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 #include "utilities/macros.hpp"
45
46 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
47 const int aligned_mask = BytesPerWord -1;
48 const int hdr_offset = oopDesc::mark_offset_in_bytes();
49 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
50 assert_different_registers(hdr, obj, disp_hdr, tmp);
51 int null_check_offset = -1;
52
53 verify_oop(obj);
54
55 // save object being locked into the BasicObjectLock
56 movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
57
58 null_check_offset = offset();
59
60 if (DiagnoseSyncOnValueBasedClasses != 0) {
61 load_klass(hdr, obj, rscratch1);
62 movl(hdr, Address(hdr, Klass::access_flags_offset()));
63 testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
64 jcc(Assembler::notZero, slow_case);
65 }
66
67 if (LockingMode == LM_LIGHTWEIGHT) {
68 #ifdef _LP64
69 const Register thread = r15_thread;
70 lightweight_lock(disp_hdr, obj, hdr, thread, tmp, slow_case);
71 #else
72 // Implicit null check.
73 movptr(hdr, Address(obj, oopDesc::mark_offset_in_bytes()));
74 // Lacking registers and thread on x86_32. Always take slow path.
75 jmp(slow_case);
76 #endif
77 } else if (LockingMode == LM_LEGACY) {
78 Label done;
79 // Load object header
80 movptr(hdr, Address(obj, hdr_offset));
81 // and mark it as unlocked
82 orptr(hdr, markWord::unlocked_value);
83 // save unlocked object header into the displaced header location on the stack
84 movptr(Address(disp_hdr, 0), hdr);
85 // test if object header is still the same (i.e. unlocked), and if so, store the
86 // displaced header address in the object header - if it is not the same, get the
87 // object header instead
88 MacroAssembler::lock(); // must be immediately before cmpxchg!
89 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
90 // if the object header was the same, we're done
91 jcc(Assembler::equal, done);
92 // if the object header was not the same, it is now in the hdr register
93 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
94 //
95 // 1) (hdr & aligned_mask) == 0
96 // 2) rsp <= hdr
97 // 3) hdr <= rsp + page_size
98 //
99 // these 3 tests can be done by evaluating the following expression:
100 //
101 // (hdr - rsp) & (aligned_mask - page_size)
102 //
103 // assuming both the stack pointer and page_size have their least
104 // significant 2 bits cleared and page_size is a power of 2
105 subptr(hdr, rsp);
106 andptr(hdr, aligned_mask - (int)os::vm_page_size());
107 // for recursive locking, the result is zero => save it in the displaced header
108 // location (null in the displaced hdr location indicates recursive locking)
109 movptr(Address(disp_hdr, 0), hdr);
110 // otherwise we don't care about the result and handle locking via runtime call
111 jcc(Assembler::notZero, slow_case);
112 // done
113 bind(done);
114 }
115
116 inc_held_monitor_count();
117
118 return null_check_offset;
119 }
120
121 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
122 const int aligned_mask = BytesPerWord -1;
123 const int hdr_offset = oopDesc::mark_offset_in_bytes();
124 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
125 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
126 Label done;
127
128 if (LockingMode != LM_LIGHTWEIGHT) {
129 // load displaced header
130 movptr(hdr, Address(disp_hdr, 0));
131 // if the loaded hdr is null we had recursive locking
132 testptr(hdr, hdr);
133 // if we had recursive locking, we are done
134 jcc(Assembler::zero, done);
135 }
136
137 // load object
138 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
139 verify_oop(obj);
140
141 if (LockingMode == LM_LIGHTWEIGHT) {
142 #ifdef _LP64
143 lightweight_unlock(obj, disp_hdr, r15_thread, hdr, slow_case);
144 #else
145 // This relies on the implementation of lightweight_unlock being able to handle
146 // that the reg_rax and thread Register parameters may alias each other.
147 get_thread(disp_hdr);
148 lightweight_unlock(obj, disp_hdr, disp_hdr, hdr, slow_case);
149 #endif
150 } else if (LockingMode == LM_LEGACY) {
151 // test if object header is pointing to the displaced header, and if so, restore
152 // the displaced header in the object - if the object header is not pointing to
153 // the displaced header, get the object header instead
154 MacroAssembler::lock(); // must be immediately before cmpxchg!
155 cmpxchgptr(hdr, Address(obj, hdr_offset));
156 // if the object header was not pointing to the displaced header,
157 // we do unlocking via runtime call
158 jcc(Assembler::notEqual, slow_case);
159 // done
160 }
161 bind(done);
162 dec_held_monitor_count();
163 }
164
165
166 // Defines obj, preserves var_size_in_bytes
167 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
168 if (UseTLAB) {
169 tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
170 } else {
171 jmp(slow_case);
172 }
173 }
174
175
176 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
177 assert_different_registers(obj, klass, len, t1, t2);
178 #ifdef _LP64
179 if (UseCompactObjectHeaders) {
180 movptr(t1, Address(klass, Klass::prototype_header_offset()));
181 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
182 } else if (UseCompressedClassPointers) { // Take care not to kill klass
183 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
184 movptr(t1, klass);
185 encode_klass_not_null(t1, rscratch1);
186 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
187 } else
188 #endif
189 {
190 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
191 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
192 }
193
194 if (len->is_valid()) {
195 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
196 #ifdef _LP64
197 int base_offset = arrayOopDesc::length_offset_in_bytes() + BytesPerInt;
198 if (!is_aligned(base_offset, BytesPerWord)) {
199 assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned");
200 // Clear gap/first 4 bytes following the length field.
201 xorl(t1, t1);
202 movl(Address(obj, base_offset), t1);
203 }
204 #endif
205 }
206 #ifdef _LP64
207 else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
208 xorptr(t1, t1);
209 store_klass_gap(obj, t1);
210 }
211 #endif
212 }
213
214
215 // preserves obj, destroys len_in_bytes
216 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
217 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
218 Label done;
219
220 // len_in_bytes is positive and ptr sized
221 subptr(len_in_bytes, hdr_size_in_bytes);
222 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
223 bind(done);
224 }
225
226
227 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
228 assert(obj == rax, "obj must be in rax, for cmpxchg");
229 assert_different_registers(obj, t1, t2); // XXX really?
230 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
231
232 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
233
234 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
235 }
236
237 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) {
238 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
239 "con_size_in_bytes is not multiple of alignment");
240 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
241 if (UseCompactObjectHeaders) {
242 assert(hdr_size_in_bytes == 8, "check object headers size");
243 }
244 initialize_header(obj, klass, noreg, t1, t2);
245
246 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
247 // clear rest of allocated space
248 const Register t1_zero = t1;
249 const Register index = t2;
250 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
251 if (var_size_in_bytes != noreg) {
252 mov(index, var_size_in_bytes);
253 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
254 } else if (con_size_in_bytes <= threshold) {
255 // use explicit null stores
256 // code size = 2 + 3*n bytes (n = number of fields to clear)
257 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
258 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
259 movptr(Address(obj, i), t1_zero);
260 } else if (con_size_in_bytes > hdr_size_in_bytes) {
261 // use loop to null out the fields
262 // code size = 16 bytes for even n (n = number of fields to clear)
263 // initialize last object field first if odd number of fields
264 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
265 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
266 // initialize last object field if constant size is odd
267 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
268 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
269 // initialize remaining object fields: rdx is a multiple of 2
270 { Label loop;
271 bind(loop);
272 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
273 t1_zero);
274 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
275 t1_zero);)
276 decrement(index);
277 jcc(Assembler::notZero, loop);
278 }
279 }
280 }
281
282 if (CURRENT_ENV->dtrace_alloc_probes()) {
283 assert(obj == rax, "must be");
284 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
285 }
286
287 verify_oop(obj);
288 }
289
290 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) {
291 assert(obj == rax, "obj must be in rax, for cmpxchg");
292 assert_different_registers(obj, len, t1, t2, klass);
293
294 // determine alignment mask
295 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
296
297 // check for negative or excessive length
298 cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
299 jcc(Assembler::above, slow_case);
300
301 const Register arr_size = t2; // okay to be the same
302 // align object end
303 movptr(arr_size, base_offset_in_bytes + MinObjAlignmentInBytesMask);
304 lea(arr_size, Address(arr_size, len, f));
305 andptr(arr_size, ~MinObjAlignmentInBytesMask);
306
307 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
308
309 initialize_header(obj, klass, len, t1, t2);
310
311 // clear rest of allocated space
312 const Register len_zero = len;
313 // Align-up to word boundary, because we clear the 4 bytes potentially
314 // following the length field in initialize_header().
315 int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
316 initialize_body(obj, arr_size, base_offset, len_zero);
317
318 if (CURRENT_ENV->dtrace_alloc_probes()) {
319 assert(obj == rax, "must be");
320 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
321 }
322
323 verify_oop(obj);
324 }
325
326 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
327 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
328 // Make sure there is enough stack space for this method's activation.
329 // Note that we do this before doing an enter(). This matches the
330 // ordering of C2's stack overflow check / rsp decrement and allows
331 // the SharedRuntime stack overflow handling to be consistent
332 // between the two compilers.
333 generate_stack_overflow_check(bang_size_in_bytes);
334
335 push(rbp);
336 if (PreserveFramePointer) {
337 mov(rbp, rsp);
338 }
339 #if !defined(_LP64) && defined(COMPILER2)
340 if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
341 // c2 leaves fpu stack dirty. Clean it on entry
342 empty_FPU_stack();
343 }
344 #endif // !_LP64 && COMPILER2
345 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
346
347 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
348 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
349 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
350 }
351
352
353 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
354 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
355 pop(rbp);
356 }
357
358
359 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
360 if (breakAtEntry || VerifyFPU) {
361 // Verified Entry first instruction should be 5 bytes long for correct
362 // patching by patch_verified_entry().
363 //
364 // Breakpoint and VerifyFPU have one byte first instruction.
365 // Also first instruction will be one byte "push(rbp)" if stack banging
366 // code is not generated (see build_frame() above).
367 // For all these cases generate long instruction first.
368 fat_nop();
369 }
370 if (breakAtEntry) int3();
371 // build frame
372 IA32_ONLY( verify_FPU(0, "method_entry"); )
373 }
374
375 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
376 // rbp, + 0: link
377 // + 1: return address
378 // + 2: argument with offset 0
379 // + 3: argument with offset 1
380 // + 4: ...
381
382 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
383 }
384
385 #ifndef PRODUCT
386
387 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
388 if (!VerifyOops) return;
389 verify_oop_addr(Address(rsp, stack_offset));
390 }
391
392 void C1_MacroAssembler::verify_not_null_oop(Register r) {
393 if (!VerifyOops) return;
394 Label not_null;
395 testptr(r, r);
396 jcc(Assembler::notZero, not_null);
397 stop("non-null oop required");
398 bind(not_null);
399 verify_oop(r);
400 }
401
402 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
403 #ifdef ASSERT
404 if (inv_rax) movptr(rax, 0xDEAD);
405 if (inv_rbx) movptr(rbx, 0xDEAD);
406 if (inv_rcx) movptr(rcx, 0xDEAD);
407 if (inv_rdx) movptr(rdx, 0xDEAD);
408 if (inv_rsi) movptr(rsi, 0xDEAD);
409 if (inv_rdi) movptr(rdi, 0xDEAD);
410 #endif
411 }
412
413 #endif // ifndef PRODUCT