1 /*
2 * Copyright (c) 1999, 2024, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2021, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "c1/c1_MacroAssembler.hpp"
28 #include "c1/c1_Runtime1.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/os.hpp"
37 #include "runtime/sharedRuntime.hpp"
38 #include "runtime/stubRoutines.hpp"
39
40 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
41 FloatRegister f0, FloatRegister f1,
42 Register result)
43 {
44 Label done;
45 if (is_float) {
46 fcmps(f0, f1);
47 } else {
48 fcmpd(f0, f1);
49 }
50 if (unordered_result < 0) {
51 // we want -1 for unordered or less than, 0 for equal and 1 for
52 // greater than.
53 cset(result, NE); // Not equal or unordered
54 cneg(result, result, LT); // Less than or unordered
55 } else {
56 // we want -1 for less than, 0 for equal and 1 for unordered or
57 // greater than.
58 cset(result, NE); // Not equal or unordered
59 cneg(result, result, LO); // Less than
60 }
61 }
62
63 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register temp, Label& slow_case) {
64 const int aligned_mask = BytesPerWord -1;
65 const int hdr_offset = oopDesc::mark_offset_in_bytes();
66 assert_different_registers(hdr, obj, disp_hdr, temp, rscratch2);
67 int null_check_offset = -1;
68
69 verify_oop(obj);
70
71 // save object being locked into the BasicObjectLock
72 str(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
73
74 null_check_offset = offset();
75
76 if (DiagnoseSyncOnValueBasedClasses != 0) {
77 load_klass(hdr, obj);
78 ldrw(hdr, Address(hdr, Klass::access_flags_offset()));
79 tstw(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
80 br(Assembler::NE, slow_case);
81 }
82
83 if (LockingMode == LM_LIGHTWEIGHT) {
84 lightweight_lock(obj, hdr, temp, rscratch2, slow_case);
85 } else if (LockingMode == LM_LEGACY) {
86 Label done;
87 // Load object header
88 ldr(hdr, Address(obj, hdr_offset));
89 // and mark it as unlocked
90 orr(hdr, hdr, markWord::unlocked_value);
91 // save unlocked object header into the displaced header location on the stack
92 str(hdr, Address(disp_hdr, 0));
93 // test if object header is still the same (i.e. unlocked), and if so, store the
94 // displaced header address in the object header - if it is not the same, get the
95 // object header instead
96 lea(rscratch2, Address(obj, hdr_offset));
97 cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/nullptr);
98 // if the object header was the same, we're done
99 // if the object header was not the same, it is now in the hdr register
100 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
101 //
102 // 1) (hdr & aligned_mask) == 0
103 // 2) sp <= hdr
104 // 3) hdr <= sp + page_size
105 //
106 // these 3 tests can be done by evaluating the following expression:
107 //
108 // (hdr - sp) & (aligned_mask - page_size)
109 //
110 // assuming both the stack pointer and page_size have their least
111 // significant 2 bits cleared and page_size is a power of 2
112 mov(rscratch1, sp);
113 sub(hdr, hdr, rscratch1);
114 ands(hdr, hdr, aligned_mask - (int)os::vm_page_size());
115 // for recursive locking, the result is zero => save it in the displaced header
116 // location (null in the displaced hdr location indicates recursive locking)
117 str(hdr, Address(disp_hdr, 0));
118 // otherwise we don't care about the result and handle locking via runtime call
119 cbnz(hdr, slow_case);
120 // done
121 bind(done);
122 }
123 increment(Address(rthread, JavaThread::held_monitor_count_offset()));
124 return null_check_offset;
125 }
126
127
128 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Register temp, Label& slow_case) {
129 const int aligned_mask = BytesPerWord -1;
130 const int hdr_offset = oopDesc::mark_offset_in_bytes();
131 assert_different_registers(hdr, obj, disp_hdr, temp, rscratch2);
132 Label done;
133
134 if (LockingMode != LM_LIGHTWEIGHT) {
135 // load displaced header
136 ldr(hdr, Address(disp_hdr, 0));
137 // if the loaded hdr is null we had recursive locking
138 // if we had recursive locking, we are done
139 cbz(hdr, done);
140 }
141
142 // load object
143 ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
144 verify_oop(obj);
145
146 if (LockingMode == LM_LIGHTWEIGHT) {
147 lightweight_unlock(obj, hdr, temp, rscratch2, slow_case);
148 } else if (LockingMode == LM_LEGACY) {
149 // test if object header is pointing to the displaced header, and if so, restore
150 // the displaced header in the object - if the object header is not pointing to
151 // the displaced header, get the object header instead
152 // if the object header was not pointing to the displaced header,
153 // we do unlocking via runtime call
154 if (hdr_offset) {
155 lea(rscratch1, Address(obj, hdr_offset));
156 cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case);
157 } else {
158 cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case);
159 }
160 // done
161 bind(done);
162 }
163 decrement(Address(rthread, JavaThread::held_monitor_count_offset()));
164 }
165
166
167 // Defines obj, preserves var_size_in_bytes
168 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
169 if (UseTLAB) {
170 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
171 } else {
172 b(slow_case);
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);
178 // This assumes that all prototype bits fit in an int32_t
179 mov(t1, (int32_t)(intptr_t)markWord::prototype().value());
180 str(t1, Address(obj, oopDesc::mark_offset_in_bytes()));
181
182 if (UseCompressedClassPointers) { // Take care not to kill klass
183 encode_klass_not_null(t1, klass);
184 strw(t1, Address(obj, oopDesc::klass_offset_in_bytes()));
185 } else {
186 str(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
187 }
188
189 if (len->is_valid()) {
190 strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
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 strw(zr, Address(obj, base_offset));
196 }
197 } else if (UseCompressedClassPointers) {
198 store_klass_gap(obj, zr);
199 }
200 }
201
202 // preserves obj, destroys len_in_bytes
203 //
204 // Scratch registers: t1 = r10, t2 = r11
205 //
206 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1, Register t2) {
207 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
208 assert(t1 == r10 && t2 == r11, "must be");
209
210 Label done;
211
212 // len_in_bytes is positive and ptr sized
213 subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
214 br(Assembler::EQ, done);
215
216 // zero_words() takes ptr in r10 and count in words in r11
217 mov(rscratch1, len_in_bytes);
218 lea(t1, Address(obj, hdr_size_in_bytes));
219 lsr(t2, rscratch1, LogBytesPerWord);
220 address tpc = zero_words(t1, t2);
221
222 bind(done);
223 if (tpc == nullptr) {
224 Compilation::current()->bailout("no space for trampoline stub");
225 }
226 }
227
228
229 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
230 assert_different_registers(obj, t1, t2); // XXX really?
231 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
232
233 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
234
235 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
236 }
237
238 // Scratch registers: t1 = r10, t2 = r11
239 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) {
240 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
241 "con_size_in_bytes is not multiple of alignment");
242 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
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 index = t2;
249 if (var_size_in_bytes != noreg) {
250 mov(index, var_size_in_bytes);
251 initialize_body(obj, index, hdr_size_in_bytes, t1, t2);
252 if (Compilation::current()->bailed_out()) {
253 return;
254 }
255 } else if (con_size_in_bytes > hdr_size_in_bytes) {
256 con_size_in_bytes -= hdr_size_in_bytes;
257 lea(t1, Address(obj, hdr_size_in_bytes));
258 address tpc = zero_words(t1, con_size_in_bytes / BytesPerWord);
259 if (tpc == nullptr) {
260 Compilation::current()->bailout("no space for trampoline stub");
261 return;
262 }
263 }
264 }
265
266 membar(StoreStore);
267
268 if (CURRENT_ENV->dtrace_alloc_probes()) {
269 assert(obj == r0, "must be");
270 far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
271 }
272
273 verify_oop(obj);
274 }
275 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int base_offset_in_bytes, int f, Register klass, Label& slow_case) {
276 assert_different_registers(obj, len, t1, t2, klass);
277
278 // determine alignment mask
279 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
280
281 // check for negative or excessive length
282 mov(rscratch1, (int32_t)max_array_allocation_length);
283 cmp(len, rscratch1);
284 br(Assembler::HS, slow_case);
285
286 const Register arr_size = t2; // okay to be the same
287 // align object end
288 mov(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask);
289 add(arr_size, arr_size, len, ext::uxtw, f);
290 andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
291
292 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
293
294 initialize_header(obj, klass, len, t1, t2);
295
296 // Align-up to word boundary, because we clear the 4 bytes potentially
297 // following the length field in initialize_header().
298 int base_offset = align_up(base_offset_in_bytes, BytesPerWord);
299 // clear rest of allocated space
300 initialize_body(obj, arr_size, base_offset, t1, t2);
301 if (Compilation::current()->bailed_out()) {
302 return;
303 }
304
305 membar(StoreStore);
306
307 if (CURRENT_ENV->dtrace_alloc_probes()) {
308 assert(obj == r0, "must be");
309 far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
310 }
311
312 verify_oop(obj);
313 }
314
315 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) {
316 assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
317 // Make sure there is enough stack space for this method's activation.
318 // Note that we do this before creating a frame.
319 generate_stack_overflow_check(bang_size_in_bytes);
320 MacroAssembler::build_frame(framesize);
321
322 // Insert nmethod entry barrier into frame.
323 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
324 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */);
325 }
326
327 void C1_MacroAssembler::remove_frame(int framesize) {
328 MacroAssembler::remove_frame(framesize);
329 }
330
331
332 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
333 // If we have to make this method not-entrant we'll overwrite its
334 // first instruction with a jump. For this action to be legal we
335 // must ensure that this first instruction is a B, BL, NOP, BKPT,
336 // SVC, HVC, or SMC. Make it a NOP.
337 nop();
338 }
339
340 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
341 // rfp, + 0: link
342 // + 1: return address
343 // + 2: argument with offset 0
344 // + 3: argument with offset 1
345 // + 4: ...
346
347 ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord));
348 }
349
350 #ifndef PRODUCT
351
352 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
353 if (!VerifyOops) return;
354 verify_oop_addr(Address(sp, stack_offset));
355 }
356
357 void C1_MacroAssembler::verify_not_null_oop(Register r) {
358 if (!VerifyOops) return;
359 Label not_null;
360 cbnz(r, not_null);
361 stop("non-null oop required");
362 bind(not_null);
363 verify_oop(r);
364 }
365
366 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) {
367 #ifdef ASSERT
368 static int nn;
369 if (inv_r0) mov(r0, 0xDEAD);
370 if (inv_r19) mov(r19, 0xDEAD);
371 if (inv_r2) mov(r2, nn++);
372 if (inv_r3) mov(r3, 0xDEAD);
373 if (inv_r4) mov(r4, 0xDEAD);
374 if (inv_r5) mov(r5, 0xDEAD);
375 #endif
376 }
377 #endif // ifndef PRODUCT