1 /*
2 * Copyright (c) 1999, 2023, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2012, 2018 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 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
44 const Register temp_reg = R12_scratch2;
45 Label Lmiss;
46
47 verify_oop(receiver, FILE_AND_LINE);
48 load_klass_check_null(temp_reg, receiver, &Lmiss);
49
50 if (TrapBasedICMissChecks && TrapBasedNullChecks) {
51 trap_ic_miss_check(temp_reg, iCache);
52 } else {
53 Label Lok;
54 cmpd(CCR0, temp_reg, iCache);
55 beq(CCR0, Lok);
56 bind(Lmiss);
57 //load_const_optimized(temp_reg, SharedRuntime::get_ic_miss_stub(), R0);
58 calculate_address_from_global_toc(temp_reg, SharedRuntime::get_ic_miss_stub(), true, true, false);
59 mtctr(temp_reg);
60 bctr();
61 align(32, 12);
62 bind(Lok);
63 }
64 }
65
66
67 void C1_MacroAssembler::explicit_null_check(Register base) {
68 Unimplemented();
69 }
70
71
72 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
73 // Avoid stack bang as first instruction. It may get overwritten by patch_verified_entry.
74 const Register return_pc = R20;
75 mflr(return_pc);
76
77 // Make sure there is enough stack space for this method's activation.
78 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
79 generate_stack_overflow_check(bang_size_in_bytes);
80
81 std(return_pc, _abi0(lr), R1_SP); // SP->lr = return_pc
82 push_frame(frame_size_in_bytes, R0); // SP -= frame_size_in_bytes
83
84 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
85 bs->nmethod_entry_barrier(this, R20);
86 }
87
88
89 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
90 if (breakAtEntry) illtrap();
91 // build frame
92 }
93
94
95 void C1_MacroAssembler::lock_object(Register Rmark, Register Roop, Register Rbox, Register Rscratch, Label& slow_case) {
96 assert_different_registers(Rmark, Roop, Rbox, Rscratch);
97
98 Label done, cas_failed, slow_int;
99
100 // The following move must be the first instruction of emitted since debug
101 // information may be generated for it.
102 // Load object header.
103 ld(Rmark, oopDesc::mark_offset_in_bytes(), Roop);
104
105 verify_oop(Roop, FILE_AND_LINE);
106
107 // Save object being locked into the BasicObjectLock...
108 std(Roop, in_bytes(BasicObjectLock::obj_offset()), Rbox);
109
110 if (DiagnoseSyncOnValueBasedClasses != 0) {
111 load_klass(Rscratch, Roop);
112 lwz(Rscratch, in_bytes(Klass::access_flags_offset()), Rscratch);
113 testbitdi(CCR0, R0, Rscratch, exact_log2(JVM_ACC_IS_VALUE_BASED_CLASS));
114 bne(CCR0, slow_int);
115 }
116
117 if (LockingMode == LM_LIGHTWEIGHT) {
118 lightweight_lock(Roop, Rmark, Rscratch, slow_int);
119 } else if (LockingMode == LM_LEGACY) {
120 // ... and mark it unlocked.
121 ori(Rmark, Rmark, markWord::unlocked_value);
122
123 // Save unlocked object header into the displaced header location on the stack.
124 std(Rmark, BasicLock::displaced_header_offset_in_bytes(), Rbox);
125
126 // Compare object markWord with Rmark and if equal exchange Rscratch with object markWord.
127 assert(oopDesc::mark_offset_in_bytes() == 0, "cas must take a zero displacement");
128 cmpxchgd(/*flag=*/CCR0,
129 /*current_value=*/Rscratch,
130 /*compare_value=*/Rmark,
131 /*exchange_value=*/Rbox,
132 /*where=*/Roop/*+0==mark_offset_in_bytes*/,
133 MacroAssembler::MemBarRel | MacroAssembler::MemBarAcq,
134 MacroAssembler::cmpxchgx_hint_acquire_lock(),
135 noreg,
136 &cas_failed,
137 /*check without membar and ldarx first*/true);
138 // If compare/exchange succeeded we found an unlocked object and we now have locked it
139 // hence we are done.
140 }
141 b(done);
142
143 bind(slow_int);
144 b(slow_case); // far
145
146 if (LockingMode == LM_LEGACY) {
147 bind(cas_failed);
148 // We did not find an unlocked object so see if this is a recursive case.
149 sub(Rscratch, Rscratch, R1_SP);
150 load_const_optimized(R0, (~(os::vm_page_size()-1) | markWord::lock_mask_in_place));
151 and_(R0/*==0?*/, Rscratch, R0);
152 std(R0/*==0, perhaps*/, BasicLock::displaced_header_offset_in_bytes(), Rbox);
153 bne(CCR0, slow_int);
154 }
155
156 bind(done);
157 inc_held_monitor_count(Rmark /*tmp*/);
158 }
159
160
161 void C1_MacroAssembler::unlock_object(Register Rmark, Register Roop, Register Rbox, Label& slow_case) {
162 assert_different_registers(Rmark, Roop, Rbox);
163
164 Label slow_int, done;
165
166 Address mark_addr(Roop, oopDesc::mark_offset_in_bytes());
167 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
168
169 if (LockingMode != LM_LIGHTWEIGHT) {
170 // Test first if it is a fast recursive unlock.
171 ld(Rmark, BasicLock::displaced_header_offset_in_bytes(), Rbox);
172 cmpdi(CCR0, Rmark, 0);
173 beq(CCR0, done);
174 }
175
176 // Load object.
177 ld(Roop, in_bytes(BasicObjectLock::obj_offset()), Rbox);
178 verify_oop(Roop, FILE_AND_LINE);
179
180 if (LockingMode == LM_LIGHTWEIGHT) {
181 lightweight_unlock(Roop, Rmark, slow_int);
182 } else if (LockingMode == LM_LEGACY) {
183 // Check if it is still a light weight lock, this is is true if we see
184 // the stack address of the basicLock in the markWord of the object.
185 cmpxchgd(/*flag=*/CCR0,
186 /*current_value=*/R0,
187 /*compare_value=*/Rbox,
188 /*exchange_value=*/Rmark,
189 /*where=*/Roop,
190 MacroAssembler::MemBarRel,
191 MacroAssembler::cmpxchgx_hint_release_lock(),
192 noreg,
193 &slow_int);
194 }
195 b(done);
196 bind(slow_int);
197 b(slow_case); // far
198
199 // Done
200 bind(done);
201 dec_held_monitor_count(Rmark /*tmp*/);
202 }
203
204
205 void C1_MacroAssembler::try_allocate(
206 Register obj, // result: pointer to object after successful allocation
207 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
208 int con_size_in_bytes, // object size in bytes if known at compile time
209 Register t1, // temp register, must be global register for incr_allocated_bytes
210 Register t2, // temp register
211 Label& slow_case // continuation point if fast allocation fails
212 ) {
213 if (UseTLAB) {
214 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
215 } else {
216 b(slow_case);
217 }
218 }
219
220
221 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
222 assert_different_registers(obj, klass, len, t1, t2);
223 load_const_optimized(t1, (intx)markWord::prototype().value());
224 std(t1, oopDesc::mark_offset_in_bytes(), obj);
225 store_klass(obj, klass);
226 if (len->is_valid()) {
227 stw(len, arrayOopDesc::length_offset_in_bytes(), obj);
228 } else if (UseCompressedClassPointers) {
229 // Otherwise length is in the class gap.
230 store_klass_gap(obj);
231 }
232 }
233
234
235 void C1_MacroAssembler::initialize_body(Register base, Register index) {
236 assert_different_registers(base, index);
237 srdi(index, index, LogBytesPerWord);
238 clear_memory_doubleword(base, index);
239 }
240
241 void C1_MacroAssembler::initialize_body(Register obj, Register tmp1, Register tmp2,
242 int obj_size_in_bytes, int hdr_size_in_bytes) {
243 const int index = (obj_size_in_bytes - hdr_size_in_bytes) / HeapWordSize;
244
245 // 2x unrolled loop is shorter with more than 9 HeapWords.
246 if (index <= 9) {
247 clear_memory_unrolled(obj, index, R0, hdr_size_in_bytes);
248 } else {
249 const Register base_ptr = tmp1,
250 cnt_dwords = tmp2;
251
252 addi(base_ptr, obj, hdr_size_in_bytes); // Compute address of first element.
253 clear_memory_doubleword(base_ptr, cnt_dwords, R0, index);
254 }
255 }
256
257 void C1_MacroAssembler::allocate_object(
258 Register obj, // result: pointer to object after successful allocation
259 Register t1, // temp register
260 Register t2, // temp register
261 Register t3, // temp register
262 int hdr_size, // object header size in words
263 int obj_size, // object size in words
264 Register klass, // object klass
265 Label& slow_case // continuation point if fast allocation fails
266 ) {
267 assert_different_registers(obj, t1, t2, t3, klass);
268
269 // allocate space & initialize header
270 if (!is_simm16(obj_size * wordSize)) {
271 // Would need to use extra register to load
272 // object size => go the slow case for now.
273 b(slow_case);
274 return;
275 }
276 try_allocate(obj, noreg, obj_size * wordSize, t2, t3, slow_case);
277
278 initialize_object(obj, klass, noreg, obj_size * HeapWordSize, t1, t2);
279 }
280
281 void C1_MacroAssembler::initialize_object(
282 Register obj, // result: pointer to object after successful allocation
283 Register klass, // object klass
284 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
285 int con_size_in_bytes, // object size in bytes if known at compile time
286 Register t1, // temp register
287 Register t2 // temp register
288 ) {
289 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
290
291 initialize_header(obj, klass, noreg, t1, t2);
292
293 #ifdef ASSERT
294 {
295 lwz(t1, in_bytes(Klass::layout_helper_offset()), klass);
296 if (var_size_in_bytes != noreg) {
297 cmpw(CCR0, t1, var_size_in_bytes);
298 } else {
299 cmpwi(CCR0, t1, con_size_in_bytes);
300 }
301 asm_assert_eq("bad size in initialize_object");
302 }
303 #endif
304
305 // Initialize body.
306 if (var_size_in_bytes != noreg) {
307 // Use a loop.
308 addi(t1, obj, hdr_size_in_bytes); // Compute address of first element.
309 addi(t2, var_size_in_bytes, -hdr_size_in_bytes); // Compute size of body.
310 initialize_body(t1, t2);
311 } else if (con_size_in_bytes > hdr_size_in_bytes) {
312 // Use a loop.
313 initialize_body(obj, t1, t2, con_size_in_bytes, hdr_size_in_bytes);
314 }
315
316 if (CURRENT_ENV->dtrace_alloc_probes()) {
317 Unimplemented();
318 // assert(obj == O0, "must be");
319 // call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
320 // relocInfo::runtime_call_type);
321 }
322
323 verify_oop(obj, FILE_AND_LINE);
324 }
325
326
327 void C1_MacroAssembler::allocate_array(
328 Register obj, // result: pointer to array after successful allocation
329 Register len, // array length
330 Register t1, // temp register
331 Register t2, // temp register
332 Register t3, // temp register
333 int base_offset_in_bytes, // elements offset in bytes
334 int elt_size, // element size in bytes
335 Register klass, // object klass
336 Label& slow_case // continuation point if fast allocation fails
337 ) {
338 assert_different_registers(obj, len, t1, t2, t3, klass);
339
340 // Determine alignment mask.
341 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
342 int log2_elt_size = exact_log2(elt_size);
343
344 // Check for negative or excessive length.
345 size_t max_length = max_array_allocation_length >> log2_elt_size;
346 if (UseTLAB) {
347 size_t max_tlab = align_up(ThreadLocalAllocBuffer::max_size() >> log2_elt_size, 64*K);
348 if (max_tlab < max_length) { max_length = max_tlab; }
349 }
350 load_const_optimized(t1, max_length);
351 cmpld(CCR0, len, t1);
352 bc_far_optimized(Assembler::bcondCRbiIs1, bi0(CCR0, Assembler::greater), slow_case);
353
354 // compute array size
355 // note: If 0 <= len <= max_length, len*elt_size + header + alignment is
356 // smaller or equal to the largest integer; also, since top is always
357 // aligned, we can do the alignment here instead of at the end address
358 // computation.
359 const Register arr_size = t1;
360 Register arr_len_in_bytes = len;
361 if (elt_size != 1) {
362 sldi(t1, len, log2_elt_size);
363 arr_len_in_bytes = t1;
364 }
365 addi(arr_size, arr_len_in_bytes, base_offset_in_bytes + MinObjAlignmentInBytesMask); // Add space for header & alignment.
366 clrrdi(arr_size, arr_size, LogMinObjAlignmentInBytes); // Align array size.
367
368 // Allocate space & initialize header.
369 try_allocate(obj, arr_size, 0, t2, t3, slow_case);
370 initialize_header(obj, klass, len, t2, t3);
371
372 // Initialize body.
373 const Register base = t2;
374 const Register index = t3;
375 addi(base, obj, base_offset_in_bytes); // compute address of first element
376 addi(index, arr_size, -(base_offset_in_bytes)); // compute index = number of bytes to clear
377
378 // Zero first 4 bytes, if start offset is not word aligned.
379 if (!is_aligned(base_offset_in_bytes, BytesPerWord)) {
380 assert(is_aligned(base_offset_in_bytes, BytesPerInt), "must be 4-byte aligned");
381 li(t1, 0);
382 stw(t1, 0, base);
383 addi(base, base, BytesPerInt);
384 // Note: initialize_body will align index down, no need to correct it here.
385 }
386
387 initialize_body(base, index);
388
389 if (CURRENT_ENV->dtrace_alloc_probes()) {
390 Unimplemented();
391 //assert(obj == O0, "must be");
392 //call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
393 // relocInfo::runtime_call_type);
394 }
395
396 verify_oop(obj, FILE_AND_LINE);
397 }
398
399
400 #ifndef PRODUCT
401
402 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
403 verify_oop_addr((RegisterOrConstant)stack_offset, R1_SP, "broken oop in stack slot");
404 }
405
406 void C1_MacroAssembler::verify_not_null_oop(Register r) {
407 Label not_null;
408 cmpdi(CCR0, r, 0);
409 bne(CCR0, not_null);
410 stop("non-null oop required");
411 bind(not_null);
412 verify_oop(r, FILE_AND_LINE);
413 }
414
415 #endif // PRODUCT
416
417 void C1_MacroAssembler::null_check(Register r, Label* Lnull) {
418 if (TrapBasedNullChecks) { // SIGTRAP based
419 trap_null_check(r);
420 } else { // explicit
421 //const address exception_entry = Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id);
422 assert(Lnull != nullptr, "must have Label for explicit check");
423 cmpdi(CCR0, r, 0);
424 bc_far_optimized(Assembler::bcondCRbiIs1, bi0(CCR0, Assembler::equal), *Lnull);
425 }
426 }
427
428 address C1_MacroAssembler::call_c_with_frame_resize(address dest, int frame_resize) {
429 if (frame_resize) { resize_frame(-frame_resize, R0); }
430 #if defined(ABI_ELFv2)
431 address return_pc = call_c(dest, relocInfo::runtime_call_type);
432 #else
433 address return_pc = call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, dest), relocInfo::runtime_call_type);
434 #endif
435 if (frame_resize) { resize_frame(frame_resize, R0); }
436 return return_pc;
437 }