1 /*
2 * Copyright (c) 1999, 2017, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/assembler.hpp"
28 #include "c1/c1_CodeStubs.hpp"
29 #include "c1/c1_Defs.hpp"
30 #include "c1/c1_MacroAssembler.hpp"
31 #include "c1/c1_Runtime1.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "gc/shared/cardTable.hpp"
34 #include "gc/shared/cardTableModRefBS.hpp"
35 #include "interpreter/interpreter.hpp"
36 #include "nativeInst_aarch64.hpp"
37 #include "oops/compiledICHolder.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "prims/jvmtiExport.hpp"
40 #include "register_aarch64.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/signature.hpp"
43 #include "runtime/vframe.hpp"
44 #include "runtime/vframeArray.hpp"
45 #include "vmreg_aarch64.inline.hpp"
46 #if INCLUDE_ALL_GCS
47 #include "gc/g1/g1CardTable.hpp"
48 #include "gc/g1/g1BarrierSet.hpp"
49 #endif
50
51
52 // Implementation of StubAssembler
53
54 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, int args_size) {
55 // setup registers
56 assert(!(oop_result1->is_valid() || metadata_result->is_valid()) || oop_result1 != metadata_result, "registers must be different");
57 assert(oop_result1 != rthread && metadata_result != rthread, "registers must be different");
58 assert(args_size >= 0, "illegal args_size");
59 bool align_stack = false;
60
61 mov(c_rarg0, rthread);
62 set_num_rt_args(0); // Nothing on stack
63
64 Label retaddr;
65 set_last_Java_frame(sp, rfp, retaddr, rscratch1);
66
67 // do the call
68 lea(rscratch1, RuntimeAddress(entry));
69 blrt(rscratch1, args_size + 1, 8, 1);
70 bind(retaddr);
71 int call_offset = offset();
72 // verify callee-saved register
73 #ifdef ASSERT
74 push(r0, sp);
75 { Label L;
76 get_thread(r0);
77 cmp(rthread, r0);
78 br(Assembler::EQ, L);
79 stop("StubAssembler::call_RT: rthread not callee saved?");
80 bind(L);
81 }
82 pop(r0, sp);
83 #endif
84 reset_last_Java_frame(true);
85 maybe_isb();
86
87 // check for pending exceptions
88 { Label L;
89 // check for pending exceptions (java_thread is set upon return)
90 ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
91 cbz(rscratch1, L);
92 // exception pending => remove activation and forward to exception handler
93 // make sure that the vm_results are cleared
94 if (oop_result1->is_valid()) {
95 str(zr, Address(rthread, JavaThread::vm_result_offset()));
96 }
97 if (metadata_result->is_valid()) {
98 str(zr, Address(rthread, JavaThread::vm_result_2_offset()));
99 }
100 if (frame_size() == no_frame_size) {
101 leave();
102 far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
103 } else if (_stub_id == Runtime1::forward_exception_id) {
104 should_not_reach_here();
105 } else {
106 far_jump(RuntimeAddress(Runtime1::entry_for(Runtime1::forward_exception_id)));
107 }
108 bind(L);
109 }
110 // get oop results if there are any and reset the values in the thread
111 if (oop_result1->is_valid()) {
112 get_vm_result(oop_result1, rthread);
113 }
114 if (metadata_result->is_valid()) {
115 get_vm_result_2(metadata_result, rthread);
116 }
117 return call_offset;
118 }
119
120
121 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1) {
122 mov(c_rarg1, arg1);
123 return call_RT(oop_result1, metadata_result, entry, 1);
124 }
125
126
127 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2) {
128 if (c_rarg1 == arg2) {
129 if (c_rarg2 == arg1) {
130 mov(rscratch1, arg1);
131 mov(arg1, arg2);
132 mov(arg2, rscratch1);
133 } else {
134 mov(c_rarg2, arg2);
135 mov(c_rarg1, arg1);
136 }
137 } else {
138 mov(c_rarg1, arg1);
139 mov(c_rarg2, arg2);
140 }
141 return call_RT(oop_result1, metadata_result, entry, 2);
142 }
143
144
145 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2, Register arg3) {
146 // if there is any conflict use the stack
147 if (arg1 == c_rarg2 || arg1 == c_rarg3 ||
148 arg2 == c_rarg1 || arg1 == c_rarg3 ||
149 arg3 == c_rarg1 || arg1 == c_rarg2) {
150 stp(arg3, arg2, Address(pre(sp, 2 * wordSize)));
151 stp(arg1, zr, Address(pre(sp, -2 * wordSize)));
152 ldp(c_rarg1, zr, Address(post(sp, 2 * wordSize)));
153 ldp(c_rarg3, c_rarg2, Address(post(sp, 2 * wordSize)));
154 } else {
155 mov(c_rarg1, arg1);
156 mov(c_rarg2, arg2);
157 mov(c_rarg3, arg3);
158 }
159 return call_RT(oop_result1, metadata_result, entry, 3);
160 }
161
162 // Implementation of StubFrame
163
164 class StubFrame: public StackObj {
165 private:
166 StubAssembler* _sasm;
167
168 public:
169 StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments);
170 void load_argument(int offset_in_words, Register reg);
171
172 ~StubFrame();
173 };;
174
175 void StubAssembler::prologue(const char* name, bool must_gc_arguments) {
176 set_info(name, must_gc_arguments);
177 enter();
178 }
179
180 void StubAssembler::epilogue() {
181 leave();
182 ret(lr);
183 }
184
185 #define __ _sasm->
186
187 StubFrame::StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments) {
188 _sasm = sasm;
189 __ prologue(name, must_gc_arguments);
190 }
191
192 // load parameters that were stored with LIR_Assembler::store_parameter
193 // Note: offsets for store_parameter and load_argument must match
194 void StubFrame::load_argument(int offset_in_words, Register reg) {
195 __ load_parameter(offset_in_words, reg);
196 }
197
198
199 StubFrame::~StubFrame() {
200 __ epilogue();
201 }
202
203 #undef __
204
205
206 // Implementation of Runtime1
207
208 #define __ sasm->
209
210 const int float_regs_as_doubles_size_in_slots = pd_nof_fpu_regs_frame_map * 2;
211
212 // Stack layout for saving/restoring all the registers needed during a runtime
213 // call (this includes deoptimization)
214 // Note: note that users of this frame may well have arguments to some runtime
215 // while these values are on the stack. These positions neglect those arguments
216 // but the code in save_live_registers will take the argument count into
217 // account.
218 //
219
220 enum reg_save_layout {
221 reg_save_frame_size = 32 /* float */ + 32 /* integer */
222 };
223
224 // Save off registers which might be killed by calls into the runtime.
225 // Tries to smart of about FP registers. In particular we separate
226 // saving and describing the FPU registers for deoptimization since we
227 // have to save the FPU registers twice if we describe them. The
228 // deopt blob is the only thing which needs to describe FPU registers.
229 // In all other cases it should be sufficient to simply save their
230 // current value.
231
232 static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs];
233 static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs];
234 static int reg_save_size_in_words;
235 static int frame_size_in_bytes = -1;
236
237 static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) {
238 int frame_size_in_bytes = reg_save_frame_size * BytesPerWord;
239 sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
240 int frame_size_in_slots = frame_size_in_bytes / sizeof(jint);
241 OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
242
243 for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {
244 Register r = as_Register(i);
245 if (i <= 18 && i != rscratch1->encoding() && i != rscratch2->encoding()) {
246 int sp_offset = cpu_reg_save_offsets[i];
247 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
248 r->as_VMReg());
249 }
250 }
251
252 if (save_fpu_registers) {
253 for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
254 FloatRegister r = as_FloatRegister(i);
255 {
256 int sp_offset = fpu_reg_save_offsets[i];
257 oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
258 r->as_VMReg());
259 }
260 }
261 }
262 return oop_map;
263 }
264
265 static OopMap* save_live_registers(StubAssembler* sasm,
266 bool save_fpu_registers = true) {
267 __ block_comment("save_live_registers");
268
269 __ push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
270
271 if (save_fpu_registers) {
272 for (int i = 30; i >= 0; i -= 2)
273 __ stpd(as_FloatRegister(i), as_FloatRegister(i+1),
274 Address(__ pre(sp, -2 * wordSize)));
275 } else {
276 __ add(sp, sp, -32 * wordSize);
277 }
278
279 return generate_oop_map(sasm, save_fpu_registers);
280 }
281
282 static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
283 if (restore_fpu_registers) {
284 for (int i = 0; i < 32; i += 2)
285 __ ldpd(as_FloatRegister(i), as_FloatRegister(i+1),
286 Address(__ post(sp, 2 * wordSize)));
287 } else {
288 __ add(sp, sp, 32 * wordSize);
289 }
290
291 __ pop(RegSet::range(r0, r29), sp);
292 }
293
294 static void restore_live_registers_except_r0(StubAssembler* sasm, bool restore_fpu_registers = true) {
295
296 if (restore_fpu_registers) {
297 for (int i = 0; i < 32; i += 2)
298 __ ldpd(as_FloatRegister(i), as_FloatRegister(i+1),
299 Address(__ post(sp, 2 * wordSize)));
300 } else {
301 __ add(sp, sp, 32 * wordSize);
302 }
303
304 __ ldp(zr, r1, Address(__ post(sp, 16)));
305 __ pop(RegSet::range(r2, r29), sp);
306 }
307
308
309
310 void Runtime1::initialize_pd() {
311 int i;
312 int sp_offset = 0;
313
314 // all float registers are saved explicitly
315 assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here");
316 for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
317 fpu_reg_save_offsets[i] = sp_offset;
318 sp_offset += 2; // SP offsets are in halfwords
319 }
320
321 for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
322 Register r = as_Register(i);
323 cpu_reg_save_offsets[i] = sp_offset;
324 sp_offset += 2; // SP offsets are in halfwords
325 }
326 }
327
328
329 // target: the entry point of the method that creates and posts the exception oop
330 // has_argument: true if the exception needs an argument (passed in rscratch1)
331
332 OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
333 // make a frame and preserve the caller's caller-save registers
334 OopMap* oop_map = save_live_registers(sasm);
335 int call_offset;
336 if (!has_argument) {
337 call_offset = __ call_RT(noreg, noreg, target);
338 } else {
339 call_offset = __ call_RT(noreg, noreg, target, rscratch1);
340 }
341 OopMapSet* oop_maps = new OopMapSet();
342 oop_maps->add_gc_map(call_offset, oop_map);
343
344 __ should_not_reach_here();
345 return oop_maps;
346 }
347
348
349 OopMapSet* Runtime1::generate_handle_exception(StubID id, StubAssembler *sasm) {
350 __ block_comment("generate_handle_exception");
351
352 // incoming parameters
353 const Register exception_oop = r0;
354 const Register exception_pc = r3;
355 // other registers used in this stub
356
357 // Save registers, if required.
358 OopMapSet* oop_maps = new OopMapSet();
359 OopMap* oop_map = NULL;
360 switch (id) {
361 case forward_exception_id:
362 // We're handling an exception in the context of a compiled frame.
363 // The registers have been saved in the standard places. Perform
364 // an exception lookup in the caller and dispatch to the handler
365 // if found. Otherwise unwind and dispatch to the callers
366 // exception handler.
367 oop_map = generate_oop_map(sasm, 1 /*thread*/);
368
369 // load and clear pending exception oop into r0
370 __ ldr(exception_oop, Address(rthread, Thread::pending_exception_offset()));
371 __ str(zr, Address(rthread, Thread::pending_exception_offset()));
372
373 // load issuing PC (the return address for this stub) into r3
374 __ ldr(exception_pc, Address(rfp, 1*BytesPerWord));
375
376 // make sure that the vm_results are cleared (may be unnecessary)
377 __ str(zr, Address(rthread, JavaThread::vm_result_offset()));
378 __ str(zr, Address(rthread, JavaThread::vm_result_2_offset()));
379 break;
380 case handle_exception_nofpu_id:
381 case handle_exception_id:
382 // At this point all registers MAY be live.
383 oop_map = save_live_registers(sasm, id != handle_exception_nofpu_id);
384 break;
385 case handle_exception_from_callee_id: {
386 // At this point all registers except exception oop (r0) and
387 // exception pc (lr) are dead.
388 const int frame_size = 2 /*fp, return address*/;
389 oop_map = new OopMap(frame_size * VMRegImpl::slots_per_word, 0);
390 sasm->set_frame_size(frame_size);
391 break;
392 }
393 default:
394 __ should_not_reach_here();
395 break;
396 }
397
398 // verify that only r0 and r3 are valid at this time
399 __ invalidate_registers(false, true, true, false, true, true);
400 // verify that r0 contains a valid exception
401 __ verify_not_null_oop(exception_oop);
402
403 #ifdef ASSERT
404 // check that fields in JavaThread for exception oop and issuing pc are
405 // empty before writing to them
406 Label oop_empty;
407 __ ldr(rscratch1, Address(rthread, JavaThread::exception_oop_offset()));
408 __ cbz(rscratch1, oop_empty);
409 __ stop("exception oop already set");
410 __ bind(oop_empty);
411
412 Label pc_empty;
413 __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
414 __ cbz(rscratch1, pc_empty);
415 __ stop("exception pc already set");
416 __ bind(pc_empty);
417 #endif
418
419 // save exception oop and issuing pc into JavaThread
420 // (exception handler will load it from here)
421 __ str(exception_oop, Address(rthread, JavaThread::exception_oop_offset()));
422 __ str(exception_pc, Address(rthread, JavaThread::exception_pc_offset()));
423
424 // patch throwing pc into return address (has bci & oop map)
425 __ str(exception_pc, Address(rfp, 1*BytesPerWord));
426
427 // compute the exception handler.
428 // the exception oop and the throwing pc are read from the fields in JavaThread
429 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
430 oop_maps->add_gc_map(call_offset, oop_map);
431
432 // r0: handler address
433 // will be the deopt blob if nmethod was deoptimized while we looked up
434 // handler regardless of whether handler existed in the nmethod.
435
436 // only r0 is valid at this time, all other registers have been destroyed by the runtime call
437 __ invalidate_registers(false, true, true, true, true, true);
438
439 // patch the return address, this stub will directly return to the exception handler
440 __ str(r0, Address(rfp, 1*BytesPerWord));
441
442 switch (id) {
443 case forward_exception_id:
444 case handle_exception_nofpu_id:
445 case handle_exception_id:
446 // Restore the registers that were saved at the beginning.
447 restore_live_registers(sasm, id != handle_exception_nofpu_id);
448 break;
449 case handle_exception_from_callee_id:
450 // Pop the return address.
451 __ leave();
452 __ ret(lr); // jump to exception handler
453 break;
454 default: ShouldNotReachHere();
455 }
456
457 return oop_maps;
458 }
459
460
461 void Runtime1::generate_unwind_exception(StubAssembler *sasm) {
462 // incoming parameters
463 const Register exception_oop = r0;
464 // callee-saved copy of exception_oop during runtime call
465 const Register exception_oop_callee_saved = r19;
466 // other registers used in this stub
467 const Register exception_pc = r3;
468 const Register handler_addr = r1;
469
470 // verify that only r0, is valid at this time
471 __ invalidate_registers(false, true, true, true, true, true);
472
473 #ifdef ASSERT
474 // check that fields in JavaThread for exception oop and issuing pc are empty
475 Label oop_empty;
476 __ ldr(rscratch1, Address(rthread, JavaThread::exception_oop_offset()));
477 __ cbz(rscratch1, oop_empty);
478 __ stop("exception oop must be empty");
479 __ bind(oop_empty);
480
481 Label pc_empty;
482 __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
483 __ cbz(rscratch1, pc_empty);
484 __ stop("exception pc must be empty");
485 __ bind(pc_empty);
486 #endif
487
488 // Save our return address because
489 // exception_handler_for_return_address will destroy it. We also
490 // save exception_oop
491 __ stp(lr, exception_oop, Address(__ pre(sp, -2 * wordSize)));
492
493 // search the exception handler address of the caller (using the return address)
494 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rthread, lr);
495 // r0: exception handler address of the caller
496
497 // Only R0 is valid at this time; all other registers have been
498 // destroyed by the call.
499 __ invalidate_registers(false, true, true, true, false, true);
500
501 // move result of call into correct register
502 __ mov(handler_addr, r0);
503
504 // get throwing pc (= return address).
505 // lr has been destroyed by the call
506 __ ldp(lr, exception_oop, Address(__ post(sp, 2 * wordSize)));
507 __ mov(r3, lr);
508
509 __ verify_not_null_oop(exception_oop);
510
511 // continue at exception handler (return address removed)
512 // note: do *not* remove arguments when unwinding the
513 // activation since the caller assumes having
514 // all arguments on the stack when entering the
515 // runtime to determine the exception handler
516 // (GC happens at call site with arguments!)
517 // r0: exception oop
518 // r3: throwing pc
519 // r1: exception handler
520 __ br(handler_addr);
521 }
522
523
524
525 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
526 // use the maximum number of runtime-arguments here because it is difficult to
527 // distinguish each RT-Call.
528 // Note: This number affects also the RT-Call in generate_handle_exception because
529 // the oop-map is shared for all calls.
530 DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
531 assert(deopt_blob != NULL, "deoptimization blob must have been created");
532
533 OopMap* oop_map = save_live_registers(sasm);
534
535 __ mov(c_rarg0, rthread);
536 Label retaddr;
537 __ set_last_Java_frame(sp, rfp, retaddr, rscratch1);
538 // do the call
539 __ lea(rscratch1, RuntimeAddress(target));
540 __ blrt(rscratch1, 1, 0, 1);
541 __ bind(retaddr);
542 OopMapSet* oop_maps = new OopMapSet();
543 oop_maps->add_gc_map(__ offset(), oop_map);
544 // verify callee-saved register
545 #ifdef ASSERT
546 { Label L;
547 __ get_thread(rscratch1);
548 __ cmp(rthread, rscratch1);
549 __ br(Assembler::EQ, L);
550 __ stop("StubAssembler::call_RT: rthread not callee saved?");
551 __ bind(L);
552 }
553 #endif
554 __ reset_last_Java_frame(true);
555 __ maybe_isb();
556
557 // check for pending exceptions
558 { Label L;
559 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
560 __ cbz(rscratch1, L);
561 // exception pending => remove activation and forward to exception handler
562
563 { Label L1;
564 __ cbnz(r0, L1); // have we deoptimized?
565 __ far_jump(RuntimeAddress(Runtime1::entry_for(Runtime1::forward_exception_id)));
566 __ bind(L1);
567 }
568
569 // the deopt blob expects exceptions in the special fields of
570 // JavaThread, so copy and clear pending exception.
571
572 // load and clear pending exception
573 __ ldr(r0, Address(rthread, Thread::pending_exception_offset()));
574 __ str(zr, Address(rthread, Thread::pending_exception_offset()));
575
576 // check that there is really a valid exception
577 __ verify_not_null_oop(r0);
578
579 // load throwing pc: this is the return address of the stub
580 __ mov(r3, lr);
581
582 #ifdef ASSERT
583 // check that fields in JavaThread for exception oop and issuing pc are empty
584 Label oop_empty;
585 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
586 __ cbz(rscratch1, oop_empty);
587 __ stop("exception oop must be empty");
588 __ bind(oop_empty);
589
590 Label pc_empty;
591 __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
592 __ cbz(rscratch1, pc_empty);
593 __ stop("exception pc must be empty");
594 __ bind(pc_empty);
595 #endif
596
597 // store exception oop and throwing pc to JavaThread
598 __ str(r0, Address(rthread, JavaThread::exception_oop_offset()));
599 __ str(r3, Address(rthread, JavaThread::exception_pc_offset()));
600
601 restore_live_registers(sasm);
602
603 __ leave();
604
605 // Forward the exception directly to deopt blob. We can blow no
606 // registers and must leave throwing pc on the stack. A patch may
607 // have values live in registers so the entry point with the
608 // exception in tls.
609 __ far_jump(RuntimeAddress(deopt_blob->unpack_with_exception_in_tls()));
610
611 __ bind(L);
612 }
613
614
615 // Runtime will return true if the nmethod has been deoptimized during
616 // the patching process. In that case we must do a deopt reexecute instead.
617
618 Label reexecuteEntry, cont;
619
620 __ cbz(r0, cont); // have we deoptimized?
621
622 // Will reexecute. Proper return address is already on the stack we just restore
623 // registers, pop all of our frame but the return address and jump to the deopt blob
624 restore_live_registers(sasm);
625 __ leave();
626 __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
627
628 __ bind(cont);
629 restore_live_registers(sasm);
630 __ leave();
631 __ ret(lr);
632
633 return oop_maps;
634 }
635
636
637 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
638
639 const Register exception_oop = r0;
640 const Register exception_pc = r3;
641
642 // for better readability
643 const bool must_gc_arguments = true;
644 const bool dont_gc_arguments = false;
645
646 // default value; overwritten for some optimized stubs that are called from methods that do not use the fpu
647 bool save_fpu_registers = true;
648
649 // stub code & info for the different stubs
650 OopMapSet* oop_maps = NULL;
651 OopMap* oop_map = NULL;
652 switch (id) {
653 {
654 case forward_exception_id:
655 {
656 oop_maps = generate_handle_exception(id, sasm);
657 __ leave();
658 __ ret(lr);
659 }
660 break;
661
662 case throw_div0_exception_id:
663 { StubFrame f(sasm, "throw_div0_exception", dont_gc_arguments);
664 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
665 }
666 break;
667
668 case throw_null_pointer_exception_id:
669 { StubFrame f(sasm, "throw_null_pointer_exception", dont_gc_arguments);
670 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
671 }
672 break;
673
674 case new_instance_id:
675 case fast_new_instance_id:
676 case fast_new_instance_init_check_id:
677 {
678 Register klass = r3; // Incoming
679 Register obj = r0; // Result
680
681 if (id == new_instance_id) {
682 __ set_info("new_instance", dont_gc_arguments);
683 } else if (id == fast_new_instance_id) {
684 __ set_info("fast new_instance", dont_gc_arguments);
685 } else {
686 assert(id == fast_new_instance_init_check_id, "bad StubID");
687 __ set_info("fast new_instance init check", dont_gc_arguments);
688 }
689
690 if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
691 UseTLAB && FastTLABRefill) {
692 Label slow_path;
693 Register obj_size = r2;
694 Register t1 = r19;
695 Register t2 = r4;
696 assert_different_registers(klass, obj, obj_size, t1, t2);
697
698 __ stp(r5, r19, Address(__ pre(sp, -2 * wordSize)));
699
700 if (id == fast_new_instance_init_check_id) {
701 // make sure the klass is initialized
702 __ ldrb(rscratch1, Address(klass, InstanceKlass::init_state_offset()));
703 __ cmpw(rscratch1, InstanceKlass::fully_initialized);
704 __ br(Assembler::NE, slow_path);
705 }
706
707 #ifdef ASSERT
708 // assert object can be fast path allocated
709 {
710 Label ok, not_ok;
711 __ ldrw(obj_size, Address(klass, Klass::layout_helper_offset()));
712 __ cmp(obj_size, 0u);
713 __ br(Assembler::LE, not_ok); // make sure it's an instance (LH > 0)
714 __ tstw(obj_size, Klass::_lh_instance_slow_path_bit);
715 __ br(Assembler::EQ, ok);
716 __ bind(not_ok);
717 __ stop("assert(can be fast path allocated)");
718 __ should_not_reach_here();
719 __ bind(ok);
720 }
721 #endif // ASSERT
722
723 // if we got here then the TLAB allocation failed, so try
724 // refilling the TLAB or allocating directly from eden.
725 Label retry_tlab, try_eden;
726 __ tlab_refill(retry_tlab, try_eden, slow_path); // does not destroy r3 (klass), returns r5
727
728 __ bind(retry_tlab);
729
730 // get the instance size (size is postive so movl is fine for 64bit)
731 __ ldrw(obj_size, Address(klass, Klass::layout_helper_offset()));
732
733 __ tlab_allocate(obj, obj_size, 0, t1, t2, slow_path);
734
735 __ initialize_object(obj, klass, obj_size, 0, t1, t2, /* is_tlab_allocated */ true);
736 __ verify_oop(obj);
737 __ ldp(r5, r19, Address(__ post(sp, 2 * wordSize)));
738 __ ret(lr);
739
740 __ bind(try_eden);
741 // get the instance size (size is postive so movl is fine for 64bit)
742 __ ldrw(obj_size, Address(klass, Klass::layout_helper_offset()));
743
744 __ eden_allocate(obj, obj_size, 0, t1, slow_path);
745 __ incr_allocated_bytes(rthread, obj_size, 0, rscratch1);
746
747 __ initialize_object(obj, klass, obj_size, 0, t1, t2, /* is_tlab_allocated */ false);
748 __ verify_oop(obj);
749 __ ldp(r5, r19, Address(__ post(sp, 2 * wordSize)));
750 __ ret(lr);
751
752 __ bind(slow_path);
753 __ ldp(r5, r19, Address(__ post(sp, 2 * wordSize)));
754 }
755
756 __ enter();
757 OopMap* map = save_live_registers(sasm);
758 int call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_instance), klass);
759 oop_maps = new OopMapSet();
760 oop_maps->add_gc_map(call_offset, map);
761 restore_live_registers_except_r0(sasm);
762 __ verify_oop(obj);
763 __ leave();
764 __ ret(lr);
765
766 // r0,: new instance
767 }
768
769 break;
770
771 case counter_overflow_id:
772 {
773 Register bci = r0, method = r1;
774 __ enter();
775 OopMap* map = save_live_registers(sasm);
776 // Retrieve bci
777 __ ldrw(bci, Address(rfp, 2*BytesPerWord));
778 // And a pointer to the Method*
779 __ ldr(method, Address(rfp, 3*BytesPerWord));
780 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, counter_overflow), bci, method);
781 oop_maps = new OopMapSet();
782 oop_maps->add_gc_map(call_offset, map);
783 restore_live_registers(sasm);
784 __ leave();
785 __ ret(lr);
786 }
787 break;
788
789 case new_type_array_id:
790 case new_object_array_id:
791 {
792 Register length = r19; // Incoming
793 Register klass = r3; // Incoming
794 Register obj = r0; // Result
795
796 if (id == new_type_array_id) {
797 __ set_info("new_type_array", dont_gc_arguments);
798 } else {
799 __ set_info("new_object_array", dont_gc_arguments);
800 }
801
802 #ifdef ASSERT
803 // assert object type is really an array of the proper kind
804 {
805 Label ok;
806 Register t0 = obj;
807 __ ldrw(t0, Address(klass, Klass::layout_helper_offset()));
808 __ asrw(t0, t0, Klass::_lh_array_tag_shift);
809 int tag = ((id == new_type_array_id)
810 ? Klass::_lh_array_tag_type_value
811 : Klass::_lh_array_tag_obj_value);
812 __ mov(rscratch1, tag);
813 __ cmpw(t0, rscratch1);
814 __ br(Assembler::EQ, ok);
815 __ stop("assert(is an array klass)");
816 __ should_not_reach_here();
817 __ bind(ok);
818 }
819 #endif // ASSERT
820
821 if (UseTLAB && FastTLABRefill) {
822 Register arr_size = r4;
823 Register t1 = r2;
824 Register t2 = r5;
825 Label slow_path;
826 assert_different_registers(length, klass, obj, arr_size, t1, t2);
827
828 // check that array length is small enough for fast path.
829 __ mov(rscratch1, C1_MacroAssembler::max_array_allocation_length);
830 __ cmpw(length, rscratch1);
831 __ br(Assembler::HI, slow_path);
832
833 // if we got here then the TLAB allocation failed, so try
834 // refilling the TLAB or allocating directly from eden.
835 Label retry_tlab, try_eden;
836 const Register thread =
837 __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves r19 & r3, returns rthread
838
839 __ bind(retry_tlab);
840
841 // get the allocation size: round_up(hdr + length << (layout_helper & 0x1F))
842 // since size is positive ldrw does right thing on 64bit
843 __ ldrw(t1, Address(klass, Klass::layout_helper_offset()));
844 __ lslvw(arr_size, length, t1);
845 __ ubfx(t1, t1, Klass::_lh_header_size_shift,
846 exact_log2(Klass::_lh_header_size_mask + 1));
847 __ add(arr_size, arr_size, t1);
848 __ add(arr_size, arr_size, MinObjAlignmentInBytesMask); // align up
849 __ andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
850
851 __ tlab_allocate(obj, arr_size, 0, t1, t2, slow_path); // preserves arr_size
852
853 __ initialize_header(obj, klass, length, t1, t2);
854 __ ldrb(t1, Address(klass, in_bytes(Klass::layout_helper_offset()) + (Klass::_lh_header_size_shift / BitsPerByte)));
855 assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
856 assert(Klass::_lh_header_size_mask <= 0xFF, "bytewise");
857 __ andr(t1, t1, Klass::_lh_header_size_mask);
858 __ sub(arr_size, arr_size, t1); // body length
859 __ add(t1, t1, obj); // body start
860 if (!ZeroTLAB) {
861 __ initialize_body(t1, arr_size, 0, t2);
862 }
863 __ verify_oop(obj);
864
865 __ ret(lr);
866
867 __ bind(try_eden);
868 // get the allocation size: round_up(hdr + length << (layout_helper & 0x1F))
869 // since size is positive ldrw does right thing on 64bit
870 __ ldrw(t1, Address(klass, Klass::layout_helper_offset()));
871 // since size is postive movw does right thing on 64bit
872 __ movw(arr_size, length);
873 __ lslvw(arr_size, length, t1);
874 __ ubfx(t1, t1, Klass::_lh_header_size_shift,
875 exact_log2(Klass::_lh_header_size_mask + 1));
876 __ add(arr_size, arr_size, t1);
877 __ add(arr_size, arr_size, MinObjAlignmentInBytesMask); // align up
878 __ andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
879
880 __ eden_allocate(obj, arr_size, 0, t1, slow_path); // preserves arr_size
881 __ incr_allocated_bytes(thread, arr_size, 0, rscratch1);
882
883 __ initialize_header(obj, klass, length, t1, t2);
884 __ ldrb(t1, Address(klass, in_bytes(Klass::layout_helper_offset()) + (Klass::_lh_header_size_shift / BitsPerByte)));
885 assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
886 assert(Klass::_lh_header_size_mask <= 0xFF, "bytewise");
887 __ andr(t1, t1, Klass::_lh_header_size_mask);
888 __ sub(arr_size, arr_size, t1); // body length
889 __ add(t1, t1, obj); // body start
890 __ initialize_body(t1, arr_size, 0, t2);
891 __ verify_oop(obj);
892
893 __ ret(lr);
894
895 __ bind(slow_path);
896 }
897
898 __ enter();
899 OopMap* map = save_live_registers(sasm);
900 int call_offset;
901 if (id == new_type_array_id) {
902 call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_type_array), klass, length);
903 } else {
904 call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_object_array), klass, length);
905 }
906
907 oop_maps = new OopMapSet();
908 oop_maps->add_gc_map(call_offset, map);
909 restore_live_registers_except_r0(sasm);
910
911 __ verify_oop(obj);
912 __ leave();
913 __ ret(lr);
914
915 // r0: new array
916 }
917 break;
918
919 case new_multi_array_id:
920 { StubFrame f(sasm, "new_multi_array", dont_gc_arguments);
921 // r0,: klass
922 // r19,: rank
923 // r2: address of 1st dimension
924 OopMap* map = save_live_registers(sasm);
925 __ mov(c_rarg1, r0);
926 __ mov(c_rarg3, r2);
927 __ mov(c_rarg2, r19);
928 int call_offset = __ call_RT(r0, noreg, CAST_FROM_FN_PTR(address, new_multi_array), r1, r2, r3);
929
930 oop_maps = new OopMapSet();
931 oop_maps->add_gc_map(call_offset, map);
932 restore_live_registers_except_r0(sasm);
933
934 // r0,: new multi array
935 __ verify_oop(r0);
936 }
937 break;
938
939 case register_finalizer_id:
940 {
941 __ set_info("register_finalizer", dont_gc_arguments);
942
943 // This is called via call_runtime so the arguments
944 // will be place in C abi locations
945
946 __ verify_oop(c_rarg0);
947
948 // load the klass and check the has finalizer flag
949 Label register_finalizer;
950 Register t = r5;
951 __ load_klass(t, r0);
952 __ ldrw(t, Address(t, Klass::access_flags_offset()));
953 __ tbnz(t, exact_log2(JVM_ACC_HAS_FINALIZER), register_finalizer);
954 __ ret(lr);
955
956 __ bind(register_finalizer);
957 __ enter();
958 OopMap* oop_map = save_live_registers(sasm);
959 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), r0);
960 oop_maps = new OopMapSet();
961 oop_maps->add_gc_map(call_offset, oop_map);
962
963 // Now restore all the live registers
964 restore_live_registers(sasm);
965
966 __ leave();
967 __ ret(lr);
968 }
969 break;
970
971 case throw_class_cast_exception_id:
972 { StubFrame f(sasm, "throw_class_cast_exception", dont_gc_arguments);
973 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
974 }
975 break;
976
977 case throw_incompatible_class_change_error_id:
978 { StubFrame f(sasm, "throw_incompatible_class_cast_exception", dont_gc_arguments);
979 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
980 }
981 break;
982
983 case slow_subtype_check_id:
984 {
985 // Typical calling sequence:
986 // __ push(klass_RInfo); // object klass or other subclass
987 // __ push(sup_k_RInfo); // array element klass or other superclass
988 // __ bl(slow_subtype_check);
989 // Note that the subclass is pushed first, and is therefore deepest.
990 enum layout {
991 r0_off, r0_off_hi,
992 r2_off, r2_off_hi,
993 r4_off, r4_off_hi,
994 r5_off, r5_off_hi,
995 sup_k_off, sup_k_off_hi,
996 klass_off, klass_off_hi,
997 framesize,
998 result_off = sup_k_off
999 };
1000
1001 __ set_info("slow_subtype_check", dont_gc_arguments);
1002 __ push(RegSet::of(r0, r2, r4, r5), sp);
1003
1004 // This is called by pushing args and not with C abi
1005 // __ ldr(r4, Address(sp, (klass_off) * VMRegImpl::stack_slot_size)); // subclass
1006 // __ ldr(r0, Address(sp, (sup_k_off) * VMRegImpl::stack_slot_size)); // superclass
1007
1008 __ ldp(r4, r0, Address(sp, (sup_k_off) * VMRegImpl::stack_slot_size));
1009
1010 Label miss;
1011 __ check_klass_subtype_slow_path(r4, r0, r2, r5, NULL, &miss);
1012
1013 // fallthrough on success:
1014 __ mov(rscratch1, 1);
1015 __ str(rscratch1, Address(sp, (result_off) * VMRegImpl::stack_slot_size)); // result
1016 __ pop(RegSet::of(r0, r2, r4, r5), sp);
1017 __ ret(lr);
1018
1019 __ bind(miss);
1020 __ str(zr, Address(sp, (result_off) * VMRegImpl::stack_slot_size)); // result
1021 __ pop(RegSet::of(r0, r2, r4, r5), sp);
1022 __ ret(lr);
1023 }
1024 break;
1025
1026 case monitorenter_nofpu_id:
1027 save_fpu_registers = false;
1028 // fall through
1029 case monitorenter_id:
1030 {
1031 StubFrame f(sasm, "monitorenter", dont_gc_arguments);
1032 OopMap* map = save_live_registers(sasm, save_fpu_registers);
1033
1034 // Called with store_parameter and not C abi
1035
1036 f.load_argument(1, r0); // r0,: object
1037 f.load_argument(0, r1); // r1,: lock address
1038
1039 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), r0, r1);
1040
1041 oop_maps = new OopMapSet();
1042 oop_maps->add_gc_map(call_offset, map);
1043 restore_live_registers(sasm, save_fpu_registers);
1044 }
1045 break;
1046
1047 case monitorexit_nofpu_id:
1048 save_fpu_registers = false;
1049 // fall through
1050 case monitorexit_id:
1051 {
1052 StubFrame f(sasm, "monitorexit", dont_gc_arguments);
1053 OopMap* map = save_live_registers(sasm, save_fpu_registers);
1054
1055 // Called with store_parameter and not C abi
1056
1057 f.load_argument(0, r0); // r0,: lock address
1058
1059 // note: really a leaf routine but must setup last java sp
1060 // => use call_RT for now (speed can be improved by
1061 // doing last java sp setup manually)
1062 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), r0);
1063
1064 oop_maps = new OopMapSet();
1065 oop_maps->add_gc_map(call_offset, map);
1066 restore_live_registers(sasm, save_fpu_registers);
1067 }
1068 break;
1069
1070 case deoptimize_id:
1071 {
1072 StubFrame f(sasm, "deoptimize", dont_gc_arguments);
1073 OopMap* oop_map = save_live_registers(sasm);
1074 f.load_argument(0, c_rarg1);
1075 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize), c_rarg1);
1076
1077 oop_maps = new OopMapSet();
1078 oop_maps->add_gc_map(call_offset, oop_map);
1079 restore_live_registers(sasm);
1080 DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
1081 assert(deopt_blob != NULL, "deoptimization blob must have been created");
1082 __ leave();
1083 __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
1084 }
1085 break;
1086
1087 case throw_range_check_failed_id:
1088 { StubFrame f(sasm, "range_check_failed", dont_gc_arguments);
1089 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
1090 }
1091 break;
1092
1093 case unwind_exception_id:
1094 { __ set_info("unwind_exception", dont_gc_arguments);
1095 // note: no stubframe since we are about to leave the current
1096 // activation and we are calling a leaf VM function only.
1097 generate_unwind_exception(sasm);
1098 }
1099 break;
1100
1101 case access_field_patching_id:
1102 { StubFrame f(sasm, "access_field_patching", dont_gc_arguments);
1103 // we should set up register map
1104 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
1105 }
1106 break;
1107
1108 case load_klass_patching_id:
1109 { StubFrame f(sasm, "load_klass_patching", dont_gc_arguments);
1110 // we should set up register map
1111 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
1112 }
1113 break;
1114
1115 case load_mirror_patching_id:
1116 { StubFrame f(sasm, "load_mirror_patching", dont_gc_arguments);
1117 // we should set up register map
1118 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_mirror_patching));
1119 }
1120 break;
1121
1122 case load_appendix_patching_id:
1123 { StubFrame f(sasm, "load_appendix_patching", dont_gc_arguments);
1124 // we should set up register map
1125 oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_appendix_patching));
1126 }
1127 break;
1128
1129 case handle_exception_nofpu_id:
1130 case handle_exception_id:
1131 { StubFrame f(sasm, "handle_exception", dont_gc_arguments);
1132 oop_maps = generate_handle_exception(id, sasm);
1133 }
1134 break;
1135
1136 case handle_exception_from_callee_id:
1137 { StubFrame f(sasm, "handle_exception_from_callee", dont_gc_arguments);
1138 oop_maps = generate_handle_exception(id, sasm);
1139 }
1140 break;
1141
1142 case throw_index_exception_id:
1143 { StubFrame f(sasm, "index_range_check_failed", dont_gc_arguments);
1144 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
1145 }
1146 break;
1147
1148 case throw_array_store_exception_id:
1149 { StubFrame f(sasm, "throw_array_store_exception", dont_gc_arguments);
1150 // tos + 0: link
1151 // + 1: return address
1152 oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true);
1153 }
1154 break;
1155
1156 case predicate_failed_trap_id:
1157 {
1158 StubFrame f(sasm, "predicate_failed_trap", dont_gc_arguments);
1159
1160 OopMap* map = save_live_registers(sasm);
1161
1162 int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap));
1163 oop_maps = new OopMapSet();
1164 oop_maps->add_gc_map(call_offset, map);
1165 restore_live_registers(sasm);
1166 __ leave();
1167 DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
1168 assert(deopt_blob != NULL, "deoptimization blob must have been created");
1169
1170 __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
1171 }
1172 break;
1173
1174
1175 default:
1176 { StubFrame f(sasm, "unimplemented entry", dont_gc_arguments);
1177 __ mov(r0, (int)id);
1178 __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), r0);
1179 __ should_not_reach_here();
1180 }
1181 break;
1182 }
1183 }
1184 return oop_maps;
1185 }
1186
1187 #undef __
1188
1189 const char *Runtime1::pd_name_for_address(address entry) { Unimplemented(); return 0; }