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 "code/aotCodeCache.hpp"
 27 #include "gc/g1/g1BarrierSet.hpp"
 28 #include "gc/g1/g1BarrierSetAssembler.hpp"
 29 #include "gc/g1/g1BarrierSetRuntime.hpp"
 30 #include "gc/g1/g1CardTable.hpp"
 31 #include "gc/g1/g1HeapRegion.hpp"
 32 #include "gc/g1/g1ThreadLocalData.hpp"
 33 #include "interpreter/interp_masm.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   // Determine and save the live input values
240   __ push_call_clobbered_registers();
241 
242   // Calling the runtime using the regular call_VM_leaf mechanism generates
243   // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
244   // that checks that the *(ebp+frame::interpreter_frame_last_sp) == nullptr.
245   //
246   // If we care generating the pre-barrier without a frame (e.g. in the
247   // intrinsified Reference.get() routine) then ebp might be pointing to
248   // the caller frame and so this check will most likely fail at runtime.
249   //
250   // Expanding the call directly bypasses the generation of the check.
251   // So when we do not have have a full interpreter frame on the stack
252   // expand_call should be passed true.
253 
254   if (expand_call) {
255     assert(pre_val != c_rarg1, "smashed arg");
256     if (c_rarg1 != thread) {
257       __ mov(c_rarg1, thread);
258     }
259     if (c_rarg0 != pre_val) {
260       __ mov(c_rarg0, pre_val);
261     }
262     __ MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), 2);
263   } else {
264     __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
265   }
266 
267   __ pop_call_clobbered_registers();
268 
269   __ bind(done);
270 }
271 
272 #if INCLUDE_CDS
273 // return a register that differs from reg1, reg2, reg3 and reg4
274 
275 static Register pick_different_reg(Register reg1, Register reg2 = noreg, Register reg3= noreg, Register reg4 = noreg) {
276   RegSet available = (RegSet::of(rscratch1, rscratch2, rax, rbx) + rdx -
277                       RegSet::of(reg1, reg2, reg3, reg4));
278   return *(available.begin());
279 }
280 #endif // INCLUDE_CDS
281 
282 static void generate_post_barrier(MacroAssembler* masm,
283                                   const Register store_addr,
284                                   const Register new_val,
285                                   const Register tmp1,
286                                   bool new_val_may_be_null) {
287 
288   assert_different_registers(store_addr, new_val, tmp1, noreg);
289 
290   Register thread = r15_thread;
291 
292   Label L_done;
293   // Does store cross heap regions?
294 #if INCLUDE_CDS
295   // AOT code needs to load the barrier grain shift from the aot
296   // runtime constants area in the code cache otherwise we can compile
297   // it as an immediate operand
298 
299   if (AOTCodeCache::is_on_for_dump()) {
300     address grain_shift_addr = AOTRuntimeConstants::grain_shift_address();
301     Register save = pick_different_reg(rcx, tmp1, new_val, store_addr);
302     __ push(save);
303     __ movptr(save, store_addr);
304     __ xorptr(save, new_val);
305     __ push(rcx);
306     __ lea(rcx, ExternalAddress(grain_shift_addr));
307     __ movl(rcx, Address(rcx, 0));
308     __ shrptr(save);
309     __ pop(rcx);
310     __ pop(save);
311     __ jcc(Assembler::equal, L_done);
312   } else
313 #endif // INCLUDE_CDS
314   {
315     __ movptr(tmp1, store_addr);                                    // tmp1 := store address
316     __ xorptr(tmp1, new_val);                                       // tmp1 := store address ^ new value
317     __ shrptr(tmp1, G1HeapRegion::LogOfHRGrainBytes);               // ((store address ^ new value) >> LogOfHRGrainBytes) == 0?
318     __ jccb(Assembler::equal, L_done);
319   }
320 
321   // Crosses regions, storing null?
322   if (new_val_may_be_null) {
323     __ testptr(new_val, new_val);                                 // new value == null?
324     __ jccb(Assembler::equal, L_done);
325   }
326 
327   __ movptr(tmp1, store_addr);                                    // tmp1 := store address
328   __ shrptr(tmp1, CardTable::card_shift());                       // tmp1 := card address relative to card table base
329 
330   Address card_table_addr(thread, in_bytes(G1ThreadLocalData::card_table_base_offset()));
331   __ addptr(tmp1, card_table_addr);                               // tmp1 := card address
332   if (UseCondCardMark) {
333     __ cmpb(Address(tmp1, 0), G1CardTable::clean_card_val());     // *(card address) == clean_card_val?
334     __ jccb(Assembler::notEqual, L_done);
335   }
336   // Storing a region crossing, non-null oop, card is clean.
337   // Dirty card.
338   __ movb(Address(tmp1, 0), G1CardTable::dirty_card_val());       // *(card address) := dirty_card_val
339   __ bind(L_done);
340 }
341 
342 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
343                                                   Register store_addr,
344                                                   Register new_val,
345                                                   Register tmp) {
346   generate_post_barrier(masm, store_addr, new_val, tmp, true /* new_val_may_be_null */);
347 }
348 
349 #if defined(COMPILER2)
350 
351 static void generate_c2_barrier_runtime_call(MacroAssembler* masm, G1BarrierStubC2* stub, const Register arg, const address runtime_path) {
352   SaveLiveRegisters save_registers(masm, stub);
353   if (c_rarg0 != arg) {
354     __ mov(c_rarg0, arg);
355   }
356   __ mov(c_rarg1, r15_thread);
357   // rax is a caller-saved, non-argument-passing register, so it does not
358   // interfere with c_rarg0 or c_rarg1. If it contained any live value before
359   // entering this stub, it is saved at this point, and restored after the
360   // call. If it did not contain any live value, it is free to be used. In
361   // either case, it is safe to use it here as a call scratch register.
362   __ call(RuntimeAddress(runtime_path), rax);
363 }
364 
365 void G1BarrierSetAssembler::g1_write_barrier_pre_c2(MacroAssembler* masm,
366                                                     Register obj,
367                                                     Register pre_val,
368                                                     Register tmp,
369                                                     G1PreBarrierStubC2* stub) {
370   const Register thread = r15_thread;
371 
372   assert(pre_val != noreg, "check this code");
373   if (obj != noreg) {
374     assert_different_registers(obj, pre_val, tmp);
375   }
376 
377   stub->initialize_registers(obj, pre_val, thread, tmp);
378 
379   generate_pre_barrier_fast_path(masm, thread);
380   // If marking is active (*(mark queue active address) != 0), jump to stub (slow path)
381   __ jcc(Assembler::notEqual, *stub->entry());
382 
383   __ bind(*stub->continuation());
384 }
385 
386 void G1BarrierSetAssembler::generate_c2_pre_barrier_stub(MacroAssembler* masm,
387                                                          G1PreBarrierStubC2* stub) const {
388   Assembler::InlineSkippedInstructionsCounter skip_counter(masm);
389   Register obj = stub->obj();
390   Register pre_val = stub->pre_val();
391   Register thread = stub->thread();
392   Register tmp = stub->tmp1();
393   assert(stub->tmp2() == noreg, "not needed in this platform");
394 
395   __ bind(*stub->entry());
396   generate_pre_barrier_slow_path(masm, obj, pre_val, thread, tmp, *stub->continuation());
397 
398   generate_c2_barrier_runtime_call(masm, stub, pre_val, CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry));
399   __ jmp(*stub->continuation());
400 }
401 
402 void G1BarrierSetAssembler::g1_write_barrier_post_c2(MacroAssembler* masm,
403                                                      Register store_addr,
404                                                      Register new_val,
405                                                      Register tmp,
406                                                      bool new_val_may_be_null) {
407   generate_post_barrier(masm, store_addr, new_val, tmp, new_val_may_be_null);
408 }
409 
410 #endif // COMPILER2
411 
412 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
413                                          Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
414   bool in_heap = (decorators & IN_HEAP) != 0;
415   bool as_normal = (decorators & AS_NORMAL) != 0;
416 
417   bool needs_pre_barrier = as_normal;
418   bool needs_post_barrier = val != noreg && in_heap;
419 
420   // flatten object address if needed
421   // We do it regardless of precise because we need the registers
422   if (dst.index() == noreg && dst.disp() == 0) {
423     if (dst.base() != tmp1) {
424       __ movptr(tmp1, dst.base());
425     }
426   } else {
427     __ lea(tmp1, dst);
428   }
429 
430   if (needs_pre_barrier) {
431     g1_write_barrier_pre(masm /*masm*/,
432                          tmp1 /* obj */,
433                          tmp2 /* pre_val */,
434                          tmp3  /* tmp */,
435                          val != noreg /* tosca_live */,
436                          false /* expand_call */);
437   }
438   if (val == noreg) {
439     BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg, noreg);
440   } else {
441     Register new_val = val;
442     if (needs_post_barrier) {
443       // G1 barrier needs uncompressed oop for region cross check.
444       if (UseCompressedOops) {
445         new_val = tmp2;
446         __ movptr(new_val, val);
447       }
448     }
449     BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg, noreg);
450     if (needs_post_barrier) {
451       g1_write_barrier_post(masm /*masm*/,
452                             tmp1 /* store_adr */,
453                             new_val /* new_val */,
454                             tmp3 /* tmp */);
455     }
456   }
457 }
458 
459 #ifdef COMPILER1
460 
461 #undef __
462 #define __ ce->masm()->
463 
464 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
465   G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
466   // At this point we know that marking is in progress.
467   // If do_load() is true then we have to emit the
468   // load of the previous value; otherwise it has already
469   // been loaded into _pre_val.
470 
471   __ bind(*stub->entry());
472   assert(stub->pre_val()->is_register(), "Precondition.");
473 
474   Register pre_val_reg = stub->pre_val()->as_register();
475 
476   if (stub->do_load()) {
477     ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/);
478   }
479 
480   __ testptr(pre_val_reg, pre_val_reg);
481   __ jcc(Assembler::equal, *stub->continuation());
482   ce->store_parameter(stub->pre_val()->as_register(), 0);
483   __ call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
484   __ jmp(*stub->continuation());
485 
486 }
487 
488 #undef __
489 
490 void G1BarrierSetAssembler::g1_write_barrier_post_c1(MacroAssembler* masm,
491                                                      Register store_addr,
492                                                      Register new_val,
493                                                      Register thread,
494                                                      Register tmp1,
495                                                      Register tmp2 /* unused on x86 */) {
496   generate_post_barrier(masm, store_addr, new_val, tmp1, true /* new_val_may_be_null */);
497 }
498 
499 #define __ sasm->
500 
501 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
502   // Generated code assumes that buffer index is pointer sized.
503   STATIC_ASSERT(in_bytes(SATBMarkQueue::byte_width_of_index()) == sizeof(intptr_t));
504 
505   __ prologue("g1_pre_barrier", false);
506   // arg0 : previous value of memory
507 
508   __ push_ppx(rax);
509   __ push_ppx(rdx);
510 
511   const Register pre_val = rax;
512   const Register thread = r15_thread;
513   const Register tmp = rdx;
514 
515   Address queue_active(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
516   Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
517   Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
518 
519   Label L_done, L_runtime;
520 
521   // Is marking still active?
522   if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
523     __ cmpl(queue_active, 0);
524   } else {
525     assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
526     __ cmpb(queue_active, 0);
527   }
528   __ jcc(Assembler::equal, L_done);
529 
530   // Can we store original value in the thread's buffer?
531 
532   __ movptr(tmp, queue_index);
533   __ testptr(tmp, tmp);
534   __ jccb(Assembler::zero, L_runtime);
535   __ subptr(tmp, wordSize);
536   __ movptr(queue_index, tmp);
537   __ addptr(tmp, buffer);
538 
539   // prev_val (rax)
540   __ load_parameter(0, pre_val);
541   __ movptr(Address(tmp, 0), pre_val);
542   __ jmp(L_done);
543 
544   __ bind(L_runtime);
545 
546   __ push_call_clobbered_registers();
547 
548   // load the pre-value
549   __ load_parameter(0, rcx);
550   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), rcx, thread);
551 
552   __ pop_call_clobbered_registers();
553 
554   __ bind(L_done);
555 
556   __ pop_ppx(rdx);
557   __ pop_ppx(rax);
558 
559   __ epilogue();
560 }
561 
562 #undef __
563 
564 #endif // COMPILER1