1 /*
2 * Copyright (c) 1999, 2024, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2012, 2024 SAP SE. 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 "asm/macroAssembler.inline.hpp"
28 #include "c1/c1_MacroAssembler.hpp"
29 #include "c1/c1_Runtime1.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 #include "utilities/align.hpp"
40 #include "utilities/macros.hpp"
41 #include "utilities/powerOfTwo.hpp"
42
43
44 void C1_MacroAssembler::explicit_null_check(Register base) {
45 Unimplemented();
46 }
47
48
49 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
50 // Avoid stack bang as first instruction. It may get overwritten by patch_verified_entry.
51 const Register return_pc = R20;
52 mflr(return_pc);
53
54 // Make sure there is enough stack space for this method's activation.
55 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
56 generate_stack_overflow_check(bang_size_in_bytes);
57
58 std(return_pc, _abi0(lr), R1_SP); // SP->lr = return_pc
59 push_frame(frame_size_in_bytes, R0); // SP -= frame_size_in_bytes
60
61 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
62 bs->nmethod_entry_barrier(this, R20);
63 }
64
65
66 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
67 if (breakAtEntry) illtrap();
68 // build frame
69 }
70
71
72 void C1_MacroAssembler::lock_object(Register Rmark, Register Roop, Register Rbox, Register Rscratch, Label& slow_case) {
73 assert_different_registers(Rmark, Roop, Rbox, Rscratch);
74
75 Label done, cas_failed, slow_int;
76
77 // The following move must be the first instruction of emitted since debug
78 // information may be generated for it.
79 // Load object header.
80 ld(Rmark, oopDesc::mark_offset_in_bytes(), Roop);
81
82 verify_oop(Roop, FILE_AND_LINE);
83
84 // Save object being locked into the BasicObjectLock...
85 std(Roop, in_bytes(BasicObjectLock::obj_offset()), Rbox);
86
87 if (DiagnoseSyncOnValueBasedClasses != 0) {
88 load_klass(Rscratch, Roop);
89 lbz(Rscratch, in_bytes(Klass::misc_flags_offset()), Rscratch);
90 testbitdi(CCR0, R0, Rscratch, exact_log2(KlassFlags::_misc_is_value_based_class));
91 bne(CCR0, slow_int);
92 }
93
94 if (LockingMode == LM_LIGHTWEIGHT) {
95 lightweight_lock(Rbox, Roop, Rmark, Rscratch, slow_int);
96 } else if (LockingMode == LM_LEGACY) {
97 // ... and mark it unlocked.
98 ori(Rmark, Rmark, markWord::unlocked_value);
99
100 // Save unlocked object header into the displaced header location on the stack.
101 std(Rmark, BasicLock::displaced_header_offset_in_bytes(), Rbox);
102
103 // Compare object markWord with Rmark and if equal exchange Rscratch with object markWord.
104 assert(oopDesc::mark_offset_in_bytes() == 0, "cas must take a zero displacement");
105 cmpxchgd(/*flag=*/CCR0,
106 /*current_value=*/Rscratch,
107 /*compare_value=*/Rmark,
108 /*exchange_value=*/Rbox,
109 /*where=*/Roop/*+0==mark_offset_in_bytes*/,
110 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq,
111 MacroAssembler::cmpxchgx_hint_acquire_lock(),
112 noreg,
113 &cas_failed,
114 /*check without membar and ldarx first*/true);
115 // If compare/exchange succeeded we found an unlocked object and we now have locked it
116 // hence we are done.
117 } else {
118 assert(false, "Unhandled LockingMode:%d", LockingMode);
119 }
120 b(done);
121
122 bind(slow_int);
123 b(slow_case); // far
124
125 if (LockingMode == LM_LEGACY) {
126 bind(cas_failed);
127 // We did not find an unlocked object so see if this is a recursive case.
128 sub(Rscratch, Rscratch, R1_SP);
129 load_const_optimized(R0, (~(os::vm_page_size()-1) | markWord::lock_mask_in_place));
130 and_(R0/*==0?*/, Rscratch, R0);
131 std(R0/*==0, perhaps*/, BasicLock::displaced_header_offset_in_bytes(), Rbox);
132 bne(CCR0, slow_int);
133 }
134
135 bind(done);
136 inc_held_monitor_count(Rmark /*tmp*/);
137 }
138
139
140 void C1_MacroAssembler::unlock_object(Register Rmark, Register Roop, Register Rbox, Label& slow_case) {
141 assert_different_registers(Rmark, Roop, Rbox);
142
143 Label slow_int, done;
144
145 Address mark_addr(Roop, oopDesc::mark_offset_in_bytes());
146 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
147
148 if (LockingMode != LM_LIGHTWEIGHT) {
149 // Test first if it is a fast recursive unlock.
150 ld(Rmark, BasicLock::displaced_header_offset_in_bytes(), Rbox);
151 cmpdi(CCR0, Rmark, 0);
152 beq(CCR0, done);
153 }
154
155 // Load object.
156 ld(Roop, in_bytes(BasicObjectLock::obj_offset()), Rbox);
157 verify_oop(Roop, FILE_AND_LINE);
158
159 if (LockingMode == LM_LIGHTWEIGHT) {
160 lightweight_unlock(Roop, Rmark, slow_int);
161 } else if (LockingMode == LM_LEGACY) {
162 // Check if it is still a light weight lock, this is is true if we see
163 // the stack address of the basicLock in the markWord of the object.
164 cmpxchgd(/*flag=*/CCR0,
165 /*current_value=*/R0,
166 /*compare_value=*/Rbox,
167 /*exchange_value=*/Rmark,
168 /*where=*/Roop,
169 MacroAssembler::MemBarRel,
170 MacroAssembler::cmpxchgx_hint_release_lock(),
171 noreg,
172 &slow_int);
173 } else {
174 assert(false, "Unhandled LockingMode:%d", LockingMode);
175 }
176 b(done);
177 bind(slow_int);
178 b(slow_case); // far
179
180 // Done
181 bind(done);
182 dec_held_monitor_count(Rmark /*tmp*/);
183 }
184
185
186 void C1_MacroAssembler::try_allocate(
187 Register obj, // result: pointer to object after successful allocation
188 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
189 int con_size_in_bytes, // object size in bytes if known at compile time
190 Register t1, // temp register
191 Register t2, // temp register
192 Label& slow_case // continuation point if fast allocation fails
193 ) {
194 if (UseTLAB) {
195 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
196 } else {
197 b(slow_case);
198 }
199 }
200
201
202 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
203 assert_different_registers(obj, klass, len, t1, t2);
204 load_const_optimized(t1, (intx)markWord::prototype().value());
205 std(t1, oopDesc::mark_offset_in_bytes(), obj);
206 store_klass(obj, klass);
207 if (len->is_valid()) {
208 stw(len, arrayOopDesc::length_offset_in_bytes(), obj);
209 } else if (UseCompressedClassPointers) {
210 // Otherwise length is in the class gap.
211 store_klass_gap(obj);
212 }
213 }
214
215
216 void C1_MacroAssembler::initialize_body(Register base, Register index) {
217 assert_different_registers(base, index);
218 srdi(index, index, LogBytesPerWord);
219 clear_memory_doubleword(base, index);
220 }
221
222 void C1_MacroAssembler::initialize_body(Register obj, Register tmp1, Register tmp2,
223 int obj_size_in_bytes, int hdr_size_in_bytes) {
224 const int index = (obj_size_in_bytes - hdr_size_in_bytes) / HeapWordSize;
225
226 // 2x unrolled loop is shorter with more than 9 HeapWords.
227 if (index <= 9) {
228 clear_memory_unrolled(obj, index, R0, hdr_size_in_bytes);
229 } else {
230 const Register base_ptr = tmp1,
231 cnt_dwords = tmp2;
232
233 addi(base_ptr, obj, hdr_size_in_bytes); // Compute address of first element.
234 clear_memory_doubleword(base_ptr, cnt_dwords, R0, index);
235 }
236 }
237
238 void C1_MacroAssembler::allocate_object(
239 Register obj, // result: pointer to object after successful allocation
240 Register t1, // temp register
241 Register t2, // temp register
242 Register t3, // temp register
243 int hdr_size, // object header size in words
244 int obj_size, // object size in words
245 Register klass, // object klass
246 Label& slow_case // continuation point if fast allocation fails
247 ) {
248 assert_different_registers(obj, t1, t2, t3, klass);
249
250 // allocate space & initialize header
251 if (!is_simm16(obj_size * wordSize)) {
252 // Would need to use extra register to load
253 // object size => go the slow case for now.
254 b(slow_case);
255 return;
256 }
257 try_allocate(obj, noreg, obj_size * wordSize, t2, t3, slow_case);
258
259 initialize_object(obj, klass, noreg, obj_size * HeapWordSize, t1, t2);
260 }
261
262 void C1_MacroAssembler::initialize_object(
263 Register obj, // result: pointer to object after successful allocation
264 Register klass, // object klass
265 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
266 int con_size_in_bytes, // object size in bytes if known at compile time
267 Register t1, // temp register
268 Register t2 // temp register
269 ) {
270 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
271
272 initialize_header(obj, klass, noreg, t1, t2);
273
274 #ifdef ASSERT
275 {
276 lwz(t1, in_bytes(Klass::layout_helper_offset()), klass);
277 if (var_size_in_bytes != noreg) {
278 cmpw(CCR0, t1, var_size_in_bytes);
279 } else {
280 cmpwi(CCR0, t1, con_size_in_bytes);
281 }
282 asm_assert_eq("bad size in initialize_object");
283 }
284 #endif
285
286 // Initialize body.
287 if (var_size_in_bytes != noreg) {
288 // Use a loop.
289 addi(t1, obj, hdr_size_in_bytes); // Compute address of first element.
290 addi(t2, var_size_in_bytes, -hdr_size_in_bytes); // Compute size of body.
291 initialize_body(t1, t2);
292 } else if (con_size_in_bytes > hdr_size_in_bytes) {
293 // Use a loop.
294 initialize_body(obj, t1, t2, con_size_in_bytes, hdr_size_in_bytes);
295 }
296
297 if (CURRENT_ENV->dtrace_alloc_probes()) {
298 Unimplemented();
299 // assert(obj == O0, "must be");
300 // call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)),
301 // relocInfo::runtime_call_type);
302 }
303
304 verify_oop(obj, FILE_AND_LINE);
305 }
306
307
308 void C1_MacroAssembler::allocate_array(
309 Register obj, // result: pointer to array after successful allocation
310 Register len, // array length
311 Register t1, // temp register
312 Register t2, // temp register
313 Register t3, // temp register
314 int base_offset_in_bytes, // elements offset in bytes
315 int elt_size, // element size in bytes
316 Register klass, // object klass
317 Label& slow_case, // continuation point if fast allocation fails
318 bool zero_array // zero the allocated array or not
319 ) {
320 assert_different_registers(obj, len, t1, t2, t3, klass);
321
322 // Determine alignment mask.
323 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
324 int log2_elt_size = exact_log2(elt_size);
325
326 // Check for negative or excessive length.
327 size_t max_length = max_array_allocation_length >> log2_elt_size;
328 if (UseTLAB) {
329 size_t max_tlab = align_up(ThreadLocalAllocBuffer::max_size() >> log2_elt_size, 64*K);
330 if (max_tlab < max_length) { max_length = max_tlab; }
331 }
332 load_const_optimized(t1, max_length);
333 cmpld(CCR0, len, t1);
334 bc_far_optimized(Assembler::bcondCRbiIs1, bi0(CCR0, Assembler::greater), slow_case);
335
336 // compute array size
337 // note: If 0 <= len <= max_length, len*elt_size + header + alignment is
338 // smaller or equal to the largest integer; also, since top is always
339 // aligned, we can do the alignment here instead of at the end address
340 // computation.
341 const Register arr_size = t1;
342 Register arr_len_in_bytes = len;
343 if (elt_size != 1) {
344 sldi(t1, len, log2_elt_size);
345 arr_len_in_bytes = t1;
346 }
347 addi(arr_size, arr_len_in_bytes, base_offset_in_bytes + MinObjAlignmentInBytesMask); // Add space for header & alignment.
348 clrrdi(arr_size, arr_size, LogMinObjAlignmentInBytes); // Align array size.
349
350 // Allocate space & initialize header.
351 try_allocate(obj, arr_size, 0, t2, t3, slow_case);
352 initialize_header(obj, klass, len, t2, t3);
353
354 if (zero_array) {
355 // Initialize body.
356 const Register base = t2;
357 const Register index = t3;
358 addi(base, obj, base_offset_in_bytes); // compute address of first element
359 addi(index, arr_size, -(base_offset_in_bytes)); // compute index = number of bytes to clear
360
361 // Zero first 4 bytes, if start offset is not word aligned.
362 if (!is_aligned(base_offset_in_bytes, BytesPerWord)) {
363 assert(is_aligned(base_offset_in_bytes, BytesPerInt), "must be 4-byte aligned");
364 li(t1, 0);
365 stw(t1, 0, base);
366 addi(base, base, BytesPerInt);
367 // Note: initialize_body will align index down, no need to correct it here.
368 }
369
370 initialize_body(base, index);
371 }
372
373 if (CURRENT_ENV->dtrace_alloc_probes()) {
374 Unimplemented();
375 //assert(obj == O0, "must be");
376 //call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(C1StubId::dtrace_object_alloc_id)),
377 // relocInfo::runtime_call_type);
378 }
379
380 verify_oop(obj, FILE_AND_LINE);
381 }
382
383
384 #ifndef PRODUCT
385
386 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
387 verify_oop_addr((RegisterOrConstant)stack_offset, R1_SP, "broken oop in stack slot");
388 }
389
390 void C1_MacroAssembler::verify_not_null_oop(Register r) {
391 Label not_null;
392 cmpdi(CCR0, r, 0);
393 bne(CCR0, not_null);
394 stop("non-null oop required");
395 bind(not_null);
396 verify_oop(r, FILE_AND_LINE);
397 }
398
399 #endif // PRODUCT
400
401 void C1_MacroAssembler::null_check(Register r, Label* Lnull) {
402 if (TrapBasedNullChecks) { // SIGTRAP based
403 trap_null_check(r);
404 } else { // explicit
405 //const address exception_entry = Runtime1::entry_for(C1StubId::throw_null_pointer_exception_id);
406 assert(Lnull != nullptr, "must have Label for explicit check");
407 cmpdi(CCR0, r, 0);
408 bc_far_optimized(Assembler::bcondCRbiIs1, bi0(CCR0, Assembler::equal), *Lnull);
409 }
410 }