1 /*
2 * Copyright (c) 1999, 2024, 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 "precompiled.hpp"
26 #include "c1/c1_MacroAssembler.hpp"
27 #include "c1/c1_Runtime1.hpp"
28 #include "code/compiledIC.hpp"
29 #include "compiler/compilerDefinitions.inline.hpp"
30 #include "gc/shared/barrierSet.hpp"
31 #include "gc/shared/barrierSetAssembler.hpp"
32 #include "gc/shared/collectedHeap.hpp"
33 #include "gc/shared/tlab_globals.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/markWord.hpp"
37 #include "runtime/basicLock.hpp"
38 #include "runtime/globals.hpp"
39 #include "runtime/os.hpp"
40 #include "runtime/sharedRuntime.hpp"
41 #include "runtime/stubRoutines.hpp"
42 #include "utilities/checkedCast.hpp"
43 #include "utilities/globalDefinitions.hpp"
44
45 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
46 const int aligned_mask = BytesPerWord -1;
47 const int hdr_offset = oopDesc::mark_offset_in_bytes();
48 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
49 assert_different_registers(hdr, obj, disp_hdr, tmp);
50 int null_check_offset = -1;
51
52 verify_oop(obj);
53
54 // save object being locked into the BasicObjectLock
55 movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
56
57 null_check_offset = offset();
58
59 if (DiagnoseSyncOnValueBasedClasses != 0) {
60 load_klass(hdr, obj, rscratch1);
61 testb(Address(hdr, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
62 jcc(Assembler::notZero, slow_case);
63 }
64
65 if (LockingMode == LM_LIGHTWEIGHT) {
66 #ifdef _LP64
67 const Register thread = r15_thread;
68 lightweight_lock(disp_hdr, obj, hdr, thread, tmp, slow_case);
69 #else
70 // Implicit null check.
71 movptr(hdr, Address(obj, oopDesc::mark_offset_in_bytes()));
72 // Lacking registers and thread on x86_32. Always take slow path.
73 jmp(slow_case);
74 #endif
75 } else if (LockingMode == LM_LEGACY) {
76 Label done;
77 // Load object header
78 movptr(hdr, Address(obj, hdr_offset));
79 // and mark it as unlocked
80 orptr(hdr, markWord::unlocked_value);
81 // save unlocked object header into the displaced header location on the stack
82 movptr(Address(disp_hdr, 0), hdr);
83 // test if object header is still the same (i.e. unlocked), and if so, store the
84 // displaced header address in the object header - if it is not the same, get the
85 // object header instead
86 MacroAssembler::lock(); // must be immediately before cmpxchg!
87 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
88 // if the object header was the same, we're done
89 jcc(Assembler::equal, done);
90 // if the object header was not the same, it is now in the hdr register
91 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
92 //
93 // 1) (hdr & aligned_mask) == 0
94 // 2) rsp <= hdr
95 // 3) hdr <= rsp + page_size
96 //
97 // these 3 tests can be done by evaluating the following expression:
98 //
99 // (hdr - rsp) & (aligned_mask - page_size)
100 //
101 // assuming both the stack pointer and page_size have their least
102 // significant 2 bits cleared and page_size is a power of 2
103 subptr(hdr, rsp);
104 andptr(hdr, aligned_mask - (int)os::vm_page_size());
105 // for recursive locking, the result is zero => save it in the displaced header
106 // location (null in the displaced hdr location indicates recursive locking)
107 movptr(Address(disp_hdr, 0), hdr);
108 // otherwise we don't care about the result and handle locking via runtime call
109 jcc(Assembler::notZero, slow_case);
110 // done
111 bind(done);
112 }
113
114 inc_held_monitor_count();
115
116 return null_check_offset;
117 }
118
119 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
120 const int aligned_mask = BytesPerWord -1;
121 const int hdr_offset = oopDesc::mark_offset_in_bytes();
122 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
123 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
124 Label done;
125
126 if (LockingMode != LM_LIGHTWEIGHT) {
127 // load displaced header
128 movptr(hdr, Address(disp_hdr, 0));
129 // if the loaded hdr is null we had recursive locking
130 testptr(hdr, hdr);
131 // if we had recursive locking, we are done
132 jcc(Assembler::zero, done);
133 }
134
135 // load object
136 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
137 verify_oop(obj);
138
139 if (LockingMode == LM_LIGHTWEIGHT) {
140 #ifdef _LP64
141 lightweight_unlock(obj, disp_hdr, r15_thread, hdr, slow_case);
142 #else
143 // Lacking registers and thread on x86_32. Always take slow path.
144 jmp(slow_case);
145 #endif
146 } else if (LockingMode == LM_LEGACY) {
147 // test if object header is pointing to the displaced header, and if so, restore
148 // the displaced header in the object - if the object header is not pointing to
149 // the displaced header, get the object header instead
150 MacroAssembler::lock(); // must be immediately before cmpxchg!
151 cmpxchgptr(hdr, Address(obj, hdr_offset));
152 // if the object header was not pointing to the displaced header,
153 // we do unlocking via runtime call
154 jcc(Assembler::notEqual, slow_case);
155 // done
156 }
157 bind(done);
158 dec_held_monitor_count();
159 }
160
161
162 // Defines obj, preserves var_size_in_bytes
163 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
164 if (UseTLAB) {
165 tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
166 } else {
167 jmp(slow_case);
168 }
169 }
170
171
172 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
173 assert_different_registers(obj, klass, len);
174 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
175 #ifdef _LP64
176 if (UseCompressedClassPointers) { // Take care not to kill klass
177 movptr(t1, klass);
178 encode_klass_not_null(t1, rscratch1);
179 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
180 } else
181 #endif
182 {
183 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
184 }
185
186 if (len->is_valid()) {
187 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
188 #ifdef _LP64
189 int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
190 if (!is_aligned(base_offset, BytesPerWord)) {
191 assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
192 // Clear gap/first 4 bytes following the length field.
193 xorl(t1, t1);
194 movl(Address(obj, base_offset), t1);
195 }
196 #endif
197 }
198 #ifdef _LP64
199 else if (UseCompressedClassPointers) {
200 xorptr(t1, t1);
201 store_klass_gap(obj, t1);
202 }
203 #endif
204 }
205
206
207 // preserves obj, destroys len_in_bytes
208 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
209 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
210 Label done;
211
212 // len_in_bytes is positive and ptr sized
213 subptr(len_in_bytes, hdr_size_in_bytes);
214 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
215 bind(done);
216 }
217
218
219 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
220 assert(obj == rax, "obj must be in rax, for cmpxchg");
221 assert_different_registers(obj, t1, t2); // XXX really?
222 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
223
224 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
225
226 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
227 }
228
229 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) {
230 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
231 "con_size_in_bytes is not multiple of alignment");
232 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
233
234 initialize_header(obj, klass, noreg, t1, t2);
235
236 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
237 // clear rest of allocated space
238 const Register t1_zero = t1;
239 const Register index = t2;
240 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
241 if (var_size_in_bytes != noreg) {
242 mov(index, var_size_in_bytes);
243 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
244 } else if (con_size_in_bytes <= threshold) {
245 // use explicit null stores
246 // code size = 2 + 3*n bytes (n = number of fields to clear)
247 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
248 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
249 movptr(Address(obj, i), t1_zero);
250 } else if (con_size_in_bytes > hdr_size_in_bytes) {
251 // use loop to null out the fields
252 // code size = 16 bytes for even n (n = number of fields to clear)
253 // initialize last object field first if odd number of fields
254 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
255 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
256 // initialize last object field if constant size is odd
257 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
258 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
259 // initialize remaining object fields: rdx is a multiple of 2
260 { Label loop;
261 bind(loop);
262 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
263 t1_zero);
264 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
265 t1_zero);)
266 decrement(index);
267 jcc(Assembler::notZero, loop);
268 }
269 }
270 }
271
272 if (CURRENT_ENV->dtrace_alloc_probes()) {
273 assert(obj == rax, "must be");
274 call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
275 }
276
277 verify_oop(obj);
278 }
279
280 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) {
281 assert(obj == rax, "obj must be in rax, for cmpxchg");
282 assert_different_registers(obj, len, t1, t2, klass);
283
284 // determine alignment mask
285 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
286
287 // check for negative or excessive length
288 cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
289 jcc(Assembler::above, slow_case);
290
291 const Register arr_size = t2; // okay to be the same
292 // align object end
293 movptr(arr_size, base_offset_in_bytes + MinObjAlignmentInBytesMask);
294 lea(arr_size, Address(arr_size, len, f));
295 andptr(arr_size, ~MinObjAlignmentInBytesMask);
296
297 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
298
299 initialize_header(obj, klass, len, t1, t2);
300
301 // clear rest of allocated space
302 if (zero_array) {
303 const Register len_zero = len;
304 // Align-up to word boundary, because we clear the 4 bytes potentially
305 // following the length field in initialize_header().
306 int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
307 initialize_body(obj, arr_size, base_offset, len_zero);
308 }
309
310 if (CURRENT_ENV->dtrace_alloc_probes()) {
311 assert(obj == rax, "must be");
312 call(RuntimeAddress(Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)));
313 }
314
315 verify_oop(obj);
316 }
317
318 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
319 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
320 // Make sure there is enough stack space for this method's activation.
321 // Note that we do this before doing an enter(). This matches the
322 // ordering of C2's stack overflow check / rsp decrement and allows
323 // the SharedRuntime stack overflow handling to be consistent
324 // between the two compilers.
325 generate_stack_overflow_check(bang_size_in_bytes);
326
327 push(rbp);
328 if (PreserveFramePointer) {
329 mov(rbp, rsp);
330 }
331 #if !defined(_LP64) && defined(COMPILER2)
332 if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
333 // c2 leaves fpu stack dirty. Clean it on entry
334 empty_FPU_stack();
335 }
336 #endif // !_LP64 && COMPILER2
337 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
338
339 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
340 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
341 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
342 }
343
344
345 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
346 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
347 pop(rbp);
348 }
349
350
351 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
352 if (breakAtEntry || VerifyFPU) {
353 // Verified Entry first instruction should be 5 bytes long for correct
354 // patching by patch_verified_entry().
355 //
356 // Breakpoint and VerifyFPU have one byte first instruction.
357 // Also first instruction will be one byte "push(rbp)" if stack banging
358 // code is not generated (see build_frame() above).
359 // For all these cases generate long instruction first.
360 fat_nop();
361 }
362 if (breakAtEntry) int3();
363 // build frame
364 IA32_ONLY( verify_FPU(0, "method_entry"); )
365 }
366
367 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
368 // rbp, + 0: link
369 // + 1: return address
370 // + 2: argument with offset 0
371 // + 3: argument with offset 1
372 // + 4: ...
373
374 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
375 }
376
377 #ifndef PRODUCT
378
379 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
380 if (!VerifyOops) return;
381 verify_oop_addr(Address(rsp, stack_offset));
382 }
383
384 void C1_MacroAssembler::verify_not_null_oop(Register r) {
385 if (!VerifyOops) return;
386 Label not_null;
387 testptr(r, r);
388 jcc(Assembler::notZero, not_null);
389 stop("non-null oop required");
390 bind(not_null);
391 verify_oop(r);
392 }
393
394 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
395 #ifdef ASSERT
396 if (inv_rax) movptr(rax, 0xDEAD);
397 if (inv_rbx) movptr(rbx, 0xDEAD);
398 if (inv_rcx) movptr(rcx, 0xDEAD);
399 if (inv_rdx) movptr(rdx, 0xDEAD);
400 if (inv_rsi) movptr(rsi, 0xDEAD);
401 if (inv_rdi) movptr(rdi, 0xDEAD);
402 #endif
403 }
404
405 #endif // ifndef PRODUCT