1 /*
2 * Copyright (c) 1999, 2018, 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 "classfile/systemDictionary.hpp"
29 #include "gc/shared/collectedHeap.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "oops/arrayOop.hpp"
32 #include "oops/markOop.hpp"
33 #include "runtime/basicLock.hpp"
34 #include "runtime/biasedLocking.hpp"
35 #include "runtime/os.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38
39 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
40 const int aligned_mask = BytesPerWord -1;
41 const int hdr_offset = oopDesc::mark_offset_in_bytes();
42 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
43 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
44 Label done;
45 int null_check_offset = -1;
46
47 verify_oop(obj);
48
49 // save object being locked into the BasicObjectLock
50 movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj);
51
52 if (UseBiasedLocking) {
53 assert(scratch != noreg, "should have scratch register at this point");
54 null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case);
55 } else {
56 null_check_offset = offset();
57 }
58
59 // Load object header
60 movptr(hdr, Address(obj, hdr_offset));
61 // and mark it as unlocked
62 orptr(hdr, markOopDesc::unlocked_value);
63 // save unlocked object header into the displaced header location on the stack
64 movptr(Address(disp_hdr, 0), hdr);
65 // test if object header is still the same (i.e. unlocked), and if so, store the
66 // displaced header address in the object header - if it is not the same, get the
67 // object header instead
68 MacroAssembler::lock(); // must be immediately before cmpxchg!
69 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
70 // if the object header was the same, we're done
71 if (PrintBiasedLockingStatistics) {
72 cond_inc32(Assembler::equal,
73 ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
74 }
75 jcc(Assembler::equal, done);
76 // if the object header was not the same, it is now in the hdr register
77 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
78 //
79 // 1) (hdr & aligned_mask) == 0
80 // 2) rsp <= hdr
81 // 3) hdr <= rsp + page_size
82 //
83 // these 3 tests can be done by evaluating the following expression:
84 //
85 // (hdr - rsp) & (aligned_mask - page_size)
86 //
87 // assuming both the stack pointer and page_size have their least
88 // significant 2 bits cleared and page_size is a power of 2
89 subptr(hdr, rsp);
90 andptr(hdr, aligned_mask - os::vm_page_size());
91 // for recursive locking, the result is zero => save it in the displaced header
92 // location (NULL in the displaced hdr location indicates recursive locking)
93 movptr(Address(disp_hdr, 0), hdr);
94 // otherwise we don't care about the result and handle locking via runtime call
95 jcc(Assembler::notZero, slow_case);
96 // done
97 bind(done);
98 return null_check_offset;
99 }
100
101
102 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
103 const int aligned_mask = BytesPerWord -1;
104 const int hdr_offset = oopDesc::mark_offset_in_bytes();
105 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
106 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
107 Label done;
108
109 if (UseBiasedLocking) {
110 // load object
111 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
112 biased_locking_exit(obj, hdr, done);
113 }
114
115 // load displaced header
116 movptr(hdr, Address(disp_hdr, 0));
117 // if the loaded hdr is NULL we had recursive locking
118 testptr(hdr, hdr);
119 // if we had recursive locking, we are done
120 jcc(Assembler::zero, done);
121 if (!UseBiasedLocking) {
122 // load object
123 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
124 }
125 verify_oop(obj);
126 // test if object header is pointing to the displaced header, and if so, restore
127 // the displaced header in the object - if the object header is not pointing to
128 // the displaced header, get the object header instead
129 MacroAssembler::lock(); // must be immediately before cmpxchg!
130 cmpxchgptr(hdr, Address(obj, hdr_offset));
131 // if the object header was not pointing to the displaced header,
132 // we do unlocking via runtime call
133 jcc(Assembler::notEqual, slow_case);
134 // done
135 bind(done);
136 }
137
138
139 // Defines obj, preserves var_size_in_bytes
140 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
141 if (UseTLAB) {
142 tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
143 } else {
144 eden_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
145 }
146 }
147
148
149 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
150 assert_different_registers(obj, klass, len);
151 if (UseBiasedLocking && !len->is_valid()) {
152 assert_different_registers(obj, klass, len, t1, t2);
153 movptr(t1, Address(klass, Klass::prototype_header_offset()));
154 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
155 } else {
156 // This assumes that all prototype bits fit in an int32_t
157 movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markOopDesc::prototype());
158 }
159 #ifdef _LP64
160 if (UseCompressedClassPointers) { // Take care not to kill klass
161 movptr(t1, klass);
162 encode_klass_not_null(t1);
163 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
164 } else
165 #endif
166 {
167 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
168 }
169
170 if (len->is_valid()) {
171 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
172 }
173 #ifdef _LP64
174 else if (UseCompressedClassPointers) {
175 xorptr(t1, t1);
176 store_klass_gap(obj, t1);
177 }
178 #endif
179 }
180
181
182 // preserves obj, destroys len_in_bytes
183 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
184 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
185 Label done;
186
187 // len_in_bytes is positive and ptr sized
188 subptr(len_in_bytes, hdr_size_in_bytes);
189 jcc(Assembler::zero, done);
190 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
191 bind(done);
192 }
193
194
195 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
196 assert(obj == rax, "obj must be in rax, for cmpxchg");
197 assert_different_registers(obj, t1, t2); // XXX really?
198 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
199
200 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
201
202 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
203 }
204
205 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) {
206 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
207 "con_size_in_bytes is not multiple of alignment");
208 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
209
210 initialize_header(obj, klass, noreg, t1, t2);
211
212 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
213 // clear rest of allocated space
214 const Register t1_zero = t1;
215 const Register index = t2;
216 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
217 if (var_size_in_bytes != noreg) {
218 mov(index, var_size_in_bytes);
219 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
220 } else if (con_size_in_bytes <= threshold) {
221 // use explicit null stores
222 // code size = 2 + 3*n bytes (n = number of fields to clear)
223 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
224 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
225 movptr(Address(obj, i), t1_zero);
226 } else if (con_size_in_bytes > hdr_size_in_bytes) {
227 // use loop to null out the fields
228 // code size = 16 bytes for even n (n = number of fields to clear)
229 // initialize last object field first if odd number of fields
230 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
231 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
232 // initialize last object field if constant size is odd
233 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
234 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
235 // initialize remaining object fields: rdx is a multiple of 2
236 { Label loop;
237 bind(loop);
238 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
239 t1_zero);
240 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
241 t1_zero);)
242 decrement(index);
243 jcc(Assembler::notZero, loop);
244 }
245 }
246 }
247
248 if (CURRENT_ENV->dtrace_alloc_probes()) {
249 assert(obj == rax, "must be");
250 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
251 }
252
253 verify_oop(obj);
254 }
255
256 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
257 assert(obj == rax, "obj must be in rax, for cmpxchg");
258 assert_different_registers(obj, len, t1, t2, klass);
259
260 // determine alignment mask
261 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
262
263 // check for negative or excessive length
264 cmpptr(len, (int32_t)max_array_allocation_length);
265 jcc(Assembler::above, slow_case);
266
267 const Register arr_size = t2; // okay to be the same
268 // align object end
269 movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
270 lea(arr_size, Address(arr_size, len, f));
271 andptr(arr_size, ~MinObjAlignmentInBytesMask);
272
273 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
274
275 initialize_header(obj, klass, len, t1, t2);
276
277 // clear rest of allocated space
278 const Register len_zero = len;
279 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
280
281 if (CURRENT_ENV->dtrace_alloc_probes()) {
282 assert(obj == rax, "must be");
283 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
284 }
285
286 verify_oop(obj);
287 }
288
289
290
291 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
292 verify_oop(receiver);
293 // explicit NULL check not needed since load from [klass_offset] causes a trap
294 // check against inline cache
295 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
296 int start_offset = offset();
297
298 if (UseCompressedClassPointers) {
299 load_klass(rscratch1, receiver);
300 cmpptr(rscratch1, iCache);
301 } else {
302 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
303 }
304 // if icache check fails, then jump to runtime routine
305 // Note: RECEIVER must still contain the receiver!
306 jump_cc(Assembler::notEqual,
307 RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
308 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
309 assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
310 }
311
312
313 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
314 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
315 // Make sure there is enough stack space for this method's activation.
316 // Note that we do this before doing an enter(). This matches the
317 // ordering of C2's stack overflow check / rsp decrement and allows
318 // the SharedRuntime stack overflow handling to be consistent
319 // between the two compilers.
320 generate_stack_overflow_check(bang_size_in_bytes);
321
322 push(rbp);
323 if (PreserveFramePointer) {
324 mov(rbp, rsp);
325 }
326 #ifdef TIERED
327 // c2 leaves fpu stack dirty. Clean it on entry
328 if (UseSSE < 2 ) {
329 empty_FPU_stack();
330 }
331 #endif // TIERED
332 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
333 }
334
335
336 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
337 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
338 pop(rbp);
339 }
340
341
342 void C1_MacroAssembler::verified_entry() {
343 if (C1Breakpoint || VerifyFPU || !UseStackBanging) {
344 // Verified Entry first instruction should be 5 bytes long for correct
345 // patching by patch_verified_entry().
346 //
347 // C1Breakpoint and VerifyFPU have one byte first instruction.
348 // Also first instruction will be one byte "push(rbp)" if stack banging
349 // code is not generated (see build_frame() above).
350 // For all these cases generate long instruction first.
351 fat_nop();
352 }
353 if (C1Breakpoint)int3();
354 // build frame
355 verify_FPU(0, "method_entry");
356 }
357
358 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
359 // rbp, + 0: link
360 // + 1: return address
361 // + 2: argument with offset 0
362 // + 3: argument with offset 1
363 // + 4: ...
364
365 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
366 }
367
368 #ifndef PRODUCT
369
370 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
371 if (!VerifyOops) return;
372 verify_oop_addr(Address(rsp, stack_offset));
373 }
374
375 void C1_MacroAssembler::verify_not_null_oop(Register r) {
376 if (!VerifyOops) return;
377 Label not_null;
378 testptr(r, r);
379 jcc(Assembler::notZero, not_null);
380 stop("non-null oop required");
381 bind(not_null);
382 verify_oop(r);
383 }
384
385 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
386 #ifdef ASSERT
387 if (inv_rax) movptr(rax, 0xDEAD);
388 if (inv_rbx) movptr(rbx, 0xDEAD);
389 if (inv_rcx) movptr(rcx, 0xDEAD);
390 if (inv_rdx) movptr(rdx, 0xDEAD);
391 if (inv_rsi) movptr(rsi, 0xDEAD);
392 if (inv_rdi) movptr(rdi, 0xDEAD);
393 #endif
394 }
395
396 #endif // ifndef PRODUCT