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 "gc/shared/barrierSet.hpp"
29 #include "gc/shared/barrierSetAssembler.hpp"
30 #include "gc/shared/collectedHeap.hpp"
31 #include "gc/shared/tlab_globals.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "oops/arrayOop.hpp"
34 #include "oops/markWord.hpp"
35 #include "runtime/basicLock.hpp"
36 #include "runtime/biasedLocking.hpp"
37 #include "runtime/globals.hpp"
38 #include "runtime/os.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/stubRoutines.hpp"
41 #include "utilities/globalDefinitions.hpp"
42
43 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
44 const Register rklass_decode_tmp = LP64_ONLY(rscratch1) NOT_LP64(noreg);
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, scratch);
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_in_bytes()), obj);
55
56 null_check_offset = offset();
57
58 if (DiagnoseSyncOnValueBasedClasses != 0) {
59 load_klass(hdr, obj, rklass_decode_tmp);
60 movl(hdr, Address(hdr, Klass::access_flags_offset()));
61 testl(hdr, JVM_ACC_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 #else
69 const Register thread = disp_hdr;
70 get_thread(thread);
71 #endif
72 lightweight_lock(obj, hdr, thread, scratch, slow_case);
73 } else {
74 Label done;
75
76 if (UseBiasedLocking) {
77 assert(scratch != noreg, "should have scratch register at this point");
78 biased_locking_enter(disp_hdr, obj, hdr, scratch, rklass_decode_tmp, false, done, &slow_case);
79 }
80
81 // Load object header
82 movptr(hdr, Address(obj, hdr_offset));
83 // and mark it as unlocked
84 orptr(hdr, markWord::unlocked_value);
85 // save unlocked object header into the displaced header location on the stack
86 movptr(Address(disp_hdr, 0), hdr);
87 // test if object header is still the same (i.e. unlocked), and if so, store the
88 // displaced header address in the object header - if it is not the same, get the
89 // object header instead
90 MacroAssembler::lock(); // must be immediately before cmpxchg!
91 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
92 // if the object header was the same, we're done
93 if (PrintBiasedLockingStatistics) {
94 cond_inc32(Assembler::equal,
95 ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
96 }
97 jcc(Assembler::equal, done);
98 // if the object header was not the same, it is now in the hdr register
99 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
100 //
101 // 1) (hdr & aligned_mask) == 0
102 // 2) rsp <= hdr
103 // 3) hdr <= rsp + page_size
104 //
105 // these 3 tests can be done by evaluating the following expression:
106 //
107 // (hdr - rsp) & (aligned_mask - page_size)
108 //
109 // assuming both the stack pointer and page_size have their least
110 // significant 2 bits cleared and page_size is a power of 2
111 subptr(hdr, rsp);
112 andptr(hdr, aligned_mask - os::vm_page_size());
113 // for recursive locking, the result is zero => save it in the displaced header
114 // location (NULL in the displaced hdr location indicates recursive locking)
115 movptr(Address(disp_hdr, 0), hdr);
116 // otherwise we don't care about the result and handle locking via runtime call
117 jcc(Assembler::notZero, slow_case);
118 // done
119 bind(done);
120 }
121 return null_check_offset;
122 }
123
124
125 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
126 const int aligned_mask = BytesPerWord -1;
127 const int hdr_offset = oopDesc::mark_offset_in_bytes();
128 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
129 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
130
131 if (LockingMode == LM_LIGHTWEIGHT) {
132 #ifdef _LP64
133 lightweight_unlock(obj, disp_hdr, r15_thread, hdr, slow_case);
134 #else
135 // This relies on the implementation of lightweight_unlock being able to handle
136 // that the reg_rax and thread Register parameters may alias each other.
137 get_thread(disp_hdr);
138 lightweight_unlock(obj, disp_hdr, disp_hdr, hdr, slow_case);
139 #endif
140 } else {
141 Label done;
142
143 if (UseBiasedLocking) {
144 // load object
145 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
146 biased_locking_exit(obj, hdr, done);
147 }
148
149 // load displaced header
150 movptr(hdr, Address(disp_hdr, 0));
151 // if the loaded hdr is NULL we had recursive locking
152 testptr(hdr, hdr);
153 // if we had recursive locking, we are done
154 jcc(Assembler::zero, done);
155 if (!UseBiasedLocking) {
156 // load object
157 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
158 }
159 verify_oop(obj);
160 // test if object header is pointing to the displaced header, and if so, restore
161 // the displaced header in the object - if the object header is not pointing to
162 // the displaced header, get the object header instead
163 MacroAssembler::lock(); // must be immediately before cmpxchg!
164 cmpxchgptr(hdr, Address(obj, hdr_offset));
165 // if the object header was not pointing to the displaced header,
166 // we do unlocking via runtime call
167 jcc(Assembler::notEqual, slow_case);
168 // done
169 bind(done);
170 }
171 }
172
173
174 // Defines obj, preserves var_size_in_bytes
175 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
176 if (UseTLAB) {
177 tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
178 } else {
179 eden_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
180 }
181 }
182
183
184 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
185 assert_different_registers(obj, klass, len, t1, t2);
186 if (UseCompactObjectHeaders || (UseBiasedLocking && !len->is_valid())) {
187 movptr(t1, Address(klass, Klass::prototype_header_offset()));
188 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
189 } else {
190 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
191 }
192
193 if (!UseCompactObjectHeaders) {
194 #ifdef _LP64
195 if (UseCompressedClassPointers) { // Take care not to kill klass
196 movptr(t1, klass);
197 encode_klass_not_null(t1, rscratch1);
198 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
199 } else
200 #endif
201 {
202 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
203 }
204 }
205 if (len->is_valid()) {
206 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
207 }
208 #ifdef _LP64
209 else if (UseCompressedClassPointers && !UseCompactObjectHeaders) {
210 xorptr(t1, t1);
211 store_klass_gap(obj, t1);
212 }
213 #endif
214 }
215
216
217 // preserves obj, destroys len_in_bytes
218 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
219 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
220 Label done;
221
222 // len_in_bytes is positive and ptr sized
223 subptr(len_in_bytes, hdr_size_in_bytes);
224 jcc(Assembler::zero, done);
225 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
226 bind(done);
227 }
228
229
230 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
231 assert(obj == rax, "obj must be in rax, for cmpxchg");
232 assert_different_registers(obj, t1, t2); // XXX really?
233 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
234
235 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
236
237 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
238 }
239
240 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) {
241 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
242 "con_size_in_bytes is not multiple of alignment");
243 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
244
245 initialize_header(obj, klass, noreg, t1, t2);
246
247 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
248 // clear rest of allocated space
249 const Register t1_zero = t1;
250 const Register index = t2;
251 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
252 int hdr_size_aligned = align_up(hdr_size_in_bytes, BytesPerWord); // klass gap is already cleared by init_header().
253 if (var_size_in_bytes != noreg) {
254 mov(index, var_size_in_bytes);
255 initialize_body(obj, index, hdr_size_aligned, t1_zero);
256 } else if (con_size_in_bytes <= threshold) {
257 // use explicit null stores
258 // code size = 2 + 3*n bytes (n = number of fields to clear)
259 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
260 for (int i = hdr_size_aligned; i < con_size_in_bytes; i += BytesPerWord)
261 movptr(Address(obj, i), t1_zero);
262 } else if (con_size_in_bytes > hdr_size_aligned) {
263 // use loop to null out the fields
264 // code size = 16 bytes for even n (n = number of fields to clear)
265 // initialize last object field first if odd number of fields
266 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
267 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
268 // initialize last object field if constant size is odd
269 if (((con_size_in_bytes - hdr_size_aligned) & 4) != 0)
270 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
271 // initialize remaining object fields: rdx is a multiple of 2
272 { Label loop;
273 bind(loop);
274 movptr(Address(obj, index, Address::times_8, hdr_size_aligned - (1*BytesPerWord)),
275 t1_zero);
276 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_aligned - (2*BytesPerWord)),
277 t1_zero);)
278 decrement(index);
279 jcc(Assembler::notZero, loop);
280 }
281 }
282 }
283
284 if (CURRENT_ENV->dtrace_alloc_probes()) {
285 assert(obj == rax, "must be");
286 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
287 }
288
289 verify_oop(obj);
290 }
291
292 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) {
293 assert(obj == rax, "obj must be in rax, for cmpxchg");
294 assert_different_registers(obj, len, t1, t2, klass);
295
296 // determine alignment mask
297 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
298
299 // check for negative or excessive length
300 cmpptr(len, (int32_t)max_array_allocation_length);
301 jcc(Assembler::above, slow_case);
302
303 const Register arr_size = t2; // okay to be the same
304 // align object end
305 movptr(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
306 lea(arr_size, Address(arr_size, len, f));
307 andptr(arr_size, ~MinObjAlignmentInBytesMask);
308
309 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
310
311 initialize_header(obj, klass, len, t1, t2);
312
313 // clear rest of allocated space
314 const Register len_zero = len;
315 initialize_body(obj, arr_size, base_offset_in_bytes, len_zero);
316
317 if (CURRENT_ENV->dtrace_alloc_probes()) {
318 assert(obj == rax, "must be");
319 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
320 }
321
322 verify_oop(obj);
323 }
324
325
326
327 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
328 verify_oop(receiver);
329 // explicit NULL check not needed since load from [klass_offset] causes a trap
330 // check against inline cache
331 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
332 int start_offset = offset();
333 Register tmp_load_klass = LP64_ONLY(rscratch2) NOT_LP64(noreg);
334
335 if (UseCompressedClassPointers) {
336 load_klass(rscratch1, receiver, tmp_load_klass);
337 cmpptr(rscratch1, iCache);
338 } else {
339 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
340 }
341 // if icache check fails, then jump to runtime routine
342 // Note: RECEIVER must still contain the receiver!
343 jump_cc(Assembler::notEqual,
344 RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
345 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
346 assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
347 }
348
349
350 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
351 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
352 // Make sure there is enough stack space for this method's activation.
353 // Note that we do this before doing an enter(). This matches the
354 // ordering of C2's stack overflow check / rsp decrement and allows
355 // the SharedRuntime stack overflow handling to be consistent
356 // between the two compilers.
357 generate_stack_overflow_check(bang_size_in_bytes);
358
359 push(rbp);
360 if (PreserveFramePointer) {
361 mov(rbp, rsp);
362 }
363 #if !defined(_LP64) && defined(COMPILER2)
364 if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
365 // c2 leaves fpu stack dirty. Clean it on entry
366 empty_FPU_stack();
367 }
368 #endif // !_LP64 && COMPILER2
369 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
370
371 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
372 bs->nmethod_entry_barrier(this);
373 }
374
375
376 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
377 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
378 pop(rbp);
379 }
380
381
382 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
383 if (breakAtEntry || VerifyFPU) {
384 // Verified Entry first instruction should be 5 bytes long for correct
385 // patching by patch_verified_entry().
386 //
387 // Breakpoint and VerifyFPU have one byte first instruction.
388 // Also first instruction will be one byte "push(rbp)" if stack banging
389 // code is not generated (see build_frame() above).
390 // For all these cases generate long instruction first.
391 fat_nop();
392 }
393 if (breakAtEntry) int3();
394 // build frame
395 IA32_ONLY( verify_FPU(0, "method_entry"); )
396 }
397
398 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
399 // rbp, + 0: link
400 // + 1: return address
401 // + 2: argument with offset 0
402 // + 3: argument with offset 1
403 // + 4: ...
404
405 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
406 }
407
408 #ifndef PRODUCT
409
410 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
411 if (!VerifyOops) return;
412 verify_oop_addr(Address(rsp, stack_offset));
413 }
414
415 void C1_MacroAssembler::verify_not_null_oop(Register r) {
416 if (!VerifyOops) return;
417 Label not_null;
418 testptr(r, r);
419 jcc(Assembler::notZero, not_null);
420 stop("non-null oop required");
421 bind(not_null);
422 verify_oop(r);
423 }
424
425 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
426 #ifdef ASSERT
427 if (inv_rax) movptr(rax, 0xDEAD);
428 if (inv_rbx) movptr(rbx, 0xDEAD);
429 if (inv_rcx) movptr(rcx, 0xDEAD);
430 if (inv_rdx) movptr(rdx, 0xDEAD);
431 if (inv_rsi) movptr(rsi, 0xDEAD);
432 if (inv_rdi) movptr(rdi, 0xDEAD);
433 #endif
434 }
435
436 #endif // ifndef PRODUCT