1 /*
2 * Copyright (c) 1999, 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 "c1/c1_MacroAssembler.hpp"
26 #include "c1/c1_Runtime1.hpp"
27 #include "code/compiledIC.hpp"
28 #include "compiler/compilerDefinitions.inline.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "gc/shared/collectedHeap.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/globals.hpp"
38 #include "runtime/os.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/stubRoutines.hpp"
41 #include "utilities/checkedCast.hpp"
42 #include "utilities/globalDefinitions.hpp"
43
44 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
45 const int aligned_mask = BytesPerWord -1;
46 const int hdr_offset = oopDesc::mark_offset_in_bytes();
47 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
48 assert_different_registers(hdr, obj, disp_hdr, tmp);
49 int null_check_offset = -1;
50
51 verify_oop(obj);
52
53 // save object being locked into the BasicObjectLock
54 movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
55
56 null_check_offset = offset();
57
58 if (LockingMode == LM_LIGHTWEIGHT) {
59 lightweight_lock(disp_hdr, obj, hdr, tmp, slow_case);
60 } else if (LockingMode == LM_LEGACY) {
61 Label done;
62
63 if (DiagnoseSyncOnValueBasedClasses != 0) {
64 load_klass(hdr, obj, rscratch1);
65 testb(Address(hdr, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
66 jcc(Assembler::notZero, slow_case);
67 }
68
69 // Load object header
70 movptr(hdr, Address(obj, hdr_offset));
71 // and mark it as unlocked
72 orptr(hdr, markWord::unlocked_value);
73 // save unlocked object header into the displaced header location on the stack
74 movptr(Address(disp_hdr, 0), hdr);
75 // test if object header is still the same (i.e. unlocked), and if so, store the
76 // displaced header address in the object header - if it is not the same, get the
77 // object header instead
78 MacroAssembler::lock(); // must be immediately before cmpxchg!
79 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
80 // if the object header was the same, we're done
81 jcc(Assembler::equal, done);
82 // if the object header was not the same, it is now in the hdr register
83 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
84 //
85 // 1) (hdr & aligned_mask) == 0
86 // 2) rsp <= hdr
87 // 3) hdr <= rsp + page_size
88 //
89 // these 3 tests can be done by evaluating the following expression:
90 //
91 // (hdr - rsp) & (aligned_mask - page_size)
92 //
93 // assuming both the stack pointer and page_size have their least
94 // significant 2 bits cleared and page_size is a power of 2
95 subptr(hdr, rsp);
96 andptr(hdr, aligned_mask - (int)os::vm_page_size());
97 // for recursive locking, the result is zero => save it in the displaced header
98 // location (null in the displaced hdr location indicates recursive locking)
99 movptr(Address(disp_hdr, 0), hdr);
100 // otherwise we don't care about the result and handle locking via runtime call
101 jcc(Assembler::notZero, slow_case);
102 // done
103 bind(done);
104 inc_held_monitor_count();
105 }
106
107 return null_check_offset;
108 }
109
110 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
111 const int aligned_mask = BytesPerWord -1;
112 const int hdr_offset = oopDesc::mark_offset_in_bytes();
113 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
114 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
115 Label done;
116
117 if (LockingMode != LM_LIGHTWEIGHT) {
118 // load displaced header
119 movptr(hdr, Address(disp_hdr, 0));
120 // if the loaded hdr is null we had recursive locking
121 testptr(hdr, hdr);
122 // if we had recursive locking, we are done
123 jcc(Assembler::zero, done);
124 }
125
126 // load object
127 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
128 verify_oop(obj);
129
130 if (LockingMode == LM_LIGHTWEIGHT) {
131 lightweight_unlock(obj, disp_hdr, hdr, slow_case);
132 } else if (LockingMode == LM_LEGACY) {
133 // test if object header is pointing to the displaced header, and if so, restore
134 // the displaced header in the object - if the object header is not pointing to
135 // the displaced header, get the object header instead
136 MacroAssembler::lock(); // must be immediately before cmpxchg!
137 cmpxchgptr(hdr, Address(obj, hdr_offset));
138 // if the object header was not pointing to the displaced header,
139 // we do unlocking via runtime call
140 jcc(Assembler::notEqual, slow_case);
141 // done
142 bind(done);
143 dec_held_monitor_count();
144 }
145 }
146
147
148 // Defines obj, preserves var_size_in_bytes
149 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
150 if (UseTLAB) {
151 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
152 } else {
153 jmp(slow_case);
154 }
155 }
156
157
158 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
159 assert_different_registers(obj, klass, len, t1, t2);
160 if (UseCompactObjectHeaders) {
161 movptr(t1, Address(klass, Klass::prototype_header_offset()));
162 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
163 } else if (UseCompressedClassPointers) { // Take care not to kill klass
164 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
165 movptr(t1, klass);
166 encode_klass_not_null(t1, rscratch1);
167 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
168 } else {
169 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
170 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
171 }
172
173 if (len->is_valid()) {
174 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
175 int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
176 if (!is_aligned(base_offset, BytesPerWord)) {
177 assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
178 // Clear gap/first 4 bytes following the length field.
179 xorl(t1, t1);
180 movl(Address(obj, base_offset), t1);
181 }
182 } else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
183 xorptr(t1, t1);
184 store_klass_gap(obj, t1);
185 }
186 }
187
188
189 // preserves obj, destroys len_in_bytes
190 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
191 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
192 Label done;
193
194 // len_in_bytes is positive and ptr sized
195 subptr(len_in_bytes, hdr_size_in_bytes);
196 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
197 bind(done);
198 }
199
200
201 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
202 assert(obj == rax, "obj must be in rax, for cmpxchg");
203 assert_different_registers(obj, t1, t2); // XXX really?
204 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
205
206 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
207
208 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
209 }
210
211 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) {
212 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
213 "con_size_in_bytes is not multiple of alignment");
214 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
215 if (UseCompactObjectHeaders) {
216 assert(hdr_size_in_bytes == 8, "check object headers size");
217 }
218 initialize_header(obj, klass, noreg, t1, t2);
219
220 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
221 // clear rest of allocated space
222 const Register t1_zero = t1;
223 const Register index = t2;
224 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
225 if (var_size_in_bytes != noreg) {
226 mov(index, var_size_in_bytes);
227 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
228 } else if (con_size_in_bytes <= threshold) {
229 // use explicit null stores
230 // code size = 2 + 3*n bytes (n = number of fields to clear)
231 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
232 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
233 movptr(Address(obj, i), t1_zero);
234 } else if (con_size_in_bytes > hdr_size_in_bytes) {
235 // use loop to null out the fields
236 // code size = 16 bytes for even n (n = number of fields to clear)
237 // initialize last object field first if odd number of fields
238 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
239 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
240 // initialize last object field if constant size is odd
241 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
242 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
243 // initialize remaining object fields: rdx is a multiple of 2
244 { Label loop;
245 bind(loop);
246 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
247 t1_zero);
248 decrement(index);
249 jcc(Assembler::notZero, loop);
250 }
251 }
252 }
253
254 if (CURRENT_ENV->dtrace_alloc_probes()) {
255 assert(obj == rax, "must be");
256 call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
257 }
258
259 verify_oop(obj);
260 }
261
262 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int base_offset_in_bytes, Address::ScaleFactor f, Register klass, Label& slow_case, bool zero_array) {
263 assert(obj == rax, "obj must be in rax, for cmpxchg");
264 assert_different_registers(obj, len, t1, t2, klass);
265
266 // determine alignment mask
267 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
268
269 // check for negative or excessive length
270 cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
271 jcc(Assembler::above, slow_case);
272
273 const Register arr_size = t2; // okay to be the same
274 // align object end
275 movptr(arr_size, base_offset_in_bytes + MinObjAlignmentInBytesMask);
276 lea(arr_size, Address(arr_size, len, f));
277 andptr(arr_size, ~MinObjAlignmentInBytesMask);
278
279 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
280
281 initialize_header(obj, klass, len, t1, t2);
282
283 // clear rest of allocated space
284 if (zero_array) {
285 const Register len_zero = len;
286 // Align-up to word boundary, because we clear the 4 bytes potentially
287 // following the length field in initialize_header().
288 int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
289 initialize_body(obj, arr_size, base_offset, len_zero);
290 }
291
292 if (CURRENT_ENV->dtrace_alloc_probes()) {
293 assert(obj == rax, "must be");
294 call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
295 }
296
297 verify_oop(obj);
298 }
299
300 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
301 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
302 // Make sure there is enough stack space for this method's activation.
303 // Note that we do this before doing an enter(). This matches the
304 // ordering of C2's stack overflow check / rsp decrement and allows
305 // the SharedRuntime stack overflow handling to be consistent
306 // between the two compilers.
307 generate_stack_overflow_check(bang_size_in_bytes);
308
309 push(rbp);
310 if (PreserveFramePointer) {
311 mov(rbp, rsp);
312 }
313 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
314
315 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
316 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
317 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
318 }
319
320
321 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
322 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
323 pop(rbp);
324 }
325
326
327 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
328 if (breakAtEntry) {
329 // Verified Entry first instruction should be 5 bytes long for correct
330 // patching by patch_verified_entry().
331 //
332 // Breakpoint has one byte first instruction.
333 // Also first instruction will be one byte "push(rbp)" if stack banging
334 // code is not generated (see build_frame() above).
335 // For all these cases generate long instruction first.
336 fat_nop();
337 }
338 if (breakAtEntry) int3();
339 // build frame
340 }
341
342 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
343 // rbp, + 0: link
344 // + 1: return address
345 // + 2: argument with offset 0
346 // + 3: argument with offset 1
347 // + 4: ...
348
349 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
350 }
351
352 #ifndef PRODUCT
353
354 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
355 if (!VerifyOops) return;
356 verify_oop_addr(Address(rsp, stack_offset));
357 }
358
359 void C1_MacroAssembler::verify_not_null_oop(Register r) {
360 if (!VerifyOops) return;
361 Label not_null;
362 testptr(r, r);
363 jcc(Assembler::notZero, not_null);
364 stop("non-null oop required");
365 bind(not_null);
366 verify_oop(r);
367 }
368
369 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
370 #ifdef ASSERT
371 if (inv_rax) movptr(rax, 0xDEAD);
372 if (inv_rbx) movptr(rbx, 0xDEAD);
373 if (inv_rcx) movptr(rcx, 0xDEAD);
374 if (inv_rdx) movptr(rdx, 0xDEAD);
375 if (inv_rsi) movptr(rsi, 0xDEAD);
376 if (inv_rdi) movptr(rdi, 0xDEAD);
377 #endif
378 }
379
380 #endif // ifndef PRODUCT