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 (DiagnoseSyncOnValueBasedClasses != 0) {
59 load_klass(hdr, obj, rscratch1);
60 testb(Address(hdr, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
61 jcc(Assembler::notZero, slow_case);
62 }
63
64 if (LockingMode == LM_LIGHTWEIGHT) {
65 #ifdef _LP64
66 const Register thread = r15_thread;
67 lightweight_lock(disp_hdr, obj, hdr, thread, tmp, slow_case);
68 #else
69 // Implicit null check.
70 movptr(hdr, Address(obj, oopDesc::mark_offset_in_bytes()));
71 // Lacking registers and thread on x86_32. Always take slow path.
72 jmp(slow_case);
73 #endif
74 } else if (LockingMode == LM_LEGACY) {
75 Label done;
76 // Load object header
77 movptr(hdr, Address(obj, hdr_offset));
78 // and mark it as unlocked
79 orptr(hdr, markWord::unlocked_value);
80 // save unlocked object header into the displaced header location on the stack
81 movptr(Address(disp_hdr, 0), hdr);
82 // test if object header is still the same (i.e. unlocked), and if so, store the
83 // displaced header address in the object header - if it is not the same, get the
84 // object header instead
85 MacroAssembler::lock(); // must be immediately before cmpxchg!
86 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
87 // if the object header was the same, we're done
88 jcc(Assembler::equal, done);
89 // if the object header was not the same, it is now in the hdr register
90 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
91 //
92 // 1) (hdr & aligned_mask) == 0
93 // 2) rsp <= hdr
94 // 3) hdr <= rsp + page_size
95 //
96 // these 3 tests can be done by evaluating the following expression:
97 //
98 // (hdr - rsp) & (aligned_mask - page_size)
99 //
100 // assuming both the stack pointer and page_size have their least
101 // significant 2 bits cleared and page_size is a power of 2
102 subptr(hdr, rsp);
103 andptr(hdr, aligned_mask - (int)os::vm_page_size());
104 // for recursive locking, the result is zero => save it in the displaced header
105 // location (null in the displaced hdr location indicates recursive locking)
106 movptr(Address(disp_hdr, 0), hdr);
107 // otherwise we don't care about the result and handle locking via runtime call
108 jcc(Assembler::notZero, slow_case);
109 // done
110 bind(done);
111 inc_held_monitor_count();
112 }
113
114 return null_check_offset;
115 }
116
117 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
118 const int aligned_mask = BytesPerWord -1;
119 const int hdr_offset = oopDesc::mark_offset_in_bytes();
120 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
121 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
122 Label done;
123
124 if (LockingMode != LM_LIGHTWEIGHT) {
125 // load displaced header
126 movptr(hdr, Address(disp_hdr, 0));
127 // if the loaded hdr is null we had recursive locking
128 testptr(hdr, hdr);
129 // if we had recursive locking, we are done
130 jcc(Assembler::zero, done);
131 }
132
133 // load object
134 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
135 verify_oop(obj);
136
137 if (LockingMode == LM_LIGHTWEIGHT) {
138 #ifdef _LP64
139 lightweight_unlock(obj, disp_hdr, r15_thread, hdr, slow_case);
140 #else
141 // Lacking registers and thread on x86_32. Always take slow path.
142 jmp(slow_case);
143 #endif
144 } else if (LockingMode == LM_LEGACY) {
145 // test if object header is pointing to the displaced header, and if so, restore
146 // the displaced header in the object - if the object header is not pointing to
147 // the displaced header, get the object header instead
148 MacroAssembler::lock(); // must be immediately before cmpxchg!
149 cmpxchgptr(hdr, Address(obj, hdr_offset));
150 // if the object header was not pointing to the displaced header,
151 // we do unlocking via runtime call
152 jcc(Assembler::notEqual, slow_case);
153 // done
154 bind(done);
155 dec_held_monitor_count();
156 }
157 }
158
159
160 // Defines obj, preserves var_size_in_bytes
161 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
162 if (UseTLAB) {
163 tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
164 } else {
165 jmp(slow_case);
166 }
167 }
168
169
170 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
171 assert_different_registers(obj, klass, len, t1, t2);
172 #ifdef _LP64
173 if (UseCompactObjectHeaders) {
174 movptr(t1, Address(klass, Klass::prototype_header_offset()));
175 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
176 } else if (UseCompressedClassPointers) { // Take care not to kill klass
177 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
178 movptr(t1, klass);
179 encode_klass_not_null(t1, rscratch1);
180 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
181 } else
182 #endif
183 {
184 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
185 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
186 }
187
188 if (len->is_valid()) {
189 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
190 #ifdef _LP64
191 int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
192 if (!is_aligned(base_offset, BytesPerWord)) {
193 assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
194 // Clear gap/first 4 bytes following the length field.
195 xorl(t1, t1);
196 movl(Address(obj, base_offset), t1);
197 }
198 #endif
199 }
200 #ifdef _LP64
201 else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
202 xorptr(t1, t1);
203 store_klass_gap(obj, t1);
204 }
205 #endif
206 }
207
208
209 // preserves obj, destroys len_in_bytes
210 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
211 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
212 Label done;
213
214 // len_in_bytes is positive and ptr sized
215 subptr(len_in_bytes, hdr_size_in_bytes);
216 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
217 bind(done);
218 }
219
220
221 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
222 assert(obj == rax, "obj must be in rax, for cmpxchg");
223 assert_different_registers(obj, t1, t2); // XXX really?
224 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
225
226 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
227
228 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
229 }
230
231 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) {
232 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
233 "con_size_in_bytes is not multiple of alignment");
234 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
235 if (UseCompactObjectHeaders) {
236 assert(hdr_size_in_bytes == 8, "check object headers size");
237 }
238 initialize_header(obj, klass, noreg, t1, t2);
239
240 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
241 // clear rest of allocated space
242 const Register t1_zero = t1;
243 const Register index = t2;
244 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
245 if (var_size_in_bytes != noreg) {
246 mov(index, var_size_in_bytes);
247 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
248 } else if (con_size_in_bytes <= threshold) {
249 // use explicit null stores
250 // code size = 2 + 3*n bytes (n = number of fields to clear)
251 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
252 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
253 movptr(Address(obj, i), t1_zero);
254 } else if (con_size_in_bytes > hdr_size_in_bytes) {
255 // use loop to null out the fields
256 // code size = 16 bytes for even n (n = number of fields to clear)
257 // initialize last object field first if odd number of fields
258 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
259 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
260 // initialize last object field if constant size is odd
261 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
262 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
263 // initialize remaining object fields: rdx is a multiple of 2
264 { Label loop;
265 bind(loop);
266 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
267 t1_zero);
268 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
269 t1_zero);)
270 decrement(index);
271 jcc(Assembler::notZero, loop);
272 }
273 }
274 }
275
276 if (CURRENT_ENV->dtrace_alloc_probes()) {
277 assert(obj == rax, "must be");
278 call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
279 }
280
281 verify_oop(obj);
282 }
283
284 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) {
285 assert(obj == rax, "obj must be in rax, for cmpxchg");
286 assert_different_registers(obj, len, t1, t2, klass);
287
288 // determine alignment mask
289 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
290
291 // check for negative or excessive length
292 cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
293 jcc(Assembler::above, slow_case);
294
295 const Register arr_size = t2; // okay to be the same
296 // align object end
297 movptr(arr_size, base_offset_in_bytes + MinObjAlignmentInBytesMask);
298 lea(arr_size, Address(arr_size, len, f));
299 andptr(arr_size, ~MinObjAlignmentInBytesMask);
300
301 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
302
303 initialize_header(obj, klass, len, t1, t2);
304
305 // clear rest of allocated space
306 if (zero_array) {
307 const Register len_zero = len;
308 // Align-up to word boundary, because we clear the 4 bytes potentially
309 // following the length field in initialize_header().
310 int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
311 initialize_body(obj, arr_size, base_offset, len_zero);
312 }
313
314 if (CURRENT_ENV->dtrace_alloc_probes()) {
315 assert(obj == rax, "must be");
316 call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
317 }
318
319 verify_oop(obj);
320 }
321
322 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
323 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
324 // Make sure there is enough stack space for this method's activation.
325 // Note that we do this before doing an enter(). This matches the
326 // ordering of C2's stack overflow check / rsp decrement and allows
327 // the SharedRuntime stack overflow handling to be consistent
328 // between the two compilers.
329 generate_stack_overflow_check(bang_size_in_bytes);
330
331 push(rbp);
332 if (PreserveFramePointer) {
333 mov(rbp, rsp);
334 }
335 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
336
337 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
338 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
339 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
340 }
341
342
343 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
344 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
345 pop(rbp);
346 }
347
348
349 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
350 if (breakAtEntry || VerifyFPU) {
351 // Verified Entry first instruction should be 5 bytes long for correct
352 // patching by patch_verified_entry().
353 //
354 // Breakpoint and VerifyFPU have one byte first instruction.
355 // Also first instruction will be one byte "push(rbp)" if stack banging
356 // code is not generated (see build_frame() above).
357 // For all these cases generate long instruction first.
358 fat_nop();
359 }
360 if (breakAtEntry) int3();
361 // build frame
362 IA32_ONLY( verify_FPU(0, "method_entry"); )
363 }
364
365 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
366 // rbp, + 0: link
367 // + 1: return address
368 // + 2: argument with offset 0
369 // + 3: argument with offset 1
370 // + 4: ...
371
372 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
373 }
374
375 #ifndef PRODUCT
376
377 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
378 if (!VerifyOops) return;
379 verify_oop_addr(Address(rsp, stack_offset));
380 }
381
382 void C1_MacroAssembler::verify_not_null_oop(Register r) {
383 if (!VerifyOops) return;
384 Label not_null;
385 testptr(r, r);
386 jcc(Assembler::notZero, not_null);
387 stop("non-null oop required");
388 bind(not_null);
389 verify_oop(r);
390 }
391
392 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
393 #ifdef ASSERT
394 if (inv_rax) movptr(rax, 0xDEAD);
395 if (inv_rbx) movptr(rbx, 0xDEAD);
396 if (inv_rcx) movptr(rcx, 0xDEAD);
397 if (inv_rdx) movptr(rdx, 0xDEAD);
398 if (inv_rsi) movptr(rsi, 0xDEAD);
399 if (inv_rdi) movptr(rdi, 0xDEAD);
400 #endif
401 }
402
403 #endif // ifndef PRODUCT