1 /*
  2  * Copyright (c) 1999, 2026, Oracle and/or its affiliates. All rights reserved.
  3  * Copyright (c) 2014, Red Hat Inc. All rights reserved.
  4  * Copyright (c) 2020, 2022, Huawei Technologies Co., Ltd. All rights reserved.
  5  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  6  *
  7  * This code is free software; you can redistribute it and/or modify it
  8  * under the terms of the GNU General Public License version 2 only, as
  9  * published by the Free Software Foundation.
 10  *
 11  * This code is distributed in the hope that it will be useful, but WITHOUT
 12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 14  * version 2 for more details (a copy is included in the LICENSE file that
 15  * accompanied this code).
 16  *
 17  * You should have received a copy of the GNU General Public License version
 18  * 2 along with this work; if not, write to the Free Software Foundation,
 19  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 20  *
 21  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 22  * or visit www.oracle.com if you need additional information or have any
 23  * questions.
 24  *
 25  */
 26 
 27 #include "c1/c1_LIR.hpp"
 28 #include "c1/c1_MacroAssembler.hpp"
 29 #include "c1/c1_Runtime1.hpp"
 30 #include "classfile/systemDictionary.hpp"
 31 #include "gc/shared/barrierSetAssembler.hpp"
 32 #include "gc/shared/collectedHeap.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 freg0, FloatRegister freg1,
 43                                   Register result) {
 44   if (is_float) {
 45     float_compare(result, freg0, freg1, unordered_result);
 46   } else {
 47     double_compare(result, freg0, freg1, unordered_result);
 48   }
 49 }
 50 
 51 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register basic_lock, Register temp, Label& slow_case) {
 52   assert_different_registers(hdr, obj, basic_lock, temp, t0, t1);
 53   int null_check_offset = -1;
 54 
 55   verify_oop(obj);
 56 
 57   // save object being locked into the BasicObjectLock
 58   sd(obj, Address(basic_lock, BasicObjectLock::obj_offset()));
 59 
 60   null_check_offset = offset();
 61 
 62   fast_lock(basic_lock, obj, hdr, temp, t1, slow_case);
 63 
 64   return null_check_offset;
 65 }
 66 
 67 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register basic_lock, Register temp, Label& slow_case) {
 68   assert_different_registers(hdr, obj, basic_lock, temp, t0, t1);
 69 
 70   // load object
 71   ld(obj, Address(basic_lock, BasicObjectLock::obj_offset()));
 72   verify_oop(obj);
 73 
 74   fast_unlock(obj, hdr, temp, t1, slow_case);
 75 }
 76 
 77 // Defines obj, preserves var_size_in_bytes
 78 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register tmp1, Register tmp2, Label& slow_case) {
 79   if (UseTLAB) {
 80     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, tmp1, tmp2, slow_case, /* is_far */ true);
 81   } else {
 82     j(slow_case);
 83   }
 84 }
 85 
 86 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register tmp1, Register tmp2) {
 87   assert_different_registers(obj, klass, len, tmp1, tmp2);
 88   if (UseCompactObjectHeaders) {
 89     ld(tmp1, Address(klass, Klass::prototype_header_offset()));
 90     sd(tmp1, Address(obj, oopDesc::mark_offset_in_bytes()));
 91   } else {
 92     // This assumes that all prototype bits fitr in an int32_t
 93     mv(tmp1, checked_cast<int32_t>(markWord::prototype().value()));
 94     sd(tmp1, Address(obj, oopDesc::mark_offset_in_bytes()));
 95     encode_klass_not_null(tmp1, klass, tmp2);
 96     sw(tmp1, Address(obj, oopDesc::klass_offset_in_bytes()));
 97   }
 98 
 99   if (len->is_valid()) {
100     sw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
101     int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
102     if (!is_aligned(base_offset, BytesPerWord)) {
103       assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
104       // Clear gap/first 4 bytes following the length field.
105       sw(zr, Address(obj, base_offset));
106     }
107   } else if (!UseCompactObjectHeaders) {
108     store_klass_gap(obj, zr);
109   }
110 }
111 
112 // preserves obj, destroys len_in_bytes
113 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register tmp) {
114   assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
115   Label done;
116 
117   // len_in_bytes is positive and ptr sized
118   subi(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
119   beqz(len_in_bytes, done);
120 
121   // Preserve obj
122   if (hdr_size_in_bytes) {
123     addi(obj, obj, hdr_size_in_bytes);
124   }
125   zero_memory(obj, len_in_bytes, tmp);
126   if (hdr_size_in_bytes) {
127     subi(obj, obj, hdr_size_in_bytes);
128   }
129 
130   bind(done);
131 }
132 
133 void C1_MacroAssembler::allocate_object(Register obj, Register tmp1, Register tmp2, int header_size, int object_size, Register klass, Label& slow_case) {
134   assert_different_registers(obj, tmp1, tmp2);
135   assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
136 
137   try_allocate(obj, noreg, object_size * BytesPerWord, tmp1, tmp2, slow_case);
138 
139   initialize_object(obj, klass, noreg, object_size * HeapWordSize, tmp1, tmp2, UseTLAB);
140 }
141 
142 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register tmp1, Register tmp2, bool is_tlab_allocated) {
143   assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
144          "con_size_in_bytes is not multiple of alignment");
145   const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
146 
147   initialize_header(obj, klass, noreg, tmp1, tmp2);
148 
149   if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
150     // clear rest of allocated space
151     const Register index = tmp2;
152     // 16: multiplier for threshold
153     const int threshold = 16 * BytesPerWord;    // approximate break even point for code size (see comments below)
154     if (var_size_in_bytes != noreg) {
155       mv(index, var_size_in_bytes);
156       initialize_body(obj, index, hdr_size_in_bytes, tmp1);
157     } else if (con_size_in_bytes <= threshold) {
158       // use explicit null stores
159       int i = hdr_size_in_bytes;
160       if (i < con_size_in_bytes && (con_size_in_bytes % (2 * BytesPerWord))) { // 2: multiplier for BytesPerWord
161         sd(zr, Address(obj, i));
162         i += BytesPerWord;
163       }
164       for (; i < con_size_in_bytes; i += BytesPerWord) {
165         sd(zr, Address(obj, i));
166       }
167     } else if (con_size_in_bytes > hdr_size_in_bytes) {
168       block_comment("zero memory");
169       // use loop to null out the fields
170       int words = (con_size_in_bytes - hdr_size_in_bytes) / BytesPerWord;
171       mv(index, words / 8); // 8: byte size
172 
173       const int unroll = 8; // Number of sd(zr) instructions we'll unroll
174       int remainder = words % unroll;
175       la(t0, Address(obj, hdr_size_in_bytes + remainder * BytesPerWord));
176 
177       Label entry_point, loop;
178       j(entry_point);
179 
180       bind(loop);
181       subi(index, index, 1);
182       for (int i = -unroll; i < 0; i++) {
183         if (-i == remainder) {
184           bind(entry_point);
185         }
186         sd(zr, Address(t0, i * wordSize));
187       }
188       if (remainder == 0) {
189         bind(entry_point);
190       }
191       addi(t0, t0, unroll * wordSize);
192       bnez(index, loop);
193     }
194   }
195 
196   membar(MacroAssembler::StoreStore);
197 
198   if (CURRENT_ENV->dtrace_alloc_probes()) {
199     assert(obj == x10, "must be");
200     far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_dtrace_object_alloc_id)));
201   }
202 
203   verify_oop(obj);
204 }
205 
206 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register tmp1, Register tmp2, int base_offset_in_bytes, int f, Register klass, Label& slow_case, bool zero_array) {
207   assert_different_registers(obj, len, tmp1, tmp2, klass);
208 
209   // determine alignment mask
210   assert(!(BytesPerWord & 1), "must be multiple of 2 for masking code to work");
211 
212   // check for negative or excessive length
213   mv(t0, (int32_t)max_array_allocation_length);
214   bgeu(len, t0, slow_case, /* is_far */ true);
215 
216   const Register arr_size = tmp2; // okay to be the same
217   // align object end
218   mv(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
219   shadd(arr_size, len, arr_size, t0, f);
220   andi(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
221 
222   try_allocate(obj, arr_size, 0, tmp1, tmp2, slow_case);
223 
224   initialize_header(obj, klass, len, tmp1, tmp2);
225 
226   // Align-up to word boundary, because we clear the 4 bytes potentially
227   // following the length field in initialize_header().
228   int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
229 
230   // clear rest of allocated space
231   const Register len_zero = len;
232   if (zero_array) {
233     initialize_body(obj, arr_size, base_offset, len_zero);
234   }
235 
236   membar(MacroAssembler::StoreStore);
237 
238   if (CURRENT_ENV->dtrace_alloc_probes()) {
239     assert(obj == x10, "must be");
240     far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_dtrace_object_alloc_id)));
241   }
242 
243   verify_oop(obj);
244 }
245 
246 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) {
247   assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
248   // Make sure there is enough stack space for this method's activation.
249   // Note that we do this before creating a frame.
250   generate_stack_overflow_check(bang_size_in_bytes);
251   MacroAssembler::build_frame(framesize);
252 
253   // Insert nmethod entry barrier into frame.
254   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
255   bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
256 }
257 
258 void C1_MacroAssembler::remove_frame(int framesize) {
259   MacroAssembler::remove_frame(framesize);
260 }
261 
262 
263 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
264   // If we have to make this method not-entrant we'll overwrite its
265   // first instruction with a jump. For this action to be legal we
266   // must ensure that this first instruction is a J, JAL or NOP.
267   // Make it a NOP.
268   IncompressibleScope scope(this); // keep the nop as 4 bytes for patching.
269   assert_alignment(pc());
270   nop();  // 4 bytes
271 }
272 
273 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
274   //  fp + -2: link
275   //     + -1: return address
276   //     +  0: argument with offset 0
277   //     +  1: argument with offset 1
278   //     +  2: ...
279   ld(reg, Address(fp, offset_in_words * BytesPerWord));
280 }
281 
282 #ifndef PRODUCT
283 
284 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
285   if (!VerifyOops) {
286     return;
287   }
288   verify_oop_addr(Address(sp, stack_offset));
289 }
290 
291 void C1_MacroAssembler::verify_not_null_oop(Register r) {
292   if (!VerifyOops) return;
293   Label not_null;
294   bnez(r, not_null);
295   stop("non-null oop required");
296   bind(not_null);
297   verify_oop(r);
298 }
299 
300 void C1_MacroAssembler::invalidate_registers(bool inv_x10, bool inv_x9, bool inv_x12, bool inv_x13, bool inv_x14, bool inv_x15) {
301 #ifdef ASSERT
302   static int nn;
303   if (inv_x10) { mv(x10, 0xDEAD); }
304   if (inv_x9)  { mv(x9, 0xDEAD);  }
305   if (inv_x12) { mv(x12, nn++);   }
306   if (inv_x13) { mv(x13, 0xDEAD); }
307   if (inv_x14) { mv(x14, 0xDEAD); }
308   if (inv_x15) { mv(x15, 0xDEAD); }
309 #endif // ASSERT
310 }
311 #endif // ifndef PRODUCT
312 
313 typedef void (C1_MacroAssembler::*c1_cond_branch_insn)(Register op1, Register op2, Label& label, bool is_far);
314 typedef void (C1_MacroAssembler::*c1_float_cond_branch_insn)(FloatRegister op1, FloatRegister op2,
315               Label& label, bool is_far, bool is_unordered);
316 
317 static c1_cond_branch_insn c1_cond_branch[] =
318 {
319   /* SHORT branches */
320   (c1_cond_branch_insn)&MacroAssembler::beq,
321   (c1_cond_branch_insn)&MacroAssembler::bne,
322   (c1_cond_branch_insn)&MacroAssembler::blt,
323   (c1_cond_branch_insn)&MacroAssembler::ble,
324   (c1_cond_branch_insn)&MacroAssembler::bge,
325   (c1_cond_branch_insn)&MacroAssembler::bgt,
326   (c1_cond_branch_insn)&MacroAssembler::bleu, // lir_cond_belowEqual
327   (c1_cond_branch_insn)&MacroAssembler::bgeu  // lir_cond_aboveEqual
328 };
329 
330 static c1_float_cond_branch_insn c1_float_cond_branch[] =
331 {
332   /* FLOAT branches */
333   (c1_float_cond_branch_insn)&MacroAssembler::float_beq,
334   (c1_float_cond_branch_insn)&MacroAssembler::float_bne,
335   (c1_float_cond_branch_insn)&MacroAssembler::float_blt,
336   (c1_float_cond_branch_insn)&MacroAssembler::float_ble,
337   (c1_float_cond_branch_insn)&MacroAssembler::float_bge,
338   (c1_float_cond_branch_insn)&MacroAssembler::float_bgt,
339   nullptr, // lir_cond_belowEqual
340   nullptr, // lir_cond_aboveEqual
341 
342   /* DOUBLE branches */
343   (c1_float_cond_branch_insn)&MacroAssembler::double_beq,
344   (c1_float_cond_branch_insn)&MacroAssembler::double_bne,
345   (c1_float_cond_branch_insn)&MacroAssembler::double_blt,
346   (c1_float_cond_branch_insn)&MacroAssembler::double_ble,
347   (c1_float_cond_branch_insn)&MacroAssembler::double_bge,
348   (c1_float_cond_branch_insn)&MacroAssembler::double_bgt,
349   nullptr, // lir_cond_belowEqual
350   nullptr  // lir_cond_aboveEqual
351 };
352 
353 void C1_MacroAssembler::c1_cmp_branch(int cmpFlag, Register op1, Register op2, Label& label,
354                                       BasicType type, bool is_far) {
355   if (type == T_OBJECT || type == T_ARRAY) {
356     assert(cmpFlag == lir_cond_equal || cmpFlag == lir_cond_notEqual, "Should be equal or notEqual");
357     if (cmpFlag == lir_cond_equal) {
358       beq(op1, op2, label, is_far);
359     } else {
360       bne(op1, op2, label, is_far);
361     }
362   } else {
363     assert(cmpFlag >= 0 && cmpFlag < (int)(sizeof(c1_cond_branch) / sizeof(c1_cond_branch[0])),
364            "invalid c1 conditional branch index");
365     (this->*c1_cond_branch[cmpFlag])(op1, op2, label, is_far);
366   }
367 }
368 
369 void C1_MacroAssembler::c1_float_cmp_branch(int cmpFlag, FloatRegister op1, FloatRegister op2, Label& label,
370                                             bool is_far, bool is_unordered) {
371   assert(cmpFlag >= 0 &&
372          cmpFlag < (int)(sizeof(c1_float_cond_branch) / sizeof(c1_float_cond_branch[0])),
373          "invalid c1 float conditional branch index");
374   (this->*c1_float_cond_branch[cmpFlag])(op1, op2, label, is_far, is_unordered);
375 }