1 /*
  2  * Copyright (c) 2018, 2025, 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 "asm/macroAssembler.inline.hpp"
 26 #include "gc/g1/g1BarrierSet.hpp"
 27 #include "gc/g1/g1BarrierSetAssembler.hpp"
 28 #include "gc/g1/g1BarrierSetRuntime.hpp"
 29 #include "gc/g1/g1CardTable.hpp"
 30 #include "gc/g1/g1HeapRegion.hpp"
 31 #include "gc/g1/g1ThreadLocalData.hpp"
 32 #include "interpreter/interp_masm.hpp"
 33 #include "runtime/arguments.hpp"
 34 #include "runtime/sharedRuntime.hpp"
 35 #include "utilities/debug.hpp"
 36 #include "utilities/macros.hpp"
 37 #ifdef COMPILER1
 38 #include "c1/c1_LIRAssembler.hpp"
 39 #include "c1/c1_MacroAssembler.hpp"
 40 #include "gc/g1/c1/g1BarrierSetC1.hpp"
 41 #endif // COMPILER1
 42 #ifdef COMPILER2
 43 #include "gc/g1/c2/g1BarrierSetC2.hpp"
 44 #endif // COMPILER2
 45 
 46 #define __ masm->
 47 
 48 void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators,
 49                                                             Register addr, Register count) {
 50   bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
 51 
 52   if (!dest_uninitialized) {
 53     Register thread = r15_thread;
 54 
 55     Label filtered;
 56     Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
 57     // Is marking active?
 58     if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
 59       __ cmpl(in_progress, 0);
 60     } else {
 61       assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
 62       __ cmpb(in_progress, 0);
 63     }
 64 
 65     __ jcc(Assembler::equal, filtered);
 66 
 67     __ push_call_clobbered_registers(false /* save_fpu */);
 68     if (count == c_rarg0) {
 69       if (addr == c_rarg1) {
 70         // exactly backwards!!
 71         __ xchgptr(c_rarg1, c_rarg0);
 72       } else {
 73         __ movptr(c_rarg1, count);
 74         __ movptr(c_rarg0, addr);
 75       }
 76     } else {
 77       __ movptr(c_rarg0, addr);
 78       __ movptr(c_rarg1, count);
 79     }
 80     if (UseCompressedOops) {
 81       __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_narrow_oop_entry), 2);
 82     } else {
 83       __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry), 2);
 84     }
 85     __ pop_call_clobbered_registers(false /* save_fpu */);
 86 
 87     __ bind(filtered);
 88   }
 89 }
 90 
 91 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
 92                                                              Register addr, Register count, Register tmp) {
 93   Label L_done;
 94 
 95   __ testptr(count, count);
 96   __ jccb(Assembler::zero, L_done);
 97 
 98   // Calculate end address in "count".
 99   Address::ScaleFactor scale = UseCompressedOops ? Address::times_4 : Address::times_8;
100   __ leaq(count, Address(addr, count, scale));
101 
102   // Calculate start card address in "addr".
103   __ shrptr(addr, CardTable::card_shift());
104 
105   Register thread = r15_thread;
106 
107   __ movptr(tmp, Address(thread, in_bytes(G1ThreadLocalData::card_table_base_offset())));
108   __ addptr(addr, tmp);
109 
110   // Calculate address of card of last word in the array.
111   __ subptr(count, 1);
112   __ shrptr(count, CardTable::card_shift());
113   __ addptr(count, tmp);
114 
115   Label L_loop;
116   // Iterate from start card to end card (inclusive).
117   __ bind(L_loop);
118 
119   Label L_is_clean_card;
120   if (UseCondCardMark) {
121     __ cmpb(Address(addr, 0), G1CardTable::clean_card_val());
122     __ jccb(Assembler::equal, L_is_clean_card);
123   } else {
124    __ movb(Address(addr, 0), G1CardTable::dirty_card_val());
125   }
126 
127   Label L_next_card;
128   __ bind(L_next_card);
129   __ addptr(addr, sizeof(CardTable::CardValue));
130   __ cmpptr(addr, count);
131   __ jccb(Assembler::belowEqual, L_loop);
132   __ jmpb(L_done);
133 
134   __ bind(L_is_clean_card);
135   // Card was clean. Dirty card and go to next.
136   __ movb(Address(addr, 0), G1CardTable::dirty_card_val());
137   __ jmpb(L_next_card);
138 
139   __ bind(L_done);
140 }
141 
142 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
143                                     Register dst, Address src, Register tmp1) {
144   bool on_oop = is_reference_type(type);
145   bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
146   bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
147   bool on_reference = on_weak || on_phantom;
148   CardTableBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1);
149   if (on_oop && on_reference) {
150     // Generate the G1 pre-barrier code to log the value of
151     // the referent field in an SATB buffer.
152     g1_write_barrier_pre(masm /* masm */,
153                          noreg /* obj */,
154                          dst /* pre_val */,
155                          tmp1 /* tmp */,
156                          true /* tosca_live */,
157                          true /* expand_call */);
158   }
159 }
160 
161 static void generate_pre_barrier_fast_path(MacroAssembler* masm,
162                                            const Register thread) {
163   Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
164   // Is marking active?
165   if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
166     __ cmpl(in_progress, 0);
167   } else {
168     assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
169     __ cmpb(in_progress, 0);
170   }
171 }
172 
173 static void generate_pre_barrier_slow_path(MacroAssembler* masm,
174                                            const Register obj,
175                                            const Register pre_val,
176                                            const Register thread,
177                                            const Register tmp,
178                                            Label& L_done) {
179   Address index_addr(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
180   Address buffer_addr(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
181 
182   // This code assumes that buffer index is pointer sized.
183   STATIC_ASSERT(in_bytes(SATBMarkQueue::byte_width_of_index()) == sizeof(intptr_t));
184 
185   Label L_runtime;
186 
187   // Do we need to load the previous value?
188   if (obj != noreg) {
189     __ load_heap_oop(pre_val, Address(obj, 0), noreg, AS_RAW);
190   }
191 
192   // Is the previous value null?
193   __ testptr(pre_val, pre_val);
194   __ jcc(Assembler::equal, L_done);
195 
196   // Can we store a value in the given thread's buffer?
197   // (The index field is typed as size_t.)
198   __ movptr(tmp, index_addr);             // temp := *(index address)
199   __ testptr(tmp, tmp);                   // index == 0?
200   __ jccb(Assembler::zero, L_runtime);    // jump to runtime if index == 0 (full buffer)
201 
202   // The buffer is not full, store value into it.
203   __ subptr(tmp, wordSize);               // temp := next index
204   __ movptr(index_addr, tmp);             // *(index address) := next index
205   __ addptr(tmp, buffer_addr);            // temp := buffer address + next index
206   __ movptr(Address(tmp, 0), pre_val);    // *(buffer address + next index) := value
207 
208   // Jump out if done, or fall-through to runtime.
209   // "L_done" is far away, so jump cannot be short.
210   __ jmp(L_done);
211   __ bind(L_runtime);
212 }
213 
214 void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm,
215                                                  Register obj,
216                                                  Register pre_val,
217                                                  Register tmp,
218                                                  bool tosca_live,
219                                                  bool expand_call) {
220   // If expand_call is true then we expand the call_VM_leaf macro
221   // directly to skip generating the check by
222   // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
223   const Register thread = r15_thread;
224 
225   Label done;
226 
227   assert(pre_val != noreg, "check this code");
228 
229   if (obj != noreg) {
230     assert_different_registers(obj, pre_val, tmp);
231     assert(pre_val != rax, "check this code");
232   }
233 
234   generate_pre_barrier_fast_path(masm, thread);
235   // If marking is not active (*(mark queue active address) == 0), jump to done
236   __ jcc(Assembler::equal, done);
237   generate_pre_barrier_slow_path(masm, obj, pre_val, thread, tmp, done);
238 
239   if (Arguments::is_valhalla_enabled() && InlineTypePassFieldsAsArgs) {
240     // Barriers might be emitted when converting between (scalarized) calling conventions for inline
241     // types. Save all argument registers before calling into the runtime.
242     // TODO 8366717: use push_set() (see JDK-8283327 push/pop_call_clobbered_registers & aarch64 )
243     __ pusha();
244     __ subptr(rsp, 64);
245     __ movdbl(Address(rsp, 0),  j_farg0);
246     __ movdbl(Address(rsp, 8),  j_farg1);
247     __ movdbl(Address(rsp, 16), j_farg2);
248     __ movdbl(Address(rsp, 24), j_farg3);
249     __ movdbl(Address(rsp, 32), j_farg4);
250     __ movdbl(Address(rsp, 40), j_farg5);
251     __ movdbl(Address(rsp, 48), j_farg6);
252     __ movdbl(Address(rsp, 56), j_farg7);
253   } else {
254     // Determine and save the live input values
255     __ push_call_clobbered_registers();
256   }
257 
258   // Calling the runtime using the regular call_VM_leaf mechanism generates
259   // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
260   // that checks that the *(ebp+frame::interpreter_frame_last_sp) == nullptr.
261   //
262   // If we care generating the pre-barrier without a frame (e.g. in the
263   // intrinsified Reference.get() routine) then ebp might be pointing to
264   // the caller frame and so this check will most likely fail at runtime.
265   //
266   // Expanding the call directly bypasses the generation of the check.
267   // So when we do not have have a full interpreter frame on the stack
268   // expand_call should be passed true.
269 
270   if (expand_call) {
271     assert(pre_val != c_rarg1, "smashed arg");
272     if (c_rarg1 != thread) {
273       __ mov(c_rarg1, thread);
274     }
275     if (c_rarg0 != pre_val) {
276       __ mov(c_rarg0, pre_val);
277     }
278     __ MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), 2);
279   } else {
280     __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
281   }
282 
283   if (Arguments::is_valhalla_enabled() && InlineTypePassFieldsAsArgs) {
284     // Restore registers
285     __ movdbl(j_farg0, Address(rsp, 0));
286     __ movdbl(j_farg1, Address(rsp, 8));
287     __ movdbl(j_farg2, Address(rsp, 16));
288     __ movdbl(j_farg3, Address(rsp, 24));
289     __ movdbl(j_farg4, Address(rsp, 32));
290     __ movdbl(j_farg5, Address(rsp, 40));
291     __ movdbl(j_farg6, Address(rsp, 48));
292     __ movdbl(j_farg7, Address(rsp, 56));
293     __ addptr(rsp, 64);
294     __ popa();
295   } else {
296     __ pop_call_clobbered_registers();
297   }
298 
299   __ bind(done);
300 }
301 
302 static void generate_post_barrier(MacroAssembler* masm,
303                                   const Register store_addr,
304                                   const Register new_val,
305                                   const Register tmp1,
306                                   bool new_val_may_be_null) {
307 
308   assert_different_registers(store_addr, new_val, tmp1, noreg);
309 
310   Register thread = r15_thread;
311 
312   Label L_done;
313   // Does store cross heap regions?
314   __ movptr(tmp1, store_addr);                                    // tmp1 := store address
315   __ xorptr(tmp1, new_val);                                       // tmp1 := store address ^ new value
316   __ shrptr(tmp1, G1HeapRegion::LogOfHRGrainBytes);               // ((store address ^ new value) >> LogOfHRGrainBytes) == 0?
317   __ jccb(Assembler::equal, L_done);
318 
319   // Crosses regions, storing null?
320   if (new_val_may_be_null) {
321     __ testptr(new_val, new_val);                                 // new value == null?
322     __ jccb(Assembler::equal, L_done);
323   }
324 
325   __ movptr(tmp1, store_addr);                                    // tmp1 := store address
326   __ shrptr(tmp1, CardTable::card_shift());                       // tmp1 := card address relative to card table base
327 
328   Address card_table_addr(thread, in_bytes(G1ThreadLocalData::card_table_base_offset()));
329   __ addptr(tmp1, card_table_addr);                               // tmp1 := card address
330   if (UseCondCardMark) {
331     __ cmpb(Address(tmp1, 0), G1CardTable::clean_card_val());     // *(card address) == clean_card_val?
332     __ jccb(Assembler::notEqual, L_done);
333   }
334   // Storing a region crossing, non-null oop, card is clean.
335   // Dirty card.
336   __ movb(Address(tmp1, 0), G1CardTable::dirty_card_val());       // *(card address) := dirty_card_val
337   __ bind(L_done);
338 }
339 
340 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
341                                                   Register store_addr,
342                                                   Register new_val,
343                                                   Register tmp) {
344   generate_post_barrier(masm, store_addr, new_val, tmp, true /* new_val_may_be_null */);
345 }
346 
347 #if defined(COMPILER2)
348 
349 static void generate_c2_barrier_runtime_call(MacroAssembler* masm, G1BarrierStubC2* stub, const Register arg, const address runtime_path) {
350   SaveLiveRegisters save_registers(masm, stub);
351   if (c_rarg0 != arg) {
352     __ mov(c_rarg0, arg);
353   }
354   __ mov(c_rarg1, r15_thread);
355   // rax is a caller-saved, non-argument-passing register, so it does not
356   // interfere with c_rarg0 or c_rarg1. If it contained any live value before
357   // entering this stub, it is saved at this point, and restored after the
358   // call. If it did not contain any live value, it is free to be used. In
359   // either case, it is safe to use it here as a call scratch register.
360   __ call(RuntimeAddress(runtime_path), rax);
361 }
362 
363 void G1BarrierSetAssembler::g1_write_barrier_pre_c2(MacroAssembler* masm,
364                                                     Register obj,
365                                                     Register pre_val,
366                                                     Register tmp,
367                                                     G1PreBarrierStubC2* stub) {
368   const Register thread = r15_thread;
369 
370   assert(pre_val != noreg, "check this code");
371   if (obj != noreg) {
372     assert_different_registers(obj, pre_val, tmp);
373   }
374 
375   stub->initialize_registers(obj, pre_val, thread, tmp);
376 
377   generate_pre_barrier_fast_path(masm, thread);
378   // If marking is active (*(mark queue active address) != 0), jump to stub (slow path)
379   __ jcc(Assembler::notEqual, *stub->entry());
380 
381   __ bind(*stub->continuation());
382 }
383 
384 void G1BarrierSetAssembler::generate_c2_pre_barrier_stub(MacroAssembler* masm,
385                                                          G1PreBarrierStubC2* stub) const {
386   Assembler::InlineSkippedInstructionsCounter skip_counter(masm);
387   Register obj = stub->obj();
388   Register pre_val = stub->pre_val();
389   Register thread = stub->thread();
390   Register tmp = stub->tmp1();
391   assert(stub->tmp2() == noreg, "not needed in this platform");
392 
393   __ bind(*stub->entry());
394   generate_pre_barrier_slow_path(masm, obj, pre_val, thread, tmp, *stub->continuation());
395 
396   generate_c2_barrier_runtime_call(masm, stub, pre_val, CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry));
397   __ jmp(*stub->continuation());
398 }
399 
400 void G1BarrierSetAssembler::g1_write_barrier_post_c2(MacroAssembler* masm,
401                                                      Register store_addr,
402                                                      Register new_val,
403                                                      Register tmp,
404                                                      bool new_val_may_be_null) {
405   generate_post_barrier(masm, store_addr, new_val, tmp, new_val_may_be_null);
406 }
407 
408 #endif // COMPILER2
409 
410 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
411                                          Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
412   bool in_heap = (decorators & IN_HEAP) != 0;
413   bool as_normal = (decorators & AS_NORMAL) != 0;
414   bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
415 
416   bool needs_pre_barrier = as_normal && !dest_uninitialized;
417   bool needs_post_barrier = val != noreg && in_heap;
418 
419   // flatten object address if needed
420   // We do it regardless of precise because we need the registers
421   if (dst.index() == noreg && dst.disp() == 0) {
422     if (dst.base() != tmp1) {
423       __ movptr(tmp1, dst.base());
424     }
425   } else {
426     __ lea(tmp1, dst);
427   }
428 
429   if (needs_pre_barrier) {
430     g1_write_barrier_pre(masm /*masm*/,
431                          tmp1 /* obj */,
432                          tmp2 /* pre_val */,
433                          tmp3  /* tmp */,
434                          val != noreg /* tosca_live */,
435                          false /* expand_call */);
436   }
437   if (val == noreg) {
438     BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg, noreg);
439   } else {
440     Register new_val = val;
441     if (needs_post_barrier) {
442       // G1 barrier needs uncompressed oop for region cross check.
443       if (UseCompressedOops) {
444         new_val = tmp2;
445         __ movptr(new_val, val);
446       }
447     }
448     BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg, noreg);
449     if (needs_post_barrier) {
450       g1_write_barrier_post(masm /*masm*/,
451                             tmp1 /* store_adr */,
452                             new_val /* new_val */,
453                             tmp3 /* tmp */);
454     }
455   }
456 }
457 
458 #ifdef COMPILER1
459 
460 #undef __
461 #define __ ce->masm()->
462 
463 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
464   G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
465   // At this point we know that marking is in progress.
466   // If do_load() is true then we have to emit the
467   // load of the previous value; otherwise it has already
468   // been loaded into _pre_val.
469 
470   __ bind(*stub->entry());
471   assert(stub->pre_val()->is_register(), "Precondition.");
472 
473   Register pre_val_reg = stub->pre_val()->as_register();
474 
475   if (stub->do_load()) {
476     ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/);
477   }
478 
479   __ testptr(pre_val_reg, pre_val_reg);
480   __ jcc(Assembler::equal, *stub->continuation());
481   ce->store_parameter(stub->pre_val()->as_register(), 0);
482   __ call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
483   __ jmp(*stub->continuation());
484 
485 }
486 
487 #undef __
488 
489 void G1BarrierSetAssembler::g1_write_barrier_post_c1(MacroAssembler* masm,
490                                                      Register store_addr,
491                                                      Register new_val,
492                                                      Register thread,
493                                                      Register tmp1,
494                                                      Register tmp2 /* unused on x86 */) {
495   generate_post_barrier(masm, store_addr, new_val, tmp1, true /* new_val_may_be_null */);
496 }
497 
498 #define __ sasm->
499 
500 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
501   // Generated code assumes that buffer index is pointer sized.
502   STATIC_ASSERT(in_bytes(SATBMarkQueue::byte_width_of_index()) == sizeof(intptr_t));
503 
504   __ prologue("g1_pre_barrier", false);
505   // arg0 : previous value of memory
506 
507   __ push_ppx(rax);
508   __ push_ppx(rdx);
509 
510   const Register pre_val = rax;
511   const Register thread = r15_thread;
512   const Register tmp = rdx;
513 
514   Address queue_active(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
515   Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
516   Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
517 
518   Label L_done, L_runtime;
519 
520   // Is marking still active?
521   if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
522     __ cmpl(queue_active, 0);
523   } else {
524     assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
525     __ cmpb(queue_active, 0);
526   }
527   __ jcc(Assembler::equal, L_done);
528 
529   // Can we store original value in the thread's buffer?
530 
531   __ movptr(tmp, queue_index);
532   __ testptr(tmp, tmp);
533   __ jccb(Assembler::zero, L_runtime);
534   __ subptr(tmp, wordSize);
535   __ movptr(queue_index, tmp);
536   __ addptr(tmp, buffer);
537 
538   // prev_val (rax)
539   __ load_parameter(0, pre_val);
540   __ movptr(Address(tmp, 0), pre_val);
541   __ jmp(L_done);
542 
543   __ bind(L_runtime);
544 
545   __ push_call_clobbered_registers();
546 
547   // load the pre-value
548   __ load_parameter(0, rcx);
549   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), rcx, thread);
550 
551   __ pop_call_clobbered_registers();
552 
553   __ bind(L_done);
554 
555   __ pop_ppx(rdx);
556   __ pop_ppx(rax);
557 
558   __ epilogue();
559 }
560 
561 #undef __
562 
563 #endif // COMPILER1