1 /*
2 * Copyright (c) 2016, 2018, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2016, 2018 SAP SE. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 // Major contributions by AHa, AS, JL, ML.
27
28 #include "precompiled.hpp"
29 #include "asm/macroAssembler.inline.hpp"
30 #include "gc/shared/barrierSet.hpp"
31 #include "gc/shared/barrierSetAssembler.hpp"
32 #include "interp_masm_s390.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "interpreter/interpreterRuntime.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/markOop.hpp"
37 #include "prims/jvmtiExport.hpp"
38 #include "prims/jvmtiThreadState.hpp"
39 #include "runtime/basicLock.hpp"
40 #include "runtime/biasedLocking.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/safepointMechanism.hpp"
43 #include "runtime/sharedRuntime.hpp"
44 #include "runtime/thread.inline.hpp"
45
46 // Implementation of InterpreterMacroAssembler.
47 // This file specializes the assembler with interpreter-specific macros.
48
49 #ifdef PRODUCT
50 #define BLOCK_COMMENT(str)
51 #define BIND(label) bind(label);
52 #else
53 #define BLOCK_COMMENT(str) block_comment(str)
54 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
55 #endif
56
jump_to_entry(address entry,Register Rscratch)57 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) {
58 assert(entry != NULL, "Entry must have been generated by now");
59 assert(Rscratch != Z_R0, "Can't use R0 for addressing");
60 branch_optimized(Assembler::bcondAlways, entry);
61 }
62
empty_expression_stack(void)63 void InterpreterMacroAssembler::empty_expression_stack(void) {
64 get_monitors(Z_R1_scratch);
65 add2reg(Z_esp, -Interpreter::stackElementSize, Z_R1_scratch);
66 }
67
68 // Dispatch code executed in the prolog of a bytecode which does not do it's
69 // own dispatch.
dispatch_prolog(TosState state,int bcp_incr)70 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
71 // On z/Architecture we are short on registers, therefore we do not preload the
72 // dispatch address of the next bytecode.
73 }
74
75 // Dispatch code executed in the epilog of a bytecode which does not do it's
76 // own dispatch.
dispatch_epilog(TosState state,int step)77 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
78 dispatch_next(state, step);
79 }
80
dispatch_next(TosState state,int bcp_incr,bool generate_poll)81 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) {
82 z_llgc(Z_bytecode, bcp_incr, Z_R0, Z_bcp); // Load next bytecode.
83 add2reg(Z_bcp, bcp_incr); // Advance bcp. Add2reg produces optimal code.
84 dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
85 }
86
87 // Common code to dispatch and dispatch_only.
88 // Dispatch value in Lbyte_code and increment Lbcp.
89
dispatch_base(TosState state,address * table,bool generate_poll)90 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table, bool generate_poll) {
91 verify_FPU(1, state);
92
93 #ifdef ASSERT
94 address reentry = NULL;
95 { Label OK;
96 // Check if the frame pointer in Z_fp is correct.
97 z_cg(Z_fp, 0, Z_SP);
98 z_bre(OK);
99 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp: " FILE_AND_LINE);
100 bind(OK);
101 }
102 { Label OK;
103 // check if the locals pointer in Z_locals is correct
104 z_cg(Z_locals, _z_ijava_state_neg(locals), Z_fp);
105 z_bre(OK);
106 reentry = stop_chain_static(reentry, "invalid locals pointer Z_locals: " FILE_AND_LINE);
107 bind(OK);
108 }
109 #endif
110
111 // TODO: Maybe implement +VerifyActivationFrameSize here.
112 // verify_thread(); // Too slow. We will just verify on method entry & exit.
113 verify_oop(Z_tos, state);
114
115 // Dispatch table to use.
116 load_absolute_address(Z_tmp_1, (address)table); // Z_tmp_1 = table;
117
118 if (SafepointMechanism::uses_thread_local_poll() && generate_poll) {
119 address *sfpt_tbl = Interpreter::safept_table(state);
120 if (table != sfpt_tbl) {
121 Label dispatch;
122 const Address poll_byte_addr(Z_thread, in_bytes(Thread::polling_page_offset()) + 7 /* Big Endian */);
123 // Armed page has poll_bit set, if poll bit is cleared just continue.
124 z_tm(poll_byte_addr, SafepointMechanism::poll_bit());
125 z_braz(dispatch);
126 load_absolute_address(Z_tmp_1, (address)sfpt_tbl); // Z_tmp_1 = table;
127 bind(dispatch);
128 }
129 }
130
131 // 0 <= Z_bytecode < 256 => Use a 32 bit shift, because it is shorter than sllg.
132 // Z_bytecode must have been loaded zero-extended for this approach to be correct.
133 z_sll(Z_bytecode, LogBytesPerWord, Z_R0); // Multiply by wordSize.
134 z_lg(Z_tmp_1, 0, Z_bytecode, Z_tmp_1); // Get entry addr.
135
136 z_br(Z_tmp_1);
137 }
138
dispatch_only(TosState state,bool generate_poll)139 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
140 dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
141 }
142
dispatch_only_normal(TosState state)143 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
144 dispatch_base(state, Interpreter::normal_table(state));
145 }
146
dispatch_via(TosState state,address * table)147 void InterpreterMacroAssembler::dispatch_via(TosState state, address *table) {
148 // Load current bytecode.
149 z_llgc(Z_bytecode, Address(Z_bcp, (intptr_t)0));
150 dispatch_base(state, table);
151 }
152
153 // The following call_VM*_base() methods overload and mask the respective
154 // declarations/definitions in class MacroAssembler. They are meant as a "detour"
155 // to perform additional, template interpreter specific tasks before actually
156 // calling their MacroAssembler counterparts.
157
call_VM_leaf_base(address entry_point)158 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point) {
159 bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated.
160 // interpreter specific
161 // Note: No need to save/restore bcp (Z_R13) pointer since these are callee
162 // saved registers and no blocking/ GC can happen in leaf calls.
163
164 // super call
165 MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation);
166 }
167
call_VM_leaf_base(address entry_point,bool allow_relocation)168 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, bool allow_relocation) {
169 // interpreter specific
170 // Note: No need to save/restore bcp (Z_R13) pointer since these are callee
171 // saved registers and no blocking/ GC can happen in leaf calls.
172
173 // super call
174 MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation);
175 }
176
call_VM_base(Register oop_result,Register last_java_sp,address entry_point,bool check_exceptions)177 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp,
178 address entry_point, bool check_exceptions) {
179 bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated.
180 // interpreter specific
181
182 save_bcp();
183 save_esp();
184 // super call
185 MacroAssembler::call_VM_base(oop_result, last_java_sp,
186 entry_point, allow_relocation, check_exceptions);
187 restore_bcp();
188 }
189
call_VM_base(Register oop_result,Register last_java_sp,address entry_point,bool allow_relocation,bool check_exceptions)190 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp,
191 address entry_point, bool allow_relocation,
192 bool check_exceptions) {
193 // interpreter specific
194
195 save_bcp();
196 save_esp();
197 // super call
198 MacroAssembler::call_VM_base(oop_result, last_java_sp,
199 entry_point, allow_relocation, check_exceptions);
200 restore_bcp();
201 }
202
check_and_handle_popframe(Register scratch_reg)203 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
204 if (JvmtiExport::can_pop_frame()) {
205 BLOCK_COMMENT("check_and_handle_popframe {");
206 Label L;
207 // Initiate popframe handling only if it is not already being
208 // processed. If the flag has the popframe_processing bit set, it
209 // means that this code is called *during* popframe handling - we
210 // don't want to reenter.
211 // TODO: Check if all four state combinations could be visible.
212 // If (processing and !pending) is an invisible/impossible state,
213 // there is optimization potential by testing both bits at once.
214 // Then, All_Zeroes and All_Ones means skip, Mixed means doit.
215 testbit(Address(Z_thread, JavaThread::popframe_condition_offset()),
216 exact_log2(JavaThread::popframe_pending_bit));
217 z_bfalse(L);
218 testbit(Address(Z_thread, JavaThread::popframe_condition_offset()),
219 exact_log2(JavaThread::popframe_processing_bit));
220 z_btrue(L);
221
222 // Call Interpreter::remove_activation_preserving_args_entry() to get the
223 // address of the same-named entrypoint in the generated interpreter code.
224 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
225 // The above call should (as its only effect) return the contents of the field
226 // _remove_activation_preserving_args_entry in Z_RET.
227 // We just jump there to have the work done.
228 z_br(Z_RET);
229 // There is no way for control to fall thru here.
230
231 bind(L);
232 BLOCK_COMMENT("} check_and_handle_popframe");
233 }
234 }
235
236
load_earlyret_value(TosState state)237 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
238 Register RjvmtiState = Z_R1_scratch;
239 int tos_off = in_bytes(JvmtiThreadState::earlyret_tos_offset());
240 int oop_off = in_bytes(JvmtiThreadState::earlyret_oop_offset());
241 int val_off = in_bytes(JvmtiThreadState::earlyret_value_offset());
242 int state_off = in_bytes(JavaThread::jvmti_thread_state_offset());
243
244 z_lg(RjvmtiState, state_off, Z_thread);
245
246 switch (state) {
247 case atos: z_lg(Z_tos, oop_off, RjvmtiState);
248 store_const(Address(RjvmtiState, oop_off), 0L, 8, 8, Z_R0_scratch);
249 break;
250 case ltos: z_lg(Z_tos, val_off, RjvmtiState); break;
251 case btos: // fall through
252 case ztos: // fall through
253 case ctos: // fall through
254 case stos: // fall through
255 case itos: z_llgf(Z_tos, val_off, RjvmtiState); break;
256 case ftos: z_le(Z_ftos, val_off, RjvmtiState); break;
257 case dtos: z_ld(Z_ftos, val_off, RjvmtiState); break;
258 case vtos: /* nothing to do */ break;
259 default : ShouldNotReachHere();
260 }
261
262 // Clean up tos value in the jvmti thread state.
263 store_const(Address(RjvmtiState, val_off), 0L, 8, 8, Z_R0_scratch);
264 // Set tos state field to illegal value.
265 store_const(Address(RjvmtiState, tos_off), ilgl, 4, 1, Z_R0_scratch);
266 }
267
check_and_handle_earlyret(Register scratch_reg)268 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
269 if (JvmtiExport::can_force_early_return()) {
270 BLOCK_COMMENT("check_and_handle_earlyret {");
271 Label L;
272 // arg regs are save, because we are just behind the call in call_VM_base
273 Register jvmti_thread_state = Z_ARG2;
274 Register tmp = Z_ARG3;
275 load_and_test_long(jvmti_thread_state, Address(Z_thread, JavaThread::jvmti_thread_state_offset()));
276 z_bre(L); // if (thread->jvmti_thread_state() == NULL) exit;
277
278 // Initiate earlyret handling only if it is not already being processed.
279 // If the flag has the earlyret_processing bit set, it means that this code
280 // is called *during* earlyret handling - we don't want to reenter.
281
282 assert((JvmtiThreadState::earlyret_pending != 0) && (JvmtiThreadState::earlyret_inactive == 0),
283 "must fix this check, when changing the values of the earlyret enum");
284 assert(JvmtiThreadState::earlyret_pending == 1, "must fix this check, when changing the values of the earlyret enum");
285
286 load_and_test_int(tmp, Address(jvmti_thread_state, JvmtiThreadState::earlyret_state_offset()));
287 z_brz(L); // if (thread->jvmti_thread_state()->_earlyret_state != JvmtiThreadState::earlyret_pending) exit;
288
289 // Call Interpreter::remove_activation_early_entry() to get the address of the
290 // same-named entrypoint in the generated interpreter code.
291 assert(sizeof(TosState) == 4, "unexpected size");
292 z_l(Z_ARG1, Address(jvmti_thread_state, JvmtiThreadState::earlyret_tos_offset()));
293 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Z_ARG1);
294 // The above call should (as its only effect) return the contents of the field
295 // _remove_activation_preserving_args_entry in Z_RET.
296 // We just jump there to have the work done.
297 z_br(Z_RET);
298 // There is no way for control to fall thru here.
299
300 bind(L);
301 BLOCK_COMMENT("} check_and_handle_earlyret");
302 }
303 }
304
super_call_VM_leaf(address entry_point,Register arg_1,Register arg_2)305 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2) {
306 lgr_if_needed(Z_ARG1, arg_1);
307 assert(arg_2 != Z_ARG1, "smashed argument");
308 lgr_if_needed(Z_ARG2, arg_2);
309 MacroAssembler::call_VM_leaf_base(entry_point, true);
310 }
311
get_cache_index_at_bcp(Register index,int bcp_offset,size_t index_size)312 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, int bcp_offset, size_t index_size) {
313 Address param(Z_bcp, bcp_offset);
314
315 BLOCK_COMMENT("get_cache_index_at_bcp {");
316 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
317 if (index_size == sizeof(u2)) {
318 load_sized_value(index, param, 2, false /*signed*/);
319 } else if (index_size == sizeof(u4)) {
320
321 load_sized_value(index, param, 4, false);
322
323 // Check if the secondary index definition is still ~x, otherwise
324 // we have to change the following assembler code to calculate the
325 // plain index.
326 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
327 not_(index); // Convert to plain index.
328 } else if (index_size == sizeof(u1)) {
329 z_llgc(index, param);
330 } else {
331 ShouldNotReachHere();
332 }
333 BLOCK_COMMENT("}");
334 }
335
336
get_cache_and_index_at_bcp(Register cache,Register cpe_offset,int bcp_offset,size_t index_size)337 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register cpe_offset,
338 int bcp_offset, size_t index_size) {
339 BLOCK_COMMENT("get_cache_and_index_at_bcp {");
340 assert_different_registers(cache, cpe_offset);
341 get_cache_index_at_bcp(cpe_offset, bcp_offset, index_size);
342 z_lg(cache, Address(Z_fp, _z_ijava_state_neg(cpoolCache)));
343 // Convert from field index to ConstantPoolCache offset in bytes.
344 z_sllg(cpe_offset, cpe_offset, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord));
345 BLOCK_COMMENT("}");
346 }
347
348 // Kills Z_R0_scratch.
get_cache_and_index_and_bytecode_at_bcp(Register cache,Register cpe_offset,Register bytecode,int byte_no,int bcp_offset,size_t index_size)349 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
350 Register cpe_offset,
351 Register bytecode,
352 int byte_no,
353 int bcp_offset,
354 size_t index_size) {
355 BLOCK_COMMENT("get_cache_and_index_and_bytecode_at_bcp {");
356 get_cache_and_index_at_bcp(cache, cpe_offset, bcp_offset, index_size);
357
358 // We want to load (from CP cache) the bytecode that corresponds to the passed-in byte_no.
359 // It is located at (cache + cpe_offset + base_offset + indices_offset + (8-1) (last byte in DW) - (byte_no+1).
360 // Instead of loading, shifting and masking a DW, we just load that one byte of interest with z_llgc (unsigned).
361 const int base_ix_off = in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset());
362 const int off_in_DW = (8-1) - (1+byte_no);
363 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
364 assert(ConstantPoolCacheEntry::bytecode_1_mask == 0xff, "");
365 load_sized_value(bytecode, Address(cache, cpe_offset, base_ix_off+off_in_DW), 1, false /*signed*/);
366
367 BLOCK_COMMENT("}");
368 }
369
370 // Load object from cpool->resolved_references(index).
load_resolved_reference_at_index(Register result,Register index)371 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) {
372 assert_different_registers(result, index);
373 get_constant_pool(result);
374
375 // Convert
376 // - from field index to resolved_references() index and
377 // - from word index to byte offset.
378 // Since this is a java object, it is potentially compressed.
379 Register tmp = index; // reuse
380 z_sllg(index, index, LogBytesPerHeapOop); // Offset into resolved references array.
381 // Load pointer for resolved_references[] objArray.
382 z_lg(result, ConstantPool::cache_offset_in_bytes(), result);
383 z_lg(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result);
384 resolve_oop_handle(result); // Load resolved references array itself.
385 #ifdef ASSERT
386 NearLabel index_ok;
387 z_lgf(Z_R0, Address(result, arrayOopDesc::length_offset_in_bytes()));
388 z_sllg(Z_R0, Z_R0, LogBytesPerHeapOop);
389 compare64_and_branch(tmp, Z_R0, Assembler::bcondLow, index_ok);
390 stop("resolved reference index out of bounds", 0x09256);
391 bind(index_ok);
392 #endif
393 z_agr(result, index); // Address of indexed array element.
394 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp, noreg);
395 }
396
397 // load cpool->resolved_klass_at(index)
load_resolved_klass_at_offset(Register cpool,Register offset,Register iklass)398 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register cpool, Register offset, Register iklass) {
399 // int value = *(Rcpool->int_at_addr(which));
400 // int resolved_klass_index = extract_low_short_from_int(value);
401 z_llgh(offset, Address(cpool, offset, sizeof(ConstantPool) + 2)); // offset = resolved_klass_index (s390 is big-endian)
402 z_sllg(offset, offset, LogBytesPerWord); // Convert 'index' to 'offset'
403 z_lg(iklass, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); // iklass = cpool->_resolved_klasses
404 z_lg(iklass, Address(iklass, offset, Array<Klass*>::base_offset_in_bytes()));
405 }
406
get_cache_entry_pointer_at_bcp(Register cache,Register tmp,int bcp_offset,size_t index_size)407 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
408 Register tmp,
409 int bcp_offset,
410 size_t index_size) {
411 BLOCK_COMMENT("get_cache_entry_pointer_at_bcp {");
412 get_cache_and_index_at_bcp(cache, tmp, bcp_offset, index_size);
413 add2reg_with_index(cache, in_bytes(ConstantPoolCache::base_offset()), tmp, cache);
414 BLOCK_COMMENT("}");
415 }
416
417 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
418 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2.
gen_subtype_check(Register Rsub_klass,Register Rsuper_klass,Register Rtmp1,Register Rtmp2,Label & ok_is_subtype)419 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
420 Register Rsuper_klass,
421 Register Rtmp1,
422 Register Rtmp2,
423 Label &ok_is_subtype) {
424 // Profile the not-null value's klass.
425 profile_typecheck(Rtmp1, Rsub_klass, Rtmp2);
426
427 // Do the check.
428 check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
429
430 // Profile the failure of the check.
431 profile_typecheck_failed(Rtmp1, Rtmp2);
432 }
433
434 // Pop topmost element from stack. It just disappears.
435 // Useful if consumed previously by access via stackTop().
popx(int len)436 void InterpreterMacroAssembler::popx(int len) {
437 add2reg(Z_esp, len*Interpreter::stackElementSize);
438 debug_only(verify_esp(Z_esp, Z_R1_scratch));
439 }
440
441 // Get Address object of stack top. No checks. No pop.
442 // Purpose: - Provide address of stack operand to exploit reg-mem operations.
443 // - Avoid RISC-like mem2reg - reg-reg-op sequence.
stackTop()444 Address InterpreterMacroAssembler::stackTop() {
445 return Address(Z_esp, Interpreter::expr_offset_in_bytes(0));
446 }
447
pop_i(Register r)448 void InterpreterMacroAssembler::pop_i(Register r) {
449 z_l(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
pop_ptr(Register r)455 void InterpreterMacroAssembler::pop_ptr(Register r) {
456 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
457 add2reg(Z_esp, Interpreter::stackElementSize);
458 assert_different_registers(r, Z_R1_scratch);
459 debug_only(verify_esp(Z_esp, Z_R1_scratch));
460 }
461
pop_l(Register r)462 void InterpreterMacroAssembler::pop_l(Register r) {
463 z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
464 add2reg(Z_esp, 2*Interpreter::stackElementSize);
465 assert_different_registers(r, Z_R1_scratch);
466 debug_only(verify_esp(Z_esp, Z_R1_scratch));
467 }
468
pop_f(FloatRegister f)469 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
470 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), false);
471 add2reg(Z_esp, Interpreter::stackElementSize);
472 debug_only(verify_esp(Z_esp, Z_R1_scratch));
473 }
474
pop_d(FloatRegister f)475 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
476 mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), true);
477 add2reg(Z_esp, 2*Interpreter::stackElementSize);
478 debug_only(verify_esp(Z_esp, Z_R1_scratch));
479 }
480
push_i(Register r)481 void InterpreterMacroAssembler::push_i(Register r) {
482 assert_different_registers(r, Z_R1_scratch);
483 debug_only(verify_esp(Z_esp, Z_R1_scratch));
484 z_st(r, Address(Z_esp));
485 add2reg(Z_esp, -Interpreter::stackElementSize);
486 }
487
push_ptr(Register r)488 void InterpreterMacroAssembler::push_ptr(Register r) {
489 z_stg(r, Address(Z_esp));
490 add2reg(Z_esp, -Interpreter::stackElementSize);
491 }
492
push_l(Register r)493 void InterpreterMacroAssembler::push_l(Register r) {
494 assert_different_registers(r, Z_R1_scratch);
495 debug_only(verify_esp(Z_esp, Z_R1_scratch));
496 int offset = -Interpreter::stackElementSize;
497 z_stg(r, Address(Z_esp, offset));
498 clear_mem(Address(Z_esp), Interpreter::stackElementSize);
499 add2reg(Z_esp, 2 * offset);
500 }
501
push_f(FloatRegister f)502 void InterpreterMacroAssembler::push_f(FloatRegister f) {
503 debug_only(verify_esp(Z_esp, Z_R1_scratch));
504 freg2mem_opt(f, Address(Z_esp), false);
505 add2reg(Z_esp, -Interpreter::stackElementSize);
506 }
507
push_d(FloatRegister d)508 void InterpreterMacroAssembler::push_d(FloatRegister d) {
509 debug_only(verify_esp(Z_esp, Z_R1_scratch));
510 int offset = -Interpreter::stackElementSize;
511 freg2mem_opt(d, Address(Z_esp, offset));
512 add2reg(Z_esp, 2 * offset);
513 }
514
push(TosState state)515 void InterpreterMacroAssembler::push(TosState state) {
516 verify_oop(Z_tos, state);
517 switch (state) {
518 case atos: push_ptr(); break;
519 case btos: push_i(); break;
520 case ztos:
521 case ctos:
522 case stos: push_i(); break;
523 case itos: push_i(); break;
524 case ltos: push_l(); break;
525 case ftos: push_f(); break;
526 case dtos: push_d(); break;
527 case vtos: /* nothing to do */ break;
528 default : ShouldNotReachHere();
529 }
530 }
531
pop(TosState state)532 void InterpreterMacroAssembler::pop(TosState state) {
533 switch (state) {
534 case atos: pop_ptr(Z_tos); break;
535 case btos: pop_i(Z_tos); break;
536 case ztos:
537 case ctos:
538 case stos: pop_i(Z_tos); break;
539 case itos: pop_i(Z_tos); break;
540 case ltos: pop_l(Z_tos); break;
541 case ftos: pop_f(Z_ftos); break;
542 case dtos: pop_d(Z_ftos); break;
543 case vtos: /* nothing to do */ break;
544 default : ShouldNotReachHere();
545 }
546 verify_oop(Z_tos, state);
547 }
548
549 // Helpers for swap and dup.
load_ptr(int n,Register val)550 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
551 z_lg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n)));
552 }
553
store_ptr(int n,Register val)554 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
555 z_stg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n)));
556 }
557
prepare_to_jump_from_interpreted(Register method)558 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted(Register method) {
559 // Satisfy interpreter calling convention (see generate_normal_entry()).
560 z_lgr(Z_R10, Z_SP); // Set sender sp (aka initial caller sp, aka unextended sp).
561 // Record top_frame_sp, because the callee might modify it, if it's compiled.
562 z_stg(Z_SP, _z_ijava_state_neg(top_frame_sp), Z_fp);
563 save_bcp();
564 save_esp();
565 z_lgr(Z_method, method); // Set Z_method (kills Z_fp!).
566 }
567
568 // Jump to from_interpreted entry of a call unless single stepping is possible
569 // in this thread in which case we must call the i2i entry.
jump_from_interpreted(Register method,Register temp)570 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
571 assert_different_registers(method, Z_R10 /*used for initial_caller_sp*/, temp);
572 prepare_to_jump_from_interpreted(method);
573
574 if (JvmtiExport::can_post_interpreter_events()) {
575 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
576 // compiled code in threads for which the event is enabled. Check here for
577 // interp_only_mode if these events CAN be enabled.
578 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
579 MacroAssembler::load_and_test_int(Z_R0_scratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
580 z_bcr(bcondEqual, Z_R1_scratch); // Run compiled code if zero.
581 // Run interpreted.
582 z_lg(Z_R1_scratch, Address(method, Method::interpreter_entry_offset()));
583 z_br(Z_R1_scratch);
584 } else {
585 // Run compiled code.
586 z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
587 z_br(Z_R1_scratch);
588 }
589 }
590
591 #ifdef ASSERT
verify_esp(Register Resp,Register Rtemp)592 void InterpreterMacroAssembler::verify_esp(Register Resp, Register Rtemp) {
593 // About to read or write Resp[0].
594 // Make sure it is not in the monitors or the TOP_IJAVA_FRAME_ABI.
595 address reentry = NULL;
596
597 {
598 // Check if the frame pointer in Z_fp is correct.
599 NearLabel OK;
600 z_cg(Z_fp, 0, Z_SP);
601 z_bre(OK);
602 reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp");
603 bind(OK);
604 }
605 {
606 // Resp must not point into or below the operand stack,
607 // i.e. IJAVA_STATE.monitors > Resp.
608 NearLabel OK;
609 Register Rmonitors = Rtemp;
610 z_lg(Rmonitors, _z_ijava_state_neg(monitors), Z_fp);
611 compareU64_and_branch(Rmonitors, Resp, bcondHigh, OK);
612 reentry = stop_chain_static(reentry, "too many pops: Z_esp points into monitor area");
613 bind(OK);
614 }
615 {
616 // Resp may point to the last word of TOP_IJAVA_FRAME_ABI, but not below
617 // i.e. !(Z_SP + frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize > Resp).
618 NearLabel OK;
619 Register Rabi_bottom = Rtemp;
620 add2reg(Rabi_bottom, frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize, Z_SP);
621 compareU64_and_branch(Rabi_bottom, Resp, bcondNotHigh, OK);
622 reentry = stop_chain_static(reentry, "too many pushes: Z_esp points into TOP_IJAVA_FRAME_ABI");
623 bind(OK);
624 }
625 }
626
asm_assert_ijava_state_magic(Register tmp)627 void InterpreterMacroAssembler::asm_assert_ijava_state_magic(Register tmp) {
628 Label magic_ok;
629 load_const_optimized(tmp, frame::z_istate_magic_number);
630 z_cg(tmp, Address(Z_fp, _z_ijava_state_neg(magic)));
631 z_bre(magic_ok);
632 stop_static("error: wrong magic number in ijava_state access");
633 bind(magic_ok);
634 }
635 #endif // ASSERT
636
save_bcp()637 void InterpreterMacroAssembler::save_bcp() {
638 z_stg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
639 asm_assert_ijava_state_magic(Z_bcp);
640 NOT_PRODUCT(z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))));
641 }
642
restore_bcp()643 void InterpreterMacroAssembler::restore_bcp() {
644 asm_assert_ijava_state_magic(Z_bcp);
645 z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
646 }
647
save_esp()648 void InterpreterMacroAssembler::save_esp() {
649 z_stg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp)));
650 }
651
restore_esp()652 void InterpreterMacroAssembler::restore_esp() {
653 asm_assert_ijava_state_magic(Z_esp);
654 z_lg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp)));
655 }
656
get_monitors(Register reg)657 void InterpreterMacroAssembler::get_monitors(Register reg) {
658 asm_assert_ijava_state_magic(reg);
659 mem2reg_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
660 }
661
save_monitors(Register reg)662 void InterpreterMacroAssembler::save_monitors(Register reg) {
663 reg2mem_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
664 }
665
get_mdp(Register mdp)666 void InterpreterMacroAssembler::get_mdp(Register mdp) {
667 z_lg(mdp, _z_ijava_state_neg(mdx), Z_fp);
668 }
669
save_mdp(Register mdp)670 void InterpreterMacroAssembler::save_mdp(Register mdp) {
671 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
672 }
673
674 // Values that are only read (besides initialization).
restore_locals()675 void InterpreterMacroAssembler::restore_locals() {
676 asm_assert_ijava_state_magic(Z_locals);
677 z_lg(Z_locals, Address(Z_fp, _z_ijava_state_neg(locals)));
678 }
679
get_method(Register reg)680 void InterpreterMacroAssembler::get_method(Register reg) {
681 asm_assert_ijava_state_magic(reg);
682 z_lg(reg, Address(Z_fp, _z_ijava_state_neg(method)));
683 }
684
get_2_byte_integer_at_bcp(Register Rdst,int bcp_offset,signedOrNot is_signed)685 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(Register Rdst, int bcp_offset,
686 signedOrNot is_signed) {
687 // Rdst is an 8-byte return value!!!
688
689 // Unaligned loads incur only a small penalty on z/Architecture. The penalty
690 // is a few (2..3) ticks, even when the load crosses a cache line
691 // boundary. In case of a cache miss, the stall could, of course, be
692 // much longer.
693
694 switch (is_signed) {
695 case Signed:
696 z_lgh(Rdst, bcp_offset, Z_R0, Z_bcp);
697 break;
698 case Unsigned:
699 z_llgh(Rdst, bcp_offset, Z_R0, Z_bcp);
700 break;
701 default:
702 ShouldNotReachHere();
703 }
704 }
705
706
get_4_byte_integer_at_bcp(Register Rdst,int bcp_offset,setCCOrNot set_cc)707 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(Register Rdst, int bcp_offset,
708 setCCOrNot set_cc) {
709 // Rdst is an 8-byte return value!!!
710
711 // Unaligned loads incur only a small penalty on z/Architecture. The penalty
712 // is a few (2..3) ticks, even when the load crosses a cache line
713 // boundary. In case of a cache miss, the stall could, of course, be
714 // much longer.
715
716 // Both variants implement a sign-extending int2long load.
717 if (set_cc == set_CC) {
718 load_and_test_int2long(Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
719 } else {
720 mem2reg_signed_opt( Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
721 }
722 }
723
get_constant_pool(Register Rdst)724 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
725 get_method(Rdst);
726 mem2reg_opt(Rdst, Address(Rdst, Method::const_offset()));
727 mem2reg_opt(Rdst, Address(Rdst, ConstMethod::constants_offset()));
728 }
729
get_cpool_and_tags(Register Rcpool,Register Rtags)730 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
731 get_constant_pool(Rcpool);
732 mem2reg_opt(Rtags, Address(Rcpool, ConstantPool::tags_offset_in_bytes()));
733 }
734
735 // Unlock if synchronized method.
736 //
737 // Unlock the receiver if this is a synchronized method.
738 // Unlock any Java monitors from syncronized blocks.
739 //
740 // If there are locked Java monitors
741 // If throw_monitor_exception
742 // throws IllegalMonitorStateException
743 // Else if install_monitor_exception
744 // installs IllegalMonitorStateException
745 // Else
746 // no error processing
unlock_if_synchronized_method(TosState state,bool throw_monitor_exception,bool install_monitor_exception)747 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
748 bool throw_monitor_exception,
749 bool install_monitor_exception) {
750 NearLabel unlocked, unlock, no_unlock;
751
752 {
753 Register R_method = Z_ARG2;
754 Register R_do_not_unlock_if_synchronized = Z_ARG3;
755
756 // Get the value of _do_not_unlock_if_synchronized into G1_scratch.
757 const Address do_not_unlock_if_synchronized(Z_thread,
758 JavaThread::do_not_unlock_if_synchronized_offset());
759 load_sized_value(R_do_not_unlock_if_synchronized, do_not_unlock_if_synchronized, 1, false /*unsigned*/);
760 z_mvi(do_not_unlock_if_synchronized, false); // Reset the flag.
761
762 // Check if synchronized method.
763 get_method(R_method);
764 verify_oop(Z_tos, state);
765 push(state); // Save tos/result.
766 testbit(method2_(R_method, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
767 z_bfalse(unlocked);
768
769 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
770 // is set.
771 compareU64_and_branch(R_do_not_unlock_if_synchronized, (intptr_t)0L, bcondNotEqual, no_unlock);
772 }
773
774 // unlock monitor
775
776 // BasicObjectLock will be first in list, since this is a
777 // synchronized method. However, need to check that the object has
778 // not been unlocked by an explicit monitorexit bytecode.
779 const Address monitor(Z_fp, -(frame::z_ijava_state_size + (int) sizeof(BasicObjectLock)));
780 // We use Z_ARG2 so that if we go slow path it will be the correct
781 // register for unlock_object to pass to VM directly.
782 load_address(Z_ARG2, monitor); // Address of first monitor.
783 z_lg(Z_ARG3, Address(Z_ARG2, BasicObjectLock::obj_offset_in_bytes()));
784 compareU64_and_branch(Z_ARG3, (intptr_t)0L, bcondNotEqual, unlock);
785
786 if (throw_monitor_exception) {
787 // Entry already unlocked need to throw an exception.
788 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
789 should_not_reach_here();
790 } else {
791 // Monitor already unlocked during a stack unroll.
792 // If requested, install an illegal_monitor_state_exception.
793 // Continue with stack unrolling.
794 if (install_monitor_exception) {
795 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
796 }
797 z_bru(unlocked);
798 }
799
800 bind(unlock);
801
802 unlock_object(Z_ARG2);
803
804 bind(unlocked);
805
806 // I0, I1: Might contain return value
807
808 // Check that all monitors are unlocked.
809 {
810 NearLabel loop, exception, entry, restart;
811 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
812 // We use Z_ARG2 so that if we go slow path it will be the correct
813 // register for unlock_object to pass to VM directly.
814 Register R_current_monitor = Z_ARG2;
815 Register R_monitor_block_bot = Z_ARG1;
816 const Address monitor_block_top(Z_fp, _z_ijava_state_neg(monitors));
817 const Address monitor_block_bot(Z_fp, -frame::z_ijava_state_size);
818
819 bind(restart);
820 // Starting with top-most entry.
821 z_lg(R_current_monitor, monitor_block_top);
822 // Points to word before bottom of monitor block.
823 load_address(R_monitor_block_bot, monitor_block_bot);
824 z_bru(entry);
825
826 // Entry already locked, need to throw exception.
827 bind(exception);
828
829 if (throw_monitor_exception) {
830 // Throw exception.
831 MacroAssembler::call_VM(noreg,
832 CAST_FROM_FN_PTR(address, InterpreterRuntime::
833 throw_illegal_monitor_state_exception));
834 should_not_reach_here();
835 } else {
836 // Stack unrolling. Unlock object and install illegal_monitor_exception.
837 // Unlock does not block, so don't have to worry about the frame.
838 // We don't have to preserve c_rarg1 since we are going to throw an exception.
839 unlock_object(R_current_monitor);
840 if (install_monitor_exception) {
841 call_VM(noreg, CAST_FROM_FN_PTR(address,
842 InterpreterRuntime::
843 new_illegal_monitor_state_exception));
844 }
845 z_bru(restart);
846 }
847
848 bind(loop);
849 // Check if current entry is used.
850 load_and_test_long(Z_R0_scratch, Address(R_current_monitor, BasicObjectLock::obj_offset_in_bytes()));
851 z_brne(exception);
852
853 add2reg(R_current_monitor, entry_size); // Otherwise advance to next entry.
854 bind(entry);
855 compareU64_and_branch(R_current_monitor, R_monitor_block_bot, bcondNotEqual, loop);
856 }
857
858 bind(no_unlock);
859 pop(state);
860 verify_oop(Z_tos, state);
861 }
862
narrow(Register result,Register ret_type)863 void InterpreterMacroAssembler::narrow(Register result, Register ret_type) {
864 get_method(ret_type);
865 z_lg(ret_type, Address(ret_type, in_bytes(Method::const_offset())));
866 z_lb(ret_type, Address(ret_type, in_bytes(ConstMethod::result_type_offset())));
867
868 Label notBool, notByte, notChar, done;
869
870 // common case first
871 compareU32_and_branch(ret_type, T_INT, bcondEqual, done);
872
873 compareU32_and_branch(ret_type, T_BOOLEAN, bcondNotEqual, notBool);
874 z_nilf(result, 0x1);
875 z_bru(done);
876
877 bind(notBool);
878 compareU32_and_branch(ret_type, T_BYTE, bcondNotEqual, notByte);
879 z_lbr(result, result);
880 z_bru(done);
881
882 bind(notByte);
883 compareU32_and_branch(ret_type, T_CHAR, bcondNotEqual, notChar);
884 z_nilf(result, 0xffff);
885 z_bru(done);
886
887 bind(notChar);
888 // compareU32_and_branch(ret_type, T_SHORT, bcondNotEqual, notShort);
889 z_lhr(result, result);
890
891 // Nothing to do for T_INT
892 bind(done);
893 }
894
895 // remove activation
896 //
897 // Unlock the receiver if this is a synchronized method.
898 // Unlock any Java monitors from syncronized blocks.
899 // Remove the activation from the stack.
900 //
901 // If there are locked Java monitors
902 // If throw_monitor_exception
903 // throws IllegalMonitorStateException
904 // Else if install_monitor_exception
905 // installs IllegalMonitorStateException
906 // Else
907 // no error processing
remove_activation(TosState state,Register return_pc,bool throw_monitor_exception,bool install_monitor_exception,bool notify_jvmti)908 void InterpreterMacroAssembler::remove_activation(TosState state,
909 Register return_pc,
910 bool throw_monitor_exception,
911 bool install_monitor_exception,
912 bool notify_jvmti) {
913 BLOCK_COMMENT("remove_activation {");
914 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
915
916 // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
917 notify_method_exit(false, state, notify_jvmti ? NotifyJVMTI : SkipNotifyJVMTI);
918
919 if (StackReservedPages > 0) {
920 BLOCK_COMMENT("reserved_stack_check:");
921 // Test if reserved zone needs to be enabled.
922 Label no_reserved_zone_enabling;
923
924 // Compare frame pointers. There is no good stack pointer, as with stack
925 // frame compression we can get different SPs when we do calls. A subsequent
926 // call could have a smaller SP, so that this compare succeeds for an
927 // inner call of the method annotated with ReservedStack.
928 z_lg(Z_R0, Address(Z_SP, (intptr_t)_z_abi(callers_sp)));
929 z_clg(Z_R0, Address(Z_thread, JavaThread::reserved_stack_activation_offset())); // Compare with frame pointer in memory.
930 z_brl(no_reserved_zone_enabling);
931
932 // Enable reserved zone again, throw stack overflow exception.
933 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), Z_thread);
934 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
935
936 should_not_reach_here();
937
938 bind(no_reserved_zone_enabling);
939 }
940
941 verify_oop(Z_tos, state);
942 verify_thread();
943
944 pop_interpreter_frame(return_pc, Z_ARG2, Z_ARG3);
945 BLOCK_COMMENT("} remove_activation");
946 }
947
948 // lock object
949 //
950 // Registers alive
951 // monitor - Address of the BasicObjectLock to be used for locking,
952 // which must be initialized with the object to lock.
953 // object - Address of the object to be locked.
lock_object(Register monitor,Register object)954 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
955
956 if (UseHeavyMonitors) {
957 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
958 return;
959 }
960
961 // template code:
962 //
963 // markOop displaced_header = obj->mark().set_unlocked();
964 // monitor->lock()->set_displaced_header(displaced_header);
965 // if (Atomic::cmpxchg(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
966 // // We stored the monitor address into the object's mark word.
967 // } else if (THREAD->is_lock_owned((address)displaced_header))
968 // // Simple recursive case.
969 // monitor->lock()->set_displaced_header(NULL);
970 // } else {
971 // // Slow path.
972 // InterpreterRuntime::monitorenter(THREAD, monitor);
973 // }
974
975 const Register displaced_header = Z_ARG5;
976 const Register object_mark_addr = Z_ARG4;
977 const Register current_header = Z_ARG5;
978
979 NearLabel done;
980 NearLabel slow_case;
981
982 // markOop displaced_header = obj->mark().set_unlocked();
983
984 // Load markOop from object into displaced_header.
985 z_lg(displaced_header, oopDesc::mark_offset_in_bytes(), object);
986
987 if (UseBiasedLocking) {
988 biased_locking_enter(object, displaced_header, Z_R1, Z_R0, done, &slow_case);
989 }
990
991 // Set displaced_header to be (markOop of object | UNLOCK_VALUE).
992 z_oill(displaced_header, markOopDesc::unlocked_value);
993
994 // monitor->lock()->set_displaced_header(displaced_header);
995
996 // Initialize the box (Must happen before we update the object mark!).
997 z_stg(displaced_header, BasicObjectLock::lock_offset_in_bytes() +
998 BasicLock::displaced_header_offset_in_bytes(), monitor);
999
1000 // if (Atomic::cmpxchg(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
1001
1002 // Store stack address of the BasicObjectLock (this is monitor) into object.
1003 add2reg(object_mark_addr, oopDesc::mark_offset_in_bytes(), object);
1004
1005 z_csg(displaced_header, monitor, 0, object_mark_addr);
1006 assert(current_header==displaced_header, "must be same register"); // Identified two registers from z/Architecture.
1007
1008 z_bre(done);
1009
1010 // } else if (THREAD->is_lock_owned((address)displaced_header))
1011 // // Simple recursive case.
1012 // monitor->lock()->set_displaced_header(NULL);
1013
1014 // We did not see an unlocked object so try the fast recursive case.
1015
1016 // Check if owner is self by comparing the value in the markOop of object
1017 // (current_header) with the stack pointer.
1018 z_sgr(current_header, Z_SP);
1019
1020 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1021
1022 // The prior sequence "LGR, NGR, LTGR" can be done better
1023 // (Z_R1 is temp and not used after here).
1024 load_const_optimized(Z_R0, (~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place));
1025 z_ngr(Z_R0, current_header); // AND sets CC (result eq/ne 0)
1026
1027 // If condition is true we are done and hence we can store 0 in the displaced
1028 // header indicating it is a recursive lock and be done.
1029 z_brne(slow_case);
1030 z_release(); // Membar unnecessary on zarch AND because the above csg does a sync before and after.
1031 z_stg(Z_R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() +
1032 BasicLock::displaced_header_offset_in_bytes(), monitor);
1033 z_bru(done);
1034
1035 // } else {
1036 // // Slow path.
1037 // InterpreterRuntime::monitorenter(THREAD, monitor);
1038
1039 // None of the above fast optimizations worked so we have to get into the
1040 // slow case of monitor enter.
1041 bind(slow_case);
1042 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
1043
1044 // }
1045
1046 bind(done);
1047 }
1048
1049 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1050 //
1051 // Registers alive
1052 // monitor - address of the BasicObjectLock to be used for locking,
1053 // which must be initialized with the object to lock.
1054 //
1055 // Throw IllegalMonitorException if object is not locked by current thread.
unlock_object(Register monitor,Register object)1056 void InterpreterMacroAssembler::unlock_object(Register monitor, Register object) {
1057
1058 if (UseHeavyMonitors) {
1059 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1060 return;
1061 }
1062
1063 // else {
1064 // template code:
1065 //
1066 // if ((displaced_header = monitor->displaced_header()) == NULL) {
1067 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL.
1068 // monitor->set_obj(NULL);
1069 // } else if (Atomic::cmpxchg(displaced_header, obj->mark_addr(), monitor) == monitor) {
1070 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1071 // monitor->set_obj(NULL);
1072 // } else {
1073 // // Slow path.
1074 // InterpreterRuntime::monitorexit(THREAD, monitor);
1075 // }
1076
1077 const Register displaced_header = Z_ARG4;
1078 const Register current_header = Z_R1;
1079 Address obj_entry(monitor, BasicObjectLock::obj_offset_in_bytes());
1080 Label done;
1081
1082 if (object == noreg) {
1083 // In the template interpreter, we must assure that the object
1084 // entry in the monitor is cleared on all paths. Thus we move
1085 // loading up to here, and clear the entry afterwards.
1086 object = Z_ARG3; // Use Z_ARG3 if caller didn't pass object.
1087 z_lg(object, obj_entry);
1088 }
1089
1090 assert_different_registers(monitor, object, displaced_header, current_header);
1091
1092 // if ((displaced_header = monitor->displaced_header()) == NULL) {
1093 // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL.
1094 // monitor->set_obj(NULL);
1095
1096 clear_mem(obj_entry, sizeof(oop));
1097
1098 if (UseBiasedLocking) {
1099 // The object address from the monitor is in object.
1100 assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0");
1101 biased_locking_exit(object, displaced_header, done);
1102 }
1103
1104 // Test first if we are in the fast recursive case.
1105 MacroAssembler::load_and_test_long(displaced_header,
1106 Address(monitor, BasicObjectLock::lock_offset_in_bytes() +
1107 BasicLock::displaced_header_offset_in_bytes()));
1108 z_bre(done); // displaced_header == 0 -> goto done
1109
1110 // } else if (Atomic::cmpxchg(displaced_header, obj->mark_addr(), monitor) == monitor) {
1111 // // We swapped the unlocked mark in displaced_header into the object's mark word.
1112 // monitor->set_obj(NULL);
1113
1114 // If we still have a lightweight lock, unlock the object and be done.
1115
1116 // The markword is expected to be at offset 0.
1117 assert(oopDesc::mark_offset_in_bytes() == 0, "unlock_object: review code below");
1118
1119 // We have the displaced header in displaced_header. If the lock is still
1120 // lightweight, it will contain the monitor address and we'll store the
1121 // displaced header back into the object's mark word.
1122 z_lgr(current_header, monitor);
1123 z_csg(current_header, displaced_header, 0, object);
1124 z_bre(done);
1125
1126 // } else {
1127 // // Slow path.
1128 // InterpreterRuntime::monitorexit(THREAD, monitor);
1129
1130 // The lock has been converted into a heavy lock and hence
1131 // we need to get into the slow case.
1132 z_stg(object, obj_entry); // Restore object entry, has been cleared above.
1133 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1134
1135 // }
1136
1137 bind(done);
1138 }
1139
test_method_data_pointer(Register mdp,Label & zero_continue)1140 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) {
1141 assert(ProfileInterpreter, "must be profiling interpreter");
1142 load_and_test_long(mdp, Address(Z_fp, _z_ijava_state_neg(mdx)));
1143 z_brz(zero_continue);
1144 }
1145
1146 // Set the method data pointer for the current bcp.
set_method_data_pointer_for_bcp()1147 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1148 assert(ProfileInterpreter, "must be profiling interpreter");
1149 Label set_mdp;
1150 Register mdp = Z_ARG4;
1151 Register method = Z_ARG5;
1152
1153 get_method(method);
1154 // Test MDO to avoid the call if it is NULL.
1155 load_and_test_long(mdp, method2_(method, method_data));
1156 z_brz(set_mdp);
1157
1158 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), method, Z_bcp);
1159 // Z_RET: mdi
1160 // Mdo is guaranteed to be non-zero here, we checked for it before the call.
1161 assert(method->is_nonvolatile(), "choose nonvolatile reg or reload from frame");
1162 z_lg(mdp, method2_(method, method_data)); // Must reload, mdp is volatile reg.
1163 add2reg_with_index(mdp, in_bytes(MethodData::data_offset()), Z_RET, mdp);
1164
1165 bind(set_mdp);
1166 save_mdp(mdp);
1167 }
1168
verify_method_data_pointer()1169 void InterpreterMacroAssembler::verify_method_data_pointer() {
1170 assert(ProfileInterpreter, "must be profiling interpreter");
1171 #ifdef ASSERT
1172 NearLabel verify_continue;
1173 Register bcp_expected = Z_ARG3;
1174 Register mdp = Z_ARG4;
1175 Register method = Z_ARG5;
1176
1177 test_method_data_pointer(mdp, verify_continue); // If mdp is zero, continue
1178 get_method(method);
1179
1180 // If the mdp is valid, it will point to a DataLayout header which is
1181 // consistent with the bcp. The converse is highly probable also.
1182 load_sized_value(bcp_expected, Address(mdp, DataLayout::bci_offset()), 2, false /*signed*/);
1183 z_ag(bcp_expected, Address(method, Method::const_offset()));
1184 load_address(bcp_expected, Address(bcp_expected, ConstMethod::codes_offset()));
1185 compareU64_and_branch(bcp_expected, Z_bcp, bcondEqual, verify_continue);
1186 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), method, Z_bcp, mdp);
1187 bind(verify_continue);
1188 #endif // ASSERT
1189 }
1190
set_mdp_data_at(Register mdp_in,int constant,Register value)1191 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) {
1192 assert(ProfileInterpreter, "must be profiling interpreter");
1193 z_stg(value, constant, mdp_in);
1194 }
1195
increment_mdp_data_at(Register mdp_in,int constant,Register tmp,bool decrement)1196 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1197 int constant,
1198 Register tmp,
1199 bool decrement) {
1200 assert_different_registers(mdp_in, tmp);
1201 // counter address
1202 Address data(mdp_in, constant);
1203 const int delta = decrement ? -DataLayout::counter_increment : DataLayout::counter_increment;
1204 add2mem_64(Address(mdp_in, constant), delta, tmp);
1205 }
1206
set_mdp_flag_at(Register mdp_in,int flag_byte_constant)1207 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1208 int flag_byte_constant) {
1209 assert(ProfileInterpreter, "must be profiling interpreter");
1210 // Set the flag.
1211 z_oi(Address(mdp_in, DataLayout::flags_offset()), flag_byte_constant);
1212 }
1213
test_mdp_data_at(Register mdp_in,int offset,Register value,Register test_value_out,Label & not_equal_continue)1214 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1215 int offset,
1216 Register value,
1217 Register test_value_out,
1218 Label& not_equal_continue) {
1219 assert(ProfileInterpreter, "must be profiling interpreter");
1220 if (test_value_out == noreg) {
1221 z_cg(value, Address(mdp_in, offset));
1222 z_brne(not_equal_continue);
1223 } else {
1224 // Put the test value into a register, so caller can use it:
1225 z_lg(test_value_out, Address(mdp_in, offset));
1226 compareU64_and_branch(test_value_out, value, bcondNotEqual, not_equal_continue);
1227 }
1228 }
1229
update_mdp_by_offset(Register mdp_in,int offset_of_disp)1230 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) {
1231 update_mdp_by_offset(mdp_in, noreg, offset_of_disp);
1232 }
1233
update_mdp_by_offset(Register mdp_in,Register dataidx,int offset_of_disp)1234 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1235 Register dataidx,
1236 int offset_of_disp) {
1237 assert(ProfileInterpreter, "must be profiling interpreter");
1238 Address disp_address(mdp_in, dataidx, offset_of_disp);
1239 Assembler::z_ag(mdp_in, disp_address);
1240 save_mdp(mdp_in);
1241 }
1242
update_mdp_by_constant(Register mdp_in,int constant)1243 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) {
1244 assert(ProfileInterpreter, "must be profiling interpreter");
1245 add2reg(mdp_in, constant);
1246 save_mdp(mdp_in);
1247 }
1248
update_mdp_for_ret(Register return_bci)1249 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1250 assert(ProfileInterpreter, "must be profiling interpreter");
1251 assert(return_bci->is_nonvolatile(), "choose nonvolatile reg or save/restore");
1252 call_VM(noreg,
1253 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1254 return_bci);
1255 }
1256
profile_taken_branch(Register mdp,Register bumped_count)1257 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) {
1258 if (ProfileInterpreter) {
1259 Label profile_continue;
1260
1261 // If no method data exists, go to profile_continue.
1262 // Otherwise, assign to mdp.
1263 test_method_data_pointer(mdp, profile_continue);
1264
1265 // We are taking a branch. Increment the taken count.
1266 // We inline increment_mdp_data_at to return bumped_count in a register
1267 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1268 Address data(mdp, JumpData::taken_offset());
1269 z_lg(bumped_count, data);
1270 // 64-bit overflow is very unlikely. Saturation to 32-bit values is
1271 // performed when reading the counts.
1272 add2reg(bumped_count, DataLayout::counter_increment);
1273 z_stg(bumped_count, data); // Store back out
1274
1275 // The method data pointer needs to be updated to reflect the new target.
1276 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1277 bind(profile_continue);
1278 }
1279 }
1280
1281 // Kills Z_R1_scratch.
profile_not_taken_branch(Register mdp)1282 void InterpreterMacroAssembler::profile_not_taken_branch(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 taking a branch. Increment the not taken count.
1290 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()), Z_R1_scratch);
1291
1292 // The method data pointer needs to be updated to correspond to
1293 // the next bytecode.
1294 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1295 bind(profile_continue);
1296 }
1297 }
1298
1299 // Kills: Z_R1_scratch.
profile_call(Register mdp)1300 void InterpreterMacroAssembler::profile_call(Register mdp) {
1301 if (ProfileInterpreter) {
1302 Label profile_continue;
1303
1304 // If no method data exists, go to profile_continue.
1305 test_method_data_pointer(mdp, profile_continue);
1306
1307 // We are making a call. Increment the count.
1308 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1309
1310 // The method data pointer needs to be updated to reflect the new target.
1311 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1312 bind(profile_continue);
1313 }
1314 }
1315
profile_final_call(Register mdp)1316 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1317 if (ProfileInterpreter) {
1318 Label profile_continue;
1319
1320 // If no method data exists, go to profile_continue.
1321 test_method_data_pointer(mdp, profile_continue);
1322
1323 // We are making a call. Increment the count.
1324 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1325
1326 // The method data pointer needs to be updated to reflect the new target.
1327 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1328 bind(profile_continue);
1329 }
1330 }
1331
profile_virtual_call(Register receiver,Register mdp,Register reg2,bool receiver_can_be_null)1332 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1333 Register mdp,
1334 Register reg2,
1335 bool receiver_can_be_null) {
1336 if (ProfileInterpreter) {
1337 NearLabel profile_continue;
1338
1339 // If no method data exists, go to profile_continue.
1340 test_method_data_pointer(mdp, profile_continue);
1341
1342 NearLabel skip_receiver_profile;
1343 if (receiver_can_be_null) {
1344 NearLabel not_null;
1345 compareU64_and_branch(receiver, (intptr_t)0L, bcondNotEqual, not_null);
1346 // We are making a call. Increment the count for null receiver.
1347 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1348 z_bru(skip_receiver_profile);
1349 bind(not_null);
1350 }
1351
1352 // Record the receiver type.
1353 record_klass_in_profile(receiver, mdp, reg2, true);
1354 bind(skip_receiver_profile);
1355
1356 // The method data pointer needs to be updated to reflect the new target.
1357 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1358 bind(profile_continue);
1359 }
1360 }
1361
1362 // This routine creates a state machine for updating the multi-row
1363 // type profile at a virtual call site (or other type-sensitive bytecode).
1364 // The machine visits each row (of receiver/count) until the receiver type
1365 // is found, or until it runs out of rows. At the same time, it remembers
1366 // the location of the first empty row. (An empty row records null for its
1367 // receiver, and can be allocated for a newly-observed receiver type.)
1368 // Because there are two degrees of freedom in the state, a simple linear
1369 // search will not work; it must be a decision tree. Hence this helper
1370 // function is recursive, to generate the required tree structured code.
1371 // It's the interpreter, so we are trading off code space for speed.
1372 // See below for example code.
record_klass_in_profile_helper(Register receiver,Register mdp,Register reg2,int start_row,Label & done,bool is_virtual_call)1373 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1374 Register receiver, Register mdp,
1375 Register reg2, int start_row,
1376 Label& done, bool is_virtual_call) {
1377 if (TypeProfileWidth == 0) {
1378 if (is_virtual_call) {
1379 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1380 }
1381 return;
1382 }
1383
1384 int last_row = VirtualCallData::row_limit() - 1;
1385 assert(start_row <= last_row, "must be work left to do");
1386 // Test this row for both the receiver and for null.
1387 // Take any of three different outcomes:
1388 // 1. found receiver => increment count and goto done
1389 // 2. found null => keep looking for case 1, maybe allocate this cell
1390 // 3. found something else => keep looking for cases 1 and 2
1391 // Case 3 is handled by a recursive call.
1392 for (int row = start_row; row <= last_row; row++) {
1393 NearLabel next_test;
1394 bool test_for_null_also = (row == start_row);
1395
1396 // See if the receiver is receiver[n].
1397 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1398 test_mdp_data_at(mdp, recvr_offset, receiver,
1399 (test_for_null_also ? reg2 : noreg),
1400 next_test);
1401 // (Reg2 now contains the receiver from the CallData.)
1402
1403 // The receiver is receiver[n]. Increment count[n].
1404 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1405 increment_mdp_data_at(mdp, count_offset);
1406 z_bru(done);
1407 bind(next_test);
1408
1409 if (test_for_null_also) {
1410 Label found_null;
1411 // Failed the equality check on receiver[n]... Test for null.
1412 z_ltgr(reg2, reg2);
1413 if (start_row == last_row) {
1414 // The only thing left to do is handle the null case.
1415 if (is_virtual_call) {
1416 z_brz(found_null);
1417 // Receiver did not match any saved receiver and there is no empty row for it.
1418 // Increment total counter to indicate polymorphic case.
1419 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1420 z_bru(done);
1421 bind(found_null);
1422 } else {
1423 z_brnz(done);
1424 }
1425 break;
1426 }
1427 // Since null is rare, make it be the branch-taken case.
1428 z_brz(found_null);
1429
1430 // Put all the "Case 3" tests here.
1431 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
1432
1433 // Found a null. Keep searching for a matching receiver,
1434 // but remember that this is an empty (unused) slot.
1435 bind(found_null);
1436 }
1437 }
1438
1439 // In the fall-through case, we found no matching receiver, but we
1440 // observed the receiver[start_row] is NULL.
1441
1442 // Fill in the receiver field and increment the count.
1443 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1444 set_mdp_data_at(mdp, recvr_offset, receiver);
1445 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1446 load_const_optimized(reg2, DataLayout::counter_increment);
1447 set_mdp_data_at(mdp, count_offset, reg2);
1448 if (start_row > 0) {
1449 z_bru(done);
1450 }
1451 }
1452
1453 // Example state machine code for three profile rows:
1454 // // main copy of decision tree, rooted at row[1]
1455 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1456 // if (row[0].rec != NULL) {
1457 // // inner copy of decision tree, rooted at row[1]
1458 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1459 // if (row[1].rec != NULL) {
1460 // // degenerate decision tree, rooted at row[2]
1461 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1462 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1463 // row[2].init(rec); goto done;
1464 // } else {
1465 // // remember row[1] is empty
1466 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1467 // row[1].init(rec); goto done;
1468 // }
1469 // } else {
1470 // // remember row[0] is empty
1471 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1472 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1473 // row[0].init(rec); goto done;
1474 // }
1475 // done:
1476
record_klass_in_profile(Register receiver,Register mdp,Register reg2,bool is_virtual_call)1477 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1478 Register mdp, Register reg2,
1479 bool is_virtual_call) {
1480 assert(ProfileInterpreter, "must be profiling");
1481 Label done;
1482
1483 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1484
1485 bind (done);
1486 }
1487
profile_ret(Register return_bci,Register mdp)1488 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
1489 if (ProfileInterpreter) {
1490 NearLabel profile_continue;
1491 uint row;
1492
1493 // If no method data exists, go to profile_continue.
1494 test_method_data_pointer(mdp, profile_continue);
1495
1496 // Update the total ret count.
1497 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1498
1499 for (row = 0; row < RetData::row_limit(); row++) {
1500 NearLabel next_test;
1501
1502 // See if return_bci is equal to bci[n]:
1503 test_mdp_data_at(mdp,
1504 in_bytes(RetData::bci_offset(row)),
1505 return_bci, noreg,
1506 next_test);
1507
1508 // Return_bci is equal to bci[n]. Increment the count.
1509 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1510
1511 // The method data pointer needs to be updated to reflect the new target.
1512 update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row)));
1513 z_bru(profile_continue);
1514 bind(next_test);
1515 }
1516
1517 update_mdp_for_ret(return_bci);
1518
1519 bind(profile_continue);
1520 }
1521 }
1522
profile_null_seen(Register mdp)1523 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1524 if (ProfileInterpreter) {
1525 Label profile_continue;
1526
1527 // If no method data exists, go to profile_continue.
1528 test_method_data_pointer(mdp, profile_continue);
1529
1530 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1531
1532 // The method data pointer needs to be updated.
1533 int mdp_delta = in_bytes(BitData::bit_data_size());
1534 if (TypeProfileCasts) {
1535 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1536 }
1537 update_mdp_by_constant(mdp, mdp_delta);
1538
1539 bind(profile_continue);
1540 }
1541 }
1542
profile_typecheck_failed(Register mdp,Register tmp)1543 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp, Register tmp) {
1544 if (ProfileInterpreter && TypeProfileCasts) {
1545 Label profile_continue;
1546
1547 // If no method data exists, go to profile_continue.
1548 test_method_data_pointer(mdp, profile_continue);
1549
1550 int count_offset = in_bytes(CounterData::count_offset());
1551 // Back up the address, since we have already bumped the mdp.
1552 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1553
1554 // *Decrement* the counter. We expect to see zero or small negatives.
1555 increment_mdp_data_at(mdp, count_offset, tmp, true);
1556
1557 bind (profile_continue);
1558 }
1559 }
1560
profile_typecheck(Register mdp,Register klass,Register reg2)1561 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1562 if (ProfileInterpreter) {
1563 Label profile_continue;
1564
1565 // If no method data exists, go to profile_continue.
1566 test_method_data_pointer(mdp, profile_continue);
1567
1568 // The method data pointer needs to be updated.
1569 int mdp_delta = in_bytes(BitData::bit_data_size());
1570 if (TypeProfileCasts) {
1571 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1572
1573 // Record the object type.
1574 record_klass_in_profile(klass, mdp, reg2, false);
1575 }
1576 update_mdp_by_constant(mdp, mdp_delta);
1577
1578 bind(profile_continue);
1579 }
1580 }
1581
profile_switch_default(Register mdp)1582 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1583 if (ProfileInterpreter) {
1584 Label profile_continue;
1585
1586 // If no method data exists, go to profile_continue.
1587 test_method_data_pointer(mdp, profile_continue);
1588
1589 // Update the default case count.
1590 increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset()));
1591
1592 // The method data pointer needs to be updated.
1593 update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset()));
1594
1595 bind(profile_continue);
1596 }
1597 }
1598
1599 // Kills: index, scratch1, scratch2.
profile_switch_case(Register index,Register mdp,Register scratch1,Register scratch2)1600 void InterpreterMacroAssembler::profile_switch_case(Register index,
1601 Register mdp,
1602 Register scratch1,
1603 Register scratch2) {
1604 if (ProfileInterpreter) {
1605 Label profile_continue;
1606 assert_different_registers(index, mdp, scratch1, scratch2);
1607
1608 // If no method data exists, go to profile_continue.
1609 test_method_data_pointer(mdp, profile_continue);
1610
1611 // Build the base (index * per_case_size_in_bytes()) +
1612 // case_array_offset_in_bytes().
1613 z_sllg(index, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1614 add2reg(index, in_bytes(MultiBranchData::case_array_offset()));
1615
1616 // Add the calculated base to the mdp -> address of the case' data.
1617 Address case_data_addr(mdp, index);
1618 Register case_data = scratch1;
1619 load_address(case_data, case_data_addr);
1620
1621 // Update the case count.
1622 increment_mdp_data_at(case_data,
1623 in_bytes(MultiBranchData::relative_count_offset()),
1624 scratch2);
1625
1626 // The method data pointer needs to be updated.
1627 update_mdp_by_offset(mdp,
1628 index,
1629 in_bytes(MultiBranchData::relative_displacement_offset()));
1630
1631 bind(profile_continue);
1632 }
1633 }
1634
1635 // kills: R0, R1, flags, loads klass from obj (if not null)
profile_obj_type(Register obj,Address mdo_addr,Register klass,bool cmp_done)1636 void InterpreterMacroAssembler::profile_obj_type(Register obj, Address mdo_addr, Register klass, bool cmp_done) {
1637 NearLabel null_seen, init_klass, do_nothing, do_update;
1638
1639 // Klass = obj is allowed.
1640 const Register tmp = Z_R1;
1641 assert_different_registers(obj, mdo_addr.base(), tmp, Z_R0);
1642 assert_different_registers(klass, mdo_addr.base(), tmp, Z_R0);
1643
1644 z_lg(tmp, mdo_addr);
1645 if (cmp_done) {
1646 z_brz(null_seen);
1647 } else {
1648 compareU64_and_branch(obj, (intptr_t)0, Assembler::bcondEqual, null_seen);
1649 }
1650
1651 verify_oop(obj);
1652 load_klass(klass, obj);
1653
1654 // Klass seen before, nothing to do (regardless of unknown bit).
1655 z_lgr(Z_R0, tmp);
1656 assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
1657 z_nill(Z_R0, TypeEntries::type_klass_mask & 0xFFFF);
1658 compareU64_and_branch(Z_R0, klass, Assembler::bcondEqual, do_nothing);
1659
1660 // Already unknown. Nothing to do anymore.
1661 z_tmll(tmp, TypeEntries::type_unknown);
1662 z_brc(Assembler::bcondAllOne, do_nothing);
1663
1664 z_lgr(Z_R0, tmp);
1665 assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
1666 z_nill(Z_R0, TypeEntries::type_mask & 0xFFFF);
1667 compareU64_and_branch(Z_R0, (intptr_t)0, Assembler::bcondEqual, init_klass);
1668
1669 // Different than before. Cannot keep accurate profile.
1670 z_oill(tmp, TypeEntries::type_unknown);
1671 z_bru(do_update);
1672
1673 bind(init_klass);
1674 // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1675 z_ogr(tmp, klass);
1676 z_bru(do_update);
1677
1678 bind(null_seen);
1679 // Set null_seen if obj is 0.
1680 z_oill(tmp, TypeEntries::null_seen);
1681 // fallthru: z_bru(do_update);
1682
1683 bind(do_update);
1684 z_stg(tmp, mdo_addr);
1685
1686 bind(do_nothing);
1687 }
1688
profile_arguments_type(Register mdp,Register callee,Register tmp,bool is_virtual)1689 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1690 if (!ProfileInterpreter) {
1691 return;
1692 }
1693
1694 assert_different_registers(mdp, callee, tmp);
1695
1696 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1697 Label profile_continue;
1698
1699 test_method_data_pointer(mdp, profile_continue);
1700
1701 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1702
1703 z_cliy(in_bytes(DataLayout::tag_offset()) - off_to_start, mdp,
1704 is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1705 z_brne(profile_continue);
1706
1707 if (MethodData::profile_arguments()) {
1708 NearLabel done;
1709 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1710 add2reg(mdp, off_to_args);
1711
1712 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1713 if (i > 0 || MethodData::profile_return()) {
1714 // If return value type is profiled we may have no argument to profile.
1715 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1716 add2reg(tmp, -i*TypeStackSlotEntries::per_arg_count());
1717 compare64_and_branch(tmp, TypeStackSlotEntries::per_arg_count(), Assembler::bcondLow, done);
1718 }
1719 z_lg(tmp, Address(callee, Method::const_offset()));
1720 z_lgh(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1721 // Stack offset o (zero based) from the start of the argument
1722 // list. For n arguments translates into offset n - o - 1 from
1723 // the end of the argument list. But there is an extra slot at
1724 // the top of the stack. So the offset is n - o from Lesp.
1725 z_sg(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
1726 z_sllg(tmp, tmp, Interpreter::logStackElementSize);
1727 Address stack_slot_addr(tmp, Z_esp);
1728 z_ltg(tmp, stack_slot_addr);
1729
1730 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
1731 profile_obj_type(tmp, mdo_arg_addr, tmp, /*ltg did compare to 0*/ true);
1732
1733 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1734 add2reg(mdp, to_add);
1735 off_to_args += to_add;
1736 }
1737
1738 if (MethodData::profile_return()) {
1739 z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1740 add2reg(tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1741 }
1742
1743 bind(done);
1744
1745 if (MethodData::profile_return()) {
1746 // We're right after the type profile for the last
1747 // argument. Tmp is the number of cells left in the
1748 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1749 // if there's a return to profile.
1750 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1751 z_sllg(tmp, tmp, exact_log2(DataLayout::cell_size));
1752 z_agr(mdp, tmp);
1753 }
1754 z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
1755 } else {
1756 assert(MethodData::profile_return(), "either profile call args or call ret");
1757 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1758 }
1759
1760 // Mdp points right after the end of the
1761 // CallTypeData/VirtualCallTypeData, right after the cells for the
1762 // return value type if there's one.
1763 bind(profile_continue);
1764 }
1765 }
1766
profile_return_type(Register mdp,Register ret,Register tmp)1767 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1768 assert_different_registers(mdp, ret, tmp);
1769 if (ProfileInterpreter && MethodData::profile_return()) {
1770 Label profile_continue;
1771
1772 test_method_data_pointer(mdp, profile_continue);
1773
1774 if (MethodData::profile_return_jsr292_only()) {
1775 // If we don't profile all invoke bytecodes we must make sure
1776 // it's a bytecode we indeed profile. We can't go back to the
1777 // beginning of the ProfileData we intend to update to check its
1778 // type because we're right after it and we don't known its
1779 // length.
1780 NearLabel do_profile;
1781 Address bc(Z_bcp);
1782 z_lb(tmp, bc);
1783 compare32_and_branch(tmp, Bytecodes::_invokedynamic, Assembler::bcondEqual, do_profile);
1784 compare32_and_branch(tmp, Bytecodes::_invokehandle, Assembler::bcondEqual, do_profile);
1785 get_method(tmp);
1786 // Supplement to 8139891: _intrinsic_id exceeded 1-byte size limit.
1787 if (Method::intrinsic_id_size_in_bytes() == 1) {
1788 z_cli(Method::intrinsic_id_offset_in_bytes(), tmp, vmIntrinsics::_compiledLambdaForm);
1789 } else {
1790 assert(Method::intrinsic_id_size_in_bytes() == 2, "size error: check Method::_intrinsic_id");
1791 z_lh(tmp, Method::intrinsic_id_offset_in_bytes(), Z_R0, tmp);
1792 z_chi(tmp, vmIntrinsics::_compiledLambdaForm);
1793 }
1794 z_brne(profile_continue);
1795
1796 bind(do_profile);
1797 }
1798
1799 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1800 profile_obj_type(ret, mdo_ret_addr, tmp);
1801
1802 bind(profile_continue);
1803 }
1804 }
1805
profile_parameters_type(Register mdp,Register tmp1,Register tmp2)1806 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1807 if (ProfileInterpreter && MethodData::profile_parameters()) {
1808 Label profile_continue, done;
1809
1810 test_method_data_pointer(mdp, profile_continue);
1811
1812 // Load the offset of the area within the MDO used for
1813 // parameters. If it's negative we're not profiling any parameters.
1814 Address parm_di_addr(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()));
1815 load_and_test_int2long(tmp1, parm_di_addr);
1816 z_brl(profile_continue);
1817
1818 // Compute a pointer to the area for parameters from the offset
1819 // and move the pointer to the slot for the last
1820 // parameters. Collect profiling from last parameter down.
1821 // mdo start + parameters offset + array length - 1
1822
1823 // Pointer to the parameter area in the MDO.
1824 z_agr(mdp, tmp1);
1825
1826 // Offset of the current profile entry to update.
1827 const Register entry_offset = tmp1;
1828 // entry_offset = array len in number of cells.
1829 z_lg(entry_offset, Address(mdp, ArrayData::array_len_offset()));
1830 // entry_offset (number of cells) = array len - size of 1 entry
1831 add2reg(entry_offset, -TypeStackSlotEntries::per_arg_count());
1832 // entry_offset in bytes
1833 z_sllg(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1834
1835 Label loop;
1836 bind(loop);
1837
1838 Address arg_off(mdp, entry_offset, ParametersTypeData::stack_slot_offset(0));
1839 Address arg_type(mdp, entry_offset, ParametersTypeData::type_offset(0));
1840
1841 // Load offset on the stack from the slot for this parameter.
1842 z_lg(tmp2, arg_off);
1843 z_sllg(tmp2, tmp2, Interpreter::logStackElementSize);
1844 z_lcgr(tmp2); // Negate.
1845
1846 // Profile the parameter.
1847 z_ltg(tmp2, Address(Z_locals, tmp2));
1848 profile_obj_type(tmp2, arg_type, tmp2, /*ltg did compare to 0*/ true);
1849
1850 // Go to next parameter.
1851 z_aghi(entry_offset, -TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size);
1852 z_brnl(loop);
1853
1854 bind(profile_continue);
1855 }
1856 }
1857
1858 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
increment_mask_and_jump(Address counter_addr,int increment,Address mask,Register scratch,bool preloaded,branch_condition cond,Label * where)1859 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1860 int increment,
1861 Address mask,
1862 Register scratch,
1863 bool preloaded,
1864 branch_condition cond,
1865 Label *where) {
1866 assert_different_registers(counter_addr.base(), scratch);
1867 if (preloaded) {
1868 add2reg(scratch, increment);
1869 reg2mem_opt(scratch, counter_addr, false);
1870 } else {
1871 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment) && counter_addr.is_RSYform()) {
1872 z_alsi(counter_addr.disp20(), counter_addr.base(), increment);
1873 mem2reg_signed_opt(scratch, counter_addr);
1874 } else {
1875 mem2reg_signed_opt(scratch, counter_addr);
1876 add2reg(scratch, increment);
1877 reg2mem_opt(scratch, counter_addr, false);
1878 }
1879 }
1880 z_n(scratch, mask);
1881 if (where) { z_brc(cond, *where); }
1882 }
1883
1884 // Get MethodCounters object for given method. Lazily allocated if necessary.
1885 // method - Ptr to Method object.
1886 // Rcounters - Ptr to MethodCounters object associated with Method object.
1887 // skip - Exit point if MethodCounters object can't be created (OOM condition).
get_method_counters(Register Rmethod,Register Rcounters,Label & skip)1888 void InterpreterMacroAssembler::get_method_counters(Register Rmethod,
1889 Register Rcounters,
1890 Label& skip) {
1891 assert_different_registers(Rmethod, Rcounters);
1892
1893 BLOCK_COMMENT("get MethodCounters object {");
1894
1895 Label has_counters;
1896 load_and_test_long(Rcounters, Address(Rmethod, Method::method_counters_offset()));
1897 z_brnz(has_counters);
1898
1899 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), Rmethod);
1900 z_ltgr(Rcounters, Z_RET); // Runtime call returns MethodCounters object.
1901 z_brz(skip); // No MethodCounters, out of memory.
1902
1903 bind(has_counters);
1904
1905 BLOCK_COMMENT("} get MethodCounters object");
1906 }
1907
1908 // Increment invocation counter in MethodCounters object.
1909 // Return (invocation_counter+backedge_counter) as "result" in RctrSum.
1910 // Counter values are all unsigned.
increment_invocation_counter(Register Rcounters,Register RctrSum)1911 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register RctrSum) {
1912 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful");
1913 assert_different_registers(Rcounters, RctrSum);
1914
1915 int increment = InvocationCounter::count_increment;
1916 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
1917 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset());
1918
1919 BLOCK_COMMENT("Increment invocation counter {");
1920
1921 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
1922 // Increment the invocation counter in place,
1923 // then add the incremented value to the backedge counter.
1924 z_l(RctrSum, be_counter_offset, Rcounters);
1925 z_alsi(inv_counter_offset, Rcounters, increment); // Atomic increment @no extra cost!
1926 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
1927 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
1928 } else {
1929 // This path is optimized for low register consumption
1930 // at the cost of somewhat higher operand delays.
1931 // It does not need an extra temp register.
1932
1933 // Update the invocation counter.
1934 z_l(RctrSum, inv_counter_offset, Rcounters);
1935 if (RctrSum == Z_R0) {
1936 z_ahi(RctrSum, increment);
1937 } else {
1938 add2reg(RctrSum, increment);
1939 }
1940 z_st(RctrSum, inv_counter_offset, Rcounters);
1941
1942 // Mask off the state bits.
1943 z_nilf(RctrSum, InvocationCounter::count_mask_value);
1944
1945 // Add the backedge counter to the updated invocation counter to
1946 // form the result.
1947 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
1948 }
1949
1950 BLOCK_COMMENT("} Increment invocation counter");
1951
1952 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
1953 }
1954
1955
1956 // increment backedge counter in MethodCounters object.
1957 // return (invocation_counter+backedge_counter) as "result" in RctrSum
1958 // counter values are all unsigned!
increment_backedge_counter(Register Rcounters,Register RctrSum)1959 void InterpreterMacroAssembler::increment_backedge_counter(Register Rcounters, Register RctrSum) {
1960 assert(UseCompiler, "incrementing must be useful");
1961 assert_different_registers(Rcounters, RctrSum);
1962
1963 int increment = InvocationCounter::count_increment;
1964 int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
1965 int be_counter_offset = in_bytes(MethodCounters::backedge_counter_offset() + InvocationCounter::counter_offset());
1966
1967 BLOCK_COMMENT("Increment backedge counter {");
1968
1969 if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
1970 // Increment the invocation counter in place,
1971 // then add the incremented value to the backedge counter.
1972 z_l(RctrSum, inv_counter_offset, Rcounters);
1973 z_alsi(be_counter_offset, Rcounters, increment); // Atomic increment @no extra cost!
1974 z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
1975 z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
1976 } else {
1977 // This path is optimized for low register consumption
1978 // at the cost of somewhat higher operand delays.
1979 // It does not need an extra temp register.
1980
1981 // Update the invocation counter.
1982 z_l(RctrSum, be_counter_offset, Rcounters);
1983 if (RctrSum == Z_R0) {
1984 z_ahi(RctrSum, increment);
1985 } else {
1986 add2reg(RctrSum, increment);
1987 }
1988 z_st(RctrSum, be_counter_offset, Rcounters);
1989
1990 // Mask off the state bits.
1991 z_nilf(RctrSum, InvocationCounter::count_mask_value);
1992
1993 // Add the backedge counter to the updated invocation counter to
1994 // form the result.
1995 z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
1996 }
1997
1998 BLOCK_COMMENT("} Increment backedge counter");
1999
2000 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2001 }
2002
2003 // Add an InterpMonitorElem to stack (see frame_s390.hpp).
add_monitor_to_stack(bool stack_is_empty,Register Rtemp1,Register Rtemp2,Register Rtemp3)2004 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty,
2005 Register Rtemp1,
2006 Register Rtemp2,
2007 Register Rtemp3) {
2008
2009 const Register Rcurr_slot = Rtemp1;
2010 const Register Rlimit = Rtemp2;
2011 const jint delta = -frame::interpreter_frame_monitor_size() * wordSize;
2012
2013 assert((delta & LongAlignmentMask) == 0,
2014 "sizeof BasicObjectLock must be even number of doublewords");
2015 assert(2 * wordSize == -delta, "this works only as long as delta == -2*wordSize");
2016 assert(Rcurr_slot != Z_R0, "Register must be usable as base register");
2017 assert_different_registers(Rlimit, Rcurr_slot, Rtemp3);
2018
2019 get_monitors(Rlimit);
2020
2021 // Adjust stack pointer for additional monitor entry.
2022 resize_frame(RegisterOrConstant((intptr_t) delta), Z_fp, false);
2023
2024 if (!stack_is_empty) {
2025 // Must copy stack contents down.
2026 NearLabel next, done;
2027
2028 // Rtemp := addr(Tos), Z_esp is pointing below it!
2029 add2reg(Rcurr_slot, wordSize, Z_esp);
2030
2031 // Nothing to do, if already at monitor area.
2032 compareU64_and_branch(Rcurr_slot, Rlimit, bcondNotLow, done);
2033
2034 bind(next);
2035
2036 // Move one stack slot.
2037 mem2reg_opt(Rtemp3, Address(Rcurr_slot));
2038 reg2mem_opt(Rtemp3, Address(Rcurr_slot, delta));
2039 add2reg(Rcurr_slot, wordSize);
2040 compareU64_and_branch(Rcurr_slot, Rlimit, bcondLow, next); // Are we done?
2041
2042 bind(done);
2043 // Done copying stack.
2044 }
2045
2046 // Adjust expression stack and monitor pointers.
2047 add2reg(Z_esp, delta);
2048 add2reg(Rlimit, delta);
2049 save_monitors(Rlimit);
2050 }
2051
2052 // Note: Index holds the offset in bytes afterwards.
2053 // You can use this to store a new value (with Llocals as the base).
access_local_int(Register index,Register dst)2054 void InterpreterMacroAssembler::access_local_int(Register index, Register dst) {
2055 z_sllg(index, index, LogBytesPerWord);
2056 mem2reg_opt(dst, Address(Z_locals, index), false);
2057 }
2058
verify_oop(Register reg,TosState state)2059 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2060 if (state == atos) { MacroAssembler::verify_oop(reg); }
2061 }
2062
2063 // Inline assembly for:
2064 //
2065 // if (thread is in interp_only_mode) {
2066 // InterpreterRuntime::post_method_entry();
2067 // }
2068
notify_method_entry()2069 void InterpreterMacroAssembler::notify_method_entry() {
2070
2071 // JVMTI
2072 // Whenever JVMTI puts a thread in interp_only_mode, method
2073 // entry/exit events are sent for that thread to track stack
2074 // depth. If it is possible to enter interp_only_mode we add
2075 // the code to check if the event should be sent.
2076 if (JvmtiExport::can_post_interpreter_events()) {
2077 Label jvmti_post_done;
2078 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2079 z_bre(jvmti_post_done);
2080 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2081 bind(jvmti_post_done);
2082 }
2083 }
2084
2085 // Inline assembly for:
2086 //
2087 // if (thread is in interp_only_mode) {
2088 // if (!native_method) save result
2089 // InterpreterRuntime::post_method_exit();
2090 // if (!native_method) restore result
2091 // }
2092 // if (DTraceMethodProbes) {
2093 // SharedRuntime::dtrace_method_exit(thread, method);
2094 // }
2095 //
2096 // For native methods their result is stored in z_ijava_state.lresult
2097 // and z_ijava_state.fresult before coming here.
2098 // Java methods have their result stored in the expression stack.
2099 //
2100 // Notice the dependency to frame::interpreter_frame_result().
notify_method_exit(bool native_method,TosState state,NotifyMethodExitMode mode)2101 void InterpreterMacroAssembler::notify_method_exit(bool native_method,
2102 TosState state,
2103 NotifyMethodExitMode mode) {
2104 // JVMTI
2105 // Whenever JVMTI puts a thread in interp_only_mode, method
2106 // entry/exit events are sent for that thread to track stack
2107 // depth. If it is possible to enter interp_only_mode we add
2108 // the code to check if the event should be sent.
2109 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2110 Label jvmti_post_done;
2111 MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2112 z_bre(jvmti_post_done);
2113 if (!native_method) push(state); // see frame::interpreter_frame_result()
2114 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2115 if (!native_method) pop(state);
2116 bind(jvmti_post_done);
2117 }
2118
2119 #if 0
2120 // Dtrace currently not supported on z/Architecture.
2121 {
2122 SkipIfEqual skip(this, &DTraceMethodProbes, false);
2123 push(state);
2124 get_method(c_rarg1);
2125 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2126 r15_thread, c_rarg1);
2127 pop(state);
2128 }
2129 #endif
2130 }
2131
skip_if_jvmti_mode(Label & Lskip,Register Rscratch)2132 void InterpreterMacroAssembler::skip_if_jvmti_mode(Label &Lskip, Register Rscratch) {
2133 if (!JvmtiExport::can_post_interpreter_events()) {
2134 return;
2135 }
2136
2137 load_and_test_int(Rscratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2138 z_brnz(Lskip);
2139
2140 }
2141
2142 // Pop the topmost TOP_IJAVA_FRAME and set it's sender_sp as new Z_SP.
2143 // The return pc is loaded into the register return_pc.
2144 //
2145 // Registers updated:
2146 // return_pc - The return pc of the calling frame.
2147 // tmp1, tmp2 - scratch
pop_interpreter_frame(Register return_pc,Register tmp1,Register tmp2)2148 void InterpreterMacroAssembler::pop_interpreter_frame(Register return_pc, Register tmp1, Register tmp2) {
2149 // F0 Z_SP -> caller_sp (F1's)
2150 // ...
2151 // sender_sp (F1's)
2152 // ...
2153 // F1 Z_fp -> caller_sp (F2's)
2154 // return_pc (Continuation after return from F0.)
2155 // ...
2156 // F2 caller_sp
2157
2158 // Remove F0's activation. Restoring Z_SP to sender_sp reverts modifications
2159 // (a) by a c2i adapter and (b) by generate_fixed_frame().
2160 // In case (a) the new top frame F1 is an unextended compiled frame.
2161 // In case (b) F1 is converted from PARENT_IJAVA_FRAME to TOP_IJAVA_FRAME.
2162
2163 // Case (b) seems to be redundant when returning to a interpreted caller,
2164 // because then the caller's top_frame_sp is installed as sp (see
2165 // TemplateInterpreterGenerator::generate_return_entry_for ()). But
2166 // pop_interpreter_frame() is also used in exception handling and there the
2167 // frame type of the caller is unknown, therefore top_frame_sp cannot be used,
2168 // so it is important that sender_sp is the caller's sp as TOP_IJAVA_FRAME.
2169
2170 Register R_f1_sender_sp = tmp1;
2171 Register R_f2_sp = tmp2;
2172
2173 // Tirst check the for the interpreter frame's magic.
2174 asm_assert_ijava_state_magic(R_f2_sp/*tmp*/);
2175 z_lg(R_f2_sp, _z_parent_ijava_frame_abi(callers_sp), Z_fp);
2176 z_lg(R_f1_sender_sp, _z_ijava_state_neg(sender_sp), Z_fp);
2177 if (return_pc->is_valid())
2178 z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2179 // Pop F0 by resizing to R_f1_sender_sp and using R_f2_sp as fp.
2180 resize_frame_absolute(R_f1_sender_sp, R_f2_sp, false/*load fp*/);
2181
2182 #ifdef ASSERT
2183 // The return_pc in the new top frame is dead... at least that's my
2184 // current understanding; to assert this I overwrite it.
2185 load_const_optimized(Z_ARG3, 0xb00b1);
2186 z_stg(Z_ARG3, _z_parent_ijava_frame_abi(return_pc), Z_SP);
2187 #endif
2188 }
2189
verify_FPU(int stack_depth,TosState state)2190 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2191 if (VerifyFPU) {
2192 unimplemented("verfiyFPU");
2193 }
2194 }
2195