1 /*
2 * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.inline.hpp"
27 #include "interp_masm_sparc.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "logging/log.hpp"
31 #include "oops/arrayOop.hpp"
32 #include "oops/markOop.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/methodCounters.hpp"
36 #include "prims/jvmtiExport.hpp"
37 #include "prims/jvmtiThreadState.hpp"
38 #include "runtime/basicLock.hpp"
39 #include "runtime/biasedLocking.hpp"
40 #include "runtime/frame.inline.hpp"
41 #include "runtime/safepointMechanism.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/thread.inline.hpp"
44 #include "utilities/align.hpp"
45
46 // Implementation of InterpreterMacroAssembler
47
48 // This file specializes the assember with interpreter-specific macros
49
50 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS);
51 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS);
52
jump_to_entry(address entry)53 void InterpreterMacroAssembler::jump_to_entry(address entry) {
54 assert(entry, "Entry must have been generated by now");
55 AddressLiteral al(entry);
56 jump_to(al, G3_scratch);
57 delayed()->nop();
58 }
59
compute_extra_locals_size_in_bytes(Register args_size,Register locals_size,Register delta)60 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) {
61 // Note: this algorithm is also used by C1's OSR entry sequence.
62 // Any changes should also be applied to CodeEmitter::emit_osr_entry().
63 assert_different_registers(args_size, locals_size);
64 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted.
65 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words
66 // Use br/mov combination because it works on both V8 and V9 and is
67 // faster.
68 Label skip_move;
69 br(Assembler::negative, true, Assembler::pt, skip_move);
70 delayed()->mov(G0, delta);
71 bind(skip_move);
72 align_up(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned)
73 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes
74 }
75
76 // Dispatch code executed in the prolog of a bytecode which does not do it's
77 // own dispatch. The dispatch address is computed and placed in IdispatchAddress
dispatch_prolog(TosState state,int bcp_incr)78 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
79 assert_not_delayed();
80 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
81 // dispatch table to use
82 AddressLiteral tbl(Interpreter::dispatch_table(state));
83 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
84 set(tbl, G3_scratch); // compute addr of table
85 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr
86 }
87
88
89 // Dispatch code executed in the epilog of a bytecode which does not do it's
90 // own dispatch. The dispatch address in IdispatchAddress is used for the
91 // dispatch.
dispatch_epilog(TosState state,int bcp_incr)92 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
93 assert_not_delayed();
94 verify_FPU(1, state);
95 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
96 jmp( IdispatchAddress, 0 );
97 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr);
98 else delayed()->nop();
99 }
100
dispatch_next(TosState state,int bcp_incr,bool generate_poll)101 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) {
102 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
103 assert_not_delayed();
104 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
105 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, true, generate_poll);
106 }
107
108
dispatch_next_noverify_oop(TosState state,int bcp_incr)109 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) {
110 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
111 assert_not_delayed();
112 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
113 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false);
114 }
115
116
dispatch_via(TosState state,address * table)117 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
118 // load current bytecode
119 assert_not_delayed();
120 ldub( Lbcp, 0, Lbyte_code); // load next bytecode
121 dispatch_base(state, table);
122 }
123
124
call_VM_leaf_base(Register java_thread,address entry_point,int number_of_arguments)125 void InterpreterMacroAssembler::call_VM_leaf_base(
126 Register java_thread,
127 address entry_point,
128 int number_of_arguments
129 ) {
130 if (!java_thread->is_valid())
131 java_thread = L7_thread_cache;
132 // super call
133 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments);
134 }
135
136
call_VM_base(Register oop_result,Register java_thread,Register last_java_sp,address entry_point,int number_of_arguments,bool check_exception)137 void InterpreterMacroAssembler::call_VM_base(
138 Register oop_result,
139 Register java_thread,
140 Register last_java_sp,
141 address entry_point,
142 int number_of_arguments,
143 bool check_exception
144 ) {
145 if (!java_thread->is_valid())
146 java_thread = L7_thread_cache;
147 // See class ThreadInVMfromInterpreter, which assumes that the interpreter
148 // takes responsibility for setting its own thread-state on call-out.
149 // However, ThreadInVMfromInterpreter resets the state to "in_Java".
150
151 //save_bcp(); // save bcp
152 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception);
153 //restore_bcp(); // restore bcp
154 //restore_locals(); // restore locals pointer
155 }
156
157
check_and_handle_popframe(Register scratch_reg)158 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
159 if (JvmtiExport::can_pop_frame()) {
160 Label L;
161
162 // Check the "pending popframe condition" flag in the current thread
163 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg);
164
165 // Initiate popframe handling only if it is not already being processed. If the flag
166 // has the popframe_processing bit set, it means that this code is called *during* popframe
167 // handling - we don't want to reenter.
168 btst(JavaThread::popframe_pending_bit, scratch_reg);
169 br(zero, false, pt, L);
170 delayed()->nop();
171 btst(JavaThread::popframe_processing_bit, scratch_reg);
172 br(notZero, false, pt, L);
173 delayed()->nop();
174
175 // Call Interpreter::remove_activation_preserving_args_entry() to get the
176 // address of the same-named entrypoint in the generated interpreter code.
177 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
178
179 // Jump to Interpreter::_remove_activation_preserving_args_entry
180 jmpl(O0, G0, G0);
181 delayed()->nop();
182 bind(L);
183 }
184 }
185
186
load_earlyret_value(TosState state)187 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
188 Register thr_state = G4_scratch;
189 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
190 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset());
191 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset());
192 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset());
193 switch (state) {
194 case ltos: ld_long(val_addr, Otos_l); break;
195 case atos: ld_ptr(oop_addr, Otos_l);
196 st_ptr(G0, oop_addr); break;
197 case btos: // fall through
198 case ztos: // fall through
199 case ctos: // fall through
200 case stos: // fall through
201 case itos: ld(val_addr, Otos_l1); break;
202 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break;
203 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break;
204 case vtos: /* nothing to do */ break;
205 default : ShouldNotReachHere();
206 }
207 // Clean up tos value in the jvmti thread state
208 or3(G0, ilgl, G3_scratch);
209 stw(G3_scratch, tos_addr);
210 st_long(G0, val_addr);
211 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
212 }
213
214
check_and_handle_earlyret(Register scratch_reg)215 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
216 if (JvmtiExport::can_force_early_return()) {
217 Label L;
218 Register thr_state = G3_scratch;
219 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
220 br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit;
221
222 // Initiate earlyret handling only if it is not already being processed.
223 // If the flag has the earlyret_processing bit set, it means that this code
224 // is called *during* earlyret handling - we don't want to reenter.
225 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch);
226 cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L);
227
228 // Call Interpreter::remove_activation_early_entry() to get the address of the
229 // same-named entrypoint in the generated interpreter code
230 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1);
231 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1);
232
233 // Jump to Interpreter::_remove_activation_early_entry
234 jmpl(O0, G0, G0);
235 delayed()->nop();
236 bind(L);
237 }
238 }
239
240
super_call_VM_leaf(Register thread_cache,address entry_point,Register arg_1,Register arg_2)241 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) {
242 mov(arg_1, O0);
243 mov(arg_2, O1);
244 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2);
245 }
246
dispatch_base(TosState state,address * table)247 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) {
248 assert_not_delayed();
249 dispatch_Lbyte_code(state, table);
250 }
251
252
dispatch_normal(TosState state)253 void InterpreterMacroAssembler::dispatch_normal(TosState state) {
254 dispatch_base(state, Interpreter::normal_table(state));
255 }
256
257
dispatch_only(TosState state)258 void InterpreterMacroAssembler::dispatch_only(TosState state) {
259 dispatch_base(state, Interpreter::dispatch_table(state));
260 }
261
262
263 // common code to dispatch and dispatch_only
264 // dispatch value in Lbyte_code and increment Lbcp
265
dispatch_Lbyte_code(TosState state,address * table,int bcp_incr,bool verify,bool generate_poll)266 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify, bool generate_poll) {
267 verify_FPU(1, state);
268 // %%%%% maybe implement +VerifyActivationFrameSize here
269 //verify_thread(); //too slow; we will just verify on method entry & exit
270 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
271 // dispatch table to use
272 AddressLiteral tbl(table);
273 Label dispatch;
274
275 if (SafepointMechanism::uses_thread_local_poll() && generate_poll) {
276 AddressLiteral sfpt_tbl(Interpreter::safept_table(state));
277 Label no_safepoint;
278
279 if (tbl.value() != sfpt_tbl.value()) {
280 ldx(Address(G2_thread, Thread::polling_page_offset()), G3_scratch, 0);
281 // Armed page has poll_bit set, if poll bit is cleared just continue.
282 and3(G3_scratch, SafepointMechanism::poll_bit(), G3_scratch);
283
284 br_null_short(G3_scratch, Assembler::pt, no_safepoint);
285 set(sfpt_tbl, G3_scratch);
286 ba_short(dispatch);
287 }
288 bind(no_safepoint);
289 }
290
291 set(tbl, G3_scratch); // compute addr of table
292 bind(dispatch);
293 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
294 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr
295 jmp( G3_scratch, 0 );
296 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr);
297 else delayed()->nop();
298 }
299
300
301 // Helpers for expression stack
302
303 // Longs and doubles are Category 2 computational types in the
304 // JVM specification (section 3.11.1) and take 2 expression stack or
305 // local slots.
306 // Aligning them on 32 bit with tagged stacks is hard because the code generated
307 // for the dup* bytecodes depends on what types are already on the stack.
308 // If the types are split into the two stack/local slots, that is much easier
309 // (and we can use 0 for non-reference tags).
310
311 // Known good alignment in _LP64 but unknown otherwise
load_unaligned_double(Register r1,int offset,FloatRegister d)312 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) {
313 assert_not_delayed();
314
315 ldf(FloatRegisterImpl::D, r1, offset, d);
316 }
317
318 // Known good alignment in _LP64 but unknown otherwise
store_unaligned_double(FloatRegister d,Register r1,int offset)319 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) {
320 assert_not_delayed();
321
322 stf(FloatRegisterImpl::D, d, r1, offset);
323 // store something more useful here
324 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);)
325 }
326
327
328 // Known good alignment in _LP64 but unknown otherwise
load_unaligned_long(Register r1,int offset,Register rd)329 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) {
330 assert_not_delayed();
331 ldx(r1, offset, rd);
332 }
333
334 // Known good alignment in _LP64 but unknown otherwise
store_unaligned_long(Register l,Register r1,int offset)335 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) {
336 assert_not_delayed();
337
338 stx(l, r1, offset);
339 // store something more useful here
340 stx(G0, r1, offset+Interpreter::stackElementSize);
341 }
342
pop_i(Register r)343 void InterpreterMacroAssembler::pop_i(Register r) {
344 assert_not_delayed();
345 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r);
346 inc(Lesp, Interpreter::stackElementSize);
347 debug_only(verify_esp(Lesp));
348 }
349
pop_ptr(Register r,Register scratch)350 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) {
351 assert_not_delayed();
352 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r);
353 inc(Lesp, Interpreter::stackElementSize);
354 debug_only(verify_esp(Lesp));
355 }
356
pop_l(Register r)357 void InterpreterMacroAssembler::pop_l(Register r) {
358 assert_not_delayed();
359 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r);
360 inc(Lesp, 2*Interpreter::stackElementSize);
361 debug_only(verify_esp(Lesp));
362 }
363
364
pop_f(FloatRegister f,Register scratch)365 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) {
366 assert_not_delayed();
367 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f);
368 inc(Lesp, Interpreter::stackElementSize);
369 debug_only(verify_esp(Lesp));
370 }
371
372
pop_d(FloatRegister f,Register scratch)373 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) {
374 assert_not_delayed();
375 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f);
376 inc(Lesp, 2*Interpreter::stackElementSize);
377 debug_only(verify_esp(Lesp));
378 }
379
380
push_i(Register r)381 void InterpreterMacroAssembler::push_i(Register r) {
382 assert_not_delayed();
383 debug_only(verify_esp(Lesp));
384 st(r, Lesp, 0);
385 dec(Lesp, Interpreter::stackElementSize);
386 }
387
push_ptr(Register r)388 void InterpreterMacroAssembler::push_ptr(Register r) {
389 assert_not_delayed();
390 st_ptr(r, Lesp, 0);
391 dec(Lesp, Interpreter::stackElementSize);
392 }
393
394 // remember: our convention for longs in SPARC is:
395 // O0 (Otos_l1) has high-order part in first word,
396 // O1 (Otos_l2) has low-order part in second word
397
push_l(Register r)398 void InterpreterMacroAssembler::push_l(Register r) {
399 assert_not_delayed();
400 debug_only(verify_esp(Lesp));
401 // Longs are stored in memory-correct order, even if unaligned.
402 int offset = -Interpreter::stackElementSize;
403 store_unaligned_long(r, Lesp, offset);
404 dec(Lesp, 2 * Interpreter::stackElementSize);
405 }
406
407
push_f(FloatRegister f)408 void InterpreterMacroAssembler::push_f(FloatRegister f) {
409 assert_not_delayed();
410 debug_only(verify_esp(Lesp));
411 stf(FloatRegisterImpl::S, f, Lesp, 0);
412 dec(Lesp, Interpreter::stackElementSize);
413 }
414
415
push_d(FloatRegister d)416 void InterpreterMacroAssembler::push_d(FloatRegister d) {
417 assert_not_delayed();
418 debug_only(verify_esp(Lesp));
419 // Longs are stored in memory-correct order, even if unaligned.
420 int offset = -Interpreter::stackElementSize;
421 store_unaligned_double(d, Lesp, offset);
422 dec(Lesp, 2 * Interpreter::stackElementSize);
423 }
424
425
push(TosState state)426 void InterpreterMacroAssembler::push(TosState state) {
427 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
428 switch (state) {
429 case atos: push_ptr(); break;
430 case btos: // fall through
431 case ztos: // fall through
432 case ctos: // fall through
433 case stos: // fall through
434 case itos: push_i(); break;
435 case ltos: push_l(); break;
436 case ftos: push_f(); break;
437 case dtos: push_d(); break;
438 case vtos: /* nothing to do */ break;
439 default : ShouldNotReachHere();
440 }
441 }
442
443
pop(TosState state)444 void InterpreterMacroAssembler::pop(TosState state) {
445 switch (state) {
446 case atos: pop_ptr(); break;
447 case btos: // fall through
448 case ztos: // fall through
449 case ctos: // fall through
450 case stos: // fall through
451 case itos: pop_i(); break;
452 case ltos: pop_l(); break;
453 case ftos: pop_f(); break;
454 case dtos: pop_d(); break;
455 case vtos: /* nothing to do */ break;
456 default : ShouldNotReachHere();
457 }
458 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
459 }
460
461
462 // Helpers for swap and dup
load_ptr(int n,Register val)463 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
464 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val);
465 }
store_ptr(int n,Register val)466 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
467 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n));
468 }
469
470
load_receiver(Register param_count,Register recv)471 void InterpreterMacroAssembler::load_receiver(Register param_count,
472 Register recv) {
473 sll(param_count, Interpreter::logStackElementSize, param_count);
474 ld_ptr(Lesp, param_count, recv); // gets receiver oop
475 }
476
empty_expression_stack()477 void InterpreterMacroAssembler::empty_expression_stack() {
478 // Reset Lesp.
479 sub( Lmonitors, wordSize, Lesp );
480
481 // Reset SP by subtracting more space from Lesp.
482 Label done;
483 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!");
484
485 // A native does not need to do this, since its callee does not change SP.
486 ld(Lmethod, Method::access_flags_offset(), Gframe_size); // Load access flags.
487 btst(JVM_ACC_NATIVE, Gframe_size);
488 br(Assembler::notZero, false, Assembler::pt, done);
489 delayed()->nop();
490
491 // Compute max expression stack+register save area
492 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Gframe_size);
493 lduh(Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size); // Load max stack.
494 add(Gframe_size, frame::memory_parameter_word_sp_offset+Method::extra_stack_entries(), Gframe_size );
495
496 //
497 // now set up a stack frame with the size computed above
498 //
499 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below
500 sll( Gframe_size, LogBytesPerWord, Gframe_size );
501 sub( Lesp, Gframe_size, Gframe_size );
502 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary
503 debug_only(verify_sp(Gframe_size, G4_scratch));
504 sub(Gframe_size, STACK_BIAS, Gframe_size );
505 mov(Gframe_size, SP);
506
507 bind(done);
508 }
509
510
511 #ifdef ASSERT
verify_sp(Register Rsp,Register Rtemp)512 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) {
513 Label Bad, OK;
514
515 // Saved SP must be aligned.
516 btst(2*BytesPerWord-1, Rsp);
517 br(Assembler::notZero, false, Assembler::pn, Bad);
518 delayed()->nop();
519
520 // Saved SP, plus register window size, must not be above FP.
521 add(Rsp, frame::register_save_words * wordSize, Rtemp);
522 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP
523 cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad);
524
525 // Saved SP must not be ridiculously below current SP.
526 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K);
527 set(maxstack, Rtemp);
528 sub(SP, Rtemp, Rtemp);
529 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp
530 cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad);
531
532 ba_short(OK);
533
534 bind(Bad);
535 stop("on return to interpreted call, restored SP is corrupted");
536
537 bind(OK);
538 }
539
540
verify_esp(Register Resp)541 void InterpreterMacroAssembler::verify_esp(Register Resp) {
542 // about to read or write Resp[0]
543 // make sure it is not in the monitors or the register save area
544 Label OK1, OK2;
545
546 cmp(Resp, Lmonitors);
547 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1);
548 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp);
549 stop("too many pops: Lesp points into monitor area");
550 bind(OK1);
551 sub(Resp, STACK_BIAS, Resp);
552 cmp(Resp, SP);
553 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2);
554 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp);
555 stop("too many pushes: Lesp points into register window");
556 bind(OK2);
557 }
558 #endif // ASSERT
559
560 // Load compiled (i2c) or interpreter entry when calling from interpreted and
561 // do the call. Centralized so that all interpreter calls will do the same actions.
562 // If jvmti single stepping is on for a thread we must not call compiled code.
call_from_interpreter(Register target,Register scratch,Register Rret)563 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) {
564
565 // Assume we want to go compiled if available
566
567 ld_ptr(G5_method, in_bytes(Method::from_interpreted_offset()), target);
568
569 if (JvmtiExport::can_post_interpreter_events()) {
570 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
571 // compiled code in threads for which the event is enabled. Check here for
572 // interp_only_mode if these events CAN be enabled.
573 verify_thread();
574 Label skip_compiled_code;
575
576 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
577 ld(interp_only, scratch);
578 cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn);
579 delayed()->ld_ptr(G5_method, in_bytes(Method::interpreter_entry_offset()), target);
580 bind(skip_compiled_code);
581 }
582
583 // the i2c_adapters need Method* in G5_method (right? %%%)
584 // do the call
585 #ifdef ASSERT
586 {
587 Label ok;
588 br_notnull_short(target, Assembler::pt, ok);
589 stop("null entry point");
590 bind(ok);
591 }
592 #endif // ASSERT
593
594 // Adjust Rret first so Llast_SP can be same as Rret
595 add(Rret, -frame::pc_return_offset, O7);
596 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer
597 // Record SP so we can remove any stack space allocated by adapter transition
598 jmp(target, 0);
599 delayed()->mov(SP, Llast_SP);
600 }
601
if_cmp(Condition cc,bool ptr_compare)602 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) {
603 assert_not_delayed();
604
605 Label not_taken;
606 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken);
607 else br (cc, false, Assembler::pn, not_taken);
608 delayed()->nop();
609
610 TemplateTable::branch(false,false);
611
612 bind(not_taken);
613
614 profile_not_taken_branch(G3_scratch);
615 }
616
617
get_2_byte_integer_at_bcp(int bcp_offset,Register Rtmp,Register Rdst,signedOrNot is_signed,setCCOrNot should_set_CC)618 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(
619 int bcp_offset,
620 Register Rtmp,
621 Register Rdst,
622 signedOrNot is_signed,
623 setCCOrNot should_set_CC ) {
624 assert(Rtmp != Rdst, "need separate temp register");
625 assert_not_delayed();
626 switch (is_signed) {
627 default: ShouldNotReachHere();
628
629 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte
630 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte
631 }
632 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte
633 sll( Rdst, BitsPerByte, Rdst);
634 switch (should_set_CC ) {
635 default: ShouldNotReachHere();
636
637 case set_CC: orcc( Rdst, Rtmp, Rdst ); break;
638 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break;
639 }
640 }
641
642
get_4_byte_integer_at_bcp(int bcp_offset,Register Rtmp,Register Rdst,setCCOrNot should_set_CC)643 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(
644 int bcp_offset,
645 Register Rtmp,
646 Register Rdst,
647 setCCOrNot should_set_CC ) {
648 assert(Rtmp != Rdst, "need separate temp register");
649 assert_not_delayed();
650 add( Lbcp, bcp_offset, Rtmp);
651 andcc( Rtmp, 3, G0);
652 Label aligned;
653 switch (should_set_CC ) {
654 default: ShouldNotReachHere();
655
656 case set_CC: break;
657 case dont_set_CC: break;
658 }
659
660 br(Assembler::zero, true, Assembler::pn, aligned);
661 delayed()->ldsw(Rtmp, 0, Rdst);
662
663 ldub(Lbcp, bcp_offset + 3, Rdst);
664 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst);
665 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst);
666 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp);
667 or3(Rtmp, Rdst, Rdst );
668
669 bind(aligned);
670 if (should_set_CC == set_CC) tst(Rdst);
671 }
672
get_cache_index_at_bcp(Register temp,Register index,int bcp_offset,size_t index_size)673 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register temp, Register index,
674 int bcp_offset, size_t index_size) {
675 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
676 if (index_size == sizeof(u2)) {
677 get_2_byte_integer_at_bcp(bcp_offset, temp, index, Unsigned);
678 } else if (index_size == sizeof(u4)) {
679 get_4_byte_integer_at_bcp(bcp_offset, temp, index);
680 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
681 xor3(index, -1, index); // convert to plain index
682 } else if (index_size == sizeof(u1)) {
683 ldub(Lbcp, bcp_offset, index);
684 } else {
685 ShouldNotReachHere();
686 }
687 }
688
689
get_cache_and_index_at_bcp(Register cache,Register tmp,int bcp_offset,size_t index_size)690 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp,
691 int bcp_offset, size_t index_size) {
692 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
693 assert_different_registers(cache, tmp);
694 assert_not_delayed();
695 get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size);
696 // convert from field index to ConstantPoolCacheEntry index and from
697 // word index to byte offset
698 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
699 add(LcpoolCache, tmp, cache);
700 }
701
702
get_cache_and_index_and_bytecode_at_bcp(Register cache,Register temp,Register bytecode,int byte_no,int bcp_offset,size_t index_size)703 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
704 Register temp,
705 Register bytecode,
706 int byte_no,
707 int bcp_offset,
708 size_t index_size) {
709 get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size);
710 ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode);
711 const int shift_count = (1 + byte_no) * BitsPerByte;
712 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
713 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
714 "correct shift count");
715 srl(bytecode, shift_count, bytecode);
716 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
717 and3(bytecode, ConstantPoolCacheEntry::bytecode_1_mask, bytecode);
718 }
719
720
get_cache_entry_pointer_at_bcp(Register cache,Register tmp,int bcp_offset,size_t index_size)721 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp,
722 int bcp_offset, size_t index_size) {
723 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
724 assert_different_registers(cache, tmp);
725 assert_not_delayed();
726 if (index_size == sizeof(u2)) {
727 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned);
728 } else {
729 ShouldNotReachHere(); // other sizes not supported here
730 }
731 // convert from field index to ConstantPoolCacheEntry index
732 // and from word index to byte offset
733 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
734 // skip past the header
735 add(tmp, in_bytes(ConstantPoolCache::base_offset()), tmp);
736 // construct pointer to cache entry
737 add(LcpoolCache, tmp, cache);
738 }
739
740
741 // Load object from cpool->resolved_references(index)
load_resolved_reference_at_index(Register result,Register index,Register tmp)742 void InterpreterMacroAssembler::load_resolved_reference_at_index(
743 Register result, Register index, Register tmp) {
744 assert_different_registers(result, index, tmp);
745 assert_not_delayed();
746 // convert from field index to resolved_references() index and from
747 // word index to byte offset. Since this is a java object, it can be compressed
748 sll(index, LogBytesPerHeapOop, index);
749 get_constant_pool(result);
750 // load pointer for resolved_references[] objArray
751 ld_ptr(result, ConstantPool::cache_offset_in_bytes(), result);
752 ld_ptr(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result);
753 resolve_oop_handle(result, tmp);
754 // Add in the index
755 add(result, index, result);
756 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result, tmp);
757 // The resulting oop is null if the reference is not yet resolved.
758 // It is Universe::the_null_sentinel() if the reference resolved to NULL via condy.
759 }
760
761
762 // load cpool->resolved_klass_at(index)
load_resolved_klass_at_offset(Register Rcpool,Register Roffset,Register Rklass)763 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool,
764 Register Roffset, Register Rklass) {
765 // int value = *this_cp->int_at_addr(which);
766 // int resolved_klass_index = extract_low_short_from_int(value);
767 //
768 // Because SPARC is big-endian, the low_short is at (cpool->int_at_addr(which) + 2 bytes)
769 add(Roffset, Rcpool, Roffset);
770 lduh(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index
771
772 Register Rresolved_klasses = Rklass;
773 ld_ptr(Rcpool, ConstantPool::resolved_klasses_offset_in_bytes(), Rresolved_klasses);
774 sll(Roffset, LogBytesPerWord, Roffset);
775 add(Roffset, Array<Klass*>::base_offset_in_bytes(), Roffset);
776 ld_ptr(Rresolved_klasses, Roffset, Rklass);
777 }
778
779
780 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
781 // 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,Register Rtmp3,Label & ok_is_subtype)782 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
783 Register Rsuper_klass,
784 Register Rtmp1,
785 Register Rtmp2,
786 Register Rtmp3,
787 Label &ok_is_subtype ) {
788 Label not_subtype;
789
790 // Profile the not-null value's klass.
791 profile_typecheck(Rsub_klass, Rtmp1);
792
793 check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass,
794 Rtmp1, Rtmp2,
795 &ok_is_subtype, ¬_subtype, NULL);
796
797 check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass,
798 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg,
799 &ok_is_subtype, NULL);
800
801 bind(not_subtype);
802 profile_typecheck_failed(Rtmp1);
803 }
804
805 // Separate these two to allow for delay slot in middle
806 // These are used to do a test and full jump to exception-throwing code.
807
808 // %%%%% Could possibly reoptimize this by testing to see if could use
809 // a single conditional branch (i.e. if span is small enough.
810 // If you go that route, than get rid of the split and give up
811 // on the delay-slot hack.
812
throw_if_not_1_icc(Condition ok_condition,Label & ok)813 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition,
814 Label& ok ) {
815 assert_not_delayed();
816 br(ok_condition, true, pt, ok);
817 // DELAY SLOT
818 }
819
throw_if_not_1_xcc(Condition ok_condition,Label & ok)820 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition,
821 Label& ok ) {
822 assert_not_delayed();
823 bp( ok_condition, true, Assembler::xcc, pt, ok);
824 // DELAY SLOT
825 }
826
throw_if_not_1_x(Condition ok_condition,Label & ok)827 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition,
828 Label& ok ) {
829 assert_not_delayed();
830 brx(ok_condition, true, pt, ok);
831 // DELAY SLOT
832 }
833
throw_if_not_2(address throw_entry_point,Register Rscratch,Label & ok)834 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point,
835 Register Rscratch,
836 Label& ok ) {
837 assert(throw_entry_point != NULL, "entry point must be generated by now");
838 AddressLiteral dest(throw_entry_point);
839 jump_to(dest, Rscratch);
840 delayed()->nop();
841 bind(ok);
842 }
843
844
845 // And if you cannot use the delay slot, here is a shorthand:
846
throw_if_not_icc(Condition ok_condition,address throw_entry_point,Register Rscratch)847 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition,
848 address throw_entry_point,
849 Register Rscratch ) {
850 Label ok;
851 if (ok_condition != never) {
852 throw_if_not_1_icc( ok_condition, ok);
853 delayed()->nop();
854 }
855 throw_if_not_2( throw_entry_point, Rscratch, ok);
856 }
throw_if_not_xcc(Condition ok_condition,address throw_entry_point,Register Rscratch)857 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition,
858 address throw_entry_point,
859 Register Rscratch ) {
860 Label ok;
861 if (ok_condition != never) {
862 throw_if_not_1_xcc( ok_condition, ok);
863 delayed()->nop();
864 }
865 throw_if_not_2( throw_entry_point, Rscratch, ok);
866 }
throw_if_not_x(Condition ok_condition,address throw_entry_point,Register Rscratch)867 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition,
868 address throw_entry_point,
869 Register Rscratch ) {
870 Label ok;
871 if (ok_condition != never) {
872 throw_if_not_1_x( ok_condition, ok);
873 delayed()->nop();
874 }
875 throw_if_not_2( throw_entry_point, Rscratch, ok);
876 }
877
878 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res
879 // Note: res is still shy of address by array offset into object.
880
index_check_without_pop(Register array,Register index,int index_shift,Register tmp,Register res)881 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) {
882 assert_not_delayed();
883
884 verify_oop(array);
885 // Sign extend since tos (index) can be a 32bit value.
886 sra(index, G0, index);
887
888 // Check array.
889 Label ptr_ok;
890 tst(array);
891 throw_if_not_1_x(notZero, ptr_ok);
892 delayed()->ld(array, arrayOopDesc::length_offset_in_bytes(), tmp); // Check index.
893 throw_if_not_2(Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok);
894
895 Label index_ok;
896 cmp(index, tmp);
897 throw_if_not_1_icc(lessUnsigned, index_ok);
898 if (index_shift > 0) {
899 delayed()->sll(index, index_shift, index);
900 } else {
901 delayed()->add(array, index, res); // addr - const offset in index
902 }
903 // Pass the array to create more detailed exceptions.
904 // Convention: move aberrant index into Otos_i for exception message.
905 mov(index, Otos_i);
906 mov(array, G3_scratch);
907 throw_if_not_2(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok);
908
909 // add offset if didn't do it in delay slot
910 if (index_shift > 0) { add(array, index, res); } // addr - const offset in index
911 }
912
913
index_check(Register array,Register index,int index_shift,Register tmp,Register res)914 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) {
915 assert_not_delayed();
916
917 // pop array
918 pop_ptr(array);
919
920 // check array
921 index_check_without_pop(array, index, index_shift, tmp, res);
922 }
923
924
get_const(Register Rdst)925 void InterpreterMacroAssembler::get_const(Register Rdst) {
926 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst);
927 }
928
929
get_constant_pool(Register Rdst)930 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
931 get_const(Rdst);
932 ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
933 }
934
935
get_constant_pool_cache(Register Rdst)936 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
937 get_constant_pool(Rdst);
938 ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst);
939 }
940
941
get_cpool_and_tags(Register Rcpool,Register Rtags)942 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
943 get_constant_pool(Rcpool);
944 ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags);
945 }
946
947
948 // unlock if synchronized method
949 //
950 // Unlock the receiver if this is a synchronized method.
951 // Unlock any Java monitors from syncronized blocks.
952 //
953 // If there are locked Java monitors
954 // If throw_monitor_exception
955 // throws IllegalMonitorStateException
956 // Else if install_monitor_exception
957 // installs IllegalMonitorStateException
958 // Else
959 // no error processing
unlock_if_synchronized_method(TosState state,bool throw_monitor_exception,bool install_monitor_exception)960 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
961 bool throw_monitor_exception,
962 bool install_monitor_exception) {
963 Label unlocked, unlock, no_unlock;
964
965 // get the value of _do_not_unlock_if_synchronized into G1_scratch
966 const Address do_not_unlock_if_synchronized(G2_thread,
967 JavaThread::do_not_unlock_if_synchronized_offset());
968 ldbool(do_not_unlock_if_synchronized, G1_scratch);
969 stbool(G0, do_not_unlock_if_synchronized); // reset the flag
970
971 // check if synchronized method
972 const Address access_flags(Lmethod, Method::access_flags_offset());
973 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
974 push(state); // save tos
975 ld(access_flags, G3_scratch); // Load access flags.
976 btst(JVM_ACC_SYNCHRONIZED, G3_scratch);
977 br(zero, false, pt, unlocked);
978 delayed()->nop();
979
980 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
981 // is set.
982 cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock);
983 delayed()->nop();
984
985 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
986 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
987
988 //Intel: if (throw_monitor_exception) ... else ...
989 // Entry already unlocked, need to throw exception
990 //...
991
992 // pass top-most monitor elem
993 add( top_most_monitor(), O1 );
994
995 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch);
996 br_notnull_short(G3_scratch, pt, unlock);
997
998 if (throw_monitor_exception) {
999 // Entry already unlocked need to throw an exception
1000 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1001 should_not_reach_here();
1002 } else {
1003 // Monitor already unlocked during a stack unroll.
1004 // If requested, install an illegal_monitor_state_exception.
1005 // Continue with stack unrolling.
1006 if (install_monitor_exception) {
1007 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1008 }
1009 ba_short(unlocked);
1010 }
1011
1012 bind(unlock);
1013
1014 unlock_object(O1);
1015
1016 bind(unlocked);
1017
1018 // I0, I1: Might contain return value
1019
1020 // Check that all monitors are unlocked
1021 { Label loop, exception, entry, restart;
1022
1023 Register Rmptr = O0;
1024 Register Rtemp = O1;
1025 Register Rlimit = Lmonitors;
1026 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
1027 assert( (delta & LongAlignmentMask) == 0,
1028 "sizeof BasicObjectLock must be even number of doublewords");
1029
1030 #ifdef ASSERT
1031 add(top_most_monitor(), Rmptr, delta);
1032 { Label L;
1033 // ensure that Rmptr starts out above (or at) Rlimit
1034 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1035 stop("monitor stack has negative size");
1036 bind(L);
1037 }
1038 #endif
1039 bind(restart);
1040 ba(entry);
1041 delayed()->
1042 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry
1043
1044 // Entry is still locked, need to throw exception
1045 bind(exception);
1046 if (throw_monitor_exception) {
1047 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1048 should_not_reach_here();
1049 } else {
1050 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
1051 // Unlock does not block, so don't have to worry about the frame
1052 unlock_object(Rmptr);
1053 if (install_monitor_exception) {
1054 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1055 }
1056 ba_short(restart);
1057 }
1058
1059 bind(loop);
1060 cmp(Rtemp, G0); // check if current entry is used
1061 brx(Assembler::notEqual, false, pn, exception);
1062 delayed()->
1063 dec(Rmptr, delta); // otherwise advance to next entry
1064 #ifdef ASSERT
1065 { Label L;
1066 // ensure that Rmptr has not somehow stepped below Rlimit
1067 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1068 stop("ran off the end of the monitor stack");
1069 bind(L);
1070 }
1071 #endif
1072 bind(entry);
1073 cmp(Rmptr, Rlimit); // check if bottom reached
1074 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry
1075 delayed()->
1076 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp);
1077 }
1078
1079 bind(no_unlock);
1080 pop(state);
1081 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1082 }
1083
narrow(Register result)1084 void InterpreterMacroAssembler::narrow(Register result) {
1085
1086 ld_ptr(Address(Lmethod, Method::const_offset()), G3_scratch);
1087 ldub(G3_scratch, in_bytes(ConstMethod::result_type_offset()), G3_scratch);
1088
1089 Label notBool, notByte, notChar, done;
1090
1091 // common case first
1092 cmp(G3_scratch, T_INT);
1093 br(Assembler::equal, true, pn, done);
1094 delayed()->nop();
1095
1096 cmp(G3_scratch, T_BOOLEAN);
1097 br(Assembler::notEqual, true, pn, notBool);
1098 delayed()->cmp(G3_scratch, T_BYTE);
1099 and3(result, 1, result);
1100 ba(done);
1101 delayed()->nop();
1102
1103 bind(notBool);
1104 // cmp(G3_scratch, T_BYTE);
1105 br(Assembler::notEqual, true, pn, notByte);
1106 delayed()->cmp(G3_scratch, T_CHAR);
1107 sll(result, 24, result);
1108 sra(result, 24, result);
1109 ba(done);
1110 delayed()->nop();
1111
1112 bind(notByte);
1113 // cmp(G3_scratch, T_CHAR);
1114 sll(result, 16, result);
1115 br(Assembler::notEqual, true, pn, done);
1116 delayed()->sra(result, 16, result);
1117 // sll(result, 16, result);
1118 srl(result, 16, result);
1119
1120 // bind(notChar);
1121 // must be short, instructions already executed in delay slot
1122 // sll(result, 16, result);
1123 // sra(result, 16, result);
1124
1125 bind(done);
1126 }
1127
1128 // remove activation
1129 //
1130 // Unlock the receiver if this is a synchronized method.
1131 // Unlock any Java monitors from syncronized blocks.
1132 // Remove the activation from the stack.
1133 //
1134 // If there are locked Java monitors
1135 // If throw_monitor_exception
1136 // throws IllegalMonitorStateException
1137 // Else if install_monitor_exception
1138 // installs IllegalMonitorStateException
1139 // Else
1140 // no error processing
remove_activation(TosState state,bool throw_monitor_exception,bool install_monitor_exception)1141 void InterpreterMacroAssembler::remove_activation(TosState state,
1142 bool throw_monitor_exception,
1143 bool install_monitor_exception) {
1144
1145 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
1146
1147 // save result (push state before jvmti call and pop it afterwards) and notify jvmti
1148 notify_method_exit(false, state, NotifyJVMTI);
1149
1150 if (StackReservedPages > 0) {
1151 // testing if Stack Reserved Area needs to be re-enabled
1152 Label no_reserved_zone_enabling;
1153 ld_ptr(G2_thread, JavaThread::reserved_stack_activation_offset(), G3_scratch);
1154 cmp_and_brx_short(SP, G3_scratch, Assembler::lessUnsigned, Assembler::pt, no_reserved_zone_enabling);
1155
1156 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), G2_thread);
1157 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError), G2_thread);
1158 should_not_reach_here();
1159
1160 bind(no_reserved_zone_enabling);
1161 }
1162
1163 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1164 verify_thread();
1165
1166 // return tos
1167 assert(Otos_l1 == Otos_i, "adjust code below");
1168 switch (state) {
1169 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0
1170 case btos: // fall through
1171 case ztos: // fall through
1172 case ctos:
1173 case stos: // fall through
1174 case atos: // fall through
1175 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0
1176 case ftos: // fall through
1177 case dtos: // fall through
1178 case vtos: /* nothing to do */ break;
1179 default : ShouldNotReachHere();
1180 }
1181 }
1182
1183 // Lock object
1184 //
1185 // Argument - lock_reg points to the BasicObjectLock to be used for locking,
1186 // it must be initialized with the object to lock
lock_object(Register lock_reg,Register Object)1187 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) {
1188 if (UseHeavyMonitors) {
1189 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1190 }
1191 else {
1192 Register obj_reg = Object;
1193 Register mark_reg = G4_scratch;
1194 Register temp_reg = G1_scratch;
1195 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes());
1196 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1197 Label done;
1198
1199 Label slow_case;
1200
1201 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg);
1202
1203 // load markOop from object into mark_reg
1204 ld_ptr(mark_addr, mark_reg);
1205
1206 if (UseBiasedLocking) {
1207 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case);
1208 }
1209
1210 // get the address of basicLock on stack that will be stored in the object
1211 // we need a temporary register here as we do not want to clobber lock_reg
1212 // (cas clobbers the destination register)
1213 mov(lock_reg, temp_reg);
1214 // set mark reg to be (markOop of object | UNLOCK_VALUE)
1215 or3(mark_reg, markOopDesc::unlocked_value, mark_reg);
1216 // initialize the box (Must happen before we update the object mark!)
1217 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1218 // compare and exchange object_addr, markOop | 1, stack address of basicLock
1219 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1220 cas_ptr(mark_addr.base(), mark_reg, temp_reg);
1221
1222 // if the compare and exchange succeeded we are done (we saw an unlocked object)
1223 cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done);
1224
1225 // We did not see an unlocked object so try the fast recursive case
1226
1227 // Check if owner is self by comparing the value in the markOop of object
1228 // with the stack pointer
1229 sub(temp_reg, SP, temp_reg);
1230 sub(temp_reg, STACK_BIAS, temp_reg);
1231 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1232
1233 // Composite "andcc" test:
1234 // (a) %sp -vs- markword proximity check, and,
1235 // (b) verify mark word LSBs == 0 (Stack-locked).
1236 //
1237 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size())
1238 // Note that the page size used for %sp proximity testing is arbitrary and is
1239 // unrelated to the actual MMU page size. We use a 'logical' page size of
1240 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate
1241 // field of the andcc instruction.
1242 andcc (temp_reg, 0xFFFFF003, G0) ;
1243
1244 // if condition is true we are done and hence we can store 0 in the displaced
1245 // header indicating it is a recursive lock and be done
1246 brx(Assembler::zero, true, Assembler::pt, done);
1247 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1248
1249 // none of the above fast optimizations worked so we have to get into the
1250 // slow case of monitor enter
1251 bind(slow_case);
1252 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1253
1254 bind(done);
1255 }
1256 }
1257
1258 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1259 //
1260 // Argument - lock_reg points to the BasicObjectLock for lock
1261 // Throw IllegalMonitorException if object is not locked by current thread
unlock_object(Register lock_reg)1262 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1263 if (UseHeavyMonitors) {
1264 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1265 } else {
1266 Register obj_reg = G3_scratch;
1267 Register mark_reg = G4_scratch;
1268 Register displaced_header_reg = G1_scratch;
1269 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes());
1270 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1271 Label done;
1272
1273 if (UseBiasedLocking) {
1274 // load the object out of the BasicObjectLock
1275 ld_ptr(lockobj_addr, obj_reg);
1276 biased_locking_exit(mark_addr, mark_reg, done, true);
1277 st_ptr(G0, lockobj_addr); // free entry
1278 }
1279
1280 // Test first if we are in the fast recursive case
1281 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes());
1282 ld_ptr(lock_addr, displaced_header_reg);
1283 br_null(displaced_header_reg, true, Assembler::pn, done);
1284 delayed()->st_ptr(G0, lockobj_addr); // free entry
1285
1286 // See if it is still a light weight lock, if so we just unlock
1287 // the object and we are done
1288
1289 if (!UseBiasedLocking) {
1290 // load the object out of the BasicObjectLock
1291 ld_ptr(lockobj_addr, obj_reg);
1292 }
1293
1294 // we have the displaced header in displaced_header_reg
1295 // we expect to see the stack address of the basicLock in case the
1296 // lock is still a light weight lock (lock_reg)
1297 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1298 cas_ptr(mark_addr.base(), lock_reg, displaced_header_reg);
1299 cmp(lock_reg, displaced_header_reg);
1300 brx(Assembler::equal, true, Assembler::pn, done);
1301 delayed()->st_ptr(G0, lockobj_addr); // free entry
1302
1303 // The lock has been converted into a heavy lock and hence
1304 // we need to get into the slow case
1305
1306 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1307
1308 bind(done);
1309 }
1310 }
1311
1312 // Get the method data pointer from the Method* and set the
1313 // specified register to its value.
1314
set_method_data_pointer()1315 void InterpreterMacroAssembler::set_method_data_pointer() {
1316 assert(ProfileInterpreter, "must be profiling interpreter");
1317 Label get_continue;
1318
1319 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1320 test_method_data_pointer(get_continue);
1321 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1322 bind(get_continue);
1323 }
1324
1325 // Set the method data pointer for the current bcp.
1326
set_method_data_pointer_for_bcp()1327 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1328 assert(ProfileInterpreter, "must be profiling interpreter");
1329 Label zero_continue;
1330
1331 // Test MDO to avoid the call if it is NULL.
1332 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1333 test_method_data_pointer(zero_continue);
1334 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp);
1335 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1336 add(ImethodDataPtr, O0, ImethodDataPtr);
1337 bind(zero_continue);
1338 }
1339
1340 // Test ImethodDataPtr. If it is null, continue at the specified label
1341
test_method_data_pointer(Label & zero_continue)1342 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1343 assert(ProfileInterpreter, "must be profiling interpreter");
1344 br_null_short(ImethodDataPtr, Assembler::pn, zero_continue);
1345 }
1346
verify_method_data_pointer()1347 void InterpreterMacroAssembler::verify_method_data_pointer() {
1348 assert(ProfileInterpreter, "must be profiling interpreter");
1349 #ifdef ASSERT
1350 Label verify_continue;
1351 test_method_data_pointer(verify_continue);
1352
1353 // If the mdp is valid, it will point to a DataLayout header which is
1354 // consistent with the bcp. The converse is highly probable also.
1355 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch);
1356 ld_ptr(Lmethod, Method::const_offset(), O5);
1357 add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch);
1358 add(G3_scratch, O5, G3_scratch);
1359 cmp(Lbcp, G3_scratch);
1360 brx(Assembler::equal, false, Assembler::pt, verify_continue);
1361
1362 Register temp_reg = O5;
1363 delayed()->mov(ImethodDataPtr, temp_reg);
1364 // %%% should use call_VM_leaf here?
1365 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr);
1366 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1);
1367 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS);
1368 stf(FloatRegisterImpl::D, Ftos_d, d_save);
1369 mov(temp_reg->after_save(), O2);
1370 save_thread(L7_thread_cache);
1371 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none);
1372 delayed()->nop();
1373 restore_thread(L7_thread_cache);
1374 ldf(FloatRegisterImpl::D, d_save, Ftos_d);
1375 restore();
1376 bind(verify_continue);
1377 #endif // ASSERT
1378 }
1379
test_invocation_counter_for_mdp(Register invocation_count,Register method_counters,Register Rtmp,Label & profile_continue)1380 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1381 Register method_counters,
1382 Register Rtmp,
1383 Label &profile_continue) {
1384 assert(ProfileInterpreter, "must be profiling interpreter");
1385 // Control will flow to "profile_continue" if the counter is less than the
1386 // limit or if we call profile_method()
1387
1388 Label done;
1389
1390 // if no method data exists, and the counter is high enough, make one
1391 br_notnull_short(ImethodDataPtr, Assembler::pn, done);
1392
1393 // Test to see if we should create a method data oop
1394 Address profile_limit(method_counters, MethodCounters::interpreter_profile_limit_offset());
1395 ld(profile_limit, Rtmp);
1396 cmp(invocation_count, Rtmp);
1397 // Use long branches because call_VM() code and following code generated by
1398 // test_backedge_count_for_osr() is large in debug VM.
1399 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue);
1400 delayed()->nop();
1401
1402 // Build it now.
1403 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1404 set_method_data_pointer_for_bcp();
1405 ba(profile_continue);
1406 delayed()->nop();
1407 bind(done);
1408 }
1409
1410 // Store a value at some constant offset from the method data pointer.
1411
set_mdp_data_at(int constant,Register value)1412 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1413 assert(ProfileInterpreter, "must be profiling interpreter");
1414 st_ptr(value, ImethodDataPtr, constant);
1415 }
1416
increment_mdp_data_at(Address counter,Register bumped_count,bool decrement)1417 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter,
1418 Register bumped_count,
1419 bool decrement) {
1420 assert(ProfileInterpreter, "must be profiling interpreter");
1421
1422 // Load the counter.
1423 ld_ptr(counter, bumped_count);
1424
1425 if (decrement) {
1426 // Decrement the register. Set condition codes.
1427 subcc(bumped_count, DataLayout::counter_increment, bumped_count);
1428
1429 // If the decrement causes the counter to overflow, stay negative
1430 Label L;
1431 brx(Assembler::negative, true, Assembler::pn, L);
1432
1433 // Store the decremented counter, if it is still negative.
1434 delayed()->st_ptr(bumped_count, counter);
1435 bind(L);
1436 } else {
1437 // Increment the register. Set carry flag.
1438 addcc(bumped_count, DataLayout::counter_increment, bumped_count);
1439
1440 // If the increment causes the counter to overflow, pull back by 1.
1441 assert(DataLayout::counter_increment == 1, "subc works");
1442 subc(bumped_count, G0, bumped_count);
1443
1444 // Store the incremented counter.
1445 st_ptr(bumped_count, counter);
1446 }
1447 }
1448
1449 // Increment the value at some constant offset from the method data pointer.
1450
increment_mdp_data_at(int constant,Register bumped_count,bool decrement)1451 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1452 Register bumped_count,
1453 bool decrement) {
1454 // Locate the counter at a fixed offset from the mdp:
1455 Address counter(ImethodDataPtr, constant);
1456 increment_mdp_data_at(counter, bumped_count, decrement);
1457 }
1458
1459 // Increment the value at some non-fixed (reg + constant) offset from
1460 // the method data pointer.
1461
increment_mdp_data_at(Register reg,int constant,Register bumped_count,Register scratch2,bool decrement)1462 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1463 int constant,
1464 Register bumped_count,
1465 Register scratch2,
1466 bool decrement) {
1467 // Add the constant to reg to get the offset.
1468 add(ImethodDataPtr, reg, scratch2);
1469 Address counter(scratch2, constant);
1470 increment_mdp_data_at(counter, bumped_count, decrement);
1471 }
1472
1473 // Set a flag value at the current method data pointer position.
1474 // Updates a single byte of the header, to avoid races with other header bits.
1475
set_mdp_flag_at(int flag_constant,Register scratch)1476 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1477 Register scratch) {
1478 assert(ProfileInterpreter, "must be profiling interpreter");
1479 // Load the data header
1480 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch);
1481
1482 // Set the flag
1483 or3(scratch, flag_constant, scratch);
1484
1485 // Store the modified header.
1486 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset()));
1487 }
1488
1489 // Test the location at some offset from the method data pointer.
1490 // If it is not equal to value, branch to the not_equal_continue Label.
1491 // Set condition codes to match the nullness of the loaded value.
1492
test_mdp_data_at(int offset,Register value,Label & not_equal_continue,Register scratch)1493 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1494 Register value,
1495 Label& not_equal_continue,
1496 Register scratch) {
1497 assert(ProfileInterpreter, "must be profiling interpreter");
1498 ld_ptr(ImethodDataPtr, offset, scratch);
1499 cmp(value, scratch);
1500 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue);
1501 delayed()->tst(scratch);
1502 }
1503
1504 // Update the method data pointer by the displacement located at some fixed
1505 // offset from the method data pointer.
1506
update_mdp_by_offset(int offset_of_disp,Register scratch)1507 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1508 Register scratch) {
1509 assert(ProfileInterpreter, "must be profiling interpreter");
1510 ld_ptr(ImethodDataPtr, offset_of_disp, scratch);
1511 add(ImethodDataPtr, scratch, ImethodDataPtr);
1512 }
1513
1514 // Update the method data pointer by the displacement located at the
1515 // offset (reg + offset_of_disp).
1516
update_mdp_by_offset(Register reg,int offset_of_disp,Register scratch)1517 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1518 int offset_of_disp,
1519 Register scratch) {
1520 assert(ProfileInterpreter, "must be profiling interpreter");
1521 add(reg, offset_of_disp, scratch);
1522 ld_ptr(ImethodDataPtr, scratch, scratch);
1523 add(ImethodDataPtr, scratch, ImethodDataPtr);
1524 }
1525
1526 // Update the method data pointer by a simple constant displacement.
1527
update_mdp_by_constant(int constant)1528 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1529 assert(ProfileInterpreter, "must be profiling interpreter");
1530 add(ImethodDataPtr, constant, ImethodDataPtr);
1531 }
1532
1533 // Update the method data pointer for a _ret bytecode whose target
1534 // was not among our cached targets.
1535
update_mdp_for_ret(TosState state,Register return_bci)1536 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1537 Register return_bci) {
1538 assert(ProfileInterpreter, "must be profiling interpreter");
1539 push(state);
1540 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile
1541 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1542 ld_ptr(l_tmp, return_bci);
1543 pop(state);
1544 }
1545
1546 // Count a taken branch in the bytecodes.
1547
profile_taken_branch(Register scratch,Register bumped_count)1548 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1549 if (ProfileInterpreter) {
1550 Label profile_continue;
1551
1552 // If no method data exists, go to profile_continue.
1553 test_method_data_pointer(profile_continue);
1554
1555 // We are taking a branch. Increment the taken count.
1556 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count);
1557
1558 // The method data pointer needs to be updated to reflect the new target.
1559 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1560 bind (profile_continue);
1561 }
1562 }
1563
1564
1565 // Count a not-taken branch in the bytecodes.
1566
profile_not_taken_branch(Register scratch)1567 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) {
1568 if (ProfileInterpreter) {
1569 Label profile_continue;
1570
1571 // If no method data exists, go to profile_continue.
1572 test_method_data_pointer(profile_continue);
1573
1574 // We are taking a branch. Increment the not taken count.
1575 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch);
1576
1577 // The method data pointer needs to be updated to correspond to the
1578 // next bytecode.
1579 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1580 bind (profile_continue);
1581 }
1582 }
1583
1584
1585 // Count a non-virtual call in the bytecodes.
1586
profile_call(Register scratch)1587 void InterpreterMacroAssembler::profile_call(Register scratch) {
1588 if (ProfileInterpreter) {
1589 Label profile_continue;
1590
1591 // If no method data exists, go to profile_continue.
1592 test_method_data_pointer(profile_continue);
1593
1594 // We are making a call. Increment the count.
1595 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1596
1597 // The method data pointer needs to be updated to reflect the new target.
1598 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1599 bind (profile_continue);
1600 }
1601 }
1602
1603
1604 // Count a final call in the bytecodes.
1605
profile_final_call(Register scratch)1606 void InterpreterMacroAssembler::profile_final_call(Register scratch) {
1607 if (ProfileInterpreter) {
1608 Label profile_continue;
1609
1610 // If no method data exists, go to profile_continue.
1611 test_method_data_pointer(profile_continue);
1612
1613 // We are making a call. Increment the count.
1614 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1615
1616 // The method data pointer needs to be updated to reflect the new target.
1617 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1618 bind (profile_continue);
1619 }
1620 }
1621
1622
1623 // Count a virtual call in the bytecodes.
1624
profile_virtual_call(Register receiver,Register scratch,bool receiver_can_be_null)1625 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1626 Register scratch,
1627 bool receiver_can_be_null) {
1628 if (ProfileInterpreter) {
1629 Label profile_continue;
1630
1631 // If no method data exists, go to profile_continue.
1632 test_method_data_pointer(profile_continue);
1633
1634
1635 Label skip_receiver_profile;
1636 if (receiver_can_be_null) {
1637 Label not_null;
1638 br_notnull_short(receiver, Assembler::pt, not_null);
1639 // We are making a call. Increment the count for null receiver.
1640 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1641 ba_short(skip_receiver_profile);
1642 bind(not_null);
1643 }
1644
1645 // Record the receiver type.
1646 record_klass_in_profile(receiver, scratch, true);
1647 bind(skip_receiver_profile);
1648
1649 // The method data pointer needs to be updated to reflect the new target.
1650 #if INCLUDE_JVMCI
1651 if (MethodProfileWidth == 0) {
1652 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1653 }
1654 #else
1655 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1656 #endif
1657 bind(profile_continue);
1658 }
1659 }
1660
1661 #if INCLUDE_JVMCI
profile_called_method(Register method,Register scratch)1662 void InterpreterMacroAssembler::profile_called_method(Register method, Register scratch) {
1663 assert_different_registers(method, scratch);
1664 if (ProfileInterpreter && MethodProfileWidth > 0) {
1665 Label profile_continue;
1666
1667 // If no method data exists, go to profile_continue.
1668 test_method_data_pointer(profile_continue);
1669
1670 Label done;
1671 record_item_in_profile_helper(method, scratch, 0, done, MethodProfileWidth,
1672 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1673 bind(done);
1674
1675 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1676 bind(profile_continue);
1677 }
1678 }
1679 #endif // INCLUDE_JVMCI
1680
record_klass_in_profile_helper(Register receiver,Register scratch,Label & done,bool is_virtual_call)1681 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register scratch,
1682 Label& done, bool is_virtual_call) {
1683 if (TypeProfileWidth == 0) {
1684 if (is_virtual_call) {
1685 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1686 }
1687 #if INCLUDE_JVMCI
1688 else if (EnableJVMCI) {
1689 increment_mdp_data_at(in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()), scratch);
1690 }
1691 #endif
1692 } else {
1693 int non_profiled_offset = -1;
1694 if (is_virtual_call) {
1695 non_profiled_offset = in_bytes(CounterData::count_offset());
1696 }
1697 #if INCLUDE_JVMCI
1698 else if (EnableJVMCI) {
1699 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1700 }
1701 #endif
1702
1703 record_item_in_profile_helper(receiver, scratch, 0, done, TypeProfileWidth,
1704 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1705 }
1706 }
1707
record_item_in_profile_helper(Register item,Register scratch,int start_row,Label & done,int total_rows,OffsetFunction item_offset_fn,OffsetFunction item_count_offset_fn,int non_profiled_offset)1708 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item,
1709 Register scratch, int start_row, Label& done, int total_rows,
1710 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1711 int non_profiled_offset) {
1712 int last_row = total_rows - 1;
1713 assert(start_row <= last_row, "must be work left to do");
1714 // Test this row for both the item and for null.
1715 // Take any of three different outcomes:
1716 // 1. found item => increment count and goto done
1717 // 2. found null => keep looking for case 1, maybe allocate this cell
1718 // 3. found something else => keep looking for cases 1 and 2
1719 // Case 3 is handled by a recursive call.
1720 for (int row = start_row; row <= last_row; row++) {
1721 Label next_test;
1722 bool test_for_null_also = (row == start_row);
1723
1724 // See if the item is item[n].
1725 int item_offset = in_bytes(item_offset_fn(row));
1726 test_mdp_data_at(item_offset, item, next_test, scratch);
1727 // delayed()->tst(scratch);
1728
1729 // The receiver is item[n]. Increment count[n].
1730 int count_offset = in_bytes(item_count_offset_fn(row));
1731 increment_mdp_data_at(count_offset, scratch);
1732 ba_short(done);
1733 bind(next_test);
1734
1735 if (test_for_null_also) {
1736 Label found_null;
1737 // Failed the equality check on item[n]... Test for null.
1738 if (start_row == last_row) {
1739 // The only thing left to do is handle the null case.
1740 if (non_profiled_offset >= 0) {
1741 brx(Assembler::zero, false, Assembler::pn, found_null);
1742 delayed()->nop();
1743 // Item did not match any saved item and there is no empty row for it.
1744 // Increment total counter to indicate polymorphic case.
1745 increment_mdp_data_at(non_profiled_offset, scratch);
1746 ba_short(done);
1747 bind(found_null);
1748 } else {
1749 brx(Assembler::notZero, false, Assembler::pt, done);
1750 delayed()->nop();
1751 }
1752 break;
1753 }
1754 // Since null is rare, make it be the branch-taken case.
1755 brx(Assembler::zero, false, Assembler::pn, found_null);
1756 delayed()->nop();
1757
1758 // Put all the "Case 3" tests here.
1759 record_item_in_profile_helper(item, scratch, start_row + 1, done, total_rows,
1760 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1761
1762 // Found a null. Keep searching for a matching item,
1763 // but remember that this is an empty (unused) slot.
1764 bind(found_null);
1765 }
1766 }
1767
1768 // In the fall-through case, we found no matching item, but we
1769 // observed the item[start_row] is NULL.
1770
1771 // Fill in the item field and increment the count.
1772 int item_offset = in_bytes(item_offset_fn(start_row));
1773 set_mdp_data_at(item_offset, item);
1774 int count_offset = in_bytes(item_count_offset_fn(start_row));
1775 mov(DataLayout::counter_increment, scratch);
1776 set_mdp_data_at(count_offset, scratch);
1777 if (start_row > 0) {
1778 ba_short(done);
1779 }
1780 }
1781
record_klass_in_profile(Register receiver,Register scratch,bool is_virtual_call)1782 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1783 Register scratch, bool is_virtual_call) {
1784 assert(ProfileInterpreter, "must be profiling");
1785 Label done;
1786
1787 record_klass_in_profile_helper(receiver, scratch, done, is_virtual_call);
1788
1789 bind (done);
1790 }
1791
1792
1793 // Count a ret in the bytecodes.
1794
profile_ret(TosState state,Register return_bci,Register scratch)1795 void InterpreterMacroAssembler::profile_ret(TosState state,
1796 Register return_bci,
1797 Register scratch) {
1798 if (ProfileInterpreter) {
1799 Label profile_continue;
1800 uint row;
1801
1802 // If no method data exists, go to profile_continue.
1803 test_method_data_pointer(profile_continue);
1804
1805 // Update the total ret count.
1806 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1807
1808 for (row = 0; row < RetData::row_limit(); row++) {
1809 Label next_test;
1810
1811 // See if return_bci is equal to bci[n]:
1812 test_mdp_data_at(in_bytes(RetData::bci_offset(row)),
1813 return_bci, next_test, scratch);
1814
1815 // return_bci is equal to bci[n]. Increment the count.
1816 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch);
1817
1818 // The method data pointer needs to be updated to reflect the new target.
1819 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch);
1820 ba_short(profile_continue);
1821 bind(next_test);
1822 }
1823
1824 update_mdp_for_ret(state, return_bci);
1825
1826 bind (profile_continue);
1827 }
1828 }
1829
1830 // Profile an unexpected null in the bytecodes.
profile_null_seen(Register scratch)1831 void InterpreterMacroAssembler::profile_null_seen(Register scratch) {
1832 if (ProfileInterpreter) {
1833 Label profile_continue;
1834
1835 // If no method data exists, go to profile_continue.
1836 test_method_data_pointer(profile_continue);
1837
1838 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch);
1839
1840 // The method data pointer needs to be updated.
1841 int mdp_delta = in_bytes(BitData::bit_data_size());
1842 if (TypeProfileCasts) {
1843 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size());
1844 }
1845 update_mdp_by_constant(mdp_delta);
1846
1847 bind (profile_continue);
1848 }
1849 }
1850
profile_typecheck(Register klass,Register scratch)1851 void InterpreterMacroAssembler::profile_typecheck(Register klass,
1852 Register scratch) {
1853 if (ProfileInterpreter) {
1854 Label profile_continue;
1855
1856 // If no method data exists, go to profile_continue.
1857 test_method_data_pointer(profile_continue);
1858
1859 int mdp_delta = in_bytes(BitData::bit_data_size());
1860 if (TypeProfileCasts) {
1861 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size());
1862
1863 // Record the object type.
1864 record_klass_in_profile(klass, scratch, false);
1865 }
1866
1867 // The method data pointer needs to be updated.
1868 update_mdp_by_constant(mdp_delta);
1869
1870 bind (profile_continue);
1871 }
1872 }
1873
profile_typecheck_failed(Register scratch)1874 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) {
1875 if (ProfileInterpreter && TypeProfileCasts) {
1876 Label profile_continue;
1877
1878 // If no method data exists, go to profile_continue.
1879 test_method_data_pointer(profile_continue);
1880
1881 int count_offset = in_bytes(CounterData::count_offset());
1882 // Back up the address, since we have already bumped the mdp.
1883 count_offset -= in_bytes(ReceiverTypeData::receiver_type_data_size());
1884
1885 // *Decrement* the counter. We expect to see zero or small negatives.
1886 increment_mdp_data_at(count_offset, scratch, true);
1887
1888 bind (profile_continue);
1889 }
1890 }
1891
1892 // Count the default case of a switch construct.
1893
profile_switch_default(Register scratch)1894 void InterpreterMacroAssembler::profile_switch_default(Register scratch) {
1895 if (ProfileInterpreter) {
1896 Label profile_continue;
1897
1898 // If no method data exists, go to profile_continue.
1899 test_method_data_pointer(profile_continue);
1900
1901 // Update the default case count
1902 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1903 scratch);
1904
1905 // The method data pointer needs to be updated.
1906 update_mdp_by_offset(
1907 in_bytes(MultiBranchData::default_displacement_offset()),
1908 scratch);
1909
1910 bind (profile_continue);
1911 }
1912 }
1913
1914 // Count the index'th case of a switch construct.
1915
profile_switch_case(Register index,Register scratch,Register scratch2,Register scratch3)1916 void InterpreterMacroAssembler::profile_switch_case(Register index,
1917 Register scratch,
1918 Register scratch2,
1919 Register scratch3) {
1920 if (ProfileInterpreter) {
1921 Label profile_continue;
1922
1923 // If no method data exists, go to profile_continue.
1924 test_method_data_pointer(profile_continue);
1925
1926 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes()
1927 set(in_bytes(MultiBranchData::per_case_size()), scratch);
1928 smul(index, scratch, scratch);
1929 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch);
1930
1931 // Update the case count
1932 increment_mdp_data_at(scratch,
1933 in_bytes(MultiBranchData::relative_count_offset()),
1934 scratch2,
1935 scratch3);
1936
1937 // The method data pointer needs to be updated.
1938 update_mdp_by_offset(scratch,
1939 in_bytes(MultiBranchData::relative_displacement_offset()),
1940 scratch2);
1941
1942 bind (profile_continue);
1943 }
1944 }
1945
profile_obj_type(Register obj,const Address & mdo_addr,Register tmp)1946 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) {
1947 Label not_null, do_nothing, do_update;
1948
1949 assert_different_registers(obj, mdo_addr.base(), tmp);
1950
1951 verify_oop(obj);
1952
1953 ld_ptr(mdo_addr, tmp);
1954
1955 br_notnull_short(obj, pt, not_null);
1956 or3(tmp, TypeEntries::null_seen, tmp);
1957 ba_short(do_update);
1958
1959 bind(not_null);
1960 load_klass(obj, obj);
1961
1962 xor3(obj, tmp, obj);
1963 btst(TypeEntries::type_klass_mask, obj);
1964 // klass seen before, nothing to do. The unknown bit may have been
1965 // set already but no need to check.
1966 brx(zero, false, pt, do_nothing);
1967 delayed()->
1968
1969 btst(TypeEntries::type_unknown, obj);
1970 // already unknown. Nothing to do anymore.
1971 brx(notZero, false, pt, do_nothing);
1972 delayed()->
1973
1974 btst(TypeEntries::type_mask, tmp);
1975 brx(zero, true, pt, do_update);
1976 // first time here. Set profile type.
1977 delayed()->or3(tmp, obj, tmp);
1978
1979 // different than before. Cannot keep accurate profile.
1980 or3(tmp, TypeEntries::type_unknown, tmp);
1981
1982 bind(do_update);
1983 // update profile
1984 st_ptr(tmp, mdo_addr);
1985
1986 bind(do_nothing);
1987 }
1988
profile_arguments_type(Register callee,Register tmp1,Register tmp2,bool is_virtual)1989 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) {
1990 if (!ProfileInterpreter) {
1991 return;
1992 }
1993
1994 assert_different_registers(callee, tmp1, tmp2, ImethodDataPtr);
1995
1996 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1997 Label profile_continue;
1998
1999 test_method_data_pointer(profile_continue);
2000
2001 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
2002
2003 ldub(ImethodDataPtr, in_bytes(DataLayout::tag_offset()) - off_to_start, tmp1);
2004 cmp_and_br_short(tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag, notEqual, pn, profile_continue);
2005
2006 if (MethodData::profile_arguments()) {
2007 Label done;
2008 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
2009 add(ImethodDataPtr, off_to_args, ImethodDataPtr);
2010
2011 for (int i = 0; i < TypeProfileArgsLimit; i++) {
2012 if (i > 0 || MethodData::profile_return()) {
2013 // If return value type is profiled we may have no argument to profile
2014 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
2015 sub(tmp1, i*TypeStackSlotEntries::per_arg_count(), tmp1);
2016 cmp_and_br_short(tmp1, TypeStackSlotEntries::per_arg_count(), less, pn, done);
2017 }
2018 ld_ptr(Address(callee, Method::const_offset()), tmp1);
2019 lduh(Address(tmp1, ConstMethod::size_of_parameters_offset()), tmp1);
2020 // stack offset o (zero based) from the start of the argument
2021 // list, for n arguments translates into offset n - o - 1 from
2022 // the end of the argument list. But there's an extra slot at
2023 // the stop of the stack. So the offset is n - o from Lesp.
2024 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, tmp2);
2025 sub(tmp1, tmp2, tmp1);
2026
2027 // Can't use MacroAssembler::argument_address() which needs Gargs to be set up
2028 sll(tmp1, Interpreter::logStackElementSize, tmp1);
2029 ld_ptr(Lesp, tmp1, tmp1);
2030
2031 Address mdo_arg_addr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
2032 profile_obj_type(tmp1, mdo_arg_addr, tmp2);
2033
2034 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
2035 add(ImethodDataPtr, to_add, ImethodDataPtr);
2036 off_to_args += to_add;
2037 }
2038
2039 if (MethodData::profile_return()) {
2040 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
2041 sub(tmp1, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count(), tmp1);
2042 }
2043
2044 bind(done);
2045
2046 if (MethodData::profile_return()) {
2047 // We're right after the type profile for the last
2048 // argument. tmp1 is the number of cells left in the
2049 // CallTypeData/VirtualCallTypeData to reach its end. Non null
2050 // if there's a return to profile.
2051 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
2052 sll(tmp1, exact_log2(DataLayout::cell_size), tmp1);
2053 add(ImethodDataPtr, tmp1, ImethodDataPtr);
2054 }
2055 } else {
2056 assert(MethodData::profile_return(), "either profile call args or call ret");
2057 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
2058 }
2059
2060 // mdp points right after the end of the
2061 // CallTypeData/VirtualCallTypeData, right after the cells for the
2062 // return value type if there's one.
2063
2064 bind(profile_continue);
2065 }
2066 }
2067
profile_return_type(Register ret,Register tmp1,Register tmp2)2068 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
2069 assert_different_registers(ret, tmp1, tmp2);
2070 if (ProfileInterpreter && MethodData::profile_return()) {
2071 Label profile_continue, done;
2072
2073 test_method_data_pointer(profile_continue);
2074
2075 if (MethodData::profile_return_jsr292_only()) {
2076 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
2077
2078 // If we don't profile all invoke bytecodes we must make sure
2079 // it's a bytecode we indeed profile. We can't go back to the
2080 // begining of the ProfileData we intend to update to check its
2081 // type because we're right after it and we don't known its
2082 // length.
2083 Label do_profile;
2084 ldub(Lbcp, 0, tmp1);
2085 cmp_and_br_short(tmp1, Bytecodes::_invokedynamic, equal, pn, do_profile);
2086 cmp(tmp1, Bytecodes::_invokehandle);
2087 br(equal, false, pn, do_profile);
2088 delayed()->lduh(Lmethod, Method::intrinsic_id_offset_in_bytes(), tmp1);
2089 cmp_and_br_short(tmp1, vmIntrinsics::_compiledLambdaForm, notEqual, pt, profile_continue);
2090
2091 bind(do_profile);
2092 }
2093
2094 Address mdo_ret_addr(ImethodDataPtr, -in_bytes(ReturnTypeEntry::size()));
2095 mov(ret, tmp1);
2096 profile_obj_type(tmp1, mdo_ret_addr, tmp2);
2097
2098 bind(profile_continue);
2099 }
2100 }
2101
profile_parameters_type(Register tmp1,Register tmp2,Register tmp3,Register tmp4)2102 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
2103 if (ProfileInterpreter && MethodData::profile_parameters()) {
2104 Label profile_continue, done;
2105
2106 test_method_data_pointer(profile_continue);
2107
2108 // Load the offset of the area within the MDO used for
2109 // parameters. If it's negative we're not profiling any parameters.
2110 lduw(ImethodDataPtr, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), tmp1);
2111 cmp_and_br_short(tmp1, 0, less, pn, profile_continue);
2112
2113 // Compute a pointer to the area for parameters from the offset
2114 // and move the pointer to the slot for the last
2115 // parameters. Collect profiling from last parameter down.
2116 // mdo start + parameters offset + array length - 1
2117
2118 // Pointer to the parameter area in the MDO
2119 Register mdp = tmp1;
2120 add(ImethodDataPtr, tmp1, mdp);
2121
2122 // offset of the current profile entry to update
2123 Register entry_offset = tmp2;
2124 // entry_offset = array len in number of cells
2125 ld_ptr(mdp, ArrayData::array_len_offset(), entry_offset);
2126
2127 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
2128 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
2129
2130 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
2131 sub(entry_offset, TypeStackSlotEntries::per_arg_count() - (off_base / DataLayout::cell_size), entry_offset);
2132 // entry_offset in bytes
2133 sll(entry_offset, exact_log2(DataLayout::cell_size), entry_offset);
2134
2135 Label loop;
2136 bind(loop);
2137
2138 // load offset on the stack from the slot for this parameter
2139 ld_ptr(mdp, entry_offset, tmp3);
2140 sll(tmp3,Interpreter::logStackElementSize, tmp3);
2141 neg(tmp3);
2142 // read the parameter from the local area
2143 ld_ptr(Llocals, tmp3, tmp3);
2144
2145 // make entry_offset now point to the type field for this parameter
2146 int type_base = in_bytes(ParametersTypeData::type_offset(0));
2147 assert(type_base > off_base, "unexpected");
2148 add(entry_offset, type_base - off_base, entry_offset);
2149
2150 // profile the parameter
2151 Address arg_type(mdp, entry_offset);
2152 profile_obj_type(tmp3, arg_type, tmp4);
2153
2154 // go to next parameter
2155 sub(entry_offset, TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base), entry_offset);
2156 cmp_and_br_short(entry_offset, off_base, greaterEqual, pt, loop);
2157
2158 bind(profile_continue);
2159 }
2160 }
2161
2162 // add a InterpMonitorElem to stack (see frame_sparc.hpp)
2163
add_monitor_to_stack(bool stack_is_empty,Register Rtemp,Register Rtemp2)2164 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty,
2165 Register Rtemp,
2166 Register Rtemp2 ) {
2167
2168 Register Rlimit = Lmonitors;
2169 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2170 assert( (delta & LongAlignmentMask) == 0,
2171 "sizeof BasicObjectLock must be even number of doublewords");
2172
2173 sub( SP, delta, SP);
2174 sub( Lesp, delta, Lesp);
2175 sub( Lmonitors, delta, Lmonitors);
2176
2177 if (!stack_is_empty) {
2178
2179 // must copy stack contents down
2180
2181 Label start_copying, next;
2182
2183 // untested("monitor stack expansion");
2184 compute_stack_base(Rtemp);
2185 ba(start_copying);
2186 delayed()->cmp(Rtemp, Rlimit); // done? duplicated below
2187
2188 // note: must copy from low memory upwards
2189 // On entry to loop,
2190 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS)
2191 // Loop mutates Rtemp
2192
2193 bind( next);
2194
2195 st_ptr(Rtemp2, Rtemp, 0);
2196 inc(Rtemp, wordSize);
2197 cmp(Rtemp, Rlimit); // are we done? (duplicated above)
2198
2199 bind( start_copying );
2200
2201 brx( notEqual, true, pn, next );
2202 delayed()->ld_ptr( Rtemp, delta, Rtemp2 );
2203
2204 // done copying stack
2205 }
2206 }
2207
2208 // Locals
access_local_ptr(Register index,Register dst)2209 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) {
2210 assert_not_delayed();
2211 sll(index, Interpreter::logStackElementSize, index);
2212 sub(Llocals, index, index);
2213 ld_ptr(index, 0, dst);
2214 // Note: index must hold the effective address--the iinc template uses it
2215 }
2216
2217 // Just like access_local_ptr but the tag is a returnAddress
access_local_returnAddress(Register index,Register dst)2218 void InterpreterMacroAssembler::access_local_returnAddress(Register index,
2219 Register dst ) {
2220 assert_not_delayed();
2221 sll(index, Interpreter::logStackElementSize, index);
2222 sub(Llocals, index, index);
2223 ld_ptr(index, 0, dst);
2224 }
2225
access_local_int(Register index,Register dst)2226 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) {
2227 assert_not_delayed();
2228 sll(index, Interpreter::logStackElementSize, index);
2229 sub(Llocals, index, index);
2230 ld(index, 0, dst);
2231 // Note: index must hold the effective address--the iinc template uses it
2232 }
2233
2234
access_local_long(Register index,Register dst)2235 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) {
2236 assert_not_delayed();
2237 sll(index, Interpreter::logStackElementSize, index);
2238 sub(Llocals, index, index);
2239 // First half stored at index n+1 (which grows down from Llocals[n])
2240 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst);
2241 }
2242
2243
access_local_float(Register index,FloatRegister dst)2244 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) {
2245 assert_not_delayed();
2246 sll(index, Interpreter::logStackElementSize, index);
2247 sub(Llocals, index, index);
2248 ldf(FloatRegisterImpl::S, index, 0, dst);
2249 }
2250
2251
access_local_double(Register index,FloatRegister dst)2252 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) {
2253 assert_not_delayed();
2254 sll(index, Interpreter::logStackElementSize, index);
2255 sub(Llocals, index, index);
2256 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst);
2257 }
2258
2259
2260 #ifdef ASSERT
check_for_regarea_stomp(Register Rindex,int offset,Register Rlimit,Register Rscratch,Register Rscratch1)2261 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) {
2262 Label L;
2263
2264 assert(Rindex != Rscratch, "Registers cannot be same");
2265 assert(Rindex != Rscratch1, "Registers cannot be same");
2266 assert(Rlimit != Rscratch, "Registers cannot be same");
2267 assert(Rlimit != Rscratch1, "Registers cannot be same");
2268 assert(Rscratch1 != Rscratch, "Registers cannot be same");
2269
2270 // untested("reg area corruption");
2271 add(Rindex, offset, Rscratch);
2272 add(Rlimit, 64 + STACK_BIAS, Rscratch1);
2273 cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L);
2274 stop("regsave area is being clobbered");
2275 bind(L);
2276 }
2277 #endif // ASSERT
2278
2279
store_local_int(Register index,Register src)2280 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) {
2281 assert_not_delayed();
2282 sll(index, Interpreter::logStackElementSize, index);
2283 sub(Llocals, index, index);
2284 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);)
2285 st(src, index, 0);
2286 }
2287
store_local_ptr(Register index,Register src)2288 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) {
2289 assert_not_delayed();
2290 sll(index, Interpreter::logStackElementSize, index);
2291 sub(Llocals, index, index);
2292 #ifdef ASSERT
2293 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2294 #endif
2295 st_ptr(src, index, 0);
2296 }
2297
2298
2299
store_local_ptr(int n,Register src)2300 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) {
2301 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n));
2302 }
2303
store_local_long(Register index,Register src)2304 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) {
2305 assert_not_delayed();
2306 sll(index, Interpreter::logStackElementSize, index);
2307 sub(Llocals, index, index);
2308 #ifdef ASSERT
2309 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2310 #endif
2311 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1
2312 }
2313
2314
store_local_float(Register index,FloatRegister src)2315 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) {
2316 assert_not_delayed();
2317 sll(index, Interpreter::logStackElementSize, index);
2318 sub(Llocals, index, index);
2319 #ifdef ASSERT
2320 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2321 #endif
2322 stf(FloatRegisterImpl::S, src, index, 0);
2323 }
2324
2325
store_local_double(Register index,FloatRegister src)2326 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) {
2327 assert_not_delayed();
2328 sll(index, Interpreter::logStackElementSize, index);
2329 sub(Llocals, index, index);
2330 #ifdef ASSERT
2331 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2332 #endif
2333 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1));
2334 }
2335
2336
top_most_monitor_byte_offset()2337 int InterpreterMacroAssembler::top_most_monitor_byte_offset() {
2338 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2339 int rounded_vm_local_words = align_up((int)frame::interpreter_frame_vm_local_words, WordsPerLong);
2340 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS;
2341 }
2342
2343
top_most_monitor()2344 Address InterpreterMacroAssembler::top_most_monitor() {
2345 return Address(FP, top_most_monitor_byte_offset());
2346 }
2347
2348
compute_stack_base(Register Rdest)2349 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) {
2350 add( Lesp, wordSize, Rdest );
2351 }
2352
get_method_counters(Register method,Register Rcounters,Label & skip)2353 void InterpreterMacroAssembler::get_method_counters(Register method,
2354 Register Rcounters,
2355 Label& skip) {
2356 Label has_counters;
2357 Address method_counters(method, in_bytes(Method::method_counters_offset()));
2358 ld_ptr(method_counters, Rcounters);
2359 br_notnull_short(Rcounters, Assembler::pt, has_counters);
2360 call_VM(noreg, CAST_FROM_FN_PTR(address,
2361 InterpreterRuntime::build_method_counters), method);
2362 ld_ptr(method_counters, Rcounters);
2363 br_null(Rcounters, false, Assembler::pn, skip); // No MethodCounters, OutOfMemory
2364 delayed()->nop();
2365 bind(has_counters);
2366 }
2367
increment_invocation_counter(Register Rcounters,Register Rtmp,Register Rtmp2)2368 void InterpreterMacroAssembler::increment_invocation_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2369 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful");
2370 assert_different_registers(Rcounters, Rtmp, Rtmp2);
2371
2372 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2373 InvocationCounter::counter_offset());
2374 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2375 InvocationCounter::counter_offset());
2376 int delta = InvocationCounter::count_increment;
2377
2378 // Load each counter in a register
2379 ld( inv_counter, Rtmp );
2380 ld( be_counter, Rtmp2 );
2381
2382 assert( is_simm13( delta ), " delta too large.");
2383
2384 // Add the delta to the invocation counter and store the result
2385 add( Rtmp, delta, Rtmp );
2386
2387 // Mask the backedge counter
2388 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2389
2390 // Store value
2391 st( Rtmp, inv_counter);
2392
2393 // Add invocation counter + backedge counter
2394 add( Rtmp, Rtmp2, Rtmp);
2395
2396 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2397 }
2398
increment_backedge_counter(Register Rcounters,Register Rtmp,Register Rtmp2)2399 void InterpreterMacroAssembler::increment_backedge_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2400 assert(UseCompiler, "incrementing must be useful");
2401 assert_different_registers(Rcounters, Rtmp, Rtmp2);
2402
2403 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2404 InvocationCounter::counter_offset());
2405 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2406 InvocationCounter::counter_offset());
2407
2408 int delta = InvocationCounter::count_increment;
2409 // Load each counter in a register
2410 ld( be_counter, Rtmp );
2411 ld( inv_counter, Rtmp2 );
2412
2413 // Add the delta to the backedge counter
2414 add( Rtmp, delta, Rtmp );
2415
2416 // Mask the invocation counter, add to backedge counter
2417 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2418
2419 // and store the result to memory
2420 st( Rtmp, be_counter );
2421
2422 // Add backedge + invocation counter
2423 add( Rtmp, Rtmp2, Rtmp );
2424
2425 // Note that this macro must leave backedge_count + invocation_count in Rtmp!
2426 }
2427
test_backedge_count_for_osr(Register backedge_count,Register method_counters,Register branch_bcp,Register Rtmp)2428 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count,
2429 Register method_counters,
2430 Register branch_bcp,
2431 Register Rtmp ) {
2432 Label did_not_overflow;
2433 Label overflow_with_error;
2434 assert_different_registers(backedge_count, Rtmp, branch_bcp);
2435 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
2436
2437 Address limit(method_counters, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()));
2438 ld(limit, Rtmp);
2439 cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow);
2440
2441 // When ProfileInterpreter is on, the backedge_count comes from the
2442 // MethodData*, which value does not get reset on the call to
2443 // frequency_counter_overflow(). To avoid excessive calls to the overflow
2444 // routine while the method is being compiled, add a second test to make sure
2445 // the overflow function is called only once every overflow_frequency.
2446 if (ProfileInterpreter) {
2447 const int overflow_frequency = 1024;
2448 andcc(backedge_count, overflow_frequency-1, Rtmp);
2449 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow);
2450 delayed()->nop();
2451 }
2452
2453 // overflow in loop, pass branch bytecode
2454 set(6,Rtmp);
2455 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp);
2456
2457 // Was an OSR adapter generated?
2458 // O0 = osr nmethod
2459 br_null_short(O0, Assembler::pn, overflow_with_error);
2460
2461 // Has the nmethod been invalidated already?
2462 ldub(O0, nmethod::state_offset(), O2);
2463 cmp_and_br_short(O2, nmethod::in_use, Assembler::notEqual, Assembler::pn, overflow_with_error);
2464
2465 // migrate the interpreter frame off of the stack
2466
2467 mov(G2_thread, L7);
2468 // save nmethod
2469 mov(O0, L6);
2470 set_last_Java_frame(SP, noreg);
2471 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
2472 reset_last_Java_frame();
2473 mov(L7, G2_thread);
2474
2475 // move OSR nmethod to I1
2476 mov(L6, I1);
2477
2478 // OSR buffer to I0
2479 mov(O0, I0);
2480
2481 // remove the interpreter frame
2482 restore(I5_savedSP, 0, SP);
2483
2484 // Jump to the osr code.
2485 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
2486 jmp(O2, G0);
2487 delayed()->nop();
2488
2489 bind(overflow_with_error);
2490
2491 bind(did_not_overflow);
2492 }
2493
2494
2495
interp_verify_oop(Register reg,TosState state,const char * file,int line)2496 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) {
2497 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); }
2498 }
2499
2500
2501 // local helper function for the verify_oop_or_return_address macro
verify_return_address(Method * m,int bci)2502 static bool verify_return_address(Method* m, int bci) {
2503 #ifndef PRODUCT
2504 address pc = (address)(m->constMethod())
2505 + in_bytes(ConstMethod::codes_offset()) + bci;
2506 // assume it is a valid return address if it is inside m and is preceded by a jsr
2507 if (!m->contains(pc)) return false;
2508 address jsr_pc;
2509 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2510 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2511 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2512 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2513 #endif // PRODUCT
2514 return false;
2515 }
2516
2517
verify_oop_or_return_address(Register reg,Register Rtmp)2518 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2519 if (!VerifyOops) return;
2520 // the VM documentation for the astore[_wide] bytecode allows
2521 // the TOS to be not only an oop but also a return address
2522 Label test;
2523 Label skip;
2524 // See if it is an address (in the current method):
2525
2526 mov(reg, Rtmp);
2527 const int log2_bytecode_size_limit = 16;
2528 srl(Rtmp, log2_bytecode_size_limit, Rtmp);
2529 br_notnull_short( Rtmp, pt, test );
2530
2531 // %%% should use call_VM_leaf here?
2532 save_frame_and_mov(0, Lmethod, O0, reg, O1);
2533 save_thread(L7_thread_cache);
2534 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none);
2535 delayed()->nop();
2536 restore_thread(L7_thread_cache);
2537 br_notnull( O0, false, pt, skip );
2538 delayed()->restore();
2539
2540 // Perform a more elaborate out-of-line call
2541 // Not an address; verify it:
2542 bind(test);
2543 verify_oop(reg);
2544 bind(skip);
2545 }
2546
2547
verify_FPU(int stack_depth,TosState state)2548 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2549 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth);
2550 }
2551
2552
2553 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
increment_mask_and_jump(Address counter_addr,int increment,Address mask_addr,Register scratch1,Register scratch2,Condition cond,Label * where)2554 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2555 int increment, Address mask_addr,
2556 Register scratch1, Register scratch2,
2557 Condition cond, Label *where) {
2558 ld(counter_addr, scratch1);
2559 add(scratch1, increment, scratch1);
2560 ld(mask_addr, scratch2);
2561 andcc(scratch1, scratch2, G0);
2562 br(cond, false, Assembler::pn, *where);
2563 delayed()->st(scratch1, counter_addr);
2564 }
2565
2566 // Inline assembly for:
2567 //
2568 // if (thread is in interp_only_mode) {
2569 // InterpreterRuntime::post_method_entry();
2570 // }
2571 // if (DTraceMethodProbes) {
2572 // SharedRuntime::dtrace_method_entry(method, receiver);
2573 // }
2574 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2575 // SharedRuntime::rc_trace_method_entry(method, receiver);
2576 // }
2577
notify_method_entry()2578 void InterpreterMacroAssembler::notify_method_entry() {
2579
2580 // Whenever JVMTI puts a thread in interp_only_mode, method
2581 // entry/exit events are sent for that thread to track stack
2582 // depth. If it is possible to enter interp_only_mode we add
2583 // the code to check if the event should be sent.
2584 if (JvmtiExport::can_post_interpreter_events()) {
2585 Label L;
2586 Register temp_reg = O5;
2587 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2588 ld(interp_only, temp_reg);
2589 cmp_and_br_short(temp_reg, 0, equal, pt, L);
2590 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2591 bind(L);
2592 }
2593
2594 {
2595 Register temp_reg = O5;
2596 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2597 call_VM_leaf(noreg,
2598 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2599 G2_thread, Lmethod);
2600 }
2601
2602 // RedefineClasses() tracing support for obsolete method entry
2603 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2604 call_VM_leaf(noreg,
2605 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2606 G2_thread, Lmethod);
2607 }
2608 }
2609
2610
2611 // Inline assembly for:
2612 //
2613 // if (thread is in interp_only_mode) {
2614 // // save result
2615 // InterpreterRuntime::post_method_exit();
2616 // // restore result
2617 // }
2618 // if (DTraceMethodProbes) {
2619 // SharedRuntime::dtrace_method_exit(thread, method);
2620 // }
2621 //
2622 // Native methods have their result stored in d_tmp and l_tmp
2623 // Java methods have their result stored in the expression stack
2624
notify_method_exit(bool is_native_method,TosState state,NotifyMethodExitMode mode)2625 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method,
2626 TosState state,
2627 NotifyMethodExitMode mode) {
2628
2629 // Whenever JVMTI puts a thread in interp_only_mode, method
2630 // entry/exit events are sent for that thread to track stack
2631 // depth. If it is possible to enter interp_only_mode we add
2632 // the code to check if the event should be sent.
2633 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2634 Label L;
2635 Register temp_reg = O5;
2636 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2637 ld(interp_only, temp_reg);
2638 cmp_and_br_short(temp_reg, 0, equal, pt, L);
2639
2640 // Note: frame::interpreter_frame_result has a dependency on how the
2641 // method result is saved across the call to post_method_exit. For
2642 // native methods it assumes the result registers are saved to
2643 // l_scratch and d_scratch. If this changes then the interpreter_frame_result
2644 // implementation will need to be updated too.
2645
2646 save_return_value(state, is_native_method);
2647 call_VM(noreg,
2648 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2649 restore_return_value(state, is_native_method);
2650 bind(L);
2651 }
2652
2653 {
2654 Register temp_reg = O5;
2655 // Dtrace notification
2656 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2657 save_return_value(state, is_native_method);
2658 call_VM_leaf(
2659 noreg,
2660 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2661 G2_thread, Lmethod);
2662 restore_return_value(state, is_native_method);
2663 }
2664 }
2665
save_return_value(TosState state,bool is_native_call)2666 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) {
2667 if (is_native_call) {
2668 stf(FloatRegisterImpl::D, F0, d_tmp);
2669 stx(O0, l_tmp);
2670 } else {
2671 push(state);
2672 }
2673 }
2674
restore_return_value(TosState state,bool is_native_call)2675 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) {
2676 if (is_native_call) {
2677 ldf(FloatRegisterImpl::D, d_tmp, F0);
2678 ldx(l_tmp, O0);
2679 } else {
2680 pop(state);
2681 }
2682 }
2683