(Z_R0_scratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
573 z_bcr(bcondEqual, Z_R1_scratch); // Run compiled code if zero.
574 // Run interpreted.
575 z_lg(Z_R1_scratch, Address(method, Method::interpreter_entry_offset()));
576 z_br(Z_R1_scratch);
577 } else {
578 // Run compiled code.
579 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
580 z_br(Z_R1_scratch);
581 }
582 }
583
584 #ifdef ASSERT
585 void InterpreterMacroAssembler::verify_esp(Register Resp, Register Rtemp) {
586 // About to read or write Resp[0].
587 // Make sure it is not in the monitors or the TOP_IJAVA_FRAME_ABI.
588 address reentry = NULL;
589
590 {
591 // Check if the frame pointer in Z_fp is correct.
592 NearLabel OK;
593 z_cg(Z_fp, 0, Z_SP);
594 z_bre(OK);
595 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp");
596 bind(OK);
597 }
598 {
599 // Resp must not point into or below the operand stack,
600 // i.e. IJAVA_STATE.monitors > Resp.
601 NearLabel OK;
602 Register Rmonitors = Rtemp;
603 z_lg(Rmonitors, _z_ijava_state_neg(monitors), Z_fp);
604 compareU64_and_branch(Rmonitors, Resp, bcondHigh, OK);
605 reentry = stop_chain_static(reentry, "too many pops: Z_esp points into monitor area");
606 bind(OK);
607 }
608 {
609 // Resp may point to the last word of TOP_IJAVA_FRAME_ABI, but not below
610 // i.e. !(Z_SP + frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize > Resp).
611 NearLabel OK;
612 Register Rabi_bottom = Rtemp;
613 add2reg(Rabi_bottom, frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize, Z_SP);
614 compareU64_and_branch(Rabi_bottom, Resp, bcondNotHigh, OK);
615 reentry = stop_chain_static(reentry, "too many pushes: Z_esp points into TOP_IJAVA_FRAME_ABI");
616 bind(OK);
617 }
618 }
619
620 void InterpreterMacroAssembler::asm_assert_ijava_state_magic(Register tmp) {
621 Label magic_ok;
622 load_const_optimized(tmp, frame::z_istate_magic_number);
623 z_cg(tmp, Address(Z_fp, _z_ijava_state_neg(magic)));
624 z_bre(magic_ok);
625 stop_static("error: wrong magic number in ijava_state access");
626 bind(magic_ok);
627 }
628 #endif // ASSERT
629
630 void InterpreterMacroAssembler::save_bcp() {
631 z_stg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
632 asm_assert_ijava_state_magic(Z_bcp);
633 NOT_PRODUCT(z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))));
634 }
635
636 void InterpreterMacroAssembler::restore_bcp() {
637 asm_assert_ijava_state_magic(Z_bcp);
638 z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
639 }
640
641 void InterpreterMacroAssembler::save_esp() {
642 z_stg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp)));
643 }
644
645 void InterpreterMacroAssembler::restore_esp() {
646 asm_assert_ijava_state_magic(Z_esp);
647 z_lg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp)));
648 }
649
650 void InterpreterMacroAssembler::get_monitors(Register reg) {
651 asm_assert_ijava_state_magic(reg);
652 mem2reg_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
653 }
654
655 void InterpreterMacroAssembler::save_monitors(Register reg) {
656 reg2mem_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
657 }
658
659 void InterpreterMacroAssembler::get_mdp(Register mdp) {
660 z_lg(mdp, _z_ijava_state_neg(mdx), Z_fp);
661 }
662
663 void InterpreterMacroAssembler::save_mdp(Register mdp) {
664 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
665 }
666
667 // Values that are only read (besides initialization).
668 void InterpreterMacroAssembler::restore_locals() {
669 asm_assert_ijava_state_magic(Z_locals);
670 z_lg(Z_locals, Address(Z_fp, _z_ijava_state_neg(locals)));
671 }
672
673 void InterpreterMacroAssembler::get_method(Register reg) {
674 asm_assert_ijava_state_magic(reg);
675 z_lg(reg, Address(Z_fp, _z_ijava_state_neg(method)));
676 }
677
678 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(Register Rdst, int bcp_offset,
679 signedOrNot is_signed) {
680 // Rdst is an 8-byte return value!!!
681
682 // Unaligned loads incur only a small penalty on z/Architecture. The penalty
683 // is a few (2..3) ticks, even when the load crosses a cache line
684 // boundary. In case of a cache miss, the stall could, of course, be
685 // much longer.
686
687 switch (is_signed) {
688 case Signed:
689 z_lgh(Rdst, bcp_offset, Z_R0, Z_bcp);
690 break;
691 case Unsigned:
692 z_llgh(Rdst, bcp_offset, Z_R0, Z_bcp);
693 break;
694 default:
695 ShouldNotReachHere();
696 }
697 }
698
699
700 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(Register Rdst, int bcp_offset,
701 setCCOrNot set_cc) {
702 // Rdst is an 8-byte return value!!!
703
704 // Unaligned loads incur only a small penalty on z/Architecture. The penalty
705 // is a few (2..3) ticks, even when the load crosses a cache line
706 // boundary. In case of a cache miss, the stall could, of course, be
707 // much longer.
708
709 // Both variants implement a sign-extending int2long load.
710 if (set_cc == set_CC) {
711 load_and_test_int2long(Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
712 } else {
713 mem2reg_signed_opt( Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
714 }
715 }
716
717 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
718 get_method(Rdst);
719 mem2reg_opt(Rdst, Address(Rdst, Method::const_offset()));
720 mem2reg_opt(Rdst, Address(Rdst, ConstMethod::constants_offset()));
721 }
722
723 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
724 get_constant_pool(Rcpool);
725 mem2reg_opt(Rtags, Address(Rcpool, ConstantPool::tags_offset_in_bytes()));
726 }
727
728 // Unlock if synchronized method.
729 //
730 // Unlock the receiver if this is a synchronized method.
731 // Unlock any Java monitors from syncronized blocks.
732 //
733 // If there are locked Java monitors
734 // If throw_monitor_exception
735 // throws IllegalMonitorStateException
736 // Else if install_monitor_exception
737 // installs IllegalMonitorStateException
738 // Else
739 // no error processing
740 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
741 bool throw_monitor_exception,
742 bool install_monitor_exception) {
743 NearLabel unlocked, unlock, no_unlock;
744
745 {
746 Register R_method = Z_ARG2;
747 Register R_do_not_unlock_if_synchronized = Z_ARG3;
748
749 // Get the value of _do_not_unlock_if_synchronized into G1_scratch.
750 const Address do_not_unlock_if_synchronized(Z_thread,
751 JavaThread::do_not_unlock_if_synchronized_offset());
752 load_sized_value(R_do_not_unlock_if_synchronized, do_not_unlock_if_synchronized, 1, false /*unsigned*/);
753 z_mvi(do_not_unlock_if_synchronized, false); // Reset the flag.
754
755 // Check if synchronized method.
756 get_method(R_method);
757 verify_oop(Z_tos, state);
758 push(state); // Save tos/result.
759 testbit(method2_(R_method, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
760 z_bfalse(unlocked);
761
762 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
763 // is set.
764 compareU64_and_branch(R_do_not_unlock_if_synchronized, (intptr_t)0L, bcondNotEqual, no_unlock);
765 }
766
767 // unlock monitor
768
769 // BasicObjectLock will be first in list, since this is a
770 // synchronized method. However, need to check that the object has
771 // not been unlocked by an explicit monitorexit bytecode.
772 const Address monitor(Z_fp, -(frame::z_ijava_state_size + (int) sizeof(BasicObjectLock)));
773 // We use Z_ARG2 so that if we go slow path it will be the correct
774 // register for unlock_object to pass to VM directly.
775 load_address(Z_ARG2, monitor); // Address of first monitor.
776 z_lg(Z_ARG3, Address(Z_ARG2, BasicObjectLock::obj_offset_in_bytes()));
777 compareU64_and_branch(Z_ARG3, (intptr_t)0L, bcondNotEqual, unlock);
778
779 if (throw_monitor_exception) {
780 // Entry already unlocked need to throw an exception.
781 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
782 should_not_reach_here();
783 } else {
784 // Monitor already unlocked during a stack unroll.
785 // If requested, install an illegal_monitor_state_exception.
786 // Continue with stack unrolling.
787 if (install_monitor_exception) {
788 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
789 }
790 z_bru(unlocked);
791 }
792
793 bind(unlock);
794
795 unlock_object(Z_ARG2);
796
797 bind(unlocked);
798
799 // I0, I1: Might contain return value
800
801 // Check that all monitors are unlocked.
802 {
803 NearLabel loop, exception, entry, restart;
804 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
805 // We use Z_ARG2 so that if we go slow path it will be the correct
806 // register for unlock_object to pass to VM directly.
807 Register R_current_monitor = Z_ARG2;
808 Register R_monitor_block_bot = Z_ARG1;
809 const Address monitor_block_top(Z_fp, _z_ijava_state_neg(monitors));
810 const Address monitor_block_bot(Z_fp, -frame::z_ijava_state_size);
811
812 bind(restart);
813 // Starting with top-most entry.
814 z_lg(R_current_monitor, monitor_block_top);
815 // Points to word before bottom of monitor block.
816 load_address(R_monitor_block_bot, monitor_block_bot);
817 z_bru(entry);
818
819 // Entry already locked, need to throw exception.
820 bind(exception);
821
822 if (throw_monitor_exception) {
823 // Throw exception.
824 MacroAssembler::call_VM(noreg,
825 CAST_FROM_FN_PTR(address, InterpreterRuntime::
826 throw_illegal_monitor_state_exception));
827 should_not_reach_here();
828 } else {
829 // Stack unrolling. Unlock object and install illegal_monitor_exception.
830 // Unlock does not block, so don't have to worry about the frame.
831 // We don't have to preserve c_rarg1 since we are going to throw an exception.
832 unlock_object(R_current_monitor);
833 if (install_monitor_exception) {
834 call_VM(noreg, CAST_FROM_FN_PTR(address,
835 InterpreterRuntime::
836 new_illegal_monitor_state_exception));
837 }
838 z_bru(restart);
839 }
840
841 bind(loop);
842 // Check if current entry is used.
843 load_and_test_long(Z_R0_scratch, Address(R_current_monitor, BasicObjectLock::obj_offset_in_bytes()));
844 z_brne(exception);
845
846 add2reg(R_current_monitor, entry_size); // Otherwise advance to next entry.
847 bind(entry);
848 compareU64_and_branch(R_current_monitor, R_monitor_block_bot, bcondNotEqual, loop);
849 }
850
851 bind(no_unlock);
852 pop(state);
853 verify_oop(Z_tos, state);
854 }
855
856 // remove activation
857 //
858 // Unlock the receiver if this is a synchronized method.
859 // Unlock any Java monitors from syncronized blocks.
860 // Remove the activation from the stack.
861 //
862 // If there are locked Java monitors
863 // If throw_monitor_exception
864 // throws IllegalMonitorStateException
865 // Else if install_monitor_exception
866 // installs IllegalMonitorStateException
867 // Else
868 // no error processing
869 void InterpreterMacroAssembler::remove_activation(TosState state,
870 Register return_pc,
871 bool throw_monitor_exception,
872 bool install_monitor_exception,
873 bool notify_jvmti) {
874 BLOCK_COMMENT("remove_activation {");
875 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
876
877 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
878 notify_method_exit(false, state, notify_jvmti ? NotifyJVMTI : SkipNotifyJVMTI);
879
880 if (StackReservedPages > 0) {
881 BLOCK_COMMENT("reserved_stack_check:");
882 // Test if reserved zone needs to be enabled.
883 Label no_reserved_zone_enabling;
884
885 // Compare frame pointers. There is no good stack pointer, as with stack
886 // frame compression we can get different SPs when we do calls. A subsequent
887 // call could have a smaller SP, so that this compare succeeds for an
888 // inner call of the method annotated with ReservedStack.
889 z_lg(Z_R0, Address(Z_SP, (intptr_t)_z_abi(callers_sp)));
890 z_clg(Z_R0, Address(Z_thread, JavaThread::reserved_stack_activation_offset())); // Compare with frame pointer in memory.
891 z_brl(no_reserved_zone_enabling);
892
893 // Enable reserved zone again, throw stack overflow exception.
894 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), Z_thread);
895 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
896
897 should_not_reach_here();
898
899 bind(no_reserved_zone_enabling);
900 }
901
902 verify_oop(Z_tos, state);
903 verify_thread();
904
905 pop_interpreter_frame(return_pc, Z_ARG2, Z_ARG3);
906 BLOCK_COMMENT("} remove_activation");
907 }
908
909 // lock object
910 //
911 // Registers alive
912 // monitor - Address of the BasicObjectLock to be used for locking,
913 // which must be initialized with the object to lock.
914 // object - Address of the object to be locked.
915 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
916
917 if (UseHeavyMonitors) {
918 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
919 monitor, /*check_for_exceptions=*/false);
920 return;
921 }
922
923 // template code:
924 //
925 // markOop displaced_header = obj->mark().set_unlocked();
926 // monitor->lock()->set_displaced_header(displaced_header);
927 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
928 // // We stored the monitor address into the object's mark word.
929 // } else if (THREAD->is_lock_owned((address)displaced_header))
930 // // Simple recursive case.
931 // monitor->lock()->set_displaced_header(NULL);
932 // } else {
933 // // Slow path.
934 // InterpreterRuntime::monitorenter(THREAD, monitor);
935 // }
936
937 const Register displaced_header = Z_ARG5;
938 const Register object_mark_addr = Z_ARG4;
939 const Register current_header = Z_ARG5;
940
941 NearLabel done;
942 NearLabel slow_case;
943
944 // markOop displaced_header = obj->mark().set_unlocked();
945
946 // Load markOop from object into displaced_header.
947 z_lg(displaced_header, oopDesc::mark_offset_in_bytes(), object);
948
949 if (UseBiasedLocking) {
950 biased_locking_enter(object, displaced_header, Z_R1, Z_R0, done, &slow_case);
951 }
952
953 // Set displaced_header to be (markOop of object | UNLOCK_VALUE).
954 z_oill(displaced_header, markOopDesc::unlocked_value);
955
956 // monitor->lock()->set_displaced_header(displaced_header);
957
958 // Initialize the box (Must happen before we update the object mark!).
959 z_stg(displaced_header, BasicObjectLock::lock_offset_in_bytes() +
960 BasicLock::displaced_header_offset_in_bytes(), monitor);
961
962 // if (Atomic::cmpxchg_ptr(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
963
964 // Store stack address of the BasicObjectLock (this is monitor) into object.
965 add2reg(object_mark_addr, oopDesc::mark_offset_in_bytes(), object);
966
967 z_csg(displaced_header, monitor, 0, object_mark_addr);
968 assert(current_header==displaced_header, "must be same register"); // Identified two registers from z/Architecture.
969
970 z_bre(done);
971
972 // } else if (THREAD->is_lock_owned((address)displaced_header))
973 // // Simple recursive case.
974 // monitor->lock()->set_displaced_header(NULL);
975
976 // We did not see an unlocked object so try the fast recursive case.
977
978 // Check if owner is self by comparing the value in the markOop of object
979 // (current_header) with the stack pointer.
980 z_sgr(current_header, Z_SP);
981
982 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
983
984 // The prior sequence "LGR, NGR, LTGR" can be done better
985 // (Z_R1 is temp and not used after here).
986 load_const_optimized(Z_R0, (~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place));
987 z_ngr(Z_R0, current_header); // AND sets CC (result eq/ne 0)
988
989 // If condition is true we are done and hence we can store 0 in the displaced
990 // header indicating it is a recursive lock and be done.
991 z_brne(slow_case);
992 z_release(); // Membar unnecessary on zarch AND because the above csg does a sync before and after.
993 z_stg(Z_R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() +
994 BasicLock::displaced_header_offset_in_bytes(), monitor);
995 z_bru(done);
996
997 // } else {
998 // // Slow path.
999 // InterpreterRuntime::monitorenter(THREAD, monitor);
1000
1001 // None of the above fast optimizations worked so we have to get into the
1002 // slow case of monitor enter.
1003 bind(slow_case);
1004
1005 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1006 monitor, /*check_for_exceptions=*/false);
1007
1008 // }
1009
1010 bind(done);
1011 }
1012
1013 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1014 //
1015 // Registers alive
1016 // monitor - address of the BasicObjectLock to be used for locking,
1017 // which must be initialized with the object to lock.
1018 //
1019 // Throw IllegalMonitorException if object is not locked by current thread.
1020 void InterpreterMacroAssembler::unlock_object(Register monitor, Register object) {
1021
1022 if (UseHeavyMonitors) {
1023 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1024 monitor, /*check_for_exceptions=*/ true);
1025 return;
1026 }
1027
1028 // else {
1029 // template code:
1030 //
1031 // if ((displaced_header = monitor->displaced_header()) == NULL) {
1032 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL.
1033 // monitor->set_obj(NULL);
1034 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) {
1035 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1036 // monitor->set_obj(NULL);
1037 // } else {
1038 // // Slow path.
1039 // InterpreterRuntime::monitorexit(THREAD, monitor);
1040 // }
1041
1042 const Register displaced_header = Z_ARG4;
1043 const Register current_header = Z_R1;
1044 Address obj_entry(monitor, BasicObjectLock::obj_offset_in_bytes());
1045 Label done;
1046
1047 if (object == noreg) {
1048 // In the template interpreter, we must assure that the object
1049 // entry in the monitor is cleared on all paths. Thus we move
1050 // loading up to here, and clear the entry afterwards.
1051 object = Z_ARG3; // Use Z_ARG3 if caller didn't pass object.
1052 z_lg(object, obj_entry);
1053 }
1054
1055 assert_different_registers(monitor, object, displaced_header, current_header);
1056
1057 // if ((displaced_header = monitor->displaced_header()) == NULL) {
1058 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL.
1059 // monitor->set_obj(NULL);
1060
1061 clear_mem(obj_entry, sizeof(oop));
1062
1063 if (UseBiasedLocking) {
1064 // The object address from the monitor is in object.
1065 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0");
1066 biased_locking_exit(object, displaced_header, done);
1067 }
1068
1069 // Test first if we are in the fast recursive case.
1070 MacroAssembler::load_and_test_long(displaced_header,
1071 Address(monitor, BasicObjectLock::lock_offset_in_bytes() +
1072 BasicLock::displaced_header_offset_in_bytes()));
1073 z_bre(done); // displaced_header == 0 -> goto done
1074
1075 // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) {
1076 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1077 // monitor->set_obj(NULL);
1078
1079 // If we still have a lightweight lock, unlock the object and be done.
1080
1081 // The markword is expected to be at offset 0.
1082 assert(oopDesc::mark_offset_in_bytes() == 0, "unlock_object: review code below");
1083
1084 // We have the displaced header in displaced_header. If the lock is still
1085 // lightweight, it will contain the monitor address and we'll store the
1086 // displaced header back into the object's mark word.
1087 z_lgr(current_header, monitor);
1088 z_csg(current_header, displaced_header, 0, object);
1089 z_bre(done);
1090
1091 // } else {
1092 // // Slow path.
1093 // InterpreterRuntime::monitorexit(THREAD, monitor);
1094
1095 // The lock has been converted into a heavy lock and hence
1096 // we need to get into the slow case.
1097 z_stg(object, obj_entry); // Restore object entry, has been cleared above.
1098 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1099 monitor, /*check_for_exceptions=*/false);
1100
1101 // }
1102
1103 bind(done);
1104 }
1105
1106 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) {
1107 assert(ProfileInterpreter, "must be profiling interpreter");
1108 load_and_test_long(mdp, Address(Z_fp, _z_ijava_state_neg(mdx)));
1109 z_brz(zero_continue);
1110 }
1111
1112 // Set the method data pointer for the current bcp.
1113 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1114 assert(ProfileInterpreter, "must be profiling interpreter");
1115 Label set_mdp;
1116 Register mdp = Z_ARG4;
1117 Register method = Z_ARG5;
1118
1119 get_method(method);
1120 // Test MDO to avoid the call if it is NULL.
1121 load_and_test_long(mdp, method2_(method, method_data));
1122 z_brz(set_mdp);
1123
1124 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), method, Z_bcp);
1125 // Z_RET: mdi
1126 // Mdo is guaranteed to be non-zero here, we checked for it before the call.
1127 assert(method->is_nonvolatile(), "choose nonvolatile reg or reload from frame");
1128 z_lg(mdp, method2_(method, method_data)); // Must reload, mdp is volatile reg.
1129 add2reg_with_index(mdp, in_bytes(MethodData::data_offset()), Z_RET, mdp);
1130
1131 bind(set_mdp);
1132 save_mdp(mdp);
1133 }
1134
1135 void InterpreterMacroAssembler::verify_method_data_pointer() {
1136 assert(ProfileInterpreter, "must be profiling interpreter");
1137 #ifdef ASSERT
1138 NearLabel verify_continue;
1139 Register bcp_expected = Z_ARG3;
1140 Register mdp = Z_ARG4;
1141 Register method = Z_ARG5;
1142
1143 test_method_data_pointer(mdp, verify_continue); // If mdp is zero, continue
1144 get_method(method);
1145
1146 // If the mdp is valid, it will point to a DataLayout header which is
1147 // consistent with the bcp. The converse is highly probable also.
1148 load_sized_value(bcp_expected, Address(mdp, DataLayout::bci_offset()), 2, false /*signed*/);
1149 z_ag(bcp_expected, Address(method, Method::const_offset()));
1150 load_address(bcp_expected, Address(bcp_expected, ConstMethod::codes_offset()));
1151 compareU64_and_branch(bcp_expected, Z_bcp, bcondEqual, verify_continue);
1152 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), method, Z_bcp, mdp);
1153 bind(verify_continue);
1154 #endif // ASSERT
1155 }
1156
1157 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) {
1158 assert(ProfileInterpreter, "must be profiling interpreter");
1159 z_stg(value, constant, mdp_in);
1160 }
1161
1162 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1163 int constant,
1164 Register tmp,
1165 bool decrement) {
1166 assert_different_registers(mdp_in, tmp);
1167 // counter address
1168 Address data(mdp_in, constant);
1169 const int delta = decrement ? -DataLayout::counter_increment : DataLayout::counter_increment;
1170 add2mem_64(Address(mdp_in, constant), delta, tmp);
1171 }
1172
1173 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1174 int flag_byte_constant) {
1175 assert(ProfileInterpreter, "must be profiling interpreter");
1176 // Set the flag.
1177 z_oi(Address(mdp_in, DataLayout::flags_offset()), flag_byte_constant);
1178 }
1179
1180 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1181 int offset,
1182 Register value,
1183 Register test_value_out,
1184 Label& not_equal_continue) {
1185 assert(ProfileInterpreter, "must be profiling interpreter");
1186 if (test_value_out == noreg) {
1187 z_cg(value, Address(mdp_in, offset));
1188 z_brne(not_equal_continue);
1189 } else {
1190 // Put the test value into a register, so caller can use it:
1191 z_lg(test_value_out, Address(mdp_in, offset));
1192 compareU64_and_branch(test_value_out, value, bcondNotEqual, not_equal_continue);
1193 }
1194 }
1195
1196 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) {
1197 update_mdp_by_offset(mdp_in, noreg, offset_of_disp);
1198 }
1199
1200 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1201 Register dataidx,
1202 int offset_of_disp) {
1203 assert(ProfileInterpreter, "must be profiling interpreter");
1204 Address disp_address(mdp_in, dataidx, offset_of_disp);
1205 Assembler::z_ag(mdp_in, disp_address);
1206 save_mdp(mdp_in);
1207 }
1208
1209 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) {
1210 assert(ProfileInterpreter, "must be profiling interpreter");
1211 add2reg(mdp_in, constant);
1212 save_mdp(mdp_in);
1213 }
1214
1215 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1216 assert(ProfileInterpreter, "must be profiling interpreter");
1217 assert(return_bci->is_nonvolatile(), "choose nonvolatile reg or save/restore");
1218 call_VM(noreg,
1219 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1220 return_bci);
1221 }
1222
1223 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) {
1224 if (ProfileInterpreter) {
1225 Label profile_continue;
1226
1227 // If no method data exists, go to profile_continue.
1228 // Otherwise, assign to mdp.
1229 test_method_data_pointer(mdp, profile_continue);
1230
1231 // We are taking a branch. Increment the taken count.
1232 // We inline increment_mdp_data_at to return bumped_count in a register
1233 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1234 Address data(mdp, JumpData::taken_offset());
1235 z_lg(bumped_count, data);
1236 // 64-bit overflow is very unlikely. Saturation to 32-bit values is
1237 // performed when reading the counts.
1238 add2reg(bumped_count, DataLayout::counter_increment);
1239 z_stg(bumped_count, data); // Store back out
1240
1241 // The method data pointer needs to be updated to reflect the new target.
1242 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1243 bind(profile_continue);
1244 }
1245 }
1246
1247 // Kills Z_R1_scratch.
1248 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1249 if (ProfileInterpreter) {
1250 Label profile_continue;
1251
1252 // If no method data exists, go to profile_continue.
1253 test_method_data_pointer(mdp, profile_continue);
1254
1255 // We are taking a branch. Increment the not taken count.
1256 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()), Z_R1_scratch);
1257
1258 // The method data pointer needs to be updated to correspond to
1259 // the next bytecode.
1260 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1261 bind(profile_continue);
1262 }
1263 }
1264
1265 // Kills: Z_R1_scratch.
1266 void InterpreterMacroAssembler::profile_call(Register mdp) {
1267 if (ProfileInterpreter) {
1268 Label profile_continue;
1269
1270 // If no method data exists, go to profile_continue.
1271 test_method_data_pointer(mdp, profile_continue);
1272
1273 // We are making a call. Increment the count.
1274 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1275
1276 // The method data pointer needs to be updated to reflect the new target.
1277 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1278 bind(profile_continue);
1279 }
1280 }
1281
1282 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1283 if (ProfileInterpreter) {
1284 Label profile_continue;
1285
1286 // If no method data exists, go to profile_continue.
1287 test_method_data_pointer(mdp, profile_continue);
1288
1289 // We are making a call. Increment the count.
1290 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1291
1292 // The method data pointer needs to be updated to reflect the new target.
1293 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1294 bind(profile_continue);
1295 }
1296 }
1297
1298 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1299 Register mdp,
1300 Register reg2,
1301 bool receiver_can_be_null) {
1302 if (ProfileInterpreter) {
1303 NearLabel profile_continue;
1304
1305 // If no method data exists, go to profile_continue.
1306 test_method_data_pointer(mdp, profile_continue);
1307
1308 NearLabel skip_receiver_profile;
1309 if (receiver_can_be_null) {
1310 NearLabel not_null;
1311 compareU64_and_branch(receiver, (intptr_t)0L, bcondNotEqual, not_null);
1312 // We are making a call. Increment the count for null receiver.
1313 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1314 z_bru(skip_receiver_profile);
1315 bind(not_null);
1316 }
1317
1318 // Record the receiver type.
1319 record_klass_in_profile(receiver, mdp, reg2, true);
1320 bind(skip_receiver_profile);
1321
1322 // The method data pointer needs to be updated to reflect the new target.
1323 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1324 bind(profile_continue);
1325 }
1326 }
1327
1328 // This routine creates a state machine for updating the multi-row
1329 // type profile at a virtual call site (or other type-sensitive bytecode).
1330 // The machine visits each row (of receiver/count) until the receiver type
1331 // is found, or until it runs out of rows. At the same time, it remembers
1332 // the location of the first empty row. (An empty row records null for its
1333 // receiver, and can be allocated for a newly-observed receiver type.)
1334 // Because there are two degrees of freedom in the state, a simple linear
1335 // search will not work; it must be a decision tree. Hence this helper
1336 // function is recursive, to generate the required tree structured code.
1337 // It's the interpreter, so we are trading off code space for speed.
1338 // See below for example code.
1339 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1340 Register receiver, Register mdp,
1341 Register reg2, int start_row,
1342 Label& done, bool is_virtual_call) {
1343 if (TypeProfileWidth == 0) {
1344 if (is_virtual_call) {
1345 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1346 }
1347 return;
1348 }
1349
1350 int last_row = VirtualCallData::row_limit() - 1;
1351 assert(start_row <= last_row, "must be work left to do");
1352 // Test this row for both the receiver and for null.
1353 // Take any of three different outcomes:
1354 // 1. found receiver => increment count and goto done
1355 // 2. found null => keep looking for case 1, maybe allocate this cell
1356 // 3. found something else => keep looking for cases 1 and 2
1357 // Case 3 is handled by a recursive call.
1358 for (int row = start_row; row <= last_row; row++) {
1359 NearLabel next_test;
1360 bool test_for_null_also = (row == start_row);
1361
1362 // See if the receiver is receiver[n].
1363 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1364 test_mdp_data_at(mdp, recvr_offset, receiver,
1365 (test_for_null_also ? reg2 : noreg),
1366 next_test);
1367 // (Reg2 now contains the receiver from the CallData.)
1368
1369 // The receiver is receiver[n]. Increment count[n].
1370 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1371 increment_mdp_data_at(mdp, count_offset);
1372 z_bru(done);
1373 bind(next_test);
1374
1375 if (test_for_null_also) {
1376 Label found_null;
1377 // Failed the equality check on receiver[n]... Test for null.
1378 z_ltgr(reg2, reg2);
1379 if (start_row == last_row) {
1380 // The only thing left to do is handle the null case.
1381 if (is_virtual_call) {
1382 z_brz(found_null);
1383 // Receiver did not match any saved receiver and there is no empty row for it.
1384 // Increment total counter to indicate polymorphic case.
1385 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1386 z_bru(done);
1387 bind(found_null);
1388 } else {
1389 z_brnz(done);
1390 }
1391 break;
1392 }
1393 // Since null is rare, make it be the branch-taken case.
1394 z_brz(found_null);
1395
1396 // Put all the "Case 3" tests here.
1397 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
1398
1399 // Found a null. Keep searching for a matching receiver,
1400 // but remember that this is an empty (unused) slot.
1401 bind(found_null);
1402 }
1403 }
1404
1405 // In the fall-through case, we found no matching receiver, but we
1406 // observed the receiver[start_row] is NULL.
1407
1408 // Fill in the receiver field and increment the count.
1409 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1410 set_mdp_data_at(mdp, recvr_offset, receiver);
1411 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1412 load_const_optimized(reg2, DataLayout::counter_increment);
1413 set_mdp_data_at(mdp, count_offset, reg2);
1414 if (start_row > 0) {
1415 z_bru(done);
1416 }
1417 }
1418
1419 // Example state machine code for three profile rows:
1420 // // main copy of decision tree, rooted at row[1]
1421 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1422 // if (row[0].rec != NULL) {
1423 // // inner copy of decision tree, rooted at row[1]
1424 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1425 // if (row[1].rec != NULL) {
1426 // // degenerate decision tree, rooted at row[2]
1427 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1428 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1429 // row[2].init(rec); goto done;
1430 // } else {
1431 // // remember row[1] is empty
1432 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1433 // row[1].init(rec); goto done;
1434 // }
1435 // } else {
1436 // // remember row[0] is empty
1437 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1438 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1439 // row[0].init(rec); goto done;
1440 // }
1441 // done:
1442
1443 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1444 Register mdp, Register reg2,
1445 bool is_virtual_call) {
1446 assert(ProfileInterpreter, "must be profiling");
1447 Label done;
1448
1449 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1450
1451 bind (done);
1452 }
1453
1454 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
1455 if (ProfileInterpreter) {
1456 NearLabel profile_continue;
1457 uint row;
1458
1459 // If no method data exists, go to profile_continue.
1460 test_method_data_pointer(mdp, profile_continue);
1461
1462 // Update the total ret count.
1463 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1464
1465 for (row = 0; row < RetData::row_limit(); row++) {
1466 NearLabel next_test;
1467
1468 // See if return_bci is equal to bci[n]:
1469 test_mdp_data_at(mdp,
1470 in_bytes(RetData::bci_offset(row)),
1471 return_bci, noreg,
1472 next_test);
1473
1474 // Return_bci is equal to bci[n]. Increment the count.
1475 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1476
1477 // The method data pointer needs to be updated to reflect the new target.
1478 update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row)));
1479 z_bru(profile_continue);
1480 bind(next_test);
1481 }
1482
1483 update_mdp_for_ret(return_bci);
1484
1485 bind(profile_continue);
1486 }
1487 }
1488
1489 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1490 if (ProfileInterpreter) {
1491 Label profile_continue;
1492
1493 // If no method data exists, go to profile_continue.
1494 test_method_data_pointer(mdp, profile_continue);
1495
1496 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1497
1498 // The method data pointer needs to be updated.
1499 int mdp_delta = in_bytes(BitData::bit_data_size());
1500 if (TypeProfileCasts) {
1501 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1502 }
1503 update_mdp_by_constant(mdp, mdp_delta);
1504
1505 bind(profile_continue);
1506 }
1507 }
1508
1509 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp, Register tmp) {
1510 if (ProfileInterpreter && TypeProfileCasts) {
1511 Label profile_continue;
1512
1513 // If no method data exists, go to profile_continue.
1514 test_method_data_pointer(mdp, profile_continue);
1515
1516 int count_offset = in_bytes(CounterData::count_offset());
1517 // Back up the address, since we have already bumped the mdp.
1518 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1519
1520 // *Decrement* the counter. We expect to see zero or small negatives.
1521 increment_mdp_data_at(mdp, count_offset, tmp, true);
1522
1523 bind (profile_continue);
1524 }
1525 }
1526
1527 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1528 if (ProfileInterpreter) {
1529 Label profile_continue;
1530
1531 // If no method data exists, go to profile_continue.
1532 test_method_data_pointer(mdp, profile_continue);
1533
1534 // The method data pointer needs to be updated.
1535 int mdp_delta = in_bytes(BitData::bit_data_size());
1536 if (TypeProfileCasts) {
1537 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1538
1539 // Record the object type.
1540 record_klass_in_profile(klass, mdp, reg2, false);
1541 }
1542 update_mdp_by_constant(mdp, mdp_delta);
1543
1544 bind(profile_continue);
1545 }
1546 }
1547
1548 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1549 if (ProfileInterpreter) {
1550 Label profile_continue;
1551
1552 // If no method data exists, go to profile_continue.
1553 test_method_data_pointer(mdp, profile_continue);
1554
1555 // Update the default case count.
1556 increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset()));
1557
1558 // The method data pointer needs to be updated.
1559 update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset()));
1560
1561 bind(profile_continue);
1562 }
1563 }
1564
1565 // Kills: index, scratch1, scratch2.
1566 void InterpreterMacroAssembler::profile_switch_case(Register index,
1567 Register mdp,
1568 Register scratch1,
1569 Register scratch2) {
1570 if (ProfileInterpreter) {
1571 Label profile_continue;
1572 assert_different_registers(index, mdp, scratch1, scratch2);
1573
1574 // If no method data exists, go to profile_continue.
1575 test_method_data_pointer(mdp, profile_continue);
1576
1577 // Build the base (index * per_case_size_in_bytes()) +
1578 // case_array_offset_in_bytes().
1579 z_sllg(index, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1580 add2reg(index, in_bytes(MultiBranchData::case_array_offset()));
1581
1582 // Add the calculated base to the mdp -> address of the case' data.
1583 Address case_data_addr(mdp, index);
1584 Register case_data = scratch1;
1585 load_address(case_data, case_data_addr);
1586
1587 // Update the case count.
1588 increment_mdp_data_at(case_data,
1589 in_bytes(MultiBranchData::relative_count_offset()),
1590 scratch2);
1591
1592 // The method data pointer needs to be updated.
1593 update_mdp_by_offset(mdp,
1594 index,
1595 in_bytes(MultiBranchData::relative_displacement_offset()));
1596
1597 bind(profile_continue);
1598 }
1599 }
1600
1601 // kills: R0, R1, flags, loads klass from obj (if not null)
1602 void InterpreterMacroAssembler::profile_obj_type(Register obj, Address mdo_addr, Register klass, bool cmp_done) {
1603 NearLabel null_seen, init_klass, do_nothing, do_update;
1604
1605 // Klass = obj is allowed.
1606 const Register tmp = Z_R1;
1607 assert_different_registers(obj, mdo_addr.base(), tmp, Z_R0);
1608 assert_different_registers(klass, mdo_addr.base(), tmp, Z_R0);
1609
1610 z_lg(tmp, mdo_addr);
1611 if (cmp_done) {
1612 z_brz(null_seen);
1613 } else {
1614 compareU64_and_branch(obj, (intptr_t)0, Assembler::bcondEqual, null_seen);
1615 }
1616
1617 verify_oop(obj);
1618 load_klass(klass, obj);
1619
1620 // Klass seen before, nothing to do (regardless of unknown bit).
1621 z_lgr(Z_R0, tmp);
1622 assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
1623 z_nill(Z_R0, TypeEntries::type_klass_mask & 0xFFFF);
1624 compareU64_and_branch(Z_R0, klass, Assembler::bcondEqual, do_nothing);
1625
1626 // Already unknown. Nothing to do anymore.
1627 z_tmll(tmp, TypeEntries::type_unknown);
1628 z_brc(Assembler::bcondAllOne, do_nothing);
1629
1630 z_lgr(Z_R0, tmp);
1631 assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
1632 z_nill(Z_R0, TypeEntries::type_mask & 0xFFFF);
1633 compareU64_and_branch(Z_R0, (intptr_t)0, Assembler::bcondEqual, init_klass);
1634
1635 // Different than before. Cannot keep accurate profile.
1636 z_oill(tmp, TypeEntries::type_unknown);
1637 z_bru(do_update);
1638
1639 bind(init_klass);
1640 // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1641 z_ogr(tmp, klass);
1642 z_bru(do_update);
1643
1644 bind(null_seen);
1645 // Set null_seen if obj is 0.
1646 z_oill(tmp, TypeEntries::null_seen);
1647 // fallthru: z_bru(do_update);
1648
1649 bind(do_update);
1650 z_stg(tmp, mdo_addr);
1651
1652 bind(do_nothing);
1653 }
1654
1655 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1656 if (!ProfileInterpreter) {
1657 return;
1658 }
1659
1660 assert_different_registers(mdp, callee, tmp);
1661
1662 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1663 Label profile_continue;
1664
1665 test_method_data_pointer(mdp, profile_continue);
1666
1667 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1668
1669 z_cliy(in_bytes(DataLayout::tag_offset()) - off_to_start, mdp,
1670 is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1671 z_brne(profile_continue);
1672
1673 if (MethodData::profile_arguments()) {
1674 NearLabel done;
1675 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1676 add2reg(mdp, off_to_args);
1677
1678 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1679 if (i > 0 || MethodData::profile_return()) {
1680 // If return value type is profiled we may have no argument to profile.
1681 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1682 add2reg(tmp, -i*TypeStackSlotEntries::per_arg_count());
1683 compare64_and_branch(tmp, TypeStackSlotEntries::per_arg_count(), Assembler::bcondLow, done);
1684 }
1685 z_lg(tmp, Address(callee, Method::const_offset()));
1686 z_lgh(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1687 // Stack offset o (zero based) from the start of the argument
1688 // list. For n arguments translates into offset n - o - 1 from
1689 // the end of the argument list. But there is an extra slot at
1690 // the top of the stack. So the offset is n - o from Lesp.
1691 z_sg(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
1692 z_sllg(tmp, tmp, Interpreter::logStackElementSize);
1693 Address stack_slot_addr(tmp, Z_esp);
1694 z_ltg(tmp, stack_slot_addr);
1695
1696 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
1697 profile_obj_type(tmp, mdo_arg_addr, tmp, /*ltg did compare to 0*/ true);
1698
1699 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1700 add2reg(mdp, to_add);
1701 off_to_args += to_add;
1702 }
1703
1704 if (MethodData::profile_return()) {
1705 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1706 add2reg(tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1707 }
1708
1709 bind(done);
1710
1711 if (MethodData::profile_return()) {
1712 // We're right after the type profile for the last
1713 // argument. Tmp is the number of cells left in the
1714 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1715 // if there's a return to profile.
1716 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1717 z_sllg(tmp, tmp, exact_log2(DataLayout::cell_size));
1718 z_agr(mdp, tmp);
1719 }
1720 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
1721 } else {
1722 assert(MethodData::profile_return(), "either profile call args or call ret");
1723 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1724 }
1725
1726 // Mdp points right after the end of the
1727 // CallTypeData/VirtualCallTypeData, right after the cells for the
1728 // return value type if there's one.
1729 bind(profile_continue);
1730 }
1731 }
1732
1733 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1734 assert_different_registers(mdp, ret, tmp);
1735 if (ProfileInterpreter && MethodData::profile_return()) {
1736 Label profile_continue;
1737
1738 test_method_data_pointer(mdp, profile_continue);
1739
1740 if (MethodData::profile_return_jsr292_only()) {
1741 // If we don't profile all invoke bytecodes we must make sure
1742 // it's a bytecode we indeed profile. We can't go back to the
1743 // beginning of the ProfileData we intend to update to check its
1744 // type because we're right after it and we don't known its
1745 // length.
1746 NearLabel do_profile;
1747 Address bc(Z_bcp);
1748 z_lb(tmp, bc);
1749 compare32_and_branch(tmp, Bytecodes::_invokedynamic, Assembler::bcondEqual, do_profile);
1750 compare32_and_branch(tmp, Bytecodes::_invokehandle, Assembler::bcondEqual, do_profile);
1751 get_method(tmp);
1752 // Supplement to 8139891: _intrinsic_id exceeded 1-byte size limit.
1753 if (Method::intrinsic_id_size_in_bytes() == 1) {
1754 z_cli(Method::intrinsic_id_offset_in_bytes(), tmp, vmIntrinsics::_compiledLambdaForm);
1755 } else {
1756 assert(Method::intrinsic_id_size_in_bytes() == 2, "size error: check Method::_intrinsic_id");
1757 z_lh(tmp, Method::intrinsic_id_offset_in_bytes(), Z_R0, tmp);
1758 z_chi(tmp, vmIntrinsics::_compiledLambdaForm);
1759 }
1760 z_brne(profile_continue);
1761
1762 bind(do_profile);
1763 }
1764
1765 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1766 profile_obj_type(ret, mdo_ret_addr, tmp);
1767
1768 bind(profile_continue);
1769 }
1770 }
1771
1772 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1773 if (ProfileInterpreter && MethodData::profile_parameters()) {
1774 Label profile_continue, done;
1775
1776 test_method_data_pointer(mdp, profile_continue);
1777
1778 // Load the offset of the area within the MDO used for
1779 // parameters. If it's negative we're not profiling any parameters.
1780 Address parm_di_addr(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()));
1781 load_and_test_int2long(tmp1, parm_di_addr);
1782 z_brl(profile_continue);
1783
1784 // Compute a pointer to the area for parameters from the offset
1785 // and move the pointer to the slot for the last
1786 // parameters. Collect profiling from last parameter down.
1787 // mdo start + parameters offset + array length - 1
1788
1789 // Pointer to the parameter area in the MDO.
1790 z_agr(mdp, tmp1);
1791
1792 // Offset of the current profile entry to update.
1793 const Register entry_offset = tmp1;
1794 // entry_offset = array len in number of cells.
1795 z_lg(entry_offset, Address(mdp, ArrayData::array_len_offset()));
1796 // entry_offset (number of cells) = array len - size of 1 entry
1797 add2reg(entry_offset, -TypeStackSlotEntries::per_arg_count());
1798 // entry_offset in bytes
1799 z_sllg(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1800
1801 Label loop;
1802 bind(loop);
1803
1804 Address arg_off(mdp, entry_offset, ParametersTypeData::stack_slot_offset(0));
1805 Address arg_type(mdp, entry_offset, ParametersTypeData::type_offset(0));
1806
1807 // Load offset on the stack from the slot for this parameter.
1808 z_lg(tmp2, arg_off);
1809 z_sllg(tmp2, tmp2, Interpreter::logStackElementSize);
1810 z_lcgr(tmp2); // Negate.
1811
1812 // Profile the parameter.
1813 z_ltg(tmp2, Address(Z_locals, tmp2));
1814 profile_obj_type(tmp2, arg_type, tmp2, /*ltg did compare to 0*/ true);
1815
1816 // Go to next parameter.
1817 z_aghi(entry_offset, -TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size);
1818 z_brnl(loop);
1819
1820 bind(profile_continue);
1821 }
1822 }
1823
1824 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1825 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1826 int increment,
1827 Address mask,
1828 Register scratch,
1829 bool preloaded,
1830 branch_condition cond,
1831 Label *where) {
1832 assert_different_registers(counter_addr.base(), scratch);
1833 if (preloaded) {
1834 add2reg(scratch, increment);
1835 reg2mem_opt(scratch, counter_addr, false);
1836 } else {
1837 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment) && counter_addr.is_RSYform()) {
1838 z_alsi(counter_addr.disp20(), counter_addr.base(), increment);
1839 mem2reg_signed_opt(scratch, counter_addr);
1840 } else {
1841 mem2reg_signed_opt(scratch, counter_addr);
1842 add2reg(scratch, increment);
1843 reg2mem_opt(scratch, counter_addr, false);
1844 }
1845 }
1846 z_n(scratch, mask);
1847 if (where) { z_brc(cond, *where); }
1848 }
1849
1850 // Get MethodCounters object for given method. Lazily allocated if necessary.
1851 // method - Ptr to Method object.
1852 // Rcounters - Ptr to MethodCounters object associated with Method object.
1853 // skip - Exit point if MethodCounters object can't be created (OOM condition).
1854 void InterpreterMacroAssembler::get_method_counters(Register Rmethod,
1855 Register Rcounters,
1856 Label& skip) {
1857 assert_different_registers(Rmethod, Rcounters);
1858
1859 BLOCK_COMMENT("get MethodCounters object {");
1860
1861 Label has_counters;
1862 load_and_test_long(Rcounters, Address(Rmethod, Method::method_counters_offset()));
1863 z_brnz(has_counters);
1864
1865 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), Rmethod, false);
1866 z_ltgr(Rcounters, Z_RET); // Runtime call returns MethodCounters object.
1867 z_brz(skip); // No MethodCounters, out of memory.
1868
1869 bind(has_counters);
1870
1871 BLOCK_COMMENT("} get MethodCounters object");
1872 }
1873
1874 // Increment invocation counter in MethodCounters object.
1875 // Return (invocation_counter+backedge_counter) as "result" in RctrSum.
1876 // Counter values are all unsigned.
1877 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register RctrSum) {
1878 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful");
1879 assert_different_registers(Rcounters, RctrSum);
1880
1881 int increment = InvocationCounter::count_increment;
1882 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
1883 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset());
1884
1885 BLOCK_COMMENT("Increment invocation counter {");
1886
1887 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
1888 // Increment the invocation counter in place,
1889 // then add the incremented value to the backedge counter.
1890 z_l(RctrSum, be_counter_offset, Rcounters);
1891 z_alsi(inv_counter_offset, Rcounters, increment); // Atomic increment @no extra cost!
1892 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
1893 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
1894 } else {
1895 // This path is optimized for low register consumption
1896 // at the cost of somewhat higher operand delays.
1897 // It does not need an extra temp register.
1898
1899 // Update the invocation counter.
1900 z_l(RctrSum, inv_counter_offset, Rcounters);
1901 if (RctrSum == Z_R0) {
1902 z_ahi(RctrSum, increment);
1903 } else {
1904 add2reg(RctrSum, increment);
1905 }
1906 z_st(RctrSum, inv_counter_offset, Rcounters);
1907
1908 // Mask off the state bits.
1909 z_nilf(RctrSum, InvocationCounter::count_mask_value);
1910
1911 // Add the backedge counter to the updated invocation counter to
1912 // form the result.
1913 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
1914 }
1915
1916 BLOCK_COMMENT("} Increment invocation counter");
1917
1918 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
1919 }
1920
1921
1922 // increment backedge counter in MethodCounters object.
1923 // return (invocation_counter+backedge_counter) as "result" in RctrSum
1924 // counter values are all unsigned!
1925 void InterpreterMacroAssembler::increment_backedge_counter(Register Rcounters, Register RctrSum) {
1926 assert(UseCompiler, "incrementing must be useful");
1927 assert_different_registers(Rcounters, RctrSum);
1928
1929 int increment = InvocationCounter::count_increment;
1930 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
1931 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset());
1932
1933 BLOCK_COMMENT("Increment backedge counter {");
1934
1935 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
1936 // Increment the invocation counter in place,
1937 // then add the incremented value to the backedge counter.
1938 z_l(RctrSum, inv_counter_offset, Rcounters);
1939 z_alsi(be_counter_offset, Rcounters, increment); // Atomic increment @no extra cost!
1940 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
1941 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
1942 } else {
1943 // This path is optimized for low register consumption
1944 // at the cost of somewhat higher operand delays.
1945 // It does not need an extra temp register.
1946
1947 // Update the invocation counter.
1948 z_l(RctrSum, be_counter_offset, Rcounters);
1949 if (RctrSum == Z_R0) {
1950 z_ahi(RctrSum, increment);
1951 } else {
1952 add2reg(RctrSum, increment);
1953 }
1954 z_st(RctrSum, be_counter_offset, Rcounters);
1955
1956 // Mask off the state bits.
1957 z_nilf(RctrSum, InvocationCounter::count_mask_value);
1958
1959 // Add the backedge counter to the updated invocation counter to
1960 // form the result.
1961 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
1962 }
1963
1964 BLOCK_COMMENT("} Increment backedge counter");
1965
1966 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
1967 }
1968
1969 // Add an InterpMonitorElem to stack (see frame_s390.hpp).
1970 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty,
1971 Register Rtemp1,
1972 Register Rtemp2,
1973 Register Rtemp3) {
1974
1975 const Register Rcurr_slot = Rtemp1;
1976 const Register Rlimit = Rtemp2;
1977 const jint delta = -frame::interpreter_frame_monitor_size() * wordSize;
1978
1979 assert((delta & LongAlignmentMask) == 0,
1980 "sizeof BasicObjectLock must be even number of doublewords");
1981 assert(2 * wordSize == -delta, "this works only as long as delta == -2*wordSize");
1982 assert(Rcurr_slot != Z_R0, "Register must be usable as base register");
1983 assert_different_registers(Rlimit, Rcurr_slot, Rtemp3);
1984
1985 get_monitors(Rlimit);
1986
1987 // Adjust stack pointer for additional monitor entry.
1988 resize_frame(RegisterOrConstant((intptr_t) delta), Z_fp, false);
1989
1990 if (!stack_is_empty) {
1991 // Must copy stack contents down.
1992 NearLabel next, done;
1993
1994 // Rtemp := addr(Tos), Z_esp is pointing below it!
1995 add2reg(Rcurr_slot, wordSize, Z_esp);
1996
1997 // Nothing to do, if already at monitor area.
1998 compareU64_and_branch(Rcurr_slot, Rlimit, bcondNotLow, done);
1999
2000 bind(next);
2001
2002 // Move one stack slot.
2003 mem2reg_opt(Rtemp3, Address(Rcurr_slot));
2004 reg2mem_opt(Rtemp3, Address(Rcurr_slot, delta));
2005 add2reg(Rcurr_slot, wordSize);
2006 compareU64_and_branch(Rcurr_slot, Rlimit, bcondLow, next); // Are we done?
2007
2008 bind(done);
2009 // Done copying stack.
2010 }
2011
2012 // Adjust expression stack and monitor pointers.
2013 add2reg(Z_esp, delta);
2014 add2reg(Rlimit, delta);
2015 save_monitors(Rlimit);
2016 }
2017
2018 // Note: Index holds the offset in bytes afterwards.
2019 // You can use this to store a new value (with Llocals as the base).
2020 void InterpreterMacroAssembler::access_local_int(Register index, Register dst) {
2021 z_sllg(index, index, LogBytesPerWord);
2022 mem2reg_opt(dst, Address(Z_locals, index), false);
2023 }
2024
2025 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2026 if (state == atos) { MacroAssembler::verify_oop(reg); }
2027 }
2028
2029 // Inline assembly for:
2030 //
2031 // if (thread is in interp_only_mode) {
2032 // InterpreterRuntime::post_method_entry();
2033 // }
2034
2035 void InterpreterMacroAssembler::notify_method_entry() {
2036
2037 // JVMTI
2038 // Whenever JVMTI puts a thread in interp_only_mode, method
2039 // entry/exit events are sent for that thread to track stack
2040 // depth. If it is possible to enter interp_only_mode we add
2041 // the code to check if the event should be sent.
2042 if (JvmtiExport::can_post_interpreter_events()) {
2043 Label jvmti_post_done;
2044 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2045 z_bre(jvmti_post_done);
2046 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), /*check_exceptions=*/false);
2047 bind(jvmti_post_done);
2048 }
2049 }
2050
2051 // Inline assembly for:
2052 //
2053 // if (thread is in interp_only_mode) {
2054 // if (!native_method) save result
2055 // InterpreterRuntime::post_method_exit();
2056 // if (!native_method) restore result
2057 // }
2058 // if (DTraceMethodProbes) {
2059 // SharedRuntime::dtrace_method_exit(thread, method);
2060 // }
2061 //
2062 // For native methods their result is stored in z_ijava_state.lresult
2063 // and z_ijava_state.fresult before coming here.
2064 // Java methods have their result stored in the expression stack.
2065 //
2066 // Notice the dependency to frame::interpreter_frame_result().
2067 void InterpreterMacroAssembler::notify_method_exit(bool native_method,
2068 TosState state,
2069 NotifyMethodExitMode mode) {
2070 // JVMTI
2071 // Whenever JVMTI puts a thread in interp_only_mode, method
2072 // entry/exit events are sent for that thread to track stack
2073 // depth. If it is possible to enter interp_only_mode we add
2074 // the code to check if the event should be sent.
2075 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2076 Label jvmti_post_done;
2077 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2078 z_bre(jvmti_post_done);
2079 if (!native_method) push(state); // see frame::interpreter_frame_result()
2080 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), /*check_exceptions=*/false);
2081 if (!native_method) pop(state);
2082 bind(jvmti_post_done);
2083 }
2084
2085 #if 0
2086 // Dtrace currently not supported on z/Architecture.
2087 {
2088 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2089 push(state);
2090 get_method(c_rarg1);
2091 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2092 r15_thread, c_rarg1);
2093 pop(state);
2094 }
2095 #endif
2096 }
2097
2098 void InterpreterMacroAssembler::skip_if_jvmti_mode(Label &Lskip, Register Rscratch) {
2099 if (!JvmtiExport::can_post_interpreter_events()) {
2100 return;
2101 }
2102
2103 load_and_test_int(Rscratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2104 z_brnz(Lskip);
2105
2106 }
2107
2108 // Pop the topmost TOP_IJAVA_FRAME and set it's sender_sp as new Z_SP.
2109 // The return pc is loaded into the register return_pc.
2110 //
2111 // Registers updated:
2112 // return_pc - The return pc of the calling frame.
2113 // tmp1, tmp2 - scratch
2114 void InterpreterMacroAssembler::pop_interpreter_frame(Register return_pc, Register tmp1, Register tmp2) {
2115 // F0 Z_SP -> caller_sp (F1's)
2116 // ...
2117 // sender_sp (F1's)
2118 // ...
2119 // F1 Z_fp -> caller_sp (F2's)
2120 // return_pc (Continuation after return from F0.)
2121 // ...
2122 // F2 caller_sp
2123
2124 // Remove F0's activation. Restoring Z_SP to sender_sp reverts modifications
2125 // (a) by a c2i adapter and (b) by generate_fixed_frame().
2126 // In case (a) the new top frame F1 is an unextended compiled frame.
2127 // In case (b) F1 is converted from PARENT_IJAVA_FRAME to TOP_IJAVA_FRAME.
2128
2129 // Case (b) seems to be redundant when returning to a interpreted caller,
2130 // because then the caller's top_frame_sp is installed as sp (see
2131 // TemplateInterpreterGenerator::generate_return_entry_for ()). But
2132 // pop_interpreter_frame() is also used in exception handling and there the
2133 // frame type of the caller is unknown, therefore top_frame_sp cannot be used,
2134 // so it is important that sender_sp is the caller's sp as TOP_IJAVA_FRAME.
2135
2136 Register R_f1_sender_sp = tmp1;
2137 Register R_f2_sp = tmp2;
2138
2139 // Tirst check the for the interpreter frame's magic.
2140 asm_assert_ijava_state_magic(R_f2_sp/*tmp*/);
2141 z_lg(R_f2_sp, _z_parent_ijava_frame_abi(callers_sp), Z_fp);
2142 z_lg(R_f1_sender_sp, _z_ijava_state_neg(sender_sp), Z_fp);
2143 if (return_pc->is_valid())
2144 z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2145 // Pop F0 by resizing to R_f1_sender_sp and using R_f2_sp as fp.
2146 resize_frame_absolute(R_f1_sender_sp, R_f2_sp, false/*load fp*/);
2147
2148 #ifdef ASSERT
2149 // The return_pc in the new top frame is dead... at least that's my
2150 // current understanding; to assert this I overwrite it.
2151 load_const_optimized(Z_ARG3, 0xb00b1);
2152 z_stg(Z_ARG3, _z_parent_ijava_frame_abi(return_pc), Z_SP);
2153 #endif
2154 }
2155
2156 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2157 if (VerifyFPU) {
2158 unimplemented("verfiyFPU");
2159 }
2160 }
2161