1 /*
  2  * Copyright (c) 1999, 2026, 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/arguments.hpp"
 37 #include "runtime/basicLock.hpp"
 38 #include "runtime/os.hpp"
 39 #include "runtime/sharedRuntime.hpp"
 40 #include "runtime/stubRoutines.hpp"
 41 
 42 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
 43                                   FloatRegister f0, FloatRegister f1,
 44                                   Register result)
 45 {
 46   Label done;
 47   if (is_float) {
 48     fcmps(f0, f1);
 49   } else {
 50     fcmpd(f0, f1);
 51   }
 52   if (unordered_result < 0) {
 53     // we want -1 for unordered or less than, 0 for equal and 1 for
 54     // greater than.
 55     cset(result, NE);  // Not equal or unordered
 56     cneg(result, result, LT);  // Less than or unordered
 57   } else {
 58     // we want -1 for less than, 0 for equal and 1 for unordered or
 59     // greater than.
 60     cset(result, NE);  // Not equal or unordered
 61     cneg(result, result, LO);  // Less than
 62   }
 63 }
 64 
 65 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register basic_lock, Register temp, Label& slow_case) {
 66   assert_different_registers(hdr, obj, basic_lock, temp, rscratch2);
 67   int null_check_offset = -1;
 68 
 69   verify_oop(obj);
 70 
 71   // save object being locked into the BasicObjectLock
 72   str(obj, Address(basic_lock, BasicObjectLock::obj_offset()));
 73 
 74   null_check_offset = offset();
 75 
 76   fast_lock(basic_lock, obj, hdr, temp, rscratch2, slow_case);
 77 
 78   return null_check_offset;
 79 }
 80 
 81 
 82 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register basic_lock, Register temp, Label& slow_case) {
 83   assert_different_registers(hdr, obj, basic_lock, temp, rscratch2);
 84 
 85   // load object
 86   ldr(obj, Address(basic_lock, BasicObjectLock::obj_offset()));
 87   verify_oop(obj);
 88 
 89   fast_unlock(obj, hdr, temp, rscratch2, slow_case);
 90 }
 91 
 92 
 93 // Defines obj, preserves var_size_in_bytes
 94 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
 95   if (UseTLAB) {
 96     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 97   } else {
 98     b(slow_case);
 99   }
100 }
101 
102 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
103   assert_different_registers(obj, klass, len);
104 
105   if (UseCompactObjectHeaders || Arguments::is_valhalla_enabled()) {
106     // COH: Markword contains class pointer which is only known at runtime.
107     // Valhalla: Could have value class which has a different prototype header to a normal object.
108     // In both cases, we need to fetch dynamically.
109     ldr(t1, Address(klass, Klass::prototype_header_offset()));
110     str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
111   } else {
112     // Otherwise: Can use the statically computed prototype header which is the same for every object.
113     mov(t1, checked_cast<int32_t>(markWord::prototype().value()));
114     str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
115   }
116 
117   if (!UseCompactObjectHeaders) {
118     // COH: Markword already contains class pointer. Nothing else to do.
119     // Otherwise: Store encoded klass pointer following the markword
120     encode_klass_not_null(t1, klass); // Take care not to kill klass
121     strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
122   }
123 
124   if (len->is_valid()) {
125     strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
126     int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
127     if (!is_aligned(base_offset, BytesPerWord)) {
128       assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
129       // Clear gap/first 4 bytes following the length field.
130       strw(zr, Address(obj, base_offset));
131     }
132   } else if (!UseCompactObjectHeaders) {
133     store_klass_gap(obj, zr);
134   }
135 }
136 
137 // preserves obj, destroys len_in_bytes
138 //
139 // Scratch registers: t1 = r10, t2 = r11
140 //
141 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1, Register t2) {
142   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
143   assert(t1 == r10 && t2 == r11, "must be");
144 
145   Label done;
146 
147   // len_in_bytes is positive and ptr sized
148   subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
149   br(Assembler::EQ, done);
150 
151   // zero_words() takes ptr in r10 and count in words in r11
152   mov(rscratch1, len_in_bytes);
153   lea(t1, Address(obj, hdr_size_in_bytes));
154   lsr(t2, rscratch1, LogBytesPerWord);
155   address tpc = zero_words(t1, t2);
156 
157   bind(done);
158   if (tpc == nullptr) {
159     Compilation::current()->bailout("no space for trampoline stub");
160   }
161 }
162 
163 
164 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
165   assert_different_registers(obj, t1, t2); // XXX really?
166   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
167 
168   try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
169 
170   initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
171 }
172 
173 // Scratch registers: t1 = r10, t2 = r11
174 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) {
175   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
176          "con_size_in_bytes is not multiple of alignment");
177   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
178 
179   initialize_header(obj, klass, noreg, t1, t2);
180 
181   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
182      // clear rest of allocated space
183      const Register index = t2;
184      if (var_size_in_bytes != noreg) {
185        mov(index, var_size_in_bytes);
186        initialize_body(obj, index, hdr_size_in_bytes, t1, t2);
187        if (Compilation::current()->bailed_out()) {
188          return;
189        }
190      } else if (con_size_in_bytes > hdr_size_in_bytes) {
191        con_size_in_bytes -= hdr_size_in_bytes;
192        lea(t1, Address(obj, hdr_size_in_bytes));
193        address tpc = zero_words(t1, con_size_in_bytes / BytesPerWord);
194        if (tpc == nullptr) {
195          Compilation::current()->bailout("no space for trampoline stub");
196          return;
197        }
198      }
199   }
200 
201   membar(StoreStore);
202 
203   if (CURRENT_ENV->dtrace_alloc_probes()) {
204     assert(obj == r0, "must be");
205     far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_dtrace_object_alloc_id)));
206   }
207 
208   verify_oop(obj);
209 }
210 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) {
211   assert_different_registers(obj, len, t1, t2, klass);
212 
213   // determine alignment mask
214   assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
215 
216   // check for negative or excessive length
217   mov(rscratch1, (int32_t)max_array_allocation_length);
218   cmp(len, rscratch1);
219   br(Assembler::HS, slow_case);
220 
221   const Register arr_size = t2; // okay to be the same
222   // align object end
223   mov(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
224   add(arr_size, arr_size, len, ext::uxtw, f);
225   andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
226 
227   try_allocate(obj, arr_size, 0, t1, t2, slow_case);
228 
229   initialize_header(obj, klass, len, t1, t2);
230 
231   // Align-up to word boundary, because we clear the 4 bytes potentially
232   // following the length field in initialize_header().
233   int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
234   // clear rest of allocated space
235   if (zero_array) {
236     initialize_body(obj, arr_size, base_offset, t1, t2);
237   }
238   if (Compilation::current()->bailed_out()) {
239     return;
240   }
241 
242   membar(StoreStore);
243 
244   if (CURRENT_ENV->dtrace_alloc_probes()) {
245     assert(obj == r0, "must be");
246     far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_dtrace_object_alloc_id)));
247   }
248 
249   verify_oop(obj);
250 }
251 
252 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) {
253   MacroAssembler::build_frame(frame_size_in_bytes);
254 
255   if (needs_stack_repair) {
256     save_stack_increment(sp_inc, frame_size_in_bytes);
257   }
258   if (reset_orig_pc) {
259     // Zero orig_pc to detect deoptimization during buffering in the entry points
260     str(zr, Address(sp, sp_offset_for_orig_pc));
261   }
262 }
263 
264 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) {
265   // Make sure there is enough stack space for this method's activation.
266   // Note that we do this before creating a frame.
267   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
268   generate_stack_overflow_check(bang_size_in_bytes);
269 
270   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, has_scalarized_args, needs_stack_repair);
271 
272   // Insert nmethod entry barrier into frame.
273   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
274   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
275 
276   if (verified_inline_entry_label != nullptr) {
277     // Jump here from the scalarized entry points that already created the frame.
278     bind(*verified_inline_entry_label);
279   }
280 }
281 
282 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
283   // If we have to make this method not-entrant we'll overwrite its
284   // first instruction with a jump.  For this action to be legal we
285   // must ensure that this first instruction is a B, BL, NOP, BKPT,
286   // SVC, HVC, or SMC.  Make it a NOP.
287   nop();
288   if (C1Breakpoint) brk(1);
289 }
290 
291 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) {
292   assert(InlineTypePassFieldsAsArgs, "sanity");
293   // Make sure there is enough stack space for this method's activation.
294   assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
295   generate_stack_overflow_check(bang_size_in_bytes);
296 
297   GrowableArray<SigEntry>* sig    = ces->sig();
298   GrowableArray<SigEntry>* sig_cc = is_inline_ro_entry ? ces->sig_cc_ro() : ces->sig_cc();
299   VMRegPair* regs      = ces->regs();
300   VMRegPair* regs_cc   = is_inline_ro_entry ? ces->regs_cc_ro() : ces->regs_cc();
301   int args_on_stack    = ces->args_on_stack();
302   int args_on_stack_cc = is_inline_ro_entry ? ces->args_on_stack_cc_ro() : ces->args_on_stack_cc();
303 
304   assert(sig->length() <= sig_cc->length(), "Zero-sized inline class not allowed!");
305   BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sig_cc->length());
306   int args_passed = sig->length();
307   int args_passed_cc = SigEntry::fill_sig_bt(sig_cc, sig_bt);
308 
309   // Create a temp frame so we can call into the runtime. It must be properly set up to accommodate GC.
310   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, 0, true, ces->c1_needs_stack_repair());
311 
312   // The runtime call might safepoint, make sure nmethod entry barrier is executed
313   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
314   // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
315   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
316 
317   // FIXME -- call runtime only if we cannot in-line allocate all the incoming inline type args.
318   mov(r19, (intptr_t) ces->method());
319   if (is_inline_ro_entry) {
320     far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_buffer_inline_args_no_receiver_id)));
321   } else {
322     far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_buffer_inline_args_id)));
323   }
324   int rt_call_offset = offset();
325 
326   // The runtime call returns the new array in r20 instead of the usual r0
327   // because r0 is also j_rarg7 which may be holding a live argument here.
328   Register val_array = r20;
329 
330   // Remove the temp frame
331   MacroAssembler::remove_frame(frame_size_in_bytes);
332 
333   // Check if we need to extend the stack for packing
334   int sp_inc = 0;
335   if (args_on_stack > args_on_stack_cc) {
336     sp_inc = extend_stack_for_inline_args(args_on_stack);
337   }
338 
339   shuffle_inline_args(true, is_inline_ro_entry, sig_cc,
340                       args_passed_cc, args_on_stack_cc, regs_cc, // from
341                       args_passed, args_on_stack, regs,          // to
342                       sp_inc, val_array);
343 
344   // Create the real frame. Below jump will then skip over the stack banging and frame
345   // setup code in the verified_inline_entry (which has a different real_frame_size).
346   build_frame_helper(frame_size_in_bytes, sp_offset_for_orig_pc, sp_inc, false, ces->c1_needs_stack_repair());
347 
348   b(verified_inline_entry_label);
349   return rt_call_offset;
350 }
351 
352 
353 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
354   // rfp, + 0: link
355   //     + 1: return address
356   //     + 2: argument with offset 0
357   //     + 3: argument with offset 1
358   //     + 4: ...
359 
360   ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
361 }
362 
363 #ifndef PRODUCT
364 
365 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
366   if (!VerifyOops) return;
367   verify_oop_addr(Address(sp, stack_offset));
368 }
369 
370 void C1_MacroAssembler::verify_not_null_oop(Register r) {
371   if (!VerifyOops) return;
372   Label not_null;
373   cbnz(r, not_null);
374   stop("non-null oop required");
375   bind(not_null);
376   verify_oop(r);
377 }
378 
379 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
380 #ifdef ASSERT
381   static int nn;
382   if (inv_r0) mov(r0, 0xDEAD);
383   if (inv_r19) mov(r19, 0xDEAD);
384   if (inv_r2) mov(r2, nn++);
385   if (inv_r3) mov(r3, 0xDEAD);
386   if (inv_r4) mov(r4, 0xDEAD);
387   if (inv_r5) mov(r5, 0xDEAD);
388 #endif
389 }
390 #endif // ifndef PRODUCT