1 /*
2 * Copyright (c) 1999, 2020, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, Red Hat Inc. All rights reserved.
4 * Copyright (c) 2020, 2022, Huawei Technologies Co., Ltd. All rights reserved.
5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 *
7 * This code is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 only, as
9 * published by the Free Software Foundation.
10 *
11 * This code is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * version 2 for more details (a copy is included in the LICENSE file that
15 * accompanied this code).
16 *
17 * You should have received a copy of the GNU General Public License version
18 * 2 along with this work; if not, write to the Free Software Foundation,
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 *
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
22 * or visit www.oracle.com if you need additional information or have any
23 * questions.
24 *
25 */
26
27 #include "precompiled.hpp"
28 #include "c1/c1_LIR.hpp"
29 #include "c1/c1_MacroAssembler.hpp"
30 #include "c1/c1_Runtime1.hpp"
31 #include "classfile/systemDictionary.hpp"
32 #include "gc/shared/barrierSetAssembler.hpp"
33 #include "gc/shared/collectedHeap.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/markWord.hpp"
37 #include "runtime/basicLock.hpp"
38 #include "runtime/os.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/stubRoutines.hpp"
41
42 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result,
43 FloatRegister freg0, FloatRegister freg1,
44 Register result)
45 {
46 if (is_float) {
47 float_compare(result, freg0, freg1, unordered_result);
48 } else {
49 double_compare(result, freg0, freg1, unordered_result);
50 }
51 }
52
53 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
54 const int aligned_mask = BytesPerWord - 1;
55 const int hdr_offset = oopDesc::mark_offset_in_bytes();
56 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
57 Label done;
58 int null_check_offset = -1;
59
60 verify_oop(obj);
61
62 // save object being locked into the BasicObjectLock
63 sd(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
64
65 null_check_offset = offset();
66
67 if (DiagnoseSyncOnValueBasedClasses != 0) {
68 load_klass(hdr, obj);
69 lwu(hdr, Address(hdr, Klass::access_flags_offset()));
70 andi(t0, hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
71 bnez(t0, slow_case, true /* is_far */);
72 }
73
74 if (UseBiasedLocking) {
75 assert(tmp != noreg, "should have tmp register at this point");
76 biased_locking_enter(disp_hdr, obj, hdr, tmp, false, done, &slow_case);
77 }
78
79 // Load object header
80 ld(hdr, Address(obj, hdr_offset));
81 // and mark it as unlocked
82 ori(hdr, hdr, markWord::unlocked_value);
83 // save unlocked object header into the displaced header location on the stack
84 sd(hdr, Address(disp_hdr, 0));
85 // test if object header is still the same (i.e. unlocked), and if so, store the
86 // displaced header address in the object header - if it is not the same, get the
87 // object header instead
88 la(t1, Address(obj, hdr_offset));
89 cmpxchgptr(hdr, disp_hdr, t1, t0, done, /*fallthough*/NULL);
90 // if the object header was the same, we're done
91 // if the object header was not the same, it is now in the hdr register
92 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
93 //
94 // 1) (hdr & aligned_mask) == 0
95 // 2) sp <= hdr
96 // 3) hdr <= sp + page_size
97 //
98 // these 3 tests can be done by evaluating the following expression:
99 //
100 // (hdr -sp) & (aligned_mask - page_size)
101 //
102 // assuming both the stack pointer and page_size have their least
103 // significant 2 bits cleared and page_size is a power of 2
104 sub(hdr, hdr, sp);
105 li(t0, aligned_mask - os::vm_page_size());
106 andr(hdr, hdr, t0);
107 // for recursive locking, the result is zero => save it in the displaced header
108 // location (NULL in the displaced hdr location indicates recursive locking)
109 sd(hdr, Address(disp_hdr, 0));
110 // otherwise we don't care about the result and handle locking via runtime call
111 bnez(hdr, slow_case, /* is_far */ true);
112 bind(done);
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(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
120 Label done;
121
122 if (UseBiasedLocking) {
123 // load object
124 ld(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
125 biased_locking_exit(obj, hdr, done);
126 }
127
128 // load displaced header
129 ld(hdr, Address(disp_hdr, 0));
130 // if the loaded hdr is NULL we had recursive locking
131 // if we had recursive locking, we are done
132 beqz(hdr, done);
133 if (!UseBiasedLocking) {
134 // load object
135 ld(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
136 }
137 verify_oop(obj);
138 // test if object header is pointing to the displaced header, and if so, restore
139 // the displaced header in the object - if the object header is not pointing to
140 // the displaced header, get the object header instead
141 // if the object header was not pointing to the displaced header,
142 // we do unlocking via runtime call
143 if (hdr_offset) {
144 la(t0, Address(obj, hdr_offset));
145 cmpxchgptr(disp_hdr, hdr, t0, t1, done, &slow_case);
146 } else {
147 cmpxchgptr(disp_hdr, hdr, obj, t1, done, &slow_case);
148 }
149 bind(done);
150 }
151
152 // Defines obj, preserves var_size_in_bytes
153 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register tmp1, Register tmp2, Label& slow_case) {
154 if (UseTLAB) {
155 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, tmp1, tmp2, slow_case, /* is_far */ true);
156 } else {
157 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, tmp1, slow_case, /* is_far */ true);
158 }
159 }
160
161 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register tmp1, Register tmp2) {
162 assert_different_registers(obj, klass, len);
163 if (UseBiasedLocking & !len->is_valid()) {
164 assert_different_registers(obj, klass, len, tmp1, tmp2);
165 ld(tmp1, Address(klass, Klass::prototype_header_offset()));
166 } else {
167 // This assumes that all prototype bits fitr in an int32_t
168 mv(tmp1, (int32_t)(intptr_t)markWord::prototype().value());
169 }
170 sd(tmp1, Address(obj, oopDesc::mark_offset_in_bytes()));
171
172 if (UseCompressedClassPointers) { // Take care not to kill klass
173 encode_klass_not_null(tmp1, klass, tmp2);
174 sw(tmp1, Address(obj, oopDesc::klass_offset_in_bytes()));
175 } else {
176 sd(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
177 }
178
179 if (len->is_valid()) {
180 sw(len, Address(obj, arrayOopDesc::length_offset_in_bytes()));
181 } else if (UseCompressedClassPointers) {
182 store_klass_gap(obj, zr);
183 }
184 }
185
186 // preserves obj, destroys len_in_bytes
187 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register tmp) {
188 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
189 Label done;
190
191 // len_in_bytes is positive and ptr sized
192 sub(len_in_bytes, len_in_bytes, hdr_size_in_bytes);
193 beqz(len_in_bytes, done);
194
195 // Preserve obj
196 if (hdr_size_in_bytes) {
197 add(obj, obj, hdr_size_in_bytes);
198 }
199 zero_memory(obj, len_in_bytes, tmp);
200 if (hdr_size_in_bytes) {
201 sub(obj, obj, hdr_size_in_bytes);
202 }
203
204 bind(done);
205 }
206
207 void C1_MacroAssembler::allocate_object(Register obj, Register tmp1, Register tmp2, int header_size, int object_size, Register klass, Label& slow_case) {
208 assert_different_registers(obj, tmp1, tmp2);
209 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
210
211 try_allocate(obj, noreg, object_size * BytesPerWord, tmp1, tmp2, slow_case);
212
213 initialize_object(obj, klass, noreg, object_size * HeapWordSize, tmp1, tmp2, UseTLAB);
214 }
215
216 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register tmp1, Register tmp2, bool is_tlab_allocated) {
217 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
218 "con_size_in_bytes is not multiple of alignment");
219 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
220
221 initialize_header(obj, klass, noreg, tmp1, tmp2);
222
223 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
224 // clear rest of allocated space
225 const Register index = tmp2;
226 // 16: multipler for threshold
227 const int threshold = 16 * BytesPerWord; // approximate break even point for code size (see comments below)
228 if (var_size_in_bytes != noreg) {
229 mv(index, var_size_in_bytes);
230 initialize_body(obj, index, hdr_size_in_bytes, tmp1);
231 } else if (con_size_in_bytes <= threshold) {
232 // use explicit null stores
233 int i = hdr_size_in_bytes;
234 if (i < con_size_in_bytes && (con_size_in_bytes % (2 * BytesPerWord))) { // 2: multipler for BytesPerWord
235 sd(zr, Address(obj, i));
236 i += BytesPerWord;
237 }
238 for (; i < con_size_in_bytes; i += BytesPerWord) {
239 sd(zr, Address(obj, i));
240 }
241 } else if (con_size_in_bytes > hdr_size_in_bytes) {
242 block_comment("zero memory");
243 // use loop to null out the fields
244 int words = (con_size_in_bytes - hdr_size_in_bytes) / BytesPerWord;
245 mv(index, words / 8); // 8: byte size
246
247 const int unroll = 8; // Number of sd(zr) instructions we'll unroll
248 int remainder = words % unroll;
249 la(t0, Address(obj, hdr_size_in_bytes + remainder * BytesPerWord));
250
251 Label entry_point, loop;
252 j(entry_point);
253
254 bind(loop);
255 sub(index, index, 1);
256 for (int i = -unroll; i < 0; i++) {
257 if (-i == remainder) {
258 bind(entry_point);
259 }
260 sd(zr, Address(t0, i * wordSize));
261 }
262 if (remainder == 0) {
263 bind(entry_point);
264 }
265 add(t0, t0, unroll * wordSize);
266 bnez(index, loop);
267 }
268 }
269
270 membar(MacroAssembler::StoreStore);
271
272 if (CURRENT_ENV->dtrace_alloc_probes()) {
273 assert(obj == x10, "must be");
274 far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
275 }
276
277 verify_oop(obj);
278 }
279
280 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register tmp1, Register tmp2, int header_size, int f, Register klass, Label& slow_case) {
281 assert_different_registers(obj, len, tmp1, tmp2, klass);
282
283 // determine alignment mask
284 assert(!(BytesPerWord & 1), "must be multiple of 2 for masking code to work");
285
286 // check for negative or excessive length
287 mv(t0, (int32_t)max_array_allocation_length);
288 bgeu(len, t0, slow_case, /* is_far */ true);
289
290 const Register arr_size = tmp2; // okay to be the same
291 // align object end
292 mv(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
293 shadd(arr_size, len, arr_size, t0, f);
294 andi(arr_size, arr_size, ~(uint)MinObjAlignmentInBytesMask);
295
296 try_allocate(obj, arr_size, 0, tmp1, tmp2, slow_case);
297
298 initialize_header(obj, klass, len, tmp1, tmp2);
299
300 // clear rest of allocated space
301 const Register len_zero = len;
302 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
303
304 membar(MacroAssembler::StoreStore);
305
306 if (CURRENT_ENV->dtrace_alloc_probes()) {
307 assert(obj == x10, "must be");
308 far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
309 }
310
311 verify_oop(obj);
312 }
313
314 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache, Label &L) {
315 verify_oop(receiver);
316 // explicit NULL check not needed since load from [klass_offset] causes a trap
317 // check against inline cache
318 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
319 assert_different_registers(receiver, iCache, t0, t2);
320 cmp_klass(receiver, iCache, t0, t2 /* call-clobbered t2 as a tmp */, L);
321 }
322
323 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) {
324 assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
325 // Make sure there is enough stack space for this method's activation.
326 // Note that we do this before creating a frame.
327 generate_stack_overflow_check(bang_size_in_bytes);
328 MacroAssembler::build_frame(framesize);
329
330 // Insert nmethod entry barrier into frame.
331 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
332 bs->nmethod_entry_barrier(this);
333 }
334
335 void C1_MacroAssembler::remove_frame(int framesize) {
336 MacroAssembler::remove_frame(framesize);
337 }
338
339
340 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
341 // If we have to make this method not-entrant we'll overwrite its
342 // first instruction with a jump. For this action to be legal we
343 // must ensure that this first instruction is a J, JAL or NOP.
344 // Make it a NOP.
345 assert_alignment(pc());
346 nop();
347 }
348
349 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
350 // fp + -2: link
351 // + -1: return address
352 // + 0: argument with offset 0
353 // + 1: argument with offset 1
354 // + 2: ...
355 ld(reg, Address(fp, offset_in_words * BytesPerWord));
356 }
357
358 #ifndef PRODUCT
359
360 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
361 if (!VerifyOops) {
362 return;
363 }
364 verify_oop_addr(Address(sp, stack_offset), "oop");
365 }
366
367 void C1_MacroAssembler::verify_not_null_oop(Register r) {
368 if (!VerifyOops) return;
369 Label not_null;
370 bnez(r, not_null);
371 stop("non-null oop required");
372 bind(not_null);
373 verify_oop(r);
374 }
375
376 void C1_MacroAssembler::invalidate_registers(bool inv_x10, bool inv_x9, bool inv_x12, bool inv_x13, bool inv_x14, bool inv_x15) {
377 #ifdef ASSERT
378 static int nn;
379 if (inv_x10) { mv(x10, 0xDEAD); }
380 if (inv_x9) { mv(x9, 0xDEAD); }
381 if (inv_x12) { mv(x12, nn++); }
382 if (inv_x13) { mv(x13, 0xDEAD); }
383 if (inv_x14) { mv(x14, 0xDEAD); }
384 if (inv_x15) { mv(x15, 0xDEAD); }
385 #endif // ASSERT
386 }
387 #endif // ifndef PRODUCT
388
389 typedef void (C1_MacroAssembler::*c1_cond_branch_insn)(Register op1, Register op2, Label& label, bool is_far);
390 typedef void (C1_MacroAssembler::*c1_float_cond_branch_insn)(FloatRegister op1, FloatRegister op2,
391 Label& label, bool is_far, bool is_unordered);
392
393 static c1_cond_branch_insn c1_cond_branch[] =
394 {
395 /* SHORT branches */
396 (c1_cond_branch_insn)&Assembler::beq,
397 (c1_cond_branch_insn)&Assembler::bne,
398 (c1_cond_branch_insn)&Assembler::blt,
399 (c1_cond_branch_insn)&Assembler::ble,
400 (c1_cond_branch_insn)&Assembler::bge,
401 (c1_cond_branch_insn)&Assembler::bgt,
402 (c1_cond_branch_insn)&Assembler::bleu, // lir_cond_belowEqual
403 (c1_cond_branch_insn)&Assembler::bgeu // lir_cond_aboveEqual
404 };
405
406 static c1_float_cond_branch_insn c1_float_cond_branch[] =
407 {
408 /* FLOAT branches */
409 (c1_float_cond_branch_insn)&MacroAssembler::float_beq,
410 (c1_float_cond_branch_insn)&MacroAssembler::float_bne,
411 (c1_float_cond_branch_insn)&MacroAssembler::float_blt,
412 (c1_float_cond_branch_insn)&MacroAssembler::float_ble,
413 (c1_float_cond_branch_insn)&MacroAssembler::float_bge,
414 (c1_float_cond_branch_insn)&MacroAssembler::float_bgt,
415 NULL, // lir_cond_belowEqual
416 NULL, // lir_cond_aboveEqual
417
418 /* DOUBLE branches */
419 (c1_float_cond_branch_insn)&MacroAssembler::double_beq,
420 (c1_float_cond_branch_insn)&MacroAssembler::double_bne,
421 (c1_float_cond_branch_insn)&MacroAssembler::double_blt,
422 (c1_float_cond_branch_insn)&MacroAssembler::double_ble,
423 (c1_float_cond_branch_insn)&MacroAssembler::double_bge,
424 (c1_float_cond_branch_insn)&MacroAssembler::double_bgt,
425 NULL, // lir_cond_belowEqual
426 NULL // lir_cond_aboveEqual
427 };
428
429 void C1_MacroAssembler::c1_cmp_branch(int cmpFlag, Register op1, Register op2, Label& label,
430 BasicType type, bool is_far) {
431 if (type == T_OBJECT || type == T_ARRAY) {
432 assert(cmpFlag == lir_cond_equal || cmpFlag == lir_cond_notEqual, "Should be equal or notEqual");
433 if (cmpFlag == lir_cond_equal) {
434 beq(op1, op2, label, is_far);
435 } else {
436 bne(op1, op2, label, is_far);
437 }
438 } else {
439 assert(cmpFlag >= 0 && cmpFlag < (int)(sizeof(c1_cond_branch) / sizeof(c1_cond_branch[0])),
440 "invalid c1 conditional branch index");
441 (this->*c1_cond_branch[cmpFlag])(op1, op2, label, is_far);
442 }
443 }
444
445 void C1_MacroAssembler::c1_float_cmp_branch(int cmpFlag, FloatRegister op1, FloatRegister op2, Label& label,
446 bool is_far, bool is_unordered) {
447 assert(cmpFlag >= 0 &&
448 cmpFlag < (int)(sizeof(c1_float_cond_branch) / sizeof(c1_float_cond_branch[0])),
449 "invalid c1 float conditional branch index");
450 (this->*c1_float_cond_branch[cmpFlag])(op1, op2, label, is_far, is_unordered);
451 }