1 /*
2 * Copyright (c) 1999, 2023, 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 "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/os.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/stubRoutines.hpp"
40 #include "utilities/checkedCast.hpp"
41
42 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
43 const int aligned_mask = BytesPerWord -1;
44 const int hdr_offset = oopDesc::mark_offset_in_bytes();
45 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
46 assert_different_registers(hdr, obj, disp_hdr, tmp);
47 int null_check_offset = -1;
48
49 verify_oop(obj);
50
51 // save object being locked into the BasicObjectLock
52 movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
53
54 null_check_offset = offset();
55
56 if (DiagnoseSyncOnValueBasedClasses != 0) {
57 load_klass(hdr, obj, rscratch1);
58 movl(hdr, Address(hdr, Klass::access_flags_offset()));
59 testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
60 jcc(Assembler::notZero, slow_case);
61 }
62
63 // Load object header
64 movptr(hdr, Address(obj, hdr_offset));
65
66 if (LockingMode == LM_LIGHTWEIGHT) {
67 #ifdef _LP64
68 const Register thread = r15_thread;
69 #else
70 const Register thread = disp_hdr;
71 get_thread(thread);
72 #endif
73 lightweight_lock(obj, hdr, thread, tmp, slow_case);
74 } else if (LockingMode == LM_LEGACY) {
75 Label done;
76 // and mark it as unlocked
77 orptr(hdr, markWord::unlocked_value);
78 // save unlocked object header into the displaced header location on the stack
79 movptr(Address(disp_hdr, 0), hdr);
80 // test if object header is still the same (i.e. unlocked), and if so, store the
81 // displaced header address in the object header - if it is not the same, get the
82 // object header instead
83 MacroAssembler::lock(); // must be immediately before cmpxchg!
84 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
85 // if the object header was the same, we're done
86 jcc(Assembler::equal, done);
87 // if the object header was not the same, it is now in the hdr register
88 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
89 //
90 // 1) (hdr & aligned_mask) == 0
91 // 2) rsp <= hdr
92 // 3) hdr <= rsp + page_size
93 //
94 // these 3 tests can be done by evaluating the following expression:
95 //
96 // (hdr - rsp) & (aligned_mask - page_size)
97 //
98 // assuming both the stack pointer and page_size have their least
99 // significant 2 bits cleared and page_size is a power of 2
100 subptr(hdr, rsp);
101 andptr(hdr, aligned_mask - (int)os::vm_page_size());
102 // for recursive locking, the result is zero => save it in the displaced header
103 // location (null in the displaced hdr location indicates recursive locking)
104 movptr(Address(disp_hdr, 0), hdr);
105 // otherwise we don't care about the result and handle locking via runtime call
106 jcc(Assembler::notZero, slow_case);
107 // done
108 bind(done);
109 }
110
111 inc_held_monitor_count();
112
113 return null_check_offset;
114 }
115
116 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
117 const int aligned_mask = BytesPerWord -1;
118 const int hdr_offset = oopDesc::mark_offset_in_bytes();
119 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
120 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
121 Label done;
122
123 if (LockingMode != LM_LIGHTWEIGHT) {
124 // load displaced header
125 movptr(hdr, Address(disp_hdr, 0));
126 // if the loaded hdr is null we had recursive locking
127 testptr(hdr, hdr);
128 // if we had recursive locking, we are done
129 jcc(Assembler::zero, done);
130 }
131
132 // load object
133 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
134 verify_oop(obj);
135
136 if (LockingMode == LM_LIGHTWEIGHT) {
137 movptr(disp_hdr, Address(obj, hdr_offset));
138 andptr(disp_hdr, ~(int32_t)markWord::lock_mask_in_place);
139 lightweight_unlock(obj, disp_hdr, hdr, slow_case);
140 } else if (LockingMode == LM_LEGACY) {
141 // test if object header is pointing to the displaced header, and if so, restore
142 // the displaced header in the object - if the object header is not pointing to
143 // the displaced header, get the object header instead
144 MacroAssembler::lock(); // must be immediately before cmpxchg!
145 cmpxchgptr(hdr, Address(obj, hdr_offset));
146 // if the object header was not pointing to the displaced header,
147 // we do unlocking via runtime call
148 jcc(Assembler::notEqual, slow_case);
149 // done
150 }
151 bind(done);
152 dec_held_monitor_count();
153 }
154
155
156 // Defines obj, preserves var_size_in_bytes
157 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
158 if (UseTLAB) {
159 tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
160 } else {
161 jmp(slow_case);
162 }
163 }
164
165
166 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
167 assert_different_registers(obj, klass, len);
168 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
169 #ifdef _LP64
170 if (UseCompressedClassPointers) { // Take care not to kill klass
171 movptr(t1, klass);
172 encode_klass_not_null(t1, rscratch1);
173 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
174 } else
175 #endif
176 {
177 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
178 }
179
180 if (len->is_valid()) {
181 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
182 }
183 #ifdef _LP64
184 else if (UseCompressedClassPointers) {
185 xorptr(t1, t1);
186 store_klass_gap(obj, t1);
187 }
188 #endif
189 }
190
191
192 // preserves obj, destroys len_in_bytes
193 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
194 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
195 Label done;
196
197 // len_in_bytes is positive and ptr sized
198 subptr(len_in_bytes, hdr_size_in_bytes);
199 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
200 bind(done);
201 }
202
203
204 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
205 assert(obj == rax, "obj must be in rax, for cmpxchg");
206 assert_different_registers(obj, t1, t2); // XXX really?
207 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
208
209 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
210
211 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
212 }
213
214 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) {
215 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
216 "con_size_in_bytes is not multiple of alignment");
217 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
218
219 initialize_header(obj, klass, noreg, t1, t2);
220
221 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
222 // clear rest of allocated space
223 const Register t1_zero = t1;
224 const Register index = t2;
225 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
226 if (var_size_in_bytes != noreg) {
227 mov(index, var_size_in_bytes);
228 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
229 } else if (con_size_in_bytes <= threshold) {
230 // use explicit null stores
231 // code size = 2 + 3*n bytes (n = number of fields to clear)
232 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
233 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
234 movptr(Address(obj, i), t1_zero);
235 } else if (con_size_in_bytes > hdr_size_in_bytes) {
236 // use loop to null out the fields
237 // code size = 16 bytes for even n (n = number of fields to clear)
238 // initialize last object field first if odd number of fields
239 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
240 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
241 // initialize last object field if constant size is odd
242 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
243 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
244 // initialize remaining object fields: rdx is a multiple of 2
245 { Label loop;
246 bind(loop);
247 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
248 t1_zero);
249 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
250 t1_zero);)
251 decrement(index);
252 jcc(Assembler::notZero, loop);
253 }
254 }
255 }
256
257 if (CURRENT_ENV->dtrace_alloc_probes()) {
258 assert(obj == rax, "must be");
259 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
260 }
261
262 verify_oop(obj);
263 }
264
265 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
266 assert(obj == rax, "obj must be in rax, for cmpxchg");
267 assert_different_registers(obj, len, t1, t2, klass);
268
269 // determine alignment mask
270 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
271
272 // check for negative or excessive length
273 cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
274 jcc(Assembler::above, slow_case);
275
276 const Register arr_size = t2; // okay to be the same
277 // align object end
278 movptr(arr_size, header_size * BytesPerWord + MinObjAlignmentInBytesMask);
279 lea(arr_size, Address(arr_size, len, f));
280 andptr(arr_size, ~MinObjAlignmentInBytesMask);
281
282 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
283
284 initialize_header(obj, klass, len, t1, t2);
285
286 // clear rest of allocated space
287 const Register len_zero = len;
288 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
289
290 if (CURRENT_ENV->dtrace_alloc_probes()) {
291 assert(obj == rax, "must be");
292 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
293 }
294
295 verify_oop(obj);
296 }
297
298
299
300 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
301 verify_oop(receiver);
302 // explicit null check not needed since load from [klass_offset] causes a trap
303 // check against inline cache
304 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
305 int start_offset = offset();
306
307 if (UseCompressedClassPointers) {
308 load_klass(rscratch1, receiver, rscratch2);
309 cmpptr(rscratch1, iCache);
310 } else {
311 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
312 }
313 // if icache check fails, then jump to runtime routine
314 // Note: RECEIVER must still contain the receiver!
315 jump_cc(Assembler::notEqual,
316 RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
317 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
318 assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
319 }
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 #if !defined(_LP64) && defined(COMPILER2)
336 if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
337 // c2 leaves fpu stack dirty. Clean it on entry
338 empty_FPU_stack();
339 }
340 #endif // !_LP64 && COMPILER2
341 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
342
343 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
344 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
345 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
346 }
347
348
349 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
350 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
351 pop(rbp);
352 }
353
354
355 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
356 if (breakAtEntry || VerifyFPU) {
357 // Verified Entry first instruction should be 5 bytes long for correct
358 // patching by patch_verified_entry().
359 //
360 // Breakpoint and VerifyFPU have one byte first instruction.
361 // Also first instruction will be one byte "push(rbp)" if stack banging
362 // code is not generated (see build_frame() above).
363 // For all these cases generate long instruction first.
364 fat_nop();
365 }
366 if (breakAtEntry) int3();
367 // build frame
368 IA32_ONLY( verify_FPU(0, "method_entry"); )
369 }
370
371 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
372 // rbp, + 0: link
373 // + 1: return address
374 // + 2: argument with offset 0
375 // + 3: argument with offset 1
376 // + 4: ...
377
378 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
379 }
380
381 #ifndef PRODUCT
382
383 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
384 if (!VerifyOops) return;
385 verify_oop_addr(Address(rsp, stack_offset));
386 }
387
388 void C1_MacroAssembler::verify_not_null_oop(Register r) {
389 if (!VerifyOops) return;
390 Label not_null;
391 testptr(r, r);
392 jcc(Assembler::notZero, not_null);
393 stop("non-null oop required");
394 bind(not_null);
395 verify_oop(r);
396 }
397
398 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
399 #ifdef ASSERT
400 if (inv_rax) movptr(rax, 0xDEAD);
401 if (inv_rbx) movptr(rbx, 0xDEAD);
402 if (inv_rcx) movptr(rcx, 0xDEAD);
403 if (inv_rdx) movptr(rdx, 0xDEAD);
404 if (inv_rsi) movptr(rsi, 0xDEAD);
405 if (inv_rdi) movptr(rdi, 0xDEAD);
406 #endif
407 }
408
409 #endif // ifndef PRODUCT