1 /*
  2  * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
  3  * Copyright (c) 2014, 2021, 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 "c1/c1_MacroAssembler.hpp"
 27 #include "c1/c1_Runtime1.hpp"
 28 #include "gc/shared/barrierSetAssembler.hpp"
 29 #include "gc/shared/collectedHeap.hpp"
 30 #include "gc/shared/barrierSet.hpp"
 31 #include "gc/shared/barrierSetAssembler.hpp"
 32 #include "gc/shared/tlab_globals.hpp"
 33 #include "interpreter/interpreter.hpp"
 34 #include "oops/arrayOop.hpp"
 35 #include "oops/markWord.hpp"
 36 #include "runtime/basicLock.hpp"
 37 #include "runtime/os.hpp"
 38 #include "runtime/sharedRuntime.hpp"
 39 #include "runtime/stubRoutines.hpp"
 40 
 41 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
 42                                   FloatRegister f0, FloatRegister f1,
 43                                   Register result)
 44 {
 45   Label done;
 46   if (is_float) {
 47     fcmps(f0, f1);
 48   } else {
 49     fcmpd(f0, f1);
 50   }
 51   if (unordered_result < 0) {
 52     // we want -1 for unordered or less than, 0 for equal and 1 for
 53     // greater than.
 54     cset(result, NE);  // Not equal or unordered
 55     cneg(result, result, LT);  // Less than or unordered
 56   } else {
 57     // we want -1 for less than, 0 for equal and 1 for unordered or
 58     // greater than.
 59     cset(result, NE);  // Not equal or unordered
 60     cneg(result, result, LO);  // Less than
 61   }
 62 }
 63 
 64 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register temp, Label& slow_case) {
 65   const int aligned_mask = BytesPerWord -1;
 66   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 67   assert_different_registers(hdr, obj, disp_hdr, temp, rscratch2);
 68   int null_check_offset = -1;
 69 
 70   verify_oop(obj);
 71 
 72   // save object being locked into the BasicObjectLock
 73   str(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
 74 
 75   null_check_offset = offset();
 76 
 77   if (LockingMode == LM_LIGHTWEIGHT) {
 78     lightweight_lock(disp_hdr, obj, hdr, temp, rscratch2, slow_case);
 79   } else if (LockingMode == LM_LEGACY) {
 80 
 81     if (DiagnoseSyncOnValueBasedClasses != 0) {
 82       load_klass(hdr, obj);
 83       ldrb(hdr, Address(hdr, Klass::misc_flags_offset()));
 84       tst(hdr, KlassFlags::_misc_is_value_based_class);
 85       br(Assembler::NE, slow_case);
 86     }
 87 
 88     Label done;
 89     // Load object header
 90     ldr(hdr, Address(obj, hdr_offset));
 91     // and mark it as unlocked
 92     orr(hdr, hdr, markWord::unlocked_value);
 93 
 94     if (EnableValhalla) {
 95       // Mask always_locked bit such that we go to the slow path if object is an inline type
 96       andr(hdr, hdr, ~markWord::inline_type_bit_in_place);
 97     }
 98 
 99     // save unlocked object header into the displaced header location on the stack
100     str(hdr, Address(disp_hdr, 0));
101     // test if object header is still the same (i.e. unlocked), and if so, store the
102     // displaced header address in the object header - if it is not the same, get the
103     // object header instead
104     lea(rscratch2, Address(obj, hdr_offset));
105     cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/nullptr);
106     // if the object header was the same, we're done
107     // if the object header was not the same, it is now in the hdr register
108     // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
109     //
110     // 1) (hdr & aligned_mask) == 0
111     // 2) sp <= hdr
112     // 3) hdr <= sp + page_size
113     //
114     // these 3 tests can be done by evaluating the following expression:
115     //
116     // (hdr - sp) & (aligned_mask - page_size)
117     //
118     // assuming both the stack pointer and page_size have their least
119     // significant 2 bits cleared and page_size is a power of 2
120     mov(rscratch1, sp);
121     sub(hdr, hdr, rscratch1);
122     ands(hdr, hdr, aligned_mask - (int)os::vm_page_size());
123     // for recursive locking, the result is zero => save it in the displaced header
124     // location (null in the displaced hdr location indicates recursive locking)
125     str(hdr, Address(disp_hdr, 0));
126     // otherwise we don't care about the result and handle locking via runtime call
127     cbnz(hdr, slow_case);
128     // done
129     bind(done);
130     inc_held_monitor_count(rscratch1);
131   }
132   return null_check_offset;
133 }
134 
135 
136 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Register temp, Label& slow_case) {
137   const int aligned_mask = BytesPerWord -1;
138   const int hdr_offset = oopDesc::mark_offset_in_bytes();
139   assert_different_registers(hdr, obj, disp_hdr, temp, rscratch2);
140   Label done;
141 
142   if (LockingMode != LM_LIGHTWEIGHT) {
143     // load displaced header
144     ldr(hdr, Address(disp_hdr, 0));
145     // if the loaded hdr is null we had recursive locking
146     // if we had recursive locking, we are done
147     cbz(hdr, done);
148   }
149 
150   // load object
151   ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
152   verify_oop(obj);
153 
154   if (LockingMode == LM_LIGHTWEIGHT) {
155     lightweight_unlock(obj, hdr, temp, rscratch2, slow_case);
156   } else if (LockingMode == LM_LEGACY) {
157     // test if object header is pointing to the displaced header, and if so, restore
158     // the displaced header in the object - if the object header is not pointing to
159     // the displaced header, get the object header instead
160     // if the object header was not pointing to the displaced header,
161     // we do unlocking via runtime call
162     if (hdr_offset) {
163       lea(rscratch1, Address(obj, hdr_offset));
164       cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case);
165     } else {
166       cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
167     }
168     // done
169     bind(done);
170     dec_held_monitor_count(rscratch1);
171   }
172 }
173 
174 
175 // Defines obj, preserves var_size_in_bytes
176 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
177   if (UseTLAB) {
178     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
179   } else {
180     b(slow_case);
181   }
182 }
183 
184 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
185   assert_different_registers(obj, klass, len);
186 
187   if (UseCompactObjectHeaders || EnableValhalla) {
188     ldr(t1, Address(klass, Klass::prototype_header_offset()));
189     str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
190   } else {
191     mov(t1, checked_cast<int32_t>(markWord::prototype().value()));
192     str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
193   }
194 
195   if (!UseCompactObjectHeaders) {
196     if (UseCompressedClassPointers) { // Take care not to kill klass
197       encode_klass_not_null(t1, klass);
198       strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
199     } else {
200       str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
201     }
202   }
203 
204   if (len->is_valid()) {
205     strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
206     int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
207     if (!is_aligned(base_offset, BytesPerWord)) {
208       assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
209       // Clear gap/first 4 bytes following the length field.
210       strw(zr, Address(obj, base_offset));
211     }
212   } else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
213     store_klass_gap(obj, zr);
214   }
215 }
216 
217 // preserves obj, destroys len_in_bytes
218 //
219 // Scratch registers: t1 = r10, t2 = r11
220 //
221 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1, Register t2) {
222   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
223   assert(t1 == r10 && t2 == r11, "must be");
224 
225   Label done;
226 
227   // len_in_bytes is positive and ptr sized
228   subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
229   br(Assembler::EQ, done);
230 
231   // zero_words() takes ptr in r10 and count in words in r11
232   mov(rscratch1, len_in_bytes);
233   lea(t1, Address(obj, hdr_size_in_bytes));
234   lsr(t2, rscratch1, LogBytesPerWord);
235   address tpc = zero_words(t1, t2);
236 
237   bind(done);
238   if (tpc == nullptr) {
239     Compilation::current()->bailout("no space for trampoline stub");
240   }
241 }
242 
243 
244 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
245   assert_different_registers(obj, t1, t2); // XXX really?
246   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
247 
248   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
249 
250   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
251 }
252 
253 // Scratch registers: t1 = r10, t2 = r11
254 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) {
255   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
256          "con_size_in_bytes is not multiple of alignment");
257   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
258 
259   initialize_header(obj, klass, noreg, t1, t2);
260 
261   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
262      // clear rest of allocated space
263      const Register index = t2;
264      if (var_size_in_bytes != noreg) {
265        mov(index, var_size_in_bytes);
266        initialize_body(obj, index, hdr_size_in_bytes, t1, t2);
267        if (Compilation::current()->bailed_out()) {
268          return;
269        }
270      } else if (con_size_in_bytes > hdr_size_in_bytes) {
271        con_size_in_bytes -= hdr_size_in_bytes;
272        lea(t1, Address(obj, hdr_size_in_bytes));
273        address tpc = zero_words(t1, con_size_in_bytes / BytesPerWord);
274        if (tpc == nullptr) {
275          Compilation::current()->bailout("no space for trampoline stub");
276          return;
277        }
278      }
279   }
280 
281   membar(StoreStore);
282 
283   if (CURRENT_ENV->dtrace_alloc_probes()) {
284     assert(obj == r0, "must be");
285     far_call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
286   }
287 
288   verify_oop(obj);
289 }
290 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int base_offset_in_bytes, int f, Register klass, Label& slow_case, bool zero_array) {
291   assert_different_registers(obj, len, t1, t2, klass);
292 
293   // determine alignment mask
294   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
295 
296   // check for negative or excessive length
297   mov(rscratch1, (int32_t)max_array_allocation_length);
298   cmp(len, rscratch1);
299   br(Assembler::HS, slow_case);
300 
301   const Register arr_size = t2; // okay to be the same
302   // align object end
303   mov(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
304   add(arr_size, arr_size, len, ext::uxtw, f);
305   andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
306 
307   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
308 
309   initialize_header(obj, klass, len, t1, t2);
310 
311   // Align-up to word boundary, because we clear the 4 bytes potentially
312   // following the length field in initialize_header().
313   int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
314   // clear rest of allocated space
315   if (zero_array) {
316     initialize_body(obj, arr_size, base_offset, t1, t2);
317   }
318   if (Compilation::current()->bailed_out()) {
319     return;
320   }
321 
322   membar(StoreStore);
323 
324   if (CURRENT_ENV->dtrace_alloc_probes()) {
325     assert(obj == r0, "must be");
326     far_call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
327   }
328 
329   verify_oop(obj);
330 }
331 
332 void C1_MacroAssembler::build_frame_helper(int frame_size_in_bytes, int sp_offset_for_orig_pc, int sp_inc, bool reset_orig_pc, bool needs_stack_repair) {
333   MacroAssembler::build_frame(frame_size_in_bytes);
334 
335   if (needs_stack_repair) {
336     save_stack_increment(sp_inc, frame_size_in_bytes);
337   }
338   if (reset_orig_pc) {
339     // Zero orig_pc to detect deoptimization during buffering in the entry points
340     str(zr, Address(sp, sp_offset_for_orig_pc));
341   }
342 }
343 
344 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes, int sp_offset_for_orig_pc, bool needs_stack_repair, bool has_scalarized_args, Label* verified_inline_entry_label) {
345   // Make sure there is enough stack space for this method's activation.
346   // Note that we do this before creating a frame.
347   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
348   generate_stack_overflow_check(bang_size_in_bytes);
349 
350   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, has_scalarized_args, needs_stack_repair);
351 
352   // Insert nmethod entry barrier into frame.
353   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
354   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
355 
356   if (verified_inline_entry_label != nullptr) {
357     // Jump here from the scalarized entry points that already created the frame.
358     bind(*verified_inline_entry_label);
359   }
360 }
361 
362 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
363   // If we have to make this method not-entrant we'll overwrite its
364   // first instruction with a jump.  For this action to be legal we
365   // must ensure that this first instruction is a B, BL, NOP, BKPT,
366   // SVC, HVC, or SMC.  Make it a NOP.
367   nop();
368   if (C1Breakpoint) brk(1);
369 }
370 
371 int C1_MacroAssembler::scalarized_entry(const CompiledEntrySignature* ces, int frame_size_in_bytes, int bang_size_in_bytes, int sp_offset_for_orig_pc, Label& verified_inline_entry_label, bool is_inline_ro_entry) {
372   assert(InlineTypePassFieldsAsArgs, "sanity");
373   // Make sure there is enough stack space for this method's activation.
374   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
375   generate_stack_overflow_check(bang_size_in_bytes);
376 
377   GrowableArray<SigEntry>* sig    = ces->sig();
378   GrowableArray<SigEntry>* sig_cc = is_inline_ro_entry ? ces->sig_cc_ro() : ces->sig_cc();
379   VMRegPair* regs      = ces->regs();
380   VMRegPair* regs_cc   = is_inline_ro_entry ? ces->regs_cc_ro() : ces->regs_cc();
381   int args_on_stack    = ces->args_on_stack();
382   int args_on_stack_cc = is_inline_ro_entry ? ces->args_on_stack_cc_ro() : ces->args_on_stack_cc();
383 
384   assert(sig->length() <= sig_cc->length(), "Zero-sized inline class not allowed!");
385   BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sig_cc->length());
386   int args_passed = sig->length();
387   int args_passed_cc = SigEntry::fill_sig_bt(sig_cc, sig_bt);
388 
389   // Create a temp frame so we can call into the runtime. It must be properly set up to accommodate GC.
390   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, true, ces->c1_needs_stack_repair());
391 
392   // The runtime call might safepoint, make sure nmethod entry barrier is executed
393   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
394   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
395   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
396 
397   // FIXME -- call runtime only if we cannot in-line allocate all the incoming inline type args.
398   mov(r19, (intptr_t) ces->method());
399   if (is_inline_ro_entry) {
400     far_call(RuntimeAddress(Runtime1::entry_for(C1StubId::buffer_inline_args_no_receiver_id)));
401   } else {
402     far_call(RuntimeAddress(Runtime1::entry_for(C1StubId::buffer_inline_args_id)));
403   }
404   int rt_call_offset = offset();
405 
406   // The runtime call returns the new array in r20 instead of the usual r0
407   // because r0 is also j_rarg7 which may be holding a live argument here.
408   Register val_array = r20;
409 
410   // Remove the temp frame
411   MacroAssembler::remove_frame(frame_size_in_bytes);
412 
413   // Check if we need to extend the stack for packing
414   int sp_inc = 0;
415   if (args_on_stack > args_on_stack_cc) {
416     sp_inc = extend_stack_for_inline_args(args_on_stack);
417   }
418 
419   shuffle_inline_args(true, is_inline_ro_entry, sig_cc,
420                       args_passed_cc, args_on_stack_cc, regs_cc, // from
421                       args_passed, args_on_stack, regs,          // to
422                       sp_inc, val_array);
423 
424   // Create the real frame. Below jump will then skip over the stack banging and frame
425   // setup code in the verified_inline_entry (which has a different real_frame_size).
426   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, sp_inc, false, ces->c1_needs_stack_repair());
427 
428   b(verified_inline_entry_label);
429   return rt_call_offset;
430 }
431 
432 
433 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
434   // rfp, + 0: link
435   //     + 1: return address
436   //     + 2: argument with offset 0
437   //     + 3: argument with offset 1
438   //     + 4: ...
439 
440   ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
441 }
442 
443 #ifndef PRODUCT
444 
445 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
446   if (!VerifyOops) return;
447   verify_oop_addr(Address(sp, stack_offset));
448 }
449 
450 void C1_MacroAssembler::verify_not_null_oop(Register r) {
451   if (!VerifyOops) return;
452   Label not_null;
453   cbnz(r, not_null);
454   stop("non-null oop required");
455   bind(not_null);
456   verify_oop(r);
457 }
458 
459 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
460 #ifdef ASSERT
461   static int nn;
462   if (inv_r0) mov(r0, 0xDEAD);
463   if (inv_r19) mov(r19, 0xDEAD);
464   if (inv_r2) mov(r2, nn++);
465   if (inv_r3) mov(r3, 0xDEAD);
466   if (inv_r4) mov(r4, 0xDEAD);
467   if (inv_r5) mov(r5, 0xDEAD);
468 #endif
469 }
470 #endif // ifndef PRODUCT