1 /*
2 * Copyright (c) 1999, 2021, 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/os.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/stubRoutines.hpp"
40
41 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
42 const Register rklass_decode_tmp = LP64_ONLY(rscratch1) NOT_LP64(noreg);
43 const int aligned_mask = BytesPerWord -1;
44 const int hdr_offset = oopDesc::mark_offset_in_bytes();
45 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
46 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
47 Label done;
48 int null_check_offset = -1;
49
50 verify_oop(obj);
51
52 // save object being locked into the BasicObjectLock
53 movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj);
54
55 null_check_offset = offset();
56
57 if (DiagnoseSyncOnValueBasedClasses != 0) {
58 load_klass(hdr, obj, rklass_decode_tmp);
59 movl(hdr, Address(hdr, Klass::access_flags_offset()));
60 testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
61 jcc(Assembler::notZero, slow_case);
62 }
63
64 if (UseBiasedLocking) {
65 assert(scratch != noreg, "should have scratch register at this point");
66 biased_locking_enter(disp_hdr, obj, hdr, scratch, rklass_decode_tmp, false, done, &slow_case);
67 }
68
69 // Load object header
70 movptr(hdr, Address(obj, hdr_offset));
71 // and mark it as unlocked
72 orptr(hdr, markWord::unlocked_value);
73 // save unlocked object header into the displaced header location on the stack
74 movptr(Address(disp_hdr, 0), hdr);
75 // test if object header is still the same (i.e. unlocked), and if so, store the
76 // displaced header address in the object header - if it is not the same, get the
77 // object header instead
78 MacroAssembler::lock(); // must be immediately before cmpxchg!
79 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
80 // if the object header was the same, we're done
81 if (PrintBiasedLockingStatistics) {
82 cond_inc32(Assembler::equal,
83 ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
84 }
85 jcc(Assembler::equal, done);
86 // if the object header was not the same, it is now in the hdr register
87 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
88 //
89 // 1) (hdr & aligned_mask) == 0
90 // 2) rsp <= hdr
91 // 3) hdr <= rsp + page_size
92 //
93 // these 3 tests can be done by evaluating the following expression:
94 //
95 // (hdr - rsp) & (aligned_mask - page_size)
96 //
97 // assuming both the stack pointer and page_size have their least
98 // significant 2 bits cleared and page_size is a power of 2
99 subptr(hdr, rsp);
100 andptr(hdr, aligned_mask - os::vm_page_size());
101 // for recursive locking, the result is zero => save it in the displaced header
102 // location (NULL in the displaced hdr location indicates recursive locking)
103 movptr(Address(disp_hdr, 0), hdr);
104 // otherwise we don't care about the result and handle locking via runtime call
105 jcc(Assembler::notZero, slow_case);
106 // done
107 bind(done);
108 return null_check_offset;
109 }
110
111
112 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
113 const int aligned_mask = BytesPerWord -1;
114 const int hdr_offset = oopDesc::mark_offset_in_bytes();
115 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
116 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
117 Label done;
118
119 if (UseBiasedLocking) {
120 // load object
121 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
122 biased_locking_exit(obj, hdr, done);
123 }
124
125 // load displaced header
126 movptr(hdr, Address(disp_hdr, 0));
127 // if the loaded hdr is NULL we had recursive locking
128 testptr(hdr, hdr);
129 // if we had recursive locking, we are done
130 jcc(Assembler::zero, done);
131 if (!UseBiasedLocking) {
132 // load object
133 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
134 }
135 verify_oop(obj);
136 // test if object header is pointing to the displaced header, and if so, restore
137 // the displaced header in the object - if the object header is not pointing to
138 // the displaced header, get the object header instead
139 MacroAssembler::lock(); // must be immediately before cmpxchg!
140 cmpxchgptr(hdr, Address(obj, hdr_offset));
141 // if the object header was not pointing to the displaced header,
142 // we do unlocking via runtime call
143 jcc(Assembler::notEqual, slow_case);
144 // done
145 bind(done);
146 }
147
148
149 // Defines obj, preserves var_size_in_bytes
150 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
151 if (UseTLAB) {
152 tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
153 } else {
154 eden_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
155 }
156 }
157
158
159 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
160 assert_different_registers(obj, klass, len);
161 Register tmp_encode_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
162 if (UseBiasedLocking && !len->is_valid()) {
163 assert_different_registers(obj, klass, len, t1, t2);
164 movptr(t1, Address(klass, Klass::prototype_header_offset()));
165 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
166 } else {
167 // This assumes that all prototype bits fit in an int32_t
168 movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markWord::prototype().value());
169 }
170 #ifdef _LP64
171 if (UseCompressedClassPointers) { // Take care not to kill klass
172 movptr(t1, klass);
173 encode_klass_not_null(t1, tmp_encode_klass);
174 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
175 } else
176 #endif
177 {
178 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
179 }
180
181 if (len->is_valid()) {
182 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
183 }
184 #ifdef _LP64
185 else if (UseCompressedClassPointers) {
186 xorptr(t1, t1);
187 store_klass_gap(obj, t1);
188 }
189 #endif
190 }
191
192
193 // preserves obj, destroys len_in_bytes
194 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
195 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
196 Label done;
197
198 // len_in_bytes is positive and ptr sized
199 subptr(len_in_bytes, hdr_size_in_bytes);
200 jcc(Assembler::zero, done);
201 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
202 bind(done);
203 }
204
205
206 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
207 assert(obj == rax, "obj must be in rax, for cmpxchg");
208 assert_different_registers(obj, t1, t2); // XXX really?
209 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
210
211 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
212
213 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
214 }
215
216 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) {
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, t1, t2);
222
223 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
224 // clear rest of allocated space
225 const Register t1_zero = t1;
226 const Register index = t2;
227 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
228 if (var_size_in_bytes != noreg) {
229 mov(index, var_size_in_bytes);
230 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
231 } else if (con_size_in_bytes <= threshold) {
232 // use explicit null stores
233 // code size = 2 + 3*n bytes (n = number of fields to clear)
234 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
235 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
236 movptr(Address(obj, i), t1_zero);
237 } else if (con_size_in_bytes > hdr_size_in_bytes) {
238 // use loop to null out the fields
239 // code size = 16 bytes for even n (n = number of fields to clear)
240 // initialize last object field first if odd number of fields
241 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
242 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
243 // initialize last object field if constant size is odd
244 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
245 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
246 // initialize remaining object fields: rdx is a multiple of 2
247 { Label loop;
248 bind(loop);
249 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
250 t1_zero);
251 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
252 t1_zero);)
253 decrement(index);
254 jcc(Assembler::notZero, loop);
255 }
256 }
257 }
258
259 if (CURRENT_ENV->dtrace_alloc_probes()) {
260 assert(obj == rax, "must be");
261 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
262 }
263
264 verify_oop(obj);
265 }
266
267 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
268 assert(obj == rax, "obj must be in rax, for cmpxchg");
269 assert_different_registers(obj, len, t1, t2, klass);
270
271 // determine alignment mask
272 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
273
274 // check for negative or excessive length
275 cmpptr(len, (int32_t)max_array_allocation_length);
276 jcc(Assembler::above, slow_case);
277
278 const Register arr_size = t2; // okay to be the same
279 // align object end
280 movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
281 lea(arr_size, Address(arr_size, len, f));
282 andptr(arr_size, ~MinObjAlignmentInBytesMask);
283
284 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
285
286 initialize_header(obj, klass, len, t1, t2);
287
288 // clear rest of allocated space
289 const Register len_zero = len;
290 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
291
292 if (CURRENT_ENV->dtrace_alloc_probes()) {
293 assert(obj == rax, "must be");
294 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
295 }
296
297 verify_oop(obj);
298 }
299
300
301
302 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
303 verify_oop(receiver);
304 // explicit NULL check not needed since load from [klass_offset] causes a trap
305 // check against inline cache
306 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
307 int start_offset = offset();
308 Register tmp_load_klass = LP64_ONLY(rscratch2) NOT_LP64(noreg);
309
310 if (UseCompressedClassPointers) {
311 load_klass(rscratch1, receiver, tmp_load_klass);
312 cmpptr(rscratch1, iCache);
313 } else {
314 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
315 }
316 // if icache check fails, then jump to runtime routine
317 // Note: RECEIVER must still contain the receiver!
318 jump_cc(Assembler::notEqual,
319 RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
320 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
321 assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
322 }
323
324
325 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
326 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
327 // Make sure there is enough stack space for this method's activation.
328 // Note that we do this before doing an enter(). This matches the
329 // ordering of C2's stack overflow check / rsp decrement and allows
330 // the SharedRuntime stack overflow handling to be consistent
331 // between the two compilers.
332 generate_stack_overflow_check(bang_size_in_bytes);
333
334 push(rbp);
335 if (PreserveFramePointer) {
336 mov(rbp, rsp);
337 }
338 #if !defined(_LP64) && defined(COMPILER2)
339 if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
340 // c2 leaves fpu stack dirty. Clean it on entry
341 empty_FPU_stack();
342 }
343 #endif // !_LP64 && COMPILER2
344 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
345
346 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
347 bs->nmethod_entry_barrier(this);
348 }
349
350
351 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
352 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
353 pop(rbp);
354 }
355
356
357 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
358 if (breakAtEntry || VerifyFPU) {
359 // Verified Entry first instruction should be 5 bytes long for correct
360 // patching by patch_verified_entry().
361 //
362 // Breakpoint and VerifyFPU have one byte first instruction.
363 // Also first instruction will be one byte "push(rbp)" if stack banging
364 // code is not generated (see build_frame() above).
365 // For all these cases generate long instruction first.
366 fat_nop();
367 }
368 if (breakAtEntry) int3();
369 // build frame
370 IA32_ONLY( verify_FPU(0, "method_entry"); )
371 }
372
373 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
374 // rbp, + 0: link
375 // + 1: return address
376 // + 2: argument with offset 0
377 // + 3: argument with offset 1
378 // + 4: ...
379
380 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
381 }
382
383 #ifndef PRODUCT
384
385 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
386 if (!VerifyOops) return;
387 verify_oop_addr(Address(rsp, stack_offset));
388 }
389
390 void C1_MacroAssembler::verify_not_null_oop(Register r) {
391 if (!VerifyOops) return;
392 Label not_null;
393 testptr(r, r);
394 jcc(Assembler::notZero, not_null);
395 stop("non-null oop required");
396 bind(not_null);
397 verify_oop(r);
398 }
399
400 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
401 #ifdef ASSERT
402 if (inv_rax) movptr(rax, 0xDEAD);
403 if (inv_rbx) movptr(rbx, 0xDEAD);
404 if (inv_rcx) movptr(rcx, 0xDEAD);
405 if (inv_rdx) movptr(rdx, 0xDEAD);
406 if (inv_rsi) movptr(rsi, 0xDEAD);
407 if (inv_rdi) movptr(rdi, 0xDEAD);
408 #endif
409 }
410
411 #endif // ifndef PRODUCT