1 /*
2 * Copyright (c) 2018, 2019, 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 #include "precompiled.hpp"
25 #include "asm/macroAssembler.inline.hpp"
26 #include "code/codeBlob.hpp"
27 #include "code/vmreg.inline.hpp"
28 #include "gc/z/zBarrier.inline.hpp"
29 #include "gc/z/zBarrierSet.hpp"
30 #include "gc/z/zBarrierSetAssembler.hpp"
31 #include "gc/z/zBarrierSetRuntime.hpp"
32 #include "memory/resourceArea.hpp"
33 #include "runtime/sharedRuntime.hpp"
34 #include "utilities/macros.hpp"
35 #ifdef COMPILER1
36 #include "c1/c1_LIRAssembler.hpp"
37 #include "c1/c1_MacroAssembler.hpp"
38 #include "gc/z/c1/zBarrierSetC1.hpp"
39 #endif // COMPILER1
40 #ifdef COMPILER2
41 #include "gc/z/c2/zBarrierSetC2.hpp"
42 #endif // COMPILER2
43
44 #ifdef PRODUCT
45 #define BLOCK_COMMENT(str) /* nothing */
46 #else
47 #define BLOCK_COMMENT(str) __ block_comment(str)
48 #endif
49
50 #undef __
51 #define __ masm->
52
53 static void call_vm(MacroAssembler* masm,
54 address entry_point,
55 Register arg0,
56 Register arg1) {
57 // Setup arguments
58 if (arg1 == c_rarg0) {
59 if (arg0 == c_rarg1) {
60 __ xchgptr(c_rarg1, c_rarg0);
61 } else {
62 __ movptr(c_rarg1, arg1);
63 __ movptr(c_rarg0, arg0);
64 }
65 } else {
66 if (arg0 != c_rarg0) {
67 __ movptr(c_rarg0, arg0);
68 }
69 if (arg1 != c_rarg1) {
70 __ movptr(c_rarg1, arg1);
71 }
72 }
73
74 // Call VM
75 __ MacroAssembler::call_VM_leaf_base(entry_point, 2);
76 }
77
78 void ZBarrierSetAssembler::load_at(MacroAssembler* masm,
79 DecoratorSet decorators,
80 BasicType type,
81 Register dst,
82 Address src,
83 Register tmp1,
84 Register tmp_thread) {
85 if (!ZBarrierSet::barrier_needed(decorators, type)) {
86 // Barrier not needed
87 BarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp_thread);
88 return;
89 }
90
91 BLOCK_COMMENT("ZBarrierSetAssembler::load_at {");
92
93 // Allocate scratch register
94 Register scratch = tmp1;
95 if (tmp1 == noreg) {
96 scratch = r12;
97 __ push(scratch);
98 }
99
100 assert_different_registers(dst, scratch);
101
102 Label done;
103
104 //
105 // Fast Path
106 //
107
108 // Load address
109 __ lea(scratch, src);
110
111 // Load oop at address
112 __ movptr(dst, Address(scratch, 0));
113
114 // Test address bad mask
115 __ testptr(dst, address_bad_mask_from_thread(r15_thread));
116 __ jcc(Assembler::zero, done);
117
118 //
119 // Slow path
120 //
121
122 // Save registers
123 __ push(rax);
124 __ push(rcx);
125 __ push(rdx);
126 __ push(rdi);
127 __ push(rsi);
128 __ push(r8);
129 __ push(r9);
130 __ push(r10);
131 __ push(r11);
132
133 // We may end up here from generate_native_wrapper, then the method may have
134 // floats as arguments, and we must spill them before calling the VM runtime
135 // leaf. From the interpreter all floats are passed on the stack.
136 assert(Argument::n_float_register_parameters_j == 8, "Assumption");
137 const int xmm_size = wordSize * 2;
138 const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
139 __ subptr(rsp, xmm_spill_size);
140 __ movdqu(Address(rsp, xmm_size * 7), xmm7);
141 __ movdqu(Address(rsp, xmm_size * 6), xmm6);
142 __ movdqu(Address(rsp, xmm_size * 5), xmm5);
143 __ movdqu(Address(rsp, xmm_size * 4), xmm4);
144 __ movdqu(Address(rsp, xmm_size * 3), xmm3);
145 __ movdqu(Address(rsp, xmm_size * 2), xmm2);
146 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
147 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
148
149 // Call VM
150 call_vm(masm, ZBarrierSetRuntime::load_barrier_on_oop_field_preloaded_addr(decorators), dst, scratch);
151
152 // Restore registers
153 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
154 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
155 __ movdqu(xmm2, Address(rsp, xmm_size * 2));
156 __ movdqu(xmm3, Address(rsp, xmm_size * 3));
157 __ movdqu(xmm4, Address(rsp, xmm_size * 4));
158 __ movdqu(xmm5, Address(rsp, xmm_size * 5));
159 __ movdqu(xmm6, Address(rsp, xmm_size * 6));
160 __ movdqu(xmm7, Address(rsp, xmm_size * 7));
161 __ addptr(rsp, xmm_spill_size);
162
163 __ pop(r11);
164 __ pop(r10);
165 __ pop(r9);
166 __ pop(r8);
167 __ pop(rsi);
168 __ pop(rdi);
169 __ pop(rdx);
170 __ pop(rcx);
171
172 if (dst == rax) {
173 __ addptr(rsp, wordSize);
174 } else {
175 __ movptr(dst, rax);
176 __ pop(rax);
177 }
178
179 __ bind(done);
180
181 // Restore scratch register
182 if (tmp1 == noreg) {
183 __ pop(scratch);
184 }
185
186 BLOCK_COMMENT("} ZBarrierSetAssembler::load_at");
187 }
188
189 #ifdef ASSERT
190
191 void ZBarrierSetAssembler::store_at(MacroAssembler* masm,
192 DecoratorSet decorators,
193 BasicType type,
194 Address dst,
195 Register src,
196 Register tmp1,
197 Register tmp2) {
198 BLOCK_COMMENT("ZBarrierSetAssembler::store_at {");
199
200 // Verify oop store
201 if (is_reference_type(type)) {
202 // Note that src could be noreg, which means we
203 // are storing null and can skip verification.
204 if (src != noreg) {
205 Label done;
206 __ testptr(src, address_bad_mask_from_thread(r15_thread));
207 __ jcc(Assembler::zero, done);
208 __ stop("Verify oop store failed");
209 __ should_not_reach_here();
210 __ bind(done);
211 }
212 }
213
214 // Store value
215 BarrierSetAssembler::store_at(masm, decorators, type, dst, src, tmp1, tmp2);
216
217 BLOCK_COMMENT("} ZBarrierSetAssembler::store_at");
218 }
219
220 #endif // ASSERT
221
222 void ZBarrierSetAssembler::arraycopy_prologue(MacroAssembler* masm,
223 DecoratorSet decorators,
224 BasicType type,
225 Register src,
226 Register dst,
227 Register count) {
228 if (!ZBarrierSet::barrier_needed(decorators, type)) {
229 // Barrier not needed
230 return;
231 }
232
233 BLOCK_COMMENT("ZBarrierSetAssembler::arraycopy_prologue {");
234
235 // Save registers
236 __ pusha();
237
238 // Call VM
239 call_vm(masm, ZBarrierSetRuntime::load_barrier_on_oop_array_addr(), src, count);
240
241 // Restore registers
242 __ popa();
243
244 BLOCK_COMMENT("} ZBarrierSetAssembler::arraycopy_prologue");
245 }
246
247 void ZBarrierSetAssembler::try_resolve_jobject_in_native(MacroAssembler* masm,
248 Register jni_env,
249 Register obj,
250 Register tmp,
251 Label& slowpath) {
252 BLOCK_COMMENT("ZBarrierSetAssembler::try_resolve_jobject_in_native {");
253
254 // Resolve jobject
255 BarrierSetAssembler::try_resolve_jobject_in_native(masm, jni_env, obj, tmp, slowpath);
256
257 // Test address bad mask
258 __ testptr(obj, address_bad_mask_from_jni_env(jni_env));
259 __ jcc(Assembler::notZero, slowpath);
260
261 BLOCK_COMMENT("} ZBarrierSetAssembler::try_resolve_jobject_in_native");
262 }
263
264 #ifdef COMPILER1
265
266 #undef __
267 #define __ ce->masm()->
268
269 void ZBarrierSetAssembler::generate_c1_load_barrier_test(LIR_Assembler* ce,
270 LIR_Opr ref) const {
271 __ testptr(ref->as_register(), address_bad_mask_from_thread(r15_thread));
272 }
273
274 void ZBarrierSetAssembler::generate_c1_load_barrier_stub(LIR_Assembler* ce,
275 ZLoadBarrierStubC1* stub) const {
276 // Stub entry
277 __ bind(*stub->entry());
278
279 Register ref = stub->ref()->as_register();
280 Register ref_addr = noreg;
281 Register tmp = noreg;
282
283 if (stub->tmp()->is_valid()) {
284 // Load address into tmp register
285 ce->leal(stub->ref_addr(), stub->tmp());
286 ref_addr = tmp = stub->tmp()->as_pointer_register();
287 } else {
288 // Address already in register
289 ref_addr = stub->ref_addr()->as_address_ptr()->base()->as_pointer_register();
290 }
291
292 assert_different_registers(ref, ref_addr, noreg);
293
294 // Save rax unless it is the result or tmp register
295 if (ref != rax && tmp != rax) {
296 __ push(rax);
297 }
298
299 // Setup arguments and call runtime stub
300 __ subptr(rsp, 2 * BytesPerWord);
301 ce->store_parameter(ref_addr, 1);
302 ce->store_parameter(ref, 0);
303 __ call(RuntimeAddress(stub->runtime_stub()));
304 __ addptr(rsp, 2 * BytesPerWord);
305
306 // Verify result
307 __ verify_oop(rax);
308
309 // Move result into place
310 if (ref != rax) {
311 __ movptr(ref, rax);
312 }
313
314 // Restore rax unless it is the result or tmp register
315 if (ref != rax && tmp != rax) {
316 __ pop(rax);
317 }
318
319 // Stub exit
320 __ jmp(*stub->continuation());
321 }
322
323 #undef __
324 #define __ sasm->
325
326 void ZBarrierSetAssembler::generate_c1_load_barrier_runtime_stub(StubAssembler* sasm,
327 DecoratorSet decorators) const {
328 // Enter and save registers
329 __ enter();
330 __ save_live_registers_no_oop_map(true /* save_fpu_registers */);
331
332 // Setup arguments
333 __ load_parameter(1, c_rarg1);
334 __ load_parameter(0, c_rarg0);
335
336 // Call VM
337 __ call_VM_leaf(ZBarrierSetRuntime::load_barrier_on_oop_field_preloaded_addr(decorators), c_rarg0, c_rarg1);
338
339 // Restore registers and return
340 __ restore_live_registers_except_rax(true /* restore_fpu_registers */);
341 __ leave();
342 __ ret(0);
343 }
344
345 #endif // COMPILER1
346
347 #ifdef COMPILER2
348
349 OptoReg::Name ZBarrierSetAssembler::refine_register(const Node* node, OptoReg::Name opto_reg) {
350 if (!OptoReg::is_reg(opto_reg)) {
351 return OptoReg::Bad;
352 }
353
354 const VMReg vm_reg = OptoReg::as_VMReg(opto_reg);
355 if (vm_reg->is_XMMRegister()) {
356 opto_reg &= ~15;
357 switch (node->ideal_reg()) {
358 case Op_VecX:
359 opto_reg |= 2;
360 break;
361 case Op_VecY:
362 opto_reg |= 4;
363 break;
364 case Op_VecZ:
365 opto_reg |= 8;
366 break;
367 default:
368 opto_reg |= 1;
369 break;
370 }
371 }
372
373 return opto_reg;
374 }
375
376 // We use the vec_spill_helper from the x86.ad file to avoid reinventing this wheel
377 extern int vec_spill_helper(CodeBuffer *cbuf, bool do_size, bool is_load,
378 int stack_offset, int reg, uint ireg, outputStream* st);
379
380 #undef __
381 #define __ _masm->
382
383 class ZSaveLiveRegisters {
384 private:
385 struct XMMRegisterData {
386 XMMRegister _reg;
387 int _size;
388
389 // Used by GrowableArray::find()
390 bool operator == (const XMMRegisterData& other) {
391 return _reg == other._reg;
392 }
393 };
394
395 MacroAssembler* const _masm;
396 GrowableArray<Register> _gp_registers;
397 GrowableArray<XMMRegisterData> _xmm_registers;
398 int _spill_size;
399 int _spill_offset;
400
401 static int xmm_compare_register_size(XMMRegisterData* left, XMMRegisterData* right) {
402 if (left->_size == right->_size) {
403 return 0;
404 }
405
406 return (left->_size < right->_size) ? -1 : 1;
407 }
408
409 static int xmm_slot_size(OptoReg::Name opto_reg) {
410 // The low order 4 bytes denote what size of the XMM register is live
411 return (opto_reg & 15) << 3;
412 }
413
414 static uint xmm_ideal_reg_for_size(int reg_size) {
415 switch (reg_size) {
416 case 8:
417 return Op_VecD;
418 case 16:
419 return Op_VecX;
420 case 32:
421 return Op_VecY;
422 case 64:
423 return Op_VecZ;
424 default:
425 fatal("Invalid register size %d", reg_size);
426 return 0;
427 }
428 }
429
430 bool xmm_needs_vzeroupper() const {
431 return _xmm_registers.is_nonempty() && _xmm_registers.at(0)._size > 16;
432 }
433
434 void xmm_register_save(const XMMRegisterData& reg_data) {
435 const OptoReg::Name opto_reg = OptoReg::as_OptoReg(reg_data._reg->as_VMReg());
436 const uint ideal_reg = xmm_ideal_reg_for_size(reg_data._size);
437 _spill_offset -= reg_data._size;
438 vec_spill_helper(__ code(), false /* do_size */, false /* is_load */, _spill_offset, opto_reg, ideal_reg, tty);
439 }
440
441 void xmm_register_restore(const XMMRegisterData& reg_data) {
442 const OptoReg::Name opto_reg = OptoReg::as_OptoReg(reg_data._reg->as_VMReg());
443 const uint ideal_reg = xmm_ideal_reg_for_size(reg_data._size);
444 vec_spill_helper(__ code(), false /* do_size */, true /* is_load */, _spill_offset, opto_reg, ideal_reg, tty);
445 _spill_offset += reg_data._size;
446 }
447
448 void gp_register_save(Register reg) {
449 _spill_offset -= 8;
450 __ movq(Address(rsp, _spill_offset), reg);
451 }
452
453 void gp_register_restore(Register reg) {
454 __ movq(reg, Address(rsp, _spill_offset));
455 _spill_offset += 8;
456 }
457
458 void initialize(ZLoadBarrierStubC2* stub) {
459 // Create mask of caller saved registers that need to
460 // be saved/restored if live
461 RegMask caller_saved;
462 caller_saved.Insert(OptoReg::as_OptoReg(rax->as_VMReg()));
463 caller_saved.Insert(OptoReg::as_OptoReg(rcx->as_VMReg()));
464 caller_saved.Insert(OptoReg::as_OptoReg(rdx->as_VMReg()));
465 caller_saved.Insert(OptoReg::as_OptoReg(rsi->as_VMReg()));
466 caller_saved.Insert(OptoReg::as_OptoReg(rdi->as_VMReg()));
467 caller_saved.Insert(OptoReg::as_OptoReg(r8->as_VMReg()));
468 caller_saved.Insert(OptoReg::as_OptoReg(r9->as_VMReg()));
469 caller_saved.Insert(OptoReg::as_OptoReg(r10->as_VMReg()));
470 caller_saved.Insert(OptoReg::as_OptoReg(r11->as_VMReg()));
471 caller_saved.Remove(OptoReg::as_OptoReg(stub->ref()->as_VMReg()));
472
473 // Create mask of live registers
474 RegMask live = stub->live();
475 if (stub->tmp() != noreg) {
476 live.Insert(OptoReg::as_OptoReg(stub->tmp()->as_VMReg()));
477 }
478
479 int gp_spill_size = 0;
480 int xmm_spill_size = 0;
481
482 // Record registers that needs to be saved/restored
483 while (live.is_NotEmpty()) {
484 const OptoReg::Name opto_reg = live.find_first_elem();
485 const VMReg vm_reg = OptoReg::as_VMReg(opto_reg);
486
487 live.Remove(opto_reg);
488
489 if (vm_reg->is_Register()) {
490 if (caller_saved.Member(opto_reg)) {
491 _gp_registers.append(vm_reg->as_Register());
492 gp_spill_size += 8;
493 }
494 } else if (vm_reg->is_XMMRegister()) {
495 // We encode in the low order 4 bits of the opto_reg, how large part of the register is live
496 const VMReg vm_reg_base = OptoReg::as_VMReg(opto_reg & ~15);
497 const int reg_size = xmm_slot_size(opto_reg);
498 const XMMRegisterData reg_data = { vm_reg_base->as_XMMRegister(), reg_size };
499 const int reg_index = _xmm_registers.find(reg_data);
500 if (reg_index == -1) {
501 // Not previously appended
502 _xmm_registers.append(reg_data);
503 xmm_spill_size += reg_size;
504 } else {
505 // Previously appended, update size
506 const int reg_size_prev = _xmm_registers.at(reg_index)._size;
507 if (reg_size > reg_size_prev) {
508 _xmm_registers.at_put(reg_index, reg_data);
509 xmm_spill_size += reg_size - reg_size_prev;
510 }
511 }
512 } else {
513 fatal("Unexpected register type");
514 }
515 }
516
517 // Sort by size, largest first
518 _xmm_registers.sort(xmm_compare_register_size);
519
520 // On Windows, the caller reserves stack space for spilling register arguments
521 const int arg_spill_size = frame::arg_reg_save_area_bytes;
522
523 // Stack pointer must be 16 bytes aligned for the call
524 _spill_offset = _spill_size = align_up(xmm_spill_size + gp_spill_size + arg_spill_size, 16);
525 }
526
527 public:
528 ZSaveLiveRegisters(MacroAssembler* masm, ZLoadBarrierStubC2* stub) :
529 _masm(masm),
530 _gp_registers(),
531 _xmm_registers(),
532 _spill_size(0),
533 _spill_offset(0) {
534
535 //
536 // Stack layout after registers have been spilled:
537 //
538 // | ... | original rsp, 16 bytes aligned
539 // ------------------
540 // | zmm0 high |
541 // | ... |
542 // | zmm0 low | 16 bytes aligned
543 // | ... |
544 // | ymm1 high |
545 // | ... |
546 // | ymm1 low | 16 bytes aligned
547 // | ... |
548 // | xmmN high |
549 // | ... |
550 // | xmmN low | 8 bytes aligned
551 // | reg0 | 8 bytes aligned
552 // | reg1 |
553 // | ... |
554 // | regN | new rsp, if 16 bytes aligned
555 // | <padding> | else new rsp, 16 bytes aligned
556 // ------------------
557 //
558
559 // Figure out what registers to save/restore
560 initialize(stub);
561
562 // Allocate stack space
563 if (_spill_size > 0) {
564 __ subptr(rsp, _spill_size);
565 }
566
567 // Save XMM/YMM/ZMM registers
568 for (int i = 0; i < _xmm_registers.length(); i++) {
569 xmm_register_save(_xmm_registers.at(i));
570 }
571
572 if (xmm_needs_vzeroupper()) {
573 __ vzeroupper();
574 }
575
576 // Save general purpose registers
577 for (int i = 0; i < _gp_registers.length(); i++) {
578 gp_register_save(_gp_registers.at(i));
579 }
580 }
581
582 ~ZSaveLiveRegisters() {
583 // Restore general purpose registers
584 for (int i = _gp_registers.length() - 1; i >= 0; i--) {
585 gp_register_restore(_gp_registers.at(i));
586 }
587
588 __ vzeroupper();
589
590 // Restore XMM/YMM/ZMM registers
591 for (int i = _xmm_registers.length() - 1; i >= 0; i--) {
592 xmm_register_restore(_xmm_registers.at(i));
593 }
594
595 // Free stack space
596 if (_spill_size > 0) {
597 __ addptr(rsp, _spill_size);
598 }
599 }
600 };
601
602 class ZSetupArguments {
603 private:
604 MacroAssembler* const _masm;
605 const Register _ref;
606 const Address _ref_addr;
607
608 public:
609 ZSetupArguments(MacroAssembler* masm, ZLoadBarrierStubC2* stub) :
610 _masm(masm),
611 _ref(stub->ref()),
612 _ref_addr(stub->ref_addr()) {
613
614 // Setup arguments
615 if (_ref_addr.base() == noreg) {
616 // No self healing
617 if (_ref != c_rarg0) {
618 __ movq(c_rarg0, _ref);
619 }
620 __ xorq(c_rarg1, c_rarg1);
621 } else {
622 // Self healing
623 if (_ref == c_rarg0) {
624 __ lea(c_rarg1, _ref_addr);
625 } else if (_ref != c_rarg1) {
626 __ lea(c_rarg1, _ref_addr);
627 __ movq(c_rarg0, _ref);
628 } else if (_ref_addr.base() != c_rarg0 && _ref_addr.index() != c_rarg0) {
629 __ movq(c_rarg0, _ref);
630 __ lea(c_rarg1, _ref_addr);
631 } else {
632 __ xchgq(c_rarg0, c_rarg1);
633 if (_ref_addr.base() == c_rarg0) {
634 __ lea(c_rarg1, Address(c_rarg1, _ref_addr.index(), _ref_addr.scale(), _ref_addr.disp()));
635 } else if (_ref_addr.index() == c_rarg0) {
636 __ lea(c_rarg1, Address(_ref_addr.base(), c_rarg1, _ref_addr.scale(), _ref_addr.disp()));
637 } else {
638 ShouldNotReachHere();
639 }
640 }
641 }
642 }
643
644 ~ZSetupArguments() {
645 // Transfer result
646 if (_ref != rax) {
647 __ movq(_ref, rax);
648 }
649 }
650 };
651
652 #undef __
653 #define __ masm->
654
655 void ZBarrierSetAssembler::generate_c2_load_barrier_stub(MacroAssembler* masm, ZLoadBarrierStubC2* stub) const {
656 BLOCK_COMMENT("ZLoadBarrierStubC2");
657
658 // Stub entry
659 __ bind(*stub->entry());
660
661 {
662 ZSaveLiveRegisters save_live_registers(masm, stub);
663 ZSetupArguments setup_arguments(masm, stub);
664 __ call(RuntimeAddress(stub->slow_path()));
665 }
666
667 // Stub exit
668 __ jmp(*stub->continuation());
669 }
670
671 #undef __
672
673 #endif // COMPILER2